]> git.ipfire.org Git - people/pmueller/ipfire-2.x.git/blame - src/patches/60031_xen3-patch-2.6.22.patch1
projects / people / pmueller / ipfire-2.x.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Fix core28 updater kernel version
[people/pmueller/ipfire-2.x.git] / src / patches / 60031_xen3-patch-2.6.22.patch1
CommitLineData
cc90b958
BS		 1From: www.kernel.org
2Subject: Update to 2.6.22
3Patch-mainline: 2.6.22
4
5Automatically created from "patches.kernel.org/patch-2.6.22" by xen-port-patches.py
6
7Acked-by: jbeulich@novell.com
8
9---
10 arch/x86/Kconfig | 4
11 arch/x86/ia32/ia32entry-xen.S | 18 -
12 arch/x86/kernel/Makefile | 2
13 arch/x86/kernel/acpi/sleep_64-xen.c | 26 -
14 arch/x86/kernel/apic_32-xen.c | 1
15 arch/x86/kernel/apic_64-xen.c | 1
16 arch/x86/kernel/asm-offsets_32.c | 5
17 arch/x86/kernel/cpu/common-xen.c | 224 ++++---------
18 arch/x86/kernel/cpu/mtrr/main-xen.c | 2
19 arch/x86/kernel/e820_32-xen.c | 46 +-
20 arch/x86/kernel/e820_64-xen.c | 28 -
21 arch/x86/kernel/early_printk-xen.c | 27 -
22 arch/x86/kernel/entry_32-xen.S | 30 -
23 arch/x86/kernel/entry_64-xen.S | 7
24 arch/x86/kernel/genapic_64-xen.c | 108 +-----
25 arch/x86/kernel/genapic_xen_64.c | 3
26 arch/x86/kernel/head64-xen.c | 32 +
27 arch/x86/kernel/head_32-xen.S | 101 ------
28 arch/x86/kernel/head_64-xen.S | 52 ---
29 arch/x86/kernel/io_apic_32-xen.c | 43 --
30 arch/x86/kernel/io_apic_64-xen.c | 39 --
31 arch/x86/kernel/ioport_32-xen.c | 2
32 arch/x86/kernel/ioport_64-xen.c | 2
33 arch/x86/kernel/irq_32-xen.c | 3
34 arch/x86/kernel/irq_64-xen.c | 34 +-
35 arch/x86/kernel/ldt_32-xen.c | 1
36 arch/x86/kernel/ldt_64-xen.c | 1
37 arch/x86/kernel/microcode-xen.c | 2
38 arch/x86/kernel/mpparse_32-xen.c | 3
39 arch/x86/kernel/mpparse_64-xen.c | 3
40 arch/x86/kernel/pci-dma-xen.c | 29 +
41 arch/x86/kernel/process_32-xen.c | 27 +
42 arch/x86/kernel/process_64-xen.c | 16
43 arch/x86/kernel/quirks-xen.c | 63 ---
44 arch/x86/kernel/setup64-xen.c | 17 -
45 arch/x86/kernel/setup_64-xen.c | 30 -
46 arch/x86/kernel/smp_32-xen.c | 191 ++++-------
47 arch/x86/kernel/smp_64-xen.c | 29 -
48 arch/x86/kernel/time_32-xen.c | 165 ++++++----
49 arch/x86/kernel/traps_32-xen.c | 46 +-
50 arch/x86/kernel/traps_64-xen.c | 55 +--
51 arch/x86/kernel/vsyscall_64-xen.c | 73 +++-
52 arch/x86/mm/fault_32-xen.c | 42 +-
53 arch/x86/mm/fault_64-xen.c | 15
54 arch/x86/mm/highmem_32-xen.c | 14
55 arch/x86/mm/init_32-xen.c | 157 ++++++---
56 arch/x86/mm/init_64-xen.c | 132 ++++----
57 arch/x86/mm/ioremap_32-xen.c | 1
58 arch/x86/mm/pageattr_64-xen.c | 27 +
59 arch/x86/mm/pgtable_32-xen.c | 206 +++++++-----
60 drivers/char/tpm/tpm_xen.c | 2
61 drivers/pci/msi-xen.c | 127 +++++--
62 drivers/xen/blkfront/blkfront.c | 2
63 drivers/xen/char/mem.c | 1
64 drivers/xen/core/hypervisor_sysfs.c | 2
65 drivers/xen/core/smpboot.c | 45 +-
66 drivers/xen/core/xen_sysfs.c | 24 -
67 drivers/xen/netback/netback.c | 14
68 drivers/xen/netfront/netfront.c | 2
69 drivers/xen/pciback/xenbus.c | 2
70 drivers/xen/pcifront/xenbus.c | 4
71 drivers/xen/scsifront/xenbus.c | 2
72 drivers/xen/sfc_netback/accel_fwd.c | 7
73 drivers/xen/sfc_netback/accel_solarflare.c | 2
74 drivers/xen/sfc_netfront/accel_tso.c | 28 -
75 drivers/xen/sfc_netfront/accel_vi.c | 4
76 drivers/xen/sfc_netfront/accel_xenbus.c | 4
77 fs/aio.c | 7
78 include/asm-x86/mach-xen/asm/desc_32.h | 119 ++++---
79 include/asm-x86/mach-xen/asm/desc_64.h | 30 -
80 include/asm-x86/mach-xen/asm/dma-mapping_64.h | 2
81 include/asm-x86/mach-xen/asm/fixmap_32.h | 9
82 include/asm-x86/mach-xen/asm/fixmap_64.h | 1
83 include/asm-x86/mach-xen/asm/highmem.h | 6
84 include/asm-x86/mach-xen/asm/io_32.h | 13
85 include/asm-x86/mach-xen/asm/irqflags_32.h | 75 ++--
86 include/asm-x86/mach-xen/asm/irqflags_64.h | 19 -
87 include/asm-x86/mach-xen/asm/mmu_context_32.h | 29 +
88 include/asm-x86/mach-xen/asm/mmu_context_64.h | 3
89 include/asm-x86/mach-xen/asm/page_64.h | 61 +--
90 include/asm-x86/mach-xen/asm/pgalloc_32.h | 3
91 include/asm-x86/mach-xen/asm/pgalloc_64.h | 15
92 include/asm-x86/mach-xen/asm/pgtable-3level-defs.h | 2
93 include/asm-x86/mach-xen/asm/pgtable-3level.h | 61 ++-
94 include/asm-x86/mach-xen/asm/pgtable_32.h | 80 ++--
95 include/asm-x86/mach-xen/asm/pgtable_64.h | 83 ++---
96 include/asm-x86/mach-xen/asm/processor_32.h | 141 +++-----
97 include/asm-x86/mach-xen/asm/processor_64.h | 55 ---
98 include/asm-x86/mach-xen/asm/segment_32.h | 10
99 include/asm-x86/mach-xen/asm/smp_32.h | 117 +++++--
100 include/asm-x86/mach-xen/asm/smp_64.h | 20 -
101 include/asm-x86/mach-xen/asm/system_32.h | 342 ++++-----------------
102 include/asm-x86/mach-xen/asm/system_64.h | 106 ------
103 include/asm-x86/mach-xen/asm/tlbflush_32.h | 11
104 include/asm-x86/mach-xen/asm/tlbflush_64.h | 2
105 include/linux/pci.h | 2
106 lib/swiotlb-xen.c | 1
107 net/core/dev.c | 15
108 scripts/Makefile.xen.awk | 2
109 99 files changed, 1771 insertions(+), 2128 deletions(-)
110
111--- a/arch/x86/ia32/ia32entry-xen.S
112+++ b/arch/x86/ia32/ia32entry-xen.S
113@@ -431,11 +431,7 @@ ia32_sys_call_table:
114 .quad sys_symlink
115 .quad sys_lstat
116 .quad sys_readlink /* 85 */
117-#ifdef CONFIG_IA32_AOUT
118 .quad sys_uselib
119-#else
120- .quad quiet_ni_syscall
121-#endif
122 .quad sys_swapon
123 .quad sys_reboot
124 .quad compat_sys_old_readdir
125@@ -574,7 +570,7 @@ ia32_sys_call_table:
126 .quad quiet_ni_syscall /* tux */
127 .quad quiet_ni_syscall /* security */
128 .quad sys_gettid
129- .quad sys_readahead /* 225 */
130+ .quad sys32_readahead /* 225 */
131 .quad sys_setxattr
132 .quad sys_lsetxattr
133 .quad sys_fsetxattr
134@@ -599,7 +595,7 @@ ia32_sys_call_table:
135 .quad compat_sys_io_getevents
136 .quad compat_sys_io_submit
137 .quad sys_io_cancel
138- .quad sys_fadvise64 /* 250 */
139+ .quad sys32_fadvise64 /* 250 */
140 .quad quiet_ni_syscall /* free_huge_pages */
141 .quad sys_exit_group
142 .quad sys32_lookup_dcookie
143@@ -663,10 +659,14 @@ ia32_sys_call_table:
144 .quad compat_sys_set_robust_list
145 .quad compat_sys_get_robust_list
146 .quad sys_splice
147- .quad sys_sync_file_range
148- .quad sys_tee
149+ .quad sys32_sync_file_range
150+ .quad sys_tee /* 315 */
151 .quad compat_sys_vmsplice
152 .quad compat_sys_move_pages
153 .quad sys_getcpu
154 .quad sys_epoll_pwait
155-ia32_syscall_end:
156+ .quad compat_sys_utimensat /* 320 */
157+ .quad compat_sys_signalfd
158+ .quad compat_sys_timerfd
159+ .quad sys_eventfd
160+ia32_syscall_end:
161--- a/arch/x86/Kconfig
162+++ b/arch/x86/Kconfig
163@@ -1429,7 +1429,7 @@ config PHYSICAL_START
164
165 config RELOCATABLE
166 bool "Build a relocatable kernel (EXPERIMENTAL)"
167- depends on EXPERIMENTAL && !X86_XEN
168+ depends on EXPERIMENTAL && !X86_XEN && !X86_64_XEN
169 help
170 This builds a kernel image that retains relocation information
171 so it can be loaded someplace besides the default 1MB.
172@@ -1483,7 +1483,6 @@ config COMPAT_VDSO
173 def_bool y
174 prompt "Compat VDSO support"
175 depends on X86_32 || IA32_EMULATION
176- depends on !X86_XEN
177 help
178 Map the 32-bit VDSO to the predictable old-style address too.
179 ---help---
180@@ -1662,6 +1661,7 @@ config PCI
181 bool "PCI support"
182 default y
183 select ARCH_SUPPORTS_MSI if (X86_LOCAL_APIC && X86_IO_APIC)
184+ select ARCH_SUPPORTS_MSI if (XEN_UNPRIVILEGED_GUEST && XEN_PCIDEV_FRONTEND)
185 help
186 Find out whether you have a PCI motherboard. PCI is the name of a
187 bus system, i.e. the way the CPU talks to the other stuff inside
188--- a/arch/x86/kernel/acpi/sleep_64-xen.c
189+++ b/arch/x86/kernel/acpi/sleep_64-xen.c
190@@ -60,19 +60,6 @@ unsigned long acpi_video_flags;
191 extern char wakeup_start, wakeup_end;
192
193 extern unsigned long acpi_copy_wakeup_routine(unsigned long);
194-
195-static pgd_t low_ptr;
196-
197-static void init_low_mapping(void)
198-{
199- pgd_t *slot0 = pgd_offset(current->mm, 0UL);
200- low_ptr = *slot0;
201- /* FIXME: We're playing with the current task's page tables here, which
202- * is potentially dangerous on SMP systems.
203- */
204- set_pgd(slot0, *pgd_offset(current->mm, PAGE_OFFSET));
205- local_flush_tlb();
206-}
207 #endif
208
209 /**
210@@ -84,8 +71,6 @@ static void init_low_mapping(void)
211 int acpi_save_state_mem(void)
212 {
213 #ifndef CONFIG_ACPI_PV_SLEEP
214- init_low_mapping();
215-
216 memcpy((void *)acpi_wakeup_address, &wakeup_start,
217 &wakeup_end - &wakeup_start);
218 acpi_copy_wakeup_routine(acpi_wakeup_address);
219@@ -98,10 +83,6 @@ int acpi_save_state_mem(void)
220 */
221 void acpi_restore_state_mem(void)
222 {
223-#ifndef CONFIG_ACPI_PV_SLEEP
224- set_pgd(pgd_offset(current->mm, 0UL), low_ptr);
225- local_flush_tlb();
226-#endif
227 }
228
229 /**
230@@ -115,10 +96,11 @@ void acpi_restore_state_mem(void)
231 void __init acpi_reserve_bootmem(void)
232 {
233 #ifndef CONFIG_ACPI_PV_SLEEP
234- acpi_wakeup_address = (unsigned long)alloc_bootmem_low(PAGE_SIZE);
235- if ((&wakeup_end - &wakeup_start) > PAGE_SIZE)
236+ acpi_wakeup_address = (unsigned long)alloc_bootmem_low(PAGE_SIZE*2);
237+ if ((&wakeup_end - &wakeup_start) > (PAGE_SIZE*2))
238 printk(KERN_CRIT
239- "ACPI: Wakeup code way too big, will crash on attempt to suspend\n");
240+ "ACPI: Wakeup code way too big, will crash on attempt"
241+ " to suspend\n");
242 #endif
243 }
244
245--- a/arch/x86/kernel/apic_32-xen.c
246+++ b/arch/x86/kernel/apic_32-xen.c
247@@ -19,7 +19,6 @@
248 #include <linux/mm.h>
249 #include <linux/delay.h>
250 #include <linux/bootmem.h>
251-#include <linux/smp_lock.h>
252 #include <linux/interrupt.h>
253 #include <linux/mc146818rtc.h>
254 #include <linux/kernel_stat.h>
255--- a/arch/x86/kernel/apic_64-xen.c
256+++ b/arch/x86/kernel/apic_64-xen.c
257@@ -19,7 +19,6 @@
258 #include <linux/mm.h>
259 #include <linux/delay.h>
260 #include <linux/bootmem.h>
261-#include <linux/smp_lock.h>
262 #include <linux/interrupt.h>
263 #include <linux/mc146818rtc.h>
264 #include <linux/kernel_stat.h>
265--- a/arch/x86/kernel/asm-offsets_32.c
266+++ b/arch/x86/kernel/asm-offsets_32.c
267@@ -109,11 +109,6 @@ void foo(void)
268
269 OFFSET(crypto_tfm_ctx_offset, crypto_tfm, __crt_ctx);
270
271-#ifdef CONFIG_XEN
272- BLANK();
273- OFFSET(XEN_START_mfn_list, start_info, mfn_list);
274-#endif
275-
276 #ifdef CONFIG_PARAVIRT
277 BLANK();
278 OFFSET(PARAVIRT_enabled, pv_info, paravirt_enabled);
279--- a/arch/x86/kernel/cpu/common-xen.c
280+++ b/arch/x86/kernel/cpu/common-xen.c
281@@ -22,16 +22,40 @@
282 #define phys_pkg_id(a,b) a
283 #endif
284 #endif
285-#include <asm/pda.h>
286 #include <asm/hypervisor.h>
287
288 #include "cpu.h"
289
290-DEFINE_PER_CPU(struct Xgt_desc_struct, cpu_gdt_descr);
291-EXPORT_PER_CPU_SYMBOL(cpu_gdt_descr);
292+DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
293+ [GDT_ENTRY_KERNEL_CS] = { 0x0000ffff, 0x00cf9a00 },
294+ [GDT_ENTRY_KERNEL_DS] = { 0x0000ffff, 0x00cf9200 },
295+ [GDT_ENTRY_DEFAULT_USER_CS] = { 0x0000ffff, 0x00cffa00 },
296+ [GDT_ENTRY_DEFAULT_USER_DS] = { 0x0000ffff, 0x00cff200 },
297+#ifndef CONFIG_XEN
298+ /*
299+ * Segments used for calling PnP BIOS have byte granularity.
300+ * They code segments and data segments have fixed 64k limits,
301+ * the transfer segment sizes are set at run time.
302+ */
303+ [GDT_ENTRY_PNPBIOS_CS32] = { 0x0000ffff, 0x00409a00 },/* 32-bit code */
304+ [GDT_ENTRY_PNPBIOS_CS16] = { 0x0000ffff, 0x00009a00 },/* 16-bit code */
305+ [GDT_ENTRY_PNPBIOS_DS] = { 0x0000ffff, 0x00009200 }, /* 16-bit data */
306+ [GDT_ENTRY_PNPBIOS_TS1] = { 0x00000000, 0x00009200 },/* 16-bit data */
307+ [GDT_ENTRY_PNPBIOS_TS2] = { 0x00000000, 0x00009200 },/* 16-bit data */
308+ /*
309+ * The APM segments have byte granularity and their bases
310+ * are set at run time. All have 64k limits.
311+ */
312+ [GDT_ENTRY_APMBIOS_BASE] = { 0x0000ffff, 0x00409a00 },/* 32-bit code */
313+ /* 16-bit code */
314+ [GDT_ENTRY_APMBIOS_BASE+1] = { 0x0000ffff, 0x00009a00 },
315+ [GDT_ENTRY_APMBIOS_BASE+2] = { 0x0000ffff, 0x00409200 }, /* data */
316
317-struct i386_pda *_cpu_pda[NR_CPUS] __read_mostly;
318-EXPORT_SYMBOL(_cpu_pda);
319+ [GDT_ENTRY_ESPFIX_SS] = { 0x00000000, 0x00c09200 },
320+#endif
321+ [GDT_ENTRY_PERCPU] = { 0x00000000, 0x00000000 },
322+} };
323+EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
324
325 static int cachesize_override __cpuinitdata = -1;
326 static int disable_x86_fxsr __cpuinitdata;
327@@ -373,7 +397,7 @@ __setup("serialnumber", x86_serial_nr_se
328 /*
329 * This does the hard work of actually picking apart the CPU stuff...
330 */
331-void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
332+static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
333 {
334 int i;
335
336@@ -484,15 +508,22 @@ void __cpuinit identify_cpu(struct cpuin
337
338 /* Init Machine Check Exception if available. */
339 mcheck_init(c);
340+}
341
342- if (c == &boot_cpu_data)
343- sysenter_setup();
344+void __init identify_boot_cpu(void)
345+{
346+ identify_cpu(&boot_cpu_data);
347+ sysenter_setup();
348 enable_sep_cpu();
349+ mtrr_bp_init();
350+}
351
352- if (c == &boot_cpu_data)
353- mtrr_bp_init();
354- else
355- mtrr_ap_init();
356+void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
357+{
358+ BUG_ON(c == &boot_cpu_data);
359+ identify_cpu(c);
360+ enable_sep_cpu();
361+ mtrr_ap_init();
362 }
363
364 #ifdef CONFIG_X86_HT
365@@ -606,136 +637,47 @@ void __init early_cpu_init(void)
366 #endif
367 }
368
369-/* Make sure %gs is initialized properly in idle threads */
370+/* Make sure %fs is initialized properly in idle threads */
371 struct pt_regs * __devinit idle_regs(struct pt_regs *regs)
372 {
373 memset(regs, 0, sizeof(struct pt_regs));
374- regs->xfs = __KERNEL_PDA;
375+ regs->xfs = __KERNEL_PERCPU;
376 return regs;
377 }
378
379-static __cpuinit int alloc_gdt(int cpu)
380+/* Current gdt points %fs at the "master" per-cpu area: after this,
381+ * it's on the real one. */
382+void switch_to_new_gdt(void)
383 {
384- struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
385- struct desc_struct *gdt;
386- struct i386_pda *pda;
387-
388- gdt = (struct desc_struct *)cpu_gdt_descr->address;
389- pda = cpu_pda(cpu);
390-
391- /*
392- * This is a horrible hack to allocate the GDT. The problem
393- * is that cpu_init() is called really early for the boot CPU
394- * (and hence needs bootmem) but much later for the secondary
395- * CPUs, when bootmem will have gone away
396- */
397- if (NODE_DATA(0)->bdata->node_bootmem_map) {
398- BUG_ON(gdt != NULL || pda != NULL);
399-
400- gdt = alloc_bootmem_pages(PAGE_SIZE);
401- pda = alloc_bootmem(sizeof(*pda));
402- /* alloc_bootmem(_pages) panics on failure, so no check */
403-
404- memset(gdt, 0, PAGE_SIZE);
405- memset(pda, 0, sizeof(*pda));
406- } else {
407- /* GDT and PDA might already have been allocated if
408- this is a CPU hotplug re-insertion. */
409- if (gdt == NULL)
410- gdt = (struct desc_struct *)get_zeroed_page(GFP_KERNEL);
411-
412- if (pda == NULL)
413- pda = kmalloc_node(sizeof(*pda), GFP_KERNEL, cpu_to_node(cpu));
414-
415- if (unlikely(!gdt || !pda)) {
416- free_pages((unsigned long)gdt, 0);
417- kfree(pda);
418- return 0;
419- }
420- }
421-
422- cpu_gdt_descr->address = (unsigned long)gdt;
423- cpu_pda(cpu) = pda;
424-
425- return 1;
426-}
427-
428-/* Initial PDA used by boot CPU */
429-struct i386_pda boot_pda = {
430- ._pda = &boot_pda,
431- .cpu_number = 0,
432- .pcurrent = &init_task,
433-};
434-
435-static inline void set_kernel_fs(void)
436-{
437- /* Set %fs for this CPU's PDA. Memory clobber is to create a
438- barrier with respect to any PDA operations, so the compiler
439- doesn't move any before here. */
440- asm volatile ("mov %0, %%fs" : : "r" (__KERNEL_PDA) : "memory");
441-}
442-
443-/* Initialize the CPU's GDT and PDA. The boot CPU does this for
444- itself, but secondaries find this done for them. */
445-__cpuinit int init_gdt(int cpu, struct task_struct *idle)
446-{
447- struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
448- struct desc_struct *gdt;
449- struct i386_pda *pda;
450-
451- /* For non-boot CPUs, the GDT and PDA should already have been
452- allocated. */
453- if (!alloc_gdt(cpu)) {
454- printk(KERN_CRIT "CPU%d failed to allocate GDT or PDA\n", cpu);
455- return 0;
456- }
457-
458- gdt = (struct desc_struct *)cpu_gdt_descr->address;
459- pda = cpu_pda(cpu);
460-
461- BUG_ON(gdt == NULL || pda == NULL);
462-
463- /*
464- * Initialize the per-CPU GDT with the boot GDT,
465- * and set up the GDT descriptor:
466- */
467- memcpy(gdt, cpu_gdt_table, GDT_SIZE);
468- cpu_gdt_descr->size = GDT_SIZE - 1;
469-
470- pack_descriptor((u32 *)&gdt[GDT_ENTRY_PDA].a,
471- (u32 *)&gdt[GDT_ENTRY_PDA].b,
472- (unsigned long)pda, sizeof(*pda) - 1,
473- 0x80 | DESCTYPE_S | 0x2, 0); /* present read-write data segment */
474-
475- memset(pda, 0, sizeof(*pda));
476- pda->_pda = pda;
477- pda->cpu_number = cpu;
478- pda->pcurrent = idle;
479-
480- return 1;
481-}
482-
483-void __cpuinit cpu_set_gdt(int cpu)
484-{
485- struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
486+ struct Xgt_desc_struct gdt_descr;
487 unsigned long va, frames[16];
488 int f;
489
490- for (va = cpu_gdt_descr->address, f = 0;
491- va < cpu_gdt_descr->address + cpu_gdt_descr->size;
492+ gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
493+ gdt_descr.size = GDT_SIZE - 1;
494+
495+ for (va = gdt_descr.address, f = 0;
496+ va < gdt_descr.address + gdt_descr.size;
497 va += PAGE_SIZE, f++) {
498 frames[f] = virt_to_mfn(va);
499 make_lowmem_page_readonly(
500 (void *)va, XENFEAT_writable_descriptor_tables);
501 }
502- BUG_ON(HYPERVISOR_set_gdt(frames, (cpu_gdt_descr->size + 1) / 8));
503-
504- set_kernel_fs();
505+ if (HYPERVISOR_set_gdt(frames, (gdt_descr.size + 1) / 8))
506+ BUG();
507+ asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
508 }
509
510-/* Common CPU init for both boot and secondary CPUs */
511-static void __cpuinit _cpu_init(int cpu, struct task_struct *curr)
512+/*
513+ * cpu_init() initializes state that is per-CPU. Some data is already
514+ * initialized (naturally) in the bootstrap process, such as the GDT
515+ * and IDT. We reload them nevertheless, this function acts as a
516+ * 'CPU state barrier', nothing should get across.
517+ */
518+void __cpuinit cpu_init(void)
519 {
520+ int cpu = smp_processor_id();
521+ struct task_struct *curr = current;
522 #ifndef CONFIG_X86_NO_TSS
523 struct tss_struct * t = &per_cpu(init_tss, cpu);
524 #endif
525@@ -757,6 +699,8 @@ static void __cpuinit _cpu_init(int cpu,
526 set_in_cr4(X86_CR4_TSD);
527 }
528
529+ switch_to_new_gdt();
530+
531 /*
532 * Set up and load the per-CPU TSS and LDT
533 */
534@@ -794,38 +738,6 @@ static void __cpuinit _cpu_init(int cpu,
535 mxcsr_feature_mask_init();
536 }
537
538-/* Entrypoint to initialize secondary CPU */
539-void __cpuinit secondary_cpu_init(void)
540-{
541- int cpu = smp_processor_id();
542- struct task_struct *curr = current;
543-
544- _cpu_init(cpu, curr);
545-}
546-
547-/*
548- * cpu_init() initializes state that is per-CPU. Some data is already
549- * initialized (naturally) in the bootstrap process, such as the GDT
550- * and IDT. We reload them nevertheless, this function acts as a
551- * 'CPU state barrier', nothing should get across.
552- */
553-void __cpuinit cpu_init(void)
554-{
555- int cpu = smp_processor_id();
556- struct task_struct *curr = current;
557-
558- /* Set up the real GDT and PDA, so we can transition from the
559- boot versions. */
560- if (!init_gdt(cpu, curr)) {
561- /* failed to allocate something; not much we can do... */
562- for (;;)
563- local_irq_enable();
564- }
565-
566- cpu_set_gdt(cpu);
567- _cpu_init(cpu, curr);
568-}
569-
570 #ifdef CONFIG_HOTPLUG_CPU
571 void __cpuinit cpu_uninit(void)
572 {
573--- a/arch/x86/kernel/cpu/mtrr/main-xen.c
574+++ b/arch/x86/kernel/cpu/mtrr/main-xen.c
575@@ -167,7 +167,7 @@ mtrr_del(int reg, unsigned long base, un
576 EXPORT_SYMBOL(mtrr_add);
577 EXPORT_SYMBOL(mtrr_del);
578
579-void __init mtrr_bp_init(void)
580+__init void mtrr_bp_init(void)
581 {
582 }
583
584--- a/arch/x86/kernel/e820_32-xen.c
585+++ b/arch/x86/kernel/e820_32-xen.c
586@@ -162,26 +162,27 @@ static struct resource standard_io_resou
587
588 static int __init romsignature(const unsigned char *rom)
589 {
590+ const unsigned short * const ptr = (const unsigned short *)rom;
591 unsigned short sig;
592
593- return probe_kernel_address((const unsigned short *)rom, sig) == 0 &&
594- sig == ROMSIGNATURE;
595+ return probe_kernel_address(ptr, sig) == 0 && sig == ROMSIGNATURE;
596 }
597
598-static int __init romchecksum(unsigned char *rom, unsigned long length)
599+static int __init romchecksum(const unsigned char *rom, unsigned long length)
600 {
601- unsigned char sum;
602+ unsigned char sum, c;
603
604- for (sum = 0; length; length--)
605- sum += *rom++;
606- return sum == 0;
607+ for (sum = 0; length && probe_kernel_address(rom++, c) == 0; length--)
608+ sum += c;
609+ return !length && !sum;
610 }
611
612 static void __init probe_roms(void)
613 {
614+ const unsigned char *rom;
615 unsigned long start, length, upper;
616- unsigned char *rom;
617- int i;
618+ unsigned char c;
619+ int i;
620
621 #ifdef CONFIG_XEN
622 /* Nothing to do if not running in dom0. */
623@@ -198,8 +199,11 @@ static void __init probe_roms(void)
624
625 video_rom_resource.start = start;
626
627+ if (probe_kernel_address(rom + 2, c) != 0)
628+ continue;
629+
630 /* 0 < length <= 0x7f * 512, historically */
631- length = rom[2] * 512;
632+ length = c * 512;
633
634 /* if checksum okay, trust length byte */
635 if (length && romchecksum(rom, length))
636@@ -233,8 +237,11 @@ static void __init probe_roms(void)
637 if (!romsignature(rom))
638 continue;
639
640+ if (probe_kernel_address(rom + 2, c) != 0)
641+ continue;
642+
643 /* 0 < length <= 0x7f * 512, historically */
644- length = rom[2] * 512;
645+ length = c * 512;
646
647 /* but accept any length that fits if checksum okay */
648 if (!length || start + length > upper || !romchecksum(rom, length))
649@@ -249,7 +256,7 @@ static void __init probe_roms(void)
650 }
651
652 #ifdef CONFIG_XEN
653-static struct e820map machine_e820 __initdata;
654+static struct e820map machine_e820;
655 #define e820 machine_e820
656 #endif
657
658@@ -409,10 +416,8 @@ int __init sanitize_e820_map(struct e820
659 ____________________33__
660 ______________________4_
661 */
662- printk("sanitize start\n");
663 /* if there's only one memory region, don't bother */
664 if (*pnr_map < 2) {
665- printk("sanitize bail 0\n");
666 return -1;
667 }
668
669@@ -421,7 +426,6 @@ int __init sanitize_e820_map(struct e820
670 /* bail out if we find any unreasonable addresses in bios map */
671 for (i=0; i<old_nr; i++)
672 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) {
673- printk("sanitize bail 1\n");
674 return -1;
675 }
676
677@@ -517,7 +521,6 @@ int __init sanitize_e820_map(struct e820
678 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
679 *pnr_map = new_nr;
680
681- printk("sanitize end\n");
682 return 0;
683 }
684
685@@ -552,7 +555,6 @@ int __init copy_e820_map(struct e820entr
686 unsigned long long size = biosmap->size;
687 unsigned long long end = start + size;
688 unsigned long type = biosmap->type;
689- printk("copy_e820_map() start: %016Lx size: %016Lx end: %016Lx type: %ld\n", start, size, end, type);
690
691 /* Overflow in 64 bits? Ignore the memory map. */
692 if (start > end)
693@@ -564,17 +566,11 @@ int __init copy_e820_map(struct e820entr
694 * Not right. Fix it up.
695 */
696 if (type == E820_RAM) {
697- printk("copy_e820_map() type is E820_RAM\n");
698 if (start < 0x100000ULL && end > 0xA0000ULL) {
699- printk("copy_e820_map() lies in range...\n");
700- if (start < 0xA0000ULL) {
701- printk("copy_e820_map() start < 0xA0000ULL\n");
702+ if (start < 0xA0000ULL)
703 add_memory_region(start, 0xA0000ULL-start, type);
704- }
705- if (end <= 0x100000ULL) {
706- printk("copy_e820_map() end <= 0x100000ULL\n");
707+ if (end <= 0x100000ULL)
708 continue;
709- }
710 start = 0x100000ULL;
711 size = end - start;
712 }
713--- a/arch/x86/kernel/e820_64-xen.c
714+++ b/arch/x86/kernel/e820_64-xen.c
715@@ -17,6 +17,8 @@
716 #include <linux/kexec.h>
717 #include <linux/module.h>
718 #include <linux/mm.h>
719+#include <linux/suspend.h>
720+#include <linux/pfn.h>
721
722 #include <asm/pgtable.h>
723 #include <asm/page.h>
724@@ -28,7 +30,7 @@
725
726 struct e820map e820 __initdata;
727 #ifdef CONFIG_XEN
728-struct e820map machine_e820 __initdata;
729+struct e820map machine_e820;
730 #endif
731
732 /*
733@@ -291,22 +293,6 @@ void __init e820_reserve_resources(struc
734 }
735
736 #ifndef CONFIG_XEN
737-/* Mark pages corresponding to given address range as nosave */
738-static void __init
739-e820_mark_nosave_range(unsigned long start, unsigned long end)
740-{
741- unsigned long pfn, max_pfn;
742-
743- if (start >= end)
744- return;
745-
746- printk("Nosave address range: %016lx - %016lx\n", start, end);
747- max_pfn = end >> PAGE_SHIFT;
748- for (pfn = start >> PAGE_SHIFT; pfn < max_pfn; pfn++)
749- if (pfn_valid(pfn))
750- SetPageNosave(pfn_to_page(pfn));
751-}
752-
753 /*
754 * Find the ranges of physical addresses that do not correspond to
755 * e820 RAM areas and mark the corresponding pages as nosave for software
756@@ -325,13 +311,13 @@ void __init e820_mark_nosave_regions(voi
757 struct e820entry *ei = &e820.map[i];
758
759 if (paddr < ei->addr)
760- e820_mark_nosave_range(paddr,
761- round_up(ei->addr, PAGE_SIZE));
762+ register_nosave_region(PFN_DOWN(paddr),
763+ PFN_UP(ei->addr));
764
765 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
766 if (ei->type != E820_RAM)
767- e820_mark_nosave_range(round_up(ei->addr, PAGE_SIZE),
768- paddr);
769+ register_nosave_region(PFN_UP(ei->addr),
770+ PFN_DOWN(paddr));
771
772 if (paddr >= (end_pfn << PAGE_SHIFT))
773 break;
774--- a/arch/x86/kernel/early_printk-xen.c
775+++ b/arch/x86/kernel/early_printk-xen.c
776@@ -11,11 +11,10 @@
777
778 #ifdef __i386__
779 #include <asm/setup.h>
780-#define VGABASE (__ISA_IO_base + 0xb8000)
781 #else
782 #include <asm/bootsetup.h>
783-#define VGABASE ((void __iomem *)0xffffffff800b8000UL)
784 #endif
785+#define VGABASE (__ISA_IO_base + 0xb8000)
786
787 #ifndef CONFIG_XEN
788 static int max_ypos = 25, max_xpos = 80;
789@@ -93,9 +92,9 @@ static int early_serial_putc(unsigned ch
790 static void early_serial_write(struct console *con, const char *s, unsigned n)
791 {
792 while (*s && n-- > 0) {
793- early_serial_putc(*s);
794 if (*s == '\n')
795 early_serial_putc('\r');
796+ early_serial_putc(*s);
797 s++;
798 }
799 }
800@@ -205,7 +204,7 @@ static noinline long simnow(long cmd, lo
801 return ret;
802 }
803
804-void __init simnow_init(char *str)
805+static void __init simnow_init(char *str)
806 {
807 char *fn = "klog";
808 if (*str == '=')
809@@ -277,22 +276,12 @@ static int __init setup_early_printk(cha
810 early_console = &simnow_console;
811 keep_early = 1;
812 }
813+
814+ if (keep_early)
815+ early_console->flags &= ~CON_BOOT;
816+ else
817+ early_console->flags |= CON_BOOT;
818 register_console(early_console);
819 return 0;
820 }
821-
822 early_param("earlyprintk", setup_early_printk);
823-
824-void __init disable_early_printk(void)
825-{
826- if (!early_console_initialized || !early_console)
827- return;
828- if (!keep_early) {
829- printk("disabling early console\n");
830- unregister_console(early_console);
831- early_console_initialized = 0;
832- } else {
833- printk("keeping early console\n");
834- }
835-}
836-
837--- a/arch/x86/kernel/entry_32-xen.S
838+++ b/arch/x86/kernel/entry_32-xen.S
839@@ -15,7 +15,7 @@
840 * I changed all the .align's to 4 (16 byte alignment), as that's faster
841 * on a 486.
842 *
843- * Stack layout in 'ret_from_system_call':
844+ * Stack layout in 'syscall_exit':
845 * ptrace needs to have all regs on the stack.
846 * if the order here is changed, it needs to be
847 * updated in fork.c:copy_process, signal.c:do_signal,
848@@ -135,7 +135,7 @@ NMI_MASK = 0x80000000
849 movl $(__USER_DS), %edx; \
850 movl %edx, %ds; \
851 movl %edx, %es; \
852- movl $(__KERNEL_PDA), %edx; \
853+ movl $(__KERNEL_PERCPU), %edx; \
854 movl %edx, %fs
855
856 #define RESTORE_INT_REGS \
857@@ -308,16 +308,12 @@ sysenter_past_esp:
858 pushl $(__USER_CS)
859 CFI_ADJUST_CFA_OFFSET 4
860 /*CFI_REL_OFFSET cs, 0*/
861-#ifndef CONFIG_COMPAT_VDSO
862 /*
863 * Push current_thread_info()->sysenter_return to the stack.
864 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
865 * pushed above; +8 corresponds to copy_thread's esp0 setting.
866 */
867 pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp)
868-#else
869- pushl $SYSENTER_RETURN
870-#endif
871 CFI_ADJUST_CFA_OFFSET 4
872 CFI_REL_OFFSET eip, 0
873
874@@ -345,7 +341,7 @@ sysenter_past_esp:
875 jae syscall_badsys
876 call *sys_call_table(,%eax,4)
877 movl %eax,PT_EAX(%esp)
878- DISABLE_INTERRUPTS(CLBR_ECX|CLBR_EDX)
879+ DISABLE_INTERRUPTS(CLBR_ANY)
880 TRACE_IRQS_OFF
881 movl TI_flags(%ebp), %ecx
882 testw $_TIF_ALLWORK_MASK, %cx
883@@ -400,10 +396,6 @@ ENTRY(system_call)
884 CFI_ADJUST_CFA_OFFSET 4
885 SAVE_ALL
886 GET_THREAD_INFO(%ebp)
887- testl $TF_MASK,PT_EFLAGS(%esp)
888- jz no_singlestep
889- orl $_TIF_SINGLESTEP,TI_flags(%ebp)
890-no_singlestep:
891 # system call tracing in operation / emulation
892 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
893 testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
894@@ -418,6 +410,10 @@ syscall_exit:
895 # setting need_resched or sigpending
896 # between sampling and the iret
897 TRACE_IRQS_OFF
898+ testl $TF_MASK,PT_EFLAGS(%esp) # If tracing set singlestep flag on exit
899+ jz no_singlestep
900+ orl $_TIF_SINGLESTEP,TI_flags(%ebp)
901+no_singlestep:
902 movl TI_flags(%ebp), %ecx
903 testw $_TIF_ALLWORK_MASK, %cx # current->work
904 jne syscall_exit_work
905@@ -635,9 +631,7 @@ END(syscall_badsys)
906 #ifndef CONFIG_XEN
907 #define FIXUP_ESPFIX_STACK \
908 /* since we are on a wrong stack, we cant make it a C code :( */ \
909- movl %fs:PDA_cpu, %ebx; \
910- PER_CPU(cpu_gdt_descr, %ebx); \
911- movl GDS_address(%ebx), %ebx; \
912+ PER_CPU(gdt_page, %ebx); \
913 GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah); \
914 addl %esp, %eax; \
915 pushl $__KERNEL_DS; \
916@@ -710,7 +704,7 @@ ENTRY(name) \
917 SAVE_ALL; \
918 TRACE_IRQS_OFF \
919 movl %esp,%eax; \
920- call smp_/**/name; \
921+ call smp_##name; \
922 jmp ret_from_intr; \
923 CFI_ENDPROC; \
924 ENDPROC(name)
925@@ -718,10 +712,6 @@ ENDPROC(name)
926 /* The include is where all of the SMP etc. interrupts come from */
927 #include "entry_arch.h"
928
929-/* This alternate entry is needed because we hijack the apic LVTT */
930-#if defined(CONFIG_VMI) && defined(CONFIG_X86_LOCAL_APIC)
931-BUILD_INTERRUPT(apic_vmi_timer_interrupt,LOCAL_TIMER_VECTOR)
932-#endif
933 #else
934 #define UNWIND_ESPFIX_STACK
935 #endif
936@@ -764,7 +754,7 @@ error_code:
937 pushl %fs
938 CFI_ADJUST_CFA_OFFSET 4
939 /*CFI_REL_OFFSET fs, 0*/
940- movl $(__KERNEL_PDA), %ecx
941+ movl $(__KERNEL_PERCPU), %ecx
942 movl %ecx, %fs
943 UNWIND_ESPFIX_STACK
944 popl %ecx
945--- a/arch/x86/kernel/entry_64-xen.S
946+++ b/arch/x86/kernel/entry_64-xen.S
947@@ -1254,3 +1254,10 @@ ENTRY(call_softirq)
948 ret
949 CFI_ENDPROC
950 ENDPROC(call_softirq)
951+
952+KPROBE_ENTRY(ignore_sysret)
953+ CFI_STARTPROC
954+ mov $-ENOSYS,%eax
955+ HYPERVISOR_IRET 0
956+ CFI_ENDPROC
957+ENDPROC(ignore_sysret)
958--- a/arch/x86/kernel/genapic_64-xen.c
959+++ b/arch/x86/kernel/genapic_64-xen.c
960@@ -11,123 +11,57 @@
961 #include <linux/threads.h>
962 #include <linux/cpumask.h>
963 #include <linux/string.h>
964+#include <linux/module.h>
965 #include <linux/kernel.h>
966 #include <linux/ctype.h>
967 #include <linux/init.h>
968-#include <linux/module.h>
969
970 #include <asm/smp.h>
971 #include <asm/ipi.h>
972+#include <asm/genapic.h>
973
974-#if defined(CONFIG_ACPI)
975+#ifdef CONFIG_ACPI
976 #include <acpi/acpi_bus.h>
977 #endif
978
979 /* which logical CPU number maps to which CPU (physical APIC ID) */
980-u8 x86_cpu_to_apicid[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = BAD_APICID };
981+u8 x86_cpu_to_apicid[NR_CPUS] __read_mostly
982+ = { [0 ... NR_CPUS-1] = BAD_APICID };
983 EXPORT_SYMBOL(x86_cpu_to_apicid);
984-u8 x86_cpu_to_log_apicid[NR_CPUS] = { [0 ... NR_CPUS-1] = BAD_APICID };
985
986-extern struct genapic apic_cluster;
987-extern struct genapic apic_flat;
988-extern struct genapic apic_physflat;
989+u8 x86_cpu_to_log_apicid[NR_CPUS] = { [0 ... NR_CPUS-1] = BAD_APICID };
990
991 #ifndef CONFIG_XEN
992-struct genapic *genapic = &apic_flat;
993-struct genapic *genapic_force;
994+struct genapic __read_mostly *genapic = &apic_flat;
995 #else
996 extern struct genapic apic_xen;
997-struct genapic *genapic = &apic_xen;
998+struct genapic __read_mostly *genapic = &apic_xen;
999 #endif
1000
1001
1002 /*
1003 * Check the APIC IDs in bios_cpu_apicid and choose the APIC mode.
