[PATCH 119/207] drm/amdkfd: Support debugger in Navi1x trap handler

Alex Deucher alexdeucher at gmail.com
Mon Jun 1 18:19:26 UTC 2020


From: Jay Cornwall <jay.cornwall at amd.com>

- Preserve scalar GPRs ttmp[4:11] and ttmp13
- Add single step exception during context save workaround
- Remove incorrect PC adjustment during context save

Signed-off-by: Jay Cornwall <jay.cornwall at amd.com>
Reviewed-by: Yong Zhao <Yong.Zhao at amd.com>
Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
---
 .../gpu/drm/amd/amdkfd/cwsr_trap_handler.h    | 1368 +++++++++--------
 .../amd/amdkfd/cwsr_trap_handler_gfx10.asm    |  141 +-
 2 files changed, 789 insertions(+), 720 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
index 7e9e2d606682..577d901fdb63 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
@@ -680,389 +680,404 @@ static const uint32_t cwsr_trap_gfx9_hex[] = {
 };
 
 static const uint32_t cwsr_trap_nv1x_hex[] = {
-	0xbf820001, 0xbf8201bc,
+	0xbf820001, 0xbf8201cd,
 	0xb0804004, 0xb978f802,
-	0x8a788678, 0xb971f803,
-	0x876eff71, 0x00000400,
-	0xbf850033, 0x876eff71,
-	0x00000100, 0xbf840002,
-	0x8878ff78, 0x00002000,
-	0x8a77ff77, 0xff000000,
-	0xb96ef807, 0x876fff6e,
-	0x02000000, 0x8f6f866f,
-	0x88776f77, 0x876fff6e,
-	0x003f8000, 0x8f6f896f,
-	0x88776f77, 0x8a6eff6e,
-	0x023f8000, 0xb9eef807,
-	0xb97af812, 0xb97bf813,
-	0x8ffa887a, 0xf4051bbd,
-	0xfa000000, 0xbf8cc07f,
-	0xf4051ebd, 0xfa000008,
-	0xbf8cc07f, 0x87ee6e6e,
-	0xbf840001, 0xbe80206e,
-	0xb971f803, 0x8771ff71,
-	0x000001ff, 0xbf850002,
-	0x806c846c, 0x826d806d,
-	0x876dff6d, 0x0000ffff,
-	0x906e8977, 0x876fff6e,
-	0x003f8000, 0x906e8677,
-	0x876eff6e, 0x02000000,
-	0x886e6f6e, 0xb9eef807,
-	0x87fe7e7e, 0x87ea6a6a,
-	0xb9f8f802, 0xbe80226c,
-	0xb971f803, 0x8771ff71,
-	0x00000100, 0xbf840006,
-	0xbef60380, 0xb9f60203,
-	0x876dff6d, 0x0000ffff,
-	0x80ec886c, 0x82ed806d,
-	0xbef60380, 0xb9f60283,
-	0xb972f816, 0xb9762c07,
-	0x8f769a76, 0x886d766d,
-	0xb97603c7, 0x8f769976,
-	0x886d766d, 0xb9760647,
-	0x8f769876, 0x886d766d,
-	0xb976f807, 0x8776ff76,
-	0x00007fff, 0xb9f6f807,
+	0x8a788678, 0xb96ef801,
+	0x876eff6e, 0x00000800,
+	0xbf840003, 0x876eff78,
+	0x00002000, 0xbf840009,
+	0xb97bf803, 0x876eff7b,
+	0x00000400, 0xbf850033,
+	0x876eff7b, 0x00000100,
+	0xbf840002, 0x8878ff78,
+	0x00002000, 0x8a77ff77,
+	0xff000000, 0xb96ef807,
+	0x876fff6e, 0x02000000,
+	0x8f6f866f, 0x88776f77,
+	0x876fff6e, 0x003f8000,
+	0x8f6f896f, 0x88776f77,
+	0x8a6eff6e, 0x023f8000,
