All of lore.kernel.org
 help / color / mirror / Atom feed
From: Eduardo Habkost <ehabkost@redhat.com>
To: Babu Moger <babu.moger@amd.com>
Cc: Thomas.Lendacky@amd.com, brijesh.singh@amd.com,
	kvm@vger.kernel.org, rkrcmar@redhat.com, kash@tripleback.net,
	mtosatti@redhat.com, Gary.Hook@amd.com, qemu-devel@nongnu.org,
	pbonzini@redhat.com, rth@twiddle.net
Subject: Re: [PATCH v4 2/5] target/i386: Populate AMD Processor Cache Information
Date: Fri, 16 Mar 2018 15:00:06 -0300	[thread overview]
Message-ID: <20180316180006.GH28578@localhost.localdomain> (raw)
In-Reply-To: <1520888449-4352-3-git-send-email-babu.moger@amd.com>

On Mon, Mar 12, 2018 at 05:00:46PM -0400, Babu Moger wrote:
> From: Stanislav Lanci <pixo@polepetko.eu>
> 
> Add information for cpuid 0x8000001D leaf. Populate cache topology information
> for different cache types(Data Cache, Instruction Cache, L2 and L3) supported
> by 0x8000001D leaf. Please refer Processor Programming Reference (PPR) for AMD
> Family 17h Model for more details.
> 
> Signed-off-by: Stanislav Lanci <pixo@polepetko.eu>
> Signed-off-by: Babu Moger <babu.moger@amd.com>

The new CPUID leaves don't seem to match the existing AMD cache information
leaves.  Is this intentional?  Why?

Details below:

> ---
>  target/i386/cpu.c | 65 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  target/i386/kvm.c | 29 ++++++++++++++++++++++---
>  2 files changed, 91 insertions(+), 3 deletions(-)
> 
> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> index 42dd381..5fdbedd 100644
> --- a/target/i386/cpu.c
> +++ b/target/i386/cpu.c
[...]
> @@ -3590,6 +3594,67 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,
>              *edx = 0;
>          }
>          break;
> +    case 0x8000001D: /* AMD TOPOEXT cache info */
> +        switch (count) {

Copying macro definitions here, for reference:

> /* L1 data cache: */
> #define L1D_LINE_SIZE         64
> #define L1D_ASSOCIATIVITY      8
> #define L1D_SETS              64
> #define L1D_PARTITIONS         1
> /* Size = LINE_SIZE*ASSOCIATIVITY*SETS*PARTITIONS = 32KiB */
> #define L1D_DESCRIPTOR CPUID_2_L1D_32KB_8WAY_64B
> /*FIXME: CPUID leaf 0x80000005 is inconsistent with leaves 2 & 4 */
> #define L1D_LINES_PER_TAG      1
> #define L1D_SIZE_KB_AMD       64
> #define L1D_ASSOCIATIVITY_AMD  2

So, we already have:

CPUID[2]: 32KB 8-way cache, 64-byte lines
CPUID[4]: 8-way cache, 64-byte lines, 64 sets, 1 partition (32 KB)
CPUID[0x80000005]: 64 KB 2-way cache, 1 line per tag


> +        case 0: /* L1 dcache info */
> +            *eax |= TYPE_DCACHE | \
> +                    CACHE_LEVEL(1) | \
> +                    CACHE_SELF_INIT_LEVEL | \
> +                    ((cs->nr_threads - 1) << 14);
> +            *ebx = (L1D_LINE_SIZE - 1) | \
> +                   ((L1D_PARTITIONS - 1) << 12) | \
> +                   ((L1D_ASSOCIATIVITY - 1) << 22);
> +            *ecx = L1D_SETS - 1;
> +            *edx = 0;
> +            break;

This adds:
CPUID[0x8000001D]: 8-way cache, 64-byte lines, 64 sets, 1 partition (32 KiB)

This agrees with CPUID[2] and CPUID[4] (Intel leaves, reserved on AMD), but not
with CPUID[0x80000005].