1004 */
1005-void __init clustered_apic_check(void)
1006+void __init setup_apic_routing(void)
1007 {
1008 #ifndef CONFIG_XEN
1009- long i;
1010- u8 clusters, max_cluster;
1011- u8 id;
1012- u8 cluster_cnt[NUM_APIC_CLUSTERS];
1013- int max_apic = 0;
1014-
1015- /* genapic selection can be forced because of certain quirks.
1016- */
1017- if (genapic_force) {
1018- genapic = genapic_force;
1019- goto print;
1020- }
1021-
1022-#if defined(CONFIG_ACPI)
1023+#ifdef CONFIG_ACPI
1024 /*
1025- * Some x86_64 machines use physical APIC mode regardless of how many
1026- * procs/clusters are present (x86_64 ES7000 is an example).
1027+ * Quirk: some x86_64 machines can only use physical APIC mode
1028+ * regardless of how many processors are present (x86_64 ES7000
1029+ * is an example).
1030 */
1031- if (acpi_gbl_FADT.header.revision > FADT2_REVISION_ID)
1032- if (acpi_gbl_FADT.flags & ACPI_FADT_APIC_PHYSICAL) {
1033- genapic = &apic_cluster;
1034- goto print;
1035- }
1036-#endif
1037-
1038- memset(cluster_cnt, 0, sizeof(cluster_cnt));
1039- for (i = 0; i < NR_CPUS; i++) {
1040- id = bios_cpu_apicid[i];
1041- if (id == BAD_APICID)
1042- continue;
1043- if (id > max_apic)
1044- max_apic = id;
1045- cluster_cnt[APIC_CLUSTERID(id)]++;
1046- }
1047-
1048- /* Don't use clustered mode on AMD platforms. */
1049- if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
1050+ if (acpi_gbl_FADT.header.revision > FADT2_REVISION_ID &&
1051+ (acpi_gbl_FADT.flags & ACPI_FADT_APIC_PHYSICAL))
1052 genapic = &apic_physflat;
1053-#ifndef CONFIG_HOTPLUG_CPU
1054- /* In the CPU hotplug case we cannot use broadcast mode
1055- because that opens a race when a CPU is removed.
1056- Stay at physflat mode in this case.
1057- It is bad to do this unconditionally though. Once
1058- we have ACPI platform support for CPU hotplug
1059- we should detect hotplug capablity from ACPI tables and
1060- only do this when really needed. -AK */
1061- if (max_apic <= 8)
1062- genapic = &apic_flat;
1063-#endif
1064- goto print;
1065- }
1066-
1067- clusters = 0;
1068- max_cluster = 0;
1069-
1070- for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
1071- if (cluster_cnt[i] > 0) {
1072- ++clusters;
1073- if (cluster_cnt[i] > max_cluster)
1074- max_cluster = cluster_cnt[i];
1075- }
1076- }
1077+ else
1078+#endif
1079
1080- /*
1081- * If we have clusters <= 1 and CPUs <= 8 in cluster 0, then flat mode,
1082- * else if max_cluster <= 4 and cluster_cnt[15] == 0, clustered logical
1083- * else physical mode.
1084- * (We don't use lowest priority delivery + HW APIC IRQ steering, so
1085- * can ignore the clustered logical case and go straight to physical.)
1086- */
1087- if (clusters <= 1 && max_cluster <= 8 && cluster_cnt[0] == max_cluster) {
1088-#ifdef CONFIG_HOTPLUG_CPU
1089- /* Don't use APIC shortcuts in CPU hotplug to avoid races */
1090- genapic = &apic_physflat;
1091-#else
1092+ if (cpus_weight(cpu_possible_map) <= 8)
1093 genapic = &apic_flat;
1094-#endif
1095- } else
1096- genapic = &apic_cluster;
1097+ else
1098+ genapic = &apic_physflat;
1099
1100-print:
1101 #else
1102 /* hardcode to xen apic functions */
1103 genapic = &apic_xen;
1104@@ -135,7 +69,7 @@ print:
1105 printk(KERN_INFO "Setting APIC routing to %s\n", genapic->name);
1106 }
1107
1108-/* Same for both flat and clustered. */
1109+/* Same for both flat and physical. */
1110
1111 #ifdef CONFIG_XEN
1112 extern void xen_send_IPI_shortcut(unsigned int shortcut, int vector, unsigned int dest);
1113--- a/arch/x86/kernel/genapic_xen_64.c
1114+++ b/arch/x86/kernel/genapic_xen_64.c
1115@@ -21,9 +21,8 @@
1116 #include <asm/ipi.h>
1117 #else
1118 #include <asm/apic.h>
1119-#include <asm/apicdef.h>
1120-#include <asm/genapic.h>
1121 #endif
1122+#include <asm/genapic.h>
1123 #include <xen/evtchn.h>
1124
1125 DECLARE_PER_CPU(int, ipi_to_irq[NR_IPIS]);
1126--- a/arch/x86/kernel/head_32-xen.S
1127+++ b/arch/x86/kernel/head_32-xen.S
1128@@ -37,7 +37,8 @@ ENTRY(startup_32)
1129 /* Set up the stack pointer */
1130 movl $(init_thread_union+THREAD_SIZE),%esp
1131
1132- call setup_pda
1133+ movl %ss,%eax
1134+ movl %eax,%fs # gets reset once there's real percpu
1135
1136 /* get vendor info */
1137 xorl %eax,%eax # call CPUID with 0 -> return vendor ID
1138@@ -64,55 +65,11 @@ ENTRY(startup_32)
1139 xorl %eax,%eax # Clear GS
1140 movl %eax,%gs
1141
1142- movl $(__KERNEL_PDA),%eax
1143- mov %eax,%fs
1144-
1145 cld # gcc2 wants the direction flag cleared at all times
1146
1147 pushl $0 # fake return address for unwinder
1148 jmp start_kernel
1149
1150-/*
1151- * Point the GDT at this CPU's PDA. This will be
1152- * cpu_gdt_table and boot_pda.
1153- */
1154-ENTRY(setup_pda)
1155- /* get the PDA pointer */
1156- movl $boot_pda, %eax
1157-
1158- /* slot the PDA address into the GDT */
1159- mov $cpu_gdt_table, %ecx
1160- mov %ax, (__KERNEL_PDA+0+2)(%ecx) /* base & 0x0000ffff */
1161- shr $16, %eax
1162- mov %al, (__KERNEL_PDA+4+0)(%ecx) /* base & 0x00ff0000 */
1163- mov %ah, (__KERNEL_PDA+4+3)(%ecx) /* base & 0xff000000 */
1164-
1165- # %esi still points to start_info, and no registers
1166- # need to be preserved.
1167-
1168- movl XEN_START_mfn_list(%esi), %ebx
1169- movl $(cpu_gdt_table - __PAGE_OFFSET), %eax
1170- shrl $PAGE_SHIFT, %eax
1171- movl (%ebx,%eax,4), %ecx
1172- pushl %ecx # frame number for set_gdt below
1173-
1174- xorl %esi, %esi
1175- xorl %edx, %edx
1176- shldl $PAGE_SHIFT, %ecx, %edx
1177- shll $PAGE_SHIFT, %ecx
1178- orl $0x61, %ecx
1179- movl $cpu_gdt_table, %ebx
1180- movl $__HYPERVISOR_update_va_mapping, %eax
1181- int $0x82
1182-
1183- movl $(PAGE_SIZE_asm / 8), %ecx
1184- movl %esp, %ebx
1185- movl $__HYPERVISOR_set_gdt, %eax
1186- int $0x82
1187-
1188- popl %ecx
1189- ret
1190-
1191 #define HYPERCALL_PAGE_OFFSET 0x1000
1192 .org HYPERCALL_PAGE_OFFSET
1193 ENTRY(hypercall_page)
1194@@ -138,60 +95,6 @@ ENTRY(empty_zero_page)
1195 */
1196 .data
1197
1198-/*
1199- * The Global Descriptor Table contains 28 quadwords, per-CPU.
1200- */
1201- .section .data.page_aligned, "aw"
1202- .align PAGE_SIZE_asm
1203-ENTRY(cpu_gdt_table)
1204- .quad 0x0000000000000000 /* NULL descriptor */
1205- .quad 0x0000000000000000 /* 0x0b reserved */
1206- .quad 0x0000000000000000 /* 0x13 reserved */
1207- .quad 0x0000000000000000 /* 0x1b reserved */
1208- .quad 0x0000000000000000 /* 0x20 unused */
1209- .quad 0x0000000000000000 /* 0x28 unused */
1210- .quad 0x0000000000000000 /* 0x33 TLS entry 1 */
1211- .quad 0x0000000000000000 /* 0x3b TLS entry 2 */
1212- .quad 0x0000000000000000 /* 0x43 TLS entry 3 */
1213- .quad 0x0000000000000000 /* 0x4b reserved */
1214- .quad 0x0000000000000000 /* 0x53 reserved */
1215- .quad 0x0000000000000000 /* 0x5b reserved */
1216-
1217- .quad 0x00cf9a000000ffff /* 0x60 kernel 4GB code at 0x00000000 */
1218- .quad 0x00cf92000000ffff /* 0x68 kernel 4GB data at 0x00000000 */
1219- .quad 0x00cffa000000ffff /* 0x73 user 4GB code at 0x00000000 */
1220- .quad 0x00cff2000000ffff /* 0x7b user 4GB data at 0x00000000 */
1221-
1222- .quad 0x0000000000000000 /* 0x80 TSS descriptor */
1223- .quad 0x0000000000000000 /* 0x88 LDT descriptor */
1224-
1225- /*
1226- * Segments used for calling PnP BIOS have byte granularity.
1227- * They code segments and data segments have fixed 64k limits,
1228- * the transfer segment sizes are set at run time.
1229- */
1230- .quad 0x0000000000000000 /* 0x90 32-bit code */
1231- .quad 0x0000000000000000 /* 0x98 16-bit code */
1232- .quad 0x0000000000000000 /* 0xa0 16-bit data */
1233- .quad 0x0000000000000000 /* 0xa8 16-bit data */
1234- .quad 0x0000000000000000 /* 0xb0 16-bit data */
1235-
1236- /*
1237- * The APM segments have byte granularity and their bases
1238- * are set at run time. All have 64k limits.
1239- */
1240- .quad 0x0000000000000000 /* 0xb8 APM CS code */
1241- .quad 0x0000000000000000 /* 0xc0 APM CS 16 code (16 bit) */
1242- .quad 0x0000000000000000 /* 0xc8 APM DS data */
1243-
1244- .quad 0x0000000000000000 /* 0xd0 - ESPFIX SS */
1245- .quad 0x00cf92000000ffff /* 0xd8 - PDA */
1246- .quad 0x0000000000000000 /* 0xe0 - unused */
1247- .quad 0x0000000000000000 /* 0xe8 - unused */
1248- .quad 0x0000000000000000 /* 0xf0 - unused */
1249- .quad 0x0000000000000000 /* 0xf8 - GDT entry 31: double-fault TSS */
1250- .align PAGE_SIZE_asm
1251-
1252 #if CONFIG_XEN_COMPAT <= 0x030002
1253 /*
1254 * __xen_guest information
1255--- a/arch/x86/kernel/head64-xen.c
1256+++ b/arch/x86/kernel/head64-xen.c
1257@@ -25,13 +25,21 @@
1258 #include <asm/setup.h>
1259 #include <asm/desc.h>
1260 #include <asm/pgtable.h>
1261+#include <asm/tlbflush.h>
1262 #include <asm/sections.h>
1263
1264 unsigned long start_pfn;
1265
1266+#ifndef CONFIG_XEN
1267+static void __init zap_identity_mappings(void)
1268+{
1269+ pgd_t *pgd = pgd_offset_k(0UL);
1270+ pgd_clear(pgd);
1271+ __flush_tlb();
1272+}
1273+
1274 /* Don't add a printk in there. printk relies on the PDA which is not initialized
1275 yet. */
1276-#if 0
1277 static void __init clear_bss(void)
1278 {
1279 memset(__bss_start, 0,
1280@@ -40,26 +48,25 @@ static void __init clear_bss(void)
1281 #endif
1282
1283 #define NEW_CL_POINTER 0x228 /* Relative to real mode data */
1284-#define OLD_CL_MAGIC_ADDR 0x90020
1285+#define OLD_CL_MAGIC_ADDR 0x20
1286 #define OLD_CL_MAGIC 0xA33F
1287-#define OLD_CL_BASE_ADDR 0x90000
1288-#define OLD_CL_OFFSET 0x90022
1289+#define OLD_CL_OFFSET 0x22
1290
1291 static void __init copy_bootdata(char *real_mode_data)
1292 {
1293 #ifndef CONFIG_XEN
1294- int new_data;
1295+ unsigned long new_data;
1296 char * command_line;
1297
1298 memcpy(x86_boot_params, real_mode_data, BOOT_PARAM_SIZE);
1299- new_data = *(int *) (x86_boot_params + NEW_CL_POINTER);
1300+ new_data = *(u32 *) (x86_boot_params + NEW_CL_POINTER);
1301 if (!new_data) {
1302- if (OLD_CL_MAGIC != * (u16 *) OLD_CL_MAGIC_ADDR) {
1303+ if (OLD_CL_MAGIC != *(u16 *)(real_mode_data + OLD_CL_MAGIC_ADDR)) {
1304 return;
1305 }
1306- new_data = OLD_CL_BASE_ADDR + * (u16 *) OLD_CL_OFFSET;
1307+ new_data = __pa(real_mode_data) + *(u16 *)(real_mode_data + OLD_CL_OFFSET);
1308 }
1309- command_line = (char *) ((u64)(new_data));
1310+ command_line = __va(new_data);
1311 memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
1312 #else
1313 int max_cmdline;
1314@@ -101,10 +108,13 @@ void __init x86_64_start_kernel(char * r
1315 while ((1UL << machine_to_phys_order) < machine_to_phys_nr_ents )
1316 machine_to_phys_order++;
1317
1318-#if 0
1319+#ifndef CONFIG_XEN
1320 /* clear bss before set_intr_gate with early_idt_handler */
1321 clear_bss();
1322
1323+ /* Make NULL pointers segfault */
1324+ zap_identity_mappings();
1325+
1326 for (i = 0; i < IDT_ENTRIES; i++)
1327 set_intr_gate(i, early_idt_handler);
1328 asm volatile("lidt %0" :: "m" (idt_descr));
1329@@ -116,7 +126,7 @@ void __init x86_64_start_kernel(char * r
1330 cpu_pda(i) = &boot_cpu_pda[i];
1331
1332 pda_init(0);
1333- copy_bootdata(real_mode_data);
1334+ copy_bootdata(__va(real_mode_data));
1335 #ifdef CONFIG_SMP
1336 cpu_set(0, cpu_online_map);
1337 #endif
1338--- a/arch/x86/kernel/head_64-xen.S
1339+++ b/arch/x86/kernel/head_64-xen.S
1340@@ -5,6 +5,7 @@
1341 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
1342 * Copyright (C) 2000 Karsten Keil <kkeil@suse.de>
1343 * Copyright (C) 2001,2002 Andi Kleen <ak@suse.de>
1344+ * Copyright (C) 2005 Eric Biederman <ebiederm@xmission.com>
1345 * Jun Nakajima <jun.nakajima@intel.com>
1346 * Modified for Xen
1347 */
1348@@ -34,27 +35,15 @@ startup_64:
1349 pushq $0 # fake return address
1350 jmp x86_64_start_kernel
1351
1352-#ifdef CONFIG_ACPI_SLEEP
1353-.org 0xf00
1354- .globl pGDT32
1355-pGDT32:
1356- .word gdt_end-cpu_gdt_table-1
1357- .long cpu_gdt_table-__START_KERNEL_map
1358-#endif
1359-ENTRY(stext)
1360-ENTRY(_stext)
1361+.balign PAGE_SIZE
1362
1363- $page = 0
1364 #define NEXT_PAGE(name) \
1365- $page = $page + 1; \
1366- .org $page * 0x1000; \
1367- phys_##name = $page * 0x1000 + __PHYSICAL_START; \
1368+ .balign PAGE_SIZE; \
1369+ phys_##name = . - .bootstrap.text; \
1370 ENTRY(name)
1371
1372 NEXT_PAGE(init_level4_pgt)
1373- /* This gets initialized in x86_64_start_kernel */
1374 .fill 512,8,0
1375-NEXT_PAGE(init_level4_user_pgt)
1376 /*
1377 * We update two pgd entries to make kernel and user pgd consistent
1378 * at pgd_populate(). It can be used for kernel modules. So we place
1379@@ -101,14 +90,6 @@ NEXT_PAGE(hypercall_page)
1380 #undef NEXT_PAGE
1381
1382 .data
1383-/* Just dummy symbol to allow compilation. Not used in sleep path */
1384-#ifdef CONFIG_ACPI_SLEEP
1385- .align PAGE_SIZE
1386-ENTRY(wakeup_level4_pgt)
1387- .fill 512,8,0
1388-#endif
1389-
1390- .data
1391
1392 .align 16
1393 .globl cpu_gdt_descr
1394@@ -136,13 +117,13 @@ gdt:
1395
1396 ENTRY(cpu_gdt_table)
1397 .quad 0x0000000000000000 /* NULL descriptor */
1398+ .quad 0x00cf9b000000ffff /* __KERNEL32_CS */
1399+ .quad 0x00af9b000000ffff /* __KERNEL_CS */
1400+ .quad 0x00cf93000000ffff /* __KERNEL_DS */
1401+ .quad 0x00cffb000000ffff /* __USER32_CS */
1402+ .quad 0x00cff3000000ffff /* __USER_DS, __USER32_DS */
1403+ .quad 0x00affb000000ffff /* __USER_CS */
1404 .quad 0x0 /* unused */
1405- .quad 0x00af9a000000ffff /* __KERNEL_CS */
1406- .quad 0x00cf92000000ffff /* __KERNEL_DS */
1407- .quad 0x00cffa000000ffff /* __USER32_CS */
1408- .quad 0x00cff2000000ffff /* __USER_DS, __USER32_DS */
1409- .quad 0x00affa000000ffff /* __USER_CS */
1410- .quad 0x00cf9a000000ffff /* __KERNEL32_CS */
1411 .quad 0,0 /* TSS */
1412 .quad 0,0 /* LDT */
1413 .quad 0,0,0 /* three TLS descriptors */
1414@@ -165,14 +146,11 @@ ENTRY(empty_zero_page)
1415 * __xen_guest information
1416 */
1417 .macro utoh value
1418- .if (\value) < 0 || (\value) >= 0x10
1419- utoh (((\value)>>4)&0x0fffffffffffffff)
1420- .endif
1421- .if ((\value) & 0xf) < 10
1422- .byte '0' + ((\value) & 0xf)
1423- .else
1424- .byte 'A' + ((\value) & 0xf) - 10
1425- .endif
1426+ i = 64
1427+ .rept 16
1428+ i = i - 4
1429+ .byte '0' + ((((\value) >> i) & 0xf) > 9) * ('0' - 'A' + 10) + (((\value) >> i) & 0xf)
1430+ .endr
1431 .endm
1432
1433 .section __xen_guest
1434--- a/arch/x86/kernel/io_apic_32-xen.c
1435+++ b/arch/x86/kernel/io_apic_32-xen.c
1436@@ -25,7 +25,6 @@
1437 #include <linux/init.h>
1438 #include <linux/delay.h>
1439 #include <linux/sched.h>
1440-#include <linux/smp_lock.h>
1441 #include <linux/mc146818rtc.h>
1442 #include <linux/compiler.h>
1443 #include <linux/acpi.h>
1444@@ -35,6 +34,7 @@
1445 #include <linux/msi.h>
1446 #include <linux/htirq.h>
1447 #include <linux/freezer.h>
1448+#include <linux/kthread.h>
1449
1450 #include <asm/io.h>
1451 #include <asm/smp.h>
1452@@ -710,8 +710,6 @@ static int balanced_irq(void *unused)
1453 unsigned long prev_balance_time = jiffies;
1454 long time_remaining = balanced_irq_interval;
1455
1456- daemonize("kirqd");
1457-
1458 /* push everything to CPU 0 to give us a starting point. */
1459 for (i = 0 ; i < NR_IRQS ; i++) {
1460 irq_desc[i].pending_mask = cpumask_of_cpu(0);
1461@@ -771,10 +769,9 @@ static int __init balanced_irq_init(void
1462 }
1463
1464 printk(KERN_INFO "Starting balanced_irq\n");
1465- if (kernel_thread(balanced_irq, NULL, CLONE_KERNEL) >= 0)
1466+ if (!IS_ERR(kthread_run(balanced_irq, NULL, "kirqd")))
1467 return 0;
1468- else
1469- printk(KERN_ERR "balanced_irq_init: failed to spawn balanced_irq");
1470+ printk(KERN_ERR "balanced_irq_init: failed to spawn balanced_irq");
1471 failed:
1472 for_each_possible_cpu(i) {
1473 kfree(irq_cpu_data[i].irq_delta);
1474@@ -1455,10 +1452,6 @@ static void __init setup_ExtINT_IRQ0_pin
1475 enable_8259A_irq(0);
1476 }
1477
1478-static inline void UNEXPECTED_IO_APIC(void)
1479-{
1480-}
1481-
1482 void __init print_IO_APIC(void)
1483 {
1484 int apic, i;
1485@@ -1498,34 +1491,12 @@ void __init print_IO_APIC(void)
1486 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1487 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1488 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1489- if (reg_00.bits.ID >= get_physical_broadcast())
1490- UNEXPECTED_IO_APIC();
1491- if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
1492- UNEXPECTED_IO_APIC();
1493
1494 printk(KERN_DEBUG ".... register #01: %08X\n", reg_01.raw);
1495 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1496- if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
1497- (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
1498- (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
1499- (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
1500- (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
1501- (reg_01.bits.entries != 0x2E) &&
1502- (reg_01.bits.entries != 0x3F)
1503- )
1504- UNEXPECTED_IO_APIC();
1505
1506 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1507 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1508- if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
1509- (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
1510- (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
1511- (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
1512- (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
1513- )
1514- UNEXPECTED_IO_APIC();
1515- if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
1516- UNEXPECTED_IO_APIC();
1517
1518 /*
1519 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1520@@ -1535,8 +1506,6 @@ void __init print_IO_APIC(void)
1521 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1522 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1523 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1524- if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
1525- UNEXPECTED_IO_APIC();
1526 }
1527
1528 /*
1529@@ -1548,8 +1517,6 @@ void __init print_IO_APIC(void)
1530 reg_03.raw != reg_01.raw) {
1531 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1532 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1533- if (reg_03.bits.__reserved_1)
1534- UNEXPECTED_IO_APIC();
1535 }
1536
1537 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1538@@ -2686,19 +2653,19 @@ int arch_setup_msi_irq(struct pci_dev *d
1539 if (irq < 0)
1540 return irq;
1541
1542- set_irq_msi(irq, desc);
1543 ret = msi_compose_msg(dev, irq, &msg);
1544 if (ret < 0) {
1545 destroy_irq(irq);
1546 return ret;
1547 }
1548
1549+ set_irq_msi(irq, desc);
1550 write_msi_msg(irq, &msg);
1551
1552 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq,
1553 "edge");
1554
1555- return irq;
1556+ return 0;
1557 }
1558
1559 void arch_teardown_msi_irq(unsigned int irq)
1560--- a/arch/x86/kernel/io_apic_64-xen.c
1561+++ b/arch/x86/kernel/io_apic_64-xen.c
1562@@ -25,7 +25,6 @@
1563 #include <linux/init.h>
1564 #include <linux/delay.h>
1565 #include <linux/sched.h>
1566-#include <linux/smp_lock.h>
1567 #include <linux/pci.h>
1568 #include <linux/mc146818rtc.h>
1569 #include <linux/acpi.h>
1570@@ -904,10 +903,6 @@ static void __init setup_ExtINT_IRQ0_pin
1571 enable_8259A_irq(0);
1572 }
1573
1574-void __init UNEXPECTED_IO_APIC(void)
1575-{
1576-}
1577-
1578 void __apicdebuginit print_IO_APIC(void)
1579 {
1580 int apic, i;
1581@@ -943,40 +938,16 @@ void __apicdebuginit print_IO_APIC(void)
1582 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
1583 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1584 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1585- if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
1586- UNEXPECTED_IO_APIC();
1587
1588 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1589 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1590- if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
1591- (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
1592- (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
1593- (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
1594- (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
1595- (reg_01.bits.entries != 0x2E) &&
1596- (reg_01.bits.entries != 0x3F) &&
1597- (reg_01.bits.entries != 0x03)
1598- )
1599- UNEXPECTED_IO_APIC();
1600
1601 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1602 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1603- if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
1604- (reg_01.bits.version != 0x02) && /* 82801BA IO-APICs (ICH2) */
1605- (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
1606- (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
1607- (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
1608- (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
1609- )
1610- UNEXPECTED_IO_APIC();
1611- if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
1612- UNEXPECTED_IO_APIC();
1613
1614 if (reg_01.bits.version >= 0x10) {
1615 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1616 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1617- if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
1618- UNEXPECTED_IO_APIC();
1619 }
1620
1621 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1622@@ -1408,8 +1379,7 @@ static void irq_complete_move(unsigned i
1623
1624 vector = ~get_irq_regs()->orig_rax;
1625 me = smp_processor_id();
1626- if ((vector == cfg->vector) &&
1627- cpu_isset(smp_processor_id(), cfg->domain)) {
1628+ if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) {
1629 cpumask_t cleanup_mask;
1630
1631 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1632@@ -1444,7 +1414,7 @@ static void ack_apic_level(unsigned int
1633
1634 /*
1635 * We must acknowledge the irq before we move it or the acknowledge will
1636- * not propogate properly.
1637+ * not propagate properly.
1638 */
1639 ack_APIC_irq();
1640
1641@@ -1527,6 +1497,7 @@ static void ack_lapic_irq (unsigned int
1642 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1643
1644 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1645+ .name = "local-APIC",
1646 .typename = "local-APIC-edge",
1647 .startup = NULL, /* startup_irq() not used for IRQ0 */
1648 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1649@@ -1998,18 +1969,18 @@ int arch_setup_msi_irq(struct pci_dev *d
1650 if (irq < 0)
1651 return irq;
1652
1653- set_irq_msi(irq, desc);
1654 ret = msi_compose_msg(dev, irq, &msg);
1655 if (ret < 0) {
1656 destroy_irq(irq);
1657 return ret;
1658 }
1659
1660+ set_irq_msi(irq, desc);
1661 write_msi_msg(irq, &msg);
1662
1663 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
1664
1665- return irq;
1666+ return 0;
1667 }
1668
1669 void arch_teardown_msi_irq(unsigned int irq)
1670--- a/arch/x86/kernel/ioport_32-xen.c
1671+++ b/arch/x86/kernel/ioport_32-xen.c
1672@@ -12,10 +12,10 @@
1673 #include <linux/types.h>
1674 #include <linux/ioport.h>
1675 #include <linux/smp.h>
1676-#include <linux/smp_lock.h>
1677 #include <linux/stddef.h>
1678 #include <linux/slab.h>
1679 #include <linux/thread_info.h>
1680+#include <linux/syscalls.h>
1681 #include <xen/interface/physdev.h>
1682
1683 /* Set EXTENT bits starting at BASE in BITMAP to value TURN_ON. */
1684--- a/arch/x86/kernel/ioport_64-xen.c
1685+++ b/arch/x86/kernel/ioport_64-xen.c
1686@@ -13,10 +13,10 @@
1687 #include <linux/ioport.h>
1688 #include <linux/mm.h>
1689 #include <linux/smp.h>
1690-#include <linux/smp_lock.h>
1691 #include <linux/stddef.h>
1692 #include <linux/slab.h>
1693 #include <linux/thread_info.h>
1694+#include <linux/syscalls.h>
1695 #include <xen/interface/physdev.h>
1696
1697 /* Set EXTENT bits starting at BASE in BITMAP to value TURN_ON. */
1698--- a/arch/x86/kernel/irq_32-xen.c
1699+++ b/arch/x86/kernel/irq_32-xen.c
1700@@ -24,6 +24,9 @@
1701 DEFINE_PER_CPU(irq_cpustat_t, irq_stat) ____cacheline_internodealigned_in_smp;
1702 EXPORT_PER_CPU_SYMBOL(irq_stat);
1703
1704+DEFINE_PER_CPU(struct pt_regs *, irq_regs);
1705+EXPORT_PER_CPU_SYMBOL(irq_regs);
1706+
1707 /*
1708 * 'what should we do if we get a hw irq event on an illegal vector'.
1709 * each architecture has to answer this themselves.
1710--- a/arch/x86/kernel/irq_64-xen.c
1711+++ b/arch/x86/kernel/irq_64-xen.c
1712@@ -32,7 +32,7 @@ atomic_t irq_err_count;
1713 */
1714 static inline void stack_overflow_check(struct pt_regs *regs)
1715 {
1716- u64 curbase = (u64) current->thread_info;
1717+ u64 curbase = (u64)task_stack_page(current);
1718 static unsigned long warned = -60*HZ;
1719
1720 if (regs->rsp >= curbase && regs->rsp <= curbase + THREAD_SIZE &&
1721@@ -145,17 +145,43 @@ void fixup_irqs(cpumask_t map)
1722
1723 for (irq = 0; irq < NR_IRQS; irq++) {
1724 cpumask_t mask;
1725+ int break_affinity = 0;
1726+ int set_affinity = 1;
1727+
1728 if (irq == 2)
1729 continue;
1730
1731+ /* interrupt's are disabled at this point */
1732+ spin_lock(&irq_desc[irq].lock);
1733+
1734+ if (!irq_has_action(irq) ||
1735+ cpus_equal(irq_desc[irq].affinity, map)) {
1736+ spin_unlock(&irq_desc[irq].lock);
1737+ continue;
1738+ }
1739+
1740 cpus_and(mask, irq_desc[irq].affinity, map);
1741- if (any_online_cpu(mask) == NR_CPUS) {
1742- /*printk("Breaking affinity for irq %i\n", irq);*/
1743+ if (cpus_empty(mask)) {
1744+ break_affinity = 1;
1745 mask = map;
1746 }
1747+
1748+ if (irq_desc[irq].chip->mask)
1749+ irq_desc[irq].chip->mask(irq);
1750+
1751 if (irq_desc[irq].chip->set_affinity)
1752 irq_desc[irq].chip->set_affinity(irq, mask);
1753- else if (irq_desc[irq].action && !(warned++))
1754+ else if (!(warned++))
1755+ set_affinity = 0;
1756+
1757+ if (irq_desc[irq].chip->unmask)
1758+ irq_desc[irq].chip->unmask(irq);
1759+
1760+ spin_unlock(&irq_desc[irq].lock);
1761+
1762+ if (break_affinity && set_affinity)
1763+ /*printk("Broke affinity for irq %i\n", irq)*/;
1764+ else if (!set_affinity)
1765 printk("Cannot set affinity for irq %i\n", irq);
1766 }
1767
1768--- a/arch/x86/kernel/ldt_32-xen.c
1769+++ b/arch/x86/kernel/ldt_32-xen.c
1770@@ -10,7 +10,6 @@
1771 #include <linux/string.h>
1772 #include <linux/mm.h>
1773 #include <linux/smp.h>
1774-#include <linux/smp_lock.h>
1775 #include <linux/vmalloc.h>
1776 #include <linux/slab.h>
1777
1778--- a/arch/x86/kernel/ldt_64-xen.c
1779+++ b/arch/x86/kernel/ldt_64-xen.c
1780@@ -13,7 +13,6 @@
1781 #include <linux/string.h>
1782 #include <linux/mm.h>
1783 #include <linux/smp.h>
1784-#include <linux/smp_lock.h>
1785 #include <linux/vmalloc.h>
1786 #include <linux/slab.h>
1787
1788--- a/arch/x86/kernel/Makefile
1789+++ b/arch/x86/kernel/Makefile
1790@@ -127,4 +127,4 @@ endif
1791 disabled-obj-$(CONFIG_XEN) := early-quirks.o hpet.o i8253.o i8259_$(BITS).o reboot.o \
1792 smpboot_$(BITS).o tsc_$(BITS).o tsc_sync.o
1793 disabled-obj-$(CONFIG_XEN_UNPRIVILEGED_GUEST) += mpparse_64.o
1794-%/head_$(BITS).o %/head_$(BITS).s: $(if $(CONFIG_XEN),EXTRA_AFLAGS,dummy) :=
1795+%/head_64.o %/head_64.s: $(if $(CONFIG_XEN),EXTRA_AFLAGS,dummy) :=
1796--- a/arch/x86/kernel/microcode-xen.c
1797+++ b/arch/x86/kernel/microcode-xen.c
1798@@ -135,7 +135,7 @@ static int __init microcode_dev_init (vo
1799 return 0;
1800 }
1801
1802-static void __exit microcode_dev_exit (void)
1803+static void microcode_dev_exit (void)
1804 {
1805 misc_deregister(&microcode_dev);
1806 }
1807--- a/arch/x86/kernel/mpparse_32-xen.c
1808+++ b/arch/x86/kernel/mpparse_32-xen.c
1809@@ -18,7 +18,6 @@
1810 #include <linux/acpi.h>
1811 #include <linux/delay.h>
1812 #include <linux/bootmem.h>
1813-#include <linux/smp_lock.h>
1814 #include <linux/kernel_stat.h>
1815 #include <linux/mc146818rtc.h>
1816 #include <linux/bitops.h>
1817@@ -484,7 +483,7 @@ static int __init smp_read_mpc(struct mp
1818 }
1819 ++mpc_record;
1820 }
1821- clustered_apic_check();
1822+ setup_apic_routing();
1823 if (!num_processors)
1824 printk(KERN_ERR "SMP mptable: no processors registered!\n");
1825 return num_processors;
1826--- a/arch/x86/kernel/mpparse_64-xen.c
1827+++ b/arch/x86/kernel/mpparse_64-xen.c
1828@@ -17,7 +17,6 @@
1829 #include <linux/init.h>
1830 #include <linux/delay.h>
1831 #include <linux/bootmem.h>
1832-#include <linux/smp_lock.h>
1833 #include <linux/kernel_stat.h>
1834 #include <linux/mc146818rtc.h>
1835 #include <linux/acpi.h>
1836@@ -307,7 +306,7 @@ static int __init smp_read_mpc(struct mp
1837 }
1838 }
1839 }
1840- clustered_apic_check();
1841+ setup_apic_routing();
1842 if (!num_processors)
1843 printk(KERN_ERR "MPTABLE: no processors registered!\n");
1844 return num_processors;
1845--- a/arch/x86/kernel/pci-dma-xen.c
1846+++ b/arch/x86/kernel/pci-dma-xen.c
1847@@ -13,6 +13,7 @@
1848 #include <linux/pci.h>
1849 #include <linux/module.h>
1850 #include <linux/version.h>
1851+#include <linux/pci.h>
1852 #include <asm/io.h>
1853 #include <xen/balloon.h>
1854 #include <xen/gnttab.h>
1855@@ -278,7 +279,7 @@ int dma_declare_coherent_memory(struct d
1856 {
1857 void __iomem *mem_base = NULL;
1858 int pages = size >> PAGE_SHIFT;
1859- int bitmap_size = (pages + 31)/32;
1860+ int bitmap_size = BITS_TO_LONGS(pages) * sizeof(long);
1861
1862 if ((flags & (DMA_MEMORY_MAP | DMA_MEMORY_IO)) == 0)
1863 goto out;
1864@@ -351,6 +352,32 @@ void *dma_mark_declared_memory_occupied(
1865 EXPORT_SYMBOL(dma_mark_declared_memory_occupied);
1866 #endif /* ARCH_HAS_DMA_DECLARE_COHERENT_MEMORY */
1867
1868+#if defined(CONFIG_PCI) && !defined(CONFIG_XEN)
1869+/* Many VIA bridges seem to corrupt data for DAC. Disable it here */
1870+
1871+int forbid_dac;
1872+EXPORT_SYMBOL(forbid_dac);
1873+
1874+static __devinit void via_no_dac(struct pci_dev *dev)
1875+{
1876+ if ((dev->class >> 8) == PCI_CLASS_BRIDGE_PCI && forbid_dac == 0) {
1877+ printk(KERN_INFO "PCI: VIA PCI bridge detected. Disabling DAC.\n");
1878+ forbid_dac = 1;
1879+ }
1880+}
1881+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_VIA, PCI_ANY_ID, via_no_dac);
1882+
1883+static int check_iommu(char *s)
1884+{
1885+ if (!strcmp(s, "usedac")) {
1886+ forbid_dac = -1;
1887+ return 1;
1888+ }
1889+ return 0;
1890+}
1891+__setup("iommu=", check_iommu);
1892+#endif
1893+
1894 dma_addr_t
1895 dma_map_single(struct device *dev, void *ptr, size_t size,
1896 enum dma_data_direction direction)
1897--- a/arch/x86/kernel/process_32-xen.c
1898+++ b/arch/x86/kernel/process_32-xen.c
1899@@ -21,7 +21,6 @@
1900 #include <linux/mm.h>
1901 #include <linux/elfcore.h>
1902 #include <linux/smp.h>
1903-#include <linux/smp_lock.h>
1904 #include <linux/stddef.h>
1905 #include <linux/slab.h>
1906 #include <linux/vmalloc.h>
1907@@ -39,6 +38,7 @@
1908 #include <linux/random.h>
1909 #include <linux/personality.h>
1910 #include <linux/tick.h>
1911+#include <linux/percpu.h>
1912
1913 #include <asm/uaccess.h>
1914 #include <asm/pgtable.h>
1915@@ -61,7 +61,6 @@
1916
1917 #include <asm/tlbflush.h>
1918 #include <asm/cpu.h>
1919-#include <asm/pda.h>
1920
1921 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
1922
1923@@ -70,6 +69,12 @@ static int hlt_counter;
1924 unsigned long boot_option_idle_override = 0;
1925 EXPORT_SYMBOL(boot_option_idle_override);
1926
1927+DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1928+EXPORT_PER_CPU_SYMBOL(current_task);
1929+
1930+DEFINE_PER_CPU(int, cpu_number);
1931+EXPORT_PER_CPU_SYMBOL(cpu_number);
1932+
1933 /*
1934 * Return saved PC of a blocked thread.