+	0xb9eef807, 0xb97af812,
+	0xb97bf813, 0x8ffa887a,
+	0xf4051bbd, 0xfa000000,
+	0xbf8cc07f, 0xf4051ebd,
+	0xfa000008, 0xbf8cc07f,
+	0x87ee6e6e, 0xbf840001,
+	0xbe80206e, 0xb97bf803,
+	0x877bff7b, 0x000001ff,
+	0xbf850002, 0x806c846c,
+	0x826d806d, 0x876dff6d,
+	0x0000ffff, 0x906e8977,
+	0x876fff6e, 0x003f8000,
+	0x906e8677, 0x876eff6e,
+	0x02000000, 0x886e6f6e,
+	0xb9eef807, 0x87fe7e7e,
+	0x87ea6a6a, 0xb9f8f802,
+	0xbe80226c, 0x876dff6d,
+	0x0000ffff, 0xbefa0380,
+	0xb9fa0283, 0xb97a2c07,
+	0x8f7a9a7a, 0x886d7a6d,
+	0xb97a03c7, 0x8f7a997a,
+	0x886d7a6d, 0xb97a0647,
+	0x8f7a987a, 0x886d7a6d,
+	0xb97af807, 0x877aff7a,
+	0x00007fff, 0xb9faf807,
 	0xbeee037e, 0xbeef037f,
 	0xbefe0480, 0xbf900004,
 	0xbf8e0002, 0xbf88fffe,
+	0xb97b02dc, 0x8f7b997b,
+	0x887b7b7f, 0xb97a2a05,
+	0x807a817a, 0xbf0d997b,
+	0xbf850002, 0x8f7a897a,
+	0xbf820001, 0x8f7a8a7a,
+	0x877bff7f, 0x0000ffff,
+	0x807aff7a, 0x00000200,
+	0x807a7e7a, 0x827b807b,
+	0xf4491c3d, 0xfa000050,
+	0xf4491d3d, 0xfa000060,
+	0xf4411e7d, 0xfa000074,
 	0xbef4037e, 0x8775ff7f,
 	0x0000ffff, 0x8875ff75,
 	0x00040000, 0xbef60380,
 	0xbef703ff, 0x10807fac,
-	0x8776ff7f, 0x08000000,
-	0x90768376, 0x88777677,
-	0x8776ff7f, 0x70000000,
-	0x90768176, 0x88777677,
-	0xbefb037c, 0xbefa0380,
+	0x877aff7f, 0x08000000,
+	0x907a837a, 0x88777a77,
+	0x877aff7f, 0x70000000,
+	0x907a817a, 0x88777a77,
+	0xbef1037c, 0xbef00380,
 	0xb97302dc, 0x8f739973,
-	0x8873737f, 0xbefe03c1,
-	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbf850002,
-	0xbeff0380, 0xbf820002,
-	0xbeff03c1, 0xbf82000b,
-	0xbef603ff, 0x01000000,
-	0xe0704000, 0x7a5d0000,
-	0xe0704080, 0x7a5d0100,
-	0xe0704100, 0x7a5d0200,
-	0xe0704180, 0x7a5d0300,
-	0xbf82000a, 0xbef603ff,
-	0x01000000, 0xe0704000,
-	0x7a5d0000, 0xe0704100,
-	0x7a5d0100, 0xe0704200,
-	0x7a5d0200, 0xe0704300,
-	0x7a5d0300, 0xb97a2a05,
-	0x807a817a, 0x907c9973,
+	0x8873737f, 0xb97bf816,
+	0xba80f816, 0x00000000,
+	0xbefe03c1, 0x907c9973,
 	0x877c817c, 0xbf06817c,
-	0xbf850002, 0x8f7a897a,
-	0xbf820001, 0x8f7a8a7a,
-	0xb9761e06, 0x8f768a76,
-	0x807a767a, 0x807aff7a,
-	0x00000200, 0xbef603ff,
-	0x01000000, 0xbefe037c,
-	0xbefc037a, 0xf4611efa,
-	0xf8000000, 0x807a847a,
-	0xbefc037e, 0xbefe037c,
-	0xbefc037a, 0xf4611b3a,
-	0xf8000000, 0x807a847a,
-	0xbefc037e, 0xbefe037c,
-	0xbefc037a, 0xf4611b7a,
-	0xf8000000, 0x807a847a,
-	0xbefc037e, 0xbefe037c,
-	0xbefc037a, 0xf4611bba,
-	0xf8000000, 0x807a847a,
-	0xbefc037e, 0xbefe037c,
-	0xbefc037a, 0xf4611bfa,
-	0xf8000000, 0x807a847a,
