Re: [PATCH v1 6/9] Add CPUID enumeration for RDT

qemu-devel

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH v1 6/9] Add CPUID enumeration for RDT

From:	Jonathan Cameron
Subject:	Re: [PATCH v1 6/9] Add CPUID enumeration for RDT
Date:	Fri, 26 Jul 2024 13:09:19 +0100

On Fri, 19 Jul 2024 16:29:26 +0000
Hendrik Wuethrich <whendrik@google.com> wrote:

> From: ‪Hendrik Wüthrich <whendrik@google.com>
> 
> Add CPUID enumeration for intel RDT monitoring and allocation, as well
> as the flags used in the enumeration code.
> 
> Signed-off-by: Hendrik Wüthrich <whendrik@google.com>
> ---
>  hw/i386/rdt.c         | 29 ++++++++++++++
>  include/hw/i386/rdt.h | 29 ++++++++++++++
>  target/i386/cpu.c     | 91 +++++++++++++++++++++++++++++++++++++++++++
>  target/i386/cpu.h     |  5 +++
>  4 files changed, 154 insertions(+)
> 
> diff --git a/hw/i386/rdt.c b/hw/i386/rdt.c
> index 0d0e5751fc..5ad05f996a 100644
> --- a/hw/i386/rdt.c
> +++ b/hw/i386/rdt.c
> @@ -17,8 +17,18 @@
>  #define MAX_L2_MASK_COUNT      48
>  #define MAX_MBA_THRTL_COUNT    31
>  
> +/* RDT L3 Allocation features */
> +#define CPUID_10_1_EAX_CBM_LENGTH       0xf
> +#define CPUID_10_1_EBX_CBM              0x0
> +#define CPUID_10_1_ECX_CDP              0x0 // to enable, it would be (1U << 
> 2)
>  #define CPUID_10_1_EDX_COS_MAX          MAX_L3_MASK_COUNT
> +/* RDT L2 Allocation features*/
> +#define CPUID_10_2_EAX_CBM_LENGTH       0xf
> +#define CPUID_10_2_EBX_CBM              0x0
>  #define CPUID_10_2_EDX_COS_MAX          MAX_L2_MASK_COUNT
> +/* RDT MBA features */
> +#define CPUID_10_3_EAX_THRTL_MAX        89
> +#define CPUID_10_3_ECX_LINEAR_RESPONSE (1U << 2)
>  #define CPUID_10_3_EDX_COS_MAX          MAX_MBA_THRTL_COUNT
>  
>  #define TYPE_RDT "rdt"
> @@ -61,8 +71,27 @@ struct RDTState {
>  
>  struct RDTStateClass { };
>  
> +uint32_t rdt_get_cpuid_15_0_edx_l3(void) { return 
> CPUID_15_1_EDX_L3_OCCUPANCY | CPUID_15_1_EDX_L3_TOTAL_BW | 
> CPUID_15_1_EDX_L3_LOCAL_BW; }
> +
> +uint32_t rdt_cpuid_15_1_edx_l3_total_bw_enabled(void) { return 
> CPUID_15_1_EDX_L3_TOTAL_BW; }
> +uint32_t rdt_cpuid_15_1_edx_l3_local_bw_enabled(void) { return 
> CPUID_15_1_EDX_L3_LOCAL_BW; }
> +uint32_t rdt_cpuid_15_1_edx_l3_occupancy_enabled(void) { return 
> CPUID_15_1_EDX_L3_OCCUPANCY; }
> +
> +uint32_t rdt_cpuid_10_0_ebx_l3_cat_enabled(void) { return 
> CPUID_10_0_EBX_L3_CAT; }
> +uint32_t rdt_cpuid_10_0_ebx_l2_cat_enabled(void) { return 
> CPUID_10_0_EBX_L2_CAT; }
> +uint32_t rdt_cpuid_10_0_ebx_l2_mba_enabled(void) { return 
> CPUID_10_0_EBX_MBA; }
> +
> +uint32_t rdt_get_cpuid_10_1_eax_cbm_length(void) { return 
> CPUID_10_1_EAX_CBM_LENGTH; }
> +uint32_t rdt_cpuid_10_1_ebx_cbm_enabled(void) { return CPUID_10_1_EBX_CBM; }
> +uint32_t rdt_cpuid_10_1_ecx_cdp_enabled(void) { return CPUID_10_1_ECX_CDP; }
>  uint32_t rdt_get_cpuid_10_1_edx_cos_max(void) { return 
> CPUID_10_1_EDX_COS_MAX; }
> +
> +uint32_t rdt_get_cpuid_10_2_eax_cbm_length(void) { return 
> CPUID_10_2_EAX_CBM_LENGTH; }
> +uint32_t rdt_cpuid_10_2_ebx_cbm_enabled(void) { return CPUID_10_2_EBX_CBM; }
>  uint32_t rdt_get_cpuid_10_2_edx_cos_max(void) { return 
> CPUID_10_2_EDX_COS_MAX; }
> +
> +uint32_t rdt_get_cpuid_10_3_eax_thrtl_max(void) { return 
> CPUID_10_3_EAX_THRTL_MAX; }
> +uint32_t rdt_cpuid_10_3_eax_linear_response_enabled(void) { return 
> CPUID_10_3_ECX_LINEAR_RESPONSE; }
>  uint32_t rdt_get_cpuid_10_3_edx_cos_max(void) { return 
> CPUID_10_3_EDX_COS_MAX; }
>  
>  bool rdt_associate_rmid_cos(uint64_t msr_ia32_pqr_assoc) {
> diff --git a/include/hw/i386/rdt.h b/include/hw/i386/rdt.h
> index 51d36822f0..74aba33995 100644
> --- a/include/hw/i386/rdt.h
