* [PATCH] drm/amdkfd: Fix named barrier restore in gfx12.1 trap handler
@ 2026-07-06 22:00 Jay Cornwall
2026-07-07 10:42 ` Indic, Vladimir
0 siblings, 1 reply; 2+ messages in thread
From: Jay Cornwall @ 2026-07-06 22:00 UTC (permalink / raw)
To: amd-gfx
Cc: Jay Cornwall, Lancelot Six, Vladimir Indic, Tishko Araz,
Max Erenberg
Barrier state layout has non-zero fields above the signal count.
Mask these off to restore the correct signal count.
Signed-off-by: Jay Cornwall <jay.cornwall@amd.com>
Cc: Lancelot Six <lancelot.six@amd.com>
Cc: Vladimir Indic <vladimir.indic@amd.com>
Cc: Tishko Araz <tishko.araz@amd.com>
Cc: Max Erenberg <max.erenberg@amd.com>
---
.../gpu/drm/amd/amdkfd/cwsr_trap_handler.h | 150 +++++++++---------
.../amd/amdkfd/cwsr_trap_handler_gfx12.asm | 2 +-
2 files changed, 77 insertions(+), 75 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
index 54fa76f374c9..42e4a5773ad6 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
@@ -3929,7 +3929,7 @@ static const uint32_t cwsr_trap_gfx12_hex[] = {
0x00030000, 0x807d847d,
0x8070ff70, 0x00000400,
0xbf0a7b7d, 0xbfa2ffe9,
- 0xbfa0014c, 0xbef4007e,
+ 0xbfa0014d, 0xbef4007e,
0x8b75ff7f, 0x0000ffff,
0xbef1007f, 0xb8f20742,
0x84729972, 0x8b6eff7f,
@@ -4070,36 +4070,36 @@ static const uint32_t cwsr_trap_gfx12_hex[] = {
0xb96ef812, 0xf4601bbb,
0xf8000030, 0xbf8a0000,
0xb96ef813, 0x8b6eff7f,
- 0x04000000, 0xbfa1000b,
+ 0x04000000, 0xbfa1000c,
0xf4601bbb, 0xf8000038,
0xbf8a0000, 0xbf0d806e,
- 0xbfa10006, 0x856e906e,
- 0x8b6e6e6e, 0xbfa10003,
- 0xbe804ec1, 0x816ec16e,
- 0xbfa0fffb, 0xbefd006f,
- 0xbefe0070, 0xbeff0071,
- 0xb97b2011, 0x857b867b,
- 0xb97b0191, 0x857b827b,
- 0xb97bba11, 0xb973f801,
- 0xb8ee3b05, 0x806e816e,
- 0xbf0d9972, 0xbfa20002,
- 0x846e896e, 0xbfa00001,
- 0x846e8a6e, 0x806eff6e,
- 0x00000240, 0x806e746e,
- 0x826f8075, 0xf4605c37,
- 0xf8000010, 0xf4605d37,
- 0xf8000020, 0xf4601e77,
- 0xf8000034, 0xbf8a0000,
- 0x8b6dff6d, 0x0000ffff,
- 0x8bfe7e7e, 0x8bea6a6a,
- 0x936eff77, 0x0002001a,
- 0xb96ef81a, 0xb97af804,
+ 0xbfa10007, 0x936eff6e,
+ 0x00070010, 0x8b6e6e6e,
+ 0xbfa10003, 0xbe804ec1,
+ 0x816ec16e, 0xbfa0fffb,
+ 0xbefd006f, 0xbefe0070,
+ 0xbeff0071, 0xb97b2011,
+ 0x857b867b, 0xb97b0191,
+ 0x857b827b, 0xb97bba11,
+ 0xb973f801, 0xb8ee3b05,
+ 0x806e816e, 0xbf0d9972,
+ 0xbfa20002, 0x846e896e,
+ 0xbfa00001, 0x846e8a6e,
+ 0x806eff6e, 0x00000240,
+ 0x806e746e, 0x826f8075,
+ 0xf4605c37, 0xf8000010,
+ 0xf4605d37, 0xf8000020,
+ 0xf4601e77, 0xf8000034,
+ 0xbf8a0000, 0x8b6dff6d,
+ 0x0000ffff, 0x8bfe7e7e,