+	0xbf850002, 0xbeff0380,
+	0xbf820002, 0xbeff03c1,
+	0xbf82000b, 0xbef603ff,
+	0x01000000, 0xe0704000,
+	0x705d0000, 0xe0704080,
+	0x705d0100, 0xe0704100,
+	0x705d0200, 0xe0704180,
+	0x705d0300, 0xbf82000a,
+	0xbef603ff, 0x01000000,
+	0xe0704000, 0x705d0000,
+	0xe0704100, 0x705d0100,
+	0xe0704200, 0x705d0200,
+	0xe0704300, 0x705d0300,
+	0xb9702a05, 0x80708170,
+	0xbf0d9973, 0xbf850002,
+	0x8f708970, 0xbf820001,
+	0x8f708a70, 0xb97a1e06,
+	0x8f7a8a7a, 0x80707a70,
+	0x8070ff70, 0x00000200,
+	0xbef603ff, 0x01000000,
+	0xbefe037c, 0xbefc0370,
+	0xf4611c7a, 0xf8000000,
+	0x80708470, 0xbefc037e,
+	0xbefe037c, 0xbefc0370,
+	0xf4611b3a, 0xf8000000,
+	0x80708470, 0xbefc037e,
+	0xbefe037c, 0xbefc0370,
+	0xf4611b7a, 0xf8000000,
+	0x80708470, 0xbefc037e,
+	0xbefe037c, 0xbefc0370,
+	0xf4611bba, 0xf8000000,
+	0x80708470, 0xbefc037e,
+	0xbefe037c, 0xbefc0370,
+	0xf4611bfa, 0xf8000000,
+	0x80708470, 0xbefc037e,
+	0xbefe037c, 0xbefc0370,
+	0xf4611e3a, 0xf8000000,
+	0x80708470, 0xbefc037e,
+	0xb97af803, 0xbefe037c,
+	0xbefc0370, 0xf4611eba,
+	0xf8000000, 0x80708470,
 	0xbefc037e, 0xbefe037c,
-	0xbefc037a, 0xf4611e3a,
-	0xf8000000, 0x807a847a,
-	0xbefc037e, 0xb971f803,
-	0xbefe037c, 0xbefc037a,
+	0xbefc0370, 0xf4611efa,
+	0xf8000000, 0x80708470,
+	0xbefc037e, 0xb971f801,
+	0xbefe037c, 0xbefc0370,
 	0xf4611c7a, 0xf8000000,
-	0x807a847a, 0xbefc037e,
-	0xbefe037c, 0xbefc037a,
-	0xf4611cba, 0xf8000000,
-	0x807a847a, 0xbefc037e,
-	0xb97bf801, 0xbefe037c,
-	0xbefc037a, 0xf4611efa,
-	0xf8000000, 0x807a847a,
-	0xbefc037e, 0xb97bf814,
-	0xbefe037c, 0xbefc037a,
-	0xf4611efa, 0xf8000000,
-	0x807a847a, 0xbefc037e,
-	0xb97bf815, 0xbefe037c,
-	0xbefc037a, 0xf4611efa,
-	0xf8000000, 0x807a847a,
-	0xbefc037e, 0xb97a2a05,
-	0x807a817a, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850002, 0x8f7a897a,
-	0xbf820001, 0x8f7a8a7a,
-	0xb9761e06, 0x8f768a76,
-	0x807a767a, 0xbef603ff,
-	0x01000000, 0xbef20374,
-	0x80747a74, 0x82758075,
-	0xbefc0380, 0xbf800000,
-	0xbe802f00, 0xbe822f02,
-	0xbe842f04, 0xbe862f06,
-	0xbe882f08, 0xbe8a2f0a,
-	0xbe8c2f0c, 0xbe8e2f0e,
-	0xf469003a, 0xfa000000,
-	0xf469013a, 0xfa000010,
-	0xf469023a, 0xfa000020,
-	0xf469033a, 0xfa000030,
-	0x8074c074, 0x82758075,
-	0x807c907c, 0xbf0aff7c,
-	0x00000060, 0xbf85ffea,
-	0xbe802f00, 0xbe822f02,
-	0xbe842f04, 0xbe862f06,
-	0xbe882f08, 0xbe8a2f0a,
-	0xf469003a, 0xfa000000,
-	0xf469013a, 0xfa000010,
-	0xf469023a, 0xfa000020,
-	0x8074b074, 0x82758075,
-	0xbef40372, 0xbefe03c1,
