diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h | 676 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm | 8 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 23 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_events.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c | 26 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 187 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.h | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c | 1 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c | 5 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c | 9 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 24 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process.c | 125 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c | 6 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 233 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.h | 9 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 55 |
17 files changed, 709 insertions, 688 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h index d7cd5fa313ff..d1caaf0e6a7c 100644 --- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h +++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h @@ -674,7 +674,7 @@ static const uint32_t cwsr_trap_gfx9_hex[] = { 0x86ea6a6a, 0x8f6e837a, 0xb96ee0c2, 0xbf800002, 0xb97a0002, 0xbf8a0000, - 0xbe801f6c, 0xbf810000, + 0xbe801f6c, 0xbf9b0000, }; static const uint32_t cwsr_trap_nv1x_hex[] = { @@ -1091,7 +1091,7 @@ static const uint32_t cwsr_trap_nv1x_hex[] = { 0xb9eef807, 0x876dff6d, 0x0000ffff, 0x87fe7e7e, 0x87ea6a6a, 0xb9faf802, - 0xbe80226c, 0xbf810000, + 0xbe80226c, 0xbf9b0000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, 0x00000000, @@ -1574,7 +1574,7 @@ static const uint32_t cwsr_trap_arcturus_hex[] = { 0x86ea6a6a, 0x8f6e837a, 0xb96ee0c2, 0xbf800002, 0xb97a0002, 0xbf8a0000, - 0xbe801f6c, 0xbf810000, + 0xbe801f6c, 0xbf9b0000, }; static const uint32_t cwsr_trap_aldebaran_hex[] = { @@ -2065,11 +2065,11 @@ static const uint32_t cwsr_trap_aldebaran_hex[] = { 0x86ea6a6a, 0x8f6e837a, 0xb96ee0c2, 0xbf800002, 0xb97a0002, 0xbf8a0000, - 0xbe801f6c, 0xbf810000, + 0xbe801f6c, 0xbf9b0000, }; static const uint32_t cwsr_trap_gfx10_hex[] = { - 0xbf820001, 0xbf820220, + 0xbf820001, 0xbf820221, 0xb0804004, 0xb978f802, 0x8a78ff78, 0x00020006, 0xb97bf803, 0x876eff78, @@ -2118,391 +2118,391 @@ static const uint32_t cwsr_trap_gfx10_hex[] = { 0xbf900004, 0xbf8cc07f, 0x877aff7f, 0x04000000, 0x8f7a857a, 0x886d7a6d, - 0xbefa037e, 0x877bff7f, - 0x0000ffff, 0xbefe03c1, - 0xbeff03c1, 0xdc5f8000, - 0x007a0000, 0x7e000280, - 0xbefe037a, 0xbeff037b, - 0xb97b02dc, 0x8f7b997b, - 0xb97a3a05, 0x807a817a, - 0xbf0d997b, 0xbf850002, - 0x8f7a897a, 0xbf820001, - 0x8f7a8a7a, 0xb97b1e06, - 0x8f7b8a7b, 0x807a7b7a, + 0x7e008200, 0xbefa037e, 0x877bff7f, 0x0000ffff, - 0x807aff7a, 0x00000200, - 0x807a7e7a, 0x827b807b, - 0xd7610000, 0x00010870, - 0xd7610000, 0x00010a71, - 0xd7610000, 0x00010c72, - 0xd7610000, 0x00010e73, - 0xd7610000, 0x00011074, - 0xd7610000, 0x00011275, - 0xd7610000, 0x00011476, - 0xd7610000, 0x00011677, - 0xd7610000, 0x00011a79, - 0xd7610000, 0x00011c7e, - 0xd7610000, 0x00011e7f, - 0xbefe03ff, 0x00003fff, - 0xbeff0380, 0xdc5f8040, - 0x007a0000, 0xd760007a, - 0x00011d00, 0xd760007b, - 0x00011f00, 0xbefe037a, - 0xbeff037b, 0xbef4037e, - 0x8775ff7f, 0x0000ffff, - 0x8875ff75, 0x00040000, - 0xbef60380, 0xbef703ff, - 0x10807fac, 0xbef1037c, - 0xbef00380, 0xb97302dc, - 0x8f739973, 0xbefe03c1, - 0x907c9973, 0x877c817c, - 0xbf06817c, 0xbf850002, - 0xbeff0380, 0xbf820002, - 0xbeff03c1, 0xbf820009, + 0xbefe03c1, 0xbeff03c1, + 0xdc5f8000, 0x007a0000, + 0x7e000280, 0xbefe037a, + 0xbeff037b, 0xb97b02dc, + 0x8f7b997b, 0xb97a3a05, + 0x807a817a, 0xbf0d997b, + 0xbf850002, 0x8f7a897a, + 0xbf820001, 0x8f7a8a7a, + 0xb97b1e06, 0x8f7b8a7b, + 0x807a7b7a, 0x877bff7f, + 0x0000ffff, 0x807aff7a, + 0x00000200, 0x807a7e7a, + 0x827b807b, 0xd7610000, + 0x00010870, 0xd7610000, + 0x00010a71, 0xd7610000, + 0x00010c72, 0xd7610000, + 0x00010e73, 0xd7610000, + 0x00011074, 0xd7610000, + 0x00011275, 0xd7610000, + 0x00011476, 0xd7610000, + 0x00011677, 0xd7610000, + 0x00011a79, 0xd7610000, + 0x00011c7e, 0xd7610000, + 0x00011e7f, 0xbefe03ff, + 0x00003fff, 0xbeff0380, + 0xdc5f8040, 0x007a0000, + 0xd760007a, 0x00011d00, + 0xd760007b, 0x00011f00, + 0xbefe037a, 0xbeff037b, + 0xbef4037e, 0x8775ff7f, + 0x0000ffff, 0x8875ff75, + 0x00040000, 0xbef60380, + 0xbef703ff, 0x10807fac, + 0xbef1037c, 0xbef00380, + 0xb97302dc, 0x8f739973, + 0xbefe03c1, 0x907c9973, + 0x877c817c, 0xbf06817c, + 0xbf850002, 0xbeff0380, + 0xbf820002, 0xbeff03c1, + 0xbf820009, 0xbef603ff, + 0x01000000, 0xe0704080, + 0x705d0100, 0xe0704100, + 0x705d0200, 0xe0704180, + 0x705d0300, 0xbf820008, 0xbef603ff, 0x01000000, - 0xe0704080, 0x705d0100, - 0xe0704100, 0x705d0200, - 0xe0704180, 0x705d0300, - 0xbf820008, 0xbef603ff, - 0x01000000, 0xe0704100, - 0x705d0100, 0xe0704200, - 0x705d0200, 0xe0704300, - 0x705d0300, 0xb9703a05, - 0x80708170, 0xbf0d9973, - 0xbf850002, 0x8f708970, - 0xbf820001, 0x8f708a70, - 0xb97a1e06, 0x8f7a8a7a, - 0x80707a70, 0x8070ff70, - 0x00000200, 0xbef603ff, - 0x01000000, 0x7e000280, - 0x7e020280, 0x7e040280, - 0xbefc0380, 0xd7610002, - 0x0000f871, 0x807c817c, - 0xd7610002, 0x0000f86c, - 0x807c817c, 0x8a7aff6d, - 0x80000000, 0xd7610002, - 0x0000f87a, 0x807c817c, - 0xd7610002, 0x0000f86e, - 0x807c817c, 0xd7610002, - 0x0000f86f, 0x807c817c, - 0xd7610002, 0x0000f878, - 0x807c817c, 0xb97af803, - 0xd7610002, 0x0000f87a, - 0x807c817c, 0xd7610002, - 0x0000f87b, 0x807c817c, - 0xb971f801, 0xd7610002, - 0x0000f871, 0x807c817c, - 0xb971f814, 0xd7610002, - 0x0000f871, 0x807c817c, - 0xb971f815, 0xd7610002, - 0x0000f871, 0x807c817c, - 0xbefe03ff, 0x0000ffff, - 0xbeff0380, 0xe0704000, - 0x705d0200, 0xbefe03c1, + 0xe0704100, 0x705d0100, + 0xe0704200, 0x705d0200, + 0xe0704300, 0x705d0300, 0xb9703a05, 0x80708170, 0xbf0d9973, 0xbf850002, 0x8f708970, 0xbf820001, 0x8f708a70, 0xb97a1e06, 0x8f7a8a7a, 0x80707a70, + 0x8070ff70, 0x00000200, 0xbef603ff, 0x01000000, - 0xbef90380, 0xbefc0380, - 0xbf800000, 0xbe802f00, - 0xbe822f02, 0xbe842f04, - 0xbe862f06, 0xbe882f08, - 0xbe8a2f0a, 0xbe8c2f0c, - 0xbe8e2f0e, 0xd7610002, - 0x0000f200, 0x80798179, - 0xd7610002, 0x0000f201, + 0x7e000280, 0x7e020280, + 0x7e040280, 0xbefc0380, + 0xd7610002, 0x0000f871, + 0x807c817c, 0xd7610002, + 0x0000f86c, 0x807c817c, + 0x8a7aff6d, 0x80000000, + 0xd7610002, 0x0000f87a, + 0x807c817c, 0xd7610002, + 0x0000f86e, 0x807c817c, + 0xd7610002, 0x0000f86f, + 0x807c817c, 0xd7610002, + 0x0000f878, 0x807c817c, + 0xb97af803, 0xd7610002, + 0x0000f87a, 0x807c817c, + 0xd7610002, 0x0000f87b, + 0x807c817c, 0xb971f801, + 0xd7610002, 0x0000f871, + 0x807c817c, 0xb971f814, + 0xd7610002, 0x0000f871, + 0x807c817c, 0xb971f815, + 0xd7610002, 0x0000f871, + 0x807c817c, 0xbefe03ff, + 0x0000ffff, 0xbeff0380, + 0xe0704000, 0x705d0200, + 0xbefe03c1, 0xb9703a05, + 0x80708170, 0xbf0d9973, + 0xbf850002, 0x8f708970, + 0xbf820001, 0x8f708a70, + 0xb97a1e06, 0x8f7a8a7a, + 0x80707a70, 0xbef603ff, + 0x01000000, 0xbef90380, + 0xbefc0380, 0xbf800000, + 0xbe802f00, 0xbe822f02, + 0xbe842f04, 0xbe862f06, + 0xbe882f08, 0xbe8a2f0a, + 0xbe8c2f0c, 0xbe8e2f0e, + 0xd7610002, 0x0000f200, 0x80798179, 0xd7610002, - 0x0000f202, 0x80798179, - 0xd7610002, 0x0000f203, + 0x0000f201, 0x80798179, + 0xd7610002, 0x0000f202, 0x80798179, 0xd7610002, - 0x0000f204, 0x80798179, - 0xd7610002, 0x0000f205, + 0x0000f203, 0x80798179, + 0xd7610002, 0x0000f204, 0x80798179, 0xd7610002, - 0x0000f206, 0x80798179, - 0xd7610002, 0x0000f207, + 0x0000f205, 0x80798179, + 0xd7610002, 0x0000f206, 0x80798179, 0xd7610002, - 0x0000f208, 0x80798179, - 0xd7610002, 0x0000f209, + 0x0000f207, 0x80798179, + 0xd7610002, 0x0000f208, 0x80798179, 0xd7610002, - 0x0000f20a, 0x80798179, - 0xd7610002, 0x0000f20b, + 0x0000f209, 0x80798179, + 0xd7610002, 0x0000f20a, 0x80798179, 0xd7610002, - 0x0000f20c, 0x80798179, - 0xd7610002, 0x0000f20d, + 0x0000f20b, 0x80798179, + 0xd7610002, 0x0000f20c, 0x80798179, 0xd7610002, - 0x0000f20e, 0x80798179, - 0xd7610002, 0x0000f20f, - 0x80798179, 0xbf06a079, - 0xbf840006, 0xe0704000, - 0x705d0200, 0x8070ff70, - 0x00000080, 0xbef90380, - 0x7e040280, 0x807c907c, - 0xbf0aff7c, 0x00000060, - 0xbf85ffbc, 0xbe802f00, - 0xbe822f02, 0xbe842f04, - 0xbe862f06, 0xbe882f08, - 0xbe8a2f0a, 0xd7610002, - 0x0000f200, 0x80798179, - 0xd7610002, 0x0000f201, + 0x0000f20d, 0x80798179, + 0xd7610002, 0x0000f20e, 0x80798179, 0xd7610002, - 0x0000f202, 0x80798179, - 0xd7610002, 0x0000f203, + 0x0000f20f, 0x80798179, + 0xbf06a079, 0xbf840006, + 0xe0704000, 0x705d0200, + 0x8070ff70, 0x00000080, + 0xbef90380, 0x7e040280, + 0x807c907c, 0xbf0aff7c, + 0x00000060, 0xbf85ffbc, + 0xbe802f00, 0xbe822f02, + 0xbe842f04, 0xbe862f06, + 0xbe882f08, 0xbe8a2f0a, + 0xd7610002, 0x0000f200, 0x80798179, 0xd7610002, - 0x0000f204, 0x80798179, - 0xd7610002, 0x0000f205, + 0x0000f201, 0x80798179, + 0xd7610002, 0x0000f202, 0x80798179, 0xd7610002, - 0x0000f206, 0x80798179, - 0xd7610002, 0x0000f207, + 0x0000f203, 0x80798179, + 0xd7610002, 0x0000f204, 0x80798179, 0xd7610002, - 0x0000f208, 0x80798179, - 0xd7610002, 0x0000f209, + 0x0000f205, 0x80798179, + 0xd7610002, 0x0000f206, 0x80798179, 0xd7610002, - 0x0000f20a, 0x80798179, - 0xd7610002, 0x0000f20b, - 0x80798179, 0xe0704000, - 0x705d0200, 0xbefe03c1, - 0x907c9973, 0x877c817c, - 0xbf06817c, 0xbf850002, - 0xbeff0380, 0xbf820001, - 0xbeff03c1, 0xb97b4306, - 0x877bc17b, 0xbf840044, - 0xbf8a0000, 0x877aff6d, - 0x80000000, 0xbf840040, - 0x8f7b867b, 0x8f7b827b, - 0xbef6037b, 0xb9703a05, - 0x80708170, 0xbf0d9973, - 0xbf850002, 