]>
git.ipfire.org Git - people/ms/u-boot.git/blob - arch/x86/cpu/mp_init.c
2 * Copyright (C) 2015 Google, Inc
4 * SPDX-License-Identifier: GPL-2.0+
6 * Based on code from the coreboot file of the same name
15 #include <asm/atomic.h>
17 #include <asm/interrupt.h>
18 #include <asm/lapic.h>
19 #include <asm/microcode.h>
23 #include <asm/processor.h>
25 #include <dm/device-internal.h>
26 #include <dm/uclass-internal.h>
29 #include <linux/linkage.h>
31 DECLARE_GLOBAL_DATA_PTR
;
33 /* Total CPUs include BSP */
36 /* This also needs to match the sipi.S assembly code for saved MSR encoding */
44 struct mp_flight_plan
{
46 struct mp_flight_record
*records
;
49 static struct mp_flight_plan mp_info
;
57 static inline void barrier_wait(atomic_t
*b
)
59 while (atomic_read(b
) == 0)
64 static inline void release_barrier(atomic_t
*b
)
70 static inline void stop_this_cpu(void)
72 /* Called by an AP when it is ready to halt and wait for a new task */
77 /* Returns 1 if timeout waiting for APs. 0 if target APs found */
78 static int wait_for_aps(atomic_t
*val
, int target
, int total_delay
,
84 while (atomic_read(val
) != target
) {
86 delayed
+= delay_step
;
87 if (delayed
>= total_delay
) {
96 static void ap_do_flight_plan(struct udevice
*cpu
)
100 for (i
= 0; i
< mp_info
.num_records
; i
++) {
101 struct mp_flight_record
*rec
= &mp_info
.records
[i
];
103 atomic_inc(&rec
->cpus_entered
);
104 barrier_wait(&rec
->barrier
);
106 if (rec
->ap_call
!= NULL
)
107 rec
->ap_call(cpu
, rec
->ap_arg
);
111 static int find_cpu_by_apic_id(int apic_id
, struct udevice
**devp
)
116 for (uclass_find_first_device(UCLASS_CPU
, &dev
);
118 uclass_find_next_device(&dev
)) {
119 struct cpu_platdata
*plat
= dev_get_parent_platdata(dev
);
121 if (plat
->cpu_id
== apic_id
) {
131 * By the time APs call ap_init() caching has been setup, and microcode has
134 static void ap_init(unsigned int cpu_index
)
140 /* Ensure the local apic is enabled */
144 ret
= find_cpu_by_apic_id(apic_id
, &dev
);
146 debug("Unknown CPU apic_id %x\n", apic_id
);
150 debug("AP: slot %d apic_id %x, dev %s\n", cpu_index
, apic_id
,
151 dev
? dev
->name
: "(apic_id not found)");
153 /* Walk the flight plan */
154 ap_do_flight_plan(dev
);
162 static const unsigned int fixed_mtrrs
[NUM_FIXED_MTRRS
] = {
163 MTRR_FIX_64K_00000_MSR
, MTRR_FIX_16K_80000_MSR
, MTRR_FIX_16K_A0000_MSR
,
164 MTRR_FIX_4K_C0000_MSR
, MTRR_FIX_4K_C8000_MSR
, MTRR_FIX_4K_D0000_MSR
,
165 MTRR_FIX_4K_D8000_MSR
, MTRR_FIX_4K_E0000_MSR
, MTRR_FIX_4K_E8000_MSR
,
166 MTRR_FIX_4K_F0000_MSR
, MTRR_FIX_4K_F8000_MSR
,
169 static inline struct saved_msr
*save_msr(int index
, struct saved_msr
*entry
)
173 msr
= msr_read(index
);
174 entry
->index
= index
;
178 /* Return the next entry */
183 static int save_bsp_msrs(char *start
, int size
)
187 struct saved_msr
*msr_entry
;
191 /* Determine number of MTRRs need to be saved */
192 msr
= msr_read(MTRR_CAP_MSR
);
193 num_var_mtrrs
= msr
.lo
& 0xff;
195 /* 2 * num_var_mtrrs for base and mask. +1 for IA32_MTRR_DEF_TYPE */
196 msr_count
= 2 * num_var_mtrrs
+ NUM_FIXED_MTRRS
+ 1;
198 if ((msr_count
* sizeof(struct saved_msr
)) > size
) {
199 printf("Cannot mirror all %d msrs\n", msr_count
);
203 msr_entry
= (void *)start
;
204 for (i
= 0; i
< NUM_FIXED_MTRRS
; i
++)
205 msr_entry
= save_msr(fixed_mtrrs
[i
], msr_entry
);
207 for (i
= 0; i
< num_var_mtrrs
; i
++) {
208 msr_entry
= save_msr(MTRR_PHYS_BASE_MSR(i
), msr_entry
);
209 msr_entry
= save_msr(MTRR_PHYS_MASK_MSR(i
), msr_entry
);
212 msr_entry
= save_msr(MTRR_DEF_TYPE_MSR
, msr_entry
);
217 static int load_sipi_vector(atomic_t
**ap_countp
, int num_cpus
)
219 struct sipi_params_16bit
*params16
;
220 struct sipi_params
*params
;