-	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbf850002,
-	0xbeff0380, 0xbf820001,
-	0xbeff03c1, 0xb9714306,
-	0x8771c171, 0xbf840046,
-	0xbf8a0000, 0x8776ff73,
-	0x04000000, 0xbf840042,
-	0x8f718671, 0x8f718271,
-	0xbef60371, 0xb97a2a05,
-	0x807a817a, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850002, 0x8f7a897a,
-	0xbf820001, 0x8f7a8a7a,
-	0xb9761e06, 0x8f768a76,
-	0x807a767a, 0x807aff7a,
-	0x00000200, 0x807aff7a,
-	0x00000080, 0xbef603ff,
-	0x01000000, 0xd7650000,
-	0x000100c1, 0xd7660000,
-	0x000200c1, 0x16000084,
-	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbefc0380,
-	0xbf850012, 0xbe8303ff,
-	0x00000080, 0xbf800000,
-	0xbf800000, 0xbf800000,
-	0xd8d80000, 0x01000000,
-	0xbf8c0000, 0xe0704000,
-	0x7a5d0100, 0x807c037c,
-	0x807a037a, 0xd5250000,
-	0x0001ff00, 0x00000080,
-	0xbf0a717c, 0xbf85fff4,
-	0xbf820011, 0xbe8303ff,
-	0x00000100, 0xbf800000,
-	0xbf800000, 0xbf800000,
-	0xd8d80000, 0x01000000,
-	0xbf8c0000, 0xe0704000,
-	0x7a5d0100, 0x807c037c,
-	0x807a037a, 0xd5250000,
-	0x0001ff00, 0x00000100,
-	0xbf0a717c, 0xbf85fff4,
-	0xbefe03c1, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850004, 0xbefa03ff,
-	0x00000200, 0xbeff0380,
-	0xbf820003, 0xbefa03ff,
-	0x00000400, 0xbeff03c1,
-	0xb9712a05, 0x80718171,
-	0x8f718271, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850017, 0xbef603ff,
-	0x01000000, 0xbefc0384,
-	0xbf0a717c, 0xbf840037,
-	0x7e008700, 0x7e028701,
-	0x7e048702, 0x7e068703,
-	0xe0704000, 0x7a5d0000,
-	0xe0704080, 0x7a5d0100,
-	0xe0704100, 0x7a5d0200,
-	0xe0704180, 0x7a5d0300,
-	0x807c847c, 0x807aff7a,
-	0x00000200, 0xbf0a717c,
-	0xbf85ffef, 0xbf820025,
+	0x80708470, 0xbefc037e,
+	0xb971f814, 0xbefe037c,
+	0xbefc0370, 0xf4611c7a,
+	0xf8000000, 0x80708470,
+	0xbefc037e, 0xb971f815,
+	0xbefe037c, 0xbefc0370,
+	0xf4611c7a, 0xf8000000,
+	0x80708470, 0xbefc037e,
+	0xb9702a05, 0x80708170,
+	0xbf0d9973, 0xbf850002,
+	0x8f708970, 0xbf820001,
+	0x8f708a70, 0xb97a1e06,
+	0x8f7a8a7a, 0x80707a70,
 	0xbef603ff, 0x01000000,
-	0xbefc0384, 0xbf0a717c,
-	0xbf840020, 0x7e008700,
-	0x7e028701, 0x7e048702,
-	0x7e068703, 0xe0704000,
-	0x7a5d0000, 0xe0704100,
-	0x7a5d0100, 0xe0704200,
-	0x7a5d0200, 0xe0704300,
-	0x7a5d0300, 0x807c847c,
-	0x807aff7a, 0x00000400,
-	0xbf0a717c, 0xbf85ffef,
-	0xb9711e06, 0x8771c171,
-	0xbf84000c, 0x8f718371,
-	0x80717c71, 0xbefe03c1,
-	0xbeff0380, 0x7e008700,
-	0xe0704000, 0x7a5d0000,
-	0x807c817c, 0x807aff7a,
-	0x00000080, 0xbf0a717c,
-	0xbf85fff8, 0xbf820142,