0x8f708970, - 0xbf820001, 0x8f708a70, - 0xb97a1e06, 0x8f7a8a7a, - 0x80707a70, 0x8070ff70, - 0x00000200, 0x8070ff70, - 0x00000080, 0xbef603ff, - 0x01000000, 0xd7650000, - 0x000100c1, 0xd7660000, - 0x000200c1, 0x16000084, - 0x907c9973, 0x877c817c, - 0xbf06817c, 0xbefc0380, - 0xbf850012, 0xbe8303ff, - 0x00000080, 0xbf800000, - 0xbf800000, 0xbf800000, - 0xd8d80000, 0x01000000, - 0xbf8c0000, 0xe0704000, - 0x705d0100, 0x807c037c, - 0x80700370, 0xd5250000, - 0x0001ff00, 0x00000080, - 0xbf0a7b7c, 0xbf85fff4, - 0xbf820011, 0xbe8303ff, - 0x00000100, 0xbf800000, - 0xbf800000, 0xbf800000, - 0xd8d80000, 0x01000000, - 0xbf8c0000, 0xe0704000, - 0x705d0100, 0x807c037c, - 0x80700370, 0xd5250000, - 0x0001ff00, 0x00000100, - 0xbf0a7b7c, 0xbf85fff4, + 0x0000f207, 0x80798179, + 0xd7610002, 0x0000f208, + 0x80798179, 0xd7610002, + 0x0000f209, 0x80798179, + 0xd7610002, 0x0000f20a, + 0x80798179, 0xd7610002, + 0x0000f20b, 0x80798179, + 0xe0704000, 0x705d0200, 0xbefe03c1, 0x907c9973, 0x877c817c, 0xbf06817c, - 0xbf850004, 0xbef003ff, - 0x00000200, 0xbeff0380, - 0xbf820003, 0xbef003ff, - 0x00000400, 0xbeff03c1, - 0xb97b3a05, 0x807b817b, - 0x8f7b827b, 0x907c9973, + 0xbf850002, 0xbeff0380, + 0xbf820001, 0xbeff03c1, + 0xb97b4306, 0x877bc17b, + 0xbf840044, 0xbf8a0000, + 0x877aff6d, 0x80000000, + 0xbf840040, 0x8f7b867b, + 0x8f7b827b, 0xbef6037b, + 0xb9703a05, 0x80708170, + 0xbf0d9973, 0xbf850002, + 0x8f708970, 0xbf820001, + 0x8f708a70, 0xb97a1e06, + 0x8f7a8a7a, 0x80707a70, + 0x8070ff70, 0x00000200, + 0x8070ff70, 0x00000080, + 0xbef603ff, 0x01000000, + 0xd7650000, 0x000100c1, + 0xd7660000, 0x000200c1, + 0x16000084, 0x907c9973, 0x877c817c, 0xbf06817c, - 0xbf850017, 0xbef603ff, - 0x01000000, 0xbefc0384, - 0xbf0a7b7c, 0xbf840037, - 0x7e008700, 0x7e028701, - 0x7e048702, 0x7e068703, - 0xe0704000, 0x705d0000, - 0xe0704080, 0x705d0100, - 0xe0704100, 0x705d0200, - 0xe0704180, 0x705d0300, - 0x807c847c, 0x8070ff70, - 0x00000200, 0xbf0a7b7c, - 0xbf85ffef, 0xbf820025, + 0xbefc0380, 0xbf850012, + 0xbe8303ff, 0x00000080, + 0xbf800000, 0xbf800000, + 0xbf800000, 0xd8d80000, + 0x01000000, 0xbf8c0000, + 0xe0704000, 0x705d0100, + 0x807c037c, 0x80700370, + 0xd5250000, 0x0001ff00, + 0x00000080, 0xbf0a7b7c, + 0xbf85fff4, 0xbf820011, + 0xbe8303ff, 0x00000100, + 0xbf800000, 0xbf800000, + 0xbf800000, 0xd8d80000, + 0x01000000, 0xbf8c0000, + 0xe0704000, 0x705d0100, + 0x807c037c, 0x80700370, + 0xd5250000, 0x0001ff00, + 0x00000100, 0xbf0a7b7c, + 0xbf85fff4, 0xbefe03c1, + 0x907c9973, 0x877c817c, + 0xbf06817c, 0xbf850004, + 0xbef003ff, 0x00000200, + 0xbeff0380, 0xbf820003, + 0xbef003ff, 0x00000400, + 0xbeff03c1, 0xb97b3a05, + 0x807b817b, 0x8f7b827b, + 0x907c9973, 0x877c817c, + 0xbf06817c, 0xbf850017, 0xbef603ff, 0x01000000, 0xbefc0384, 0xbf0a7b7c, - 0xbf840011, 0x7e008700, + 0xbf840037, 0x7e008700, 0x7e028701, 0x7e048702, 0x7e068703, 0xe0704000, - 0x705d0000, 0xe0704100, - 0x705d0100, 0xe0704200, - 0x705d0200, 0xe0704300, + 0x705d0000, 0xe0704080, + 0x705d0100, 0xe0704100, + 0x705d0200, 0xe0704180, 0x705d0300, 0x807c847c, - 0x8070ff70, 0x00000400, + 0x8070ff70, 0x00000200, 0xbf0a7b7c, 0xbf85ffef, - 0xb97b1e06, 0x877bc17b, - 0xbf84000c, 0x8f7b837b, - 0x807b7c7b, 0xbefe03c1, - 0xbeff0380, 0x7e008700, + 0xbf820025, 0xbef603ff, + 0x01000000, 0xbefc0384, + 0xbf0a7b7c, 0xbf840011, + 0x7e008700, 0x7e028701, + 0x7e048702, 0x7e068703, 0xe0704000, 0x705d0000, - 0x807c817c, 0x8070ff70, - 0x00000080, 0xbf0a7b7c, - 0xbf85fff8, 0xbf82013b, - 0xbef4037e, 0x8775ff7f, - 0x0000ffff, 0x8875ff75, - 0x00040000, 0xbef60380, - 0xbef703ff, 0x10807fac, - 0xb97202dc, 0x8f729972, - 0x876eff7f, 0x04000000, - 0xbf840034, 0xbefe03c1, - 0x907c9972, 0x877c817c, - 0xbf06817c, 0xbf850002, - 0xbeff0380, 0xbf820001, - 0xbeff03c1, 0xb96f4306, - 0x876fc16f, 0xbf840029, - 0x8f6f866f, 0x8f6f826f, - 0xbef6036f, 0xb9783a05, - 0x80788178, 0xbf0d9972, - 0xbf850002, 0x8f788978, - 0xbf820001, 0x8f788a78, - 0xb96e1e06, 0x8f6e8a6e, - 0x80786e78, 0x8078ff78, - 0x00000200, 0x8078ff78, - 0x00000080, 0xbef603ff, - 0x01000000, 0x907c9972, - 0x877c817c, 0xbf06817c, - 0xbefc0380, 0xbf850009, - 0xe0310000, 0x781d0000, - 0x807cff7c, 0x00000080, - 0x8078ff78, 0x00000080, - 0xbf0a6f7c, 0xbf85fff8, - 0xbf820008, 0xe0310000, - 0x781d0000, 0x807cff7c, - 0x00000100, 0x8078ff78, - 0x00000100, 0xbf0a6f7c, - 0xbf85fff8, 0xbef80380, + 0xe0704100, 0x705d0100, + 0xe0704200, 0x705d0200, + 0xe0704300, 0x705d0300, + 0x807c847c, 0x8070ff70, + 0x00000400, 0xbf0a7b7c, + 0xbf85ffef, 0xb97b1e06, + 0x877bc17b, 0xbf84000c, + 0x8f7b837b, 0x807b7c7b, + 0xbefe03c1, 0xbeff0380, + 0x7e008700, 0xe0704000, + 0x705d0000, 0x807c817c, + 0x8070ff70, 0x00000080, + 0xbf0a7b7c, 0xbf85fff8, + 0xbf82013b, 0xbef4037e, + 0x8775ff7f, 0x0000ffff, + 0x8875ff75, 0x00040000, + 0xbef60380, 0xbef703ff, + 0x10807fac, 0xb97202dc, + 0x8f729972, 0x876eff7f, + 0x04000000, 0xbf840034, 0xbefe03c1, 0x907c9972, 0x877c817c, 0xbf06817c, 0xbf850002, 0xbeff0380, 0xbf820001, 0xbeff03c1, - 0xb96f3a05, 0x806f816f, - 0x8f6f826f, 0x907c9972, - 0x877c817c, 0xbf06817c, - 0xbf850024, 0xbef603ff, - 0x01000000, 0xbeee0378, + 0xb96f4306, 0x876fc16f, + 0xbf840029, 0x8f6f866f, + 0x8f6f826f, 0xbef6036f, + 0xb9783a05, 0x80788178, + 0xbf0d9972, 0xbf850002, + 0x8f788978, 0xbf820001, + 0x8f788a78, 0xb96e1e06, + 0x8f6e8a6e, 0x80786e78, 0x8078ff78, 0x00000200, - 0xbefc0384, 0xbf0a6f7c, - 0xbf840050, 0xe0304000, - 0x785d0000, 0xe0304080, - 0x785d0100, 0xe0304100, - 0x785d0200, 0xe0304180, - 0x785d0300, 0xbf8c3f70, - 0x7e008500, 0x7e028501, - 0x7e048502, 0x7e068503, - 0x807c847c, 0x8078ff78, - 0x00000200, 0xbf0a6f7c, - 0xbf85ffee, 0xe0304000, - 0x6e5d0000, 0xe0304080, - 0x6e5d0100, 0xe0304100, - 0x6e5d0200, 0xe0304180, - 0x6e5d0300, 0xbf8c3f70, - 0xbf820034, 0xbef603ff, - 0x01000000, 0xbeee0378, - 0x8078ff78, 0x00000400, - 0xbefc0384, 0xbf0a6f7c, - 0xbf840012, 0xe0304000, - 0x785d0000, 0xe0304100, - 0x785d0100, 0xe0304200, - 0x785d0200, 0xe0304300, - 0x785d0300, 0xbf8c3f70, - 0x7e008500, 0x7e028501, - 0x7e048502, 0x7e068503, - 0x807c847c, 0x8078ff78, - 0x00000400, 0xbf0a6f7c, - 0xbf85ffee, 0xb96f1e06, - 0x876fc16f, 0xbf84000e, - 0x8f6f836f, 0x806f7c6f, - 0xbefe03c1, 0xbeff0380, + 0x8078ff78, 0x00000080, + 0xbef603ff, 0x01000000, + 0x907c9972, 0x877c817c, + 0xbf06817c, 0xbefc0380, + 0xbf850009, 0xe0310000, + 0x781d0000, 0x807cff7c, + 0x00000080, 0x8078ff78, + 0x00000080, 0xbf0a6f7c, + 0xbf85fff8, 0xbf820008, + 0xe0310000, 0x781d0000, + 0x807cff7c, 0x00000100, + 0x8078ff78, 0x00000100, + 0xbf0a6f7c, 0xbf85fff8, + 0xbef80380, 0xbefe03c1, + 0x907c9972, 0x877c817c, + 0xbf06817c, 0xbf850002, + 0xbeff0380, 0xbf820001, + 0xbeff03c1, 0xb96f3a05, + 0x806f816f, 0x8f6f826f, + 0x907c9972, 0x877c817c, + 0xbf06817c, 0xbf850024, + 0xbef603ff, 0x01000000, + 0xbeee0378, 0x8078ff78, + 0x00000200, 0xbefc0384, + 0xbf0a6f7c, 0xbf840050, 0xe0304000, 0x785d0000, + 0xe0304080, 0x785d0100, + 0xe0304100, 0x785d0200, + 0xe0304180, 0x785d0300, 0xbf8c3f70, 0x7e008500, - 0x807c817c, 0x8078ff78, - 0x00000080, 0xbf0a6f7c, - 0xbf85fff7, 0xbeff03c1, + 0x7e028501, 0x7e048502, + 0x7e068503, 0x807c847c, + 0x8078ff78, 0x00000200, + 0xbf0a6f7c, 0xbf85ffee, 0xe0304000, 0x6e5d0000, - 0xe0304100, 0x6e5d0100, - 0xe0304200, 0x6e5d0200, - 0xe0304300, 0x6e5d0300, - 0xbf8c3f70, 0xb9783a05, - 0x80788178, 0xbf0d9972, - 0xbf850002, 0x8f788978, - 0xbf820001, 0x8f788a78, - 0xb96e1e06, 0x8f6e8a6e, - 0x80786e78, 0x8078ff78, - 0x00000200, 0x80f8ff78, - 0x00000050, 0xbef603ff, - 0x01000000, 0xbefc03ff, - 0x0000006c, 0x80f89078, - 0xf429003a, 0xf0000000, - 0xbf8cc07f, 0x80fc847c, - 0xbf800000, 0xbe803100, - 0xbe823102, 0x80f8a078, - 0xf42d003a, 0xf0000000, - 0xbf8cc07f, 0x80fc887c, - 0xbf800000, 0xbe803100, - 0xbe823102, 0xbe843104, - 0xbe863106, 0x80f8c078, - 0xf431003a, 0xf0000000, - 0xbf8cc07f, 0x80fc907c, - 0xbf800000, 0xbe803100, - 0xbe823102, 0xbe843104, - 0xbe863106, 0xbe883108, - 0xbe8a310a, 0xbe8c310c, - 0xbe8e310e, 0xbf06807c, - 0xbf84fff0, 0xba80f801, - 0x00000000, 0xbf8a0000, + 0xe0304080, 0x6e5d0100, + 0xe0304100, 0x6e5d0200, + 0xe0304180, 0x6e5d0300, + 0xbf8c3f70, 0xbf820034, + 0xbef603ff, 0x01000000, + 0xbeee0378, 0x8078ff78, + 0x00000400, 0xbefc0384, + 0xbf0a6f7c, 0xbf840012, + 0xe0304000, 0x785d0000, + 0xe0304100, 0x785d0100, + 0xe0304200, 0x785d0200, + 0xe0304300, 0x785d0300, + 0xbf8c3f70, 0x7e008500, + 0x7e028501, 0x7e048502, + 0x7e068503, 0x807c847c, + 0x8078ff78, 0x00000400, + 0xbf0a6f7c, 0xbf85ffee, + 0xb96f1e06, 0x876fc16f, + 0xbf84000e, 0x8f6f836f, + 0x806f7c6f, 0xbefe03c1, + 0xbeff0380, 0xe0304000, + 0x785d0000, 0xbf8c3f70, + 0x7e008500, 0x807c817c, + 0x8078ff78, 0x00000080, + 0xbf0a6f7c, 0xbf85fff7, + 0xbeff03c1, 0xe0304000, + 0x6e5d0000, 0xe0304100, + 0x6e5d0100, 0xe0304200, + 0x6e5d0200, 0xe0304300, + 0x6e5d0300, 0xbf8c3f70, 0xb9783a05, 0x80788178, 0xbf0d9972, 0xbf850002, 0x8f788978, 0xbf820001, 0x8f788a78, 0xb96e1e06, 0x8f6e8a6e, 0x80786e78, 0x8078ff78, 0x00000200, + 0x80f8ff78, 0x00000050, 0xbef603ff, 0x01000000, - 0xf4211bfa, 0xf0000000, - 0x80788478, 0xf4211b3a, + 0xbefc03ff, 0x0000006c, + 0x80f89078, 0xf429003a, + 0xf0000000, 0xbf8cc07f, + 0x80fc847c, 0xbf800000, + 0xbe803100, 0xbe823102, + 0x80f8a078, 0xf42d003a, + 0xf0000000, 0xbf8cc07f, + 0x80fc887c, 0xbf800000, + 0xbe803100, 0xbe823102, + 0xbe843104, 0xbe863106, + 0x80f8c078, 0xf431003a, + 0xf0000000, 0xbf8cc07f, + 0x80fc907c, 0xbf800000, + 0xbe803100, 0xbe823102, + 0xbe843104, 0xbe863106, + 0xbe883108, 0xbe8a310a, + 0xbe8c310c, 0xbe8e310e, + 0xbf06807c, 0xbf84fff0, + 0xba80f801, 0x00000000, + 0xbf8a0000, 0xb9783a05, + 0x80788178, 0xbf0d9972, + 0xbf850002, 0x8f788978, + 0xbf820001, 0x8f788a78, + 0xb96e1e06, 0x8f6e8a6e, + 