1935 */
1936@@ -168,6 +173,7 @@ void cpu_idle(void)
1937 if (__get_cpu_var(cpu_idle_state))
1938 __get_cpu_var(cpu_idle_state) = 0;
1939
1940+ check_pgt_cache();
1941 rmb();
1942 idle = xen_idle; /* no alternatives */
1943
1944@@ -218,18 +224,19 @@ void __devinit select_idle_routine(const
1945 {
1946 }
1947
1948-static int __init idle_setup (char *str)
1949+static int __init idle_setup(char *str)
1950 {
1951- if (!strncmp(str, "poll", 4)) {
1952+ if (!strcmp(str, "poll")) {
1953 printk("using polling idle threads.\n");
1954 pm_idle = poll_idle;
1955 }
1956+ else
1957+ return -1;
1958
1959 boot_option_idle_override = 1;
1960- return 1;
1961+ return 0;
1962 }
1963-
1964-__setup("idle=", idle_setup);
1965+early_param("idle", idle_setup);
1966
1967 void show_regs(struct pt_regs * regs)
1968 {
1969@@ -282,7 +289,7 @@ int kernel_thread(int (*fn)(void *), voi
1970
1971 regs.xds = __USER_DS;
1972 regs.xes = __USER_DS;
1973- regs.xfs = __KERNEL_PDA;
1974+ regs.xfs = __KERNEL_PERCPU;
1975 regs.orig_eax = -1;
1976 regs.eip = (unsigned long) kernel_thread_helper;
1977 regs.xcs = __KERNEL_CS | get_kernel_rpl();
1978@@ -562,7 +569,7 @@ struct task_struct fastcall * __switch_t
1979 * multicall to indicate FPU task switch, rather than
1980 * synchronously trapping to Xen.
1981 */
1982- if (prev_p->thread_info->status & TS_USEDFPU) {
1983+ if (task_thread_info(prev_p)->status & TS_USEDFPU) {
1984 __save_init_fpu(prev_p); /* _not_ save_init_fpu() */
1985 mcl->op = __HYPERVISOR_fpu_taskswitch;
1986 mcl->args[0] = 1;
1987@@ -669,7 +676,7 @@ struct task_struct fastcall * __switch_t
1988 if (prev->gs | next->gs)
1989 loadsegment(gs, next->gs);
1990
1991- write_pda(pcurrent, next_p);
1992+ x86_write_percpu(current_task, next_p);
1993
1994 return prev_p;
1995 }
1996--- a/arch/x86/kernel/process_64-xen.c
1997+++ b/arch/x86/kernel/process_64-xen.c
1998@@ -39,6 +39,7 @@
1999 #include <linux/random.h>
2000 #include <linux/notifier.h>
2001 #include <linux/kprobes.h>
2002+#include <linux/kdebug.h>
2003
2004 #include <asm/uaccess.h>
2005 #include <asm/pgtable.h>
2006@@ -49,7 +50,6 @@
2007 #include <asm/mmu_context.h>
2008 #include <asm/pda.h>
2009 #include <asm/prctl.h>
2010-#include <asm/kdebug.h>
2011 #include <xen/interface/platform.h>
2012 #include <xen/interface/physdev.h>
2013 #include <xen/interface/vcpu.h>
2014@@ -232,16 +232,18 @@ void __cpuinit select_idle_routine(const
2015
2016 static int __init idle_setup (char *str)
2017 {
2018- if (!strncmp(str, "poll", 4)) {
2019+ if (!strcmp(str, "poll")) {
2020 printk("using polling idle threads.\n");
2021 pm_idle = poll_idle;
2022- }
2023+ } else if (!strcmp(str, "mwait"))
2024+ force_mwait = 1;
2025+ else
2026+ return -1;
2027
2028 boot_option_idle_override = 1;
2029- return 1;
2030+ return 0;
2031 }
2032-
2033-__setup("idle=", idle_setup);
2034+early_param("idle", idle_setup);
2035
2036 /* Prints also some state that isn't saved in the pt_regs */
2037 void __show_regs(struct pt_regs * regs)
2038@@ -546,7 +548,7 @@ __switch_to(struct task_struct *prev_p,
2039 * The AMD workaround requires it to be after DS reload, or
2040 * after DS has been cleared, which we do in __prepare_arch_switch.
2041 */
2042- if (prev_p->thread_info->status & TS_USEDFPU) {
2043+ if (task_thread_info(prev_p)->status & TS_USEDFPU) {
2044 __save_init_fpu(prev_p); /* _not_ save_init_fpu() */
2045 mcl->op = __HYPERVISOR_fpu_taskswitch;
2046 mcl->args[0] = 1;
2047--- a/arch/x86/kernel/quirks-xen.c
2048+++ b/arch/x86/kernel/quirks-xen.c
2049@@ -3,12 +3,10 @@
2050 */
2051 #include <linux/pci.h>
2052 #include <linux/irq.h>
2053-#include <asm/pci-direct.h>
2054-#include <asm/genapic.h>
2055-#include <asm/cpu.h>
2056
2057 #if defined(CONFIG_X86_IO_APIC) && (defined(CONFIG_SMP) || defined(CONFIG_XEN)) && defined(CONFIG_PCI)
2058-static void __devinit verify_quirk_intel_irqbalance(struct pci_dev *dev)
2059+
2060+static void __devinit quirk_intel_irqbalance(struct pci_dev *dev)
2061 {
2062 u8 config, rev;
2063 u32 word;
2064@@ -16,7 +14,7 @@ static void __devinit verify_quirk_intel
2065 /* BIOS may enable hardware IRQ balancing for
2066 * E7520/E7320/E7525(revision ID 0x9 and below)
2067 * based platforms.
2068- * For those platforms, make sure that the genapic is set to 'flat'
2069+ * Disable SW irqbalance/affinity on those platforms.
2070 */
2071 pci_read_config_byte(dev, PCI_CLASS_REVISION, &rev);
2072 if (rev > 0x9)
2073@@ -30,59 +28,20 @@ static void __devinit verify_quirk_intel
2074 raw_pci_ops->read(0, 0, 0x40, 0x4c, 2, &word);
2075
2076 if (!(word & (1 << 13))) {
2077-#ifndef CONFIG_XEN
2078-#ifdef CONFIG_X86_64
2079- if (genapic != &apic_flat)
2080- panic("APIC mode must be flat on this system\n");
2081-#elif defined(CONFIG_X86_GENERICARCH)
2082- if (genapic != &apic_default)
2083- panic("APIC mode must be default(flat) on this system. Use apic=default\n");
2084-#endif
2085-#endif
2086- }
2087-
2088- /* put back the original value for config space*/
2089- if (!(config & 0x2))
2090- pci_write_config_byte(dev, 0xf4, config);
2091-}
2092-
2093-void __init quirk_intel_irqbalance(void)
2094-{
2095- u8 config, rev;
2096- u32 word;
2097-
2098- /* BIOS may enable hardware IRQ balancing for
2099- * E7520/E7320/E7525(revision ID 0x9 and below)
2100- * based platforms.
2101- * Disable SW irqbalance/affinity on those platforms.
2102- */
2103- rev = read_pci_config_byte(0, 0, 0, PCI_CLASS_REVISION);
2104- if (rev > 0x9)
2105- return;
2106-
2107- printk(KERN_INFO "Intel E7520/7320/7525 detected.");
2108-
2109- /* enable access to config space */
2110- config = read_pci_config_byte(0, 0, 0, 0xf4);
2111- write_pci_config_byte(0, 0, 0, 0xf4, config|0x2);
2112-
2113- /* read xTPR register */
2114- word = read_pci_config_16(0, 0, 0x40, 0x4c);
2115-
2116- if (!(word & (1 << 13))) {
2117 struct xen_platform_op op;
2118- printk(KERN_INFO "Disabling irq balancing and affinity\n");
2119+
2120+ printk(KERN_INFO "Intel E7520/7320/7525 detected. "
2121+ "Disabling irq balancing and affinity\n");
2122 op.cmd = XENPF_platform_quirk;
2123 op.u.platform_quirk.quirk_id = QUIRK_NOIRQBALANCING;
2124 WARN_ON(HYPERVISOR_platform_op(&op));
2125 }
2126
2127- /* put back the original value for config space */
2128+ /* put back the original value for config space*/
2129 if (!(config & 0x2))
2130- write_pci_config_byte(0, 0, 0, 0xf4, config);
2131+ pci_write_config_byte(dev, 0xf4, config);
2132 }
2133-DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7320_MCH, verify_quirk_intel_irqbalance);
2134-DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7525_MCH, verify_quirk_intel_irqbalance);
2135-DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7520_MCH, verify_quirk_intel_irqbalance);
2136-
2137+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7320_MCH, quirk_intel_irqbalance);
2138+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7525_MCH, quirk_intel_irqbalance);
2139+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7520_MCH, quirk_intel_irqbalance);
2140 #endif
2141--- a/arch/x86/kernel/setup_64-xen.c
2142+++ b/arch/x86/kernel/setup_64-xen.c
2143@@ -120,6 +120,8 @@ int bootloader_type;
2144
2145 unsigned long saved_video_mode;
2146
2147+int force_mwait __cpuinitdata;
2148+
2149 /*
2150 * Early DMI memory
2151 */
2152@@ -253,10 +255,10 @@ static void discover_ebda(void)
2153 * there is a real-mode segmented pointer pointing to the
2154 * 4K EBDA area at 0x40E
2155 */
2156- ebda_addr = *(unsigned short *)EBDA_ADDR_POINTER;
2157+ ebda_addr = *(unsigned short *)__va(EBDA_ADDR_POINTER);
2158 ebda_addr <<= 4;
2159
2160- ebda_size = *(unsigned short *)(unsigned long)ebda_addr;
2161+ ebda_size = *(unsigned short *)__va(ebda_addr);
2162
2163 /* Round EBDA up to pages */
2164 if (ebda_size == 0)
2165@@ -410,15 +412,8 @@ void __init setup_arch(char **cmdline_p)
2166 #endif
2167
2168 #ifdef CONFIG_SMP
2169- /*
2170- * But first pinch a few for the stack/trampoline stuff
2171- * FIXME: Don't need the extra page at 4K, but need to fix
2172- * trampoline before removing it. (see the GDT stuff)
2173- */
2174- reserve_bootmem_generic(PAGE_SIZE, PAGE_SIZE);
2175-
2176 /* Reserve SMP trampoline */
2177- reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, PAGE_SIZE);
2178+ reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, 2*PAGE_SIZE);
2179 #endif
2180 #endif
2181
2182@@ -570,8 +565,6 @@ void __init setup_arch(char **cmdline_p)
2183 early_quirks();
2184 #endif
2185
2186- zap_low_mappings(0);
2187-
2188 /*
2189 * set this early, so we dont allocate cpu0
2190 * if MADT list doesnt list BSP first
2191@@ -864,6 +857,10 @@ static void __cpuinit init_amd(struct cp
2192
2193 /* RDTSC can be speculated around */
2194 clear_bit(X86_FEATURE_SYNC_RDTSC, &c->x86_capability);
2195+
2196+ /* Family 10 doesn't support C states in MWAIT so don't use it */
2197+ if (c->x86 == 0x10 && !force_mwait)
2198+ clear_bit(X86_FEATURE_MWAIT, &c->x86_capability);
2199 }
2200
2201 static void __cpuinit detect_ht(struct cpuinfo_x86 *c)
2202@@ -1146,9 +1143,7 @@ void __cpuinit identify_cpu(struct cpuin
2203 #ifdef CONFIG_X86_MCE
2204 mcheck_init(c);
2205 #endif
2206- if (c == &boot_cpu_data)
2207- mtrr_bp_init();
2208- else
2209+ if (c != &boot_cpu_data)
2210 mtrr_ap_init();
2211 #ifdef CONFIG_NUMA
2212 numa_add_cpu(smp_processor_id());
2213@@ -1239,9 +1234,8 @@ static int show_cpuinfo(struct seq_file
2214 "stc",
2215 "100mhzsteps",
2216 "hwpstate",
2217- NULL, /* tsc invariant mapped to constant_tsc */
2218- NULL,
2219- /* nothing */ /* constant_tsc - moved to flags */
2220+ "", /* tsc invariant mapped to constant_tsc */
2221+ /* nothing */
2222 };
2223
2224
2225--- a/arch/x86/kernel/setup64-xen.c
2226+++ b/arch/x86/kernel/setup64-xen.c
2227@@ -113,9 +113,9 @@ void __init setup_per_cpu_areas(void)
2228 if (!NODE_DATA(cpu_to_node(i))) {
2229 printk("cpu with no node %d, num_online_nodes %d\n",
2230 i, num_online_nodes());
2231- ptr = alloc_bootmem(size);
2232+ ptr = alloc_bootmem_pages(size);
2233 } else {
2234- ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
2235+ ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
2236 }
2237 if (!ptr)
2238 panic("Cannot allocate cpu data for CPU %d\n", i);
2239@@ -208,6 +208,8 @@ char boot_exception_stacks[(N_EXCEPTION_
2240 __attribute__((section(".bss.page_aligned")));
2241 #endif
2242
2243+extern asmlinkage void ignore_sysret(void);
2244+
2245 /* May not be marked __init: used by software suspend */
2246 void syscall_init(void)
2247 {
2248@@ -219,12 +221,22 @@ void syscall_init(void)
2249 */
2250 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
2251 wrmsrl(MSR_LSTAR, system_call);
2252+ wrmsrl(MSR_CSTAR, ignore_sysret);
2253
2254 /* Flags to clear on syscall */
2255 wrmsrl(MSR_SYSCALL_MASK, EF_TF|EF_DF|EF_IE|0x3000);
2256 #endif
2257 #ifdef CONFIG_IA32_EMULATION
2258 syscall32_cpu_init ();
2259+#else
2260+ {
2261+ static const struct callback_register cstar = {
2262+ .type = CALLBACKTYPE_syscall32,
2263+ .address = (unsigned long)ignore_sysret
2264+ };
2265+ if (HYPERVISOR_callback_op(CALLBACKOP_register, &cstar))
2266+ printk(KERN_WARN "Unable to register CSTAR callback\n");
2267+ }
2268 #endif
2269 }
2270
2271@@ -262,7 +274,6 @@ void __cpuinit cpu_init (void)
2272 /* CPU 0 is initialised in head64.c */
2273 if (cpu != 0) {
2274 pda_init(cpu);
2275- zap_low_mappings(cpu);
2276 }
2277 #ifndef CONFIG_X86_NO_TSS
2278 else
2279--- a/arch/x86/kernel/smp_32-xen.c
2280+++ b/arch/x86/kernel/smp_32-xen.c
2281@@ -13,7 +13,6 @@
2282 #include <linux/mm.h>
2283 #include <linux/delay.h>
2284 #include <linux/spinlock.h>
2285-#include <linux/smp_lock.h>
2286 #include <linux/kernel_stat.h>
2287 #include <linux/mc146818rtc.h>
2288 #include <linux/cache.h>
2289@@ -216,7 +215,6 @@ static cpumask_t flush_cpumask;
2290 static struct mm_struct * flush_mm;
2291 static unsigned long flush_va;
2292 static DEFINE_SPINLOCK(tlbstate_lock);
2293-#define FLUSH_ALL 0xffffffff
2294
2295 /*
2296 * We cannot call mmdrop() because we are in interrupt context,
2297@@ -298,7 +296,7 @@ irqreturn_t smp_invalidate_interrupt(int
2298
2299 if (flush_mm == per_cpu(cpu_tlbstate, cpu).active_mm) {
2300 if (per_cpu(cpu_tlbstate, cpu).state == TLBSTATE_OK) {
2301- if (flush_va == FLUSH_ALL)
2302+ if (flush_va == TLB_FLUSH_ALL)
2303 local_flush_tlb();
2304 else
2305 __flush_tlb_one(flush_va);
2306@@ -314,9 +312,11 @@ out:
2307 return IRQ_HANDLED;
2308 }
2309
2310-static void flush_tlb_others(cpumask_t cpumask, struct mm_struct *mm,
2311- unsigned long va)
2312+void native_flush_tlb_others(const cpumask_t *cpumaskp, struct mm_struct *mm,
2313+ unsigned long va)
2314 {
2315+ cpumask_t cpumask = *cpumaskp;
2316+
2317 /*
2318 * A couple of (to be removed) sanity checks:
2319 *
2320@@ -327,10 +327,12 @@ static void flush_tlb_others(cpumask_t c
2321 BUG_ON(cpu_isset(smp_processor_id(), cpumask));
2322 BUG_ON(!mm);
2323
2324+#ifdef CONFIG_HOTPLUG_CPU
2325 /* If a CPU which we ran on has gone down, OK. */
2326 cpus_and(cpumask, cpumask, cpu_online_map);
2327- if (cpus_empty(cpumask))
2328+ if (unlikely(cpus_empty(cpumask)))
2329 return;
2330+#endif
2331
2332 /*
2333 * i'm not happy about this global shared spinlock in the
2334@@ -341,17 +343,7 @@ static void flush_tlb_others(cpumask_t c
2335
2336 flush_mm = mm;
2337 flush_va = va;
2338-#if NR_CPUS <= BITS_PER_LONG
2339- atomic_set_mask(cpumask, &flush_cpumask);
2340-#else
2341- {
2342- int k;
2343- unsigned long *flush_mask = (unsigned long *)&flush_cpumask;
2344- unsigned long *cpu_mask = (unsigned long *)&cpumask;
2345- for (k = 0; k < BITS_TO_LONGS(NR_CPUS); ++k)
2346- atomic_set_mask(cpu_mask[k], &flush_mask[k]);
2347- }
2348-#endif
2349+ cpus_or(flush_cpumask, cpumask, flush_cpumask);
2350 /*
2351 * We have to send the IPI only to
2352 * CPUs affected.
2353@@ -378,7 +370,7 @@ void flush_tlb_current_task(void)
2354
2355 local_flush_tlb();
2356 if (!cpus_empty(cpu_mask))
2357- flush_tlb_others(cpu_mask, mm, FLUSH_ALL);
2358+ flush_tlb_others(cpu_mask, mm, TLB_FLUSH_ALL);
2359 preempt_enable();
2360 }
2361
2362@@ -397,7 +389,7 @@ void flush_tlb_mm (struct mm_struct * mm
2363 leave_mm(smp_processor_id());
2364 }
2365 if (!cpus_empty(cpu_mask))
2366- flush_tlb_others(cpu_mask, mm, FLUSH_ALL);
2367+ flush_tlb_others(cpu_mask, mm, TLB_FLUSH_ALL);
2368
2369 preempt_enable();
2370 }
2371@@ -446,7 +438,7 @@ void flush_tlb_all(void)
2372 * it goes straight through and wastes no time serializing
2373 * anything. Worst case is that we lose a reschedule ...
2374 */
2375-void smp_send_reschedule(int cpu)
2376+void xen_smp_send_reschedule(int cpu)
2377 {
2378 WARN_ON(cpu_is_offline(cpu));
2379 send_IPI_mask(cpumask_of_cpu(cpu), RESCHEDULE_VECTOR);
2380@@ -478,36 +470,79 @@ void unlock_ipi_call_lock(void)
2381
2382 static struct call_data_struct *call_data;
2383
2384+static void __smp_call_function(void (*func) (void *info), void *info,
2385+ int nonatomic, int wait)
2386+{
2387+ struct call_data_struct data;
2388+ int cpus = num_online_cpus() - 1;
2389+
2390+ if (!cpus)
2391+ return;
2392+
2393+ data.func = func;
2394+ data.info = info;
2395+ atomic_set(&data.started, 0);
2396+ data.wait = wait;
2397+ if (wait)
2398+ atomic_set(&data.finished, 0);
2399+
2400+ call_data = &data;
2401+ mb();
2402+
2403+ /* Send a message to all other CPUs and wait for them to respond */
2404+ send_IPI_allbutself(CALL_FUNCTION_VECTOR);
2405+
2406+ /* Wait for response */
2407+ while (atomic_read(&data.started) != cpus)
2408+ cpu_relax();
2409+
2410+ if (wait)
2411+ while (atomic_read(&data.finished) != cpus)
2412+ cpu_relax();
2413+}
2414+
2415+
2416 /**
2417- * smp_call_function(): Run a function on all other CPUs.
2418+ * smp_call_function_mask(): Run a function on a set of other CPUs.
2419+ * @mask: The set of cpus to run on. Must not include the current cpu.
2420 * @func: The function to run. This must be fast and non-blocking.
2421 * @info: An arbitrary pointer to pass to the function.
2422- * @nonatomic: currently unused.
2423 * @wait: If true, wait (atomically) until function has completed on other CPUs.
2424 *
2425- * Returns 0 on success, else a negative status code. Does not return until
2426- * remote CPUs are nearly ready to execute <<func>> or are or have executed.
2427+ * Returns 0 on success, else a negative status code.
2428+ *
2429+ * If @wait is true, then returns once @func has returned; otherwise
2430+ * it returns just before the target cpu calls @func.
2431 *
2432 * You must not call this function with disabled interrupts or from a
2433 * hardware interrupt handler or from a bottom half handler.
2434 */
2435-int smp_call_function (void (*func) (void *info), void *info, int nonatomic,
2436- int wait)
2437+int
2438+xen_smp_call_function_mask(cpumask_t mask,
2439+ void (*func)(void *), void *info,
2440+ int wait)
2441 {
2442 struct call_data_struct data;
2443+ cpumask_t allbutself;
2444 int cpus;
2445
2446+ /* Can deadlock when called with interrupts disabled */
2447+ WARN_ON(irqs_disabled());
2448+
2449 /* Holding any lock stops cpus from going down. */
2450 spin_lock(&call_lock);
2451- cpus = num_online_cpus() - 1;
2452+
2453+ allbutself = cpu_online_map;
2454+ cpu_clear(smp_processor_id(), allbutself);
2455+
2456+ cpus_and(mask, mask, allbutself);
2457+ cpus = cpus_weight(mask);
2458+
2459 if (!cpus) {
2460 spin_unlock(&call_lock);
2461 return 0;
2462 }
2463
2464- /* Can deadlock when called with interrupts disabled */
2465- WARN_ON(irqs_disabled());
2466-
2467 data.func = func;
2468 data.info = info;
2469 atomic_set(&data.started, 0);
2470@@ -517,9 +552,12 @@ int smp_call_function (void (*func) (voi
2471
2472 call_data = &data;
2473 mb();
2474-
2475- /* Send a message to all other CPUs and wait for them to respond */
2476- send_IPI_allbutself(CALL_FUNCTION_VECTOR);
2477+
2478+ /* Send a message to other CPUs */
2479+ if (cpus_equal(mask, allbutself))
2480+ send_IPI_allbutself(CALL_FUNCTION_VECTOR);
2481+ else
2482+ send_IPI_mask(mask, CALL_FUNCTION_VECTOR);
2483
2484 /* Wait for response */
2485 while (atomic_read(&data.started) != cpus)
2486@@ -532,15 +570,14 @@ int smp_call_function (void (*func) (voi
2487
2488 return 0;
2489 }
2490-EXPORT_SYMBOL(smp_call_function);
2491
2492 static void stop_this_cpu (void * dummy)
2493 {
2494+ local_irq_disable();
2495 /*
2496 * Remove this CPU:
2497 */
2498 cpu_clear(smp_processor_id(), cpu_online_map);
2499- local_irq_disable();
2500 disable_all_local_evtchn();
2501 if (cpu_data[smp_processor_id()].hlt_works_ok)
2502 for(;;) halt();
2503@@ -551,13 +588,18 @@ static void stop_this_cpu (void * dummy)
2504 * this function calls the 'stop' function on all other CPUs in the system.
2505 */
2506
2507-void smp_send_stop(void)
2508+void xen_smp_send_stop(void)
2509 {
2510- smp_call_function(stop_this_cpu, NULL, 1, 0);
2511+ /* Don't deadlock on the call lock in panic */
2512+ int nolock = !spin_trylock(&call_lock);
2513+ unsigned long flags;
2514
2515- local_irq_disable();
2516+ local_irq_save(flags);
2517+ __smp_call_function(stop_this_cpu, NULL, 0, 0);
2518+ if (!nolock)
2519+ spin_unlock(&call_lock);
2520 disable_all_local_evtchn();
2521- local_irq_enable();
2522+ local_irq_restore(flags);
2523 }
2524
2525 /*
2526@@ -598,74 +640,3 @@ irqreturn_t smp_call_function_interrupt(
2527
2528 return IRQ_HANDLED;
2529 }
2530-
2531-/*
2532- * this function sends a 'generic call function' IPI to one other CPU
2533- * in the system.
2534- *
2535- * cpu is a standard Linux logical CPU number.
2536- */
2537-static void
2538-__smp_call_function_single(int cpu, void (*func) (void *info), void *info,
2539- int nonatomic, int wait)
2540-{
2541- struct call_data_struct data;
2542- int cpus = 1;
2543-
2544- data.func = func;
2545- data.info = info;
2546- atomic_set(&data.started, 0);
2547- data.wait = wait;
2548- if (wait)
2549- atomic_set(&data.finished, 0);
2550-
2551- call_data = &data;
2552- wmb();
2553- /* Send a message to all other CPUs and wait for them to respond */
2554- send_IPI_mask(cpumask_of_cpu(cpu), CALL_FUNCTION_VECTOR);
2555-
2556- /* Wait for response */
2557- while (atomic_read(&data.started) != cpus)
2558- cpu_relax();
2559-
2560- if (!wait)
2561- return;
2562-
2563- while (atomic_read(&data.finished) != cpus)
2564- cpu_relax();
2565-}
2566-
2567-/*
2568- * smp_call_function_single - Run a function on another CPU
2569- * @func: The function to run. This must be fast and non-blocking.
2570- * @info: An arbitrary pointer to pass to the function.
2571- * @nonatomic: Currently unused.
2572- * @wait: If true, wait until function has completed on other CPUs.
2573- *
2574- * Retrurns 0 on success, else a negative status code.
2575- *
2576- * Does not return until the remote CPU is nearly ready to execute <func>
2577- * or is or has executed.
2578- */
2579-
2580-int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
2581- int nonatomic, int wait)
2582-{
2583- /* prevent preemption and reschedule on another processor */
2584- int me = get_cpu();
2585- if (cpu == me) {
2586- WARN_ON(1);
2587- put_cpu();
2588- return -EBUSY;
2589- }
2590-
2591- /* Can deadlock when called with interrupts disabled */
2592- WARN_ON(irqs_disabled());
2593-
2594- spin_lock_bh(&call_lock);
2595- __smp_call_function_single(cpu, func, info, nonatomic, wait);
2596- spin_unlock_bh(&call_lock);
2597- put_cpu();
2598- return 0;
2599-}
2600-EXPORT_SYMBOL(smp_call_function_single);
2601--- a/arch/x86/kernel/smp_64-xen.c
2602+++ b/arch/x86/kernel/smp_64-xen.c
2603@@ -14,7 +14,6 @@
2604 #include <linux/mm.h>
2605 #include <linux/delay.h>
2606 #include <linux/spinlock.h>
2607-#include <linux/smp_lock.h>
2608 #include <linux/smp.h>
2609 #include <linux/kernel_stat.h>
2610 #include <linux/mc146818rtc.h>
2611@@ -457,44 +456,36 @@ int smp_call_function (void (*func) (voi
2612 }
2613 EXPORT_SYMBOL(smp_call_function);
2614
2615-void smp_stop_cpu(void)
2616+static void stop_this_cpu(void *dummy)
2617 {
2618- unsigned long flags;
2619+ local_irq_disable();
2620 /*
2621 * Remove this CPU:
2622 */
2623 cpu_clear(smp_processor_id(), cpu_online_map);
2624- local_irq_save(flags);
2625 disable_all_local_evtchn();
2626- local_irq_restore(flags);
2627-}
2628-
2629-static void smp_really_stop_cpu(void *dummy)
2630-{
2631- smp_stop_cpu();
2632 for (;;)
2633 halt();
2634 }
2635
2636 void smp_send_stop(void)
2637 {
2638- int nolock = 0;
2639+ int nolock;
2640+ unsigned long flags;
2641+
2642 #ifndef CONFIG_XEN
2643 if (reboot_force)
2644 return;
2645 #endif
2646+
2647 /* Don't deadlock on the call lock in panic */
2648- if (!spin_trylock(&call_lock)) {
2649- /* ignore locking because we have panicked anyways */
2650- nolock = 1;
2651- }
2652- __smp_call_function(smp_really_stop_cpu, NULL, 0, 0);
2653+ nolock = !spin_trylock(&call_lock);
2654+ local_irq_save(flags);
2655+ __smp_call_function(stop_this_cpu, NULL, 0, 0);
2656 if (!nolock)
2657 spin_unlock(&call_lock);
2658-
2659- local_irq_disable();
2660 disable_all_local_evtchn();
2661- local_irq_enable();
2662+ local_irq_restore(flags);
2663 }
2664
2665 /*
2666--- a/arch/x86/kernel/time_32-xen.c
2667+++ b/arch/x86/kernel/time_32-xen.c
2668@@ -80,7 +80,6 @@
2669 #include <asm/i8253.h>
2670 DEFINE_SPINLOCK(i8253_lock);
2671 EXPORT_SYMBOL(i8253_lock);
2672-int pit_latch_buggy; /* extern */
2673 #else
2674 volatile unsigned long __jiffies __section_jiffies = INITIAL_JIFFIES;
2675 #endif
2676@@ -218,6 +217,26 @@ static inline u64 scale_delta(u64 delta,
2677 return product;
2678 }
2679
2680+static inline u64 get64(volatile u64 *ptr)
2681+{
2682+#ifndef CONFIG_64BIT
2683+ return cmpxchg64(ptr, 0, 0);
2684+#else
2685+ return *ptr;
2686+#define cmpxchg64 cmpxchg
2687+#endif
2688+}
2689+
2690+static inline u64 get64_local(volatile u64 *ptr)
2691+{
2692+#ifndef CONFIG_64BIT
2693+ return cmpxchg64_local(ptr, 0, 0);
2694+#else
2695+ return *ptr;
2696+#define cmpxchg64_local cmpxchg_local
2697+#endif
2698+}
2699+
2700 static void init_cpu_khz(void)
2701 {
2702 u64 __cpu_khz = 1000000ULL << 32;
2703@@ -399,7 +418,7 @@ static int set_rtc_mmss(unsigned long no
2704 return retval;
2705 }
2706
2707-unsigned long long sched_clock(void)
2708+static unsigned long long local_clock(void)
2709 {
2710 unsigned int cpu = get_cpu();
2711 struct shadow_time_info *shadow = &per_cpu(shadow_time, cpu);
2712@@ -420,6 +439,61 @@ unsigned long long sched_clock(void)
2713 return time;
2714 }
2715
2716+/*
2717+ * Runstate accounting
2718+ */
2719+static void get_runstate_snapshot(struct vcpu_runstate_info *res)
2720+{
2721+ u64 state_time;
2722+ struct vcpu_runstate_info *state;
2723+
2724+ BUG_ON(preemptible());
2725+
2726+ state = &__get_cpu_var(runstate);
2727+
2728+ do {
2729+ state_time = get64_local(&state->state_entry_time);
2730+ *res = *state;
2731+ } while (get64_local(&state->state_entry_time) != state_time);
2732+
2733+ WARN_ON_ONCE(res->state != RUNSTATE_running);
2734+}
2735+
2736+/*
2737+ * Xen sched_clock implementation. Returns the number of unstolen
2738+ * nanoseconds, which is nanoseconds the VCPU spent in RUNNING+BLOCKED
2739+ * states.
2740+ */
2741+unsigned long long sched_clock(void)
2742+{
2743+ struct vcpu_runstate_info runstate;
2744+ cycle_t now;
2745+ u64 ret;
2746+ s64 offset;
2747+
2748+ /*
2749+ * Ideally sched_clock should be called on a per-cpu basis
2750+ * anyway, so preempt should already be disabled, but that's
2751+ * not current practice at the moment.
2752+ */
2753+ preempt_disable();
2754+
2755+ now = local_clock();
2756+
2757+ get_runstate_snapshot(&runstate);
2758+
2759+ offset = now - runstate.state_entry_time;
2760+ if (offset < 0)
2761+ offset = 0;
2762+
2763+ ret = offset + runstate.time[RUNSTATE_running]
2764+ + runstate.time[RUNSTATE_blocked];
2765+
2766+ preempt_enable();
2767+
2768+ return ret;
2769+}
2770+
2771 unsigned long profile_pc(struct pt_regs *regs)
2772 {
2773 unsigned long pc = instruction_pointer(regs);
2774@@ -467,10 +541,9 @@ EXPORT_SYMBOL(profile_pc);
2775 irqreturn_t timer_interrupt(int irq, void *dev_id)
2776 {
2777 s64 delta, delta_cpu, stolen, blocked;
2778- u64 sched_time;
2779 unsigned int i, cpu = smp_processor_id();
2780 struct shadow_time_info *shadow = &per_cpu(shadow_time, cpu);
2781- struct vcpu_runstate_info *runstate = &per_cpu(runstate, cpu);
2782+ struct vcpu_runstate_info runstate;
2783
2784 /*
2785 * Here we are in the timer irq handler. We just have irqs locally
2786@@ -490,20 +563,7 @@ irqreturn_t timer_interrupt(int irq, voi
2787 delta -= processed_system_time;
2788 delta_cpu -= per_cpu(processed_system_time, cpu);
2789
2790- /*
2791- * Obtain a consistent snapshot of stolen/blocked cycles. We
2792- * can use state_entry_time to detect if we get preempted here.
2793- */
2794- do {
2795- sched_time = runstate->state_entry_time;
2796- barrier();
2797- stolen = runstate->time[RUNSTATE_runnable] +
2798- runstate->time[RUNSTATE_offline] -
2799- per_cpu(processed_stolen_time, cpu);
2800- blocked = runstate->time[RUNSTATE_blocked] -
2801- per_cpu(processed_blocked_time, cpu);
2802- barrier();
2803- } while (sched_time != runstate->state_entry_time);
2804+ get_runstate_snapshot(&runstate);
2805 } while (!time_values_up_to_date(cpu));
2806
2807 if ((unlikely(delta < -(s64)permitted_clock_jitter) ||
2808@@ -545,6 +605,9 @@ irqreturn_t timer_interrupt(int irq, voi
2809 * HACK: Passing NULL to account_steal_time()
2810 * ensures that the ticks are accounted as stolen.
2811 */
2812+ stolen = runstate.time[RUNSTATE_runnable]
2813+ + runstate.time[RUNSTATE_offline]
2814+ - per_cpu(processed_stolen_time, cpu);
2815 if ((stolen > 0) && (delta_cpu > 0)) {
2816 delta_cpu -= stolen;
2817 if (unlikely(delta_cpu < 0))
2818@@ -560,6 +623,8 @@ irqreturn_t timer_interrupt(int irq, voi
2819 * HACK: Passing idle_task to account_steal_time()
2820 * ensures that the ticks are accounted as idle/wait.
2821 */
2822+ blocked = runstate.time[RUNSTATE_blocked]
2823+ - per_cpu(processed_blocked_time, cpu);
2824 if ((blocked > 0) && (delta_cpu > 0)) {
2825 delta_cpu -= blocked;
2826 if (unlikely(delta_cpu < 0))
2827@@ -596,7 +661,7 @@ irqreturn_t timer_interrupt(int irq, voi
2828 return IRQ_HANDLED;
2829 }
2830
2831-void mark_tsc_unstable(void)
2832+void mark_tsc_unstable(char *reason)
2833 {
2834 #ifndef CONFIG_XEN /* XXX Should tell the hypervisor about this fact. */
2835 tsc_unstable = 1;
2836@@ -604,17 +669,13 @@ void mark_tsc_unstable(void)
2837 }
2838 EXPORT_SYMBOL_GPL(mark_tsc_unstable);
2839
2840+static cycle_t cs_last;
2841+
2842 static cycle_t xen_clocksource_read(void)
2843 {
2844 #ifdef CONFIG_SMP
2845- static cycle_t last_ret;
2846-#ifndef CONFIG_64BIT
2847- cycle_t last = cmpxchg64(&last_ret, 0, 0);
2848-#else
2849- cycle_t last = last_ret;
2850-#define cmpxchg64 cmpxchg
2851-#endif
2852- cycle_t ret = sched_clock();
2853+ cycle_t last = get64(&cs_last);
2854+ cycle_t ret = local_clock();
2855
2856 if (unlikely((s64)(ret - last) < 0)) {
2857 if (last - ret > permitted_clock_jitter
2858@@ -633,17 +694,25 @@ static cycle_t xen_clocksource_read(void
2859 }
2860
2861 for (;;) {
2862- cycle_t cur = cmpxchg64(&last_ret, last, ret);
2863+ cycle_t cur = cmpxchg64(&cs_last, last, ret);
2864
2865 if (cur == last || (s64)(ret - cur) < 0)
2866 return ret;
2867 last = cur;
2868 }
2869 #else
2870- return sched_clock();
2871+ return local_clock();
2872 #endif
2873 }
2874
2875+static void xen_clocksource_resume(void)
2876+{
2877+ extern void time_resume(void);
2878+
2879+ time_resume();
2880+ cs_last = local_clock();
2881+}
2882+
2883 static struct clocksource clocksource_xen = {
2884 .name = "xen",
2885 .rating = 400,
2886@@ -652,6 +721,7 @@ static struct clocksource clocksource_xe
2887 .mult = 1 << XEN_SHIFT, /* time directly in nanoseconds */
2888 .shift = XEN_SHIFT,
2889 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
2890+ .resume = xen_clocksource_resume,
2891 };
2892
2893 static void init_missing_ticks_accounting(unsigned int cpu)
2894@@ -740,35 +810,6 @@ void notify_arch_cmos_timer(void)
2895 mod_timer(&sync_xen_wallclock_timer, jiffies + 1);
2896 }
2897
2898-static int timer_resume(struct sys_device *dev)
2899-{
2900- extern void time_resume(void);
2901- time_resume();
2902- return 0;
2903-}
2904-
2905-static struct sysdev_class timer_sysclass = {
2906- .resume = timer_resume,
2907- set_kset_name("timer"),
2908-};
2909-
2910-
2911-/* XXX this driverfs stuff should probably go elsewhere later -john */
2912-static struct sys_device device_timer = {
2913- .id = 0,
2914- .cls = &timer_sysclass,
2915-};
2916-
2917-static int time_init_device(void)
2918-{
2919- int error = sysdev_class_register(&timer_sysclass);
2920- if (!error)
2921- error = sysdev_register(&device_timer);
2922- return error;
2923-}
2924-
2925-device_initcall(time_init_device);
2926-
2927 extern void (*late_time_init)(void);
2928
2929 /* Dynamically-mapped IRQ. */
2930@@ -899,21 +940,21 @@ static void start_hz_timer(void)
2931 cpu_clear(smp_processor_id(), nohz_cpu_mask);
2932 }
2933
2934-void raw_safe_halt(void)
2935+void xen_safe_halt(void)
2936 {
2937 stop_hz_timer();
2938 /* Blocking includes an implicit local_irq_enable(). */
2939 HYPERVISOR_block();
2940 start_hz_timer();
2941 }
2942-EXPORT_SYMBOL(raw_safe_halt);
2943+EXPORT_SYMBOL(xen_safe_halt);
2944
2945-void halt(void)
2946+void xen_halt(void)
2947 {
2948 if (irqs_disabled())
2949 VOID(HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL));
2950 }
2951-EXPORT_SYMBOL(halt);
2952+EXPORT_SYMBOL(xen_halt);
2953
2954 /* No locking required. Interrupts are disabled on all CPUs. */
2955 void time_resume(void)
2956--- a/arch/x86/kernel/traps_32-xen.c
2957+++ b/arch/x86/kernel/traps_32-xen.c
2958@@ -52,7 +52,7 @@
2959 #include <asm/unwind.h>
2960 #include <asm/smp.h>
2961 #include <asm/arch_hooks.h>
2962-#include <asm/kdebug.h>
2963+#include <linux/kdebug.h>
2964 #include <asm/stacktrace.h>
2965
2966 #include <linux/module.h>
2967@@ -101,20 +101,6 @@ asmlinkage void machine_check(void);
2968
2969 int kstack_depth_to_print = 24;
2970 static unsigned int code_bytes = 64;
2971-ATOMIC_NOTIFIER_HEAD(i386die_chain);
2972-
2973-int register_die_notifier(struct notifier_block *nb)
2974-{
2975- vmalloc_sync_all();
2976- return atomic_notifier_chain_register(&i386die_chain, nb);
2977-}
2978-EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
2979-
2980-int unregister_die_notifier(struct notifier_block *nb)
2981-{
2982- return atomic_notifier_chain_unregister(&i386die_chain, nb);
2983-}
2984-EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
2985
2986 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
2987 {
2988@@ -325,7 +311,7 @@ void show_registers(struct pt_regs *regs
2989 regs->xds & 0xffff, regs->xes & 0xffff, regs->xfs & 0xffff, gs, ss);
2990 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
2991 TASK_COMM_LEN, current->comm, current->pid,
2992- current_thread_info(), current, current->thread_info);
2993+ current_thread_info(), current, task_thread_info(current));
2994 /*
2995 * When in-kernel, we also print out the stack and code at the
2996 * time of the fault..