221 static char msr_save
[512];
228 /* Copy in the code */
229 code_len
= ap_start16_code_end
- ap_start16
;
230 debug("Copying SIPI code to %x: %d bytes\n", AP_DEFAULT_BASE
,
232 memcpy((void *)AP_DEFAULT_BASE
, ap_start16
, code_len
);
234 addr
= AP_DEFAULT_BASE
+ (ulong
)sipi_params_16bit
- (ulong
)ap_start16
;
235 params16
= (struct sipi_params_16bit
*)addr
;
236 params16
->ap_start
= (uint32_t)ap_start
;
237 params16
->gdt
= (uint32_t)gd
->arch
.gdt
;
238 params16
->gdt_limit
= X86_GDT_SIZE
- 1;
239 debug("gdt = %x, gdt_limit = %x\n", params16
->gdt
, params16
->gdt_limit
);
241 params
= (struct sipi_params
*)sipi_params
;
242 debug("SIPI 32-bit params at %p\n", params
);
243 params
->idt_ptr
= (uint32_t)x86_get_idt();
245 params
->stack_size
= CONFIG_AP_STACK_SIZE
;
246 size
= params
->stack_size
* num_cpus
;
247 stack
= memalign(4096, size
);
250 params
->stack_top
= (u32
)(stack
+ size
);
251 #if !defined(CONFIG_QEMU) && !defined(CONFIG_HAVE_FSP)
252 params
->microcode_ptr
= ucode_base
;
253 debug("Microcode at %x\n", params
->microcode_ptr
);
255 params
->msr_table_ptr
= (u32
)msr_save
;
256 ret
= save_bsp_msrs(msr_save
, sizeof(msr_save
));
259 params
->msr_count
= ret
;
261 params
->c_handler
= (uint32_t)&ap_init
;
263 *ap_countp
= ¶ms
->ap_count
;
264 atomic_set(*ap_countp
, 0);
265 debug("SIPI vector is ready\n");
270 static int check_cpu_devices(int expected_cpus
)
274 for (i
= 0; i
< expected_cpus
; i
++) {
278 ret
= uclass_find_device(UCLASS_CPU
, i
, &dev
);
280 debug("Cannot find CPU %d in device tree\n", i
);
288 /* Returns 1 for timeout. 0 on success */
289 static int apic_wait_timeout(int total_delay
, const char *msg
)
293 if (!(lapic_read(LAPIC_ICR
) & LAPIC_ICR_BUSY
))
296 debug("Waiting for %s...", msg
);
297 while (lapic_read(LAPIC_ICR
) & LAPIC_ICR_BUSY
) {
300 if (total
>= total_delay
) {
301 debug("timed out: aborting\n");
310 static int start_aps(int ap_count
, atomic_t
*num_aps
)
313 /* Max location is 4KiB below 1MiB */
314 const int max_vector_loc
= ((1 << 20) - (1 << 12)) >> 12;
319 /* The vector is sent as a 4k aligned address in one byte */
320 sipi_vector
= AP_DEFAULT_BASE
>> 12;
322 if (sipi_vector
> max_vector_loc
) {
323 printf("SIPI vector too large! 0x%08x\n",
328 debug("Attempting to start %d APs\n", ap_count
);
330 if (apic_wait_timeout(1000, "ICR not to be busy"))
333 /* Send INIT IPI to all but self */
334 lapic_write(LAPIC_ICR2
, SET_LAPIC_DEST_FIELD(0));
335 lapic_write(LAPIC_ICR
, LAPIC_DEST_ALLBUT
| LAPIC_INT_ASSERT
|
337 debug("Waiting for 10ms after sending INIT\n");
341 if (apic_wait_timeout(1000, "ICR not to be busy"))
344 lapic_write(LAPIC_ICR2
, SET_LAPIC_DEST_FIELD(0));
345 lapic_write(LAPIC_ICR
, LAPIC_DEST_ALLBUT
| LAPIC_INT_ASSERT
|
346 LAPIC_DM_STARTUP
| sipi_vector
);
347 if (apic_wait_timeout(10000, "first SIPI to complete"))
350 /* Wait for CPUs to check in up to 200 us */
351 wait_for_aps(num_aps
, ap_count
, 200, 15);
354 if (apic_wait_timeout(1000, "ICR not to be busy"))
357 lapic_write(LAPIC_ICR2
, SET_LAPIC_DEST_FIELD(0));
358 lapic_write(LAPIC_ICR
, LAPIC_DEST_ALLBUT
| LAPIC_INT_ASSERT
|
359 LAPIC_DM_STARTUP
| sipi_vector
);
360 if (apic_wait_timeout(10000, "second SIPI to complete"))
363 /* Wait for CPUs to check in */
364 if (wait_for_aps(num_aps
, ap_count
, 10000, 50)) {
365 debug("Not all APs checked in: %d/%d\n",
366 atomic_read(num_aps
), ap_count
);
373 static int bsp_do_flight_plan(struct udevice
*cpu
, struct mp_params
*mp_params
)
377 const int timeout_us
= 100000;
378 const int step_us
= 100;
379 int num_aps
= num_cpus
- 1;
381 for (i
= 0; i
< mp_params
->num_records
; i
++) {
382 struct mp_flight_record
*rec
= &mp_params
->flight_plan
[i
];