-	0xbef4037e, 0x8775ff7f,
-	0x0000ffff, 0x8875ff75,
-	0x00040000, 0xbef60380,
-	0xbef703ff, 0x10807fac,
-	0x8772ff7f, 0x08000000,
-	0x90728372, 0x88777277,
-	0x8772ff7f, 0x70000000,
-	0x90728172, 0x88777277,
-	0xb97302dc, 0x8f739973,
-	0x8873737f, 0x8772ff7f,
-	0x04000000, 0xbf840036,
+	0xbefb0374, 0x80747074,
+	0x82758075, 0xbefc0380,
+	0xbf800000, 0xbe802f00,
+	0xbe822f02, 0xbe842f04,
+	0xbe862f06, 0xbe882f08,
+	0xbe8a2f0a, 0xbe8c2f0c,
+	0xbe8e2f0e, 0xf469003a,
+	0xfa000000, 0xf469013a,
+	0xfa000010, 0xf469023a,
+	0xfa000020, 0xf469033a,
+	0xfa000030, 0x8074c074,
+	0x82758075, 0x807c907c,
+	0xbf0aff7c, 0x00000060,
+	0xbf85ffea, 0xbe802f00,
+	0xbe822f02, 0xbe842f04,
+	0xbe862f06, 0xbe882f08,
+	0xbe8a2f0a, 0xf469003a,
+	0xfa000000, 0xf469013a,
+	0xfa000010, 0xf469023a,
+	0xfa000020, 0x8074b074,
+	0x82758075, 0xbef4037b,
 	0xbefe03c1, 0x907c9973,
 	0x877c817c, 0xbf06817c,
 	0xbf850002, 0xbeff0380,
 	0xbf820001, 0xbeff03c1,
-	0xb96f4306, 0x876fc16f,
-	0xbf84002b, 0x8f6f866f,
-	0x8f6f826f, 0xbef6036f,
-	0xb9782a05, 0x80788178,
-	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbf850002,
-	0x8f788978, 0xbf820001,
-	0x8f788a78, 0xb9721e06,
-	0x8f728a72, 0x80787278,
-	0x8078ff78, 0x00000200,
-	0x8078ff78, 0x00000080,
+	0xb97b4306, 0x877bc17b,
+	0xbf840044, 0xbf8a0000,
+	0x877aff73, 0x04000000,
+	0xbf840040, 0x8f7b867b,
+	0x8f7b827b, 0xbef6037b,
+	0xb9702a05, 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, 0xbefe03c1,
 	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbefc0380,
-	0xbf850009, 0xe0310000,
-	0x781d0000, 0x807cff7c,
-	0x00000080, 0x8078ff78,
-	0x00000080, 0xbf0a6f7c,
-	0xbf85fff8, 0xbf820008,
-	0xe0310000, 0x781d0000,
-	0x807cff7c, 0x00000100,
-	0x8078ff78, 0x00000100,
-	0xbf0a6f7c, 0xbf85fff8,
-	0xbef80380, 0xbefe03c1,
+	0xbf06817c, 0xbf850004,
+	0xbef003ff, 0x00000200,
+	0xbeff0380, 0xbf820003,
+	0xbef003ff, 0x00000400,
+	0xbeff03c1, 0xb97b2a05,
+	0x807b817b, 0x8f7b827b,
 	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, 0xbef603ff,
+	0x01000000, 0xbefc0384,
+	0xbf0a7b7c, 0xbf840020,
+	0x7e008700, 0x7e028701,
+	0x7e048702, 0x7e068703,
+	0xe0704000, 0x705d0000,
+	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,
+	0xbf82014f, 0xbef4037e,
+	0x8775ff7f, 0x0000ffff,
+	0x8875ff75, 0x00040000,
+	0xbef60380, 0xbef703ff,
+	0x10807fac, 0x876eff7f,
+	0x08000000, 0x906e836e,
+	0x88776e77, 0x876eff7f,
+	0x70000000, 0x906e816e,
+	0x88776e77, 0xb97202dc,
+	0x8f729972, 0x8872727f,
+	0x876eff7f, 0x04000000,
+	0xbf840034, 0xbefe03c1,