> +++ b/include/hw/i386/rdt.h
> @@ -4,15 +4,44 @@
>  #include <stdbool.h>
>  #include <stdint.h>
>  
> +/* RDT L3 Cache Monitoring Technology */
> +#define CPUID_15_0_EDX_L3               (1U << 1)
> +#define CPUID_15_1_EDX_L3_OCCUPANCY     (1U << 0)
> +#define CPUID_15_1_EDX_L3_TOTAL_BW      (1U << 1)
> +#define CPUID_15_1_EDX_L3_LOCAL_BW      (1U << 2)
> +
> +/* RDT Cache Allocation Technology */
> +#define CPUID_10_0_EBX_L3_CAT           (1U << 1)
> +#define CPUID_10_0_EBX_L2_CAT           (1U << 2)
> +#define CPUID_10_0_EBX_MBA              (1U << 3)
> +#define CPUID_10_0_EDX CPUID_10_0_EBX_L3_CAT | CPUID_10_0_EBX_L2_CAT | 
> CPUID_10_0_EBX_MBA
> +
>  typedef struct RDTState RDTState;
>  typedef struct RDTStateInstance RDTStateInstance;
>  typedef struct RDTMonitor RDTMonitor;
>  typedef struct RDTAllocation RDTAllocation;
>  
> +uint32_t rdt_get_cpuid_15_0_edx_l3(void);
> +
> +uint32_t rdt_cpuid_15_1_edx_l3_total_bw_enabled(void);
> +uint32_t rdt_cpuid_15_1_edx_l3_local_bw_enabled(void);
> +uint32_t rdt_cpuid_15_1_edx_l3_occupancy_enabled(void);
> +
> +uint32_t rdt_cpuid_10_0_ebx_l3_cat_enabled(void);
> +uint32_t rdt_cpuid_10_0_ebx_l2_cat_enabled(void);
> +uint32_t rdt_cpuid_10_0_ebx_l2_mba_enabled(void);
> +
> +uint32_t rdt_get_cpuid_10_1_eax_cbm_length(void);
> +uint32_t rdt_cpuid_10_1_ebx_cbm_enabled(void);
> +uint32_t rdt_cpuid_10_1_ecx_cdp_enabled(void);
>  uint32_t rdt_get_cpuid_10_1_edx_cos_max(void);
>  
> +uint32_t rdt_get_cpuid_10_2_eax_cbm_length(void);
> +uint32_t rdt_cpuid_10_2_ebx_cbm_enabled(void);
>  uint32_t rdt_get_cpuid_10_2_edx_cos_max(void);
>  
> +uint32_t rdt_get_cpuid_10_3_eax_thrtl_max(void);
> +uint32_t rdt_cpuid_10_3_eax_linear_response_enabled(void);
>  uint32_t rdt_get_cpuid_10_3_edx_cos_max(void);
>  
>  bool rdt_associate_rmid_cos(uint64_t msr_ia32_pqr_assoc);
> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> index 4688d140c2..c61981bf82 100644
> --- a/target/i386/cpu.c
> +++ b/target/i386/cpu.c
> @@ -42,6 +42,7 @@
>  #include "hw/boards.h"
>  #include "hw/i386/sgx-epc.h"
>  #endif
> +#include "hw/i386/rdt.h"
>  
>  #include "disas/capstone.h"
>  #include "cpu-internal.h"
> @@ -6629,6 +6630,96 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, 
> uint32_t count,
>          assert(!(*eax & ~0x1f));
>          *ebx &= 0xffff; /* The count doesn't need to be reliable. */
>          break;
> +#ifndef CONFIG_USER_ONLY
> +    case 0xF:
> +        /* Shared Resource Monitoring Enumeration Leaf */
> +        *eax = 0;
> +        *ebx = 0;
> +        *ecx = 0;
> +        *edx = 0;
> +        if (!(env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_PQM))
> +            break;
> +        assert(cpu->rdt);
> +        /* Non-zero count is ResId */
> +        switch (count) {
> +            /* Monitoring Resource Type Enumeration */
> +            case 0:
> +                *edx = env->features[FEAT_RDT_15_0_EDX];
> +                *ebx = rdt_max_rmid(cpu->rdt);
> +                break;
> +            /* L3 Cache Monitoring Capability Enumeration Data */
> +            case 1:
> +                /* Upscaling Factor */
> +                *ebx = 1;
> +                /* MaxRMID */
> +                *ecx = rdt_max_rmid(cpu->rdt);
> +                /* Set L3 Total BW */
comments fairly obvious from naming of functions so I'd be tempted
to drop them all and have
                   *ecx = rdt_max_rmid(cpu->rdt);
                   *edx = rdt_cpuid_15_1_edx_l3_total_bw_enabled() |
                          rdt_cpuid_15_1_edx_l3_local_bw_enabled() |
                          rdt_cpuid_15_1_edx_l3_occupancy_enabled();