2997@@ -482,8 +468,6 @@ static void __kprobes do_trap(int trapnr
2998 siginfo_t *info)
2999 {
3000 struct task_struct *tsk = current;
3001- tsk->thread.error_code = error_code;
3002- tsk->thread.trap_no = trapnr;
3003
3004 if (regs->eflags & VM_MASK) {
3005 if (vm86)
3006@@ -495,6 +479,18 @@ static void __kprobes do_trap(int trapnr
3007 goto kernel_trap;
3008
3009 trap_signal: {
3010+ /*
3011+ * We want error_code and trap_no set for userspace faults and
3012+ * kernelspace faults which result in die(), but not
3013+ * kernelspace faults which are fixed up. die() gives the
3014+ * process no chance to handle the signal and notice the
3015+ * kernel fault information, so that won't result in polluting
3016+ * the information about previously queued, but not yet
3017+ * delivered, faults. See also do_general_protection below.
3018+ */
3019+ tsk->thread.error_code = error_code;
3020+ tsk->thread.trap_no = trapnr;
3021+
3022 if (info)
3023 force_sig_info(signr, info, tsk);
3024 else
3025@@ -503,8 +499,11 @@ static void __kprobes do_trap(int trapnr
3026 }
3027
3028 kernel_trap: {
3029- if (!fixup_exception(regs))
3030+ if (!fixup_exception(regs)) {
3031+ tsk->thread.error_code = error_code;
3032+ tsk->thread.trap_no = trapnr;
3033 die(str, regs, error_code);
3034+ }
3035 return;
3036 }
3037
3038@@ -578,9 +577,6 @@ DO_ERROR_INFO(32, SIGSEGV, "iret excepti
3039 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
3040 long error_code)
3041 {
3042- current->thread.error_code = error_code;
3043- current->thread.trap_no = 13;
3044-
3045 if (regs->eflags & VM_MASK)
3046 goto gp_in_vm86;
3047
3048@@ -599,6 +595,8 @@ gp_in_vm86:
3049
3050 gp_in_kernel:
3051 if (!fixup_exception(regs)) {
3052+ current->thread.error_code = error_code;
3053+ current->thread.trap_no = 13;
3054 if (notify_die(DIE_GPF, "general protection fault", regs,
3055 error_code, 13, SIGSEGV) == NOTIFY_STOP)
3056 return;
3057@@ -987,9 +985,7 @@ fastcall void do_spurious_interrupt_bug(
3058 fastcall unsigned long patch_espfix_desc(unsigned long uesp,
3059 unsigned long kesp)
3060 {
3061- int cpu = smp_processor_id();
3062- struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
3063- struct desc_struct *gdt = (struct desc_struct *)cpu_gdt_descr->address;
3064+ struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
3065 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
3066 unsigned long new_kesp = kesp - base;
3067 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
3068--- a/arch/x86/kernel/traps_64-xen.c
3069+++ b/arch/x86/kernel/traps_64-xen.c
3070@@ -32,6 +32,7 @@
3071 #include <linux/unwind.h>
3072 #include <linux/uaccess.h>
3073 #include <linux/bug.h>
3074+#include <linux/kdebug.h>
3075
3076 #include <asm/system.h>
3077 #include <asm/io.h>
3078@@ -39,7 +40,6 @@
3079 #include <asm/debugreg.h>
3080 #include <asm/desc.h>
3081 #include <asm/i387.h>
3082-#include <asm/kdebug.h>
3083 #include <asm/processor.h>
3084 #include <asm/unwind.h>
3085 #include <asm/smp.h>
3086@@ -71,22 +71,6 @@ asmlinkage void alignment_check(void);
3087 asmlinkage void machine_check(void);
3088 asmlinkage void spurious_interrupt_bug(void);
3089
3090-ATOMIC_NOTIFIER_HEAD(die_chain);
3091-EXPORT_SYMBOL(die_chain);
3092-
3093-int register_die_notifier(struct notifier_block *nb)
3094-{
3095- vmalloc_sync_all();
3096- return atomic_notifier_chain_register(&die_chain, nb);
3097-}
3098-EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
3099-
3100-int unregister_die_notifier(struct notifier_block *nb)
3101-{
3102- return atomic_notifier_chain_unregister(&die_chain, nb);
3103-}
3104-EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
3105-
3106 static inline void conditional_sti(struct pt_regs *regs)
3107 {
3108 if (regs->eflags & X86_EFLAGS_IF)
3109@@ -428,8 +412,7 @@ void show_registers(struct pt_regs *regs
3110 const int cpu = smp_processor_id();
3111 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
3112
3113- rsp = regs->rsp;
3114-
3115+ rsp = regs->rsp;
3116 printk("CPU %d ", cpu);
3117 __show_regs(regs);
3118 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
3119@@ -440,7 +423,6 @@ void show_registers(struct pt_regs *regs
3120 * time of the fault..
3121 */
3122 if (in_kernel) {
3123-
3124 printk("Stack: ");
3125 _show_stack(NULL, regs, (unsigned long*)rsp);
3126
3127@@ -485,13 +467,14 @@ static unsigned int die_nest_count;
3128
3129 unsigned __kprobes long oops_begin(void)
3130 {
3131- int cpu = smp_processor_id();
3132+ int cpu;
3133 unsigned long flags;
3134
3135 oops_enter();
3136
3137 /* racy, but better than risking deadlock. */
3138 local_irq_save(flags);
3139+ cpu = smp_processor_id();
3140 if (!spin_trylock(&die_lock)) {
3141 if (cpu == die_owner)
3142 /* nested oops. should stop eventually */;
3143@@ -585,10 +568,20 @@ static void __kprobes do_trap(int trapnr
3144 {
3145 struct task_struct *tsk = current;
3146
3147- tsk->thread.error_code = error_code;
3148- tsk->thread.trap_no = trapnr;
3149-
3150 if (user_mode(regs)) {
3151+ /*
3152+ * We want error_code and trap_no set for userspace
3153+ * faults and kernelspace faults which result in
3154+ * die(), but not kernelspace faults which are fixed
3155+ * up. die() gives the process no chance to handle
3156+ * the signal and notice the kernel fault information,
3157+ * so that won't result in polluting the information
3158+ * about previously queued, but not yet delivered,
3159+ * faults. See also do_general_protection below.
3160+ */
3161+ tsk->thread.error_code = error_code;
3162+ tsk->thread.trap_no = trapnr;
3163+
3164 if (exception_trace && unhandled_signal(tsk, signr))
3165 printk(KERN_INFO
3166 "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
3167@@ -609,8 +602,11 @@ static void __kprobes do_trap(int trapnr
3168 fixup = search_exception_tables(regs->rip);
3169 if (fixup)
3170 regs->rip = fixup->fixup;
3171- else
3172+ else {
3173+ tsk->thread.error_code = error_code;
3174+ tsk->thread.trap_no = trapnr;
3175 die(str, regs, error_code);
3176+ }
3177 return;
3178 }
3179 }
3180@@ -686,10 +682,10 @@ asmlinkage void __kprobes do_general_pro
3181
3182 conditional_sti(regs);
3183
3184- tsk->thread.error_code = error_code;
3185- tsk->thread.trap_no = 13;
3186-
3187 if (user_mode(regs)) {
3188+ tsk->thread.error_code = error_code;
3189+ tsk->thread.trap_no = 13;
3190+
3191 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
3192 printk(KERN_INFO
3193 "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
3194@@ -708,6 +704,9 @@ asmlinkage void __kprobes do_general_pro
3195 regs->rip = fixup->fixup;
3196 return;
3197 }
3198+
3199+ tsk->thread.error_code = error_code;
3200+ tsk->thread.trap_no = 13;
3201 if (notify_die(DIE_GPF, "general protection fault", regs,
3202 error_code, 13, SIGSEGV) == NOTIFY_STOP)
3203 return;
3204--- a/arch/x86/kernel/vsyscall_64-xen.c
3205+++ b/arch/x86/kernel/vsyscall_64-xen.c
3206@@ -45,14 +45,34 @@
3207
3208 #define __vsyscall(nr) __attribute__ ((unused,__section__(".vsyscall_" #nr)))
3209 #define __syscall_clobber "r11","rcx","memory"
3210+#define __pa_vsymbol(x) \
3211+ ({unsigned long v; \
3212+ extern char __vsyscall_0; \
3213+ asm("" : "=r" (v) : "0" (x)); \
3214+ ((v - VSYSCALL_FIRST_PAGE) + __pa_symbol(&__vsyscall_0)); })
3215
3216+/*
3217+ * vsyscall_gtod_data contains data that is :
3218+ * - readonly from vsyscalls
3219+ * - writen by timer interrupt or systcl (/proc/sys/kernel/vsyscall64)
3220+ * Try to keep this structure as small as possible to avoid cache line ping pongs
3221+ */
3222 struct vsyscall_gtod_data_t {
3223- seqlock_t lock;
3224- int sysctl_enabled;
3225- struct timeval wall_time_tv;
3226+ seqlock_t lock;
3227+
3228+ /* open coded 'struct timespec' */
3229+ time_t wall_time_sec;
3230+ u32 wall_time_nsec;
3231+
3232+ int sysctl_enabled;
3233 struct timezone sys_tz;
3234- cycle_t offset_base;
3235- struct clocksource clock;
3236+ struct { /* extract of a clocksource struct */
3237+ cycle_t (*vread)(void);
3238+ cycle_t cycle_last;
3239+ cycle_t mask;
3240+ u32 mult;
3241+ u32 shift;
3242+ } clock;
3243 };
3244 int __vgetcpu_mode __section_vgetcpu_mode;
3245
3246@@ -68,9 +88,13 @@ void update_vsyscall(struct timespec *wa
3247
3248 write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
3249 /* copy vsyscall data */
3250- vsyscall_gtod_data.clock = *clock;
3251- vsyscall_gtod_data.wall_time_tv.tv_sec = wall_time->tv_sec;
3252- vsyscall_gtod_data.wall_time_tv.tv_usec = wall_time->tv_nsec/1000;
3253+ vsyscall_gtod_data.clock.vread = clock->vread;
3254+ vsyscall_gtod_data.clock.cycle_last = clock->cycle_last;
3255+ vsyscall_gtod_data.clock.mask = clock->mask;
3256+ vsyscall_gtod_data.clock.mult = clock->mult;
3257+ vsyscall_gtod_data.clock.shift = clock->shift;
3258+ vsyscall_gtod_data.wall_time_sec = wall_time->tv_sec;
3259+ vsyscall_gtod_data.wall_time_nsec = wall_time->tv_nsec;
3260 vsyscall_gtod_data.sys_tz = sys_tz;
3261 write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
3262 }
3263@@ -105,7 +129,8 @@ static __always_inline long time_syscall
3264 static __always_inline void do_vgettimeofday(struct timeval * tv)
3265 {
3266 cycle_t now, base, mask, cycle_delta;
3267- unsigned long seq, mult, shift, nsec_delta;
3268+ unsigned seq;
3269+ unsigned long mult, shift, nsec;
3270 cycle_t (*vread)(void);
3271 do {
3272 seq = read_seqbegin(&__vsyscall_gtod_data.lock);
3273@@ -121,21 +146,20 @@ static __always_inline void do_vgettimeo
3274 mult = __vsyscall_gtod_data.clock.mult;
3275 shift = __vsyscall_gtod_data.clock.shift;
3276
3277- *tv = __vsyscall_gtod_data.wall_time_tv;
3278-
3279+ tv->tv_sec = __vsyscall_gtod_data.wall_time_sec;
3280+ nsec = __vsyscall_gtod_data.wall_time_nsec;
3281 } while (read_seqretry(&__vsyscall_gtod_data.lock, seq));
3282
3283 /* calculate interval: */
3284 cycle_delta = (now - base) & mask;
3285 /* convert to nsecs: */
3286- nsec_delta = (cycle_delta * mult) >> shift;
3287+ nsec += (cycle_delta * mult) >> shift;
3288
3289- /* convert to usecs and add to timespec: */
3290- tv->tv_usec += nsec_delta / NSEC_PER_USEC;
3291- while (tv->tv_usec > USEC_PER_SEC) {
3292+ while (nsec >= NSEC_PER_SEC) {
3293 tv->tv_sec += 1;
3294- tv->tv_usec -= USEC_PER_SEC;
3295+ nsec -= NSEC_PER_SEC;
3296 }
3297+ tv->tv_usec = nsec / NSEC_PER_USEC;
3298 }
3299
3300 int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz)
3301@@ -151,11 +175,16 @@ int __vsyscall(0) vgettimeofday(struct t
3302 * unlikely */
3303 time_t __vsyscall(1) vtime(time_t *t)
3304 {
3305+ struct timeval tv;
3306+ time_t result;
3307 if (unlikely(!__vsyscall_gtod_data.sysctl_enabled))
3308 return time_syscall(t);
3309- else if (t)
3310- *t = __vsyscall_gtod_data.wall_time_tv.tv_sec;
3311- return __vsyscall_gtod_data.wall_time_tv.tv_sec;
3312+
3313+ vgettimeofday(&tv, 0);
3314+ result = tv.tv_sec;
3315+ if (t)
3316+ *t = result;
3317+ return result;
3318 }
3319
3320 /* Fast way to get current CPU and node.
3321@@ -224,10 +253,10 @@ static int vsyscall_sysctl_change(ctl_ta
3322 return ret;
3323 /* gcc has some trouble with __va(__pa()), so just do it this
3324 way. */
3325- map1 = ioremap(__pa_symbol(&vsysc1), 2);
3326+ map1 = ioremap(__pa_vsymbol(&vsysc1), 2);
3327 if (!map1)
3328 return -ENOMEM;
3329- map2 = ioremap(__pa_symbol(&vsysc2), 2);
3330+ map2 = ioremap(__pa_vsymbol(&vsysc2), 2);
3331 if (!map2) {
3332 ret = -ENOMEM;
3333 goto out;
3334@@ -304,7 +333,7 @@ static int __cpuinit
3335 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
3336 {
3337 long cpu = (long)arg;
3338- if (action == CPU_ONLINE)
3339+ if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
3340 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 0, 1);
3341 return NOTIFY_DONE;
3342 }
3343--- a/arch/x86/mm/fault_32-xen.c
3344+++ b/arch/x86/mm/fault_32-xen.c
3345@@ -14,19 +14,20 @@
3346 #include <linux/mman.h>
3347 #include <linux/mm.h>
3348 #include <linux/smp.h>
3349-#include <linux/smp_lock.h>
3350 #include <linux/interrupt.h>
3351 #include <linux/init.h>
3352 #include <linux/tty.h>
3353 #include <linux/vt_kern.h> /* For unblank_screen() */
3354 #include <linux/highmem.h>
3355+#include <linux/bootmem.h> /* for max_low_pfn */
3356+#include <linux/vmalloc.h>
3357 #include <linux/module.h>
3358 #include <linux/kprobes.h>
3359 #include <linux/uaccess.h>
3360+#include <linux/kdebug.h>
3361
3362 #include <asm/system.h>
3363 #include <asm/desc.h>
3364-#include <asm/kdebug.h>
3365 #include <asm/segment.h>
3366
3367 extern void die(const char *,struct pt_regs *,long);
3368@@ -259,25 +260,20 @@ static void dump_fault_path(unsigned lon
3369 unsigned long page;
3370
3371 page = read_cr3();
3372- page = ((unsigned long *) __va(page))[address >> 22];
3373- if (oops_may_print())
3374- printk(KERN_ALERT "*pde = ma %08lx pa %08lx\n", page,
3375- machine_to_phys(page));
3376+ page = ((unsigned long *) __va(page))[address >> PGDIR_SHIFT];
3377+ printk(KERN_ALERT "*pde = ma %08lx pa %08lx\n", page,
3378+ machine_to_phys(page));
3379 /*
3380 * We must not directly access the pte in the highpte
3381 * case if the page table is located in highmem.
3382 * And lets rather not kmap-atomic the pte, just in case
3383 * it's allocated already.
3384 */
3385-#ifdef CONFIG_HIGHPTE
3386- if ((page >> PAGE_SHIFT) >= highstart_pfn)
3387- return;
3388-#endif
3389- if ((page & 1) && oops_may_print()) {
3390- page &= PAGE_MASK;
3391- address &= 0x003ff000;
3392- page = machine_to_phys(page);
3393- page = ((unsigned long *) __va(page))[address >> PAGE_SHIFT];
3394+ if ((machine_to_phys(page) >> PAGE_SHIFT) < max_low_pfn
3395+ && (page & _PAGE_PRESENT)) {
3396+ page = machine_to_phys(page & PAGE_MASK);
3397+ page = ((unsigned long *) __va(page))[(address >> PAGE_SHIFT)
3398+ & (PTRS_PER_PTE - 1)];
3399 printk(KERN_ALERT "*pte = ma %08lx pa %08lx\n", page,
3400 machine_to_phys(page));
3401 }
3402@@ -581,6 +577,11 @@ bad_area:
3403 bad_area_nosemaphore:
3404 /* User mode accesses just cause a SIGSEGV */
3405 if (error_code & 4) {
3406+ /*
3407+ * It's possible to have interrupts off here.
3408+ */
3409+ local_irq_enable();
3410+
3411 /*
3412 * Valid to do another page fault here because this one came
3413 * from user space.
3414@@ -633,7 +634,7 @@ no_context:
3415 bust_spinlocks(1);
3416
3417 if (oops_may_print()) {
3418- #ifdef CONFIG_X86_PAE
3419+#ifdef CONFIG_X86_PAE
3420 if (error_code & 16) {
3421 pte_t *pte = lookup_address(address);
3422
3423@@ -642,7 +643,7 @@ no_context:
3424 "NX-protected page - exploit attempt? "
3425 "(uid: %d)\n", current->uid);
3426 }
3427- #endif
3428+#endif
3429 if (address < PAGE_SIZE)
3430 printk(KERN_ALERT "BUG: unable to handle kernel NULL "
3431 "pointer dereference");
3432@@ -652,8 +653,8 @@ no_context:
3433 printk(" at virtual address %08lx\n",address);
3434 printk(KERN_ALERT " printing eip:\n");
3435 printk("%08lx\n", regs->eip);
3436+ dump_fault_path(address);
3437 }
3438- dump_fault_path(address);
3439 tsk->thread.cr2 = address;
3440 tsk->thread.trap_no = 14;
3441 tsk->thread.error_code = error_code;
3442@@ -694,7 +695,6 @@ do_sigbus:
3443 force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
3444 }
3445
3446-#if !HAVE_SHARED_KERNEL_PMD
3447 void vmalloc_sync_all(void)
3448 {
3449 /*
3450@@ -710,6 +710,9 @@ void vmalloc_sync_all(void)
3451 static unsigned long start = TASK_SIZE;
3452 unsigned long address;
3453
3454+ if (SHARED_KERNEL_PMD)
3455+ return;
3456+
3457 BUILD_BUG_ON(TASK_SIZE & ~PGDIR_MASK);
3458 for (address = start;
3459 address >= TASK_SIZE && address < hypervisor_virt_start;
3460@@ -739,4 +742,3 @@ void vmalloc_sync_all(void)
3461 start = address + (1UL << PMD_SHIFT);
3462 }
3463 }
3464-#endif
3465--- a/arch/x86/mm/fault_64-xen.c
3466+++ b/arch/x86/mm/fault_64-xen.c
3467@@ -15,22 +15,22 @@
3468 #include <linux/mman.h>
3469 #include <linux/mm.h>
3470 #include <linux/smp.h>
3471-#include <linux/smp_lock.h>
3472 #include <linux/interrupt.h>
3473 #include <linux/init.h>
3474 #include <linux/tty.h>
3475 #include <linux/vt_kern.h> /* For unblank_screen() */
3476 #include <linux/compiler.h>
3477+#include <linux/vmalloc.h>
3478 #include <linux/module.h>
3479 #include <linux/kprobes.h>
3480 #include <linux/uaccess.h>
3481+#include <linux/kdebug.h>
3482
3483 #include <asm/system.h>
3484 #include <asm/pgalloc.h>
3485 #include <asm/smp.h>
3486 #include <asm/tlbflush.h>
3487 #include <asm/proto.h>
3488-#include <asm/kdebug.h>
3489 #include <asm-generic/sections.h>
3490
3491 /* Page fault error code bits */
3492@@ -537,6 +537,12 @@ bad_area:
3493 bad_area_nosemaphore:
3494 /* User mode accesses just cause a SIGSEGV */
3495 if (error_code & PF_USER) {
3496+
3497+ /*
3498+ * It's possible to have interrupts off here.
3499+ */
3500+ local_irq_enable();
3501+
3502 if (is_prefetch(regs, address, error_code))
3503 return;
3504
3505@@ -646,7 +652,7 @@ do_sigbus:
3506 }
3507
3508 DEFINE_SPINLOCK(pgd_lock);
3509-struct page *pgd_list;
3510+LIST_HEAD(pgd_list);
3511
3512 void vmalloc_sync_all(void)
3513 {
3514@@ -666,8 +672,7 @@ void vmalloc_sync_all(void)
3515 if (pgd_none(*pgd_ref))
3516 continue;
3517 spin_lock(&pgd_lock);
3518- for (page = pgd_list; page;
3519- page = (struct page *)page->index) {
3520+ list_for_each_entry(page, &pgd_list, lru) {
3521 pgd_t *pgd;
3522 pgd = (pgd_t *)page_address(page) + pgd_index(address);
3523 if (pgd_none(*pgd))
3524--- a/arch/x86/mm/highmem_32-xen.c
3525+++ b/arch/x86/mm/highmem_32-xen.c
3526@@ -26,7 +26,7 @@ void kunmap(struct page *page)
3527 * However when holding an atomic kmap is is not legal to sleep, so atomic
3528 * kmaps are appropriate for short, tight code paths only.
3529 */
3530-static void *__kmap_atomic(struct page *page, enum km_type type, pgprot_t prot)
3531+void *kmap_atomic_prot(struct page *page, enum km_type type, pgprot_t prot)
3532 {
3533 enum fixed_addresses idx;
3534 unsigned long vaddr;
3535@@ -49,15 +49,7 @@ static void *__kmap_atomic(struct page *
3536
3537 void *kmap_atomic(struct page *page, enum km_type type)
3538 {
3539- return __kmap_atomic(page, type, kmap_prot);
3540-}
3541-
3542-/* Same as kmap_atomic but with PAGE_KERNEL_RO page protection. */
3543-void *kmap_atomic_pte(struct page *page, enum km_type type)
3544-{
3545- return __kmap_atomic(page, type,
3546- test_bit(PG_pinned, &page->flags)
3547- ? PAGE_KERNEL_RO : kmap_prot);
3548+ return kmap_atomic_prot(page, type, kmap_prot);
3549 }
3550
3551 void kunmap_atomic(void *kvaddr, enum km_type type)
3552@@ -80,6 +72,7 @@ void kunmap_atomic(void *kvaddr, enum km
3553 #endif
3554 }
3555
3556+ /*arch_flush_lazy_mmu_mode();*/
3557 pagefault_enable();
3558 }
3559
3560@@ -162,7 +155,6 @@ void copy_highpage(struct page *to, stru
3561 EXPORT_SYMBOL(kmap);
3562 EXPORT_SYMBOL(kunmap);
3563 EXPORT_SYMBOL(kmap_atomic);
3564-EXPORT_SYMBOL(kmap_atomic_pte);
3565 EXPORT_SYMBOL(kunmap_atomic);
3566 EXPORT_SYMBOL(kmap_atomic_to_page);
3567 EXPORT_SYMBOL(clear_highpage);
3568--- a/arch/x86/mm/init_32-xen.c
3569+++ b/arch/x86/mm/init_32-xen.c
3570@@ -22,6 +22,7 @@
3571 #include <linux/init.h>
3572 #include <linux/highmem.h>
3573 #include <linux/pagemap.h>
3574+#include <linux/pfn.h>
3575 #include <linux/poison.h>
3576 #include <linux/bootmem.h>
3577 #include <linux/slab.h>
3578@@ -65,17 +66,19 @@ static pmd_t * __init one_md_table_init(
3579 pmd_t *pmd_table;
3580
3581 #ifdef CONFIG_X86_PAE
3582- pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
3583- paravirt_alloc_pd(__pa(pmd_table) >> PAGE_SHIFT);
3584- make_lowmem_page_readonly(pmd_table, XENFEAT_writable_page_tables);
3585- set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
3586- pud = pud_offset(pgd, 0);
3587- if (pmd_table != pmd_offset(pud, 0))
3588- BUG();
3589-#else
3590+ if (!(__pgd_val(*pgd) & _PAGE_PRESENT)) {
3591+ pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
3592+
3593+ paravirt_alloc_pd(__pa(pmd_table) >> PAGE_SHIFT);
3594+ make_lowmem_page_readonly(pmd_table, XENFEAT_writable_page_tables);
3595+ set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
3596+ pud = pud_offset(pgd, 0);
3597+ if (pmd_table != pmd_offset(pud, 0))
3598+ BUG();
3599+ }
3600+#endif
3601 pud = pud_offset(pgd, 0);
3602 pmd_table = pmd_offset(pud, 0);
3603-#endif
3604
3605 return pmd_table;
3606 }
3607@@ -86,16 +89,18 @@ static pmd_t * __init one_md_table_init(
3608 */
3609 static pte_t * __init one_page_table_init(pmd_t *pmd)
3610 {
3611+#if CONFIG_XEN_COMPAT <= 0x030002
3612 if (pmd_none(*pmd)) {
3613+#else
3614+ if (!(__pmd_val(*pmd) & _PAGE_PRESENT)) {
3615+#endif
3616 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
3617+
3618 paravirt_alloc_pt(__pa(page_table) >> PAGE_SHIFT);
3619 make_lowmem_page_readonly(page_table,
3620 XENFEAT_writable_page_tables);
3621 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
3622- if (page_table != pte_offset_kernel(pmd, 0))
3623- BUG();
3624-
3625- return page_table;
3626+ BUG_ON(page_table != pte_offset_kernel(pmd, 0));
3627 }
3628
3629 return pte_offset_kernel(pmd, 0);
3630@@ -115,7 +120,6 @@ static pte_t * __init one_page_table_ini
3631 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
3632 {
3633 pgd_t *pgd;
3634- pud_t *pud;
3635 pmd_t *pmd;
3636 int pgd_idx, pmd_idx;
3637 unsigned long vaddr;
3638@@ -126,12 +130,10 @@ static void __init page_table_range_init
3639 pgd = pgd_base + pgd_idx;
3640
3641 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
3642- if (pgd_none(*pgd))
3643- one_md_table_init(pgd);
3644- pud = pud_offset(pgd, vaddr);
3645- pmd = pmd_offset(pud, vaddr);
3646+ pmd = one_md_table_init(pgd);
3647+ pmd = pmd + pmd_index(vaddr);
3648 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
3649- if (vaddr < hypervisor_virt_start && pmd_none(*pmd))
3650+ if (vaddr < hypervisor_virt_start)
3651 one_page_table_init(pmd);
3652
3653 vaddr += PMD_SIZE;
3654@@ -194,24 +196,25 @@ static void __init kernel_physical_mappi
3655 /* Map with big pages if possible, otherwise create normal page tables. */
3656 if (cpu_has_pse) {
3657 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
3658-
3659 if (is_kernel_text(address) || is_kernel_text(address2))
3660 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
3661 else
3662 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
3663+
3664 pfn += PTRS_PER_PTE;
3665 } else {
3666 pte = one_page_table_init(pmd);
3667
3668- pte += pte_ofs;
3669- for (; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
3670- /* XEN: Only map initial RAM allocation. */
3671- if ((pfn >= max_ram_pfn) || pte_present(*pte))
3672- continue;
3673- if (is_kernel_text(address))
3674- set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
3675- else
3676- set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
3677+ for (pte += pte_ofs;
3678+ pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn;
3679+ pte++, pfn++, pte_ofs++, address += PAGE_SIZE) {
3680+ /* XEN: Only map initial RAM allocation. */
3681+ if ((pfn >= max_ram_pfn) || pte_present(*pte))
3682+ continue;
3683+ if (is_kernel_text(address))
3684+ set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
3685+ else
3686+ set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
3687 }
3688 pte_ofs = 0;
3689 }
3690@@ -381,15 +384,44 @@ extern void __init remap_numa_kva(void);
3691
3692 pgd_t *swapper_pg_dir;
3693
3694+static void __init xen_pagetable_setup_start(pgd_t *base)
3695+{
3696+}
3697+
3698+static void __init xen_pagetable_setup_done(pgd_t *base)
3699+{
3700+}
3701+
3702+/*
3703+ * Build a proper pagetable for the kernel mappings. Up until this
3704+ * point, we've been running on some set of pagetables constructed by
3705+ * the boot process.
3706+ *
3707+ * If we're booting on native hardware, this will be a pagetable
3708+ * constructed in arch/i386/kernel/head.S, and not running in PAE mode
3709+ * (even if we'll end up running in PAE). The root of the pagetable
3710+ * will be swapper_pg_dir.
3711+ *
3712+ * If we're booting paravirtualized under a hypervisor, then there are
3713+ * more options: we may already be running PAE, and the pagetable may
3714+ * or may not be based in swapper_pg_dir. In any case,
3715+ * paravirt_pagetable_setup_start() will set up swapper_pg_dir
3716+ * appropriately for the rest of the initialization to work.
3717+ *
3718+ * In general, pagetable_init() assumes that the pagetable may already
3719+ * be partially populated, and so it avoids stomping on any existing
3720+ * mappings.
3721+ */
3722 static void __init pagetable_init (void)
3723 {
3724- unsigned long vaddr;
3725+ unsigned long vaddr, end;
3726 pgd_t *pgd_base = (pgd_t *)xen_start_info->pt_base;
3727
3728+ xen_pagetable_setup_start(pgd_base);
3729+
3730 /* Enable PSE if available */
3731- if (cpu_has_pse) {
3732+ if (cpu_has_pse)
3733 set_in_cr4(X86_CR4_PSE);
3734- }
3735
3736 /* Enable PGE if available */
3737 if (cpu_has_pge) {
3738@@ -406,9 +438,12 @@ static void __init pagetable_init (void)
3739 * created - mappings will be set by set_fixmap():
3740 */
3741 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
3742- page_table_range_init(vaddr, hypervisor_virt_start, pgd_base);
3743+ end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
3744+ page_table_range_init(vaddr, end, pgd_base);
3745
3746 permanent_kmaps_init(pgd_base);
3747+
3748+ xen_pagetable_setup_done(pgd_base);
3749 }
3750
3751 #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP)
3752@@ -750,34 +785,29 @@ int remove_memory(u64 start, u64 size)
3753 EXPORT_SYMBOL_GPL(remove_memory);
3754 #endif
3755
3756-struct kmem_cache *pgd_cache;
3757 struct kmem_cache *pmd_cache;
3758
3759 void __init pgtable_cache_init(void)
3760 {
3761+ size_t pgd_size = PTRS_PER_PGD*sizeof(pgd_t);
3762+
3763 if (PTRS_PER_PMD > 1) {
3764 pmd_cache = kmem_cache_create("pmd",
3765 PTRS_PER_PMD*sizeof(pmd_t),
3766 PTRS_PER_PMD*sizeof(pmd_t),
3767- 0,
3768+ SLAB_PANIC,
3769 pmd_ctor,
3770 NULL);
3771- if (!pmd_cache)
3772- panic("pgtable_cache_init(): cannot create pmd cache");
3773+ if (!SHARED_KERNEL_PMD) {
3774+ /* If we're in PAE mode and have a non-shared
3775+ kernel pmd, then the pgd size must be a
3776+ page size. This is because the pgd_list
3777+ links through the page structure, so there
3778+ can only be one pgd per page for this to
3779+ work. */
3780+ pgd_size = PAGE_SIZE;
3781+ }
3782 }
3783- pgd_cache = kmem_cache_create("pgd",
3784-#ifndef CONFIG_XEN
3785- PTRS_PER_PGD*sizeof(pgd_t),
3786- PTRS_PER_PGD*sizeof(pgd_t),
3787-#else
3788- PAGE_SIZE,
3789- PAGE_SIZE,
3790-#endif
3791- 0,
3792- pgd_ctor,
3793- PTRS_PER_PMD == 1 ? pgd_dtor : NULL);
3794- if (!pgd_cache)
3795- panic("pgtable_cache_init(): Cannot create pgd cache");
3796 }
3797
3798 /*
3799@@ -811,13 +841,26 @@ static int noinline do_test_wp_bit(void)
3800
3801 void mark_rodata_ro(void)
3802 {
3803- unsigned long addr = (unsigned long)__start_rodata;
3804-
3805- for (; addr < (unsigned long)__end_rodata; addr += PAGE_SIZE)
3806- change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RO);
3807+ unsigned long start = PFN_ALIGN(_text);
3808+ unsigned long size = PFN_ALIGN(_etext) - start;
3809
3810- printk("Write protecting the kernel read-only data: %uk\n",
3811- (__end_rodata - __start_rodata) >> 10);
3812+#ifndef CONFIG_KPROBES
3813+#ifdef CONFIG_HOTPLUG_CPU
3814+ /* It must still be possible to apply SMP alternatives. */
3815+ if (num_possible_cpus() <= 1)
3816+#endif
3817+ {
3818+ change_page_attr(virt_to_page(start),
3819+ size >> PAGE_SHIFT, PAGE_KERNEL_RX);
3820+ printk("Write protecting the kernel text: %luk\n", size >> 10);
3821+ }
3822+#endif
3823+ start += size;
3824+ size = (unsigned long)__end_rodata - start;
3825+ change_page_attr(virt_to_page(start),
3826+ size >> PAGE_SHIFT, PAGE_KERNEL_RO);
3827+ printk("Write protecting the kernel read-only data: %luk\n",
3828+ size >> 10);
3829
3830 /*
3831 * change_page_attr() requires a global_flush_tlb() call after it.
3832@@ -840,7 +883,7 @@ void free_init_pages(char *what, unsigne
3833 free_page(addr);
3834 totalram_pages++;
3835 }
3836- printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
3837+ printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
3838 }
3839
3840 void free_initmem(void)
3841--- a/arch/x86/mm/init_64-xen.c
3842+++ b/arch/x86/mm/init_64-xen.c
3843@@ -25,10 +25,12 @@
3844 #include <linux/bootmem.h>
3845 #include <linux/proc_fs.h>
3846 #include <linux/pci.h>
3847+#include <linux/pfn.h>
3848 #include <linux/poison.h>
3849 #include <linux/dma-mapping.h>
3850 #include <linux/module.h>
3851 #include <linux/memory_hotplug.h>
3852+#include <linux/nmi.h>
3853
3854 #include <asm/processor.h>
3855 #include <asm/system.h>
3856@@ -51,7 +53,7 @@
3857 #define Dprintk(x...)
3858 #endif
3859
3860-struct dma_mapping_ops* dma_ops;
3861+const struct dma_mapping_ops* dma_ops;
3862 EXPORT_SYMBOL(dma_ops);
3863
3864 #if CONFIG_XEN_COMPAT <= 0x030002
3865@@ -191,6 +193,13 @@ void show_mem(void)
3866
3867 for_each_online_pgdat(pgdat) {
3868 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
3869+ /* this loop can take a while with 256 GB and 4k pages
3870+ so update the NMI watchdog */
3871+ if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) {
3872+ touch_nmi_watchdog();
3873+ }
3874+ if (!pfn_valid(pgdat->node_start_pfn + i))
3875+ continue;
3876 page = pfn_to_page(pgdat->node_start_pfn + i);
3877 total++;
3878 if (PageReserved(page))
3879@@ -359,7 +368,7 @@ __set_fixmap (enum fixed_addresses idx,
3880 }
3881 }
3882
3883-unsigned long __initdata table_start, table_end;
3884+unsigned long __meminitdata table_start, table_end;
3885
3886 static __meminit void *alloc_static_page(unsigned long *phys)
3887 {
3888@@ -376,7 +385,7 @@ static __meminit void *alloc_static_page
3889 start_pfn++;
3890 memset((void *)va, 0, PAGE_SIZE);
3891 return (void *)va;
3892-}
3893+}
3894
3895 #define PTE_SIZE PAGE_SIZE
3896
3897@@ -412,28 +421,46 @@ static inline int make_readonly(unsigned
3898
3899 #ifndef CONFIG_XEN
3900 /* Must run before zap_low_mappings */
3901-__init void *early_ioremap(unsigned long addr, unsigned long size)
3902+__meminit void *early_ioremap(unsigned long addr, unsigned long size)
3903 {
3904- unsigned long map = round_down(addr, LARGE_PAGE_SIZE);
3905-
3906- /* actually usually some more */
3907- if (size >= LARGE_PAGE_SIZE) {
3908- return NULL;
3909+ unsigned long vaddr;
3910+ pmd_t *pmd, *last_pmd;
3911+ int i, pmds;
3912+
3913+ pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
3914+ vaddr = __START_KERNEL_map;
3915+ pmd = level2_kernel_pgt;
3916+ last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
3917+ for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
3918+ for (i = 0; i < pmds; i++) {
3919+ if (pmd_present(pmd[i]))
3920+ goto next;
3921+ }
3922+ vaddr += addr & ~PMD_MASK;
3923+ addr &= PMD_MASK;
3924+ for (i = 0; i < pmds; i++, addr += PMD_SIZE)
3925+ set_pmd(pmd + i,__pmd(addr | _KERNPG_TABLE | _PAGE_PSE));
3926+ __flush_tlb();
3927+ return (void *)vaddr;
3928+ next:
3929+ ;
3930 }
3931- set_pmd(temp_mappings[0].pmd, __pmd(map | _KERNPG_TABLE | _PAGE_PSE));
3932- map += LARGE_PAGE_SIZE;
3933- set_pmd(temp_mappings[1].pmd, __pmd(map | _KERNPG_TABLE | _PAGE_PSE));
3934- __flush_tlb();
3935- return temp_mappings[0].address + (addr & (LARGE_PAGE_SIZE-1));
3936+ printk("early_ioremap(0x%lx, %lu) failed\n", addr, size);
3937+ return NULL;
3938 }
3939
3940 /* To avoid virtual aliases later */
3941-__init void early_iounmap(void *addr, unsigned long size)
3942+__meminit void early_iounmap(void *addr, unsigned long size)
3943 {
3944- if ((void *)round_down((unsigned long)addr, LARGE_PAGE_SIZE) != temp_mappings[0].address)
3945- printk("early_iounmap: bad address %p\n", addr);
3946- set_pmd(temp_mappings[0].pmd, __pmd(0));
3947- set_pmd(temp_mappings[1].pmd, __pmd(0));
3948+ unsigned long vaddr;
3949+ pmd_t *pmd;
3950+ int i, pmds;
3951+
3952+ vaddr = (unsigned long)addr;
3953+ pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
3954+ pmd = level2_kernel_pgt + pmd_index(vaddr);
3955+ for (i = 0; i < pmds; i++)
3956+ pmd_clear(pmd + i);
3957 __flush_tlb();
3958 }
3959 #endif
3960@@ -767,14 +794,6 @@ void __meminit init_memory_mapping(unsig
3961 __flush_tlb_all();
3962 }
3963
3964-void __cpuinit zap_low_mappings(int cpu)
3965-{
3966- /* this is not required for Xen */
3967-#if 0
3968- swap_low_mappings();
3969-#endif
3970-}
3971-
3972 #ifndef CONFIG_NUMA
3973 void __init paging_init(void)
3974 {
3975@@ -960,17 +979,6 @@ void __init mem_init(void)
3976 reservedpages << (PAGE_SHIFT-10),
3977 datasize >> 10,
3978 initsize >> 10);
3979-
3980-#ifndef CONFIG_XEN
3981-#ifdef CONFIG_SMP
3982- /*
3983- * Sync boot_level4_pgt mappings with the init_level4_pgt
3984- * except for the low identity mappings which are already zapped
3985- * in init_level4_pgt. This sync-up is essential for AP's bringup
3986- */
3987- memcpy(boot_level4_pgt+1, init_level4_pgt+1, (PTRS_PER_PGD-1)*sizeof(pgd_t));
3988-#endif
3989-#endif
3990 }
3991
3992 void free_init_pages(char *what, unsigned long begin, unsigned long end)
3993@@ -980,7 +988,7 @@ void free_init_pages(char *what, unsigne
3994 if (begin >= end)
3995 return;
3996
3997- printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
3998+ printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
3999 for (addr = begin; addr < end; addr += PAGE_SIZE) {
4000 ClearPageReserved(virt_to_page(addr));
4001 init_page_count(virt_to_page(addr));
4002@@ -989,24 +997,17 @@ void free_init_pages(char *what, unsigne
4003 if (addr >= __START_KERNEL_map) {
4004 /* make_readonly() reports all kernel addresses. */
4005 __make_page_writable(__va(__pa(addr)));
4006- if (HYPERVISOR_update_va_mapping(addr, __pte(0), 0)) {
4007- pgd_t *pgd = pgd_offset_k(addr);
4008- pud_t *pud = pud_offset(pgd, addr);
4009- pmd_t *pmd = pmd_offset(pud, addr);
4010- pte_t *pte = pte_offset_kernel(pmd, addr);
4011-
4012- xen_l1_entry_update(pte, __pte(0)); /* fallback */
4013- }
4014+ change_page_attr_addr(addr, 1, __pgprot(0));
4015 }
4016 free_page(addr);
4017 totalram_pages++;
4018 }
4019+ if (addr > __START_KERNEL_map)
4020+ global_flush_tlb();
4021 }
4022
4023 void free_initmem(void)
4024 {
4025- memset(__initdata_begin, POISON_FREE_INITDATA,
4026- __initdata_end - __initdata_begin);
4027 free_init_pages("unused kernel memory",
4028 (unsigned long)(&__init_begin),
4029 (unsigned long)(&__init_end));
4030@@ -1016,13 +1017,28 @@ void free_initmem(void)
4031
4032 void mark_rodata_ro(void)
4033 {
4034- unsigned long addr = (unsigned long)__start_rodata;
4035+ unsigned long start = (unsigned long)_stext, end;
4036+
4037+#ifdef CONFIG_HOTPLUG_CPU
4038+ /* It must still be possible to apply SMP alternatives. */
4039+ if (num_possible_cpus() > 1)
4040+ start = (unsigned long)_etext;
4041+#endif
4042+
4043+#ifdef CONFIG_KPROBES
4044+ start = (unsigned long)__start_rodata;
4045+#endif
4046+
4047+ end = (unsigned long)__end_rodata;
4048+ start = (start + PAGE_SIZE - 1) & PAGE_MASK;
4049+ end &= PAGE_MASK;
4050+ if (end <= start)
4051+ return;
4052
4053- for (; addr < (unsigned long)__end_rodata; addr += PAGE_SIZE)
4054- change_page_attr_addr(addr, 1, PAGE_KERNEL_RO);
4055+ change_page_attr_addr(start, (end - start) >> PAGE_SHIFT, PAGE_KERNEL_RO);
4056
4057- printk ("Write protecting the kernel read-only data: %luk\n",
4058- (__end_rodata - __start_rodata) >> 10);
4059+ printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
4060+ (end - start) >> 10);
4061
4062 /*
4063 * change_page_attr_addr() requires a global_flush_tlb() call after it.