+	0x907c9972, 0x877c817c,
 	0xbf06817c, 0xbf850002,
 	0xbeff0380, 0xbf820001,
-	0xbeff03c1, 0xb96f2a05,
-	0x806f816f, 0x8f6f826f,
-	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbf850021,
-	0xbef603ff, 0x01000000,
-	0xbef20378, 0x8078ff78,
-	0x00000200, 0xbefc0384,
-	0xe0304000, 0x785d0000,
-	0xe0304080, 0x785d0100,
-	0xe0304100, 0x785d0200,
-	0xe0304180, 0x785d0300,
-	0xbf8c3f70, 0x7e008500,
-	0x7e028501, 0x7e048502,
-	0x7e068503, 0x807c847c,
+	0xbeff03c1, 0xb96f4306,
+	0x876fc16f, 0xbf840029,
+	0x8f6f866f, 0x8f6f826f,
+	0xbef6036f, 0xb9782a05,
+	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,
+	0xbefe03c1, 0x907c9972,
+	0x877c817c, 0xbf06817c,
+	0xbf850002, 0xbeff0380,
+	0xbf820001, 0xbeff03c1,
+	0xb96f2a05, 0x806f816f,
+	0x8f6f826f, 0x907c9972,
+	0x877c817c, 0xbf06817c,
+	0xbf850021, 0xbef603ff,
+	0x01000000, 0xbeee0378,
 	0x8078ff78, 0x00000200,
-	0xbf0a6f7c, 0xbf85ffee,
-	0xe0304000, 0x725d0000,
-	0xe0304080, 0x725d0100,
-	0xe0304100, 0x725d0200,
-	0xe0304180, 0x725d0300,
-	0xbf820032, 0xbef603ff,
-	0x01000000, 0xbef20378,
-	0x8078ff78, 0x00000400,
 	0xbefc0384, 0xe0304000,
-	0x785d0000, 0xe0304100,
-	0x785d0100, 0xe0304200,
-	0x785d0200, 0xe0304300,
+	0x785d0000, 0xe0304080,
+	0x785d0100, 0xe0304100,
+	0x785d0200, 0xe0304180,
 	0x785d0300, 0xbf8c3f70,
 	0x7e008500, 0x7e028501,
 	0x7e048502, 0x7e068503,
 	0x807c847c, 0x8078ff78,
-	0x00000400, 0xbf0a6f7c,
-	0xbf85ffee, 0xb96f1e06,
-	0x876fc16f, 0xbf84000e,
-	0x8f6f836f, 0x806f7c6f,
-	0xbefe03c1, 0xbeff0380,
+	0x00000200, 0xbf0a6f7c,
+	0xbf85ffee, 0xe0304000,
+	0x6e5d0000, 0xe0304080,
+	0x6e5d0100, 0xe0304100,
+	0x6e5d0200, 0xe0304180,
+	0x6e5d0300, 0xbf820032,
+	0xbef603ff, 0x01000000,
+	0xbeee0378, 0x8078ff78,
+	0x00000400, 0xbefc0384,
 	0xe0304000, 0x785d0000,
+	0xe0304100, 0x785d0100,
+	0xe0304200, 0x785d0200,
+	0xe0304300, 0x785d0300,
 	0xbf8c3f70, 0x7e008500,
-	0x807c817c, 0x8078ff78,
-	0x00000080, 0xbf0a6f7c,
-	0xbf85fff7, 0xbeff03c1,
-	0xe0304000, 0x725d0000,
-	0xe0304100, 0x725d0100,
-	0xe0304200, 0x725d0200,
-	0xe0304300, 0x725d0300,
-	0xbf8c3f70, 0xb9782a05,
-	0x80788178, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850002, 0x8f788978,
-	0xbf820001, 0x8f788a78,
-	0xb9721e06, 0x8f728a72,
-	0x80787278, 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, 0xb9782a05,
-	0x80788178, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850002, 0x8f788978,
-	0xbf820001, 0x8f788a78,
-	0xb9721e06, 0x8f728a72,
-	0x80787278, 0x8078ff78,
