diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_crat.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 768 |
1 files changed, 358 insertions, 410 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c index 8bfdfd062ff6..4a7180b46b71 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c @@ -26,9 +26,9 @@ #include "kfd_crat.h" #include "kfd_priv.h" #include "kfd_topology.h" -#include "kfd_iommu.h" #include "amdgpu.h" #include "amdgpu_amdkfd.h" +#include "amdgpu_xgmi.h" /* GPU Processor ID base for dGPUs for which VCRAT needs to be created. * GPU processor ID are expressed with Bit[31]=1. @@ -50,22 +50,13 @@ static inline unsigned int get_and_inc_gpu_processor_id( return current_id; } -/* Static table to describe GPU Cache information */ -struct kfd_gpu_cache_info { - uint32_t cache_size; - uint32_t cache_level; - uint32_t flags; - /* Indicates how many Compute Units share this cache - * within a SA. Value = 1 indicates the cache is not shared - */ - uint32_t num_cu_shared; -}; static struct kfd_gpu_cache_info kaveri_cache_info[] = { { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -75,6 +66,7 @@ static struct kfd_gpu_cache_info kaveri_cache_info[] = { /* Scalar L1 Instruction Cache (in SQC module) per bank */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -84,6 +76,7 @@ static struct kfd_gpu_cache_info kaveri_cache_info[] = { /* Scalar L1 Data Cache (in SQC module) per bank */ .cache_size = 8, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -99,6 +92,7 @@ static struct kfd_gpu_cache_info carrizo_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -106,8 +100,9 @@ static struct kfd_gpu_cache_info carrizo_cache_info[] = { }, { /* Scalar L1 Instruction Cache (in SQC module) per bank */ - .cache_size = 8, + .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -115,8 +110,9 @@ static struct kfd_gpu_cache_info carrizo_cache_info[] = { }, { /* Scalar L1 Data Cache (in SQC module) per bank. */ - .cache_size = 4, + .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -146,6 +142,7 @@ static struct kfd_gpu_cache_info vega10_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -155,6 +152,7 @@ static struct kfd_gpu_cache_info vega10_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -164,6 +162,7 @@ static struct kfd_gpu_cache_info vega10_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -173,6 +172,7 @@ static struct kfd_gpu_cache_info vega10_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 4096, .cache_level = 2, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -185,6 +185,7 @@ static struct kfd_gpu_cache_info raven_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -194,6 +195,7 @@ static struct kfd_gpu_cache_info raven_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -203,6 +205,7 @@ static struct kfd_gpu_cache_info raven_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -212,6 +215,7 @@ static struct kfd_gpu_cache_info raven_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 1024, .cache_level = 2, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -224,6 +228,7 @@ static struct kfd_gpu_cache_info renoir_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -233,6 +238,7 @@ static struct kfd_gpu_cache_info renoir_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -242,6 +248,7 @@ static struct kfd_gpu_cache_info renoir_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -251,6 +258,7 @@ static struct kfd_gpu_cache_info renoir_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 1024, .cache_level = 2, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -263,6 +271,7 @@ static struct kfd_gpu_cache_info vega12_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -272,6 +281,7 @@ static struct kfd_gpu_cache_info vega12_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -281,6 +291,7 @@ static struct kfd_gpu_cache_info vega12_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -290,6 +301,7 @@ static struct kfd_gpu_cache_info vega12_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 2048, .cache_level = 2, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -302,6 +314,7 @@ static struct kfd_gpu_cache_info vega20_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -311,6 +324,7 @@ static struct kfd_gpu_cache_info vega20_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -320,6 +334,7 @@ static struct kfd_gpu_cache_info vega20_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -329,6 +344,7 @@ static struct kfd_gpu_cache_info vega20_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 8192, .cache_level = 2, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -341,6 +357,7 @@ static struct kfd_gpu_cache_info aldebaran_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -350,6 +367,7 @@ static struct kfd_gpu_cache_info aldebaran_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -359,6 +377,7 @@ static struct kfd_gpu_cache_info aldebaran_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -368,6 +387,7 @@ static struct kfd_gpu_cache_info aldebaran_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 8192, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -380,6 +400,7 @@ static struct kfd_gpu_cache_info navi10_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -389,6 +410,7 @@ static struct kfd_gpu_cache_info navi10_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -398,6 +420,7 @@ static struct kfd_gpu_cache_info navi10_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -407,6 +430,7 @@ static struct kfd_gpu_cache_info navi10_cache_info[] = { /* GL1 Data Cache per SA */ .cache_size = 128, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -416,6 +440,7 @@ static struct kfd_gpu_cache_info navi10_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 