4064@@ -1175,3 +1191,11 @@ int in_gate_area_no_task(unsigned long a
4065 {
4066 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
4067 }
4068+
4069+#ifndef CONFIG_XEN
4070+void *alloc_bootmem_high_node(pg_data_t *pgdat, unsigned long size)
4071+{
4072+ return __alloc_bootmem_core(pgdat->bdata, size,
4073+ SMP_CACHE_BYTES, (4UL*1024*1024*1024), 0);
4074+}
4075+#endif
4076--- a/arch/x86/mm/ioremap_32-xen.c
4077+++ b/arch/x86/mm/ioremap_32-xen.c
4078@@ -13,6 +13,7 @@
4079 #include <linux/slab.h>
4080 #include <linux/module.h>
4081 #include <linux/io.h>
4082+#include <linux/sched.h>
4083 #include <asm/fixmap.h>
4084 #include <asm/cacheflush.h>
4085 #include <asm/tlbflush.h>
4086--- a/arch/x86/mm/pageattr_64-xen.c
4087+++ b/arch/x86/mm/pageattr_64-xen.c
4088@@ -215,13 +215,13 @@ void mm_pin_all(void)
4089 preempt_enable();
4090 }
4091
4092-void _arch_dup_mmap(struct mm_struct *mm)
4093+void arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
4094 {
4095 if (!mm->context.pinned)
4096 mm_pin(mm);
4097 }
4098
4099-void _arch_exit_mmap(struct mm_struct *mm)
4100+void arch_exit_mmap(struct mm_struct *mm)
4101 {
4102 struct task_struct *tsk = current;
4103
4104@@ -343,10 +343,11 @@ static void flush_kernel_map(void *arg)
4105 struct page *pg;
4106
4107 /* When clflush is available always use it because it is
4108- much cheaper than WBINVD */
4109- if (!cpu_has_clflush)
4110+ much cheaper than WBINVD. Disable clflush for now because
4111+ the high level code is not ready yet */
4112+ if (1 || !cpu_has_clflush)
4113 asm volatile("wbinvd" ::: "memory");
4114- list_for_each_entry(pg, l, lru) {
4115+ else list_for_each_entry(pg, l, lru) {
4116 void *adr = page_address(pg);
4117 if (cpu_has_clflush)
4118 cache_flush_page(adr);
4119@@ -460,16 +461,24 @@ __change_page_attr(unsigned long address
4120 */
4121 int change_page_attr_addr(unsigned long address, int numpages, pgprot_t prot)
4122 {
4123- int err = 0;
4124+ int err = 0, kernel_map = 0;
4125 int i;
4126
4127+ if (address >= __START_KERNEL_map
4128+ && address < __START_KERNEL_map + KERNEL_TEXT_SIZE) {
4129+ address = (unsigned long)__va(__pa(address));
4130+ kernel_map = 1;
4131+ }
4132+
4133 down_write(&init_mm.mmap_sem);
4134 for (i = 0; i < numpages; i++, address += PAGE_SIZE) {
4135 unsigned long pfn = __pa(address) >> PAGE_SHIFT;
4136
4137- err = __change_page_attr(address, pfn, prot, PAGE_KERNEL);
4138- if (err)
4139- break;
4140+ if (!kernel_map || pte_present(pfn_pte(0, prot))) {
4141+ err = __change_page_attr(address, pfn, prot, PAGE_KERNEL);
4142+ if (err)
4143+ break;
4144+ }
4145 /* Handle kernel mapping too which aliases part of the
4146 * lowmem */
4147 if (__pa(address) < KERNEL_TEXT_SIZE) {
4148--- a/arch/x86/mm/pgtable_32-xen.c
4149+++ b/arch/x86/mm/pgtable_32-xen.c
4150@@ -13,6 +13,7 @@
4151 #include <linux/pagemap.h>
4152 #include <linux/spinlock.h>
4153 #include <linux/module.h>
4154+#include <linux/quicklist.h>
4155
4156 #include <asm/system.h>
4157 #include <asm/pgtable.h>
4158@@ -218,8 +219,6 @@ void pmd_ctor(void *pmd, struct kmem_cac
4159 * against pageattr.c; it is the unique case in which a valid change
4160 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
4161 * vmalloc faults work because attached pagetables are never freed.
4162- * The locking scheme was chosen on the basis of manfred's
4163- * recommendations and having no core impact whatsoever.
4164 * -- wli
4165 */
4166 DEFINE_SPINLOCK(pgd_lock);
4167@@ -245,37 +244,54 @@ static inline void pgd_list_del(pgd_t *p
4168 set_page_private(next, (unsigned long)pprev);
4169 }
4170
4171-void pgd_ctor(void *pgd, struct kmem_cache *cache, unsigned long unused)
4172+
4173+
4174+#if (PTRS_PER_PMD == 1)
4175+/* Non-PAE pgd constructor */
4176+void pgd_ctor(void *pgd)
4177 {
4178 unsigned long flags;
4179
4180- if (PTRS_PER_PMD > 1) {
4181- if (HAVE_SHARED_KERNEL_PMD)
4182- clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD,
4183- swapper_pg_dir + USER_PTRS_PER_PGD,
4184- KERNEL_PGD_PTRS);
4185- } else {
4186- spin_lock_irqsave(&pgd_lock, flags);
4187+ /* !PAE, no pagetable sharing */
4188+ memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
4189+
4190+ spin_lock_irqsave(&pgd_lock, flags);
4191+
4192+ /* must happen under lock */
4193+ clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD,
4194+ swapper_pg_dir + USER_PTRS_PER_PGD,
4195+ KERNEL_PGD_PTRS);
4196+
4197+ paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT,
4198+ __pa(swapper_pg_dir) >> PAGE_SHIFT,
4199+ USER_PTRS_PER_PGD,
4200+ KERNEL_PGD_PTRS);
4201+ pgd_list_add(pgd);
4202+ spin_unlock_irqrestore(&pgd_lock, flags);
4203+}
4204+#else /* PTRS_PER_PMD > 1 */
4205+/* PAE pgd constructor */
4206+void pgd_ctor(void *pgd)
4207+{
4208+ /* PAE, kernel PMD may be shared */
4209+
4210+ if (SHARED_KERNEL_PMD) {
4211 clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD,
4212 swapper_pg_dir + USER_PTRS_PER_PGD,
4213 KERNEL_PGD_PTRS);
4214+ } else {
4215 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
4216-
4217- /* must happen under lock */
4218- paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT,
4219- __pa(swapper_pg_dir) >> PAGE_SHIFT,
4220- USER_PTRS_PER_PGD, PTRS_PER_PGD - USER_PTRS_PER_PGD);
4221-
4222- pgd_list_add(pgd);
4223- spin_unlock_irqrestore(&pgd_lock, flags);
4224 }
4225 }
4226+#endif /* PTRS_PER_PMD */
4227
4228-/* never called when PTRS_PER_PMD > 1 */
4229-void pgd_dtor(void *pgd, struct kmem_cache *cache, unsigned long unused)
4230+void pgd_dtor(void *pgd)
4231 {
4232 unsigned long flags; /* can be called from interrupt context */
4233
4234+ if (SHARED_KERNEL_PMD)
4235+ return;
4236+
4237 paravirt_release_pd(__pa(pgd) >> PAGE_SHIFT);
4238 spin_lock_irqsave(&pgd_lock, flags);
4239 pgd_list_del(pgd);
4240@@ -284,11 +300,46 @@ void pgd_dtor(void *pgd, struct kmem_cac
4241 pgd_test_and_unpin(pgd);
4242 }
4243
4244+#define UNSHARED_PTRS_PER_PGD \
4245+ (SHARED_KERNEL_PMD ? USER_PTRS_PER_PGD : PTRS_PER_PGD)
4246+
4247+/* If we allocate a pmd for part of the kernel address space, then
4248+ make sure its initialized with the appropriate kernel mappings.
4249+ Otherwise use a cached zeroed pmd. */
4250+static pmd_t *pmd_cache_alloc(int idx)
4251+{
4252+ pmd_t *pmd;
4253+
4254+ if (idx >= USER_PTRS_PER_PGD) {
4255+ pmd = (pmd_t *)__get_free_page(GFP_KERNEL);
4256+
4257+#ifndef CONFIG_XEN
4258+ if (pmd)
4259+ memcpy(pmd,
4260+ (void *)pgd_page_vaddr(swapper_pg_dir[idx]),
4261+ sizeof(pmd_t) * PTRS_PER_PMD);
4262+#endif
4263+ } else
4264+ pmd = kmem_cache_alloc(pmd_cache, GFP_KERNEL);
4265+
4266+ return pmd;
4267+}
4268+
4269+static void pmd_cache_free(pmd_t *pmd, int idx)
4270+{
4271+ if (idx >= USER_PTRS_PER_PGD) {
4272+ make_lowmem_page_writable(pmd, XENFEAT_writable_page_tables);
4273+ memset(pmd, 0, PTRS_PER_PMD*sizeof(pmd_t));
4274+ free_page((unsigned long)pmd);
4275+ } else
4276+ kmem_cache_free(pmd_cache, pmd);
4277+}
4278+
4279 pgd_t *pgd_alloc(struct mm_struct *mm)
4280 {
4281 int i;
4282- pgd_t *pgd = kmem_cache_alloc(pgd_cache, GFP_KERNEL);
4283- pmd_t **pmd;
4284+ pgd_t *pgd = quicklist_alloc(0, GFP_KERNEL, pgd_ctor);
4285+ pmd_t **pmds = NULL;
4286 unsigned long flags;
4287
4288 pgd_test_and_unpin(pgd);
4289@@ -296,37 +347,40 @@ pgd_t *pgd_alloc(struct mm_struct *mm)
4290 if (PTRS_PER_PMD == 1 || !pgd)
4291 return pgd;
4292
4293- if (HAVE_SHARED_KERNEL_PMD) {
4294- for (i = 0; i < USER_PTRS_PER_PGD; ++i) {
4295- pmd_t *pmd = kmem_cache_alloc(pmd_cache, GFP_KERNEL);
4296- if (!pmd)
4297- goto out_oom;
4298- paravirt_alloc_pd(__pa(pmd) >> PAGE_SHIFT);
4299- set_pgd(&pgd[i], __pgd(1 + __pa(pmd)));
4300+#ifdef CONFIG_XEN
4301+ if (!SHARED_KERNEL_PMD) {
4302+ /*
4303+ * We can race save/restore (if we sleep during a GFP_KERNEL memory
4304+ * allocation). We therefore store virtual addresses of pmds as they
4305+ * do not change across save/restore, and poke the machine addresses
4306+ * into the pgdir under the pgd_lock.
4307+ */
4308+ pmds = kmalloc(PTRS_PER_PGD * sizeof(pmd_t *), GFP_KERNEL);
4309+ if (!pmds) {
4310+ quicklist_free(0, pgd_dtor, pgd);
4311+ return NULL;
4312 }
4313- return pgd;
4314- }
4315-
4316- /*
4317- * We can race save/restore (if we sleep during a GFP_KERNEL memory
4318- * allocation). We therefore store virtual addresses of pmds as they
4319- * do not change across save/restore, and poke the machine addresses
4320- * into the pgdir under the pgd_lock.
4321- */
4322- pmd = kmalloc(PTRS_PER_PGD * sizeof(pmd_t *), GFP_KERNEL);
4323- if (!pmd) {
4324- kmem_cache_free(pgd_cache, pgd);
4325- return NULL;
4326 }
4327+#endif
4328
4329 /* Allocate pmds, remember virtual addresses. */
4330- for (i = 0; i < PTRS_PER_PGD; ++i) {
4331- pmd[i] = kmem_cache_alloc(pmd_cache, GFP_KERNEL);
4332- if (!pmd[i])
4333+ for (i = 0; i < UNSHARED_PTRS_PER_PGD; ++i) {
4334+ pmd_t *pmd = pmd_cache_alloc(i);
4335+
4336+ if (!pmd)
4337 goto out_oom;
4338+
4339 paravirt_alloc_pd(__pa(pmd) >> PAGE_SHIFT);
4340+ if (pmds)
4341+ pmds[i] = pmd;
4342+ else
4343+ set_pgd(&pgd[i], __pgd(1 + __pa(pmd)));
4344 }
4345
4346+#ifdef CONFIG_XEN
4347+ if (SHARED_KERNEL_PMD)
4348+ return pgd;
4349+
4350 spin_lock_irqsave(&pgd_lock, flags);
4351
4352 /* Protect against save/restore: move below 4GB under pgd_lock. */
4353@@ -341,44 +395,43 @@ pgd_t *pgd_alloc(struct mm_struct *mm)
4354
4355 /* Copy kernel pmd contents and write-protect the new pmds. */
4356 for (i = USER_PTRS_PER_PGD; i < PTRS_PER_PGD; i++) {
4357- unsigned long v = (unsigned long)i << PGDIR_SHIFT;
4358- pgd_t *kpgd = pgd_offset_k(v);
4359- pud_t *kpud = pud_offset(kpgd, v);
4360- pmd_t *kpmd = pmd_offset(kpud, v);
4361- memcpy(pmd[i], kpmd, PAGE_SIZE);
4362+ memcpy(pmds[i],
4363+ (void *)pgd_page_vaddr(swapper_pg_dir[i]),
4364+ sizeof(pmd_t) * PTRS_PER_PMD);
4365 make_lowmem_page_readonly(
4366- pmd[i], XENFEAT_writable_page_tables);
4367+ pmds[i], XENFEAT_writable_page_tables);
4368 }
4369
4370 /* It is safe to poke machine addresses of pmds under the pmd_lock. */
4371 for (i = 0; i < PTRS_PER_PGD; i++)
4372- set_pgd(&pgd[i], __pgd(1 + __pa(pmd[i])));
4373+ set_pgd(&pgd[i], __pgd(1 + __pa(pmds[i])));
4374
4375 /* Ensure this pgd gets picked up and pinned on save/restore. */
4376 pgd_list_add(pgd);
4377
4378 spin_unlock_irqrestore(&pgd_lock, flags);
4379
4380- kfree(pmd);
4381+ kfree(pmds);
4382+#endif
4383
4384 return pgd;
4385
4386 out_oom:
4387- if (HAVE_SHARED_KERNEL_PMD) {
4388+ if (!pmds) {
4389 for (i--; i >= 0; i--) {
4390 pgd_t pgdent = pgd[i];
4391 void* pmd = (void *)__va(pgd_val(pgdent)-1);
4392 paravirt_release_pd(__pa(pmd) >> PAGE_SHIFT);
4393- kmem_cache_free(pmd_cache, pmd);
4394+ pmd_cache_free(pmd, i);
4395 }
4396 } else {
4397 for (i--; i >= 0; i--) {
4398- paravirt_release_pd(__pa(pmd[i]) >> PAGE_SHIFT);
4399- kmem_cache_free(pmd_cache, pmd[i]);
4400+ paravirt_release_pd(__pa(pmds[i]) >> PAGE_SHIFT);
4401+ pmd_cache_free(pmds[i], i);
4402 }
4403- kfree(pmd);
4404+ kfree(pmds);
4405 }
4406- kmem_cache_free(pgd_cache, pgd);
4407+ quicklist_free(0, pgd_dtor, pgd);
4408 return NULL;
4409 }
4410
4411@@ -398,35 +451,24 @@ void pgd_free(pgd_t *pgd)
4412
4413 /* in the PAE case user pgd entries are overwritten before usage */
4414 if (PTRS_PER_PMD > 1) {
4415- for (i = 0; i < USER_PTRS_PER_PGD; ++i) {
4416+ for (i = 0; i < UNSHARED_PTRS_PER_PGD; ++i) {
4417 pgd_t pgdent = pgd[i];
4418 void* pmd = (void *)__va(pgd_val(pgdent)-1);
4419 paravirt_release_pd(__pa(pmd) >> PAGE_SHIFT);
4420- kmem_cache_free(pmd_cache, pmd);
4421+ pmd_cache_free(pmd, i);
4422 }
4423
4424- if (!HAVE_SHARED_KERNEL_PMD) {
4425- unsigned long flags;
4426- spin_lock_irqsave(&pgd_lock, flags);
4427- pgd_list_del(pgd);
4428- spin_unlock_irqrestore(&pgd_lock, flags);
4429-
4430- for (i = USER_PTRS_PER_PGD; i < PTRS_PER_PGD; i++) {
4431- pmd_t *pmd = (void *)__va(pgd_val(pgd[i])-1);
4432- make_lowmem_page_writable(
4433- pmd, XENFEAT_writable_page_tables);
4434- memset(pmd, 0, PTRS_PER_PMD*sizeof(pmd_t));
4435- kmem_cache_free(pmd_cache, pmd);
4436- }
4437-
4438- if (!xen_feature(XENFEAT_pae_pgdir_above_4gb))
4439- xen_destroy_contiguous_region(
4440- (unsigned long)pgd, 0);
4441- }
4442+ if (!xen_feature(XENFEAT_pae_pgdir_above_4gb))
4443+ xen_destroy_contiguous_region((unsigned long)pgd, 0);
4444 }
4445
4446 /* in the non-PAE case, free_pgtables() clears user pgd entries */
4447- kmem_cache_free(pgd_cache, pgd);
4448+ quicklist_free(0, pgd_dtor, pgd);
4449+}
4450+
4451+void check_pgt_cache(void)
4452+{
4453+ quicklist_trim(0, pgd_dtor, 25, 16);
4454 }
4455
4456 void make_lowmem_page_readonly(void *va, unsigned int feature)
4457@@ -723,13 +765,13 @@ void mm_pin_all(void)
4458 spin_unlock_irqrestore(&pgd_lock, flags);
4459 }
4460
4461-void _arch_dup_mmap(struct mm_struct *mm)
4462+void arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
4463 {
4464 if (!test_bit(PG_pinned, &virt_to_page(mm->pgd)->flags))
4465 mm_pin(mm);
4466 }
4467
4468-void _arch_exit_mmap(struct mm_struct *mm)
4469+void arch_exit_mmap(struct mm_struct *mm)
4470 {
4471 struct task_struct *tsk = current;
4472
4473--- a/drivers/char/tpm/tpm_xen.c
4474+++ b/drivers/char/tpm/tpm_xen.c
4475@@ -463,7 +463,7 @@ static int tpmif_connect(struct xenbus_d
4476 tp->backend_id = domid;
4477
4478 err = bind_listening_port_to_irqhandler(
4479- domid, tpmif_int, SA_SAMPLE_RANDOM, "tpmif", tp);
4480+ domid, tpmif_int, IRQF_SAMPLE_RANDOM, "tpmif", tp);
4481 if (err <= 0) {
4482 WPRINTK("bind_listening_port_to_irqhandler failed "
4483 "(err=%d)\n", err);
4484--- a/drivers/pci/msi-xen.c
4485+++ b/drivers/pci/msi-xen.c
4486@@ -12,16 +12,15 @@
4487 #include <linux/interrupt.h>
4488 #include <linux/init.h>
4489 #include <linux/ioport.h>
4490-#include <linux/smp_lock.h>
4491 #include <linux/pci.h>
4492 #include <linux/proc_fs.h>
4493 #include <linux/msi.h>
4494+#include <linux/smp.h>
4495
4496 #include <xen/evtchn.h>
4497
4498 #include <asm/errno.h>
4499 #include <asm/io.h>
4500-#include <asm/smp.h>
4501
4502 #include "pci.h"
4503 #include "msi.h"
4504@@ -154,6 +153,7 @@ int register_msi_get_owner(int (*func)(s
4505 get_owner = func;
4506 return 0;
4507 }
4508+EXPORT_SYMBOL(register_msi_get_owner);
4509
4510 int unregister_msi_get_owner(int (*func)(struct pci_dev *dev))
4511 {
4512@@ -162,6 +162,7 @@ int unregister_msi_get_owner(int (*func)
4513 get_owner = NULL;
4514 return 0;
4515 }
4516+EXPORT_SYMBOL(unregister_msi_get_owner);
4517
4518 static int msi_get_dev_owner(struct pci_dev *dev)
4519 {
4520@@ -263,11 +264,6 @@ static int msi_map_vector(struct pci_dev
4521 return msi_map_pirq_to_vector(dev, -1, entry_nr, table_base);
4522 }
4523
4524-static int msi_init(void)
4525-{
4526- return 0;
4527-}
4528-
4529 #ifdef CONFIG_PM
4530 static void __pci_restore_msi_state(struct pci_dev *dev)
4531 {
4532@@ -434,21 +430,32 @@ static int msix_capability_init(struct p
4533 }
4534
4535 /**
4536- * pci_msi_supported - check whether MSI may be enabled on device
4537+ * pci_msi_check_device - check whether MSI may be enabled on a device
4538 * @dev: pointer to the pci_dev data structure of MSI device function
4539+ * @nvec: how many MSIs have been requested ?
4540+ * @type: are we checking for MSI or MSI-X ?
4541 *
4542 * Look at global flags, the device itself, and its parent busses
4543- * to return 0 if MSI are supported for the device.
4544+ * to determine if MSI/-X are supported for the device. If MSI/-X is
4545+ * supported return 0, else return an error code.
4546 **/
4547-static
4548-int pci_msi_supported(struct pci_dev * dev)
4549+static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
4550 {
4551 struct pci_bus *bus;
4552+ int ret;
4553
4554 /* MSI must be globally enabled and supported by the device */
4555 if (!pci_msi_enable || !dev || dev->no_msi)
4556 return -EINVAL;
4557
4558+ /*
4559+ * You can't ask to have 0 or less MSIs configured.
4560+ * a) it's stupid ..
4561+ * b) the list manipulation code assumes nvec >= 1.
4562+ */
4563+ if (nvec < 1)
4564+ return -ERANGE;
4565+
4566 /* Any bridge which does NOT route MSI transactions from it's
4567 * secondary bus to it's primary bus must set NO_MSI flag on
4568 * the secondary pci_bus.
4569@@ -459,6 +466,13 @@ int pci_msi_supported(struct pci_dev * d
4570 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
4571 return -EINVAL;
4572
4573+ ret = arch_msi_check_device(dev, nvec, type);
4574+ if (ret)
4575+ return ret;
4576+
4577+ if (!pci_find_capability(dev, type))
4578+ return -EINVAL;
4579+
4580 return 0;
4581 }
4582
4583@@ -476,18 +490,15 @@ extern int pci_frontend_enable_msi(struc
4584 int pci_enable_msi(struct pci_dev* dev)
4585 {
4586 struct pci_bus *bus;
4587- int pos, temp, status;
4588-
4589- if (pci_msi_supported(dev) < 0)
4590- return -EINVAL;
4591+ int temp, status;
4592
4593 for (bus = dev->bus; bus; bus = bus->parent)
4594 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
4595 return -EINVAL;
4596
4597- status = msi_init();
4598- if (status < 0)
4599- return status;
4600+ status = pci_msi_check_device(dev, 1, PCI_CAP_ID_MSI);
4601+ if (status)
4602+ return status;
4603
4604 #ifdef CONFIG_XEN_PCIDEV_FRONTEND
4605 if (!is_initial_xendomain())
4606@@ -508,10 +519,6 @@ int pci_enable_msi(struct pci_dev* dev)
4607
4608 temp = dev->irq;
4609
4610- pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
4611- if (!pos)
4612- return -EINVAL;
4613-
4614 /* Check whether driver already requested for MSI-X irqs */
4615 if (dev->msix_enabled) {
4616 printk(KERN_INFO "PCI: %s: Can't enable MSI. "
4617@@ -526,15 +533,14 @@ int pci_enable_msi(struct pci_dev* dev)
4618
4619 return status;
4620 }
4621+EXPORT_SYMBOL(pci_enable_msi);
4622
4623 extern void pci_frontend_disable_msi(struct pci_dev* dev);
4624 void pci_disable_msi(struct pci_dev* dev)
4625 {
4626 int pirq;
4627
4628- if (!pci_msi_enable)
4629- return;
4630- if (!dev)
4631+ if (!pci_msi_enable || !dev)
4632 return;
4633
4634 #ifdef CONFIG_XEN_PCIDEV_FRONTEND
4635@@ -559,6 +565,7 @@ void pci_disable_msi(struct pci_dev* dev
4636 pci_intx(dev, 1); /* enable intx */
4637 dev->msi_enabled = 0;
4638 }
4639+EXPORT_SYMBOL(pci_disable_msi);
4640
4641 /**
4642 * pci_enable_msix - configure device's MSI-X capability structure
4643@@ -583,7 +590,7 @@ int pci_enable_msix(struct pci_dev* dev,
4644 int i, j, temp;
4645 u16 control;
4646
4647- if (!entries || pci_msi_supported(dev) < 0)
4648+ if (!entries)
4649 return -EINVAL;
4650
4651 #ifdef CONFIG_XEN_PCIDEV_FRONTEND
4652@@ -621,14 +628,11 @@ int pci_enable_msix(struct pci_dev* dev,
4653 }
4654 #endif
4655
4656- status = msi_init();
4657- if (status < 0)
4658+ status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
4659+ if (status)
4660 return status;
4661
4662 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
4663- if (!pos)
4664- return -EINVAL;
4665-
4666 pci_read_config_word(dev, msi_control_reg(pos), &control);
4667 nr_entries = multi_msix_capable(control);
4668 if (nvec > nr_entries)
4669@@ -660,6 +664,7 @@ int pci_enable_msix(struct pci_dev* dev,
4670
4671 return status;
4672 }
4673+EXPORT_SYMBOL(pci_enable_msix);
4674
4675 extern void pci_frontend_disable_msix(struct pci_dev* dev);
4676 void pci_disable_msix(struct pci_dev* dev)
4677@@ -699,6 +704,7 @@ void pci_disable_msix(struct pci_dev* de
4678 pci_intx(dev, 1); /* enable intx */
4679 dev->msix_enabled = 0;
4680 }
4681+EXPORT_SYMBOL(pci_disable_msix);
4682
4683 /**
4684 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
4685@@ -742,12 +748,57 @@ void pci_no_msi(void)
4686 pci_msi_enable = 0;
4687 }
4688
4689-EXPORT_SYMBOL(pci_enable_msi);
4690-EXPORT_SYMBOL(pci_disable_msi);
4691-EXPORT_SYMBOL(pci_enable_msix);
4692-EXPORT_SYMBOL(pci_disable_msix);
4693-#ifdef CONFIG_XEN
4694-EXPORT_SYMBOL(register_msi_get_owner);
4695-EXPORT_SYMBOL(unregister_msi_get_owner);
4696+void pci_msi_init_pci_dev(struct pci_dev *dev)
4697+{
4698+#ifndef CONFIG_XEN
4699+ INIT_LIST_HEAD(&dev->msi_list);
4700 #endif
4701+}
4702+
4703+
4704+/* Arch hooks */
4705+
4706+int __attribute__ ((weak))
4707+arch_msi_check_device(struct pci_dev* dev, int nvec, int type)
4708+{
4709+ return 0;
4710+}
4711+
4712+#ifndef CONFIG_XEN
4713+int __attribute__ ((weak))
4714+arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *entry)
4715+{
4716+ return 0;
4717+}
4718+
4719+int __attribute__ ((weak))
4720+arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
4721+{
4722+ struct msi_desc *entry;
4723+ int ret;
4724
4725+ list_for_each_entry(entry, &dev->msi_list, list) {
4726+ ret = arch_setup_msi_irq(dev, entry);
4727+ if (ret)
4728+ return ret;
4729+ }
4730+
4731+ return 0;
4732+}
4733+
4734+void __attribute__ ((weak)) arch_teardown_msi_irq(unsigned int irq)
4735+{
4736+ return;
4737+}
4738+
4739+void __attribute__ ((weak))
4740+arch_teardown_msi_irqs(struct pci_dev *dev)
4741+{
4742+ struct msi_desc *entry;
4743+
4744+ list_for_each_entry(entry, &dev->msi_list, list) {
4745+ if (entry->irq != 0)
4746+ arch_teardown_msi_irq(entry->irq);
4747+ }
4748+}
4749+#endif
4750--- a/drivers/xen/blkfront/blkfront.c
4751+++ b/drivers/xen/blkfront/blkfront.c
4752@@ -241,7 +241,7 @@ static int setup_blkring(struct xenbus_d
4753 info->ring_ref = err;
4754
4755 err = bind_listening_port_to_irqhandler(
4756- dev->otherend_id, blkif_int, SA_SAMPLE_RANDOM, "blkif", info);
4757+ dev->otherend_id, blkif_int, IRQF_SAMPLE_RANDOM, "blkif", info);
4758 if (err <= 0) {
4759 xenbus_dev_fatal(dev, err,
4760 "bind_listening_port_to_irqhandler");
4761--- a/drivers/xen/char/mem.c
4762+++ b/drivers/xen/char/mem.c
4763@@ -18,7 +18,6 @@
4764 #include <linux/raw.h>
4765 #include <linux/tty.h>
4766 #include <linux/capability.h>
4767-#include <linux/smp_lock.h>
4768 #include <linux/ptrace.h>
4769 #include <linux/device.h>
4770 #include <asm/pgalloc.h>
4771--- a/drivers/xen/core/hypervisor_sysfs.c
4772+++ b/drivers/xen/core/hypervisor_sysfs.c
4773@@ -50,7 +50,7 @@ static int __init hypervisor_subsys_init
4774 if (!is_running_on_xen())
4775 return -ENODEV;
4776
4777- hypervisor_subsys.kset.kobj.ktype = &hyp_sysfs_kobj_type;
4778+ hypervisor_subsys.kobj.ktype = &hyp_sysfs_kobj_type;
4779 return 0;
4780 }
4781
4782--- a/drivers/xen/core/smpboot.c
4783+++ b/drivers/xen/core/smpboot.c
4784@@ -165,13 +165,12 @@ static void xen_smp_intr_exit(unsigned i
4785
4786 void __cpuinit cpu_bringup(void)
4787 {
4788+ cpu_init();
4789 #ifdef __i386__
4790- cpu_set_gdt(current_thread_info()->cpu);
4791- secondary_cpu_init();
4792+ identify_secondary_cpu(cpu_data + smp_processor_id());
4793 #else
4794- cpu_init();
4795-#endif
4796 identify_cpu(cpu_data + smp_processor_id());
4797+#endif
4798 touch_softlockup_watchdog();
4799 preempt_disable();
4800 local_irq_enable();
4801@@ -191,11 +190,6 @@ static void __cpuinit cpu_initialize_con
4802 static DEFINE_SPINLOCK(ctxt_lock);
4803
4804 struct task_struct *idle = idle_task(cpu);
4805-#ifdef __x86_64__
4806- struct desc_ptr *gdt_descr = &cpu_gdt_descr[cpu];
4807-#else
4808- struct Xgt_desc_struct *gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
4809-#endif
4810
4811 if (cpu_test_and_set(cpu, cpu_initialized_map))
4812 return;
4813@@ -218,11 +212,11 @@ static void __cpuinit cpu_initialize_con
4814 smp_trap_init(ctxt.trap_ctxt);
4815
4816 ctxt.ldt_ents = 0;
4817-
4818- ctxt.gdt_frames[0] = virt_to_mfn(gdt_descr->address);
4819- ctxt.gdt_ents = gdt_descr->size / 8;
4820+ ctxt.gdt_ents = GDT_SIZE / 8;
4821
4822 #ifdef __i386__
4823+ ctxt.gdt_frames[0] = virt_to_mfn(get_cpu_gdt_table(cpu));
4824+
4825 ctxt.user_regs.cs = __KERNEL_CS;
4826 ctxt.user_regs.esp = idle->thread.esp0 - sizeof(struct pt_regs);
4827
4828@@ -235,7 +229,11 @@ static void __cpuinit cpu_initialize_con
4829 ctxt.failsafe_callback_eip = (unsigned long)failsafe_callback;
4830
4831 ctxt.ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
4832+
4833+ ctxt.user_regs.fs = __KERNEL_PERCPU;
4834 #else /* __x86_64__ */
4835+ ctxt.gdt_frames[0] = virt_to_mfn(cpu_gdt_descr[cpu].address);
4836+
4837 ctxt.user_regs.cs = __KERNEL_CS;
4838 ctxt.user_regs.esp = idle->thread.rsp0 - sizeof(struct pt_regs);
4839
4840@@ -265,9 +263,8 @@ void __init smp_prepare_cpus(unsigned in
4841 struct vcpu_get_physid cpu_id;
4842 #ifdef __x86_64__
4843 struct desc_ptr *gdt_descr;
4844-#else
4845- struct Xgt_desc_struct *gdt_descr;
4846 #endif
4847+ void *gdt_addr;
4848
4849 apicid = 0;
4850 if (HYPERVISOR_vcpu_op(VCPUOP_get_physid, 0, &cpu_id) == 0)
4851@@ -317,14 +314,12 @@ void __init smp_prepare_cpus(unsigned in
4852 }
4853 gdt_descr->size = GDT_SIZE;
4854 memcpy((void *)gdt_descr->address, cpu_gdt_table, GDT_SIZE);
4855+ gdt_addr = (void *)gdt_descr->address;
4856 #else
4857- if (unlikely(!init_gdt(cpu, idle)))
4858- continue;
4859- gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
4860+ init_gdt(cpu);
4861+ gdt_addr = get_cpu_gdt_table(cpu);
4862 #endif
4863- make_page_readonly(
4864- (void *)gdt_descr->address,
4865- XENFEAT_writable_descriptor_tables);
4866+ make_page_readonly(gdt_addr, XENFEAT_writable_descriptor_tables);
4867
4868 apicid = cpu;
4869 if (HYPERVISOR_vcpu_op(VCPUOP_get_physid, cpu, &cpu_id) == 0)
4870@@ -338,7 +333,9 @@ void __init smp_prepare_cpus(unsigned in
4871 #ifdef __x86_64__
4872 cpu_pda(cpu)->pcurrent = idle;
4873 cpu_pda(cpu)->cpunumber = cpu;
4874- clear_ti_thread_flag(idle->thread_info, TIF_FORK);
4875+ clear_ti_thread_flag(task_thread_info(idle), TIF_FORK);
4876+#else
4877+ per_cpu(current_task, cpu) = idle;
4878 #endif
4879
4880 irq_ctx_init(cpu);
4881@@ -363,8 +360,12 @@ void __init smp_prepare_cpus(unsigned in
4882 #endif
4883 }
4884
4885-void __devinit smp_prepare_boot_cpu(void)
4886+void __init smp_prepare_boot_cpu(void)
4887 {
4888+#ifdef __i386__
4889+ init_gdt(smp_processor_id());
4890+ switch_to_new_gdt();
4891+#endif
4892 prefill_possible_map();
4893 }
4894
4895--- a/drivers/xen/core/xen_sysfs.c
4896+++ b/drivers/xen/core/xen_sysfs.c
4897@@ -29,12 +29,12 @@ HYPERVISOR_ATTR_RO(type);
4898
4899 static int __init xen_sysfs_type_init(void)
4900 {
4901- return sysfs_create_file(&hypervisor_subsys.kset.kobj, &type_attr.attr);
4902+ return sysfs_create_file(&hypervisor_subsys.kobj, &type_attr.attr);
4903 }
4904
4905 static void xen_sysfs_type_destroy(void)
4906 {
4907- sysfs_remove_file(&hypervisor_subsys.kset.kobj, &type_attr.attr);
4908+ sysfs_remove_file(&hypervisor_subsys.kobj, &type_attr.attr);
4909 }
4910
4911 /* xen version attributes */
4912@@ -90,13 +90,13 @@ static struct attribute_group version_gr
4913
4914 static int __init xen_sysfs_version_init(void)
4915 {
4916- return sysfs_create_group(&hypervisor_subsys.kset.kobj,
4917+ return sysfs_create_group(&hypervisor_subsys.kobj,
4918 &version_group);
4919 }
4920
4921 static void xen_sysfs_version_destroy(void)
4922 {
4923- sysfs_remove_group(&hypervisor_subsys.kset.kobj, &version_group);
4924+ sysfs_remove_group(&hypervisor_subsys.kobj, &version_group);
4925 }
4926
4927 /* UUID */
4928@@ -126,12 +126,12 @@ HYPERVISOR_ATTR_RO(uuid);
4929
4930 static int __init xen_sysfs_uuid_init(void)
4931 {
4932- return sysfs_create_file(&hypervisor_subsys.kset.kobj, &uuid_attr.attr);
4933+ return sysfs_create_file(&hypervisor_subsys.kobj, &uuid_attr.attr);
4934 }
4935
4936 static void xen_sysfs_uuid_destroy(void)
4937 {
4938- sysfs_remove_file(&hypervisor_subsys.kset.kobj, &uuid_attr.attr);
4939+ sysfs_remove_file(&hypervisor_subsys.kobj, &uuid_attr.attr);
4940 }
4941
4942 /* xen compilation attributes */
4943@@ -204,13 +204,13 @@ static struct attribute_group xen_compil
4944
4945 int __init static xen_compilation_init(void)
4946 {
4947- return sysfs_create_group(&hypervisor_subsys.kset.kobj,
4948+ return sysfs_create_group(&hypervisor_subsys.kobj,
4949 &xen_compilation_group);
4950 }
4951
4952 static void xen_compilation_destroy(void)
4953 {
4954- sysfs_remove_group(&hypervisor_subsys.kset.kobj,
4955+ sysfs_remove_group(&hypervisor_subsys.kobj,
4956 &xen_compilation_group);
4957 }
4958
4959@@ -325,13 +325,13 @@ static struct attribute_group xen_proper
4960
4961 static int __init xen_properties_init(void)
4962 {
4963- return sysfs_create_group(&hypervisor_subsys.kset.kobj,
4964+ return sysfs_create_group(&hypervisor_subsys.kobj,
4965 &xen_properties_group);
4966 }
4967
4968 static void xen_properties_destroy(void)
4969 {
4970- sysfs_remove_group(&hypervisor_subsys.kset.kobj,
4971+ sysfs_remove_group(&hypervisor_subsys.kobj,
4972 &xen_properties_group);
4973 }
4974
4975@@ -350,13 +350,13 @@ HYPERVISOR_ATTR_RO(vmcoreinfo);
4976
4977 static int __init xen_sysfs_vmcoreinfo_init(void)
4978 {
4979- return sysfs_create_file(&hypervisor_subsys.kset.kobj,
4980+ return sysfs_create_file(&hypervisor_subsys.kobj,
4981 &vmcoreinfo_attr.attr);
4982 }
4983
4984 static void xen_sysfs_vmcoreinfo_destroy(void)
4985 {
4986- sysfs_remove_file(&hypervisor_subsys.kset.kobj, &vmcoreinfo_attr.attr);
4987+ sysfs_remove_file(&hypervisor_subsys.kobj, &vmcoreinfo_attr.attr);
4988 }
4989
4990 #endif
4991--- a/drivers/xen/netback/netback.c
4992+++ b/drivers/xen/netback/netback.c
4993@@ -179,7 +179,7 @@ static struct sk_buff *netbk_copy_skb(st
4994 goto err;
4995
4996 skb_reserve(nskb, 16 + NET_IP_ALIGN);
4997- headlen = nskb->end - nskb->data;
4998+ headlen = skb_end_pointer(nskb) - nskb->data;
4999 if (headlen > skb_headlen(skb))
5000 headlen = skb_headlen(skb);
5001 ret = skb_copy_bits(skb, 0, __skb_put(nskb, headlen), headlen);
5002@@ -225,11 +225,15 @@ static struct sk_buff *netbk_copy_skb(st
5003 len -= copy;
5004 }
5005
5006+#ifdef NET_SKBUFF_DATA_USES_OFFSET
5007+ offset = 0;
5008+#else
5009 offset = nskb->data - skb->data;
5010+#endif
5011
5012- nskb->h.raw = skb->h.raw + offset;
5013- nskb->nh.raw = skb->nh.raw + offset;
5014- nskb->mac.raw = skb->mac.raw + offset;
5015+ nskb->transport_header = skb->transport_header + offset;
5016+ nskb->network_header = skb->network_header + offset;
5017+ nskb->mac_header = skb->mac_header + offset;
5018
5019 return nskb;
5020
5021@@ -1601,7 +1605,7 @@ static int __init netback_init(void)
5022 (void)bind_virq_to_irqhandler(VIRQ_DEBUG,
5023 0,
5024 netif_be_dbg,
5025- SA_SHIRQ,
5026+ IRQF_SHARED,
5027 "net-be-dbg",
5028 &netif_be_dbg);
5029 #endif
5030--- a/drivers/xen/netfront/netfront.c
5031+++ b/drivers/xen/netfront/netfront.c
5032@@ -513,7 +513,7 @@ static int setup_device(struct xenbus_de
5033 memcpy(netdev->dev_addr, info->mac, ETH_ALEN);
5034
5035 err = bind_listening_port_to_irqhandler(
5036- dev->otherend_id, netif_int, SA_SAMPLE_RANDOM, netdev->name,
5037+ dev->otherend_id, netif_int, IRQF_SAMPLE_RANDOM, netdev->name,
5038 netdev);
5039 if (err < 0)
5040 goto fail;
5041--- a/drivers/xen/pciback/xenbus.c
5042+++ b/drivers/xen/pciback/xenbus.c
5043@@ -99,7 +99,7 @@ static int pciback_do_attach(struct pcib
5044
5045 err = bind_interdomain_evtchn_to_irqhandler(
5046 pdev->xdev->otherend_id, remote_evtchn, pciback_handle_event,
5047- SA_SAMPLE_RANDOM, "pciback", pdev);
5048+ IRQF_SAMPLE_RANDOM, "pciback", pdev);
5049 if (err < 0) {
5050 xenbus_dev_fatal(pdev->xdev, err,
5051 "Error binding event channel to IRQ");
5052--- a/drivers/xen/pcifront/xenbus.c
5053+++ b/drivers/xen/pcifront/xenbus.c
5054@@ -10,10 +10,6 @@
5055 #include <xen/gnttab.h>
5056 #include "pcifront.h"
5057
5058-#ifndef __init_refok
5059-#define __init_refok
5060-#endif
5061-
5062 #define INVALID_GRANT_REF (0)
5063 #define INVALID_EVTCHN (-1)
5064
5065--- a/drivers/xen/scsifront/xenbus.c
5066+++ b/drivers/xen/scsifront/xenbus.c
5067@@ -96,7 +96,7 @@ static int scsifront_alloc_ring(struct v
5068
5069 err = bind_listening_port_to_irqhandler(
5070 dev->otherend_id, scsifront_intr,
5071- SA_SAMPLE_RANDOM, "scsifront", info);
5072+ IRQF_SAMPLE_RANDOM, "scsifront", info);
5073
5074 if (err <= 0) {
5075 xenbus_dev_fatal(dev, err, "bind_listening_port_to_irqhandler");
5076--- a/drivers/xen/sfc_netback/accel_fwd.c
5077+++ b/drivers/xen/sfc_netback/accel_fwd.c
5078@@ -308,7 +308,7 @@ static struct netback_accel *for_a_vnic(
5079 static inline int packet_is_arp_reply(struct sk_buff *skb)
5080 {
5081 return skb->protocol == ntohs(ETH_P_ARP)
5082- && skb->nh.arph->ar_op == ntohs(ARPOP_REPLY);
5083+ && arp_hdr(skb)->ar_op == ntohs(ARPOP_REPLY);
5084 }
5085
5086
5087@@ -392,12 +392,13 @@ void netback_accel_tx_packet(struct sk_b
5088
5089 BUG_ON(fwd_priv == NULL);
5090
5091- if (is_broadcast_ether_addr(skb->mac.raw) && packet_is_arp_reply(skb)) {
5092+ if (is_broadcast_ether_addr(skb_mac_header(skb))
5093+ && packet_is_arp_reply(skb)) {
5094 /*
5095 * update our fast path forwarding to reflect this
5096 * gratuitous ARP
5097 */
5098- mac = skb->mac.raw+ETH_ALEN;
5099+ mac = skb_mac_header(skb)+ETH_ALEN;
5100
5101 DPRINTK("%s: found gratuitous ARP for " MAC_FMT "\n",
5102 __FUNCTION__, MAC_ARG(mac));
5103--- a/drivers/xen/sfc_netback/accel_solarflare.c
5104+++ b/drivers/xen/sfc_netback/accel_solarflare.c
5105@@ -114,7 +114,7 @@ bend_dl_tx_packet(struct efx_dl_device *
5106 BUG_ON(port == NULL);
5107
5108 NETBACK_ACCEL_STATS_OP(global_stats.dl_tx_packets++);
5109- if (skb->mac.raw != NULL)
5110+ if (skb_mac_header_was_set(skb))
5111 netback_accel_tx_packet(skb, port->fwd_priv);
5112 else {
5113 DPRINTK("Ignoring packet with missing mac address\n");
5114--- a/drivers/xen/sfc_netfront/accel_tso.c
5115+++ b/drivers/xen/sfc_netfront/accel_tso.c
5116@@ -33,10 +33,9 @@
5117
5118 #include "accel_tso.h"
5119
5120-#define PTR_DIFF(p1, p2) ((u8*)(p1) - (u8*)(p2))
5121-#define ETH_HDR_LEN(skb) ((skb)->nh.raw - (skb)->data)
5122-#define SKB_TCP_OFF(skb) PTR_DIFF ((skb)->h.th, (skb)->data)
5123-#define SKB_IP_OFF(skb) PTR_DIFF ((skb)->nh.iph, (skb)->data)
5124+#define ETH_HDR_LEN(skb) skb_network_offset(skb)
5125+#define SKB_TCP_OFF(skb) skb_transport_offset(skb)
5126+#define SKB_IP_OFF(skb) skb_network_offset(skb)
5127
5128 /*
5129 * Set a maximum number of buffers in each output packet to make life
5130@@ -114,9 +113,8 @@ struct netfront_accel_tso_state {
5131 static inline void tso_check_safe(struct sk_buff *skb) {
5132 EPRINTK_ON(skb->protocol != htons (ETH_P_IP));
5133 EPRINTK_ON(((struct ethhdr*) skb->data)->h_proto != htons (ETH_P_IP));
5134- EPRINTK_ON(skb->nh.iph->protocol != IPPROTO_TCP);
5135- EPRINTK_ON((SKB_TCP_OFF(skb)
5136- + (skb->h.th->doff << 2u)) > skb_headlen(skb));
5137+ EPRINTK_ON(ip_hdr(skb)->protocol != IPPROTO_TCP);
5138+ EPRINTK_ON((SKB_TCP_OFF(skb) + tcp_hdrlen(skb)) > skb_headlen(skb));
5139 }
5140
5141
5142@@ -129,17 +127,17 @@ static inline void tso_start(struct netf
5143 * All ethernet/IP/TCP headers combined size is TCP header size
5144 * plus offset of TCP header relative to start of packet.