>  
>  /* L1 instruction cache: */
>  #define L1I_LINE_SIZE         64
>  #define L1I_ASSOCIATIVITY      8
>  #define L1I_SETS              64
> +#define L1I_SETS_AMD         256
>  #define L1I_PARTITIONS         1
>  /* Size = LINE_SIZE*ASSOCIATIVITY*SETS*PARTITIONS = 32KiB */
>  #define L1I_DESCRIPTOR CPUID_2_L1I_32KB_8WAY_64B
>  /*FIXME: CPUID leaf 0x80000005 is inconsistent with leaves 2 & 4 */
>  #define L1I_LINES_PER_TAG      1
>  #define L1I_SIZE_KB_AMD       64
>  #define L1I_ASSOCIATIVITY_AMD  2

Currently we have:

CPUID[2]: 32KiB 8-way cache, 64-byte lines
CPUID[4]: 8-way cache, 64-byte lines, 64 sets, 1 partition (32 KiB)
CPUID[0x80000005]: 64 KiB 2-way cache, 1 line per tag


>  
> +        case 1: /* L1 icache info */
> +            *eax |= TYPE_ICACHE | \
> +                    CACHE_LEVEL(1) | \
> +                    CACHE_SELF_INIT_LEVEL | \
> +                    ((cs->nr_threads - 1) << 14);
> +            *ebx = (L1I_LINE_SIZE - 1) | \
> +                   ((L1I_PARTITIONS - 1) << 12) | \
> +                   ((L1I_ASSOCIATIVITY_AMD - 1) << 22);
> +            *ecx = L1I_SETS_AMD - 1;
> +            *edx = 0;
> +            break;

This adds:
CPUID[0x8000001D]: 2-way cache, 64-byte lines, 256 sets, 1 partition (32 KiB)

This doesn't match any of the existing leaves.


>  /* Level 2 unified cache: */
>  #define L2_LINE_SIZE          64
>  #define L2_ASSOCIATIVITY      16
> +#define L2_ASSOCIATIVITY_AMD   8
>  #define L2_SETS             4096
> +#define L2_SETS_AMD         1024
>  #define L2_PARTITIONS          1
>  /* Size = LINE_SIZE*ASSOCIATIVITY*SETS*PARTITIONS = 4MiB */
>  /*FIXME: CPUID leaf 2 descriptor is inconsistent with CPUID leaf 4 */
>  #define L2_DESCRIPTOR CPUID_2_L2_2MB_8WAY_64B
>  /*FIXME: CPUID leaf 0x80000006 is inconsistent with leaves 2 & 4 */
>  #define L2_LINES_PER_TAG       1
>  #define L2_SIZE_KB_AMD       512

Currently we have:
CPUID[2]: 4MiB 8-way cache, 64-byte lines
CPUID[4]: 64-byte lines, 16-way, 1 partition, 4096 sets (4 MiB)
CPUID[0x80000006]: 512 KiB, 16-way cache, 1 line per tag

>  
> +        case 2: /* L2 cache info */
> +            *eax |= TYPE_UNIFIED | \
> +                    CACHE_LEVEL(2) | \
> +                    CACHE_SELF_INIT_LEVEL | \
> +                    ((cs->nr_threads - 1) << 14);
> +            *ebx = (L2_LINE_SIZE - 1) | \
> +                   ((L2_PARTITIONS - 1) << 12) | \
> +                   ((L2_ASSOCIATIVITY_AMD - 1) << 22);
> +            *ecx = L2_SETS_AMD - 1;
> +            *edx = CACHE_INCLUSIVE;
> +            break;

This adds:
CPUID[0x8000001D]: 64-byte lines, 8-way, 1 partition, 1024 sets (512 KiB).

This doesn't match any of the existing leaves.


