src/patches/suse-2.6.27.39/patches.arch/amd_magny_cour_topology_fix.patch

   1 From: Andreas Herrmann <andreas.herrmann3@amd.com>
   2 Subject: x86: Fix CPU llc_shared_map information for AMD Magny-Cours
   3 References: fate#307306
   4 Patch-Mainline: yes
   5 Commit-ID: 4a376ec3a2599c02207cd4cbd5dbf73783548463
   6
   7 Signed-off-by: Thomas Renninger <trenn@suse.de>
   8
   9 Construct entire NodeID and use it as cpu_llc_id. Thus internal node
  10 siblings are stored in llc_shared_map.
  11
  12 Signed-off-by: Andreas Herrmann <andreas.herrmann3@amd.com>
  13 Signed-off-by: H. Peter Anvin <hpa@zytor.com>
  14
  15 ---
  16  arch/x86/kernel/cpu/amd.c    |   62 +++++++++++++++++++++++++++++++++++++++++
  17  arch/x86/kernel/cpu/amd_64.c |   64 +++++++++++++++++++++++++++++++++++++++++++
  18  include/asm-x86/cpufeature.h |    1
  19  3 files changed, 127 insertions(+)
  20
  21 Index: linux-2.6.27-SLE11_BRANCH/arch/x86/kernel/cpu/amd.c
  22 ===================================================================
  23 --- linux-2.6.27-SLE11_BRANCH.orig/arch/x86/kernel/cpu/amd.c
  24 +++ linux-2.6.27-SLE11_BRANCH/arch/x86/kernel/cpu/amd.c
  25 @@ -4,6 +4,7 @@
  26  #include <asm/io.h>
  27  #include <asm/processor.h>
  28  #include <asm/apic.h>
  29 +#include <asm/pci-direct.h>
  30
  31  #include <mach_apic.h>
  32  #include "cpu.h"
  33 @@ -38,6 +39,64 @@ static void __cpuinit early_init_amd(str
  34                 set_cpu_cap(c, X86_FEATURE_K6_MTRR);
  35  }
  36
  37 +/*
  38 + * Fixup core topology information for AMD multi-node processors.
  39 + * Assumption 1: Number of cores in each internal node is the same.
  40 + * Assumption 2: Mixed systems with both single-node and dual-node
  41 + *               processors are not supported.
  42 + */
  43 +#ifdef CONFIG_X86_HT
  44 +static void __cpuinit amd_fixup_dcm(struct cpuinfo_x86 *c)
  45 +{
  46 +#ifdef CONFIG_PCI
  47 +       u32 t, cpn;
  48 +       u8 n, n_id;
  49 +       int cpu = smp_processor_id();
  50 +
  51 +       /* fixup topology information only once for a core */
  52 +       if (cpu_has(c, X86_FEATURE_AMD_DCM))
  53 +               return;
  54 +
  55 +       /* check for multi-node processor on boot cpu */
  56 +       t = read_pci_config(0, 24, 3, 0xe8);
  57 +       if (!(t & (1 << 29)))
  58 +               return;
  59 +
  60 +       set_cpu_cap(c, X86_FEATURE_AMD_DCM);
  61 +
  62 +       /* cores per node: each internal node has half the number of cores */
  63 +       cpn = c->x86_max_cores >> 1;
  64 +
  65 +       /* even-numbered NB_id of this dual-node processor */
  66 +       n = c->phys_proc_id << 1;
  67 +
  68 +       /*
  69 +        * determine internal node id and assign cores fifty-fifty to
  70 +        * each node of the dual-node processor
  71 +        */
  72 +       t = read_pci_config(0, 24 + n, 3, 0xe8);
  73 +       n = (t>>30) & 0x3;
  74 +       if (n == 0) {
  75 +               if (c->cpu_core_id < cpn)
  76 +                       n_id = 0;
  77 +               else
  78 +                       n_id = 1;
  79 +       } else {
  80 +               if (c->cpu_core_id < cpn)
  81 +                       n_id = 1;
  82 +               else
  83 +                       n_id = 0;
  84 +       }
  85 +
  86 +       /* compute entire NodeID, use llc_shared_map to store sibling info */
  87 +       per_cpu(cpu_llc_id, cpu) = (c->phys_proc_id << 1) + n_id;
  88 +
  89 +       /* fixup core id to be in range from 0 to cpn */
  90 +       c->cpu_core_id = c->cpu_core_id % cpn;
  91 +#endif
  92 +}
  93 +#endif
  94 +
  95  static void __cpuinit init_amd(struct cpuinfo_x86 *c)
  96  {
  97         u32 l, h;
  98 @@ -250,6 +309,9 @@ static void __cpuinit init_amd(struct cp
  99                 printk(KERN_INFO "CPU %d(%d) -> Core %d\n",
 100                        cpu, c->x86_max_cores, c->cpu_core_id);
 101         }
 102 +       /* fixup topology information on multi-node processors */
 103 +       if ((c->x86 == 0x10) && (c->x86_model == 9))
 104 +               amd_fixup_dcm(c);
 105  #endif
 106
 107         if (cpuid_eax(0x80000000) >= 0x80000006) {