+ 0x8bea6a6a, 0x936eff77,
+ 0x0002001a, 0xb96ef81a,
+ 0xb97af804, 0xbe804ec2,
+ 0xbf94fffe, 0xbe804a6c,
0xbe804ec2, 0xbf94fffe,
- 0xbe804a6c, 0xbe804ec2,
- 0xbf94fffe, 0xbfb10000,
+ 0xbfb10000, 0xbf9f0000,
0xbf9f0000, 0xbf9f0000,
0xbf9f0000, 0xbf9f0000,
- 0xbf9f0000, 0x00000000,
};
static const uint32_t cwsr_trap_gfx9_5_0_hex[] = {
@@ -5066,7 +5066,7 @@ static const uint32_t cwsr_trap_gfx12_1_0_hex[] = {
0x018c0000, 0x00030000,
0x807d847d, 0x8070ff70,
0x00000400, 0xbf0a7b7d,
- 0xbfa2ffe9, 0xbfa00184,
+ 0xbfa2ffe9, 0xbfa00187,
0xbef4007e, 0x8b75ff7f,
0x01ffffff, 0xbef1007f,
0xb8f20742, 0x84729972,
@@ -5208,64 +5208,66 @@ static const uint32_t cwsr_trap_gfx12_1_0_hex[] = {
0xf4601bbb, 0xf8000030,
0xbf8a0000, 0xb96ef813,
0x8b6eff7f, 0x04000000,
- 0xbfa10022, 0xf4601bbb,
+ 0xbfa10024, 0xf4601bbb,
0xf8000038, 0xbf8a0000,
- 0xbf0d806e, 0xbfa1001d,
- 0x856e906e, 0x8b6e6e6e,
- 0xbfa10003, 0xbe804ec1,
- 0x816ec16e, 0xbfa0fffb,
- 0xbef800ff, 0x00000080,
- 0xbefd0081, 0xf4601bbb,
- 0xf0000000, 0xbfc70000,
- 0x80788478, 0x937eff6e,
- 0x00070004, 0x847e907e,
- 0x8c7d7e7d, 0xbe80517d,
- 0x917dff7d, 0x007f0000,
- 0x856e906e, 0x8b6e6e6e,
+ 0xbf0d806e, 0xbfa1001f,
+ 0x936eff6e, 0x00070010,
+ 0x8b6e6e6e, 0xbfa10003,
+ 0xbe804ec1, 0x816ec16e,
+ 0xbfa0fffb, 0xbef800ff,
+ 0x00000080, 0xbefd0081,
+ 0xf4601bbb, 0xf0000000,
+ 0xbfc70000, 0x80788478,
+ 0x937eff6e, 0x00070004,
+ 0x847e907e, 0x8c7d7e7d,
+ 0xbe80517d, 0x917dff7d,
+ 0x007f0000, 0x936eff6e,
+ 0x00070010, 0x8b6e6e6e,
0xbfa10003, 0xbe804e7d,
0x816ec16e, 0xbfa0fffb,
0x807d817d, 0xbf08907d,
- 0xbfa1ffec, 0xf4601bbb,
+ 0xbfa1ffeb, 0xf4601bbb,
0xf800003c, 0xbfc70000,
- 0xbf0d806e, 0xbfa1000c,
+ 0xbf0d806e, 0xbfa1000d,
0xbf0d9a7f, 0xbfa10002,
0xbf068180, 0xbe804fc4,
- 0xbf94fffc, 0xbfa10006,
- 0x856e906e, 0x8b6e6e6e,
- 0xbfa10003, 0xbe804ec3,
- 0x816ec16e, 0xbfa0fffb,
- 0xf4601bbb, 0xf8000040,
- 0xbfc70000, 0xb96ef81a,
- 0xbefd006f, 0xbefe0070,
- 0xbeff0071, 0xb979f822,
- 0xb97b2011, 0x857b867b,
- 0xb97b0191, 0x857b827b,
- 0xb97bba11, 0xb973f801,
- 0xb8ee3b05, 0x806e816e,
- 0xbf0d9972, 0xbfa20002,
- 0x846e896e, 0xbfa00001,
- 0x846e8a6e, 0x806eff6e,
- 0x000001c0, 0x806e746e,
- 0x826f8075, 0xf4605c37,
- 0xf8000010, 0xf4605d37,
- 0xf8000020, 0xf4601e77,
- 0xf8000034, 0xbf8a0000,
- 0x856e9677, 0xb96e04a1,
- 0x856e9577, 0xb96e0421,
- 0x856e8e77, 0xb96e3021,
- 0x8b6dff6d, 0x01ffffff,
- 0x8bfe7e7e, 0x8bea6a6a,
- 0xb97af804, 0xb8eef802,
- 0xbf0c8b6e, 0xbfa20003,
- 0xbe804fc2, 0xbf94fffe,
- 0xbfa10001, 0xbe804ec4,
- 0xbf94fffc, 0x857a897a,