>  /* Level 3 unified cache: */
>  #define L3_N_LINE_SIZE         64
>  #define L3_N_ASSOCIATIVITY     16
>  #define L3_N_SETS           16384
> +#define L3_N_SETS_AMD        8192
>  #define L3_N_PARTITIONS         1
>  #define L3_N_DESCRIPTOR CPUID_2_L3_16MB_16WAY_64B
>  #define L3_N_LINES_PER_TAG      1
>  #define L3_N_SIZE_KB_AMD    16384
>  

Currently we have:
CPUID[2]: 16MiB 16-way cache, 64-byte lines
CPUID[4]: 64-byte lines, 16-way, 16384 sets, 1 partition (16 MiB)
CPUID[0x80000006]: 16 MiB cache, 16-way, 1 line per tag


> +        case 3: /* L3 cache info */
> +            if (cpu->enable_l3_cache) {
> +                *eax |= TYPE_UNIFIED | \
> +                        CACHE_LEVEL(3) | \
> +                        CACHE_SELF_INIT_LEVEL | \
> +                        ((cs->nr_cores * cs->nr_threads - 1) << 14);
> +                *ebx = (L3_N_LINE_SIZE - 1) | \
> +                       ((L3_N_PARTITIONS - 1) << 12) | \
> +                       ((L3_N_ASSOCIATIVITY - 1) << 22);
> +                *ecx = L3_N_SETS_AMD - 1;
> +                *edx = CACHE_NO_INVD_SHARING;

This adds:
CPUID[0x8000001D]: 64-byte lines, 16-way, 1 partition, 8192 sets (8 MiB)

This doesn't match any of the existing leaves.


> +            } else {
> +                *eax = 0;
> +                *ebx = 0;
> +                *ecx = 0;
> +                *edx = 0;
> +            }
> +            break;
> +        default: /* end of info */
> +            *eax = 0;
> +            *ebx = 0;
> +            *ecx = 0;
> +            *edx = 0;
> +            break;
> +        }
> +        break;
>      case 0xC0000000:
>          *eax = env->cpuid_xlevel2;
>          *ebx = 0;
[...]

-- 
Eduardo

WARNING: multiple messages have this Message-ID (diff)
From: Eduardo Habkost <ehabkost@redhat.com>
To: Babu Moger <babu.moger@amd.com>
Cc: pbonzini@redhat.com, rth@twiddle.net, rkrcmar@redhat.com,
	Thomas.Lendacky@amd.com, brijesh.singh@amd.com,
	kvm@vger.kernel.org, kash@tripleback.net, mtosatti@redhat.com,
	Gary.Hook@amd.com, qemu-devel@nongnu.org
Subject: Re: [Qemu-devel] [PATCH v4 2/5] target/i386: Populate AMD Processor Cache Information
Date: Fri, 16 Mar 2018 15:00:06 -0300	[thread overview]
Message-ID: <20180316180006.GH28578@localhost.localdomain> (raw)
In-Reply-To: <1520888449-4352-3-git-send-email-babu.moger@amd.com>

On Mon, Mar 12, 2018 at 05:00:46PM -0400, Babu Moger wrote:
> From: Stanislav Lanci <pixo@polepetko.eu>
> 
> Add information for cpuid 0x8000001D leaf. Populate cache topology information
> for different cache types(Data Cache, Instruction Cache, L2 and L3) supported
> by 0x8000001D leaf. Please refer Processor Programming Reference (PPR) for AMD
> Family 17h Model for more details.
> 
> Signed-off-by: Stanislav Lanci <pixo@polepetko.eu>
> Signed-off-by: Babu Moger <babu.moger@amd.com>

The new CPUID leaves don't seem to match the existing AMD cache information
leaves.  Is this intentional?  Why?