-	0x00000200, 0xbef603ff,
-	0x01000000, 0xf4211bfa,
+	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,
+	0xb9782a05, 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,
+	0xb9782a05, 0x80788178,
+	0xbf0d9972, 0xbf850002,
+	0x8f788978, 0xbf820001,
+	0x8f788a78, 0xb96e1e06,
+	0x8f6e8a6e, 0x80786e78,
+	0x8078ff78, 0x00000200,
+	0xbef603ff, 0x01000000,
+	0xf4211bfa, 0xf0000000,
+	0x80788478, 0xf4211b3a,
 	0xf0000000, 0x80788478,
-	0xf4211b3a, 0xf0000000,
-	0x80788478, 0xf4211b7a,
+	0xf4211b7a, 0xf0000000,
+	0x80788478, 0xf4211c3a,
 	0xf0000000, 0x80788478,
-	0xf4211eba, 0xf0000000,
-	0x80788478, 0xf4211efa,
+	0xf4211c7a, 0xf0000000,
+	0x80788478, 0xf4211eba,
 	0xf0000000, 0x80788478,
-	0xf4211c3a, 0xf0000000,
-	0x80788478, 0xf4211c7a,
+	0xf4211efa, 0xf0000000,
+	0x80788478, 0xf4211e7a,
 	0xf0000000, 0x80788478,
-	0xf4211e7a, 0xf0000000,
-	0x80788478, 0xf4211cfa,
+	0xf4211cfa, 0xf0000000,
+	0x80788478, 0xf4211bba,
 	0xf0000000, 0x80788478,
+	0xbf8cc07f, 0xb9eef814,
 	0xf4211bba, 0xf0000000,
 	0x80788478, 0xbf8cc07f,
-	0xb9eef814, 0xf4211bba,
-	0xf0000000, 0x80788478,
-	0xbf8cc07f, 0xb9eef815,
-	0xbef2036d, 0x876dff72,
-	0x0000ffff, 0xbefc036f,
-	0xbefe037a, 0xbeff037b,
-	0x876f71ff, 0x000003ff,
+	0xb9eef815, 0xbefc036f,
+	0xbefe0370, 0xbeff0371,
+	0x876f7bff, 0x000003ff,
 	0xb9ef4803, 0xb9f9f816,
-	0x876f71ff, 0xfffff800,
+	0x876f7bff, 0xfffff800,
 	0x906f8b6f, 0xb9efa2c3,
-	0xb9f3f801, 0x876fff72,
+	0xb9f3f801, 0xb96e2a05,
+	0x806e816e, 0xbf0d9972,
+	0xbf850002, 0x8f6e896e,
+	0xbf820001, 0x8f6e8a6e,
+	0x806eff6e, 0x00000200,
+	0x806e746e, 0x826f8075,
+	0x876fff6f, 0x0000ffff,
+	0xf4091c37, 0xfa000050,
+	0xf4091d37, 0xfa000060,
+	0xf4011e77, 0xfa000074,
+	0xbf8cc07f, 0x876fff6d,
 	0xfc000000, 0x906f9a6f,
-	0x8f6f906f, 0xbef30380,
-	0x88736f73, 0x876fff72,
+	0x8f6f906f, 0xbeee0380,
+	0x886e6f6e, 0x876fff6d,
 	0x02000000, 0x906f996f,
-	0x8f6f8f6f, 0x88736f73,
-	0x876fff72, 0x01000000,
+	0x8f6f8f6f, 0x886e6f6e,
+	0x876fff6d, 0x01000000,
 	0x906f986f, 0x8f6f996f,
-	0x88736f73, 0x876fff70,
+	0x886e6f6e, 0x876fff7a,
 	0x00800000, 0x906f976f,
-	0xb9f3f807, 0x87fe7e7e,
-	0x87ea6a6a, 0xb9f0f802,
+	0xb9eef807, 0x876dff6d,
+	0x0000ffff, 0x87fe7e7e,
+	0x87ea6a6a, 0xb9faf802,
 	0xbf8a0000, 0xbe80226c,
 	0xbf810000, 0xbf9f0000,
 	0xbf9f0000, 0xbf9f0000,
@@ -1560,384 +1575,395 @@ static const uint32_t cwsr_trap_arcturus_hex[] = {
 };
 
 static const uint32_t cwsr_trap_gfx10_hex[] = {
-	0xbf820001, 0xbf8201c6,
+	0xbf820001, 0xbf8201cf,