- 0xb97a0244, 0xbe804a6c,
+ 0xbf94fffc, 0xbfa10007,
+ 0x936eff6e, 0x00070010,
+ 0x8b6e6e6e, 0xbfa10003,
+ 0xbe804ec3, 0x816ec16e,
+ 0xbfa0fffb, 0xf4601bbb,
+ 0xf8000040, 0xbfc70000,
+ 0xb96ef81a, 0xbefd006f,
+ 0xbefe0070, 0xbeff0071,
+ 0xb979f822, 0xb97b2011,
+ 0x857b867b, 0xb97b0191,
+ 0x857b827b, 0xb97bba11,
+ 0xb973f801, 0xb8ee3b05,
+ 0x806e816e, 0xbf0d9972,
+ 0xbfa20002, 0x846e896e,
+ 0xbfa00001, 0x846e8a6e,
+ 0x806eff6e, 0x000001c0,
+ 0x806e746e, 0x826f8075,
+ 0xf4605c37, 0xf8000010,
+ 0xf4605d37, 0xf8000020,
+ 0xf4601e77, 0xf8000034,
+ 0xbf8a0000, 0x856e9677,
+ 0xb96e04a1, 0x856e9577,
+ 0xb96e0421, 0x856e8e77,
+ 0xb96e3021, 0x8b6dff6d,
+ 0x01ffffff, 0x8bfe7e7e,
+ 0x8bea6a6a, 0xb97af804,
0xb8eef802, 0xbf0c8b6e,
0xbfa20003, 0xbe804fc2,
0xbf94fffe, 0xbfa10001,
0xbe804ec4, 0xbf94fffc,
- 0xbfb10000, 0xbf9f0000,
+ 0x857a897a, 0xb97a0244,
+ 0xbe804a6c, 0xb8eef802,
+ 0xbf0c8b6e, 0xbfa20003,
+ 0xbe804fc2, 0xbf94fffe,
+ 0xbfa10001, 0xbe804ec4,
+ 0xbf94fffc, 0xbfb10000,
0xbf9f0000, 0xbf9f0000,
0xbf9f0000, 0xbf9f0000,
+ 0xbf9f0000, 0x00000000,
};
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 456db8199899..0da98cbc4c30 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm
@@ -1387,7 +1387,7 @@ end
function restore_barrier_signal_count(barrier_id)
// extract the saved signal count from s_restore_tmp
- s_lshr_b32 s_restore_tmp, s_restore_tmp, BARRIER_STATE_SIGNAL_OFFSET
+ s_bfe_u32 s_restore_tmp, s_restore_tmp, (BARRIER_STATE_SIGNAL_OFFSET | (BARRIER_STATE_SIGNAL_SIZE << 0x10))
// We need to call s_barrier_signal repeatedly to restore the signal count
// of the group/cluster barrier. The member count is already initialized.
--
2.34.1
^ permalink raw reply related [flat|nested] 2+ messages in thread
* Re: [PATCH] drm/amdkfd: Fix named barrier restore in gfx12.1 trap handler
2026-07-06 22:00 [PATCH] drm/amdkfd: Fix named barrier restore in gfx12.1 trap handler Jay Cornwall
@ 2026-07-07 10:42 ` Indic, Vladimir
0 siblings, 0 replies; 2+ messages in thread
From: Indic, Vladimir @ 2026-07-07 10:42 UTC (permalink / raw)
To: Cornwall, Jay, amd-gfx@lists.freedesktop.org
Cc: Six, Lancelot, Araz, Tishko, Erenberg, Max
[-- Attachment #1: Type: text/plain, Size: 9643 bytes --]
AMD General
Hi Jay,
Thanks for the patch!
Reviewed-by: Vladimir Indic <vladimir.indic@amd.com>
The bug and fix are correct. S_GET_BARRIER_STATE returns:
{ 5'b0, NBC[2:0], 1'b0, signalCnt[6:0], 5'b0, memberCnt[6:0], 3'b0, valid }
where NBC (bits [26:24]) sits immediately above signalCnt (bits [22:16]).