Details below:

> ---
>  target/i386/cpu.c | 65 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  target/i386/kvm.c | 29 ++++++++++++++++++++++---
>  2 files changed, 91 insertions(+), 3 deletions(-)
> 
> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> index 42dd381..5fdbedd 100644
> --- a/target/i386/cpu.c
> +++ b/target/i386/cpu.c
[...]
> @@ -3590,6 +3594,67 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,
>              *edx = 0;
>          }
>          break;
> +    case 0x8000001D: /* AMD TOPOEXT cache info */
> +        switch (count) {

Copying macro definitions here, for reference:

> /* L1 data cache: */
> #define L1D_LINE_SIZE         64
> #define L1D_ASSOCIATIVITY      8
> #define L1D_SETS              64
> #define L1D_PARTITIONS         1
> /* Size = LINE_SIZE*ASSOCIATIVITY*SETS*PARTITIONS = 32KiB */
> #define L1D_DESCRIPTOR CPUID_2_L1D_32KB_8WAY_64B
> /*FIXME: CPUID leaf 0x80000005 is inconsistent with leaves 2 & 4 */
> #define L1D_LINES_PER_TAG      1
> #define L1D_SIZE_KB_AMD       64
> #define L1D_ASSOCIATIVITY_AMD  2

So, we already have:

CPUID[2]: 32KB 8-way cache, 64-byte lines
CPUID[4]: 8-way cache, 64-byte lines, 64 sets, 1 partition (32 KB)
CPUID[0x80000005]: 64 KB 2-way cache, 1 line per tag


> +        case 0: /* L1 dcache info */
> +            *eax |= TYPE_DCACHE | \
> +                    CACHE_LEVEL(1) | \
> +                    CACHE_SELF_INIT_LEVEL | \
> +                    ((cs->nr_threads - 1) << 14);
> +            *ebx = (L1D_LINE_SIZE - 1) | \
> +                   ((L1D_PARTITIONS - 1) << 12) | \
> +                   ((L1D_ASSOCIATIVITY - 1) << 22);
> +            *ecx = L1D_SETS - 1;
> +            *edx = 0;
> +            break;

This adds:
CPUID[0x8000001D]: 8-way cache, 64-byte lines, 64 sets, 1 partition (32 KiB)

This agrees with CPUID[2] and CPUID[4] (Intel leaves, reserved on AMD), but not
with CPUID[0x80000005].

>  
>  /* L1 instruction cache: */
>  #define L1I_LINE_SIZE         64
>  #define L1I_ASSOCIATIVITY      8
>  #define L1I_SETS              64
> +#define L1I_SETS_AMD         256
>  #define L1I_PARTITIONS         1
>  /* Size = LINE_SIZE*ASSOCIATIVITY*SETS*PARTITIONS = 32KiB */
>  #define L1I_DESCRIPTOR CPUID_2_L1I_32KB_8WAY_64B
>  /*FIXME: CPUID leaf 0x80000005 is inconsistent with leaves 2 & 4 */
>  #define L1I_LINES_PER_TAG      1
>  #define L1I_SIZE_KB_AMD       64
>  #define L1I_ASSOCIATIVITY_AMD  2

Currently we have:

CPUID[2]: 32KiB 8-way cache, 64-byte lines
CPUID[4]: 8-way cache, 64-byte lines, 64 sets, 1 partition (32 KiB)
CPUID[0x80000005]: 64 KiB 2-way cache, 1 line per tag


>  
> +        case 1: /* L1 icache info */
> +            *eax |= TYPE_ICACHE | \
> +                    CACHE_LEVEL(1) | \
> +                    CACHE_SELF_INIT_LEVEL | \
> +                    ((cs->nr_threads - 1) << 14);
> +            *ebx = (L1I_LINE_SIZE - 1) | \
> +                   ((L1I_PARTITIONS - 1) << 12) | \
> +                   ((L1I_ASSOCIATIVITY_AMD - 1) << 22);
> +            *ecx = L1I_SETS_AMD - 1;
> +            *edx = 0;
> +            break;

This adds:
CPUID[0x8000001D]: 2-way cache, 64-byte lines, 256 sets, 1 partition (32 KiB)

This doesn't match any of the existing leaves.