0x80786e78, 0x8078ff78, + 0x00000200, 0xbef603ff, + 0x01000000, 0xf4211bfa, 0xf0000000, 0x80788478, - 0xf4211b7a, 0xf0000000, - 0x80788478, 0xf4211c3a, + 0xf4211b3a, 0xf0000000, + 0x80788478, 0xf4211b7a, 0xf0000000, 0x80788478, - 0xf4211c7a, 0xf0000000, - 0x80788478, 0xf4211eba, + 0xf4211c3a, 0xf0000000, + 0x80788478, 0xf4211c7a, 0xf0000000, 0x80788478, - 0xf4211efa, 0xf0000000, - 0x80788478, 0xf4211e7a, + 0xf4211eba, 0xf0000000, + 0x80788478, 0xf4211efa, 0xf0000000, 0x80788478, - 0xf4211cfa, 0xf0000000, - 0x80788478, 0xf4211bba, + 0xf4211e7a, 0xf0000000, + 0x80788478, 0xf4211cfa, 0xf0000000, 0x80788478, - 0xbf8cc07f, 0xb9eef814, 0xf4211bba, 0xf0000000, 0x80788478, 0xbf8cc07f, - 0xb9eef815, 0xbefc036f, - 0xbefe0370, 0xbeff0371, - 0x876f7bff, 0x000003ff, - 0xb9ef4803, 0x876f7bff, - 0xfffff800, 0x906f8b6f, - 0xb9efa2c3, 0xb9f3f801, - 0xb96e3a05, 0x806e816e, - 0xbf0d9972, 0xbf850002, - 0x8f6e896e, 0xbf820001, - 0x8f6e8a6e, 0xb96f1e06, - 0x8f6f8a6f, 0x806e6f6e, - 0x806eff6e, 0x00000200, - 0x806e746e, 0x826f8075, - 0x876fff6f, 0x0000ffff, - 0xf4091c37, 0xfa000050, - 0xf4091d37, 0xfa000060, - 0xf4011e77, 0xfa000074, - 0xbf8cc07f, 0x876dff6d, - 0x0000ffff, 0x87fe7e7e, - 0x87ea6a6a, 0xb9faf802, - 0xbe80226c, 0xbf810000, + 0xb9eef814, 0xf4211bba, + 0xf0000000, 0x80788478, + 0xbf8cc07f, 0xb9eef815, + 0xbefc036f, 0xbefe0370, + 0xbeff0371, 0x876f7bff, + 0x000003ff, 0xb9ef4803, + 0x876f7bff, 0xfffff800, + 0x906f8b6f, 0xb9efa2c3, + 0xb9f3f801, 0xb96e3a05, + 0x806e816e, 0xbf0d9972, + 0xbf850002, 0x8f6e896e, + 0xbf820001, 0x8f6e8a6e, + 0xb96f1e06, 0x8f6f8a6f, + 0x806e6f6e, 0x806eff6e, + 0x00000200, 0x806e746e, + 0x826f8075, 0x876fff6f, + 0x0000ffff, 0xf4091c37, + 0xfa000050, 0xf4091d37, + 0xfa000060, 0xf4011e77, + 0xfa000074, 0xbf8cc07f, + 0x876dff6d, 0x0000ffff, + 0x87fe7e7e, 0x87ea6a6a, + 0xb9faf802, 0xbe80226c, + 0xbf9b0000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, - 0xbf9f0000, 0x00000000, }; static const uint32_t cwsr_trap_gfx11_hex[] = { @@ -2944,7 +2944,7 @@ static const uint32_t cwsr_trap_gfx11_hex[] = { 0xb8eef802, 0xbf0d866e, 0xbfa20002, 0xb97af802, 0xbe80486c, 0xb97af802, - 0xbe804a6c, 0xbfb00000, + 0xbe804a6c, 0xbfb10000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, 0x00000000, @@ -3436,5 +3436,5 @@ static const uint32_t cwsr_trap_gfx9_4_3_hex[] = { 0x86ea6a6a, 0x8f6e837a, 0xb96ee0c2, 0xbf800002, 0xb97a0002, 0xbf8a0000, - 0xbe801f6c, 0xbf810000, + 0xbe801f6c, 0xbf9b0000, }; diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm index fdab64624422..71b3dc0c7363 100644 --- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm +++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm @@ -369,6 +369,12 @@ L_SLEEP: s_or_b32 s_save_pc_hi, s_save_pc_hi, s_save_tmp #if NO_SQC_STORE +#if ASIC_FAMILY <= CHIP_SIENNA_CICHLID + // gfx10: If there was a VALU exception, the exception state must be + // cleared before executing the VALU instructions below. + v_clrexcp +#endif + // Trap temporaries must be saved via VGPR but all VGPRs are in use. // There is no ttmp space to hold the resource constant for VGPR save. // Save v0 by itself since it requires only two SGPRs. @@ -1098,7 +1104,7 @@ L_RETURN_WITHOUT_PRIV: s_rfe_b64 s_restore_pc_lo //Return to the main shader program and resume execution L_END_PGM: - s_endpgm + s_endpgm_saved end function write_hwreg_to_mem(s, s_rsrc, s_mem_offset) diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm index e506411ad28a..bb26338204f4 100644 --- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm +++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm @@ -921,7 +921,7 @@ L_RESTORE: /* the END */ /**************************************************************************/ L_END_PGM: - s_endpgm + s_endpgm_saved end diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index f6d4748c1980..80e90fdef291 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -1442,7 +1442,9 @@ static int kfd_ioctl_unmap_memory_from_gpu(struct file *filep, kfd_flush_tlb(peer_pdd, TLB_FLUSH_HEAVYWEIGHT); /* Remove dma mapping after tlb flush to avoid IO_PAGE_FAULT */ - amdgpu_amdkfd_gpuvm_dmaunmap_mem(mem, peer_pdd->drm_priv); + err = amdgpu_amdkfd_gpuvm_dmaunmap_mem(mem, peer_pdd->drm_priv); + if (err) + goto sync_memory_failed; } mutex_unlock(&p->mutex); @@ -1564,16 +1566,11 @@ static int kfd_ioctl_import_dmabuf(struct file *filep, { struct kfd_ioctl_import_dmabuf_args *args = data; struct kfd_process_device *pdd; - struct dma_buf *dmabuf; int idr_handle; uint64_t size; void *mem; int r; - dmabuf = dma_buf_get(args->dmabuf_fd); - if (IS_ERR(dmabuf)) - return PTR_ERR(dmabuf); - mutex_lock(&p->mutex); pdd = kfd_process_device_data_by_id(p, args->gpu_id); if (!pdd) { @@ -1587,10 +1584,10 @@ static int kfd_ioctl_import_dmabuf(struct file *filep, goto err_unlock; } - r = amdgpu_amdkfd_gpuvm_import_dmabuf(pdd->dev->adev, dmabuf, - args->va_addr, pdd->drm_priv, - (struct kgd_mem **)&mem, &size, - NULL); + r = amdgpu_amdkfd_gpuvm_import_dmabuf_fd(pdd->dev->adev, args->dmabuf_fd, + args->va_addr, pdd->drm_priv, + (struct kgd_mem **)&mem, &size, + NULL); if (r) goto err_unlock; @@ -1601,7 +1598,6 @@ static int kfd_ioctl_import_dmabuf(struct file *filep, } mutex_unlock(&p->mutex); - dma_buf_put(dmabuf); args->handle = MAKE_HANDLE(args->gpu_id, idr_handle); @@ -1612,7 +1608,6 @@ err_free: pdd->drm_priv, NULL); err_unlock: mutex_unlock(&p->mutex); - dma_buf_put(dmabuf); return r; } @@ -1855,8 +1850,8 @@ static uint32_t get_process_num_bos(struct kfd_process *p) return num_of_bos; } -static int criu_get_prime_handle(struct kgd_mem *mem, int flags, - u32 *shared_fd) +static int criu_get_prime_handle(struct kgd_mem *mem, + int flags, u32 *shared_fd) { struct dma_buf *dmabuf; int ret; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c index 0f58be65132f..739721254a5d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c @@ -880,6 +880,10 @@ static int copy_signaled_event_data(uint32_t num_events, dst = &data[i].memory_exception_data; src = &event->memory_exception_data; size = sizeof(struct kfd_hsa_memory_exception_data); + } else if (event->type == KFD_EVENT_TYPE_HW_EXCEPTION) { + dst = &data[i].memory_exception_data; + src = &event->hw_exception_data; + size = sizeof(struct kfd_hsa_hw_exception_data); } else if (event->type == KFD_EVENT_TYPE_SIGNAL && waiter->event_age_enabled) { dst = &data[i].signal_event_data.last_event_age; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c index 62b205dac63a..6604a3f99c5e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c @@ -330,12 +330,6 @@ static void kfd_init_apertures_vi(struct kfd_process_device *pdd, uint8_t id) pdd->gpuvm_limit = pdd->dev->kfd->shared_resources.gpuvm_size - 1; - /* dGPUs: the reserved space for kernel - * before SVM - */ - pdd->qpd.cwsr_base = SVM_CWSR_BASE; - pdd->qpd.ib_base = SVM_IB_BASE; - pdd->scratch_base = MAKE_SCRATCH_APP_BASE_VI(); pdd->scratch_limit = MAKE_SCRATCH_APP_LIMIT(pdd->scratch_base); } @@ -345,18 +339,18 @@ static void kfd_init_apertures_v9(struct kfd_process_device *pdd, uint8_t id) pdd->lds_base = MAKE_LDS_APP_BASE_V9(); pdd->lds_limit = MAKE_LDS_APP_LIMIT(pdd->lds_base); - pdd->gpuvm_base = PAGE_SIZE; + /* Raven needs SVM to support graphic handle, etc. Leave the small + * reserved space before SVM on Raven as well, even though we don't + * have to. + * Set gpuvm_base and gpuvm_limit to CANONICAL addresses so that they + * are used in Thunk to reserve SVM. + */ + pdd->gpuvm_base = SVM_USER_BASE; pdd->gpuvm_limit = pdd->dev->kfd->shared_resources.gpuvm_size - 1; pdd->scratch_base = MAKE_SCRATCH_APP_BASE_V9(); pdd->scratch_limit = MAKE_SCRATCH_APP_LIMIT(pdd->scratch_base); - - /* - * Place TBA/TMA on opposite side of VM hole to prevent - * stray faults from triggering SVM on these pages. - */ - pdd->qpd.cwsr_base = pdd->dev->kfd->shared_resources.gpuvm_size; } int kfd_init_apertures(struct kfd_process *process) @@ -413,6 +407,12 @@ int kfd_init_apertures(struct kfd_process *process) return -EINVAL; } } + + /* dGPUs: the reserved space for kernel + * before SVM + */ + pdd->qpd.cwsr_base = SVM_CWSR_BASE; + pdd->qpd.ib_base = SVM_IB_BASE; } dev_dbg(kfd_device, "node id %u\n", id); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 6c25dab051d5..bdc01ca9609a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -260,19 +260,6 @@ static void svm_migrate_put_sys_page(unsigned long addr) put_page(page); } -static unsigned long svm_migrate_successful_pages(struct migrate_vma *migrate) -{ - unsigned long cpages = 0; - unsigned long i; - - for (i = 0; i < migrate->npages; i++) { - if (migrate->src[i] & MIGRATE_PFN_VALID && - migrate->src[i] & MIGRATE_PFN_MIGRATE) - cpages++; - } - return cpages; -} - static unsigned long svm_migrate_unsuccessful_pages(struct migrate_vma *migrate) { unsigned long upages = 0; @@ -402,6 +389,7 @@ svm_migrate_vma_to_vram(struct kfd_node *node, struct svm_range *prange, struct dma_fence *mfence = NULL; struct migrate_vma migrate = { 0 }; unsigned long cpages = 0; + unsigned long mpages = 0; dma_addr_t *scratch; void *buf; int r = -ENOMEM; @@ -442,20 +430,21 @@ svm_migrate_vma_to_vram(struct kfd_node *node, struct svm_range *prange, goto out_free; } if (cpages != npages) - pr_debug("partial migration, 0x%lx/0x%llx pages migrated\n", + pr_debug("partial migration, 0x%lx/0x%llx pages collected\n", cpages, npages); else - pr_debug("0x%lx pages migrated\n", cpages); + pr_debug("0x%lx pages collected\n", cpages); r = svm_migrate_copy_to_vram(node, prange, &migrate, &mfence, scratch, ttm_res_offset); migrate_vma_pages(&migrate); - pr_debug("successful/cpages/npages 