384 /* Wait for APs if the record is not released */
385 if (atomic_read(&rec
->barrier
) == 0) {
386 /* Wait for the APs to check in */
387 if (wait_for_aps(&rec
->cpus_entered
, num_aps
,
388 timeout_us
, step_us
)) {
389 debug("MP record %d timeout\n", i
);
394 if (rec
->bsp_call
!= NULL
)
395 rec
->bsp_call(cpu
, rec
->bsp_arg
);
397 release_barrier(&rec
->barrier
);
402 static int init_bsp(struct udevice
**devp
)
404 char processor_name
[CPU_MAX_NAME_LEN
];
408 cpu_get_name(processor_name
);
409 debug("CPU: %s\n", processor_name
);
412 ret
= find_cpu_by_apic_id(apic_id
, devp
);
414 printf("Cannot find boot CPU, APIC ID %d\n", apic_id
);
422 static int qemu_cpu_fixup(void)
427 struct udevice
*dev
, *pdev
;
428 struct cpu_platdata
*plat
;
431 /* first we need to find '/cpus' */
432 for (device_find_first_child(dm_root(), &pdev
);
434 device_find_next_child(&pdev
)) {
435 if (!strcmp(pdev
->name
, "cpus"))
439 printf("unable to find cpus device\n");
443 /* calculate cpus that are already bound */
445 for (uclass_find_first_device(UCLASS_CPU
, &dev
);
447 uclass_find_next_device(&dev
)) {
451 /* get actual cpu number */
452 cpu_online
= qemu_fwcfg_online_cpus();
453 if (cpu_online
< 0) {
454 printf("unable to get online cpu number: %d\n", cpu_online
);
458 /* bind addtional cpus */
460 for (; cpu_num
< cpu_online
; cpu_num
++) {
462 * allocate device name here as device_bind_driver() does
463 * not copy device name, 8 bytes are enough for
464 * sizeof("cpu@") + 3 digits cpu number + '\0'
468 printf("unable to allocate device name\n");
471 sprintf(cpu
, "cpu@%d", cpu_num
);
472 ret
= device_bind_driver(pdev
, "cpu_qemu", cpu
, &dev
);
474 printf("binding cpu@%d failed: %d\n", cpu_num
, ret
);
477 plat
= dev_get_parent_platdata(dev
);
478 plat
->cpu_id
= cpu_num
;
484 int mp_init(struct mp_params
*p
)
491 /* This will cause the CPUs devices to be bound */
493 ret
= uclass_get(UCLASS_CPU
, &uc
);
498 ret
= qemu_cpu_fixup();
503 ret
= init_bsp(&cpu
);
505 debug("Cannot init boot CPU: err=%d\n", ret
);
509 if (p
== NULL
|| p
->flight_plan
== NULL
|| p
->num_records
< 1) {
510 printf("Invalid MP parameters\n");
514 num_cpus
= cpu_get_count(cpu
);
516 debug("Cannot get number of CPUs: err=%d\n", num_cpus
);
521 debug("Warning: Only 1 CPU is detected\n");
523 ret
= check_cpu_devices(num_cpus
);
525 debug("Warning: Device tree does not describe all CPUs. Extra ones will not be started correctly\n");
527 /* Copy needed parameters so that APs have a reference to the plan */
528 mp_info
.num_records
= p
->num_records
;
529 mp_info
.records
= p
->flight_plan
;
531 /* Load the SIPI vector */
532 ret
= load_sipi_vector(&ap_count
, num_cpus
);
533 if (ap_count
== NULL
)
537 * Make sure SIPI data hits RAM so the APs that come up will see
538 * the startup code even if the caches are disabled
542 /* Start the APs providing number of APs and the cpus_entered field */
543 num_aps
= num_cpus
- 1;
544 ret
= start_aps(num_aps
, ap_count
);
547 debug("%d/%d eventually checked in?\n", atomic_read(ap_count
),
552 /* Walk the flight plan for the BSP */
553 ret
= bsp_do_flight_plan(cpu
, p
);
555 debug("CPU init failed: err=%d\n", ret
);
562 int mp_init_cpu(struct udevice
*cpu
, void *unused
)
564 struct cpu_platdata
*plat
= dev_get_parent_platdata(cpu
);
567 * Multiple APs are brought up simultaneously and they may get the same
568 * seq num in the uclass_resolve_seq() during device_probe(). To avoid
569 * this, set req_seq to the reg number in the device tree in advance.
571 cpu
->req_seq
= fdtdec_get_int(gd
->fdt_blob
, dev_of_offset(cpu
), "reg",
573 plat
->ucode_version
= microcode_read_rev();
574 plat
->device_id
= gd
->arch
.x86_device
;
576 return device_probe(cpu
);