4096, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -428,6 +453,7 @@ static struct kfd_gpu_cache_info vangogh_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -437,6 +463,7 @@ static struct kfd_gpu_cache_info vangogh_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -446,6 +473,7 @@ static struct kfd_gpu_cache_info vangogh_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -455,6 +483,7 @@ static struct kfd_gpu_cache_info vangogh_cache_info[] = { /* GL1 Data Cache per SA */ .cache_size = 128, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -464,6 +493,7 @@ static struct kfd_gpu_cache_info vangogh_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 1024, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -476,6 +506,7 @@ static struct kfd_gpu_cache_info navi14_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -485,6 +516,7 @@ static struct kfd_gpu_cache_info navi14_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -494,6 +526,7 @@ static struct kfd_gpu_cache_info navi14_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -503,6 +536,7 @@ static struct kfd_gpu_cache_info navi14_cache_info[] = { /* GL1 Data Cache per SA */ .cache_size = 128, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -512,6 +546,7 @@ static struct kfd_gpu_cache_info navi14_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 2048, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -524,6 +559,7 @@ static struct kfd_gpu_cache_info sienna_cichlid_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -533,6 +569,7 @@ static struct kfd_gpu_cache_info sienna_cichlid_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -542,6 +579,7 @@ static struct kfd_gpu_cache_info sienna_cichlid_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -551,6 +589,7 @@ static struct kfd_gpu_cache_info sienna_cichlid_cache_info[] = { /* GL1 Data Cache per SA */ .cache_size = 128, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -560,6 +599,7 @@ static struct kfd_gpu_cache_info sienna_cichlid_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 4096, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -569,6 +609,7 @@ static struct kfd_gpu_cache_info sienna_cichlid_cache_info[] = { /* L3 Data Cache per GPU */ .cache_size = 128*1024, .cache_level = 3, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -581,6 +622,7 @@ static struct kfd_gpu_cache_info navy_flounder_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -590,6 +632,7 @@ static struct kfd_gpu_cache_info navy_flounder_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -599,6 +642,7 @@ static struct kfd_gpu_cache_info navy_flounder_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -608,6 +652,7 @@ static struct kfd_gpu_cache_info navy_flounder_cache_info[] = { /* GL1 Data Cache per SA */ .cache_size = 128, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -617,6 +662,7 @@ static struct kfd_gpu_cache_info navy_flounder_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 3072, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -626,6 +672,7 @@ static struct kfd_gpu_cache_info navy_flounder_cache_info[] = { /* L3 Data Cache per GPU */ .cache_size = 96*1024, .cache_level = 3, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -638,6 +685,7 @@ static struct kfd_gpu_cache_info dimgrey_cavefish_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -647,6 +695,7 @@ static struct kfd_gpu_cache_info dimgrey_cavefish_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -656,6 +705,7 @@ static struct kfd_gpu_cache_info dimgrey_cavefish_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -665,6 +715,7 @@ static struct kfd_gpu_cache_info dimgrey_cavefish_cache_info[] = { /* GL1 Data Cache per SA */ .cache_size = 128, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -674,6 +725,7 @@ static struct kfd_gpu_cache_info dimgrey_cavefish_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 2048, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -683,6 +735,7 @@ static struct kfd_gpu_cache_info dimgrey_cavefish_cache_info[] = { /* L3 Data Cache per GPU */ .cache_size = 32*1024, .cache_level = 3, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -695,6 +748,7 @@ static struct kfd_gpu_cache_info beige_goby_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -704,6 +758,7 @@ static struct kfd_gpu_cache_info beige_goby_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -713,6 +768,7 @@ static struct kfd_gpu_cache_info beige_goby_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -722,6 +778,7 @@ static struct kfd_gpu_cache_info beige_goby_cache_info[] = { /* GL1 Data Cache per SA */ .cache_size = 128, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -731,6 +788,7 @@ static struct kfd_gpu_cache_info beige_goby_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 1024, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -740,6 +798,7 @@ static struct kfd_gpu_cache_info beige_goby_cache_info[] = { /* L3 Data Cache per GPU */ .cache_size = 16*1024, .cache_level = 3, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -752,6 +811,7 @@ static struct kfd_gpu_cache_info yellow_carp_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -761,6 +821,7 @@ static struct kfd_gpu_cache_info yellow_carp_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -770,6 +831,7 @@ static struct kfd_gpu_cache_info yellow_carp_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -779,6 +841,7 @@ static struct kfd_gpu_cache_info yellow_carp_cache_info[] = { /* GL1 Data Cache per SA */ .cache_size = 128, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -788,6 +851,7 @@ static struct kfd_gpu_cache_info yellow_carp_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 