An s_lshr_b32 by 22 shifts both down together, so the extracted value is
(NBC << 2 | signalCnt) — inflated whenever named barriers are allocated
(NBC > 0), causing the restore loop to over-signal the barrier.
s_bfe_u32 with offset=22, size=8 masks NBC out precisely, leaving only the
7-bit signal count in the destination.
One note: BARRIER_STATE_SIGNAL_SIZE is defined as 8, but the signalCnt
field is 7 bits wide ([22:16]). Bit 29 is always 0 in practice (it's
reserved above NBC), so this is harmless and the extraction is still
correct. A comment (or a corrected constant value 7) would make the intent
clearer, but it's not a blocker.
Best regards,
Vladimir
________________________________
From: Cornwall, Jay <Jay.Cornwall@amd.com>
Sent: Tuesday, July 7, 2026 12:00 AM
To: amd-gfx@lists.freedesktop.org <amd-gfx@lists.freedesktop.org>
Cc: Cornwall, Jay <Jay.Cornwall@amd.com>; Six, Lancelot <Lancelot.Six@amd.com>; Indic, Vladimir <Vladimir.Indic@amd.com>; Araz, Tishko <Tishko.Araz@amd.com>; Erenberg, Max <Max.Erenberg@amd.com>
Subject: [PATCH] drm/amdkfd: Fix named barrier restore in gfx12.1 trap handler
Barrier state layout has non-zero fields above the signal count.
Mask these off to restore the correct signal count.
Signed-off-by: Jay Cornwall <jay.cornwall@amd.com>
Cc: Lancelot Six <lancelot.six@amd.com>
Cc: Vladimir Indic <vladimir.indic@amd.com>
Cc: Tishko Araz <tishko.araz@amd.com>
Cc: Max Erenberg <max.erenberg@amd.com>
---
.../gpu/drm/amd/amdkfd/cwsr_trap_handler.h | 150 +++++++++---------
.../amd/amdkfd/cwsr_trap_handler_gfx12.asm | 2 +-
2 files changed, 77 insertions(+), 75 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
index 54fa76f374c9..42e4a5773ad6 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
@@ -3929,7 +3929,7 @@ static const uint32_t cwsr_trap_gfx12_hex[] = {
0x00030000, 0x807d847d,
0x8070ff70, 0x00000400,
0xbf0a7b7d, 0xbfa2ffe9,
- 0xbfa0014c, 0xbef4007e,
+ 0xbfa0014d, 0xbef4007e,
0x8b75ff7f, 0x0000ffff,
0xbef1007f, 0xb8f20742,
0x84729972, 0x8b6eff7f,
@@ -4070,36 +4070,36 @@ static const uint32_t cwsr_trap_gfx12_hex[] = {
0xb96ef812, 0xf4601bbb,
0xf8000030, 0xbf8a0000,
0xb96ef813, 0x8b6eff7f,
- 0x04000000, 0xbfa1000b,
+ 0x04000000, 0xbfa1000c,
0xf4601bbb, 0xf8000038,
0xbf8a0000, 0xbf0d806e,
- 0xbfa10006, 0x856e906e,
- 0x8b6e6e6e, 0xbfa10003,
- 0xbe804ec1, 0x816ec16e,
- 0xbfa0fffb, 0xbefd006f,
- 0xbefe0070, 0xbeff0071,
- 0xb97b2011, 0x857b867b,
- 0xb97b0191, 0x857b827b,
- 0xb97bba11, 0xb973f801,
- 0xb8ee3b05, 0x806e816e,
- 0xbf0d9972, 0xbfa20002,
- 0x846e896e, 0xbfa00001,
- 0x846e8a6e, 0x806eff6e,
- 0x00000240, 0x806e746e,
- 0x826f8075, 0xf4605c37,
