Missing synchronization with VGPR restore leads to intermittent
VGPR trashing in the user shader.
Signed-off-by: Jay Cornwall
---
drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h | 139 +++--
.../gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm | 1 +
2 files changed, 71 insert
ttmp[4:5] hold information useful to the debugger. Use ttmp[14:15]
instead, aligning implementation with gfx9 trap handler.
Signed-off-by: Jay Cornwall
---
drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h | 6 +++---
drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm | 10 +-
VCC moved out of user SGPR allocation in gfx10. It's now stored
in SGPRs 106-107.
Also fixes incorrect SGPR read offsets.
Cc: Shaoyun Liu
Signed-off-by: Jay Cornwall
---
drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h | 452 ++---
.../gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx
Copy/paste error, first 4 VGPRs are separated by 64 dwords (256 bytes).
Cc: Shaoyun Liu
Signed-off-by: Jay Cornwall
---
drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h | 6 +++---
drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm | 6 +++---
2 files changed, 6 insertions(+), 6 delet
These moved from SGPRs in gfx9 to HWREG in gfx10.
Cc: Shaoyun Liu
Signed-off-by: Jay Cornwall
---
drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h | 56 +-
.../gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm | 14 ++
2 files changed, 48 insertions(+), 22 deletions(-)
d
Signed-off-by: Jay Cornwall
---
.../gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm | 395 +
.../gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm | 375 +--
2 files changed, 5 insertions(+), 765 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handl
When a wavefront raises TRAPSTS.XNACK_ERROR with STATUS.ALLOW_REPLAY=0
subsequent memory instructions have undefined behavior. In practice
SQC stores continue to work but TCP stores do not.
Context save is permitted to fail after XNACK error because the
wavefront will be halted and subsequently te
If the trap is entered due to MODE.DEBUG_EN=1 and SAVECTX is raised
concurrently the handler cannot identify the source of the exception.
This causes the debugger to lose single step exception notification
when a context save request arrives at the same time.
When MODE.DEBUG_EN=1 and STATUS.HALT=0