5145 */
5146- st->p.header_length = (skb->h.th->doff << 2u) + SKB_TCP_OFF(skb);
5147+ st->p.header_length = tcp_hdrlen(skb) + SKB_TCP_OFF(skb);
5148 st->p.full_packet_size = (st->p.header_length
5149 + skb_shinfo(skb)->gso_size);
5150 st->p.gso_size = skb_shinfo(skb)->gso_size;
5151
5152- st->p.ip_id = htons(skb->nh.iph->id);
5153- st->seqnum = ntohl(skb->h.th->seq);
5154+ st->p.ip_id = htons(ip_hdr(skb)->id);
5155+ st->seqnum = ntohl(tcp_hdr(skb)->seq);
5156
5157- EPRINTK_ON(skb->h.th->urg);
5158- EPRINTK_ON(skb->h.th->syn);
5159- EPRINTK_ON(skb->h.th->rst);
5160+ EPRINTK_ON(tcp_hdr(skb)->urg);
5161+ EPRINTK_ON(tcp_hdr(skb)->syn);
5162+ EPRINTK_ON(tcp_hdr(skb)->rst);
5163
5164 st->remaining_len = skb->len - st->p.header_length;
5165
5166@@ -258,8 +256,8 @@ int tso_start_new_packet(netfront_accel_
5167 /* This packet will be the last in the TSO burst. */
5168 ip_length = (st->p.header_length - ETH_HDR_LEN(skb)
5169 + st->remaining_len);
5170- tsoh_th->fin = skb->h.th->fin;
5171- tsoh_th->psh = skb->h.th->psh;
5172+ tsoh_th->fin = tcp_hdr(skb)->fin;
5173+ tsoh_th->psh = tcp_hdr(skb)->psh;
5174 }
5175
5176 tsoh_iph->tot_len = htons(ip_length);
5177--- a/drivers/xen/sfc_netfront/accel_vi.c
5178+++ b/drivers/xen/sfc_netfront/accel_vi.c
5179@@ -463,7 +463,7 @@ netfront_accel_enqueue_skb_multi(netfron
5180
5181 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5182 /* Set to zero to encourage falcon to work it out for us */
5183- *(u16*)(skb->h.raw + skb->csum_offset) = 0;
5184+ *(u16*)(skb->head + skb->csum_start + skb->csum_offset) = 0;
5185 }
5186
5187 if (multi_post_start_new_buffer(vnic, &state)) {
5188@@ -582,7 +582,7 @@ netfront_accel_enqueue_skb_single(netfro
5189
5190 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5191 /* Set to zero to encourage falcon to work it out for us */
5192- *(u16*)(skb->h.raw + skb->csum_offset) = 0;
5193+ *(u16*)(skb->head + skb->csum_start + skb->csum_offset) = 0;
5194 }
5195 NETFRONT_ACCEL_PKTBUFF_FOR_EACH_FRAGMENT
5196 (skb, idx, frag_data, frag_len, {
5197--- a/drivers/xen/sfc_netfront/accel_xenbus.c
5198+++ b/drivers/xen/sfc_netfront/accel_xenbus.c
5199@@ -356,7 +356,7 @@ static int vnic_setup_domU_shared_state(
5200 /* Create xenbus msg event channel */
5201 err = bind_listening_port_to_irqhandler
5202 (dev->otherend_id, netfront_accel_msg_channel_irq_from_bend,
5203- SA_SAMPLE_RANDOM, "vnicctrl", vnic);
5204+ IRQF_SAMPLE_RANDOM, "vnicctrl", vnic);
5205 if (err < 0) {
5206 EPRINTK("Couldn't bind msg event channel\n");
5207 goto fail_msg_irq;
5208@@ -367,7 +367,7 @@ static int vnic_setup_domU_shared_state(
5209 /* Create xenbus net event channel */
5210 err = bind_listening_port_to_irqhandler
5211 (dev->otherend_id, netfront_accel_net_channel_irq_from_bend,
5212- SA_SAMPLE_RANDOM, "vnicfront", vnic);
5213+ IRQF_SAMPLE_RANDOM, "vnicfront", vnic);
5214 if (err < 0) {
5215 EPRINTK("Couldn't bind net event channel\n");
5216 goto fail_net_irq;
5217--- a/fs/aio.c
5218+++ b/fs/aio.c
5219@@ -38,7 +38,7 @@
5220
5221 #ifdef CONFIG_EPOLL
5222 #include <linux/poll.h>
5223-#include <linux/eventpoll.h>
5224+#include <linux/anon_inodes.h>
5225 #endif
5226
5227 #if DEBUG > 1
5228@@ -1309,7 +1309,7 @@ static const struct file_operations aioq
5229
5230 /* make_aio_fd:
5231 * Create a file descriptor that can be used to poll the event queue.
5232- * Based and piggybacked on the excellent epoll code.
5233+ * Based on the excellent epoll code.
5234 */
5235
5236 static int make_aio_fd(struct kioctx *ioctx)
5237@@ -1318,7 +1318,8 @@ static int make_aio_fd(struct kioctx *io
5238 struct inode *inode;
5239 struct file *file;
5240
5241- error = ep_getfd(&fd, &inode, &file, NULL, &aioq_fops);
5242+ error = anon_inode_getfd(&fd, &inode, &file, "[aioq]",
5243+ &aioq_fops, ioctx);
5244 if (error)
5245 return error;
5246
5247--- a/include/asm-x86/mach-xen/asm/desc_32.h
5248+++ b/include/asm-x86/mach-xen/asm/desc_32.h
5249@@ -11,23 +11,24 @@
5250
5251 #include <asm/mmu.h>
5252
5253-extern struct desc_struct cpu_gdt_table[GDT_ENTRIES];
5254-
5255 struct Xgt_desc_struct {
5256 unsigned short size;
5257 unsigned long address __attribute__((packed));
5258 unsigned short pad;
5259 } __attribute__ ((packed));
5260
5261-extern struct Xgt_desc_struct idt_descr;
5262-DECLARE_PER_CPU(struct Xgt_desc_struct, cpu_gdt_descr);
5263-extern struct Xgt_desc_struct early_gdt_descr;
5264+struct gdt_page
5265+{
5266+ struct desc_struct gdt[GDT_ENTRIES];
5267+} __attribute__((aligned(PAGE_SIZE)));
5268+DECLARE_PER_CPU(struct gdt_page, gdt_page);
5269
5270 static inline struct desc_struct *get_cpu_gdt_table(unsigned int cpu)
5271 {
5272- return (struct desc_struct *)per_cpu(cpu_gdt_descr, cpu).address;
5273+ return per_cpu(gdt_page, cpu).gdt;
5274 }
5275
5276+extern struct Xgt_desc_struct idt_descr;
5277 extern struct desc_struct idt_table[];
5278 extern void set_intr_gate(unsigned int irq, void * addr);
5279
5280@@ -55,53 +56,32 @@ static inline void pack_gate(__u32 *a, _
5281 #define DESCTYPE_S 0x10 /* !system */
5282
5283 #ifndef CONFIG_XEN
5284-#define load_TR_desc() __asm__ __volatile__("ltr %w0"::"q" (GDT_ENTRY_TSS*8))
5285-
5286-#define load_gdt(dtr) __asm__ __volatile("lgdt %0"::"m" (*dtr))
5287-#define load_idt(dtr) __asm__ __volatile("lidt %0"::"m" (*dtr))
5288+#define load_TR_desc() native_load_tr_desc()
5289+#define load_gdt(dtr) native_load_gdt(dtr)
5290+#define load_idt(dtr) native_load_idt(dtr)
5291 #define load_tr(tr) __asm__ __volatile("ltr %0"::"m" (tr))
5292 #define load_ldt(ldt) __asm__ __volatile("lldt %0"::"m" (ldt))
5293
5294-#define store_gdt(dtr) __asm__ ("sgdt %0":"=m" (*dtr))
5295-#define store_idt(dtr) __asm__ ("sidt %0":"=m" (*dtr))
5296-#define store_tr(tr) __asm__ ("str %0":"=m" (tr))
5297+#define store_gdt(dtr) native_store_gdt(dtr)
5298+#define store_idt(dtr) native_store_idt(dtr)
5299+#define store_tr(tr) (tr = native_store_tr())
5300 #define store_ldt(ldt) __asm__ ("sldt %0":"=m" (ldt))
5301-#endif
5302
5303-#if TLS_SIZE != 24
5304-# error update this code.
5305-#endif
5306-
5307-static inline void load_TLS(struct thread_struct *t, unsigned int cpu)
5308-{
5309-#define C(i) if (HYPERVISOR_update_descriptor(virt_to_machine(&get_cpu_gdt_table(cpu)[GDT_ENTRY_TLS_MIN + i]), \
5310- *(u64 *)&t->tls_array[i]) \
5311- BUG()
5312- C(0); C(1); C(2);
5313-#undef C
5314-}
5315+#define load_TLS(t, cpu) native_load_tls(t, cpu)
5316+#define set_ldt native_set_ldt
5317
5318-#ifndef CONFIG_XEN
5319 #define write_ldt_entry(dt, entry, a, b) write_dt_entry(dt, entry, a, b)
5320 #define write_gdt_entry(dt, entry, a, b) write_dt_entry(dt, entry, a, b)
5321 #define write_idt_entry(dt, entry, a, b) write_dt_entry(dt, entry, a, b)
5322
5323-static inline void write_dt_entry(void *dt, int entry, __u32 entry_a, __u32 entry_b)
5324+static inline void write_dt_entry(struct desc_struct *dt,
5325+ int entry, u32 entry_low, u32 entry_high)
5326 {
5327- __u32 *lp = (__u32 *)((char *)dt + entry*8);
5328- *lp = entry_a;
5329- *(lp+1) = entry_b;
5330+ dt[entry].a = entry_low;
5331+ dt[entry].b = entry_high;
5332 }
5333-#define set_ldt native_set_ldt
5334-#else
5335-extern int write_ldt_entry(void *ldt, int entry, __u32 entry_a, __u32 entry_b);
5336-extern int write_gdt_entry(void *gdt, int entry, __u32 entry_a, __u32 entry_b);
5337-#define set_ldt xen_set_ldt
5338-#endif
5339
5340-#ifndef CONFIG_XEN
5341-static inline fastcall void native_set_ldt(const void *addr,
5342- unsigned int entries)
5343+static inline void native_set_ldt(const void *addr, unsigned int entries)
5344 {
5345 if (likely(entries == 0))
5346 __asm__ __volatile__("lldt %w0"::"q" (0));
5347@@ -116,6 +96,65 @@ static inline fastcall void native_set_l
5348 __asm__ __volatile__("lldt %w0"::"q" (GDT_ENTRY_LDT*8));
5349 }
5350 }
5351+
5352+
5353+static inline void native_load_tr_desc(void)
5354+{
5355+ asm volatile("ltr %w0"::"q" (GDT_ENTRY_TSS*8));
5356+}
5357+
5358+static inline void native_load_gdt(const struct Xgt_desc_struct *dtr)
5359+{
5360+ asm volatile("lgdt %0"::"m" (*dtr));
5361+}
5362+
5363+static inline void native_load_idt(const struct Xgt_desc_struct *dtr)
5364+{
5365+ asm volatile("lidt %0"::"m" (*dtr));
5366+}
5367+
5368+static inline void native_store_gdt(struct Xgt_desc_struct *dtr)
5369+{
5370+ asm ("sgdt %0":"=m" (*dtr));
5371+}
5372+
5373+static inline void native_store_idt(struct Xgt_desc_struct *dtr)
5374+{
5375+ asm ("sidt %0":"=m" (*dtr));
5376+}
5377+
5378+static inline unsigned long native_store_tr(void)
5379+{
5380+ unsigned long tr;
5381+ asm ("str %0":"=r" (tr));
5382+ return tr;
5383+}
5384+
5385+static inline void native_load_tls(struct thread_struct *t, unsigned int cpu)
5386+{
5387+ unsigned int i;
5388+ struct desc_struct *gdt = get_cpu_gdt_table(cpu);
5389+
5390+ for (i = 0; i < GDT_ENTRY_TLS_ENTRIES; i++)
5391+ gdt[GDT_ENTRY_TLS_MIN + i] = t->tls_array[i];
5392+}
5393+#else
5394+#define load_TLS(t, cpu) xen_load_tls(t, cpu)
5395+#define set_ldt xen_set_ldt
5396+
5397+extern int write_ldt_entry(void *ldt, int entry, __u32 entry_a, __u32 entry_b);
5398+extern int write_gdt_entry(void *gdt, int entry, __u32 entry_a, __u32 entry_b);
5399+
5400+static inline void xen_load_tls(struct thread_struct *t, unsigned int cpu)
5401+{
5402+ unsigned int i;
5403+ struct desc_struct *gdt = get_cpu_gdt_table(cpu) + GDT_ENTRY_TLS_MIN;
5404+
5405+ for (i = 0; i < GDT_ENTRY_TLS_ENTRIES; i++)
5406+ if (HYPERVISOR_update_descriptor(virt_to_machine(&gdt[i]),
5407+ *(u64 *)&t->tls_array[i]))
5408+ BUG();
5409+}
5410 #endif
5411
5412 #ifndef CONFIG_X86_NO_IDT
5413--- a/include/asm-x86/mach-xen/asm/desc_64.h
5414+++ b/include/asm-x86/mach-xen/asm/desc_64.h
5415@@ -127,16 +127,6 @@ static inline void set_ldt_desc(unsigned
5416 DESC_LDT, size * 8 - 1);
5417 }
5418
5419-static inline void set_seg_base(unsigned cpu, int entry, void *base)
5420-{
5421- struct desc_struct *d = &cpu_gdt(cpu)[entry];
5422- u32 addr = (u32)(u64)base;
5423- BUG_ON((u64)base >> 32);
5424- d->base0 = addr & 0xffff;
5425- d->base1 = (addr >> 16) & 0xff;
5426- d->base2 = (addr >> 24) & 0xff;
5427-}
5428-
5429 #define LDT_entry_a(info) \
5430 ((((info)->base_addr & 0x0000ffff) << 16) | ((info)->limit & 0x0ffff))
5431 /* Don't allow setting of the lm bit. It is useless anyways because
5432@@ -165,25 +155,15 @@ static inline void set_seg_base(unsigned
5433 (info)->useable == 0 && \
5434 (info)->lm == 0)
5435
5436-#if TLS_SIZE != 24
5437-# error update this code.
5438-#endif
5439-
5440 static inline void load_TLS(struct thread_struct *t, unsigned int cpu)
5441 {
5442-#if 0
5443+ unsigned int i;
5444 u64 *gdt = (u64 *)(cpu_gdt(cpu) + GDT_ENTRY_TLS_MIN);
5445- gdt[0] = t->tls_array[0];
5446- gdt[1] = t->tls_array[1];
5447- gdt[2] = t->tls_array[2];
5448-#endif
5449-#define C(i) \
5450- if (HYPERVISOR_update_descriptor(virt_to_machine(&cpu_gdt(cpu)[GDT_ENTRY_TLS_MIN + i]), \
5451- t->tls_array[i])) \
5452- BUG();
5453
5454- C(0); C(1); C(2);
5455-#undef C
5456+ for (i = 0; i < GDT_ENTRY_TLS_ENTRIES; i++)
5457+ if (HYPERVISOR_update_descriptor(virt_to_machine(&gdt[i]),
5458+ t->tls_array[i]))
5459+ BUG();
5460 }
5461
5462 /*
5463--- a/include/asm-x86/mach-xen/asm/dma-mapping_64.h
5464+++ b/include/asm-x86/mach-xen/asm/dma-mapping_64.h
5465@@ -51,7 +51,7 @@ struct dma_mapping_ops {
5466 };
5467
5468 extern dma_addr_t bad_dma_address;
5469-extern struct dma_mapping_ops* dma_ops;
5470+extern const struct dma_mapping_ops* dma_ops;
5471 extern int iommu_merge;
5472
5473 #if 0
5474--- a/include/asm-x86/mach-xen/asm/fixmap_32.h
5475+++ b/include/asm-x86/mach-xen/asm/fixmap_32.h
5476@@ -19,10 +19,8 @@
5477 * the start of the fixmap.
5478 */
5479 extern unsigned long __FIXADDR_TOP;
5480-#ifdef CONFIG_COMPAT_VDSO
5481-#define FIXADDR_USER_START __fix_to_virt(FIX_VDSO)
5482-#define FIXADDR_USER_END __fix_to_virt(FIX_VDSO - 1)
5483-#endif
5484+#define FIXADDR_USER_START __fix_to_virt(FIX_VDSO)
5485+#define FIXADDR_USER_END __fix_to_virt(FIX_VDSO - 1)
5486
5487 #ifndef __ASSEMBLY__
5488 #include <linux/kernel.h>
5489@@ -85,6 +83,9 @@ enum fixed_addresses {
5490 #ifdef CONFIG_PCI_MMCONFIG
5491 FIX_PCIE_MCFG,
5492 #endif
5493+#ifdef CONFIG_PARAVIRT
5494+ FIX_PARAVIRT_BOOTMAP,
5495+#endif
5496 FIX_SHARED_INFO,
5497 #define NR_FIX_ISAMAPS 256
5498 FIX_ISAMAP_END,
5499--- a/include/asm-x86/mach-xen/asm/fixmap_64.h
5500+++ b/include/asm-x86/mach-xen/asm/fixmap_64.h
5501@@ -15,7 +15,6 @@
5502 #include <asm/apicdef.h>
5503 #include <asm/page.h>
5504 #include <asm/vsyscall.h>
5505-#include <asm/vsyscall32.h>
5506 #include <asm/acpi.h>
5507
5508 /*
5509--- a/include/asm-x86/mach-xen/asm/highmem.h
5510+++ b/include/asm-x86/mach-xen/asm/highmem.h
5511@@ -67,12 +67,18 @@ extern void FASTCALL(kunmap_high(struct
5512
5513 void *kmap(struct page *page);
5514 void kunmap(struct page *page);
5515+void *kmap_atomic_prot(struct page *page, enum km_type type, pgprot_t prot);
5516 void *kmap_atomic(struct page *page, enum km_type type);
5517 void *kmap_atomic_pte(struct page *page, enum km_type type);
5518 void kunmap_atomic(void *kvaddr, enum km_type type);
5519 void *kmap_atomic_pfn(unsigned long pfn, enum km_type type);
5520 struct page *kmap_atomic_to_page(void *ptr);
5521
5522+#define kmap_atomic_pte(page, type) \
5523+ kmap_atomic_prot(page, type, \
5524+ test_bit(PG_pinned, &(page)->flags) \
5525+ ? PAGE_KERNEL_RO : kmap_prot)
5526+
5527 #define flush_cache_kmaps() do { } while (0)
5528
5529 void clear_highpage(struct page *);
5530--- a/include/asm-x86/mach-xen/asm/io_32.h
5531+++ b/include/asm-x86/mach-xen/asm/io_32.h
5532@@ -263,15 +263,18 @@ static inline void flush_write_buffers(v
5533
5534 #endif /* __KERNEL__ */
5535
5536-#define __SLOW_DOWN_IO "outb %%al,$0x80;"
5537+static inline void xen_io_delay(void)
5538+{
5539+ asm volatile("outb %%al,$0x80" : : : "memory");
5540+}
5541
5542 static inline void slow_down_io(void) {
5543- __asm__ __volatile__(
5544- __SLOW_DOWN_IO
5545+ xen_io_delay();
5546 #ifdef REALLY_SLOW_IO
5547- __SLOW_DOWN_IO __SLOW_DOWN_IO __SLOW_DOWN_IO
5548+ xen_io_delay();
5549+ xen_io_delay();
5550+ xen_io_delay();
5551 #endif
5552- : : );
5553 }
5554
5555 #ifdef CONFIG_X86_NUMAQ
5556--- a/include/asm-x86/mach-xen/asm/irqflags_32.h
5557+++ b/include/asm-x86/mach-xen/asm/irqflags_32.h
5558@@ -11,6 +11,40 @@
5559 #define _ASM_IRQFLAGS_H
5560
5561 #ifndef __ASSEMBLY__
5562+#define xen_save_fl(void) (current_vcpu_info()->evtchn_upcall_mask)
5563+
5564+#define xen_restore_fl(f) \
5565+do { \
5566+ vcpu_info_t *_vcpu; \
5567+ barrier(); \
5568+ _vcpu = current_vcpu_info(); \
5569+ if ((_vcpu->evtchn_upcall_mask = (f)) == 0) { \
5570+ barrier(); /* unmask then check (avoid races) */\
5571+ if (unlikely(_vcpu->evtchn_upcall_pending)) \
5572+ force_evtchn_callback(); \
5573+ } \
5574+} while (0)
5575+
5576+#define xen_irq_disable() \
5577+do { \
5578+ current_vcpu_info()->evtchn_upcall_mask = 1; \
5579+ barrier(); \
5580+} while (0)
5581+
5582+#define xen_irq_enable() \
5583+do { \
5584+ vcpu_info_t *_vcpu; \
5585+ barrier(); \
5586+ _vcpu = current_vcpu_info(); \
5587+ _vcpu->evtchn_upcall_mask = 0; \
5588+ barrier(); /* unmask then check (avoid races) */ \
5589+ if (unlikely(_vcpu->evtchn_upcall_pending)) \
5590+ force_evtchn_callback(); \
5591+} while (0)
5592+
5593+void xen_safe_halt(void);
5594+
5595+void xen_halt(void);
5596
5597 /*
5598 * The use of 'barrier' in the following reflects their use as local-lock
5599@@ -20,48 +54,31 @@
5600 * includes these barriers, for example.
5601 */
5602
5603-#define __raw_local_save_flags() (current_vcpu_info()->evtchn_upcall_mask)
5604+#define __raw_local_save_flags() xen_save_fl()
5605
5606-#define raw_local_irq_restore(x) \
5607-do { \
5608- vcpu_info_t *_vcpu; \
5609- barrier(); \
5610- _vcpu = current_vcpu_info(); \
5611- if ((_vcpu->evtchn_upcall_mask = (x)) == 0) { \
5612- barrier(); /* unmask then check (avoid races) */ \
5613- if (unlikely(_vcpu->evtchn_upcall_pending)) \
5614- force_evtchn_callback(); \
5615- } \
5616-} while (0)
5617+#define raw_local_irq_restore(flags) xen_restore_fl(flags)
5618
5619-#define raw_local_irq_disable() \
5620-do { \
5621- current_vcpu_info()->evtchn_upcall_mask = 1; \
5622- barrier(); \
5623-} while (0)
5624+#define raw_local_irq_disable() xen_irq_disable()
5625
5626-#define raw_local_irq_enable() \
5627-do { \
5628- vcpu_info_t *_vcpu; \
5629- barrier(); \
5630- _vcpu = current_vcpu_info(); \
5631- _vcpu->evtchn_upcall_mask = 0; \
5632- barrier(); /* unmask then check (avoid races) */ \
5633- if (unlikely(_vcpu->evtchn_upcall_pending)) \
5634- force_evtchn_callback(); \
5635-} while (0)
5636+#define raw_local_irq_enable() xen_irq_enable()
5637
5638 /*
5639 * Used in the idle loop; sti takes one instruction cycle
5640 * to complete:
5641 */
5642-void raw_safe_halt(void);
5643+static inline void raw_safe_halt(void)
5644+{
5645+ xen_safe_halt();
5646+}
5647
5648 /*
5649 * Used when interrupts are already enabled or to
5650 * shutdown the processor:
5651 */
5652-void halt(void);
5653+static inline void halt(void)
5654+{
5655+ xen_halt();
5656+}
5657
5658 /*
5659 * For spinlocks, etc:
5660--- a/include/asm-x86/mach-xen/asm/irqflags_64.h
5661+++ b/include/asm-x86/mach-xen/asm/irqflags_64.h
5662@@ -9,6 +9,7 @@
5663 */
5664 #ifndef _ASM_IRQFLAGS_H
5665 #define _ASM_IRQFLAGS_H
5666+#include <asm/processor-flags.h>
5667
5668 #ifndef __ASSEMBLY__
5669 /*
5670@@ -50,19 +51,19 @@ static inline void raw_local_irq_disable
5671 {
5672 unsigned long flags = __raw_local_save_flags();
5673
5674- raw_local_irq_restore((flags & ~(1 << 9)) | (1 << 18));
5675+ raw_local_irq_restore((flags & ~X86_EFLAGS_IF) | X86_EFLAGS_AC);
5676 }
5677
5678 static inline void raw_local_irq_enable(void)
5679 {
5680 unsigned long flags = __raw_local_save_flags();
5681
5682- raw_local_irq_restore((flags | (1 << 9)) & ~(1 << 18));
5683+ raw_local_irq_restore((flags | X86_EFLAGS_IF) & (~X86_EFLAGS_AC));
5684 }
5685
5686 static inline int raw_irqs_disabled_flags(unsigned long flags)
5687 {
5688- return !(flags & (1<<9)) || (flags & (1 << 18));
5689+ return !(flags & X86_EFLAGS_IF) || (flags & X86_EFLAGS_AC);
5690 }
5691
5692 #else /* CONFIG_X86_VSMP */
5693@@ -118,13 +119,21 @@ static inline int raw_irqs_disabled_flag
5694 * Used in the idle loop; sti takes one instruction cycle
5695 * to complete:
5696 */
5697-void raw_safe_halt(void);
5698+void xen_safe_halt(void);
5699+static inline void raw_safe_halt(void)
5700+{
5701+ xen_safe_halt();
5702+}
5703
5704 /*
5705 * Used when interrupts are already enabled or to
5706 * shutdown the processor:
5707 */
5708-void halt(void);
5709+void xen_halt(void);
5710+static inline void halt(void)
5711+{
5712+ xen_halt();
5713+}
5714
5715 #else /* __ASSEMBLY__: */
5716 # ifdef CONFIG_TRACE_IRQFLAGS
5717--- a/include/asm-x86/mach-xen/asm/mmu_context_32.h
5718+++ b/include/asm-x86/mach-xen/asm/mmu_context_32.h
5719@@ -6,6 +6,20 @@
5720 #include <asm/pgalloc.h>
5721 #include <asm/tlbflush.h>
5722
5723+void arch_exit_mmap(struct mm_struct *mm);
5724+void arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm);
5725+
5726+void mm_pin(struct mm_struct *mm);
5727+void mm_unpin(struct mm_struct *mm);
5728+void mm_pin_all(void);
5729+
5730+static inline void xen_activate_mm(struct mm_struct *prev,
5731+ struct mm_struct *next)
5732+{
5733+ if (!test_bit(PG_pinned, &virt_to_page(next->pgd)->flags))
5734+ mm_pin(next);
5735+}
5736+
5737 /*
5738 * Used for LDT copy/destruction.
5739 */
5740@@ -37,10 +51,6 @@ static inline void __prepare_arch_switch
5741 : : "r" (0) );
5742 }
5743
5744-extern void mm_pin(struct mm_struct *mm);
5745-extern void mm_unpin(struct mm_struct *mm);
5746-void mm_pin_all(void);
5747-
5748 static inline void switch_mm(struct mm_struct *prev,
5749 struct mm_struct *next,
5750 struct task_struct *tsk)
5751@@ -97,11 +107,10 @@ static inline void switch_mm(struct mm_s
5752 #define deactivate_mm(tsk, mm) \
5753 asm("movl %0,%%gs": :"r" (0));
5754
5755-static inline void activate_mm(struct mm_struct *prev, struct mm_struct *next)
5756-{
5757- if (!test_bit(PG_pinned, &virt_to_page(next->pgd)->flags))
5758- mm_pin(next);
5759- switch_mm(prev, next, NULL);
5760-}
5761+#define activate_mm(prev, next) \
5762+ do { \
5763+ xen_activate_mm(prev, next); \
5764+ switch_mm((prev),(next),NULL); \
5765+ } while(0)
5766
5767 #endif
5768--- a/include/asm-x86/mach-xen/asm/mmu_context_64.h
5769+++ b/include/asm-x86/mach-xen/asm/mmu_context_64.h
5770@@ -9,6 +9,9 @@
5771 #include <asm/pgtable.h>
5772 #include <asm/tlbflush.h>
5773
5774+void arch_exit_mmap(struct mm_struct *mm);
5775+void arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm);
5776+
5777 /*
5778 * possibly do the LDT unload here?
5779 */
5780--- a/include/asm-x86/mach-xen/asm/page_64.h
5781+++ b/include/asm-x86/mach-xen/asm/page_64.h
5782@@ -7,6 +7,7 @@
5783 #include <linux/types.h>
5784 #include <asm/bug.h>
5785 #endif
5786+#include <linux/const.h>
5787 #include <xen/interface/xen.h>
5788
5789 /*
5790@@ -19,18 +20,14 @@
5791
5792 /* PAGE_SHIFT determines the page size */
5793 #define PAGE_SHIFT 12
5794-#ifdef __ASSEMBLY__
5795-#define PAGE_SIZE (0x1 << PAGE_SHIFT)
5796-#else
5797-#define PAGE_SIZE (1UL << PAGE_SHIFT)
5798-#endif
5799+#define PAGE_SIZE (_AC(1,UL) << PAGE_SHIFT)
5800 #define PAGE_MASK (~(PAGE_SIZE-1))
5801
5802 /* See Documentation/x86_64/mm.txt for a description of the memory map. */
5803 #define __PHYSICAL_MASK_SHIFT 46
5804-#define __PHYSICAL_MASK ((1UL << __PHYSICAL_MASK_SHIFT) - 1)
5805+#define __PHYSICAL_MASK ((_AC(1,UL) << __PHYSICAL_MASK_SHIFT) - 1)
5806 #define __VIRTUAL_MASK_SHIFT 48
5807-#define __VIRTUAL_MASK ((1UL << __VIRTUAL_MASK_SHIFT) - 1)
5808+#define __VIRTUAL_MASK ((_AC(1,UL) << __VIRTUAL_MASK_SHIFT) - 1)
5809
5810 #define PHYSICAL_PAGE_MASK (~(PAGE_SIZE-1) & __PHYSICAL_MASK)
5811
5812@@ -55,10 +52,10 @@
5813 #define N_EXCEPTION_STACKS 5 /* hw limit: 7 */
5814
5815 #define LARGE_PAGE_MASK (~(LARGE_PAGE_SIZE-1))
5816-#define LARGE_PAGE_SIZE (1UL << PMD_SHIFT)
5817+#define LARGE_PAGE_SIZE (_AC(1,UL) << PMD_SHIFT)
5818
5819 #define HPAGE_SHIFT PMD_SHIFT
5820-#define HPAGE_SIZE ((1UL) << HPAGE_SHIFT)
5821+#define HPAGE_SIZE (_AC(1,UL) << HPAGE_SHIFT)
5822 #define HPAGE_MASK (~(HPAGE_SIZE - 1))
5823 #define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
5824
5825@@ -152,17 +149,23 @@ static inline pgd_t __pgd(unsigned long
5826
5827 #define __pgprot(x) ((pgprot_t) { (x) } )
5828
5829-#define __PHYSICAL_START ((unsigned long)CONFIG_PHYSICAL_START)
5830-#define __START_KERNEL (__START_KERNEL_map + __PHYSICAL_START)
5831-#define __START_KERNEL_map 0xffffffff80000000UL
5832-#define __PAGE_OFFSET 0xffff880000000000UL
5833+#endif /* !__ASSEMBLY__ */
5834
5835-#else
5836 #define __PHYSICAL_START CONFIG_PHYSICAL_START
5837+#define __KERNEL_ALIGN 0x200000
5838+
5839+/*
5840+ * Make sure kernel is aligned to 2MB address. Catching it at compile
5841+ * time is better. Change your config file and compile the kernel
5842+ * for a 2MB aligned address (CONFIG_PHYSICAL_START)
5843+ */
5844+#if (CONFIG_PHYSICAL_START % __KERNEL_ALIGN) != 0
5845+#error "CONFIG_PHYSICAL_START must be a multiple of 2MB"
5846+#endif
5847+
5848 #define __START_KERNEL (__START_KERNEL_map + __PHYSICAL_START)
5849-#define __START_KERNEL_map 0xffffffff80000000
5850-#define __PAGE_OFFSET 0xffff880000000000
5851-#endif /* !__ASSEMBLY__ */
5852+#define __START_KERNEL_map _AC(0xffffffff80000000, UL)
5853+#define __PAGE_OFFSET _AC(0xffff880000000000, UL)
5854
5855 #if CONFIG_XEN_COMPAT <= 0x030002
5856 #undef LOAD_OFFSET
5857@@ -172,20 +175,20 @@ static inline pgd_t __pgd(unsigned long
5858 /* to align the pointer to the (next) page boundary */
5859 #define PAGE_ALIGN(addr) (((addr)+PAGE_SIZE-1)&PAGE_MASK)
5860
5861-#define KERNEL_TEXT_SIZE (40UL*1024*1024)
5862-#define KERNEL_TEXT_START 0xffffffff80000000UL
5863+#define KERNEL_TEXT_SIZE (40*1024*1024)
5864+#define KERNEL_TEXT_START _AC(0xffffffff80000000, UL)
5865+
5866+#define PAGE_OFFSET __PAGE_OFFSET
5867
5868-#define PAGE_OFFSET ((unsigned long)__PAGE_OFFSET)
5869+#ifndef __ASSEMBLY__
5870+static inline unsigned long __phys_addr(unsigned long x)
5871+{
5872+ return x - (x >= __START_KERNEL_map ? __START_KERNEL_map : PAGE_OFFSET);
5873+}
5874+#endif
5875
5876-/* Note: __pa(&symbol_visible_to_c) should be always replaced with __pa_symbol.