- 0xf8000010, 0xf4605d37,
- 0xf8000020, 0xf4601e77,
- 0xf8000034, 0xbf8a0000,
- 0x8b6dff6d, 0x0000ffff,
- 0x8bfe7e7e, 0x8bea6a6a,
- 0x936eff77, 0x0002001a,
- 0xb96ef81a, 0xb97af804,
+ 0xbfa10007, 0x936eff6e,
+ 0x00070010, 0x8b6e6e6e,
+ 0xbfa10003, 0xbe804ec1,
+ 0x816ec16e, 0xbfa0fffb,
+ 0xbefd006f, 0xbefe0070,
+ 0xbeff0071, 0xb97b2011,
+ 0x857b867b, 0xb97b0191,
+ 0x857b827b, 0xb97bba11,
+ 0xb973f801, 0xb8ee3b05,
+ 0x806e816e, 0xbf0d9972,
+ 0xbfa20002, 0x846e896e,
+ 0xbfa00001, 0x846e8a6e,
+ 0x806eff6e, 0x00000240,
+ 0x806e746e, 0x826f8075,
+ 0xf4605c37, 0xf8000010,
+ 0xf4605d37, 0xf8000020,
+ 0xf4601e77, 0xf8000034,
+ 0xbf8a0000, 0x8b6dff6d,
+ 0x0000ffff, 0x8bfe7e7e,
+ 0x8bea6a6a, 0x936eff77,
+ 0x0002001a, 0xb96ef81a,
+ 0xb97af804, 0xbe804ec2,
+ 0xbf94fffe, 0xbe804a6c,
0xbe804ec2, 0xbf94fffe,
- 0xbe804a6c, 0xbe804ec2,
- 0xbf94fffe, 0xbfb10000,
+ 0xbfb10000, 0xbf9f0000,
0xbf9f0000, 0xbf9f0000,
0xbf9f0000, 0xbf9f0000,
- 0xbf9f0000, 0x00000000,
};
static const uint32_t cwsr_trap_gfx9_5_0_hex[] = {
@@ -5066,7 +5066,7 @@ static const uint32_t cwsr_trap_gfx12_1_0_hex[] = {
0x018c0000, 0x00030000,
0x807d847d, 0x8070ff70,
0x00000400, 0xbf0a7b7d,
- 0xbfa2ffe9, 0xbfa00184,
+ 0xbfa2ffe9, 0xbfa00187,
0xbef4007e, 0x8b75ff7f,
0x01ffffff, 0xbef1007f,
0xb8f20742, 0x84729972,
@@ -5208,64 +5208,66 @@ static const uint32_t cwsr_trap_gfx12_1_0_hex[] = {
0xf4601bbb, 0xf8000030,
0xbf8a0000, 0xb96ef813,
0x8b6eff7f, 0x04000000,
- 0xbfa10022, 0xf4601bbb,
+ 0xbfa10024, 0xf4601bbb,
0xf8000038, 0xbf8a0000,
- 0xbf0d806e, 0xbfa1001d,
- 0x856e906e, 0x8b6e6e6e,
- 0xbfa10003, 0xbe804ec1,
- 0x816ec16e, 0xbfa0fffb,
- 0xbef800ff, 0x00000080,
- 0xbefd0081, 0xf4601bbb,
- 0xf0000000, 0xbfc70000,
- 0x80788478, 0x937eff6e,
- 0x00070004, 0x847e907e,
- 0x8c7d7e7d, 0xbe80517d,
- 0x917dff7d, 0x007f0000,
- 0x856e906e, 0x8b6e6e6e,
+ 0xbf0d806e, 0xbfa1001f,
+ 0x936eff6e, 0x00070010,
+ 0x8b6e6e6e, 0xbfa10003,
+ 0xbe804ec1, 0x816ec16e,
+ 0xbfa0fffb, 0xbef800ff,
+ 0x00000080, 0xbefd0081,
+ 0xf4601bbb, 0xf0000000,
+ 0xbfc70000, 0x80788478,
+ 0x937eff6e, 0x00070004,
+ 0x847e907e, 0x8c7d7e7d,
+ 0xbe80517d, 0x917dff7d,
+ 0x007f0000, 0x936eff6e,
+ 0x00070010, 0x8b6e6e6e,
0xbfa10003, 0xbe804e7d,
0x816ec16e, 0xbfa0fffb,
0x807d817d, 0xbf08907d,
- 0xbfa1ffec, 0xf4601bbb,
+ 0xbfa1ffeb, 0xf4601bbb,
0xf800003c, 0xbfc70000,
- 0xbf0d806e, 0xbfa1000c,
+ 0xbf0d806e, 0xbfa1000d,
0xbf0d9a7f, 0xbfa10002,
0xbf068180, 0xbe804fc4,
- 0xbf94fffc, 0xbfa10006,
- 0x856e906e, 0x8b6e6e6e,
- 0xbfa10003, 0xbe804ec3,
- 0x816ec16e, 0xbfa0fffb,
- 0xf4601bbb, 0xf8000040,
- 0xbfc70000, 0xb96ef81a,
- 0xbefd006f, 0xbefe0070,
- 0xbeff0071, 0xb979f822,
- 0xb97b2011, 0x857b867b,