Same for the rest of the cases
> +                *edx |= rdt_cpuid_15_1_edx_l3_total_bw_enabled();
> +                /* Set L3 Local BW */
> +                *edx |= rdt_cpuid_15_1_edx_l3_local_bw_enabled();
> +                /* Set L3 Occupancy */
> +                *edx |= rdt_cpuid_15_1_edx_l3_occupancy_enabled();
> +                break;
> +            default:
> +                break;
> +        }
> +        break;
> +    case 0x10:
> +        /* Shared Resource Director Technology Allocation Enumeration Leaf */
> +        *eax = 0;
> +        *ebx = 0;
> +        *ecx = 0;
> +        *edx = 0;
> +        if (!(env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_PQE))
> +            break;
> +        assert(cpu->rdt);
> +        /* Non-zero count is ResId */
> +        switch (count) {
> +            /* Cache Allocation Technology Available Resource Types */
> +            case 0:
> +                /* Set L3 CAT */
> +                *ebx |= rdt_cpuid_10_0_ebx_l3_cat_enabled();
> +                /* Set L2 CAT */
> +                *ebx |= rdt_cpuid_10_0_ebx_l2_cat_enabled();
> +                /* Set MBA */
> +                *ebx |= rdt_cpuid_10_0_ebx_l2_mba_enabled();
> +                // *edx = env->features[FEAT_RDT_10_0_EBX];

Make sure to clean up any commented out code for v2.