5877- Otherwise you risk miscompilation. */
5878-#define __pa(x) (((unsigned long)(x)>=__START_KERNEL_map)?(unsigned long)(x) - (unsigned long)__START_KERNEL_map:(unsigned long)(x) - PAGE_OFFSET)
5879-/* __pa_symbol should be used for C visible symbols.
5880- This seems to be the official gcc blessed way to do such arithmetic. */
5881-#define __pa_symbol(x) \
5882- ({unsigned long v; \
5883- asm("" : "=r" (v) : "0" (x)); \
5884- __pa(v); })
5885+#define __pa(x) __phys_addr((unsigned long)(x))
5886+#define __pa_symbol(x) __phys_addr((unsigned long)(x))
5887
5888 #define __va(x) ((void *)((unsigned long)(x)+PAGE_OFFSET))
5889 #define __boot_va(x) __va(x)
5890--- a/include/asm-x86/mach-xen/asm/pgalloc_32.h
5891+++ b/include/asm-x86/mach-xen/asm/pgalloc_32.h
5892@@ -1,7 +1,6 @@
5893 #ifndef _I386_PGALLOC_H
5894 #define _I386_PGALLOC_H
5895
5896-#include <asm/fixmap.h>
5897 #include <linux/threads.h>
5898 #include <linux/mm.h> /* for struct page */
5899 #include <asm/io.h> /* for phys_to_virt and page_to_pseudophys */
5900@@ -69,6 +68,4 @@ do { \
5901 #define pud_populate(mm, pmd, pte) BUG()
5902 #endif
5903
5904-#define check_pgt_cache() do { } while (0)
5905-
5906 #endif /* _I386_PGALLOC_H */
5907--- a/include/asm-x86/mach-xen/asm/pgalloc_64.h
5908+++ b/include/asm-x86/mach-xen/asm/pgalloc_64.h
5909@@ -1,7 +1,6 @@
5910 #ifndef _X86_64_PGALLOC_H
5911 #define _X86_64_PGALLOC_H
5912
5913-#include <asm/fixmap.h>
5914 #include <asm/pda.h>
5915 #include <linux/threads.h>
5916 #include <linux/mm.h>
5917@@ -100,24 +99,16 @@ static inline void pgd_list_add(pgd_t *p
5918 struct page *page = virt_to_page(pgd);
5919
5920 spin_lock(&pgd_lock);
5921- page->index = (pgoff_t)pgd_list;
5922- if (pgd_list)
5923- pgd_list->private = (unsigned long)&page->index;
5924- pgd_list = page;
5925- page->private = (unsigned long)&pgd_list;
5926+ list_add(&page->lru, &pgd_list);
5927 spin_unlock(&pgd_lock);
5928 }
5929
5930 static inline void pgd_list_del(pgd_t *pgd)
5931 {
5932- struct page *next, **pprev, *page = virt_to_page(pgd);
5933+ struct page *page = virt_to_page(pgd);
5934
5935 spin_lock(&pgd_lock);
5936- next = (struct page *)page->index;
5937- pprev = (struct page **)page->private;
5938- *pprev = next;
5939- if (next)
5940- next->private = (unsigned long)pprev;
5941+ list_del(&page->lru);
5942 spin_unlock(&pgd_lock);
5943 }
5944
5945--- a/include/asm-x86/mach-xen/asm/pgtable_32.h
5946+++ b/include/asm-x86/mach-xen/asm/pgtable_32.h
5947@@ -24,11 +24,11 @@
5948 #include <linux/slab.h>
5949 #include <linux/list.h>
5950 #include <linux/spinlock.h>
5951+#include <linux/sched.h>
5952
5953 /* Is this pagetable pinned? */
5954 #define PG_pinned PG_arch_1
5955
5956-struct mm_struct;
5957 struct vm_area_struct;
5958
5959 /*
5960@@ -38,17 +38,16 @@ struct vm_area_struct;
5961 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
5962 extern unsigned long empty_zero_page[1024];
5963 extern pgd_t *swapper_pg_dir;
5964-extern struct kmem_cache *pgd_cache;
5965 extern struct kmem_cache *pmd_cache;
5966 extern spinlock_t pgd_lock;
5967 extern struct page *pgd_list;
5968+void check_pgt_cache(void);
5969
5970 void pmd_ctor(void *, struct kmem_cache *, unsigned long);
5971-void pgd_ctor(void *, struct kmem_cache *, unsigned long);
5972-void pgd_dtor(void *, struct kmem_cache *, unsigned long);
5973 void pgtable_cache_init(void);
5974 void paging_init(void);
5975
5976+
5977 /*
5978 * The Linux x86 paging architecture is 'compile-time dual-mode', it
5979 * implements both the traditional 2-level x86 page tables and the
5980@@ -165,6 +164,7 @@ void paging_init(void);
5981
5982 extern unsigned long long __PAGE_KERNEL, __PAGE_KERNEL_EXEC;
5983 #define __PAGE_KERNEL_RO (__PAGE_KERNEL & ~_PAGE_RW)
5984+#define __PAGE_KERNEL_RX (__PAGE_KERNEL_EXEC & ~_PAGE_RW)
5985 #define __PAGE_KERNEL_NOCACHE (__PAGE_KERNEL | _PAGE_PCD)
5986 #define __PAGE_KERNEL_LARGE (__PAGE_KERNEL | _PAGE_PSE)
5987 #define __PAGE_KERNEL_LARGE_EXEC (__PAGE_KERNEL_EXEC | _PAGE_PSE)
5988@@ -172,6 +172,7 @@ extern unsigned long long __PAGE_KERNEL,
5989 #define PAGE_KERNEL __pgprot(__PAGE_KERNEL)
5990 #define PAGE_KERNEL_RO __pgprot(__PAGE_KERNEL_RO)
5991 #define PAGE_KERNEL_EXEC __pgprot(__PAGE_KERNEL_EXEC)
5992+#define PAGE_KERNEL_RX __pgprot(__PAGE_KERNEL_RX)
5993 #define PAGE_KERNEL_NOCACHE __pgprot(__PAGE_KERNEL_NOCACHE)
5994 #define PAGE_KERNEL_LARGE __pgprot(__PAGE_KERNEL_LARGE)
5995 #define PAGE_KERNEL_LARGE_EXEC __pgprot(__PAGE_KERNEL_LARGE_EXEC)
5996@@ -275,7 +276,13 @@ static inline pte_t pte_mkhuge(pte_t pte
5997 */
5998 #define pte_update(mm, addr, ptep) do { } while (0)
5999 #define pte_update_defer(mm, addr, ptep) do { } while (0)
6000-#define paravirt_map_pt_hook(slot, va, pfn) do { } while (0)
6001+
6002+/* local pte updates need not use xchg for locking */
6003+static inline pte_t xen_local_ptep_get_and_clear(pte_t *ptep, pte_t res)
6004+{
6005+ xen_set_pte(ptep, __pte(0));
6006+ return res;
6007+}
6008
6009 /*
6010 * We only update the dirty/accessed state if we set
6011@@ -286,17 +293,34 @@ static inline pte_t pte_mkhuge(pte_t pte
6012 */
6013 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
6014 #define ptep_set_access_flags(vma, address, ptep, entry, dirty) \
6015-do { \
6016- if (dirty) \
6017+({ \
6018+ int __changed = !pte_same(*(ptep), entry); \
6019+ if (__changed && (dirty)) \
6020 ptep_establish(vma, address, ptep, entry); \
6021-} while (0)
6022+ __changed; \
6023+})
6024
6025-/*
6026- * We don't actually have these, but we want to advertise them so that
6027- * we can encompass the flush here.
6028- */
6029 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
6030+#define ptep_test_and_clear_dirty(vma, addr, ptep) ({ \
6031+ int __ret = 0; \
6032+ if (pte_dirty(*(ptep))) \
6033+ __ret = test_and_clear_bit(_PAGE_BIT_DIRTY, \
6034+ &(ptep)->pte_low); \
6035+ if (__ret) \
6036+ pte_update((vma)->vm_mm, addr, ptep); \
6037+ __ret; \
6038+})
6039+
6040 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
6041+#define ptep_test_and_clear_young(vma, addr, ptep) ({ \
6042+ int __ret = 0; \
6043+ if (pte_young(*(ptep))) \
6044+ __ret = test_and_clear_bit(_PAGE_BIT_ACCESSED, \
6045+ &(ptep)->pte_low); \
6046+ if (__ret) \
6047+ pte_update((vma)->vm_mm, addr, ptep); \
6048+ __ret; \
6049+})
6050
6051 /*
6052 * Rules for using ptep_establish: the pte MUST be a user pte, and
6053@@ -323,7 +347,7 @@ do { \
6054 int __dirty = pte_dirty(__pte); \
6055 __pte = pte_mkclean(__pte); \
6056 if (test_bit(PG_pinned, &virt_to_page((vma)->vm_mm->pgd)->flags)) \
6057- ptep_set_access_flags(vma, address, ptep, __pte, __dirty); \
6058+ (void)ptep_set_access_flags(vma, address, ptep, __pte, __dirty); \
6059 else if (__dirty) \
6060 (ptep)->pte_low = __pte.pte_low; \
6061 __dirty; \
6062@@ -336,7 +360,7 @@ do { \
6063 int __young = pte_young(__pte); \
6064 __pte = pte_mkold(__pte); \
6065 if (test_bit(PG_pinned, &virt_to_page((vma)->vm_mm->pgd)->flags)) \
6066- ptep_set_access_flags(vma, address, ptep, __pte, __young); \
6067+ (void)ptep_set_access_flags(vma, address, ptep, __pte, __young); \
6068 else if (__young) \
6069 (ptep)->pte_low = __pte.pte_low; \
6070 __young; \
6071@@ -349,7 +373,7 @@ static inline pte_t ptep_get_and_clear(s
6072 if (!pte_none(pte)
6073 && (mm != &init_mm
6074 || HYPERVISOR_update_va_mapping(addr, __pte(0), 0))) {
6075- pte = raw_ptep_get_and_clear(ptep, pte);
6076+ pte = xen_ptep_get_and_clear(ptep, pte);
6077 pte_update(mm, addr, ptep);
6078 }
6079 return pte;
6080@@ -491,24 +515,10 @@ extern pte_t *lookup_address(unsigned lo
6081 #endif
6082
6083 #if defined(CONFIG_HIGHPTE)
6084-#define pte_offset_map(dir, address) \
6085-({ \
6086- pte_t *__ptep; \
6087- unsigned pfn = pmd_val(*(dir)) >> PAGE_SHIFT; \
6088- __ptep = (pte_t *)kmap_atomic_pte(pfn_to_page(pfn),KM_PTE0); \
6089- paravirt_map_pt_hook(KM_PTE0,__ptep, pfn); \
6090- __ptep = __ptep + pte_index(address); \
6091- __ptep; \
6092-})
6093-#define pte_offset_map_nested(dir, address) \
6094-({ \
6095- pte_t *__ptep; \
6096- unsigned pfn = pmd_val(*(dir)) >> PAGE_SHIFT; \
6097- __ptep = (pte_t *)kmap_atomic_pte(pfn_to_page(pfn),KM_PTE1); \
6098- paravirt_map_pt_hook(KM_PTE1,__ptep, pfn); \
6099- __ptep = __ptep + pte_index(address); \
6100- __ptep; \
6101-})
6102+#define pte_offset_map(dir, address) \
6103+ ((pte_t *)kmap_atomic_pte(pmd_page(*(dir)),KM_PTE0) + pte_index(address))
6104+#define pte_offset_map_nested(dir, address) \
6105+ ((pte_t *)kmap_atomic_pte(pmd_page(*(dir)),KM_PTE1) + pte_index(address))
6106 #define pte_unmap(pte) kunmap_atomic(pte, KM_PTE0)
6107 #define pte_unmap_nested(pte) kunmap_atomic(pte, KM_PTE1)
6108 #else
6109@@ -584,10 +594,6 @@ int xen_change_pte_range(struct mm_struc
6110 #define io_remap_pfn_range(vma,from,pfn,size,prot) \
6111 direct_remap_pfn_range(vma,from,pfn,size,prot,DOMID_IO)
6112
6113-#define MK_IOSPACE_PFN(space, pfn) (pfn)
6114-#define GET_IOSPACE(pfn) 0
6115-#define GET_PFN(pfn) (pfn)
6116-
6117 #include <asm-generic/pgtable.h>
6118
6119 #endif /* _I386_PGTABLE_H */
6120--- a/include/asm-x86/mach-xen/asm/pgtable-3level-defs.h
6121+++ b/include/asm-x86/mach-xen/asm/pgtable-3level-defs.h
6122@@ -1,7 +1,7 @@
6123 #ifndef _I386_PGTABLE_3LEVEL_DEFS_H
6124 #define _I386_PGTABLE_3LEVEL_DEFS_H
6125
6126-#define HAVE_SHARED_KERNEL_PMD 0
6127+#define SHARED_KERNEL_PMD 0
6128
6129 /*
6130 * PGDIR_SHIFT determines what a top-level page table entry can map
6131--- a/include/asm-x86/mach-xen/asm/pgtable-3level.h
6132+++ b/include/asm-x86/mach-xen/asm/pgtable-3level.h
6133@@ -52,32 +52,40 @@ static inline int pte_exec_kernel(pte_t
6134 * value and then use set_pte to update it. -ben
6135 */
6136
6137-static inline void set_pte(pte_t *ptep, pte_t pte)
6138+static inline void xen_set_pte(pte_t *ptep, pte_t pte)
6139 {
6140 ptep->pte_high = pte.pte_high;
6141 smp_wmb();
6142 ptep->pte_low = pte.pte_low;
6143 }
6144-#define set_pte_atomic(pteptr,pteval) \
6145- set_64bit((unsigned long long *)(pteptr),__pte_val(pteval))
6146
6147-#define set_pte_at(_mm,addr,ptep,pteval) do { \
6148- if (((_mm) != current->mm && (_mm) != &init_mm) || \
6149- HYPERVISOR_update_va_mapping((addr), (pteval), 0)) \
6150- set_pte((ptep), (pteval)); \
6151-} while (0)
6152-
6153-#define set_pmd(pmdptr,pmdval) \
6154- xen_l2_entry_update((pmdptr), (pmdval))
6155-#define set_pud(pudptr,pudval) \
6156- xen_l3_entry_update((pudptr), (pudval))
6157+static inline void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
6158+ pte_t *ptep , pte_t pte)
6159+{
6160+ if ((mm != current->mm && mm != &init_mm) ||
6161+ HYPERVISOR_update_va_mapping(addr, pte, 0))
6162+ xen_set_pte(ptep, pte);
6163+}
6164+
6165+static inline void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
6166+{
6167+ set_64bit((unsigned long long *)(ptep),__pte_val(pte));
6168+}
6169+static inline void xen_set_pmd(pmd_t *pmdp, pmd_t pmd)
6170+{
6171+ xen_l2_entry_update(pmdp, pmd);
6172+}
6173+static inline void xen_set_pud(pud_t *pudp, pud_t pud)
6174+{
6175+ xen_l3_entry_update(pudp, pud);
6176+}
6177
6178 /*
6179 * For PTEs and PDEs, we must clear the P-bit first when clearing a page table
6180 * entry, so clear the bottom half first and enforce ordering with a compiler
6181 * barrier.
6182 */
6183-static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
6184+static inline void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
6185 {
6186 if ((mm != current->mm && mm != &init_mm)
6187 || HYPERVISOR_update_va_mapping(addr, __pte(0), 0)) {
6188@@ -87,7 +95,18 @@ static inline void pte_clear(struct mm_s
6189 }
6190 }
6191
6192-#define pmd_clear(xp) do { set_pmd(xp, __pmd(0)); } while (0)
6193+static inline void xen_pmd_clear(pmd_t *pmd)
6194+{
6195+ xen_l2_entry_update(pmd, __pmd(0));
6196+}
6197+
6198+#define set_pte(ptep, pte) xen_set_pte(ptep, pte)
6199+#define set_pte_at(mm, addr, ptep, pte) xen_set_pte_at(mm, addr, ptep, pte)
6200+#define set_pte_atomic(ptep, pte) xen_set_pte_atomic(ptep, pte)
6201+#define set_pmd(pmdp, pmd) xen_set_pmd(pmdp, pmd)
6202+#define set_pud(pudp, pud) xen_set_pud(pudp, pud)
6203+#define pte_clear(mm, addr, ptep) xen_pte_clear(mm, addr, ptep)
6204+#define pmd_clear(pmd) xen_pmd_clear(pmd)
6205
6206 /*
6207 * Pentium-II erratum A13: in PAE mode we explicitly have to flush
6208@@ -108,7 +127,8 @@ static inline void pud_clear (pud_t * pu
6209 #define pmd_offset(pud, address) ((pmd_t *) pud_page(*(pud)) + \
6210 pmd_index(address))
6211
6212-static inline pte_t raw_ptep_get_and_clear(pte_t *ptep, pte_t res)
6213+#ifdef CONFIG_SMP
6214+static inline pte_t xen_ptep_get_and_clear(pte_t *ptep, pte_t res)
6215 {
6216 uint64_t val = __pte_val(res);
6217 if (__cmpxchg64(ptep, val, 0) != val) {
6218@@ -119,6 +139,9 @@ static inline pte_t raw_ptep_get_and_cle
6219 }
6220 return res;
6221 }
6222+#else
6223+#define xen_ptep_get_and_clear(xp, pte) xen_local_ptep_get_and_clear(xp, pte)
6224+#endif
6225
6226 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
6227 #define ptep_clear_flush(vma, addr, ptep) \
6228@@ -165,13 +188,13 @@ extern unsigned long long __supported_pt
6229 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
6230 {
6231 return __pte((((unsigned long long)page_nr << PAGE_SHIFT) |
6232- pgprot_val(pgprot)) & __supported_pte_mask);
6233+ pgprot_val(pgprot)) & __supported_pte_mask);
6234 }
6235
6236 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
6237 {
6238 return __pmd((((unsigned long long)page_nr << PAGE_SHIFT) |
6239- pgprot_val(pgprot)) & __supported_pte_mask);
6240+ pgprot_val(pgprot)) & __supported_pte_mask);
6241 }
6242
6243 /*
6244@@ -191,6 +214,4 @@ static inline pmd_t pfn_pmd(unsigned lon
6245
6246 #define __pmd_free_tlb(tlb, x) do { } while (0)
6247
6248-void vmalloc_sync_all(void);
6249-
6250 #endif /* _I386_PGTABLE_3LEVEL_H */
6251--- a/include/asm-x86/mach-xen/asm/pgtable_64.h
6252+++ b/include/asm-x86/mach-xen/asm/pgtable_64.h
6253@@ -1,12 +1,14 @@
6254 #ifndef _X86_64_PGTABLE_H
6255 #define _X86_64_PGTABLE_H
6256
6257+#include <linux/const.h>
6258+#ifndef __ASSEMBLY__
6259+
6260 /*
6261 * This file contains the functions and defines necessary to modify and use
6262 * the x86-64 page table tree.
6263 */
6264 #include <asm/processor.h>
6265-#include <asm/fixmap.h>
6266 #include <asm/bitops.h>
6267 #include <linux/threads.h>
6268 #include <linux/sched.h>
6269@@ -33,11 +35,9 @@ extern pte_t *lookup_address(unsigned lo
6270 #endif
6271
6272 extern pud_t level3_kernel_pgt[512];
6273-extern pud_t level3_physmem_pgt[512];
6274 extern pud_t level3_ident_pgt[512];
6275 extern pmd_t level2_kernel_pgt[512];
6276 extern pgd_t init_level4_pgt[];
6277-extern pgd_t boot_level4_pgt[];
6278 extern unsigned long __supported_pte_mask;
6279
6280 #define swapper_pg_dir init_level4_pgt
6281@@ -52,6 +52,8 @@ extern void clear_kernel_mapping(unsigne
6282 extern unsigned long empty_zero_page[PAGE_SIZE/sizeof(unsigned long)];
6283 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
6284
6285+#endif /* !__ASSEMBLY__ */
6286+
6287 /*
6288 * PGDIR_SHIFT determines what a top-level page table entry can map
6289 */
6290@@ -76,6 +78,8 @@ extern unsigned long empty_zero_page[PAG
6291 */
6292 #define PTRS_PER_PTE 512
6293
6294+#ifndef __ASSEMBLY__
6295+
6296 #define pte_ERROR(e) \
6297 printk("%s:%d: bad pte %p(%016lx pfn %010lx).\n", __FILE__, __LINE__, \
6298 &(e), __pte_val(e), pte_pfn(e))
6299@@ -118,22 +122,23 @@ static inline void pgd_clear (pgd_t * pg
6300
6301 #define pte_pgprot(a) (__pgprot((a).pte & ~PHYSICAL_PAGE_MASK))
6302
6303-#define PMD_SIZE (1UL << PMD_SHIFT)
6304+#endif /* !__ASSEMBLY__ */
6305+
6306+#define PMD_SIZE (_AC(1,UL) << PMD_SHIFT)
6307 #define PMD_MASK (~(PMD_SIZE-1))
6308-#define PUD_SIZE (1UL << PUD_SHIFT)
6309+#define PUD_SIZE (_AC(1,UL) << PUD_SHIFT)
6310 #define PUD_MASK (~(PUD_SIZE-1))
6311-#define PGDIR_SIZE (1UL << PGDIR_SHIFT)
6312+#define PGDIR_SIZE (_AC(1,UL) << PGDIR_SHIFT)
6313 #define PGDIR_MASK (~(PGDIR_SIZE-1))
6314
6315 #define USER_PTRS_PER_PGD ((TASK_SIZE-1)/PGDIR_SIZE+1)
6316 #define FIRST_USER_ADDRESS 0
6317
6318-#ifndef __ASSEMBLY__
6319-#define MAXMEM 0x3fffffffffffUL
6320-#define VMALLOC_START 0xffffc20000000000UL
6321-#define VMALLOC_END 0xffffe1ffffffffffUL
6322-#define MODULES_VADDR 0xffffffff88000000UL
6323-#define MODULES_END 0xfffffffffff00000UL
6324+#define MAXMEM _AC(0x3fffffffffff, UL)
6325+#define VMALLOC_START _AC(0xffffc20000000000, UL)
6326+#define VMALLOC_END _AC(0xffffe1ffffffffff, UL)
6327+#define MODULES_VADDR _AC(0xffffffff88000000, UL)
6328+#define MODULES_END _AC(0xfffffffffff00000, UL)
6329 #define MODULES_LEN (MODULES_END - MODULES_VADDR)
6330
6331 #define _PAGE_BIT_PRESENT 0
6332@@ -159,16 +164,18 @@ static inline void pgd_clear (pgd_t * pg
6333 #define _PAGE_GLOBAL 0x100 /* Global TLB entry */
6334
6335 #define _PAGE_PROTNONE 0x080 /* If not present */
6336-#define _PAGE_NX (1UL<<_PAGE_BIT_NX)
6337+#define _PAGE_NX (_AC(1,UL)<<_PAGE_BIT_NX)
6338
6339 /* Mapped page is I/O or foreign and has no associated page struct. */
6340 #define _PAGE_IO 0x200
6341
6342+#ifndef __ASSEMBLY__
6343 #if CONFIG_XEN_COMPAT <= 0x030002
6344 extern unsigned int __kernel_page_user;
6345 #else
6346 #define __kernel_page_user 0
6347 #endif
6348+#endif
6349
6350 #define _PAGE_TABLE (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | _PAGE_ACCESSED | _PAGE_DIRTY)
6351 #define _KERNPG_TABLE (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY | __kernel_page_user)
6352@@ -233,6 +240,8 @@ extern unsigned int __kernel_page_user;
6353 #define __S110 PAGE_SHARED_EXEC
6354 #define __S111 PAGE_SHARED_EXEC
6355
6356+#ifndef __ASSEMBLY__
6357+
6358 static inline unsigned long pgd_bad(pgd_t pgd)
6359 {
6360 return __pgd_val(pgd) & ~(PTE_MASK | _KERNPG_TABLE | _PAGE_USER);
6361@@ -344,6 +353,20 @@ static inline pte_t pte_mkwrite(pte_t pt
6362 static inline pte_t pte_mkhuge(pte_t pte) { __pte_val(pte) |= _PAGE_PSE; return pte; }
6363 static inline pte_t pte_clrhuge(pte_t pte) { __pte_val(pte) &= ~_PAGE_PSE; return pte; }
6364
6365+static inline int ptep_test_and_clear_dirty(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
6366+{
6367+ if (!pte_dirty(*ptep))
6368+ return 0;
6369+ return test_and_clear_bit(_PAGE_BIT_DIRTY, &ptep->pte);
6370+}
6371+
6372+static inline int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
6373+{
6374+ if (!pte_young(*ptep))
6375+ return 0;
6376+ return test_and_clear_bit(_PAGE_BIT_ACCESSED, &ptep->pte);
6377+}
6378+
6379 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
6380 {
6381 pte_t pte = *ptep;
6382@@ -468,18 +491,12 @@ static inline pte_t pte_modify(pte_t pte
6383 * bit at the same time. */
6384 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
6385 #define ptep_set_access_flags(vma, address, ptep, entry, dirty) \
6386- do { \
6387- if (dirty) \
6388- ptep_establish(vma, address, ptep, entry); \
6389- } while (0)
6390-
6391-
6392-/*
6393- * i386 says: We don't actually have these, but we want to advertise
6394- * them so that we can encompass the flush here.
6395- */
6396-#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
6397-#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
6398+({ \
6399+ int __changed = !pte_same(*(ptep), entry); \
6400+ if (__changed && (dirty)) \
6401+ ptep_establish(vma, address, ptep, entry); \
6402+ __changed; \
6403+})
6404
6405 #define __HAVE_ARCH_PTEP_CLEAR_DIRTY_FLUSH
6406 #define ptep_clear_flush_dirty(vma, address, ptep) \
6407@@ -488,7 +505,7 @@ static inline pte_t pte_modify(pte_t pte
6408 int __dirty = pte_dirty(__pte); \
6409 __pte = pte_mkclean(__pte); \
6410 if ((vma)->vm_mm->context.pinned) \
6411- ptep_set_access_flags(vma, address, ptep, __pte, __dirty); \
6412+ (void)ptep_set_access_flags(vma, address, ptep, __pte, __dirty); \
6413 else if (__dirty) \
6414 set_pte(ptep, __pte); \
6415 __dirty; \
6416@@ -501,7 +518,7 @@ static inline pte_t pte_modify(pte_t pte
6417 int __young = pte_young(__pte); \
6418 __pte = pte_mkold(__pte); \
6419 if ((vma)->vm_mm->context.pinned) \
6420- ptep_set_access_flags(vma, address, ptep, __pte, __young); \
6421+ (void)ptep_set_access_flags(vma, address, ptep, __pte, __young); \
6422 else if (__young) \
6423 set_pte(ptep, __pte); \
6424 __young; \
6425@@ -515,10 +532,7 @@ static inline pte_t pte_modify(pte_t pte
6426 #define __swp_entry_to_pte(x) ((pte_t) { (x).val })
6427
6428 extern spinlock_t pgd_lock;
6429-extern struct page *pgd_list;
6430-void vmalloc_sync_all(void);
6431-
6432-#endif /* !__ASSEMBLY__ */
6433+extern struct list_head pgd_list;
6434
6435 extern int kern_addr_valid(unsigned long addr);
6436
6437@@ -557,10 +571,6 @@ int xen_change_pte_range(struct mm_struc
6438 #define io_remap_pfn_range(vma, vaddr, pfn, size, prot) \
6439 direct_remap_pfn_range(vma,vaddr,pfn,size,prot,DOMID_IO)
6440
6441-#define MK_IOSPACE_PFN(space, pfn) (pfn)
6442-#define GET_IOSPACE(pfn) 0
6443-#define GET_PFN(pfn) (pfn)
6444-
6445 #define HAVE_ARCH_UNMAPPED_AREA
6446
6447 #define pgtable_cache_init() do { } while (0)
6448@@ -574,11 +584,14 @@ int xen_change_pte_range(struct mm_struc
6449 #define kc_offset_to_vaddr(o) \
6450 (((o) & (1UL << (__VIRTUAL_MASK_SHIFT-1))) ? ((o) | (~__VIRTUAL_MASK)) : (o))
6451
6452+#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
6453+#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
6454 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
6455 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
6456 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
6457 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
6458 #define __HAVE_ARCH_PTE_SAME
6459 #include <asm-generic/pgtable.h>
6460+#endif /* !__ASSEMBLY__ */
6461
6462 #endif /* _X86_64_PGTABLE_H */
6463--- a/include/asm-x86/mach-xen/asm/processor_32.h
6464+++ b/include/asm-x86/mach-xen/asm/processor_32.h
6465@@ -21,6 +21,7 @@
6466 #include <asm/percpu.h>
6467 #include <linux/cpumask.h>
6468 #include <linux/init.h>
6469+#include <asm/processor-flags.h>
6470 #include <xen/interface/physdev.h>
6471
6472 /* flag for disabling the tsc */
6473@@ -118,7 +119,8 @@ extern char ignore_fpu_irq;
6474
6475 void __init cpu_detect(struct cpuinfo_x86 *c);
6476
6477-extern void identify_cpu(struct cpuinfo_x86 *);
6478+extern void identify_boot_cpu(void);
6479+extern void identify_secondary_cpu(struct cpuinfo_x86 *);
6480 extern void print_cpu_info(struct cpuinfo_x86 *);
6481 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
6482 extern unsigned short num_cache_leaves;
6483@@ -129,29 +131,8 @@ extern void detect_ht(struct cpuinfo_x86
6484 static inline void detect_ht(struct cpuinfo_x86 *c) {}
6485 #endif
6486
6487-/*
6488- * EFLAGS bits
6489- */
6490-#define X86_EFLAGS_CF 0x00000001 /* Carry Flag */
6491-#define X86_EFLAGS_PF 0x00000004 /* Parity Flag */
6492-#define X86_EFLAGS_AF 0x00000010 /* Auxillary carry Flag */
6493-#define X86_EFLAGS_ZF 0x00000040 /* Zero Flag */
6494-#define X86_EFLAGS_SF 0x00000080 /* Sign Flag */
6495-#define X86_EFLAGS_TF 0x00000100 /* Trap Flag */
6496-#define X86_EFLAGS_IF 0x00000200 /* Interrupt Flag */
6497-#define X86_EFLAGS_DF 0x00000400 /* Direction Flag */
6498-#define X86_EFLAGS_OF 0x00000800 /* Overflow Flag */
6499-#define X86_EFLAGS_IOPL 0x00003000 /* IOPL mask */
6500-#define X86_EFLAGS_NT 0x00004000 /* Nested Task */
6501-#define X86_EFLAGS_RF 0x00010000 /* Resume Flag */
6502-#define X86_EFLAGS_VM 0x00020000 /* Virtual Mode */
6503-#define X86_EFLAGS_AC 0x00040000 /* Alignment Check */
6504-#define X86_EFLAGS_VIF 0x00080000 /* Virtual Interrupt Flag */
6505-#define X86_EFLAGS_VIP 0x00100000 /* Virtual Interrupt Pending */
6506-#define X86_EFLAGS_ID 0x00200000 /* CPUID detection flag */
6507-
6508-static inline fastcall void xen_cpuid(unsigned int *eax, unsigned int *ebx,
6509- unsigned int *ecx, unsigned int *edx)
6510+static inline void xen_cpuid(unsigned int *eax, unsigned int *ebx,
6511+ unsigned int *ecx, unsigned int *edx)
6512 {
6513 /* ecx is often an input as well as an output. */
6514 __asm__(XEN_CPUID
6515@@ -165,21 +146,6 @@ static inline fastcall void xen_cpuid(un
6516 #define load_cr3(pgdir) write_cr3(__pa(pgdir))
6517
6518 /*
6519- * Intel CPU features in CR4
6520- */
6521-#define X86_CR4_VME 0x0001 /* enable vm86 extensions */
6522-#define X86_CR4_PVI 0x0002 /* virtual interrupts flag enable */
6523-#define X86_CR4_TSD 0x0004 /* disable time stamp at ipl 3 */
6524-#define X86_CR4_DE 0x0008 /* enable debugging extensions */
6525-#define X86_CR4_PSE 0x0010 /* enable page size extensions */
6526-#define X86_CR4_PAE 0x0020 /* enable physical address extensions */
6527-#define X86_CR4_MCE 0x0040 /* Machine check enable */
6528-#define X86_CR4_PGE 0x0080 /* enable global pages */
6529-#define X86_CR4_PCE 0x0100 /* enable performance counters at ipl 3 */
6530-#define X86_CR4_OSFXSR 0x0200 /* enable fast FPU save and restore */
6531-#define X86_CR4_OSXMMEXCPT 0x0400 /* enable unmasked SSE exceptions */
6532-
6533-/*
6534 * Save the cr4 feature set we're using (ie
6535 * Pentium 4MB enable and PPro Global page
6536 * enable), so that any CPU's that boot up
6537@@ -206,26 +172,6 @@ static inline void clear_in_cr4 (unsigne
6538 }
6539
6540 /*
6541- * NSC/Cyrix CPU configuration register indexes
6542- */
6543-
6544-#define CX86_PCR0 0x20
6545-#define CX86_GCR 0xb8
6546-#define CX86_CCR0 0xc0
6547-#define CX86_CCR1 0xc1
6548-#define CX86_CCR2 0xc2
6549-#define CX86_CCR3 0xc3
6550-#define CX86_CCR4 0xe8
6551-#define CX86_CCR5 0xe9
6552-#define CX86_CCR6 0xea
6553-#define CX86_CCR7 0xeb
6554-#define CX86_PCR1 0xf0
6555-#define CX86_DIR0 0xfe
6556-#define CX86_DIR1 0xff
6557-#define CX86_ARR_BASE 0xc4
6558-#define CX86_RCR_BASE 0xdc
6559-
6560-/*
6561 * NSC/Cyrix CPU indexed register access macros
6562 */
6563
6564@@ -351,7 +297,8 @@ typedef struct {
6565 struct thread_struct;
6566
6567 #ifndef CONFIG_X86_NO_TSS
6568-struct tss_struct {
6569+/* This is the TSS defined by the hardware. */
6570+struct i386_hw_tss {
6571 unsigned short back_link,__blh;
6572 unsigned long esp0;
6573 unsigned short ss0,__ss0h;
6574@@ -375,6 +322,11 @@ struct tss_struct {
6575 unsigned short gs, __gsh;
6576 unsigned short ldt, __ldth;
6577 unsigned short trace, io_bitmap_base;
6578+} __attribute__((packed));
6579+
6580+struct tss_struct {
6581+ struct i386_hw_tss x86_tss;
6582+
6583 /*
6584 * The extra 1 is there because the CPU will access an
6585 * additional byte beyond the end of the IO permission
6586@@ -428,10 +380,11 @@ struct thread_struct {
6587 };
6588
6589 #define INIT_THREAD { \
6590+ .esp0 = sizeof(init_stack) + (long)&init_stack, \
6591 .vm86_info = NULL, \
6592 .sysenter_cs = __KERNEL_CS, \
6593 .io_bitmap_ptr = NULL, \
6594- .fs = __KERNEL_PDA, \
6595+ .fs = __KERNEL_PERCPU, \
6596 }
6597
6598 /*
6599@@ -441,10 +394,12 @@ struct thread_struct {
6600 * be within the limit.