- 0xb97b0191, 0x857b827b,
- 0xb97bba11, 0xb973f801,
- 0xb8ee3b05, 0x806e816e,
- 0xbf0d9972, 0xbfa20002,
- 0x846e896e, 0xbfa00001,
- 0x846e8a6e, 0x806eff6e,
- 0x000001c0, 0x806e746e,
- 0x826f8075, 0xf4605c37,
- 0xf8000010, 0xf4605d37,
- 0xf8000020, 0xf4601e77,
- 0xf8000034, 0xbf8a0000,
- 0x856e9677, 0xb96e04a1,
- 0x856e9577, 0xb96e0421,
- 0x856e8e77, 0xb96e3021,
- 0x8b6dff6d, 0x01ffffff,
- 0x8bfe7e7e, 0x8bea6a6a,
- 0xb97af804, 0xb8eef802,
- 0xbf0c8b6e, 0xbfa20003,
- 0xbe804fc2, 0xbf94fffe,
- 0xbfa10001, 0xbe804ec4,
- 0xbf94fffc, 0x857a897a,
- 0xb97a0244, 0xbe804a6c,
+ 0xbf94fffc, 0xbfa10007,
+ 0x936eff6e, 0x00070010,
+ 0x8b6e6e6e, 0xbfa10003,
+ 0xbe804ec3, 0x816ec16e,
+ 0xbfa0fffb, 0xf4601bbb,
+ 0xf8000040, 0xbfc70000,
+ 0xb96ef81a, 0xbefd006f,
+ 0xbefe0070, 0xbeff0071,
+ 0xb979f822, 0xb97b2011,
+ 0x857b867b, 0xb97b0191,
+ 0x857b827b, 0xb97bba11,
+ 0xb973f801, 0xb8ee3b05,
+ 0x806e816e, 0xbf0d9972,
+ 0xbfa20002, 0x846e896e,
+ 0xbfa00001, 0x846e8a6e,
+ 0x806eff6e, 0x000001c0,
+ 0x806e746e, 0x826f8075,
+ 0xf4605c37, 0xf8000010,
+ 0xf4605d37, 0xf8000020,
+ 0xf4601e77, 0xf8000034,
+ 0xbf8a0000, 0x856e9677,
+ 0xb96e04a1, 0x856e9577,
+ 0xb96e0421, 0x856e8e77,
+ 0xb96e3021, 0x8b6dff6d,
+ 0x01ffffff, 0x8bfe7e7e,
+ 0x8bea6a6a, 0xb97af804,
0xb8eef802, 0xbf0c8b6e,
0xbfa20003, 0xbe804fc2,
0xbf94fffe, 0xbfa10001,
0xbe804ec4, 0xbf94fffc,
- 0xbfb10000, 0xbf9f0000,
+ 0x857a897a, 0xb97a0244,
+ 0xbe804a6c, 0xb8eef802,
+ 0xbf0c8b6e, 0xbfa20003,
+ 0xbe804fc2, 0xbf94fffe,
+ 0xbfa10001, 0xbe804ec4,
+ 0xbf94fffc, 0xbfb10000,
0xbf9f0000, 0xbf9f0000,
0xbf9f0000, 0xbf9f0000,
+ 0xbf9f0000, 0x00000000,
};
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 456db8199899..0da98cbc4c30 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm
@@ -1387,7 +1387,7 @@ end
function restore_barrier_signal_count(barrier_id)
// extract the saved signal count from s_restore_tmp
- s_lshr_b32 s_restore_tmp, s_restore_tmp, BARRIER_STATE_SIGNAL_OFFSET
+ s_bfe_u32 s_restore_tmp, s_restore_tmp, (BARRIER_STATE_SIGNAL_OFFSET | (BARRIER_STATE_SIGNAL_SIZE << 0x10))
// We need to call s_barrier_signal repeatedly to restore the signal count
// of the group/cluster barrier. The member count is already initialized.
--
2.34.1
[-- Attachment #2: Type: text/html, Size: 22282 bytes --]
^ permalink raw reply related [flat|nested] 2+ messages in thread
end of thread, other threads:[~2026-07-07 10:42 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-07-06 22:00 [PATCH] drm/amdkfd: Fix named barrier restore in gfx12.1 trap handler Jay Cornwall
2026-07-07 10:42 ` Indic, Vladimir
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.