>  /* Level 2 unified cache: */
>  #define L2_LINE_SIZE          64
>  #define L2_ASSOCIATIVITY      16
> +#define L2_ASSOCIATIVITY_AMD   8
>  #define L2_SETS             4096
> +#define L2_SETS_AMD         1024
>  #define L2_PARTITIONS          1
>  /* Size = LINE_SIZE*ASSOCIATIVITY*SETS*PARTITIONS = 4MiB */
>  /*FIXME: CPUID leaf 2 descriptor is inconsistent with CPUID leaf 4 */
>  #define L2_DESCRIPTOR CPUID_2_L2_2MB_8WAY_64B
>  /*FIXME: CPUID leaf 0x80000006 is inconsistent with leaves 2 & 4 */
>  #define L2_LINES_PER_TAG       1
>  #define L2_SIZE_KB_AMD       512

Currently we have:
CPUID[2]: 4MiB 8-way cache, 64-byte lines
CPUID[4]: 64-byte lines, 16-way, 1 partition, 4096 sets (4 MiB)
CPUID[0x80000006]: 512 KiB, 16-way cache, 1 line per tag

>  
> +        case 2: /* L2 cache info */
> +            *eax |= TYPE_UNIFIED | \
> +                    CACHE_LEVEL(2) | \
> +                    CACHE_SELF_INIT_LEVEL | \
> +                    ((cs->nr_threads - 1) << 14);
> +            *ebx = (L2_LINE_SIZE - 1) | \
> +                   ((L2_PARTITIONS - 1) << 12) | \
> +                   ((L2_ASSOCIATIVITY_AMD - 1) << 22);
> +            *ecx = L2_SETS_AMD - 1;
> +            *edx = CACHE_INCLUSIVE;
> +            break;

This adds:
CPUID[0x8000001D]: 64-byte lines, 8-way, 1 partition, 1024 sets (512 KiB).

This doesn't match any of the existing leaves.


>  /* Level 3 unified cache: */
>  #define L3_N_LINE_SIZE         64
>  #define L3_N_ASSOCIATIVITY     16
>  #define L3_N_SETS           16384
> +#define L3_N_SETS_AMD        8192
>  #define L3_N_PARTITIONS         1
>  #define L3_N_DESCRIPTOR CPUID_2_L3_16MB_16WAY_64B
>  #define L3_N_LINES_PER_TAG      1
>  #define L3_N_SIZE_KB_AMD    16384
>  

Currently we have:
CPUID[2]: 16MiB 16-way cache, 64-byte lines
CPUID[4]: 64-byte lines, 16-way, 16384 sets, 1 partition (16 MiB)
CPUID[0x80000006]: 16 MiB cache, 16-way, 1 line per tag


> +        case 3: /* L3 cache info */
> +            if (cpu->enable_l3_cache) {
> +                *eax |= TYPE_UNIFIED | \
> +                        CACHE_LEVEL(3) | \
> +                        CACHE_SELF_INIT_LEVEL | \
> +                        ((cs->nr_cores * cs->nr_threads - 1) << 14);
> +                *ebx = (L3_N_LINE_SIZE - 1) | \
> +                       ((L3_N_PARTITIONS - 1) << 12) | \
> +                       ((L3_N_ASSOCIATIVITY - 1) << 22);
> +                *ecx = L3_N_SETS_AMD - 1;
> +                *edx = CACHE_NO_INVD_SHARING;

This adds:
CPUID[0x8000001D]: 64-byte lines, 16-way, 1 partition, 8192 sets (8 MiB)

This doesn't match any of the existing leaves.


> +            } else {
> +                *eax = 0;
> +                *ebx = 0;
> +                *ecx = 0;
> +                *edx = 0;
> +            }
> +            break;
> +        default: /* end of info */
> +            *eax = 0;
> +            *ebx = 0;
> +            *ecx = 0;
> +            *edx = 0;
> +            break;
> +        }
> +        break;
>      case 0xC0000000:
>          *eax = env->cpuid_xlevel2;
>          *ebx = 0;
[...]