2048, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -800,6 +864,7 @@ static struct kfd_gpu_cache_info gfx1037_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -809,6 +874,7 @@ static struct kfd_gpu_cache_info gfx1037_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -818,6 +884,7 @@ static struct kfd_gpu_cache_info gfx1037_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -827,6 +894,7 @@ static struct kfd_gpu_cache_info gfx1037_cache_info[] = { /* GL1 Data Cache per SA */ .cache_size = 128, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -836,6 +904,7 @@ static struct kfd_gpu_cache_info gfx1037_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 256, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -848,6 +917,7 @@ static struct kfd_gpu_cache_info gc_10_3_6_cache_info[] = { /* TCP L1 Cache per CU */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -857,6 +927,7 @@ static struct kfd_gpu_cache_info gc_10_3_6_cache_info[] = { /* Scalar L1 Instruction Cache per SQC */ .cache_size = 32, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -866,6 +937,7 @@ static struct kfd_gpu_cache_info gc_10_3_6_cache_info[] = { /* Scalar L1 Data Cache per SQC */ .cache_size = 16, .cache_level = 1, + .cache_line_size = 64, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -875,6 +947,7 @@ static struct kfd_gpu_cache_info gc_10_3_6_cache_info[] = { /* GL1 Data Cache per SA */ .cache_size = 128, .cache_level = 1, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -884,6 +957,7 @@ static struct kfd_gpu_cache_info gc_10_3_6_cache_info[] = { /* L2 Data Cache per GPU (Total Tex Cache) */ .cache_size = 256, .cache_level = 2, + .cache_line_size = 128, .flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE), @@ -891,6 +965,59 @@ static struct kfd_gpu_cache_info gc_10_3_6_cache_info[] = { }, }; +static struct kfd_gpu_cache_info dummy_cache_info[] = { + { + /* TCP L1 Cache per CU */ + .cache_size = 16, + .cache_level = 1, + .cache_line_size = 64, + .flags = (CRAT_CACHE_FLAGS_ENABLED | + CRAT_CACHE_FLAGS_DATA_CACHE | + CRAT_CACHE_FLAGS_SIMD_CACHE), + .num_cu_shared = 1, + }, + { + /* Scalar L1 Instruction Cache per SQC */ + .cache_size = 32, + .cache_level = 1, + .cache_line_size = 64, + .flags = (CRAT_CACHE_FLAGS_ENABLED | + CRAT_CACHE_FLAGS_INST_CACHE | + CRAT_CACHE_FLAGS_SIMD_CACHE), + .num_cu_shared = 2, + }, + { + /* Scalar L1 Data Cache per SQC */ + .cache_size = 16, + .cache_level = 1, + .cache_line_size = 64, + .flags = (CRAT_CACHE_FLAGS_ENABLED | + CRAT_CACHE_FLAGS_DATA_CACHE | + CRAT_CACHE_FLAGS_SIMD_CACHE), + .num_cu_shared = 2, + }, + { + /* GL1 Data Cache per SA */ + .cache_size = 128, + .cache_level = 1, + .cache_line_size = 64, + .flags = (CRAT_CACHE_FLAGS_ENABLED | + CRAT_CACHE_FLAGS_DATA_CACHE | + CRAT_CACHE_FLAGS_SIMD_CACHE), + .num_cu_shared = 6, + }, + { + /* L2 Data Cache per GPU (Total Tex Cache) */ + .cache_size = 2048, + .cache_level = 2, + .cache_line_size = 64, + .flags = (CRAT_CACHE_FLAGS_ENABLED | + CRAT_CACHE_FLAGS_DATA_CACHE | + CRAT_CACHE_FLAGS_SIMD_CACHE), + .num_cu_shared = 6, + }, +}; + static void kfd_populated_cu_info_cpu(struct kfd_topology_device *dev, struct crat_subtype_computeunit *cu) { @@ -1071,8 +1198,12 @@ static int kfd_parse_subtype_cache(struct crat_subtype_cache *cache, props->cachelines_per_tag = cache->lines_per_tag; props->cache_assoc = cache->associativity; props->cache_latency = cache->cache_latency; + memcpy(props->sibling_map, cache->sibling_map, - sizeof(props->sibling_map)); + CRAT_SIBLINGMAP_SIZE); + + /* set the sibling_map_size as 32 for CRAT from ACPI */ + props->sibling_map_size = CRAT_SIBLINGMAP_SIZE; if (cache->flags & CRAT_CACHE_FLAGS_DATA_CACHE) props->cache_type |= HSA_CACHE_TYPE_DATA; @@ -1083,7 +1214,6 @@ static int kfd_parse_subtype_cache(struct crat_subtype_cache *cache, if (cache->flags & CRAT_CACHE_FLAGS_SIMD_CACHE) props->cache_type |= HSA_CACHE_TYPE_HSACU; - dev->cache_count++; dev->node_props.caches_count++; list_add_tail(&props->list, &dev->cache_props); @@ -1125,7 +1255,7 @@ static int kfd_parse_subtype_iolink(struct crat_subtype_iolink *iolink, if (props->iolink_type == CRAT_IOLINK_TYPE_PCIEXPRESS) props->weight = 20; else if (props->iolink_type == CRAT_IOLINK_TYPE_XGMI) - props->weight = 15 * iolink->num_hops_xgmi; + props->weight = iolink->weight_xgmi; else props->weight = node_distance(id_from, id_to); @@ -1291,127 +1421,9 @@ err: return ret; } -/* Helper function. See kfd_fill_gpu_cache_info for parameter description */ -static int fill_in_l1_pcache(struct crat_subtype_cache *pcache, - struct kfd_gpu_cache_info *pcache_info, - struct kfd_cu_info *cu_info, - int mem_available, - int cu_bitmask, - int cache_type, unsigned int cu_processor_id, - int cu_block) -{ - unsigned int cu_sibling_map_mask; - int first_active_cu; - - /* First check if enough memory is available */ - if (sizeof(struct crat_subtype_cache) > mem_available) - return -ENOMEM; - - cu_sibling_map_mask = cu_bitmask; - cu_sibling_map_mask >>= cu_block; - cu_sibling_map_mask &= - ((1 << pcache_info[cache_type].num_cu_shared) - 1); - first_active_cu = ffs(cu_sibling_map_mask); - - /* CU could be inactive. In case of shared cache find the first active - * CU. and incase of non-shared cache check if the CU is inactive. If - * inactive active skip it - */ - if (first_active_cu) { - memset(pcache, 0, sizeof(struct crat_subtype_cache)); - pcache->type = CRAT_SUBTYPE_CACHE_AFFINITY; - pcache->length = sizeof(struct crat_subtype_cache); - pcache->flags = pcache_info[cache_type].flags; - pcache->processor_id_low = cu_processor_id - + (first_active_cu - 1); - pcache->cache_level = pcache_info[cache_type].cache_level; - pcache->cache_size = pcache_info[cache_type].cache_size; - - /* Sibling map is w.r.t processor_id_low, so shift out - * inactive CU - */ - cu_sibling_map_mask = - cu_sibling_map_mask >> (first_active_cu - 1); - - pcache->sibling_map[0] = (uint8_t)(cu_sibling_map_mask & 0xFF); - pcache->sibling_map[1] = - (uint8_t)((cu_sibling_map_mask >> 8) & 0xFF); - pcache->sibling_map[2] = - (uint8_t)((cu_sibling_map_mask >> 16) & 0xFF); - pcache->sibling_map[3] = - (uint8_t)((cu_sibling_map_mask >> 24) & 0xFF); - return 0; - } - return 1; -} - -/* Helper function. See kfd_fill_gpu_cache_info for parameter description */ -static int fill_in_l2_l3_pcache(struct crat_subtype_cache *pcache, - struct kfd_gpu_cache_info *pcache_info, - struct kfd_cu_info *cu_info, - int mem_available, - int cache_type, unsigned int cu_processor_id) -{ - unsigned int cu_sibling_map_mask; - int first_active_cu; - int i, j, k; - - /* First check if enough memory is available */ - if (sizeof(struct crat_subtype_cache) > mem_available) - return -ENOMEM; - - cu_sibling_map_mask = cu_info->cu_bitmap[0][0]; - cu_sibling_map_mask &= - ((1 << pcache_info[cache_type].num_cu_shared) - 1); - first_active_cu = ffs(cu_sibling_map_mask); - - /* CU could be inactive. In case of shared cache find the first active - * CU. and incase of non-shared cache check if the CU is inactive. If - * inactive active skip it - */ - if (first_active_cu) { - memset(pcache, 0, sizeof(struct crat_subtype_cache)); - pcache->type = CRAT_SUBTYPE_CACHE_AFFINITY; - pcache->length = sizeof(struct crat_subtype_cache); - pcache->flags = pcache_info[cache_type].flags; - pcache->processor_id_low = cu_processor_id - + (first_active_cu - 1); - pcache->cache_level = pcache_info[cache_type].cache_level; - pcache->cache_size = pcache_info[cache_type].cache_size; - - /* Sibling map is w.r.t processor_id_low, so shift out - * inactive CU - */ - cu_sibling_map_mask = - cu_sibling_map_mask >> (first_active_cu - 1); - k = 0; - for (i = 0; i < cu_info->num_shader_engines; i++) { - for (j = 0; j < cu_info->num_shader_arrays_per_engine; - j++) { - pcache->sibling_map[k] = - (uint8_t)(cu_sibling_map_mask & 0xFF); - pcache->sibling_map[k+1] = - (uint8_t)((cu_sibling_map_mask >> 8) & 0xFF); - pcache->sibling_map[k+2] = - (uint8_t)((cu_sibling_map_mask >> 16) & 0xFF); - pcache->sibling_map[k+3] = - (uint8_t)((cu_sibling_map_mask >> 24) & 0xFF); - k += 4; - cu_sibling_map_mask = - cu_info->cu_bitmap[i % 4][j + i / 4]; - cu_sibling_map_mask &= ( - (1 << pcache_info[cache_type].num_cu_shared) - - 1); - } - } - return 0; - } - return 1; -} - -#define KFD_MAX_CACHE_TYPES 6 static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, + bool cache_line_size_missing, struct kfd_gpu_cache_info *pcache_info) { struct amdgpu_device *adev = kdev->adev; @@ -1424,7 +1436,10 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); - pcache_info[0].num_cu_shared = adev->gfx.config.gc_num_tcp_per_wpg / 2; + pcache_info[i].num_cu_shared = adev->gfx.config.gc_num_tcp_per_wpg / 2; + pcache_info[i].cache_line_size = adev->gfx.config.gc_tcp_cache_line_size; + if (cache_line_size_missing && !pcache_info[i].cache_line_size) + pcache_info[i].cache_line_size = 128; i++; } /* Scalar L1 Instruction Cache per SQC */ @@ -1436,6 +1451,9 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); pcache_info[i].num_cu_shared = adev->gfx.config.gc_num_sqc_per_wgp * 2; + pcache_info[i].cache_line_size = adev->gfx.config.gc_instruction_cache_line_size; + if (cache_line_size_missing && !pcache_info[i].cache_line_size) + pcache_info[i].cache_line_size = 128; i++; } /* Scalar L1 Data Cache per SQC */ @@ -1446,6 +1464,9 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); pcache_info[i].num_cu_shared = adev->gfx.config.gc_num_sqc_per_wgp * 2; + pcache_info[i].cache_line_size = adev->gfx.config.gc_scalar_data_cache_line_size; + if (cache_line_size_missing && !pcache_info[i].cache_line_size) + pcache_info[i].cache_line_size = 64; i++; } /* GL1 Data Cache per SA */ @@ -1458,6 +1479,8 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); pcache_info[i].num_cu_shared = adev->gfx.config.max_cu_per_sh; + if (cache_line_size_missing) + pcache_info[i].cache_line_size = 128; i++; } /* L2 Data Cache per GPU (Total Tex Cache) */ @@ -1468,6 +1491,9 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); pcache_info[i].num_cu_shared = adev->gfx.config.max_cu_per_sh; + pcache_info[i].cache_line_size = adev->gfx.config.gc_tcc_cache_line_size; + if (cache_line_size_missing && !pcache_info[i].cache_line_size) + pcache_info[i].cache_line_size = 128; i++; } /* L3 Data Cache per GPU */ @@ -1478,303 +1504,227 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); pcache_info[i].num_cu_shared = adev->gfx.config.max_cu_per_sh; + pcache_info[i].cache_line_size = 64; i++; } return i; } -/* kfd_fill_gpu_cache_info - Fill GPU cache info using kfd_gpu_cache_info - * tables - * - * @kdev - [IN] GPU device - * @gpu_processor_id - [IN] GPU processor ID to which these caches - * associate - * @available_size - [IN] Amount of memory available in pcache - * @cu_info - [IN] Compute Unit info obtained from KGD - * @pcache - [OUT] memory into which cache data is to be filled in. - * @size_filled - [OUT] amount of data used up in pcache. - * @num_of_entries - [OUT] number of caches added - */ -static int kfd_fill_gpu_cache_info(struct kfd_dev *kdev, - int gpu_processor_id, - int available_size, - struct kfd_cu_info *cu_info, - struct crat_subtype_cache *pcache, - int *size_filled, - int *num_of_entries) +static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, + struct kfd_gpu_cache_info *pcache_info) +{ + struct amdgpu_device *adev = kdev->adev; + int i = 0; + + /* TCP L1 Cache per