0x%lx/0x%lx/0x%lx\n", - svm_migrate_successful_pages(&migrate), cpages, migrate.npages); - svm_migrate_copy_done(adev, mfence); migrate_vma_finalize(&migrate); + mpages = cpages - svm_migrate_unsuccessful_pages(&migrate); + pr_debug("successful/cpages/npages 0x%lx/0x%lx/0x%lx\n", + mpages, cpages, migrate.npages); + kfd_smi_event_migration_end(node, p->lead_thread->pid, start >> PAGE_SHIFT, end >> PAGE_SHIFT, 0, node->id, trigger); @@ -465,12 +454,12 @@ svm_migrate_vma_to_vram(struct kfd_node *node, struct svm_range *prange, out_free: kvfree(buf); out: - if (!r && cpages) { + if (!r && mpages) { pdd = svm_range_get_pdd_by_node(prange, node); if (pdd) - WRITE_ONCE(pdd->page_in, pdd->page_in + cpages); + WRITE_ONCE(pdd->page_in, pdd->page_in + mpages); - return cpages; + return mpages; } return r; } @@ -479,6 +468,8 @@ out: * svm_migrate_ram_to_vram - migrate svm range from system to device * @prange: range structure * @best_loc: the device to migrate to + * @start_mgr: start page to migrate + * @last_mgr: last page to migrate * @mm: the process mm structure * @trigger: reason of migration * @@ -489,19 +480,20 @@ out: */ static int svm_migrate_ram_to_vram(struct svm_range *prange, uint32_t best_loc, + unsigned long start_mgr, unsigned long last_mgr, struct mm_struct *mm, uint32_t trigger) { unsigned long addr, start, end; struct vm_area_struct *vma; uint64_t ttm_res_offset; struct kfd_node *node; - unsigned long cpages = 0; + unsigned long mpages = 0; long r = 0; - if (prange->actual_loc == best_loc) { - pr_debug("svms 0x%p [0x%lx 0x%lx] already on best_loc 0x%x\n", - prange->svms, prange->start, prange->last, best_loc); - return 0; + if (start_mgr < prange->start || last_mgr > prange->last) { + pr_debug("range [0x%lx 0x%lx] out prange [0x%lx 0x%lx]\n", + start_mgr, last_mgr, prange->start, prange->last); + return -EFAULT; } node = svm_range_get_node_by_id(prange, best_loc); @@ -510,18 +502,19 @@ svm_migrate_ram_to_vram(struct svm_range *prange, uint32_t best_loc, return -ENODEV; } - pr_debug("svms 0x%p [0x%lx 0x%lx] to gpu 0x%x\n", prange->svms, - prange->start, prange->last, best_loc); + pr_debug("svms 0x%p [0x%lx 0x%lx] in [0x%lx 0x%lx] to gpu 0x%x\n", + prange->svms, start_mgr, last_mgr, prange->start, prange->last, + best_loc); - start = prange->start << PAGE_SHIFT; - end = (prange->last + 1) << PAGE_SHIFT; + start = start_mgr << PAGE_SHIFT; + end = (last_mgr + 1) << PAGE_SHIFT; r = svm_range_vram_node_new(node, prange, true); if (r) { dev_dbg(node->adev->dev, "fail %ld to alloc vram\n", r); return r; } - ttm_res_offset = prange->offset << PAGE_SHIFT; + ttm_res_offset = (start_mgr - prange->start + prange->offset) << PAGE_SHIFT; for (addr = start; addr < end;) { unsigned long next; @@ -536,16 +529,19 @@ svm_migrate_ram_to_vram(struct svm_range *prange, uint32_t best_loc, pr_debug("failed %ld to migrate\n", r); break; } else { - cpages += r; + mpages += r; } ttm_res_offset += next - addr; addr = next; } - if (cpages) { + if (mpages) { prange->actual_loc = best_loc; - svm_range_dma_unmap(prange); - } else { + prange->vram_pages += mpages; + } else if (!prange->actual_loc) { + /* if no page migrated and all pages from prange are at + * sys ram drop svm_bo got from svm_range_vram_node_new + */ svm_range_vram_node_free(prange); } @@ -578,7 +574,7 @@ svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange, pr_debug("svms 0x%p [0x%lx 0x%lx]\n", prange->svms, prange->start, prange->last); - addr = prange->start << PAGE_SHIFT; + addr = migrate->start; src = (uint64_t *)(scratch + npages); dst = scratch; @@ -663,9 +659,8 @@ out_oom: * Context: Process context, caller hold mmap read lock, prange->migrate_mutex * * Return: - * 0 - success with all pages migrated * negative values - indicate error - * positive values - partial migration, number of pages not migrated + * positive values or zero - number of pages got migrated */ static long svm_migrate_vma_to_ram(struct kfd_node *node, struct svm_range *prange, @@ -676,6 +671,7 @@ svm_migrate_vma_to_ram(struct kfd_node *node, struct svm_range *prange, uint64_t npages = (end - start) >> PAGE_SHIFT; unsigned long upages = npages; unsigned long cpages = 0; + unsigned long mpages = 0; struct amdgpu_device *adev = node->adev; struct kfd_process_device *pdd; struct dma_fence *mfence = NULL; @@ -725,10 +721,10 @@ svm_migrate_vma_to_ram(struct kfd_node *node, struct svm_range *prange, goto out_free; } if (cpages != npages) - pr_debug("partial migration, 0x%lx/0x%llx pages migrated\n", + pr_debug("partial migration, 0x%lx/0x%llx pages collected\n", cpages, npages); else - pr_debug("0x%lx pages migrated\n", cpages); + pr_debug("0x%lx pages collected\n", cpages); r = svm_migrate_copy_to_ram(adev, prange, &migrate, &mfence, scratch, npages); @@ -751,17 +747,21 @@ out_free: kvfree(buf); out: if (!r && cpages) { + mpages = cpages - upages; pdd = svm_range_get_pdd_by_node(prange, node); if (pdd) - WRITE_ONCE(pdd->page_out, pdd->page_out + cpages); + WRITE_ONCE(pdd->page_out, pdd->page_out + mpages); } - return r ? r : upages; + + return r ? r : mpages; } /** * svm_migrate_vram_to_ram - migrate svm range from device to system * @prange: range structure * @mm: process mm, use current->mm if NULL + * @start_mgr: start page need be migrated to sys ram + * @last_mgr: last page need be migrated to sys ram * @trigger: reason of migration * @fault_page: is from vmf->page, svm_migrate_to_ram(), this is CPU page fault callback * @@ -771,6 +771,7 @@ out: * 0 - OK, otherwise error code */ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm, + unsigned long start_mgr, unsigned long last_mgr, uint32_t trigger, struct page *fault_page) { struct kfd_node *node; @@ -778,26 +779,33 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm, unsigned long addr; unsigned long start; unsigned long end; - unsigned long upages = 0; + unsigned long mpages = 0; long r = 0; + /* this pragne has no any vram page to migrate to sys ram */ if (!prange->actual_loc) { pr_debug("[0x%lx 0x%lx] already migrated to ram\n", prange->start, prange->last); return 0; } + if (start_mgr < prange->start || last_mgr > prange->last) { + pr_debug("range [0x%lx 0x%lx] out prange [0x%lx 0x%lx]\n", + start_mgr, last_mgr, prange->start, prange->last); + return -EFAULT; + } + node = svm_range_get_node_by_id(prange, prange->actual_loc); if (!node) { pr_debug("failed to get kfd node by id 0x%x\n", prange->actual_loc); return -ENODEV; } pr_debug("svms 0x%p prange 0x%p [0x%lx 0x%lx] from gpu 0x%x to ram\n", - prange->svms, prange, prange->start, prange->last, + prange->svms, prange, start_mgr, last_mgr, prange->actual_loc); - start = prange->start << PAGE_SHIFT; - end = (prange->last + 1) << PAGE_SHIFT; + start = start_mgr << PAGE_SHIFT; + end = (last_mgr + 1) << PAGE_SHIFT; for (addr = start; addr < end;) { unsigned long next; @@ -816,14 +824,21 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm, pr_debug("failed %ld to migrate prange %p\n", r, prange); break; } else { - upages += r; + mpages += r; } addr = next; } - if (r >= 0 && !upages) { - svm_range_vram_node_free(prange); - prange->actual_loc = 0; + if (r >= 0) { + prange->vram_pages -= mpages; + + /* prange does not have vram page set its actual_loc to system + * and drop its svm_bo ref + */ + if (prange->vram_pages == 0 && prange->ttm_res) { + prange->actual_loc = 0; + svm_range_vram_node_free(prange); + } } return r < 0 ? r : 0; @@ -833,17 +848,23 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm, * svm_migrate_vram_to_vram - migrate svm range from device to device * @prange: range structure * @best_loc: the device to migrate to + * @start: start page need be migrated to sys ram + * @last: last page need be migrated to sys ram * @mm: process mm, use current->mm if NULL * @trigger: reason of migration * * Context: Process context, caller hold mmap read lock, svms lock, prange lock * + * migrate all vram pages in prange to sys ram, then migrate + * [start, last] pages from sys ram to gpu node best_loc. + * * Return: * 0 - OK, otherwise error code */ static int svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc, - struct mm_struct *mm, uint32_t trigger) + unsigned long start, unsigned long last, + struct mm_struct *mm, uint32_t trigger) { int r, retries = 3; @@ -855,7 +876,8 @@ svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc, pr_debug("from gpu 0x%x to gpu 0x%x\n", prange->actual_loc, best_loc); do { - r = svm_migrate_vram_to_ram(prange, mm, trigger, NULL); + r = svm_migrate_vram_to_ram(prange, mm, prange->start, prange->last, + trigger, NULL); if (r) return r; } while (prange->actual_loc && --retries); @@ -863,17 +885,21 @@ svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc, if (prange->actual_loc) return -EDEADLK; - return svm_migrate_ram_to_vram(prange, best_loc, mm, trigger); + return svm_migrate_ram_to_vram(prange, best_loc, start, last, mm, trigger); } int svm_migrate_to_vram(struct svm_range *prange, uint32_t best_loc, + unsigned long start, unsigned long last, struct mm_struct *mm, uint32_t trigger) { - if (!prange->actual_loc) - return svm_migrate_ram_to_vram(prange, best_loc, mm, trigger); + if (!prange->actual_loc || prange->actual_loc == best_loc) + return svm_migrate_ram_to_vram(prange, best_loc, start, last, + mm, trigger); + else - return svm_migrate_vram_to_vram(prange, best_loc, mm, trigger); + return svm_migrate_vram_to_vram(prange, best_loc, start, last, + mm, trigger); } @@ -889,10 +915,9 @@ svm_migrate_to_vram(struct svm_range *prange, uint32_t best_loc, */ static vm_fault_t svm_migrate_to_ram(struct vm_fault *vmf) { + unsigned long start, last, size; unsigned long addr = vmf->address; struct svm_range_bo *svm_bo; - enum svm_work_list_ops op; - struct svm_range *parent; struct svm_range *prange; struct kfd_process *p; struct mm_struct *mm; @@ -929,51 +954,31 @@ static vm_fault_t svm_migrate_to_ram(struct vm_fault *vmf) mutex_lock(&p->svms.lock); - prange = svm_range_from_addr(&p->svms, addr, &parent); + prange = svm_range_from_addr(&p->svms, addr, NULL); if (!prange) { pr_debug("failed get range svms 0x%p addr 0x%lx\n", &p->svms, addr); r = -EFAULT; goto out_unlock_svms; } - mutex_lock(&parent->migrate_mutex); - if (prange != parent) - mutex_lock_nested(&prange->migrate_mutex, 1); + mutex_lock(&prange->migrate_mutex); if (!prange->actual_loc) goto out_unlock_prange; - svm_range_lock(parent); - if (prange != parent) - mutex_lock_nested(&prange->lock, 1); - r = svm_range_split_by_granularity(p, mm, addr, parent, prange); - if (prange != parent) - mutex_unlock(&prange->lock); - svm_range_unlock(parent); - if (r) { - pr_debug("failed %d to split range by granularity\n", r); - goto out_unlock_prange; - } + /* Align migration range start and size to granularity size */ + size = 1UL << prange->granularity; + start = max(ALIGN_DOWN(addr, size), prange->start); + last = min(ALIGN(addr + 1, size) - 1, prange->last); - r = svm_migrate_vram_to_ram(prange, vmf->vma->vm_mm, - KFD_MIGRATE_TRIGGER_PAGEFAULT_CPU, - vmf->page); + r = svm_migrate_vram_to_ram(prange, vmf->vma->vm_mm, start, last, + KFD_MIGRATE_TRIGGER_PAGEFAULT_CPU, vmf->page); if (r) pr_debug("failed %d migrate svms 0x%p range 0x%p [0x%lx 0x%lx]\n", - r, prange->svms, prange, prange->start, prange->last); - - /* xnack on, update mapping on GPUs with ACCESS_IN_PLACE */ - if (p->xnack_enabled && parent == prange) - op = SVM_OP_UPDATE_RANGE_NOTIFIER_AND_MAP; - else - op = SVM_OP_UPDATE_RANGE_NOTIFIER; - svm_range_add_list_work(&p->svms, parent, mm, op); - schedule_deferred_list_work(&p->svms); + r, prange->svms, prange, start, last); out_unlock_prange: - if (prange != parent) - mutex_unlock(&prange->migrate_mutex); - mutex_unlock(&parent->migrate_mutex); + mutex_unlock(&prange->migrate_mutex); out_unlock_svms: mutex_unlock(&p->svms.lock); out_unref_process: @@ -1021,7 +1026,7 @@ int kgd2kfd_init_zone_device(struct amdgpu_device *adev) } else { res = devm_request_free_mem_region(adev->dev, &iomem_resource, size); if (IS_ERR(res)) - return -ENOMEM; + return PTR_ERR(res); pgmap->range.start = res->start; pgmap->range.end = res->end; pgmap->type = MEMORY_DEVICE_PRIVATE; @@ -1037,10 +1042,10 @@ int kgd2kfd_init_zone_device(struct amdgpu_device *adev) r = devm_memremap_pages(adev->dev, pgmap); if (IS_ERR(r)) { pr_err("failed to register HMM device memory\n"); - /* Disable SVM support capability */ - pgmap->type = 0; if (pgmap->type == MEMORY_DEVICE_PRIVATE) devm_release_mem_region(adev->dev, res->start, resource_size(res)); + /* Disable SVM support capability */ + pgmap->type = 0; return PTR_ERR(r); } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h index 487f26368164..2eebf67f9c2c 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h @@ -41,9 +41,13 @@ enum MIGRATION_COPY_DIR { }; int svm_migrate_to_vram(struct svm_range *prange, uint32_t best_loc, + unsigned long start, unsigned long last, struct mm_struct *mm, uint32_t trigger); + int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm, + unsigned long start, unsigned long last, uint32_t trigger, struct page *fault_page); + unsigned long svm_migrate_addr_to_pfn(struct amdgpu_device *adev, unsigned long addr); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c index 8b7fed913526..22cbfa1bdadd 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c @@ -170,6 +170,7 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, m->cp_hqd_pq_control = 5 << CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE__SHIFT; m->cp_hqd_pq_control |= ffs(q->queue_size / sizeof(unsigned int)) - 1 - 1; + m->cp_hqd_pq_control |= CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK; pr_debug("cp_hqd_pq_control 0x%x\n", m->cp_hqd_pq_control); m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c index 15277f1d5cf0..826bc4f6c8a7 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c @@ -55,8 +55,8 @@ static void update_cu_mask(struct mqd_manager *mm, void *mqd, m = get_mqd(mqd); if (has_wa_flag) { - uint32_t wa_mask = minfo->update_flag == UPDATE_FLAG_DBG_WA_ENABLE ? - 0xffff : 0xffffffff; + uint32_t wa_mask = + (minfo->update_flag & UPDATE_FLAG_DBG_WA_ENABLE) ? 0xffff : 0xffffffff; m->compute_static_thread_mgmt_se0 = wa_mask; m->compute_static_thread_mgmt_se1 = wa_mask; @@ -224,6 +224,7 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, m->cp_hqd_pq_control = 5 << CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE__SHIFT; m->cp_hqd_pq_control |= ffs(q->queue_size / sizeof(unsigned int)) - 1 - 1; + m->cp_hqd_pq_control |= CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK; pr_debug("cp_hqd_pq_control 0x%x\n", m->cp_hqd_pq_control); m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c index 42d881809dc7..697b6d530d12 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c @@ -303,6 +303,15 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, update_cu_mask(mm, mqd, minfo, 0); set_priority(m, q); + if (minfo && KFD_GC_VERSION(mm->dev) >= IP_VERSION(9, 4, 2)) { + if (minfo->update_flag & UPDATE_FLAG_IS_GWS) + m->compute_resource_limits |= + COMPUTE_RESOURCE_LIMITS__FORCE_SIMD_DIST_MASK; + else + m->compute_resource_limits &= + ~COMPUTE_RESOURCE_LIMITS__FORCE_SIMD_DIST_MASK; + } + q->is_active = QUEUE_IS_ACTIVE(*q); } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index 4c8e278a0d0c..80320b8603fc 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -532,6 +532,7 @@ struct queue_properties { enum mqd_update_flag { UPDATE_FLAG_DBG_WA_ENABLE = 1, UPDATE_FLAG_DBG_WA_DISABLE = 2, + UPDATE_FLAG_IS_GWS = 4, /* quirk for gfx9 IP */ }; struct mqd_update_info { @@ -748,7 +749,6 @@ struct kfd_process_device { /* VM context for GPUVM allocations */ struct file *drm_file; void *drm_priv; - atomic64_t tlb_seq; /* GPUVM allocations storage */ struct idr alloc_idr; @@ -918,7 +918,7 @@ struct kfd_process { * fence will be triggered during eviction and new one will be created * during restore */ - struct dma_fence *ef; + struct dma_fence __rcu *ef; /* Work items for evicting and restoring BOs */ struct delayed_work eviction_work; @@ -971,7 +971,7 @@ struct kfd_process { struct work_struct debug_event_workarea; /* Tracks debug per-vmid request for debug flags */ - bool dbg_flags; + u32 dbg_flags; atomic_t poison; /* Queues are in paused stated because we are in the process of doing a CRIU checkpoint */ @@ -1462,7 +1462,14 @@ void kfd_signal_reset_event(struct kfd_node *dev); void kfd_signal_poison_consumed_event(struct kfd_node *dev, u32 pasid); -void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type); +static inline void kfd_flush_tlb(struct kfd_process_device *pdd, + enum TLB_FLUSH_TYPE type) +{ + struct amdgpu_device *adev = pdd->dev->adev; + struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv); + + amdgpu_vm_flush_compute_tlb(adev, vm, type, pdd->dev->xcc_mask); +} static inline bool kfd_flush_tlb_after_unmap(struct kfd_dev *dev) { @@ -1482,10 +1489,15 @@ void kfd_dec_compute_active(struct kfd_node *dev); /* Cgroup Support */ /* Check with device cgroup if @kfd device is accessible */ -static inline int kfd_devcgroup_check_permission(struct kfd_node *kfd) +static inline int kfd_devcgroup_check_permission(struct kfd_node *node) { #if defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF) - struct drm_device *ddev = adev_to_drm(kfd->adev); + struct drm_device *ddev; + + if (node->xcp) + ddev = node->xcp->ddev; + else + ddev = adev_to_drm(node->adev); return devcgroup_check_permission(DEVCG_DEV_CHAR, DRM_MAJOR, ddev->render->index, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index 7a33e06f5c90..717a60d7a4ea 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -664,7 +664,8 @@ int kfd_process_create_wq(void) if (!kfd_process_wq) kfd_process_wq = alloc_workqueue("kfd_process_wq", 0, 0); if (!kfd_restore_wq) - kfd_restore_wq = alloc_ordered_workqueue("kfd_restore_wq", 0); + kfd_restore_wq = alloc_ordered_workqueue("kfd_restore_wq", + WQ_FREEZABLE); if (!kfd_process_wq || !kfd_restore_wq) { kfd_process_destroy_wq(); @@ -1109,6 +1110,7 @@ static void kfd_process_wq_release(struct work_struct *work) { struct kfd_process *p = container_of(work, struct kfd_process, release_work); + struct dma_fence *ef; kfd_process_dequeue_from_all_devices(p); pqm_uninit(&p->pqm); @@ -1117,7 +1119,9 @@ static void kfd_process_wq_release(struct work_struct *work) * destroyed. This allows any BOs to be freed without * triggering pointless evictions or waiting for fences. */ - dma_fence_signal(p->ef); + synchronize_rcu(); + ef = rcu_access_pointer(p->ef); + dma_fence_signal(ef); kfd_process_remove_sysfs(p); @@ -1126,7 +1130,7 @@ static void kfd_process_wq_release(struct work_struct *work) svm_range_list_fini(p); kfd_process_destroy_pdds(p); - dma_fence_put(p->ef); + dma_fence_put(ef); kfd_event_free_process(p); @@ -1642,6 +1646,7 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd, struct amdgpu_fpriv *drv_priv; struct amdgpu_vm *avm; struct kfd_process *p; + struct dma_fence *ef; struct kfd_node *dev; int ret; @@ -1661,13 +1666,13 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd, ret = amdgpu_amdkfd_gpuvm_acquire_process_vm(dev->adev, avm, &p->kgd_process_info, - &p->ef); + &ef); if (ret) { pr_err("Failed to create process VM object\n"); return ret; } + RCU_INIT_POINTER(p->ef, ef); pdd->drm_priv = drm_file->private_data; - atomic64_set(&pdd->tlb_seq, 0); ret = kfd_process_device_reserve_ib_mem(pdd); if (ret) @@ -1909,6 +1914,21 @@ kfd_process_gpuid_from_node(struct kfd_process *p, struct kfd_node *node, return -EINVAL; } +static int signal_eviction_fence(struct kfd_process *p) +{ + struct dma_fence *ef; + int ret; + + rcu_read_lock(); + ef = dma_fence_get_rcu_safe(&p->ef); + rcu_read_unlock(); + + ret = dma_fence_signal(ef); + dma_fence_put(ef); + + return ret; +} + static void evict_process_worker(struct work_struct *work) { int ret; @@ -1921,31 +1941,46 @@ static void evict_process_worker(struct work_struct *work) * lifetime of this thread, kfd_process p will be valid */ p = container_of(dwork, struct kfd_process, eviction_work); - WARN_ONCE(p->last_eviction_seqno != p->ef->seqno, - "Eviction fence mismatch\n"); - - /* Narrow window of overlap between restore and evict work - * item is possible. Once amdgpu_amdkfd_gpuvm_restore_process_bos - * unreserves KFD BOs, it is possible to evicted again. But - * restore has few more steps of finish. So lets wait for any - * previous restore work to complete - */ - flush_delayed_work(&p->restore_work); pr_debug("Started evicting pasid 0x%x\n", p->pasid); ret = kfd_process_evict_queues(p, KFD_QUEUE_EVICTION_TRIGGER_TTM); if (!ret) { - dma_fence_signal(p->ef); - dma_fence_put(p->ef); - p->ef = NULL; - queue_delayed_work(kfd_restore_wq, &p->restore_work, + /* If another thread already signaled the eviction fence, + * they are responsible stopping the queues and scheduling + * the restore work. + */ + if (!signal_eviction_fence(p)) + queue_delayed_work(kfd_restore_wq, &p->restore_work, msecs_to_jiffies(PROCESS_RESTORE_TIME_MS)); + else + kfd_process_restore_queues(p); pr_debug("Finished evicting pasid 0x%x\n", p->pasid); } else pr_err("Failed to evict queues of pasid 0x%x\n", p->pasid); } +static int restore_process_helper(struct kfd_process *p) +{ + int ret = 0; + + /* VMs may not have been acquired yet during debugging. */ + if (p->kgd_process_info) { + ret = amdgpu_amdkfd_gpuvm_restore_process_bos( + p->kgd_process_info, &p->ef); + if (ret) + return ret; + } + + ret = kfd_process_restore_queues(p); + if (!ret) + pr_debug("Finished restoring pasid 0x%x\n", p->pasid); + else + pr_err("Failed to restore queues of pasid 0x%x\n", p->pasid); + + return ret; +} + static void restore_process_worker(struct work_struct *work) { struct delayed_work *dwork; @@ -1971,24 +2006,15 @@ static void restore_process_worker(struct work_struct *work) */ p->last_restore_timestamp = get_jiffies_64(); - /* VMs may not have been acquired yet during debugging. */ - if (p->kgd_process_info) - ret = amdgpu_amdkfd_gpuvm_restore_process_bos(p->kgd_process_info, - &p->ef); + + ret = restore_process_helper(p); if (ret) { pr_debug("Failed to restore BOs of pasid 0x%x, retry after %d ms\n", p->pasid, PROCESS_BACK_OFF_TIME_MS); ret = queue_delayed_work(kfd_restore_wq, &p->restore_work, msecs_to_jiffies(PROCESS_BACK_OFF_TIME_MS)); WARN(!ret, "reschedule restore work failed\n"); - return; } - - ret = kfd_process_restore_queues(p); - if (!ret) - pr_debug("Finished restoring pasid 0x%x\n", p->pasid); - else - pr_err("Failed to restore queues of pasid 0x%x\n", p->pasid); } void kfd_suspend_all_processes(void) @@ -1999,14 +2025,9 @@ void kfd_suspend_all_processes(void) WARN(debug_evictions, "Evicting all processes"); hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) { - cancel_delayed_work_sync(&p->eviction_work); - flush_delayed_work(&p->restore_work); - if (kfd_process_evict_queues(p, KFD_QUEUE_EVICTION_TRIGGER_SUSPEND)) pr_err("Failed to suspend process 0x%x\n", p->pasid); - dma_fence_signal(p->ef); - dma_fence_put(p->ef); - p->ef = NULL; + signal_eviction_fence(p); } srcu_read_unlock(&kfd_processes_srcu, idx); } @@ -2018,7 +2039,7 @@ int kfd_resume_all_processes(void) int ret = 0, idx = srcu_read_lock(&kfd_processes_srcu); hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) { - if (!queue_delayed_work(kfd_restore_wq, &p->restore_work, 0)) { + if (restore_process_helper(p)) { pr_err("Restore process %d failed during resume\n", p->pasid); ret = -EFAULT; @@ -2059,36 +2080,6 @@ int kfd_reserved_mem_mmap(struct kfd_node *dev, struct kfd_process *process, KFD_CWSR_TBA_TMA_SIZE, vma->vm_page_prot); } -void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type) -{ - struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv); - uint64_t tlb_seq = amdgpu_vm_tlb_seq(vm); - struct kfd_node *dev = pdd->dev; - uint32_t xcc_mask = dev->xcc_mask; - int xcc = 0; - - /* - * It can be that we race and lose here, but that is extremely unlikely - * and the worst thing which could happen is that we flush the changes - * into the TLB once more which is harmless. - */ - if (atomic64_xchg(&pdd->tlb_seq, tlb_seq) == tlb_seq) - return; - - if (dev->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) { - /* Nothing to flush until a VMID is assigned, which - * only happens when the first queue is created. - */ - if (pdd->qpd.vmid) - amdgpu_amdkfd_flush_gpu_tlb_vmid(dev->adev, - pdd->qpd.vmid); - } else { - for_each_inst(xcc, xcc_mask) - amdgpu_amdkfd_flush_gpu_tlb_pasid( - dev->adev, pdd->process->pasid, type, xcc); - } -} - /* assumes caller holds process lock. */ int kfd_process_drain_interrupts(struct kfd_process_device *pdd) { diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c index 77f493262e05..4858112f9a53 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c @@ -87,12 +87,15 @@ void kfd_process_dequeue_from_device(struct kfd_process_device *pdd) return; dev->dqm->ops.process_termination(dev->dqm, &pdd->qpd); + if (dev->kfd->shared_resources.enable_mes) + amdgpu_mes_flush_shader_debugger(dev->adev, pdd->proc_ctx_gpu_addr); pdd->already_dequeued = true; } int pqm_set_gws(struct process_queue_manager *pqm, unsigned int qid, void *gws) { + struct mqd_update_info minfo = {0}; struct kfd_node *dev = NULL; struct process_queue_node *pqn; struct kfd_process_device *pdd; @@ -144,9 +147,10 @@ int pqm_set_gws(struct process_queue_manager *pqm, unsigned int qid, } pdd->qpd.num_gws = gws ? dev->adev->gds.gws_size : 0; + minfo.update_flag = gws ? UPDATE_FLAG_IS_GWS : 0; return pqn->q->device->dqm->ops.update_queue(pqn->q->device->dqm, - pqn->q, NULL); + pqn->q, &minfo); } void kfd_process_dequeue_from_all_devices(struct kfd_process *p) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index a15bfb5223e8..c50a0dc9c9c0 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -198,6 +198,7 @@ svm_range_dma_map_dev(struct amdgpu_device *adev, struct svm_range *prange, pr_debug_ratelimited("dma mapping 0x%llx for page addr 0x%lx\n", addr[i] >> PAGE_SHIFT, page_to_pfn(page)); } + return 0; } @@ -349,6 +350,7 @@ svm_range *svm_range_new(struct svm_range_list *svms, uint64_t start, INIT_LIST_HEAD(&prange->child_list); atomic_set(&prange->invalid, 0); prange->validate_timestamp = 0; + prange->vram_pages = 0; mutex_init(&prange->migrate_mutex); mutex_init(&prange->lock); @@ -395,19 +397,16 @@ static void svm_range_bo_release(struct kref *kref) prange->start, prange->last); mutex_lock(&prange->lock); prange->svm_bo = NULL; + /* prange should not hold vram page now */ + WARN_ONCE(prange->actual_loc, "prange should not hold vram page"); mutex_unlock(&prange->lock); spin_lock(&svm_bo->list_lock); } spin_unlock(&svm_bo->list_lock); - if (!dma_fence_is_signaled(&svm_bo->eviction_fence->base)) { - /* We're not in the eviction worker. - * Signal the fence and synchronize with any - * pending eviction work. - */ + if (!dma_fence_is_signaled(&svm_bo->eviction_fence->base)) + /* We're not in the eviction worker. Signal the fence. */ dma_fence_signal(&svm_bo->eviction_fence->base); - cancel_work_sync(&svm_bo->eviction_work); - } dma_fence_put(&svm_bo->eviction_fence->base); amdgpu_bo_unref(&svm_bo->bo); kfree(svm_bo); @@ -878,14 +877,29 @@ static void svm_range_debug_dump(struct svm_range_list *svms) static void * svm_range_copy_array(void *psrc, size_t size, uint64_t num_elements, - uint64_t offset) + uint64_t offset, uint64_t *vram_pages) { + unsigned char *src = (unsigned char *)psrc + offset; unsigned char *dst; + uint64_t i; dst = kvmalloc_array(num_elements, size, GFP_KERNEL); if (!dst) return NULL; - memcpy(dst, (unsigned char *)psrc + offset, num_elements * size); + + if (!vram_pages) { + memcpy(dst, src, num_elements * size); + return (void *)dst; + } + + *vram_pages = 0; + for (i = 0; i < num_elements; i++) { + dma_addr_t *temp; + temp = (dma_addr_t *)dst + i; + *temp = *((dma_addr_t *)src + i); + if (*temp&SVM_RANGE_VRAM_DOMAIN) + (*vram_pages)++; + } return (void *)dst; } @@ -899,7 +913,7 @@ svm_range_copy_dma_addrs(struct svm_range *dst, struct svm_range *src) if (!src->dma_addr[i]) continue; dst->dma_addr[i] = svm_range_copy_array(src->dma_addr[i], - sizeof(*src->dma_addr[i]), src->npages, 0); + sizeof(*src->dma_addr[i]), src->npages, 0, NULL); if (!dst->dma_addr[i]) return -ENOMEM; } @@ -910,7 +924,7 @@ svm_range_copy_dma_addrs(struct svm_range *dst, struct svm_range *src) static int svm_range_split_array(void *ppnew, void *ppold, size_t size, uint64_t old_start, uint64_t old_n, - uint64_t new_start, uint64_t new_n) + uint64_t new_start, uint64_t new_n, uint64_t *new_vram_pages) { unsigned char *new, *old, *pold; uint64_t d; @@ -922,11 +936,12 @@ svm_range_split_array(void *ppnew, void *ppold, size_t size, return 0; d = (new_start - old_start) * size; - new = svm_range_copy_array(pold, size, new_n, d); + /* get dma addr array for new range and calculte its vram page number */ + new = svm_range_copy_array(pold, size, new_n, d, new_vram_pages); if (!new) return -ENOMEM; d = (new_start == old_start) ? new_n * size : 0; - old = svm_range_copy_array(pold, size, old_n, d); + old = svm_range_copy_array(pold, size, old_n, d, NULL); if (!old) { kvfree(new); return -ENOMEM; @@ -948,10 +963,13 @@ svm_range_split_pages(struct svm_range *new, struct svm_range *old, for (i = 0; i < MAX_GPU_INSTANCE; i++) { r = svm_range_split_array(&new->dma_addr[i], &old->dma_addr[i], sizeof(*old->dma_addr[i]), old->start, - npages, new->start, new->npages); + npages, new->start, new->npages, + old->actual_loc ? &new->vram_pages : NULL); if (r) return r; } + if (old->actual_loc) + old->vram_pages -= new->vram_pages; return 0; } @@ -1097,7 +1115,7 @@ static int svm_range_split_tail(struct svm_range *prange, uint64_t new_last, struct list_head *insert_list, struct list_head *remap_list) { - struct svm_range *tail; + struct svm_range *tail = NULL; int r = svm_range_split(prange, prange->start, new_last, &tail); if (!r) { @@ -1112,7 +1130,7 @@ static int svm_range_split_head(struct svm_range *prange, uint64_t new_start, struct list_head *insert_list, struct list_head *remap_list) { - struct svm_range *head; + struct svm_range *head = NULL; int r = svm_range_split(prange, new_start, prange->last, &head); if (!r) { @@ -1135,66 +1153,6 @@ svm_range_add_child(struct svm_range *prange, struct mm_struct *mm, list_add_tail(&pchild->child_list, &prange->child_list); } -/** - * svm_range_split_by_granularity - collect ranges within granularity boundary - * - * @p: the process with svms list - * @mm: mm structure - * @addr: the vm fault address in pages, to split the prange - * @parent: parent range if prange is from child list - * @prange: prange to split - * - * Trims @prange to be a single aligned block of prange->granularity if - * possible. The head and tail are added to the child_list in @parent. - * - * Context: caller must hold mmap_read_lock and prange->lock - * - * Return: - * 0 - OK, otherwise error code - */ -int -svm_range_split_by_granularity(struct kfd_process *p, struct mm_struct *mm, - unsigned long addr, struct svm_range *parent, - struct svm_range *prange) -{ - struct svm_range *head, *tail; - unsigned long start, last, size; - int r; - - /* Align splited range start and size to granularity size, then a single - * PTE will be used for whole range, this reduces the number of PTE - * updated and the L1 TLB space used for translation. - */ - size = 1UL << prange->granularity; - start = ALIGN_DOWN(addr, size); - last = ALIGN(addr + 1, size) - 1; - - pr_debug("svms 0x%p split [0x%lx 0x%lx] to [0x%lx 0x%lx] size 0x%lx\n", - prange->svms, prange->start, prange->last, start, last, size); - - if (start > prange->start) { - r = svm_range_split(prange, start, prange->last, &head); - if (r) - return r; - svm_range_add_child(parent, mm, head, SVM_OP_ADD_RANGE); - } - - if (last < prange->last) { - r = svm_range_split(prange, prange->start, last, &tail); - if (r) - return r; - svm_range_add_child(parent, mm, tail, SVM_OP_ADD_RANGE); - } - - /* xnack on, update mapping on GPUs with ACCESS_IN_PLACE */ - if (p->xnack_enabled && prange->work_item.op == SVM_OP_ADD_RANGE) { - prange->work_item.op = SVM_OP_ADD_RANGE_AND_MAP; - pr_debug("change prange 0x%p [0x%lx 0x%lx] op %d\n", - prange, prange->start, prange->last, - SVM_OP_ADD_RANGE_AND_MAP); - } - return 0; -} static bool svm_nodes_in_same_hive(struct kfd_node *node_a, struct kfd_node *node_b) { @@ -1529,7 +1487,7 @@ static int svm_range_reserve_bos(struct svm_validate_context *ctx, bool intr) uint32_t gpuidx; int r; - drm_exec_init(&ctx->exec, intr ? DRM_EXEC_INTERRUPTIBLE_WAIT: 0); + drm_exec_init(&ctx->exec, intr ? DRM_EXEC_INTERRUPTIBLE_WAIT: 0, 0); drm_exec_until_all_locked(&ctx->exec) { for_each_set_bit(gpuidx, ctx->bitmap, MAX_GPU_INSTANCE) { pdd = kfd_process_device_from_gpuidx(ctx->process, gpuidx); @@ -1614,6 +1572,7 @@ static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx) * 5. Release page table (and SVM BO) reservation */ static int svm_range_validate_and_map(struct mm_struct *mm, + unsigned long map_start, unsigned long map_last, struct svm_range *prange, int32_t gpuidx, bool intr, bool wait, bool flush_tlb) { @@ -1694,10 +1653,12 @@ static int svm_range_validate_and_map(struct mm_struct *mm, } } - start = prange->start << PAGE_SHIFT; - end = (prange->last + 1) << PAGE_SHIFT; + start = map_start << PAGE_SHIFT; + end = (map_last + 1) << PAGE_SHIFT; for (addr = start; !r && addr < end; ) { struct hmm_range *hmm_range; + unsigned long map_start_vma; + unsigned long map_last_vma; struct vm_area_struct *vma; unsigned long next = 0; unsigned long offset; @@ -1725,7 +1686,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm, } if (!r) { - offset = (addr - start) >> PAGE_SHIFT; + offset = (addr >> PAGE_SHIFT) - prange->start; r = svm_range_dma_map(prange, ctx->bitmap, offset, npages, hmm_range->hmm_pfns); if (r) @@ -1743,9 +1704,16 @@ static int svm_range_validate_and_map(struct mm_struct *mm, r = -EAGAIN; } - if (!r) - r = svm_range_map_to_gpus(prange, offset, npages, readonly, - ctx->bitmap, wait, flush_tlb); + if (!r) { + map_start_vma = max(map_start, prange->start + offset); + map_last_vma = min(map_last, prange->start + offset + npages - 1); + if (map_start_vma <= map_last_vma) { + offset = map_start_vma - prange->start; + npages = map_last_vma - map_start_vma + 1; + r = svm_range_map_to_gpus(prange, offset, npages, readonly, + ctx->bitmap, wait, flush_tlb); + } + } if (!r && next == end) prange->mapped_to_gpu = true; @@ -1838,8 +1806,8 @@ static void svm_range_restore_work(struct work_struct *work) */ mutex_lock(&prange->migrate_mutex); - r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE, - false, true, false); + r = svm_range_validate_and_map(mm, prange->start, prange->last, prange, + MAX_GPU_INSTANCE, false, true, false); if (r) pr_debug("failed %d to map 0x%lx to gpus\n", r, prange->start); @@ -1876,7 +1844,7 @@ out_reschedule: /* If validation failed, reschedule another attempt */ if (evicted_ranges) { pr_debug("reschedule to restore svm range\n"); - schedule_delayed_work(&svms->restore_work, + queue_delayed_work(system_freezable_wq, &svms->restore_work, msecs_to_jiffies(AMDGPU_SVM_RANGE_RESTORE_DELAY_MS)); kfd_smi_event_queue_restore_rescheduled(mm); @@ -1952,7 +1920,7 @@ svm_range_evict(struct svm_range *prange, struct mm_struct *mm, pr_debug("failed to quiesce KFD\n"); pr_debug("schedule to restore svm %p ranges\n", svms); - schedule_delayed_work(&svms->restore_work, + queue_delayed_work(system_freezable_wq, &svms->restore_work, msecs_to_jiffies(AMDGPU_SVM_RANGE_RESTORE_DELAY_MS)); } else { unsigned long s, l; @@ -2007,6 +1975,7 @@ static struct svm_range *svm_range_clone(struct svm_range *old) new->actual_loc = old->actual_loc; new->granularity = old->granularity; new->mapped_to_gpu = old->mapped_to_gpu; + new->vram_pages = old->vram_pages; bitmap_copy(new->bitmap_access, old->bitmap_access, MAX_GPU_INSTANCE); bitmap_copy(new->bitmap_aip, old->bitmap_aip, MAX_GPU_INSTANCE); @@ -2371,8 +2340,10 @@ retry: mutex_unlock(&svms->lock); mmap_write_unlock(mm); - /* Pairs with mmget in svm_range_add_list_work */ - mmput(mm); + /* Pairs with mmget in svm_range_add_list_work. If dropping the + * last mm refcount, schedule release work to avoid circular locking + */ + mmput_async(mm); spin_lock(&svms->deferred_list_lock); } @@ -2683,6 +2654,7 @@ svm_range_get_range_boundaries(struct kfd_process *p, int64_t addr, { struct vm_area_struct *vma; struct interval_tree_node *node; + struct rb_node *rb_node; unsigned long start_limit, end_limit; vma = vma_lookup(p->mm, addr << PAGE_SHIFT); @@ -2702,16 +2674,15 @@ svm_range_get_range_boundaries(struct kfd_process *p, int64_t addr, if (node) { end_limit = min(end_limit, node->start); /* Last range that ends before the fault address */ - node = container_of(rb_prev(&node->rb), - struct interval_tree_node, rb); + rb_node = rb_prev(&node->rb); } else { /* Last range must end before addr because * there was no range after addr */ - node = container_of(rb_last(&p->svms.objects.rb_root), - struct interval_tree_node, rb); + rb_node = rb_last(&p->svms.objects.rb_root); } - if (node) { + if (rb_node) { + node = container_of(rb_node, struct interval_tree_node, rb); if (node->last >= addr) { WARN(1, "Overlap with prev node and page fault addr\n"); return -EFAULT; @@ -2914,6 +2885,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid, uint32_t vmid, uint32_t node_id, uint64_t addr, bool write_fault) { + unsigned long start, last, size; struct mm_struct *mm = NULL; struct svm_range_list *svms; struct svm_range *prange; @@ -3049,40 +3021,44 @@ retry_write_locked: kfd_smi_event_page_fault_start(node, p->lead_thread->pid, addr, write_fault, timestamp); - if (prange->actual_loc != best_loc) { + /* Align migration range start and size to granularity size */ + size = 1UL << prange->granularity; + start = max_t(unsigned long, ALIGN_DOWN(addr, size), prange->start); + last = min_t(unsigned long, ALIGN(addr + 1, size) - 1, prange->last); + if (prange->actual_loc != 0 || best_loc != 0) { migration = true; + if (best_loc) { - r = svm_migrate_to_vram(prange, best_loc, mm, - KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU); + r = svm_migrate_to_vram(prange, best_loc, start, last, + mm, KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU); if (r) { pr_debug("svm_migrate_to_vram failed (%d) at %llx, falling back to system memory\n", r, addr); /* Fallback to system memory if migration to * VRAM failed */ - if (prange->actual_loc) - r = svm_migrate_vram_to_ram(prange, mm, - KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU, - NULL); + if (prange->actual_loc && prange->actual_loc != best_loc) + r = svm_migrate_vram_to_ram(prange, mm, start, last, + KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU, NULL); else r = 0; } } else { - r = svm_migrate_vram_to_ram(prange, mm, - KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU, - NULL); + r = svm_migrate_vram_to_ram(prange, mm, start, last, + KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU, NULL); } if (r) { pr_debug("failed %d to migrate svms %p [0x%lx 0x%lx]\n", - r, svms, prange->start, prange->last); + r, svms, start, last); goto out_unlock_range; } } - r = svm_range_validate_and_map(mm, prange, gpuidx, false, false, false); + r = svm_range_validate_and_map(mm, start, last, prange, gpuidx, false, + false, false); if (r) pr_debug("failed %d to map svms 0x%p [0x%lx 0x%lx] to gpus\n", - r, svms, prange->start, prange->last); + r, svms, start, last); kfd_smi_event_page_fault_end(node, p->lead_thread->pid, addr, migration); @@ -3428,18 +3404,24 @@ svm_range_trigger_migration(struct mm_struct *mm, struct svm_range *prange, *migrated = false; best_loc = svm_range_best_prefetch_location(prange); - if (best_loc == KFD_IOCTL_SVM_LOCATION_UNDEFINED || - best_loc == prange->actual_loc) + /* when best_loc is a gpu node and same as prange->actual_loc + * we still need do migration as prange->actual_loc !=0 does + * not mean all pages in prange are vram. hmm migrate will pick + * up right pages during migration. + */ + if ((best_loc == KFD_IOCTL_SVM_LOCATION_UNDEFINED) || + (best_loc == 0 && prange->actual_loc == 0)) return 0; if (!best_loc) { - r = svm_migrate_vram_to_ram(prange, mm, + r = svm_migrate_vram_to_ram(prange, mm, prange->start, prange->last, KFD_MIGRATE_TRIGGER_PREFETCH, NULL); *migrated = !r; return r; } - r = svm_migrate_to_vram(prange, best_loc, mm, KFD_MIGRATE_TRIGGER_PREFETCH); + r = svm_migrate_to_vram(prange, best_loc, prange->start, prange->last, + mm, KFD_MIGRATE_TRIGGER_PREFETCH); *migrated = !r; return r; @@ -3447,13 +3429,14 @@ svm_range_trigger_migration(struct mm_struct *mm, struct svm_range *prange, int svm_range_schedule_evict_svm_bo(struct amdgpu_amdkfd_fence *fence) { - if (!fence) - return -EINVAL; - - if (dma_fence_is_signaled(&fence->base)) - return 0; - - if (fence->svm_bo) { + /* Dereferencing fence->svm_bo is safe here because the fence hasn't + * signaled yet and we're under the protection of the fence->lock. + * After the fence is signaled in svm_range_bo_release, we cannot get + * here any more. + * + * Reference is dropped in svm_range_evict_svm_bo_worker. + */ + if (svm_bo_ref_unless_zero(fence->svm_bo)) { WRITE_ONCE(fence->svm_bo->evicting, 1); schedule_work(&fence->svm_bo->eviction_work); } @@ -3468,8 +3451,6 @@ static void svm_range_evict_svm_bo_worker(struct work_struct *work) int r = 0; svm_bo = container_of(work, struct svm_range_bo, eviction_work); - if (!svm_bo_ref_unless_zero(svm_bo)) - return; /* svm_bo was freed while eviction was pending */ if (mmget_not_zero(svm_bo->eviction_fence->mm)) { mm = svm_bo->eviction_fence->mm; @@ -3494,7 +3475,11 @@ static void svm_range_evict_svm_bo_worker(struct work_struct *work) mutex_lock(&prange->migrate_mutex); do { + /* migrate all vram pages in this prange to sys ram + * after that prange->actual_loc should be zero + */ r = svm_migrate_vram_to_ram(prange, mm, + prange->start, prange->last, KFD_MIGRATE_TRIGGER_TTM_EVICTION, NULL); } while (!r && prange->actual_loc && --retries); @@ -3618,8 +3603,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, flush_tlb = !migrated && update_mapping && prange->mapped_to_gpu; - r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE, - true, true, flush_tlb); + r = svm_range_validate_and_map(mm, prange->start, prange->last, prange, + MAX_GPU_INSTANCE, true, true, flush_tlb); if (r) pr_debug("failed %d to map svm range\n", r); @@ -3633,8 +3618,8 @@ out_unlock_range: pr_debug("Remapping prange 0x%p [0x%lx 0x%lx]\n", prange, prange->start, prange->last); mutex_lock(&prange->migrate_mutex); - r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE, - true, true, prange->mapped_to_gpu); + r = svm_range_validate_and_map(mm, prange->start, prange->last, prange, + MAX_GPU_INSTANCE, true, true, prange->mapped_to_gpu); if (r) pr_debug("failed %d on remap svm range\n", r); mutex_unlock(&prange->migrate_mutex); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h index c528df1d0ba2..026863a0abcd 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h @@ -78,6 +78,7 @@ struct svm_work_list_item { * @update_list:link list node used to add to update_list * @mapping: bo_va mapping structure to create and update GPU page table * @npages: number of pages + * @vram_pages: vram pages number in this svm_range * @dma_addr: dma mapping address on each GPU for system memory physical page * @ttm_res: vram ttm resource map * @offset: range start offset within mm_nodes @@ -88,7 +89,9 @@ struct svm_work_list_item { * @flags: flags defined as KFD_IOCTL_SVM_FLAG_* * @perferred_loc: perferred location, 0 for CPU, or GPU id * @perfetch_loc: last prefetch location, 0 for CPU, or GPU id - * @actual_loc: the actual location, 0 for CPU, or GPU id + * @actual_loc: this svm_range location. 0: all pages are from sys ram; + * GPU id: this svm_range may include vram pages from GPU with + * id actual_loc. * @granularity:migration granularity, log2 num pages * @invalid: not 0 means cpu page table is invalidated * @validate_timestamp: system timestamp when range is validated @@ -112,6 +115,7 @@ struct svm_range { struct list_head list; struct list_head update_list; uint64_t npages; + uint64_t vram_pages; dma_addr_t *dma_addr[MAX_GPU_INSTANCE]; struct ttm_resource *ttm_res; uint64_t offset; @@ -168,9 +172,6 @@ struct kfd_node *svm_range_get_node_by_id(struct svm_range *prange, int svm_range_vram_node_new(struct kfd_node *node, struct svm_range *prange, bool clear); void svm_range_vram_node_free(struct svm_range *prange); -int svm_range_split_by_granularity(struct kfd_process *p, struct mm_struct *mm, - unsigned long addr, struct svm_range *parent, - struct svm_range *prange); int svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid, uint32_t vmid, uint32_t node_id, uint64_t addr, bool write_fault); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index 057284bf50bb..6ed2ec381aaa 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -1342,10 +1342,11 @@ static int kfd_create_indirect_link_prop(struct kfd_topology_device *kdev, int g num_cpu++; } + if (list_empty(&kdev->io_link_props)) + return -ENODATA; + gpu_link = list_first_entry(&kdev->io_link_props, - struct kfd_iolink_properties, list); - if (!gpu_link) - return -ENOMEM; + struct kfd_iolink_properties, list); for (i = 0; i < num_cpu; i++) { /* CPU <--> GPU */ @@ -1423,15 +1424,17 @@ static int kfd_add_peer_prop(struct kfd_topology_device *kdev, peer->gpu->adev)) return ret; + if (list_empty(&kdev->io_link_props)) + return -ENODATA; + iolink1 = list_first_entry(&kdev->io_link_props, - struct kfd_iolink_properties, list); - if (!iolink1) - return -ENOMEM; + struct kfd_iolink_properties, list); + + if (list_empty(&peer->io_link_props)) + return -ENODATA; iolink2 = list_first_entry(&peer->io_link_props, - struct kfd_iolink_properties, list); - if (!iolink2) - return -ENOMEM; + struct kfd_iolink_properties, list); props = kfd_alloc_struct(props); if (!props) @@ -1449,17 +1452,19 @@ static int kfd_add_peer_prop(struct kfd_topology_device *kdev, /* CPU->CPU link*/ cpu_dev = kfd_topology_device_by_proximity_domain(iolink1->node_to); if (cpu_dev) { - list_for_each_entry(iolink3, &cpu_dev->io_link_props, list) - if (iolink3->node_to == iolink2->node_to) - break; - - props->weight += iolink3->weight; - props->min_latency += iolink3->min_latency; - props->max_latency += iolink3->max_latency; - props->min_bandwidth = min(props->min_bandwidth, - iolink3->min_bandwidth); - props->max_bandwidth = min(props->max_bandwidth, - iolink3->max_bandwidth); + list_for_each_entry(iolink3, &cpu_dev->io_link_props, list) { + if (iolink3->node_to != iolink2->node_to) + continue; + + props->weight += iolink3->weight; + props->min_latency += iolink3->min_latency; + props->max_latency += iolink3->max_latency; + props->min_bandwidth = min(props->min_bandwidth, + iolink3->min_bandwidth); + props->max_bandwidth = min(props->max_bandwidth, + iolink3->max_bandwidth); + break; + } } else { WARN(1, "CPU node not found"); } @@ -1633,12 +1638,10 @@ static int fill_in_l2_l3_pcache(struct kfd_cache_properties **props_ext, else mode = UNKNOWN_MEMORY_PARTITION_MODE; - if (pcache->cache_level == 2) - pcache->cache_size = pcache_info[cache_type].cache_size * num_xcc; - else if (mode) - pcache->cache_size = pcache_info[cache_type].cache_size / mode; - else - pcache->cache_size = pcache_info[cache_type].cache_size; + pcache->cache_size = pcache_info[cache_type].cache_size; + /* Partition mode only affects L3 cache size */ + if (mode && pcache->cache_level == 3) + pcache->cache_size /= mode; if (pcache_info[cache_type].flags & CRAT_CACHE_FLAGS_DATA_CACHE) pcache->cache_type |= HSA_CACHE_TYPE_DATA; |