 108 Index: linux-2.6.27-SLE11_BRANCH/include/asm-x86/cpufeature.h
 109 ===================================================================
 110 --- linux-2.6.27-SLE11_BRANCH.orig/include/asm-x86/cpufeature.h
 111 +++ linux-2.6.27-SLE11_BRANCH/include/asm-x86/cpufeature.h
 112 @@ -84,6 +84,7 @@
 113  #define X86_FEATURE_AMDC1E     (3*32+21) /* AMD C1E detected */
 114  #define X86_FEATURE_XTOPOLOGY  (3*32+22) /* cpu topology enum extensions */
 115  #define X86_FEATURE_TSC_RELIABLE (3*32+23) /* TSC is known to be reliable */
 116 +#define X86_FEATURE_AMD_DCM     (3*32+27) /* multi-node processor */
 117
 118  /* Intel-defined CPU features, CPUID level 0x00000001 (ecx), word 4 */
 119  #define X86_FEATURE_XMM3       (4*32+ 0) /* Streaming SIMD Extensions-3 */
 120 Index: linux-2.6.27-SLE11_BRANCH/arch/x86/kernel/cpu/amd_64.c
 121 ===================================================================
 122 --- linux-2.6.27-SLE11_BRANCH.orig/arch/x86/kernel/cpu/amd_64.c
 123 +++ linux-2.6.27-SLE11_BRANCH/arch/x86/kernel/cpu/amd_64.c
 124 @@ -4,6 +4,7 @@
 125  #include <asm/numa_64.h>
 126  #include <asm/mmconfig.h>
 127  #include <asm/cacheflush.h>
 128 +#include <asm/pci-direct.h>
 129
 130  #include <mach_apic.h>
 131
 132 @@ -31,6 +32,64 @@ static int __cpuinit nearby_node(int api
 133  #endif
 134
 135  /*
 136 + * Fixup core topology information for AMD multi-node processors.
 137 + * Assumption 1: Number of cores in each internal node is the same.
 138 + * Assumption 2: Mixed systems with both single-node and dual-node
 139 + *               processors are not supported.
 140 + */
 141 +#ifdef CONFIG_X86_HT
 142 +static void __cpuinit amd_fixup_dcm(struct cpuinfo_x86 *c)
 143 +{
 144 +#ifdef CONFIG_PCI
 145 +       u32 t, cpn;
 146 +       u8 n, n_id;
 147 +       int cpu = smp_processor_id();
 148 +
 149 +       /* fixup topology information only once for a core */
 150 +       if (cpu_has(c, X86_FEATURE_AMD_DCM))
 151 +               return;
 152 +
 153 +       /* check for multi-node processor on boot cpu */
 154 +       t = read_pci_config(0, 24, 3, 0xe8);
 155 +       if (!(t & (1 << 29)))
 156 +               return;
 157 +
 158 +       set_cpu_cap(c, X86_FEATURE_AMD_DCM);
 159 +
 160 +       /* cores per node: each internal node has half the number of cores */
 161 +       cpn = c->x86_max_cores >> 1;
 162 +
 163 +       /* even-numbered NB_id of this dual-node processor */
 164 +       n = c->phys_proc_id << 1;
 165 +
 166 +       /*
 167 +        * determine internal node id and assign cores fifty-fifty to
 168 +        * each node of the dual-node processor
 169 +        */
 170 +       t = read_pci_config(0, 24 + n, 3, 0xe8);
 171 +       n = (t>>30) & 0x3;
 172 +       if (n == 0) {
 173 +               if (c->cpu_core_id < cpn)
 174 +                       n_id = 0;
 175 +               else
 176 +                       n_id = 1;
 177 +       } else {
 178 +               if (c->cpu_core_id < cpn)
 179 +                       n_id = 1;
 180 +               else
 181 +                       n_id = 0;
 182 +       }
 183 +
 184 +       /* compute entire NodeID, use llc_shared_map to store sibling info */
 185 +       per_cpu(cpu_llc_id, cpu) = (c->phys_proc_id << 1) + n_id;
 186 +
 187 +       /* fixup core id to be in range from 0 to cpn */
 188 +       c->cpu_core_id = c->cpu_core_id % cpn;
 189 +#endif
 190 +}
 191 +#endif
 192 +
 193 +/*
 194   * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
 195   * Assumes number of cores is a power of two.
 196   */
 197 @@ -49,6 +108,11 @@ static void __cpuinit amd_detect_cmp(str
 198         c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
 199         /* Convert the initial APIC ID into the socket ID */
 200         c->phys_proc_id = c->initial_apicid >> bits;
 201 +#ifdef CONFIG_X86_HT
 202 +       /* fixup topology information on multi-node processors */
 203 +       if ((c->x86 == 0x10) && (c->x86_model == 9))
 204 +               amd_fixup_dcm(c);
 205 +#endif
 206
 207  #ifdef CONFIG_NUMA
 208         node = c->phys_proc_id;