CU */ + if (adev->gfx.config.gc_tcp_size_per_cu) { + pcache_info[i].cache_size = adev->gfx.config.gc_tcp_size_per_cu; + pcache_info[i].cache_level = 1; + /* Cacheline size not available in IP discovery for gc943,gc944 */ + pcache_info[i].cache_line_size = 128; + pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | + CRAT_CACHE_FLAGS_DATA_CACHE | + CRAT_CACHE_FLAGS_SIMD_CACHE); + pcache_info[i].num_cu_shared = 1; + i++; + } + /* Scalar L1 Instruction Cache per SQC */ + if (adev->gfx.config.gc_l1_instruction_cache_size_per_sqc) { + pcache_info[i].cache_size = + adev->gfx.config.gc_l1_instruction_cache_size_per_sqc; + pcache_info[i].cache_level = 1; + pcache_info[i].cache_line_size = 64; + pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | + CRAT_CACHE_FLAGS_INST_CACHE | + CRAT_CACHE_FLAGS_SIMD_CACHE); + pcache_info[i].num_cu_shared = adev->gfx.config.gc_num_cu_per_sqc; + i++; + } + /* Scalar L1 Data Cache per SQC */ + if (adev->gfx.config.gc_l1_data_cache_size_per_sqc) { + pcache_info[i].cache_size = adev->gfx.config.gc_l1_data_cache_size_per_sqc; + pcache_info[i].cache_level = 1; + pcache_info[i].cache_line_size = 64; + pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | + CRAT_CACHE_FLAGS_DATA_CACHE | + CRAT_CACHE_FLAGS_SIMD_CACHE); + pcache_info[i].num_cu_shared = adev->gfx.config.gc_num_cu_per_sqc; + i++; + } + /* L2 Data Cache per GPU (Total Tex Cache) */ + if (adev->gfx.config.gc_tcc_size) { + pcache_info[i].cache_size = adev->gfx.config.gc_tcc_size; + pcache_info[i].cache_level = 2; + pcache_info[i].cache_line_size = 128; + pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | + CRAT_CACHE_FLAGS_DATA_CACHE | + CRAT_CACHE_FLAGS_SIMD_CACHE); + pcache_info[i].num_cu_shared = adev->gfx.config.max_cu_per_sh; + i++; + } + /* L3 Data Cache per GPU */ + if (adev->gmc.mall_size) { + pcache_info[i].cache_size = adev->gmc.mall_size / 1024; + pcache_info[i].cache_level = 3; + pcache_info[i].cache_line_size = 64; + pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | + CRAT_CACHE_FLAGS_DATA_CACHE | + CRAT_CACHE_FLAGS_SIMD_CACHE); + pcache_info[i].num_cu_shared = adev->gfx.config.max_cu_per_sh; + i++; + } + return i; +} + +int kfd_get_gpu_cache_info(struct kfd_node *kdev, struct kfd_gpu_cache_info **pcache_info) { - struct kfd_gpu_cache_info *pcache_info; - struct kfd_gpu_cache_info cache_info[KFD_MAX_CACHE_TYPES]; int num_of_cache_types = 0; - int i, j, k; - int ct = 0; - int mem_available = available_size; - unsigned int cu_processor_id; - int ret; - unsigned int num_cu_shared; + bool cache_line_size_missing = false; switch (kdev->adev->asic_type) { case CHIP_KAVERI: - pcache_info = kaveri_cache_info; + *pcache_info = kaveri_cache_info; num_of_cache_types = ARRAY_SIZE(kaveri_cache_info); break; case CHIP_HAWAII: - pcache_info = hawaii_cache_info; + *pcache_info = hawaii_cache_info; num_of_cache_types = ARRAY_SIZE(hawaii_cache_info); break; case CHIP_CARRIZO: - pcache_info = carrizo_cache_info; + *pcache_info = carrizo_cache_info; num_of_cache_types = ARRAY_SIZE(carrizo_cache_info); break; case CHIP_TONGA: - pcache_info = tonga_cache_info; + *pcache_info = tonga_cache_info; num_of_cache_types = ARRAY_SIZE(tonga_cache_info); break; case CHIP_FIJI: - pcache_info = fiji_cache_info; + *pcache_info = fiji_cache_info; num_of_cache_types = ARRAY_SIZE(fiji_cache_info); break; case CHIP_POLARIS10: - pcache_info = polaris10_cache_info; + *pcache_info = polaris10_cache_info; num_of_cache_types = ARRAY_SIZE(polaris10_cache_info); break; case CHIP_POLARIS11: - pcache_info = polaris11_cache_info; + *pcache_info = polaris11_cache_info; num_of_cache_types = ARRAY_SIZE(polaris11_cache_info); break; case CHIP_POLARIS12: - pcache_info = polaris12_cache_info; + *pcache_info = polaris12_cache_info; num_of_cache_types = ARRAY_SIZE(polaris12_cache_info); break; case CHIP_VEGAM: - pcache_info = vegam_cache_info; + *pcache_info = vegam_cache_info; num_of_cache_types = ARRAY_SIZE(vegam_cache_info); break; default: switch (KFD_GC_VERSION(kdev)) { case IP_VERSION(9, 0, 1): - pcache_info = vega10_cache_info; + *pcache_info = vega10_cache_info; num_of_cache_types = ARRAY_SIZE(vega10_cache_info); break; case IP_VERSION(9, 2, 1): - pcache_info = vega12_cache_info; + *pcache_info = vega12_cache_info; num_of_cache_types = ARRAY_SIZE(vega12_cache_info); break; case IP_VERSION(9, 4, 0): case IP_VERSION(9, 4, 1): - pcache_info = vega20_cache_info; + *pcache_info = vega20_cache_info; num_of_cache_types = ARRAY_SIZE(vega20_cache_info); break; case IP_VERSION(9, 4, 2): - pcache_info = aldebaran_cache_info; + *pcache_info = aldebaran_cache_info; num_of_cache_types = ARRAY_SIZE(aldebaran_cache_info); break; + case IP_VERSION(9, 4, 3): + case IP_VERSION(9, 4, 4): + case IP_VERSION(9, 5, 0): + num_of_cache_types = + kfd_fill_gpu_cache_info_from_gfx_config_v2(kdev->kfd, + *pcache_info); + break; case IP_VERSION(9, 1, 0): case IP_VERSION(9, 2, 2): - pcache_info = raven_cache_info; + *pcache_info = raven_cache_info; num_of_cache_types = ARRAY_SIZE(raven_cache_info); break; case IP_VERSION(9, 3, 0): - pcache_info = renoir_cache_info; + *pcache_info = renoir_cache_info; num_of_cache_types = ARRAY_SIZE(renoir_cache_info); break; case IP_VERSION(10, 1, 10): case IP_VERSION(10, 1, 2): case IP_VERSION(10, 1, 3): case IP_VERSION(10, 1, 4): - pcache_info = navi10_cache_info; + *pcache_info = navi10_cache_info; num_of_cache_types = ARRAY_SIZE(navi10_cache_info); break; case IP_VERSION(10, 1, 1): - pcache_info = navi14_cache_info; + *pcache_info = navi14_cache_info; num_of_cache_types = ARRAY_SIZE(navi14_cache_info); break; case IP_VERSION(10, 3, 0): - pcache_info = sienna_cichlid_cache_info; + *pcache_info = sienna_cichlid_cache_info; num_of_cache_types = ARRAY_SIZE(sienna_cichlid_cache_info); break; case IP_VERSION(10, 3, 2): - pcache_info = navy_flounder_cache_info; + *pcache_info = navy_flounder_cache_info; num_of_cache_types = ARRAY_SIZE(navy_flounder_cache_info); break; case IP_VERSION(10, 3, 4): - pcache_info = dimgrey_cavefish_cache_info; + *pcache_info = dimgrey_cavefish_cache_info; num_of_cache_types = ARRAY_SIZE(dimgrey_cavefish_cache_info); break; case IP_VERSION(10, 3, 1): - pcache_info = vangogh_cache_info; + *pcache_info = vangogh_cache_info; num_of_cache_types = ARRAY_SIZE(vangogh_cache_info); break; case IP_VERSION(10, 3, 5): - pcache_info = beige_goby_cache_info; + *pcache_info = beige_goby_cache_info; num_of_cache_types = ARRAY_SIZE(beige_goby_cache_info); break; case IP_VERSION(10, 3, 3): - pcache_info = yellow_carp_cache_info; + *pcache_info = yellow_carp_cache_info; num_of_cache_types = ARRAY_SIZE(yellow_carp_cache_info); break; case IP_VERSION(10, 3, 6): - pcache_info = gc_10_3_6_cache_info; + *pcache_info = gc_10_3_6_cache_info; num_of_cache_types = ARRAY_SIZE(gc_10_3_6_cache_info); break; case IP_VERSION(10, 3, 7): - pcache_info = gfx1037_cache_info; + *pcache_info = gfx1037_cache_info; num_of_cache_types = ARRAY_SIZE(gfx1037_cache_info); break; case IP_VERSION(11, 0, 0): case IP_VERSION(11, 0, 1): case IP_VERSION(11, 0, 2): case IP_VERSION(11, 0, 3): - pcache_info = cache_info; + case IP_VERSION(11, 0, 4): + case IP_VERSION(11, 5, 0): + case IP_VERSION(11, 5, 1): + case IP_VERSION(11, 5, 2): + case IP_VERSION(11, 5, 3): + /* Cacheline size not available in IP discovery for gc11. + * kfd_fill_gpu_cache_info_from_gfx_config to hard code it + */ + cache_line_size_missing = true; + fallthrough; + case IP_VERSION(12, 0, 0): + case IP_VERSION(12, 0, 1): num_of_cache_types = - kfd_fill_gpu_cache_info_from_gfx_config(kdev, pcache_info); + kfd_fill_gpu_cache_info_from_gfx_config(kdev->kfd, + cache_line_size_missing, + *pcache_info); break; default: - return -EINVAL; - } - } - - *size_filled = 0; - *num_of_entries = 0; - - /* For each type of cache listed in the kfd_gpu_cache_info table, - * go through all available Compute Units. - * The [i,j,k] loop will - * if kfd_gpu_cache_info.num_cu_shared = 1 - * will parse through all available CU - * If (kfd_gpu_cache_info.num_cu_shared != 1) - * then it will consider only one CU from - * the shared unit - */ - - for (ct = 0; ct < num_of_cache_types; ct++) { - cu_processor_id = gpu_processor_id; - if (pcache_info[ct].cache_level == 1) { - for (i = 0; i < cu_info->num_shader_engines; i++) { - for (j = 0; j < cu_info->num_shader_arrays_per_engine; j++) { - for (k = 0; k < cu_info->num_cu_per_sh; - k += pcache_info[ct].num_cu_shared) { - ret = fill_in_l1_pcache(pcache, - pcache_info, - cu_info, - mem_available, - cu_info->cu_bitmap[i % 4][j + i / 4], - ct, - cu_processor_id, - k); - - if (ret < 0) + *pcache_info = dummy_cache_info; + num_of_cache_types = ARRAY_SIZE(dummy_cache_info); + pr_warn("dummy cache info is used temporarily and real cache info need update later.\n"); break; - - if (!ret) { - pcache++; - (*num_of_entries)++; - mem_available -= sizeof(*pcache); - (*size_filled) += sizeof(*pcache); - } - - /* Move to next CU block */ - num_cu_shared = ((k + pcache_info[ct].num_cu_shared) <= - cu_info->num_cu_per_sh) ? - pcache_info[ct].num_cu_shared : - (cu_info->num_cu_per_sh - k); - cu_processor_id += num_cu_shared; } - } - } - } else { - ret = fill_in_l2_l3_pcache(pcache, - pcache_info, - cu_info, - mem_available, - ct, - cu_processor_id); - - if (ret < 0) - break; - - if (!ret) { - pcache++; - (*num_of_entries)++; - mem_available -= sizeof(*pcache); - (*size_filled) += sizeof(*pcache); - } - } - } - - pr_debug("Added [%d] GPU cache entries\n", *num_of_entries); - - return 0; -} - -static bool kfd_ignore_crat(void) -{ - bool ret; - - if (ignore_crat) - return true; - -#ifndef KFD_SUPPORT_IOMMU_V2 - ret = true; -#else - ret = false; -#endif - - return ret; -} - -/* - * kfd_create_crat_image_acpi - Allocates memory for CRAT image and - * copies CRAT from ACPI (if available). - * NOTE: Call kfd_destroy_crat_image to free CRAT image memory - * - * @crat_image: CRAT read from ACPI. If no CRAT in ACPI then - * crat_image will be NULL - * @size: [OUT] size of crat_image - * - * Return 0 if successful else return error code - */ -int kfd_create_crat_image_acpi(void **crat_image, size_t *size) -{ - struct acpi_table_header *crat_table; - acpi_status status; - void *pcrat_image; - int rc = 0; - - if (!crat_image) - return -EINVAL; - - *crat_image = NULL; - - if (kfd_ignore_crat()) { - pr_info("CRAT table disabled by module option\n"); - return -ENODATA; - } - - /* Fetch the CRAT table from ACPI */ - status = acpi_get_table(CRAT_SIGNATURE, 0, &crat_table); - if (status == AE_NOT_FOUND) { - pr_info("CRAT table not found\n"); - return -ENODATA; - } else if (ACPI_FAILURE(status)) { - const char *err = acpi_format_exception(status); - - pr_err("CRAT table error: %s\n", err); - return -EINVAL; - } - - pcrat_image = kvmalloc(crat_table->length, GFP_KERNEL); - if (!pcrat_image) { - rc = -ENOMEM; - goto out; } - - memcpy(pcrat_image, crat_table, crat_table->length); - *crat_image = pcrat_image; - *size = crat_table->length; -out: - acpi_put_table(crat_table); - return rc; + return num_of_cache_types; } /* Memory required to create Virtual CRAT. @@ -2029,7 +1979,7 @@ static int kfd_create_vcrat_image_cpu(void *pcrat_image, size_t *size) } static int kfd_fill_gpu_memory_affinity(int *avail_size, - struct kfd_dev *kdev, uint8_t type, uint64_t size, + struct kfd_node *kdev, uint8_t type, uint64_t size, struct crat_subtype_memory *sub_type_hdr, uint32_t proximity_domain, const struct