summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorJay Cornwall <jay.cornwall@amd.com>2025-10-23 23:28:39 +0300
committerAlex Deucher <alexander.deucher@amd.com>2026-01-06 00:59:56 +0300
commit1005ab86cf75aec3462c1c58d6efa87871f2843f (patch)
treed69a08663a026b35ebb121e5a70958e0cbe8e355
parent864a8b2c1ff9e5767d6bb19854a671c30b3b268f (diff)
downloadlinux-1005ab86cf75aec3462c1c58d6efa87871f2843f.tar.xz
drm/amdkfd: Fix VGPR bank state save in gfx12.1 trap handler
S_SETREG_IMM32_B32 does not apply a mask to the MODE bank bits. SRC2 is consequently unconditonally cleared during context save. Use S_SETREG_B32 instead to preserve SRC2. Signed-off-by: Jay Cornwall <jay.cornwall@amd.com> Reviewed-by: Lancelot Six <lancelot.six@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
-rw-r--r--drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h2
-rw-r--r--drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm3
2 files changed, 3 insertions, 2 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
index 5c6d533a5956..d82ce2f1e9b9 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
@@ -4731,7 +4731,7 @@ static const uint32_t cwsr_trap_gfx12_1_0_hex[] = {
0xb8eff822, 0xb980f822,
0x00000000, 0xb8fa2b01,
0x847a997a, 0x8c6d7a6d,
- 0xb9802b01, 0x00000000,
+ 0xbefa0080, 0xb97a2b01,
0xbefa007e, 0x8b7bff7f,
0x01ffffff, 0xbefe00c1,
0xbeff00c1, 0xee0a407a,
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm
index a807e7557e93..d59400d242d1 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm
@@ -414,7 +414,8 @@ L_HAVE_VGPRS:
s_getreg_b32 s_save_tmp, hwreg(HW_REG_WAVE_MODE, SQ_WAVE_MODE_DST_SRC0_SRC1_VGPR_MSB_SHIFT, SQ_WAVE_MODE_DST_SRC0_SRC1_VGPR_MSB_SIZE)
s_lshl_b32 s_save_tmp, s_save_tmp, S_SAVE_PC_HI_DST_SRC0_SRC1_VGPR_MSB_SHIFT
s_or_b32 s_save_pc_hi, s_save_pc_hi, s_save_tmp
- s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, SQ_WAVE_MODE_DST_SRC0_SRC1_VGPR_MSB_SHIFT, SQ_WAVE_MODE_DST_SRC0_SRC1_VGPR_MSB_SIZE), 0
+ s_mov_b32 s_save_tmp, 0
+ s_setreg_b32 hwreg(HW_REG_WAVE_MODE, SQ_WAVE_MODE_DST_SRC0_SRC1_VGPR_MSB_SHIFT, SQ_WAVE_MODE_DST_SRC0_SRC1_VGPR_MSB_SIZE), s_save_tmp
#endif
// Trap temporaries must be saved via VGPR but all VGPRs are in use.