-- 
Eduardo

  parent reply	other threads:[~2018-03-16 18:00 UTC|newest]

Thread overview: 42+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-03-12 21:00 [PATCH v4 0/5] Enable TOPOEXT to support hyperthreading on AMD CPU Babu Moger
2018-03-12 21:00 ` [Qemu-devel] " Babu Moger
2018-03-12 21:00 ` [PATCH v4 1/5] target/i386: Generalize some of the macro definitions Babu Moger
2018-03-12 21:00   ` [Qemu-devel] " Babu Moger
2018-03-15 19:07   ` Eduardo Habkost
2018-03-15 19:07     ` [Qemu-devel] " Eduardo Habkost
2018-03-12 21:00 ` [PATCH v4 2/5] target/i386: Populate AMD Processor Cache Information Babu Moger
2018-03-12 21:00   ` [Qemu-devel] " Babu Moger
2018-03-15 19:04   ` Eduardo Habkost
2018-03-15 19:04     ` [Qemu-devel] " Eduardo Habkost
2018-03-16 18:00   ` Eduardo Habkost [this message]
2018-03-16 18:00     ` Eduardo Habkost
2018-03-20 17:25     ` Moger, Babu
2018-03-20 17:25       ` [Qemu-devel] " Moger, Babu
2018-03-20 17:54       ` Eduardo Habkost
2018-03-20 17:54         ` [Qemu-devel] " Eduardo Habkost
2018-03-20 19:20         ` Moger, Babu
2018-03-20 19:20           ` [Qemu-devel] " Moger, Babu
2018-03-21 15:58         ` Moger, Babu
2018-03-21 15:58           ` [Qemu-devel] " Moger, Babu
2018-03-21 17:09           ` Eduardo Habkost
2018-03-21 17:09             ` [Qemu-devel] " Eduardo Habkost
2018-03-21 17:12             ` Kash Pande
2018-03-21 17:12               ` [Qemu-devel] " Kash Pande
2018-03-21 17:47             ` Moger, Babu
2018-03-21 17:47               ` [Qemu-devel] " Moger, Babu
2018-03-21 18:15               ` Eduardo Habkost
2018-03-21 18:15                 ` [Qemu-devel] " Eduardo Habkost
2018-03-21 20:07                 ` Moger, Babu
2018-03-21 20:07                   ` [Qemu-devel] " Moger, Babu
2018-03-21 20:29                   ` Eduardo Habkost
2018-03-21 20:29                     ` [Qemu-devel] " Eduardo Habkost
2018-03-27 21:36                     ` Moger, Babu
2018-03-27 21:36                       ` [Qemu-devel] " Moger, Babu
2018-03-12 21:00 ` [PATCH v4 3/5] target/i386: Add support for CPUID_8000_001E for AMD Babu Moger
2018-03-12 21:00   ` [Qemu-devel] " Babu Moger
2018-03-12 21:00 ` [PATCH v4 4/5] target/i386: Enable TOPOEXT feature on AMD EPYC CPU Babu Moger
2018-03-12 21:00   ` [Qemu-devel] " Babu Moger
2018-03-12 21:00 ` [PATCH v4 5/5] target/i386: Remove generic SMT thread check Babu Moger
2018-03-12 21:00   ` [Qemu-devel] " Babu Moger
2018-03-13 21:39 ` [PATCH v4 0/5] Enable TOPOEXT to support hyperthreading on AMD CPU Kash Pande
2018-03-13 21:39   ` [Qemu-devel] " Kash Pande

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20180316180006.GH28578@localhost.localdomain \
    --to=ehabkost@redhat.com \
    --cc=Gary.Hook@amd.com \
    --cc=Thomas.Lendacky@amd.com \
    --cc=babu.moger@amd.com \
    --cc=brijesh.singh@amd.com \
    --cc=kash@tripleback.net \
    --cc=kvm@vger.kernel.org \
    --cc=mtosatti@redhat.com \
    --cc=pbonzini@redhat.com \
    --cc=qemu-devel@nongnu.org \
    --cc=rkrcmar@redhat.com \
    --cc=rth@twiddle.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.