diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_crat.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 142 |
1 files changed, 132 insertions, 10 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c index cd8e459201f1..693469c18c60 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c @@ -28,6 +28,7 @@ #include "kfd_topology.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. @@ -55,6 +56,7 @@ 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), @@ -64,6 +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), @@ -73,6 +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), @@ -88,6 +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), @@ -95,8 +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), @@ -104,8 +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), @@ -135,6 +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), @@ -144,6 +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), @@ -153,6 +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), @@ -162,6 +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), @@ -174,6 +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), @@ -183,6 +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), @@ -192,6 +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), @@ -201,6 +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), @@ -213,6 +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), @@ -222,6 +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), @@ -231,6 +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), @@ -240,6 +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), @@ -252,6 +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), @@ -261,6 +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), @@ -270,6 +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), @@ -279,6 +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), @@ -291,6 +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), @@ -300,6 +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), @@ -309,6 +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), @@ -318,6 +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), @@ -330,6 +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), @@ -339,6 +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), @@ -348,6 +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), @@ -357,6 +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), @@ -369,6 +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), @@ -378,6 +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), @@ -387,6 +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), @@ -396,6 +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), @@ -405,6 +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), @@ -417,6 +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), @@ -426,6 +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), @@ -435,6 +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), @@ -444,6 +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), @@ -453,6 +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), @@ -465,6 +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), @@ -474,6 +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), @@ -483,6 +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), @@ -492,6 +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), @@ -501,6 +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), @@ -513,6 +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), @@ -522,6 +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), @@ -531,6 +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), @@ -540,6 +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), @@ -549,6 +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), @@ -558,6 +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), @@ -570,6 +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), @@ -579,6 +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), @@ -588,6 +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), @@ -597,6 +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), @@ -606,6 +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), @@ -615,6 +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), @@ -627,6 +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), @@ -636,6 +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), @@ -645,6 +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), @@ -654,6 +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), @@ -663,6 +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), @@ -672,6 +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), @@ -684,6 +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), @@ -693,6 +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), @@ -702,6 +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), @@ -711,6 +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), @@ -720,6 +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), @@ -729,6 +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), @@ -741,6 +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), @@ -750,6 +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), @@ -759,6 +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), @@ -768,6 +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), @@ -777,6 +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), @@ -789,6 +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), @@ -798,6 +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), @@ -807,6 +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), @@ -816,6 +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), @@ -825,6 +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), @@ -837,6 +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), @@ -846,6 +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), @@ -855,6 +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), @@ -864,6 +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), @@ -873,6 +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), @@ -885,6 +970,7 @@ 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), @@ -894,6 +980,7 @@ static struct kfd_gpu_cache_info dummy_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), @@ -903,6 +990,7 @@ static struct kfd_gpu_cache_info dummy_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), @@ -912,6 +1000,7 @@ static struct kfd_gpu_cache_info dummy_cache_info[] = { /* 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), @@ -921,6 +1010,7 @@ static struct kfd_gpu_cache_info dummy_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), @@ -1333,6 +1423,7 @@ err: 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; @@ -1345,7 +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 */ @@ -1357,6 +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 */ @@ -1367,6 +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 */ @@ -1379,6 +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) */ @@ -1389,6 +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 */ @@ -1399,6 +1504,7 @@ 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; @@ -1414,6 +1520,8 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, 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); @@ -1425,6 +1533,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, 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); @@ -1435,6 +1544,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, 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); @@ -1445,6 +1555,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, 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); @@ -1455,6 +1566,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, 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); @@ -1467,6 +1579,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, int kfd_get_gpu_cache_info(struct kfd_node *kdev, struct kfd_gpu_cache_info **pcache_info) { int num_of_cache_types = 0; + bool cache_line_size_missing = false; switch (kdev->adev->asic_type) { case CHIP_KAVERI: @@ -1525,6 +1638,8 @@ int kfd_get_gpu_cache_info(struct kfd_node *kdev, struct kfd_gpu_cache_info **pc 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); @@ -1587,8 +1702,19 @@ int kfd_get_gpu_cache_info(struct kfd_node *kdev, struct kfd_gpu_cache_info **pc case IP_VERSION(11, 0, 3): case IP_VERSION(11, 0, 4): case IP_VERSION(11, 5, 0): + case IP_VERSION(11, 5, 1): + case IP_VERSION(11, 5, 2): + /* 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->kfd, *pcache_info); + kfd_fill_gpu_cache_info_from_gfx_config(kdev->kfd, + cache_line_size_missing, + *pcache_info); break; default: *pcache_info = dummy_cache_info; @@ -2120,9 +2246,6 @@ 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, @@ -2136,9 +2259,6 @@ 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)); @@ -2235,6 +2355,8 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image, continue; 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 + sizeof(struct crat_subtype_iolink)); |