kfd_local_mem_info *local_mem_info) @@ -2058,13 +2008,13 @@ static int kfd_fill_gpu_memory_affinity(int *avail_size, } #ifdef CONFIG_ACPI_NUMA -static void kfd_find_numa_node_in_srat(struct kfd_dev *kdev) +static void kfd_find_numa_node_in_srat(struct kfd_node *kdev) { struct acpi_table_header *table_header = NULL; struct acpi_subtable_header *sub_header = NULL; unsigned long table_end, subtable_len; - u32 pci_id = pci_domain_nr(kdev->pdev->bus) << 16 | - pci_dev_id(kdev->pdev); + u32 pci_id = pci_domain_nr(kdev->adev->pdev->bus) << 16 | + pci_dev_id(kdev->adev->pdev); u32 bdf; acpi_status status; struct acpi_srat_cpu_affinity *cpu; @@ -2139,10 +2089,13 @@ static void kfd_find_numa_node_in_srat(struct kfd_dev *kdev) numa_node = 0; if (numa_node != NUMA_NO_NODE) - set_dev_node(&kdev->pdev->dev, numa_node); + set_dev_node(&kdev->adev->pdev->dev, numa_node); } #endif +#define KFD_CRAT_INTRA_SOCKET_WEIGHT 13 +#define KFD_CRAT_XGMI_WEIGHT 15 + /* kfd_fill_gpu_direct_io_link - Fill in direct io link from GPU * to its NUMA node * @avail_size: Available size in the memory @@ -2153,7 +2106,7 @@ static void kfd_find_numa_node_in_srat(struct kfd_dev *kdev) * Return 0 if successful else return -ve value */ static int kfd_fill_gpu_direct_io_link_to_cpu(int *avail_size, - struct kfd_dev *kdev, + struct kfd_node *kdev, struct crat_subtype_iolink *sub_type_hdr, uint32_t proximity_domain) { @@ -2173,7 +2126,13 @@ static int kfd_fill_gpu_direct_io_link_to_cpu(int *avail_size, /* Fill in IOLINK subtype. * TODO: Fill-in other fields of iolink subtype */ - if (kdev->adev->gmc.xgmi.connected_to_cpu) { + if (kdev->adev->gmc.xgmi.connected_to_cpu || + (KFD_GC_VERSION(kdev) == IP_VERSION(9, 4, 3) && + kdev->adev->smuio.funcs->get_pkg_type(kdev->adev) == + AMDGPU_PKG_TYPE_APU)) { + bool ext_cpu = KFD_GC_VERSION(kdev) != IP_VERSION(9, 4, 3); + int mem_bw = 819200, weight = ext_cpu ? KFD_CRAT_XGMI_WEIGHT : + KFD_CRAT_INTRA_SOCKET_WEIGHT; /* * with host gpu xgmi link, host can access gpu memory whether * or not pcie bar type is large, so always create bidirectional @@ -2181,13 +2140,16 @@ static int kfd_fill_gpu_direct_io_link_to_cpu(int *avail_size, */ sub_type_hdr->flags |= CRAT_IOLINK_FLAGS_BI_DIRECTIONAL; sub_type_hdr->io_interface_type = CRAT_IOLINK_TYPE_XGMI; - sub_type_hdr->num_hops_xgmi = 1; - if (KFD_GC_VERSION(kdev) == IP_VERSION(9, 4, 2)) { - sub_type_hdr->minimum_bandwidth_mbs = - amdgpu_amdkfd_get_xgmi_bandwidth_mbytes( - kdev->adev, NULL, true); - sub_type_hdr->maximum_bandwidth_mbs = - sub_type_hdr->minimum_bandwidth_mbs; + sub_type_hdr->weight_xgmi = weight; + if (ext_cpu) { + amdgpu_xgmi_get_bandwidth(kdev->adev, NULL, + AMDGPU_XGMI_BW_MODE_PER_LINK, + AMDGPU_XGMI_BW_UNIT_MBYTES, + &sub_type_hdr->minimum_bandwidth_mbs, + &sub_type_hdr->maximum_bandwidth_mbs); + } else { + sub_type_hdr->minimum_bandwidth_mbs = mem_bw; + sub_type_hdr->maximum_bandwidth_mbs = mem_bw; } } else { sub_type_hdr->io_interface_type = CRAT_IOLINK_TYPE_PCIEXPRESS; @@ -2200,14 +2162,15 @@ static int kfd_fill_gpu_direct_io_link_to_cpu(int *avail_size, sub_type_hdr->proximity_domain_from = proximity_domain; #ifdef CONFIG_ACPI_NUMA - if (kdev->pdev->dev.numa_node == NUMA_NO_NODE) + if (kdev->adev->pdev->dev.numa_node == NUMA_NO_NODE && + num_possible_nodes() > 1) kfd_find_numa_node_in_srat(kdev); #endif #ifdef CONFIG_NUMA - if (kdev->pdev->dev.numa_node == NUMA_NO_NODE) + if (kdev->adev->pdev->dev.numa_node == NUMA_NO_NODE) sub_type_hdr->proximity_domain_to = 0; else - sub_type_hdr->proximity_domain_to = kdev->pdev->dev.numa_node; + sub_type_hdr->proximity_domain_to = kdev->adev->pdev->dev.numa_node; #else sub_type_hdr->proximity_domain_to = 0; #endif @@ -2215,12 +2178,14 @@ static int kfd_fill_gpu_direct_io_link_to_cpu(int *avail_size, } static int kfd_fill_gpu_xgmi_link_to_gpu(int *avail_size, - struct kfd_dev *kdev, - struct kfd_dev *peer_kdev, + struct kfd_node *kdev, + struct kfd_node *peer_kdev, struct crat_subtype_iolink *sub_type_hdr, uint32_t proximity_domain_from, uint32_t proximity_domain_to) { + bool use_ta_info = kdev->kfd->num_nodes == 1; + *avail_size -= sizeof(struct crat_subtype_iolink); if (*avail_size < 0) return -ENOMEM; @@ -2235,12 +2200,25 @@ static int kfd_fill_gpu_xgmi_link_to_gpu(int *avail_size, sub_type_hdr->io_interface_type = CRAT_IOLINK_TYPE_XGMI; sub_type_hdr->proximity_domain_from = proximity_domain_from; sub_type_hdr->proximity_domain_to = proximity_domain_to; - sub_type_hdr->num_hops_xgmi = - amdgpu_amdkfd_get_xgmi_hops_count(kdev->adev, peer_kdev->adev); - sub_type_hdr->maximum_bandwidth_mbs = - amdgpu_amdkfd_get_xgmi_bandwidth_mbytes(kdev->adev, peer_kdev->adev, false); - sub_type_hdr->minimum_bandwidth_mbs = sub_type_hdr->maximum_bandwidth_mbs ? - amdgpu_amdkfd_get_xgmi_bandwidth_mbytes(kdev->adev, NULL, true) : 0; + + if (use_ta_info) { + sub_type_hdr->weight_xgmi = KFD_CRAT_XGMI_WEIGHT * + amdgpu_xgmi_get_hops_count(kdev->adev, peer_kdev->adev); + amdgpu_xgmi_get_bandwidth(kdev->adev, peer_kdev->adev, + AMDGPU_XGMI_BW_MODE_PER_PEER, + AMDGPU_XGMI_BW_UNIT_MBYTES, + &sub_type_hdr->minimum_bandwidth_mbs, + &sub_type_hdr->maximum_bandwidth_mbs); + } else { + bool is_single_hop = kdev->kfd == peer_kdev->kfd; + int weight = is_single_hop ? KFD_CRAT_INTRA_SOCKET_WEIGHT : + (2 * KFD_CRAT_INTRA_SOCKET_WEIGHT) + KFD_CRAT_XGMI_WEIGHT; + int mem_bw = 819200; + + sub_type_hdr->weight_xgmi = weight; + sub_type_hdr->maximum_bandwidth_mbs = is_single_hop ? mem_bw : 0; + sub_type_hdr->minimum_bandwidth_mbs = is_single_hop ? mem_bw : 0; + } return 0; } @@ -2252,19 +2230,18 @@ static int kfd_fill_gpu_xgmi_link_to_gpu(int *avail_size, * [OUT] actual size of data filled in crat_image */ static int kfd_create_vcrat_image_gpu(void *pcrat_image, - size_t *size, struct kfd_dev *kdev, + size_t *size, struct kfd_node *kdev, uint32_t proximity_domain) { struct crat_header *crat_table = (struct crat_header *)pcrat_image; + struct amdgpu_gfx_config *gfx_info = &kdev->adev->gfx.config; + struct amdgpu_cu_info *cu_info = &kdev->adev->gfx.cu_info; struct crat_subtype_generic *sub_type_hdr; struct kfd_local_mem_info local_mem_info; struct kfd_topology_device *peer_dev; struct crat_subtype_computeunit *cu; - struct kfd_cu_info cu_info; int avail_size = *size; uint32_t total_num_of_cu; - int num_of_cache_entries = 0; - int cache_mem_filled = 0; uint32_t nid = 0; int ret = 0; @@ -2275,9 +2252,6 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image, * Modify length and total_entries as subunits are added. */ avail_size -= sizeof(struct crat_header); - if (avail_size < 0) - return -ENOMEM; - memset(crat_table, 0, sizeof(struct crat_header)); memcpy(&crat_table->signature, CRAT_SIGNATURE, @@ -2291,9 +2265,6 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image, * First fill in the sub type header and then sub type data */ avail_size -= sizeof(struct crat_subtype_computeunit); - if (avail_size < 0) - return -ENOMEM; - sub_type_hdr = (struct crat_subtype_generic *)(crat_table + 1); memset(sub_type_hdr, 0, sizeof(struct crat_subtype_computeunit)); @@ -2306,29 +2277,23 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image, cu->flags |= CRAT_CU_FLAGS_GPU_PRESENT; cu->proximity_domain = proximity_domain; - amdgpu_amdkfd_get_cu_info(kdev->adev, &cu_info); - cu->num_simd_per_cu = cu_info.simd_per_cu; - cu->num_simd_cores = cu_info.simd_per_cu * cu_info.cu_active_number; - cu->max_waves_simd = cu_info.max_waves_per_simd; + cu->num_simd_per_cu = cu_info->simd_per_cu; + cu->num_simd_cores = cu_info->simd_per_cu * + (cu_info->number / kdev->kfd->num_nodes); + cu->max_waves_simd = cu_info->max_waves_per_simd; - cu->wave_front_size = cu_info.wave_front_size; - cu->array_count = cu_info.num_shader_arrays_per_engine * - cu_info.num_shader_engines; - total_num_of_cu = (cu->array_count * cu_info.num_cu_per_sh); + cu->wave_front_size = cu_info->wave_front_size; + cu->array_count = gfx_info->max_sh_per_se * + gfx_info->max_shader_engines; + total_num_of_cu = (cu->array_count * gfx_info->max_cu_per_sh); cu->processor_id_low = get_and_inc_gpu_processor_id(total_num_of_cu); - cu->num_cu_per_array = cu_info.num_cu_per_sh; - cu->max_slots_scatch_cu = cu_info.max_scratch_slots_per_cu; - cu->num_banks = cu_info.num_shader_engines; - cu->lds_size_in_kb = cu_info.lds_size; + cu->num_cu_per_array = gfx_info->max_cu_per_sh; + cu->max_slots_scatch_cu = cu_info->max_scratch_slots_per_cu; + cu->num_banks = gfx_info->max_shader_engines; + cu->lds_size_in_kb = cu_info->lds_size; cu->hsa_capability = 0; - /* Check if this node supports IOMMU. During parsing this flag will - * translate to HSA_CAP_ATS_PRESENT - */ - if (!kfd_iommu_check_device(kdev)) - cu->hsa_capability |= CRAT_CU_FLAGS_IOMMU_PRESENT; - crat_table->length += sub_type_hdr->length; crat_table->total_entries++; @@ -2341,7 +2306,7 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image, sub_type_hdr = (typeof(sub_type_hdr))((char *)sub_type_hdr + sub_type_hdr->length); - if (debug_largebar) + if (kdev->adev->debug_largebar) local_mem_info.local_mem_size_private = 0; if (local_mem_info.local_mem_size_private == 0) @@ -2365,31 +2330,12 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image, crat_table->length += sizeof(struct crat_subtype_memory); crat_table->total_entries++; - /* TODO: Fill in cache information. This information is NOT readily - * available in KGD - */ - sub_type_hdr = (typeof(sub_type_hdr))((char *)sub_type_hdr + - sub_type_hdr->length); - ret = kfd_fill_gpu_cache_info(kdev, cu->processor_id_low, - avail_size, - &cu_info, - (struct crat_subtype_cache *)sub_type_hdr, - &cache_mem_filled, - &num_of_cache_entries); - - if (ret < 0) - return ret; - - crat_table->length += cache_mem_filled; - crat_table->total_entries += num_of_cache_entries; - avail_size -= cache_mem_filled; - /* Fill in Subtype: IO_LINKS * Only direct links are added here which is Link from GPU to * its NUMA node. Indirect links are added by userspace. */ sub_type_hdr = (typeof(sub_type_hdr))((char *)sub_type_hdr + - cache_mem_filled); + sub_type_hdr->length); ret = kfd_fill_gpu_direct_io_link_to_cpu(&avail_size, kdev, (struct crat_subtype_iolink *)sub_type_hdr, proximity_domain); @@ -2408,12 +2354,14 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image, * (from other GPU to this GPU) will be added * in kfd_parse_subtype_iolink. */ - if (kdev->hive_id) { + if (kdev->kfd->hive_id) { for (nid = 0; nid < proximity_domain; ++nid) { peer_dev = kfd_topology_device_by_proximity_domain_no_lock(nid); if (!peer_dev->gpu) continue; - if (peer_dev->gpu->hive_id != kdev->hive_id) + if (peer_dev->gpu->kfd->hive_id != kdev->kfd->hive_id) + continue; + if (!amdgpu_xgmi_get_is_sharing_enabled(kdev->adev, peer_dev->gpu->adev)) continue; sub_type_hdr = (typeof(sub_type_hdr))( (char *)sub_type_hdr + @@ -2447,12 +2395,12 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image, * (COMPUTE_UNIT_CPU | COMPUTE_UNIT_GPU) - Create VCRAT for APU * -- this option is not currently implemented. * The assumption is that all AMD APUs will have CRAT - * @kdev: Valid kfd_device required if flags contain COMPUTE_UNIT_GPU + * @kdev: Valid kfd_node required if flags contain COMPUTE_UNIT_GPU * * Return 0 if successful else return -ve value */ int kfd_create_crat_image_virtual(void **crat_image, size_t *size, - int flags, struct kfd_dev *kdev, + int flags, struct kfd_node *kdev, uint32_t proximity_domain) { void *pcrat_image = NULL; |