6601 */
6602 #define INIT_TSS { \
6603- .esp0 = sizeof(init_stack) + (long)&init_stack, \
6604- .ss0 = __KERNEL_DS, \
6605- .ss1 = __KERNEL_CS, \
6606- .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, \
6607+ .x86_tss = { \
6608+ .esp0 = sizeof(init_stack) + (long)&init_stack, \
6609+ .ss0 = __KERNEL_DS, \
6610+ .ss1 = __KERNEL_CS, \
6611+ .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, \
6612+ }, \
6613 .io_bitmap = { [ 0 ... IO_BITMAP_LONGS] = ~0 }, \
6614 }
6615
6616@@ -551,38 +506,33 @@ static inline void rep_nop(void)
6617
6618 #define cpu_relax() rep_nop()
6619
6620-#define paravirt_enabled() 0
6621-#define __cpuid xen_cpuid
6622-
6623 #ifndef CONFIG_X86_NO_TSS
6624-static inline void __load_esp0(struct tss_struct *tss, struct thread_struct *thread)
6625+static inline void native_load_esp0(struct tss_struct *tss, struct thread_struct *thread)
6626 {
6627- tss->esp0 = thread->esp0;
6628+ tss->x86_tss.esp0 = thread->esp0;
6629 /* This can only happen when SEP is enabled, no need to test "SEP"arately */
6630- if (unlikely(tss->ss1 != thread->sysenter_cs)) {
6631- tss->ss1 = thread->sysenter_cs;
6632+ if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
6633+ tss->x86_tss.ss1 = thread->sysenter_cs;
6634 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
6635 }
6636 }
6637-#define load_esp0(tss, thread) \
6638- __load_esp0(tss, thread)
6639 #else
6640-#define load_esp0(tss, thread) do { \
6641+#define xen_load_esp0(tss, thread) do { \
6642 if (HYPERVISOR_stack_switch(__KERNEL_DS, (thread)->esp0)) \
6643 BUG(); \
6644 } while (0)
6645 #endif
6646
6647
6648-/*
6649- * These special macros can be used to get or set a debugging register
6650- */
6651-#define get_debugreg(var, register) \
6652- (var) = HYPERVISOR_get_debugreg(register)
6653-#define set_debugreg(value, register) \
6654- WARN_ON(HYPERVISOR_set_debugreg(register, value))
6655+static inline unsigned long xen_get_debugreg(int regno)
6656+{
6657+ return HYPERVISOR_get_debugreg(regno);
6658+}
6659
6660-#define set_iopl_mask xen_set_iopl_mask
6661+static inline void xen_set_debugreg(int regno, unsigned long value)
6662+{
6663+ WARN_ON(HYPERVISOR_set_debugreg(regno, value));
6664+}
6665
6666 /*
6667 * Set IOPL bits in EFLAGS from given mask
6668@@ -597,6 +547,21 @@ static inline void xen_set_iopl_mask(uns
6669 }
6670
6671
6672+#define paravirt_enabled() 0
6673+#define __cpuid xen_cpuid
6674+
6675+#define load_esp0 xen_load_esp0
6676+
6677+/*
6678+ * These special macros can be used to get or set a debugging register
6679+ */
6680+#define get_debugreg(var, register) \
6681+ (var) = xen_get_debugreg(register)
6682+#define set_debugreg(value, register) \
6683+ xen_set_debugreg(register, value)
6684+
6685+#define set_iopl_mask xen_set_iopl_mask
6686+
6687 /*
6688 * Generic CPUID function
6689 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
6690@@ -749,8 +714,14 @@ extern unsigned long boot_option_idle_ov
6691 extern void enable_sep_cpu(void);
6692 extern int sysenter_setup(void);
6693
6694-extern int init_gdt(int cpu, struct task_struct *idle);
6695+/* Defined in head.S */
6696+extern struct Xgt_desc_struct early_gdt_descr;
6697+
6698 extern void cpu_set_gdt(int);
6699-extern void secondary_cpu_init(void);
6700+extern void switch_to_new_gdt(void);
6701+extern void cpu_init(void);
6702+extern void init_gdt(int cpu);
6703+
6704+extern int force_mwait;
6705
6706 #endif /* __ASM_I386_PROCESSOR_H */
6707--- a/include/asm-x86/mach-xen/asm/processor_64.h
6708+++ b/include/asm-x86/mach-xen/asm/processor_64.h
6709@@ -20,6 +20,7 @@
6710 #include <asm/percpu.h>
6711 #include <linux/personality.h>
6712 #include <linux/cpumask.h>
6713+#include <asm/processor-flags.h>
6714
6715 #define TF_MASK 0x00000100
6716 #define IF_MASK 0x00000200
6717@@ -103,42 +104,6 @@ extern unsigned int init_intel_cacheinfo
6718 extern unsigned short num_cache_leaves;
6719
6720 /*
6721- * EFLAGS bits
6722- */
6723-#define X86_EFLAGS_CF 0x00000001 /* Carry Flag */
6724-#define X86_EFLAGS_PF 0x00000004 /* Parity Flag */
6725-#define X86_EFLAGS_AF 0x00000010 /* Auxillary carry Flag */
6726-#define X86_EFLAGS_ZF 0x00000040 /* Zero Flag */
6727-#define X86_EFLAGS_SF 0x00000080 /* Sign Flag */
6728-#define X86_EFLAGS_TF 0x00000100 /* Trap Flag */
6729-#define X86_EFLAGS_IF 0x00000200 /* Interrupt Flag */
6730-#define X86_EFLAGS_DF 0x00000400 /* Direction Flag */
6731-#define X86_EFLAGS_OF 0x00000800 /* Overflow Flag */
6732-#define X86_EFLAGS_IOPL 0x00003000 /* IOPL mask */
6733-#define X86_EFLAGS_NT 0x00004000 /* Nested Task */
6734-#define X86_EFLAGS_RF 0x00010000 /* Resume Flag */
6735-#define X86_EFLAGS_VM 0x00020000 /* Virtual Mode */
6736-#define X86_EFLAGS_AC 0x00040000 /* Alignment Check */
6737-#define X86_EFLAGS_VIF 0x00080000 /* Virtual Interrupt Flag */
6738-#define X86_EFLAGS_VIP 0x00100000 /* Virtual Interrupt Pending */
6739-#define X86_EFLAGS_ID 0x00200000 /* CPUID detection flag */
6740-
6741-/*
6742- * Intel CPU features in CR4
6743- */
6744-#define X86_CR4_VME 0x0001 /* enable vm86 extensions */
6745-#define X86_CR4_PVI 0x0002 /* virtual interrupts flag enable */
6746-#define X86_CR4_TSD 0x0004 /* disable time stamp at ipl 3 */
6747-#define X86_CR4_DE 0x0008 /* enable debugging extensions */
6748-#define X86_CR4_PSE 0x0010 /* enable page size extensions */
6749-#define X86_CR4_PAE 0x0020 /* enable physical address extensions */
6750-#define X86_CR4_MCE 0x0040 /* Machine check enable */
6751-#define X86_CR4_PGE 0x0080 /* enable global pages */
6752-#define X86_CR4_PCE 0x0100 /* enable performance counters at ipl 3 */
6753-#define X86_CR4_OSFXSR 0x0200 /* enable fast FPU save and restore */
6754-#define X86_CR4_OSXMMEXCPT 0x0400 /* enable unmasked SSE exceptions */
6755-
6756-/*
6757 * Save the cr4 feature set we're using (ie
6758 * Pentium 4MB enable and PPro Global page
6759 * enable), so that any CPU's that boot up
6760@@ -203,7 +168,7 @@ struct i387_fxsave_struct {
6761 u32 mxcsr;
6762 u32 mxcsr_mask;
6763 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
6764- u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 128 bytes */
6765+ u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
6766 u32 padding[24];
6767 } __attribute__ ((aligned (16)));
6768
6769@@ -436,22 +401,6 @@ static inline void prefetchw(void *x)
6770 #define cpu_relax() rep_nop()
6771
6772 /*
6773- * NSC/Cyrix CPU configuration register indexes
6774- */
6775-#define CX86_CCR0 0xc0
6776-#define CX86_CCR1 0xc1
6777-#define CX86_CCR2 0xc2
6778-#define CX86_CCR3 0xc3
6779-#define CX86_CCR4 0xe8
6780-#define CX86_CCR5 0xe9
6781-#define CX86_CCR6 0xea
6782-#define CX86_CCR7 0xeb
6783-#define CX86_DIR0 0xfe
6784-#define CX86_DIR1 0xff
6785-#define CX86_ARR_BASE 0xc4
6786-#define CX86_RCR_BASE 0xdc
6787-
6788-/*
6789 * NSC/Cyrix CPU indexed register access macros
6790 */
6791
6792--- a/include/asm-x86/mach-xen/asm/segment_32.h
6793+++ b/include/asm-x86/mach-xen/asm/segment_32.h
6794@@ -39,7 +39,7 @@
6795 * 25 - APM BIOS support
6796 *
6797 * 26 - ESPFIX small SS
6798- * 27 - PDA [ per-cpu private data area ]
6799+ * 27 - per-cpu [ offset to per-cpu data area ]
6800 * 28 - unused
6801 * 29 - unused
6802 * 30 - unused
6803@@ -74,8 +74,12 @@
6804 #define GDT_ENTRY_ESPFIX_SS (GDT_ENTRY_KERNEL_BASE + 14)
6805 #define __ESPFIX_SS (GDT_ENTRY_ESPFIX_SS * 8)
6806
6807-#define GDT_ENTRY_PDA (GDT_ENTRY_KERNEL_BASE + 15)
6808-#define __KERNEL_PDA (GDT_ENTRY_PDA * 8)
6809+#define GDT_ENTRY_PERCPU (GDT_ENTRY_KERNEL_BASE + 15)
6810+#ifdef CONFIG_SMP
6811+#define __KERNEL_PERCPU (GDT_ENTRY_PERCPU * 8)
6812+#else
6813+#define __KERNEL_PERCPU 0
6814+#endif
6815
6816 #define GDT_ENTRY_DOUBLEFAULT_TSS 31
6817
6818--- a/include/asm-x86/mach-xen/asm/smp_32.h
6819+++ b/include/asm-x86/mach-xen/asm/smp_32.h
6820@@ -8,19 +8,15 @@
6821 #include <linux/kernel.h>
6822 #include <linux/threads.h>
6823 #include <linux/cpumask.h>
6824-#include <asm/pda.h>
6825 #endif
6826
6827-#ifdef CONFIG_X86_LOCAL_APIC
6828-#ifndef __ASSEMBLY__
6829-#include <asm/fixmap.h>
6830+#if defined(CONFIG_X86_LOCAL_APIC) && !defined(__ASSEMBLY__)
6831 #include <asm/bitops.h>
6832 #include <asm/mpspec.h>
6833+#include <asm/apic.h>
6834 #ifdef CONFIG_X86_IO_APIC
6835 #include <asm/io_apic.h>
6836 #endif
6837-#include <asm/apic.h>
6838-#endif
6839 #endif
6840
6841 #define BAD_APICID 0xFFu
6842@@ -52,9 +48,76 @@ extern void cpu_exit_clear(void);
6843 extern void cpu_uninit(void);
6844 #endif
6845
6846-#ifndef CONFIG_PARAVIRT
6847+#ifndef CONFIG_XEN
6848+struct smp_ops
6849+{
6850+ void (*smp_prepare_boot_cpu)(void);
6851+ void (*smp_prepare_cpus)(unsigned max_cpus);
6852+ int (*cpu_up)(unsigned cpu);
6853+ void (*smp_cpus_done)(unsigned max_cpus);
6854+
6855+ void (*smp_send_stop)(void);
6856+ void (*smp_send_reschedule)(int cpu);
6857+ int (*smp_call_function_mask)(cpumask_t mask,
6858+ void (*func)(void *info), void *info,
6859+ int wait);
6860+};
6861+
6862+extern struct smp_ops smp_ops;
6863+
6864+static inline void smp_prepare_boot_cpu(void)
6865+{
6866+ smp_ops.smp_prepare_boot_cpu();
6867+}
6868+static inline void smp_prepare_cpus(unsigned int max_cpus)
6869+{
6870+ smp_ops.smp_prepare_cpus(max_cpus);
6871+}
6872+static inline int __cpu_up(unsigned int cpu)
6873+{
6874+ return smp_ops.cpu_up(cpu);
6875+}
6876+static inline void smp_cpus_done(unsigned int max_cpus)
6877+{
6878+ smp_ops.smp_cpus_done(max_cpus);
6879+}
6880+
6881+static inline void smp_send_stop(void)
6882+{
6883+ smp_ops.smp_send_stop();
6884+}
6885+static inline void smp_send_reschedule(int cpu)
6886+{
6887+ smp_ops.smp_send_reschedule(cpu);
6888+}
6889+static inline int smp_call_function_mask(cpumask_t mask,
6890+ void (*func) (void *info), void *info,
6891+ int wait)
6892+{
6893+ return smp_ops.smp_call_function_mask(mask, func, info, wait);
6894+}
6895+
6896+void native_smp_prepare_boot_cpu(void);
6897+void native_smp_prepare_cpus(unsigned int max_cpus);
6898+int native_cpu_up(unsigned int cpunum);
6899+void native_smp_cpus_done(unsigned int max_cpus);
6900+
6901 #define startup_ipi_hook(phys_apicid, start_eip, start_esp) \
6902 do { } while (0)
6903+
6904+#else
6905+
6906+
6907+void xen_smp_send_stop(void);
6908+void xen_smp_send_reschedule(int cpu);
6909+int xen_smp_call_function_mask(cpumask_t mask,
6910+ void (*func) (void *info), void *info,
6911+ int wait);
6912+
6913+#define smp_send_stop xen_smp_send_stop
6914+#define smp_send_reschedule xen_smp_send_reschedule
6915+#define smp_call_function_mask xen_smp_call_function_mask
6916+
6917 #endif
6918
6919 /*
6920@@ -62,7 +125,8 @@ do { } while (0)
6921 * from the initial startup. We map APIC_BASE very early in page_setup(),
6922 * so this is correct in the x86 case.
6923 */
6924-#define raw_smp_processor_id() (read_pda(cpu_number))
6925+DECLARE_PER_CPU(int, cpu_number);
6926+#define raw_smp_processor_id() (x86_read_percpu(cpu_number))
6927
6928 extern cpumask_t cpu_possible_map;
6929 #define cpu_callin_map cpu_possible_map
6930@@ -73,20 +137,6 @@ static inline int num_booting_cpus(void)
6931 return cpus_weight(cpu_possible_map);
6932 }
6933
6934-#ifdef CONFIG_X86_LOCAL_APIC
6935-
6936-#ifdef APIC_DEFINITION
6937-extern int hard_smp_processor_id(void);
6938-#else
6939-#include <mach_apicdef.h>
6940-static inline int hard_smp_processor_id(void)
6941-{
6942- /* we don't want to mark this access volatile - bad code generation */
6943- return GET_APIC_ID(*(unsigned long *)(APIC_BASE+APIC_ID));
6944-}
6945-#endif
6946-#endif
6947-
6948 #define safe_smp_processor_id() smp_processor_id()
6949 extern int __cpu_disable(void);
6950 extern void __cpu_die(unsigned int cpu);
6951@@ -102,10 +152,31 @@ extern unsigned int num_processors;
6952
6953 #define NO_PROC_ID 0xFF /* No processor magic marker */
6954
6955-#endif
6956+#endif /* CONFIG_SMP */
6957
6958 #ifndef __ASSEMBLY__
6959
6960+#ifdef CONFIG_X86_LOCAL_APIC
6961+
6962+#ifdef APIC_DEFINITION
6963+extern int hard_smp_processor_id(void);
6964+#else
6965+#include <mach_apicdef.h>
6966+static inline int hard_smp_processor_id(void)
6967+{
6968+ /* we don't want to mark this access volatile - bad code generation */
6969+ return GET_APIC_ID(*(unsigned long *)(APIC_BASE+APIC_ID));
6970+}
6971+#endif /* APIC_DEFINITION */
6972+
6973+#else /* CONFIG_X86_LOCAL_APIC */
6974+
6975+#ifndef CONFIG_SMP
6976+#define hard_smp_processor_id() 0
6977+#endif
6978+
6979+#endif /* CONFIG_X86_LOCAL_APIC */
6980+
6981 extern u8 apicid_2_node[];
6982
6983 #ifdef CONFIG_X86_LOCAL_APIC
6984--- a/include/asm-x86/mach-xen/asm/smp_64.h
6985+++ b/include/asm-x86/mach-xen/asm/smp_64.h
6986@@ -11,12 +11,11 @@
6987 extern int disable_apic;
6988
6989 #ifdef CONFIG_X86_LOCAL_APIC
6990-#include <asm/fixmap.h>
6991 #include <asm/mpspec.h>
6992+#include <asm/apic.h>
6993 #ifdef CONFIG_X86_IO_APIC
6994 #include <asm/io_apic.h>
6995 #endif
6996-#include <asm/apic.h>
6997 #include <asm/thread_info.h>
6998 #endif
6999
7000@@ -41,7 +40,6 @@ extern void lock_ipi_call_lock(void);
7001 extern void unlock_ipi_call_lock(void);
7002 extern int smp_num_siblings;
7003 extern void smp_send_reschedule(int cpu);
7004-void smp_stop_cpu(void);
7005
7006 extern cpumask_t cpu_sibling_map[NR_CPUS];
7007 extern cpumask_t cpu_core_map[NR_CPUS];
7008@@ -62,14 +60,6 @@ static inline int num_booting_cpus(void)
7009
7010 #define raw_smp_processor_id() read_pda(cpunumber)
7011
7012-#ifdef CONFIG_X86_LOCAL_APIC
7013-static inline int hard_smp_processor_id(void)
7014-{
7015- /* we don't want to mark this access volatile - bad code generation */
7016- return GET_APIC_ID(*(unsigned int *)(APIC_BASE+APIC_ID));
7017-}
7018-#endif
7019-
7020 extern int __cpu_disable(void);
7021 extern void __cpu_die(unsigned int cpu);
7022 extern void prefill_possible_map(void);
7023@@ -78,6 +68,14 @@ extern unsigned __cpuinitdata disabled_c
7024
7025 #define NO_PROC_ID 0xFF /* No processor magic marker */
7026
7027+#endif /* CONFIG_SMP */
7028+
7029+#ifdef CONFIG_X86_LOCAL_APIC
7030+static inline int hard_smp_processor_id(void)
7031+{
7032+ /* we don't want to mark this access volatile - bad code generation */
7033+ return GET_APIC_ID(*(unsigned int *)(APIC_BASE+APIC_ID));
7034+}
7035 #endif
7036
7037 /*
7038--- a/include/asm-x86/mach-xen/asm/system_32.h
7039+++ b/include/asm-x86/mach-xen/asm/system_32.h
7040@@ -4,7 +4,7 @@
7041 #include <linux/kernel.h>
7042 #include <asm/segment.h>
7043 #include <asm/cpufeature.h>
7044-#include <linux/bitops.h> /* for LOCK_PREFIX */
7045+#include <asm/cmpxchg.h>
7046 #include <asm/synch_bitops.h>
7047 #include <asm/hypervisor.h>
7048
7049@@ -90,308 +90,102 @@ __asm__ __volatile__ ("movw %%dx,%1\n\t"
7050 #define savesegment(seg, value) \
7051 asm volatile("mov %%" #seg ",%0":"=rm" (value))
7052
7053-#define read_cr0() ({ \
7054- unsigned int __dummy; \
7055- __asm__ __volatile__( \
7056- "movl %%cr0,%0\n\t" \
7057- :"=r" (__dummy)); \
7058- __dummy; \
7059-})
7060-#define write_cr0(x) \
7061- __asm__ __volatile__("movl %0,%%cr0": :"r" (x))
7062-
7063-#define read_cr2() (current_vcpu_info()->arch.cr2)
7064-#define write_cr2(x) \
7065- __asm__ __volatile__("movl %0,%%cr2": :"r" (x))
7066-
7067-#define read_cr3() ({ \
7068- unsigned int __dummy; \
7069- __asm__ ( \
7070- "movl %%cr3,%0\n\t" \
7071- :"=r" (__dummy)); \
7072- __dummy = xen_cr3_to_pfn(__dummy); \
7073- mfn_to_pfn(__dummy) << PAGE_SHIFT; \
7074-})
7075-#define write_cr3(x) ({ \
7076- unsigned int __dummy = pfn_to_mfn((x) >> PAGE_SHIFT); \
7077- __dummy = xen_pfn_to_cr3(__dummy); \
7078- __asm__ __volatile__("movl %0,%%cr3": :"r" (__dummy)); \
7079-})
7080-#define read_cr4() ({ \
7081- unsigned int __dummy; \
7082- __asm__( \
7083- "movl %%cr4,%0\n\t" \
7084- :"=r" (__dummy)); \
7085- __dummy; \
7086-})
7087-#define read_cr4_safe() ({ \
7088- unsigned int __dummy; \
7089- /* This could fault if %cr4 does not exist */ \
7090- __asm__("1: movl %%cr4, %0 \n" \
7091- "2: \n" \
7092- ".section __ex_table,\"a\" \n" \
7093- ".long 1b,2b \n" \
7094- ".previous \n" \
7095- : "=r" (__dummy): "0" (0)); \
7096- __dummy; \
7097-})
7098-
7099-#define write_cr4(x) \
7100- __asm__ __volatile__("movl %0,%%cr4": :"r" (x))
7101-
7102-#define wbinvd() \
7103- __asm__ __volatile__ ("wbinvd": : :"memory")
7104-
7105-/* Clear the 'TS' bit */
7106-#define clts() (HYPERVISOR_fpu_taskswitch(0))
7107-
7108-/* Set the 'TS' bit */
7109-#define stts() (HYPERVISOR_fpu_taskswitch(1))
7110-
7111-#endif /* __KERNEL__ */
7112-
7113-static inline unsigned long get_limit(unsigned long segment)
7114+static inline void xen_clts(void)
7115 {
7116- unsigned long __limit;
7117- __asm__("lsll %1,%0"
7118- :"=r" (__limit):"r" (segment));
7119- return __limit+1;
7120+ HYPERVISOR_fpu_taskswitch(0);
7121 }
7122
7123-#define nop() __asm__ __volatile__ ("nop")
7124-
7125-#define xchg(ptr,v) ((__typeof__(*(ptr)))__xchg((unsigned long)(v),(ptr),sizeof(*(ptr))))
7126-
7127-#define tas(ptr) (xchg((ptr),1))
7128-
7129-struct __xchg_dummy { unsigned long a[100]; };
7130-#define __xg(x) ((struct __xchg_dummy *)(x))
7131+static inline unsigned long xen_read_cr0(void)
7132+{
7133+ unsigned long val;
7134+ asm volatile("movl %%cr0,%0\n\t" :"=r" (val));
7135+ return val;
7136+}
7137
7138+static inline void xen_write_cr0(unsigned long val)
7139+{
7140+ asm volatile("movl %0,%%cr0": :"r" (val));
7141+}
7142
7143-#ifdef CONFIG_X86_CMPXCHG64
7144+#define xen_read_cr2() (current_vcpu_info()->arch.cr2)
7145
7146-/*
7147- * The semantics of XCHGCMP8B are a bit strange, this is why
7148- * there is a loop and the loading of %%eax and %%edx has to
7149- * be inside. This inlines well in most cases, the cached
7150- * cost is around ~38 cycles. (in the future we might want
7151- * to do an SIMD/3DNOW!/MMX/FPU 64-bit store here, but that
7152- * might have an implicit FPU-save as a cost, so it's not
7153- * clear which path to go.)
7154- *
7155- * cmpxchg8b must be used with the lock prefix here to allow
7156- * the instruction to be executed atomically, see page 3-102
7157- * of the instruction set reference 24319102.pdf. We need
7158- * the reader side to see the coherent 64bit value.
7159- */
7160-static inline void __set_64bit (unsigned long long * ptr,
7161- unsigned int low, unsigned int high)
7162+static inline void xen_write_cr2(unsigned long val)
7163 {
7164- __asm__ __volatile__ (
7165- "\n1:\t"
7166- "movl (%0), %%eax\n\t"
7167- "movl 4(%0), %%edx\n\t"
7168- "lock cmpxchg8b (%0)\n\t"
7169- "jnz 1b"
7170- : /* no outputs */
7171- : "D"(ptr),
7172- "b"(low),
7173- "c"(high)
7174- : "ax","dx","memory");
7175+ asm volatile("movl %0,%%cr2": :"r" (val));
7176 }
7177
7178-static inline void __set_64bit_constant (unsigned long long *ptr,
7179- unsigned long long value)
7180+static inline unsigned long xen_read_cr3(void)
7181 {
7182- __set_64bit(ptr,(unsigned int)(value), (unsigned int)((value)>>32ULL));
7183+ unsigned long val;
7184+ asm volatile("movl %%cr3,%0\n\t" :"=r" (val));
7185+ return mfn_to_pfn(xen_cr3_to_pfn(val)) << PAGE_SHIFT;
7186 }
7187-#define ll_low(x) *(((unsigned int*)&(x))+0)
7188-#define ll_high(x) *(((unsigned int*)&(x))+1)
7189
7190-static inline void __set_64bit_var (unsigned long long *ptr,
7191- unsigned long long value)
7192+static inline void xen_write_cr3(unsigned long val)
7193 {
7194- __set_64bit(ptr,ll_low(value), ll_high(value));
7195+ val = xen_pfn_to_cr3(pfn_to_mfn(val >> PAGE_SHIFT));
7196+ asm volatile("movl %0,%%cr3": :"r" (val));
7197 }
7198
7199-#define set_64bit(ptr,value) \
7200-(__builtin_constant_p(value) ? \
7201- __set_64bit_constant(ptr, value) : \
7202- __set_64bit_var(ptr, value) )
7203-
7204-#define _set_64bit(ptr,value) \
7205-(__builtin_constant_p(value) ? \
7206- __set_64bit(ptr, (unsigned int)(value), (unsigned int)((value)>>32ULL) ) : \
7207- __set_64bit(ptr, ll_low(value), ll_high(value)) )
7208-
7209-#endif
7210-
7211-/*
7212- * Note: no "lock" prefix even on SMP: xchg always implies lock anyway
7213- * Note 2: xchg has side effect, so that attribute volatile is necessary,
7214- * but generally the primitive is invalid, *ptr is output argument. --ANK
7215- */
7216-static inline unsigned long __xchg(unsigned long x, volatile void * ptr, int size)
7217+static inline unsigned long xen_read_cr4(void)
7218 {
7219- switch (size) {
7220- case 1:
7221- __asm__ __volatile__("xchgb %b0,%1"
7222- :"=q" (x)
7223- :"m" (*__xg(ptr)), "0" (x)
7224- :"memory");
7225- break;
7226- case 2:
7227- __asm__ __volatile__("xchgw %w0,%1"
7228- :"=r" (x)
7229- :"m" (*__xg(ptr)), "0" (x)
7230- :"memory");
7231- break;
7232- case 4:
7233- __asm__ __volatile__("xchgl %0,%1"
7234- :"=r" (x)
7235- :"m" (*__xg(ptr)), "0" (x)
7236- :"memory");
7237- break;
7238- }
7239- return x;
7240+ unsigned long val;
7241+ asm volatile("movl %%cr4,%0\n\t" :"=r" (val));
7242+ return val;
7243 }
7244
7245-/*
7246- * Atomic compare and exchange. Compare OLD with MEM, if identical,
7247- * store NEW in MEM. Return the initial value in MEM. Success is
7248- * indicated by comparing RETURN with OLD.
7249- */
7250-
7251-#ifdef CONFIG_X86_CMPXCHG
7252-#define __HAVE_ARCH_CMPXCHG 1
7253-#define cmpxchg(ptr,o,n)\
7254- ((__typeof__(*(ptr)))__cmpxchg((ptr),(unsigned long)(o),\
7255- (unsigned long)(n),sizeof(*(ptr))))
7256-#define sync_cmpxchg(ptr,o,n)\
7257- ((__typeof__(*(ptr)))__sync_cmpxchg((ptr),(unsigned long)(o),\
7258- (unsigned long)(n),sizeof(*(ptr))))
7259-#endif
7260+static inline unsigned long xen_read_cr4_safe(void)
7261+{
7262+ unsigned long val;
7263+ /* This could fault if %cr4 does not exist */
7264+ asm("1: movl %%cr4, %0 \n"
7265+ "2: \n"
7266+ ".section __ex_table,\"a\" \n"
7267+ ".long 1b,2b \n"
7268+ ".previous \n"
7269+ : "=r" (val): "0" (0));
7270+ return val;
7271+}
7272
7273-static inline unsigned long __cmpxchg(volatile void *ptr, unsigned long old,
7274- unsigned long new, int size)
7275+static inline void xen_write_cr4(unsigned long val)
7276 {
7277- unsigned long prev;
7278- switch (size) {
7279- case 1:
7280- __asm__ __volatile__(LOCK_PREFIX "cmpxchgb %b1,%2"
7281- : "=a"(prev)
7282- : "q"(new), "m"(*__xg(ptr)), "0"(old)
7283- : "memory");
7284- return prev;
7285- case 2:
7286- __asm__ __volatile__(LOCK_PREFIX "cmpxchgw %w1,%2"
7287- : "=a"(prev)
7288- : "r"(new), "m"(*__xg(ptr)), "0"(old)
7289- : "memory");
7290- return prev;
7291- case 4:
7292- __asm__ __volatile__(LOCK_PREFIX "cmpxchgl %1,%2"
7293- : "=a"(prev)
7294- : "r"(new), "m"(*__xg(ptr)), "0"(old)
7295- : "memory");
7296- return prev;
7297- }
7298- return old;
7299+ asm volatile("movl %0,%%cr4": :"r" (val));
7300 }
7301
7302-/*
7303- * Always use locked operations when touching memory shared with a
7304- * hypervisor, since the system may be SMP even if the guest kernel
7305- * isn't.
7306- */
7307-static inline unsigned long __sync_cmpxchg(volatile void *ptr,
7308- unsigned long old,
7309- unsigned long new, int size)
7310-{
7311- unsigned long prev;
7312- switch (size) {
7313- case 1:
7314- __asm__ __volatile__("lock; cmpxchgb %b1,%2"
7315- : "=a"(prev)
7316- : "q"(new), "m"(*__xg(ptr)), "0"(old)
7317- : "memory");
7318- return prev;
7319- case 2:
7320- __asm__ __volatile__("lock; cmpxchgw %w1,%2"
7321- : "=a"(prev)
7322- : "r"(new), "m"(*__xg(ptr)), "0"(old)
7323- : "memory");
7324- return prev;
7325- case 4:
7326- __asm__ __volatile__("lock; cmpxchgl %1,%2"
7327- : "=a"(prev)
7328- : "r"(new), "m"(*__xg(ptr)), "0"(old)
7329- : "memory");
7330- return prev;
7331- }
7332- return old;
7333+static inline void xen_wbinvd(void)
7334+{
7335+ asm volatile("wbinvd": : :"memory");
7336 }
7337
7338-#ifndef CONFIG_X86_CMPXCHG
7339-/*
7340- * Building a kernel capable running on 80386. It may be necessary to
7341- * simulate the cmpxchg on the 80386 CPU. For that purpose we define
7342- * a function for each of the sizes we support.
7343- */
7344+#define read_cr0() (xen_read_cr0())
7345+#define write_cr0(x) (xen_write_cr0(x))
7346+#define read_cr2() (xen_read_cr2())
7347+#define write_cr2(x) (xen_write_cr2(x))
7348+#define read_cr3() (xen_read_cr3())
7349+#define write_cr3(x) (xen_write_cr3(x))
7350+#define read_cr4() (xen_read_cr4())
7351+#define read_cr4_safe() (xen_read_cr4_safe())
7352+#define write_cr4(x) (xen_write_cr4(x))
7353+#define wbinvd() (xen_wbinvd())
7354
7355-extern unsigned long cmpxchg_386_u8(volatile void *, u8, u8);
7356-extern unsigned long cmpxchg_386_u16(volatile void *, u16, u16);
7357-extern unsigned long cmpxchg_386_u32(volatile void *, u32, u32);
7358-
7359-static inline unsigned long cmpxchg_386(volatile void *ptr, unsigned long old,
7360- unsigned long new, int size)
7361-{
7362- switch (size) {
7363- case 1:
7364- return cmpxchg_386_u8(ptr, old, new);
7365- case 2:
7366- return cmpxchg_386_u16(ptr, old, new);
7367- case 4:
7368- return cmpxchg_386_u32(ptr, old, new);
7369- }
7370- return old;
7371-}
7372-
7373-#define cmpxchg(ptr,o,n) \
7374-({ \
7375- __typeof__(*(ptr)) __ret; \
7376- if (likely(boot_cpu_data.x86 > 3)) \
7377- __ret = __cmpxchg((ptr), (unsigned long)(o), \
7378- (unsigned long)(n), sizeof(*(ptr))); \
7379- else \
7380- __ret = cmpxchg_386((ptr), (unsigned long)(o), \
7381- (unsigned long)(n), sizeof(*(ptr))); \
7382- __ret; \
7383-})
7384-#endif
7385+/* Clear the 'TS' bit */
7386+#define clts() (xen_clts())
7387
7388-#ifdef CONFIG_X86_CMPXCHG64
7389+/* Set the 'TS' bit */
7390+#define stts() (HYPERVISOR_fpu_taskswitch(1))
7391
7392-static inline unsigned long long __cmpxchg64(volatile void *ptr, unsigned long long old,
7393- unsigned long long new)
7394+#endif /* __KERNEL__ */
7395+
7396+static inline unsigned long get_limit(unsigned long segment)
7397 {
7398- unsigned long long prev;
7399- __asm__ __volatile__(LOCK_PREFIX "cmpxchg8b %3"
7400- : "=A"(prev)
7401- : "b"((unsigned long)new),
7402- "c"((unsigned long)(new >> 32)),
7403- "m"(*__xg(ptr)),
7404- "0"(old)
7405- : "memory");
7406- return prev;
7407-}
7408-
7409-#define cmpxchg64(ptr,o,n)\
7410- ((__typeof__(*(ptr)))__cmpxchg64((ptr),(unsigned long long)(o),\
7411- (unsigned long long)(n)))
7412+ unsigned long __limit;
7413+ __asm__("lsll %1,%0"
7414+ :"=r" (__limit):"r" (segment));
7415+ return __limit+1;
7416+}
7417+
7418+#define nop() __asm__ __volatile__ ("nop")
7419
7420-#endif
7421-
7422 /*
7423 * Force strict CPU ordering.
7424 * And yes, this is required on UP too when we're talking
7425--- a/include/asm-x86/mach-xen/asm/system_64.h
7426+++ b/include/asm-x86/mach-xen/asm/system_64.h
7427@@ -3,7 +3,7 @@
7428
7429 #include <linux/kernel.h>
7430 #include <asm/segment.h>
7431-#include <asm/alternative.h>
7432+#include <asm/cmpxchg.h>
7433
7434 #include <asm/synch_bitops.h>
7435 #include <asm/hypervisor.h>
7436@@ -43,7 +43,7 @@
7437 [threadrsp] "i" (offsetof(struct task_struct, thread.rsp)), \
7438 [ti_flags] "i" (offsetof(struct thread_info, flags)),\
7439 [tif_fork] "i" (TIF_FORK), \
7440- [thread_info] "i" (offsetof(struct task_struct, thread_info)), \
7441+ [thread_info] "i" (offsetof(struct task_struct, stack)), \
7442 [pda_pcurrent] "i" (offsetof(struct x8664_pda, pcurrent)) \
7443 : "memory", "cc" __EXTRA_CLOBBER)
7444
7445@@ -92,6 +92,12 @@ static inline void write_cr0(unsigned lo
7446 machine_to_phys(__dummy); \
7447 })
7448
7449+static inline void write_cr3(unsigned long val)
7450+{
7451+ val = phys_to_machine(val);
7452+ asm volatile("movq %0,%%cr3" :: "r" (val) : "memory");
7453+}
7454+
7455 static inline unsigned long read_cr4(void)
7456 {
7457 unsigned long cr4;
7458@@ -101,7 +107,7 @@ static inline unsigned long read_cr4(voi
7459
7460 static inline void write_cr4(unsigned long val)
7461 {
7462- asm volatile("movq %0,%%cr4" :: "r" (val));
7463+ asm volatile("movq %0,%%cr4" :: "r" (val) : "memory");
7464 }
7465
7466 #define stts() (HYPERVISOR_fpu_taskswitch(1))
7467@@ -122,100 +128,6 @@ static inline void sched_cacheflush(void
7468
7469 #define nop() __asm__ __volatile__ ("nop")
7470
7471-#define xchg(ptr,v) ((__typeof__(*(ptr)))__xchg((unsigned long)(v),(ptr),sizeof(*(ptr))))
7472-
7473-#define tas(ptr) (xchg((ptr),1))
7474-
7475-#define __xg(x) ((volatile long *)(x))
7476-
7477-static inline void set_64bit(volatile unsigned long *ptr, unsigned long val)
7478-{
7479- *ptr = val;
7480-}
7481-
7482-#define _set_64bit set_64bit
7483-
7484-/*
7485- * Note: no "lock" prefix even on SMP: xchg always implies lock anyway
7486- * Note 2: xchg has side effect, so that attribute volatile is necessary,
7487- * but generally the primitive is invalid, *ptr is output argument. --ANK
7488- */
7489-static inline unsigned long __xchg(unsigned long x, volatile void * ptr, int size)
7490-{
7491- switch (size) {
7492- case 1:
7493- __asm__ __volatile__("xchgb %b0,%1"
7494- :"=q" (x)
7495- :"m" (*__xg(ptr)), "0" (x)
7496- :"memory");
7497- break;
7498- case 2:
7499- __asm__ __volatile__("xchgw %w0,%1"
7500- :"=r" (x)
7501- :"m" (*__xg(ptr)), "0" (x)
7502- :"memory");
7503- break;
7504- case 4:
7505- __asm__ __volatile__("xchgl %k0,%1"
7506- :"=r" (x)
7507- :"m" (*__xg(ptr)), "0" (x)
7508- :"memory");
7509- break;
7510- case 8:
7511- __asm__ __volatile__("xchgq %0,%1"
7512- :"=r" (x)
7513- :"m" (*__xg(ptr)), "0" (x)
7514- :"memory");
7515- break;
7516- }
7517- return x;
7518-}
7519-
7520-/*
7521- * Atomic compare and exchange. Compare OLD with MEM, if identical,
7522- * store NEW in MEM. Return the initial value in MEM. Success is
7523- * indicated by comparing RETURN with OLD.
7524- */
7525-
7526-#define __HAVE_ARCH_CMPXCHG 1
7527-
7528-static inline unsigned long __cmpxchg(volatile void *ptr, unsigned long old,
7529- unsigned long new, int size)
7530-{
7531- unsigned long prev;
7532- switch (size) {
7533- case 1:
7534- __asm__ __volatile__(LOCK_PREFIX "cmpxchgb %b1,%2"
7535- : "=a"(prev)
7536- : "q"(new), "m"(*__xg(ptr)), "0"(old)
7537- : "memory");
7538- return prev;
7539- case 2:
7540- __asm__ __volatile__(LOCK_PREFIX "cmpxchgw %w1,%2"
7541- : "=a"(prev)
7542- : "r"(new), "m"(*__xg(ptr)), "0"(old)
7543- : "memory");
7544- return prev;
7545- case 4:
7546- __asm__ __volatile__(LOCK_PREFIX "cmpxchgl %k1,%2"
7547- : "=a"(prev)
7548- : "r"(new), "m"(*__xg(ptr)), "0"(old)
7549- : "memory");
7550- return prev;
7551- case 8:
7552- __asm__ __volatile__(LOCK_PREFIX "cmpxchgq %1,%2"
7553- : "=a"(prev)
7554- : "r"(new), "m"(*__xg(ptr)), "0"(old)
7555- : "memory");
7556- return prev;
7557- }
7558- return old;
7559-}
7560-
7561-#define cmpxchg(ptr,o,n)\
7562- ((__typeof__(*(ptr)))__cmpxchg((ptr),(unsigned long)(o),\
7563- (unsigned long)(n),sizeof(*(ptr))))
7564-
7565 #ifdef CONFIG_SMP
7566 #define smp_mb() mb()
7567 #define smp_rmb() rmb()
7568--- a/include/asm-x86/mach-xen/asm/tlbflush_32.h
7569+++ b/include/asm-x86/mach-xen/asm/tlbflush_32.h
7570@@ -29,8 +29,13 @@
7571 * and page-granular flushes are available only on i486 and up.
7572 */
7573
7574+#define TLB_FLUSH_ALL 0xffffffff
7575+
7576+
7577 #ifndef CONFIG_SMP
7578
7579+#include <linux/sched.h>
7580+
7581 #define flush_tlb() __flush_tlb()
7582 #define flush_tlb_all() __flush_tlb_all()
7583 #define local_flush_tlb() __flush_tlb()
7584@@ -55,7 +60,7 @@ static inline void flush_tlb_range(struc
7585 __flush_tlb();
7586 }
7587
7588-#else
7589+#else /* SMP */
7590
7591 #include <asm/smp.h>
7592
7593@@ -84,9 +89,7 @@ struct tlb_state
7594 char __cacheline_padding[L1_CACHE_BYTES-8];
7595 };
7596 DECLARE_PER_CPU(struct tlb_state, cpu_tlbstate);
7597-
7598-
7599-#endif
7600+#endif /* SMP */
7601
7602 #define flush_tlb_kernel_range(start, end) flush_tlb_all()
7603
7604--- a/include/asm-x86/mach-xen/asm/tlbflush_64.h
7605+++ b/include/asm-x86/mach-xen/asm/tlbflush_64.h
7606@@ -2,7 +2,9 @@
7607 #define _X8664_TLBFLUSH_H
7608
7609 #include <linux/mm.h>
7610+#include <linux/sched.h>
7611 #include <asm/processor.h>
7612+#include <asm/system.h>
7613
7614 #define __flush_tlb() xen_tlb_flush()
7615
7616--- a/include/linux/pci.h
7617+++ b/include/linux/pci.h
7618@@ -239,7 +239,7 @@ struct pci_dev {
7619 int rom_attr_enabled; /* has display of the rom attribute been enabled? */
7620 struct bin_attribute *res_attr[DEVICE_COUNT_RESOURCE]; /* sysfs file for resources */
7621 struct bin_attribute *res_attr_wc[DEVICE_COUNT_RESOURCE]; /* sysfs file for WC mapping of resources */
7622-#ifdef CONFIG_PCI_MSI
7623+#if defined(CONFIG_PCI_MSI) && !defined(CONFIG_XEN)
7624 struct list_head msi_list;
7625 #endif
7626 struct pci_vpd *vpd;
7627--- a/lib/swiotlb-xen.c
7628+++ b/lib/swiotlb-xen.c
7629@@ -723,7 +723,6 @@ swiotlb_dma_supported (struct device *hw
7630 return (mask >= ((1UL << dma_bits) - 1));
7631 }
7632
7633-EXPORT_SYMBOL(swiotlb_init);
7634 EXPORT_SYMBOL(swiotlb_map_single);
7635 EXPORT_SYMBOL(swiotlb_unmap_single);
7636 EXPORT_SYMBOL(swiotlb_map_sg);
7637--- a/net/core/dev.c
7638+++ b/net/core/dev.c
7639@@ -1744,12 +1744,17 @@ static struct netdev_queue *dev_pick_tx(
7640 inline int skb_checksum_setup(struct sk_buff *skb)
7641 {
7642 if (skb->proto_csum_blank) {
7643+ struct iphdr *iph;
7644+ unsigned char *th;
7645+
7646 if (skb->protocol != htons(ETH_P_IP))
7647 goto out;
7648- skb->h.raw = (unsigned char *)skb->nh.iph + 4*skb->nh.iph->ihl;
7649- if (skb->h.raw >= skb->tail)
7650+ iph = ip_hdr(skb);
7651+ th = skb_network_header(skb) + 4 * iph->ihl;
7652+ if (th >= skb_tail_pointer(skb))
7653 goto out;
7654- switch (skb->nh.iph->protocol) {
7655+ skb->csum_start = th - skb->head;
7656+ switch (iph->protocol) {
7657 case IPPROTO_TCP:
7658 skb->csum_offset = offsetof(struct tcphdr, check);
7659 break;
7660@@ -1760,10 +1765,10 @@ inline int skb_checksum_setup(struct sk_
7661 if (net_ratelimit())
7662 printk(KERN_ERR "Attempting to checksum a non-"
7663 "TCP/UDP packet, dropping a protocol"
7664- " %d packet", skb->nh.iph->protocol);
7665+ " %d packet", iph->protocol);
7666 goto out;
7667 }
7668- if ((skb->h.raw + skb->csum_offset + 2) > skb->tail)
7669+ if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
7670 goto out;
7671 skb->ip_summed = CHECKSUM_PARTIAL;
7672 skb->proto_csum_blank = 0;
7673--- a/scripts/Makefile.xen.awk
7674+++ b/scripts/Makefile.xen.awk
7675@@ -13,7 +13,7 @@ BEGIN {
7676 next
7677 }
7678
7679-/:[[:space:]]*%\.[cS][[:space:]]/ {
7680+/:[[:space:]]*\$\(src\)\/%\.[cS][[:space:]]/ {
7681 line = gensub(/%.([cS])/, "%-xen.\\1", "g", $0)
7682 line = gensub(/(single-used-m)/, "xen-\\1", "g", line)
7683 print line
Peter's tree of IPFire 2.x