diff options
| author | Andrey Petrov <anpetrov@fb.com> | 2019-11-15 13:19:08 -0800 |
|---|---|---|
| committer | Patrick Georgi <pgeorgi@google.com> | 2019-11-26 11:45:31 +0000 |
| commit | 515ef38db40cc44592770c00be8e4980bbaccc69 (patch) | |
| tree | 41d462dcf759f60f4d21803115eca3d5205c8d48 /src/arch/x86/smbios.c | |
| parent | 2ed6848ea3530beb234a3b9d5c843a54688d1d3c (diff) | |
arch/x86: SMBIOS: Improve core count reporting
Current code uses CPUID leaf 0x1, EBX bits 16:23 to determine number for
"core count". However, it turns out this number has little to do with
real number of cores. According to SDM vol 2A, it stays for "maximum
number of addressable IDs for logical processors in this physical
package". This does not seem to take into account fusing of giving
processor.
The new code determines 'core count' by dividing thread-level cpus by
reported logical cores. This seems to be the only way to arrive
to number of cores as it is reported in official CPU datasheet.
TEST=tested on OCP monolake
Change-Id: Id4ba9e3079f92ffe38f9104ffcfafe62582dd259
Signed-off-by: Andrey Petrov <anpetrov@fb.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/36941
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Werner Zeh <werner.zeh@siemens.com>
Diffstat (limited to 'src/arch/x86/smbios.c')
| -rw-r--r-- | src/arch/x86/smbios.c | 21 |
1 files changed, 20 insertions, 1 deletions
diff --git a/src/arch/x86/smbios.c b/src/arch/x86/smbios.c index 7e05408242..a599addb61 100644 --- a/src/arch/x86/smbios.c +++ b/src/arch/x86/smbios.c @@ -647,7 +647,26 @@ static int smbios_write_type4(unsigned long *current, int handle) t->processor_version = smbios_processor_name(t->eos); t->processor_family = (res.eax > 0) ? 0x0c : 0x6; t->processor_type = 3; /* System Processor */ - t->core_count = (res.ebx >> 16) & 0xff; + /* + * If CPUID leaf 11 is available, calculate "core count" by dividing + * SMT_ID (logical processors in a core) by Core_ID (number of cores). + * This seems to be the way to arrive to a number of cores mentioned on + * ark.intel.com. + */ + if (cpu_have_cpuid() && cpuid_get_max_func() >= 0xb) { + uint32_t leaf_b_cores = 0, leaf_b_threads = 0; + res = cpuid_ext(0xb, 1); + leaf_b_cores = res.ebx; + res = cpuid_ext(0xb, 0); + leaf_b_threads = res.ebx; + /* if hyperthreading is not available, pretend this is 1 */ + if (leaf_b_threads == 0) { + leaf_b_threads = 1; + } + t->core_count = leaf_b_cores / leaf_b_threads; + } else { + t->core_count = (res.ebx >> 16) & 0xff; + } /* Assume we enable all the cores always, capped only by MAX_CPUS */ t->core_enabled = MIN(t->core_count, CONFIG_MAX_CPUS); t->l1_cache_handle = 0xffff; |