> +                break;
> +            case 1:
> +                /* Length of capacity bitmask in -1 notation */
> +                *eax = rdt_get_cpuid_10_1_eax_cbm_length();
> +                /* Capability bit mask */
> +                *ebx = rdt_cpuid_10_1_ebx_cbm_enabled();
> +                /* Code and Data priotitization */
> +                *ecx |= rdt_cpuid_10_1_ecx_cdp_enabled();
> +                /* Support for n COS masks (zero-referenced)*/
> +                *edx =  rdt_get_cpuid_10_1_edx_cos_max();
> +                break;
> +            case 2:
> +                /* Length of capacity bitmask in -1 notation */
> +                *eax = rdt_get_cpuid_10_2_eax_cbm_length();
> +                /* Capability bit mask */
> +                *ebx = rdt_cpuid_10_2_ebx_cbm_enabled();
> +                /* Support for n COS masks (zero-referenced)*/
> +                *edx =  rdt_get_cpuid_10_2_edx_cos_max();
> +                break;
> +            case 3:
> +                /* Max throttling value -1 (89 means 90) */
> +                *eax = rdt_get_cpuid_10_3_eax_thrtl_max();
> +                /* Linear response of delay values */
> +                *ecx = rdt_cpuid_10_3_eax_linear_response_enabled();
> +                /* Max number of CLOS -1 (15 means 16) */
> +                *edx = rdt_get_cpuid_10_3_edx_cos_max();
> +                break;
> +            default:
> +                *eax = 0;
> +                *ebx = 0;
> +                *ecx = 0;
> +                *edx = 0;
already done above.
> +                break;
> +        }
> +        break;
> +#endif
>      case 0x1C:
>          if (cpu->enable_pmu && (env->features[FEAT_7_0_EDX] & 
> CPUID_7_0_EDX_ARCH_LBR)) {
>              x86_cpu_get_supported_cpuid(0x1C, 0, eax, ebx, ecx, edx);

[Prev in Thread]

Current Thread

[Next in Thread]

Re: [PATCH v1 5/9] Add RDT device interface through MSRs, (continued)
- [PATCH v1 2/9] Add state for RDT device., Hendrik Wuethrich, 2024/07/19
  - Re: [PATCH v1 2/9] Add state for RDT device., Jonathan Cameron, 2024/07/26
- [PATCH v1 9/9] Adjust level for RDT on full_cpuid_auto_level, Hendrik Wuethrich, 2024/07/19
- [PATCH v1 4/9] Add RDT functionality, Hendrik Wuethrich, 2024/07/19
  - Re: [PATCH v1 4/9] Add RDT functionality, Jonathan Cameron, 2024/07/26
- [PATCH v1 8/9] Adjust CPUID level for RDT features, Hendrik Wuethrich, 2024/07/19
- [PATCH v1 1/9] Add Intel RDT device to config., Hendrik Wuethrich, 2024/07/19
  - Re: [PATCH v1 1/9] Add Intel RDT device to config., Jonathan Cameron, 2024/07/26
- [PATCH v1 6/9] Add CPUID enumeration for RDT, Hendrik Wuethrich, 2024/07/19
  - Re: [PATCH v1 6/9] Add CPUID enumeration for RDT, Jonathan Cameron <=
- [PATCH v1 3/9] Add init and realize funciontality for RDT device., Hendrik Wuethrich, 2024/07/19
  - Re: [PATCH v1 3/9] Add init and realize funciontality for RDT device., Jonathan Cameron, 2024/07/26
  - Re: [PATCH v1 3/9] Add init and realize funciontality for RDT device., Jonathan Cameron, 2024/07/26
- [PATCH v1 7/9] Add RDT feature flags., Hendrik Wuethrich, 2024/07/19

Prev by Date: Re: [PATCH v2] tests/tcg: Skip failing ppc64 multi-threaded tests
Next by Date: Re: [RFC 1/2] target/riscv: rvv: reduce the overhead for simple RISC-V vector unit-stride loads and stores
Previous by thread: [PATCH v1 6/9] Add CPUID enumeration for RDT
Next by thread: [PATCH v1 3/9] Add init and realize funciontality for RDT device.
Index(es):
- Date
- Thread