 	0xb0804004, 0xb978f802,
-	0x8a788678, 0xb971f803,
-	0x876eff71, 0x00000400,
-	0xbf85001d, 0x876eff71,
-	0x00000100, 0xbf840002,
-	0x8878ff78, 0x00002000,
-	0xb970f812, 0xb971f813,
-	0x8ff08870, 0xf4051bb8,
-	0xfa000000, 0xbf8cc07f,
-	0xf4051c38, 0xfa000008,
-	0xbf8cc07f, 0x87ee6e6e,
-	0xbf840001, 0xbe80206e,
-	0xb971f803, 0x8771ff71,
-	0x000001ff, 0xbf850002,
-	0x806c846c, 0x826d806d,
-	0x876dff6d, 0x0000ffff,
-	0x87fe7e7e, 0x87ea6a6a,
-	0xb9f8f802, 0xbe80226c,
-	0xb971f803, 0x8771ff71,
-	0x00000100, 0xbf840006,
-	0xbef60380, 0xb9f60203,
-	0x876dff6d, 0x0000ffff,
-	0x80ec886c, 0x82ed806d,
-	0xbef60380, 0xb9f60283,
-	0xbeee037e, 0xbeef037f,
-	0xbefe0480, 0xbf900004,
-	0xbf8cc07f, 0xbef4037e,
-	0x8775ff7f, 0x0000ffff,
-	0x8875ff75, 0x00040000,
-	0xbef60380, 0xbef703ff,
-	0x10807fac, 0x8776ff7f,
-	0x08000000, 0x90768376,
-	0x88777677, 0x8776ff7f,
-	0x70000000, 0x90768176,
-	0x88777677, 0xbefb037c,
-	0xbefa0380, 0xb97302dc,
-	0x8f739973, 0x8873737f,
-	0xbefe03c1, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850002, 0xbeff0380,
-	0xbf820002, 0xbeff03c1,
-	0xbf82000b, 0xbef603ff,
-	0x01000000, 0xe0704000,
-	0x7a5d0000, 0xe0704080,
-	0x7a5d0100, 0xe0704100,
-	0x7a5d0200, 0xe0704180,
-	0x7a5d0300, 0xbf82000a,
-	0xbef603ff, 0x01000000,
-	0xe0704000, 0x7a5d0000,
-	0xe0704100, 0x7a5d0100,
-	0xe0704200, 0x7a5d0200,
-	0xe0704300, 0x7a5d0300,
-	0xb97a2a05, 0x807a817a,
+	0x8a788678, 0xb96ef801,
+	0x876eff6e, 0x00000800,
+	0xbf840003, 0x876eff78,
+	0x00002000, 0xbf840009,
+	0xb97bf803, 0x876eff7b,
+	0x00000400, 0xbf85001d,
+	0x876eff7b, 0x00000100,
+	0xbf840002, 0x8878ff78,
+	0x00002000, 0xb97af812,
+	0xb97bf813, 0x8ffa887a,
+	0xf4051bbd, 0xfa000000,
+	0xbf8cc07f, 0xf4051ebd,
+	0xfa000008, 0xbf8cc07f,
+	0x87ee6e6e, 0xbf840001,
+	0xbe80206e, 0xb97bf803,
+	0x877bff7b, 0x000001ff,
+	0xbf850002, 0x806c846c,
+	0x826d806d, 0x876dff6d,
+	0x0000ffff, 0x87fe7e7e,
+	0x87ea6a6a, 0xb9f8f802,
+	0xbe80226c, 0x876dff6d,
+	0x0000ffff, 0xbefa0380,
+	0xb9fa0283, 0xbeee037e,
+	0xbeef037f, 0xbefe0480,
+	0xbf900004, 0xbf8cc07f,
+	0xb97b02dc, 0x8f7b997b,
+	0x887b7b7f, 0xb97a2a05,
+	0x807a817a, 0xbf0d997b,
+	0xbf850002, 0x8f7a897a,
+	0xbf820001, 0x8f7a8a7a,
+	0x877bff7f, 0x0000ffff,
+	0x807aff7a, 0x00000200,
+	0x807a7e7a, 0x827b807b,
+	0xbef4037e, 0x8775ff7f,
+	0x0000ffff, 0x8875ff75,
+	0x00040000, 0xbef60380,
+	0xbef703ff, 0x10807fac,
+	0x877aff7f, 0x08000000,
+	0x907a837a, 0x88777a77,
+	0x877aff7f, 0x70000000,
+	0x907a817a, 0x88777a77,
+	0xbef1037c, 0xbef00380,
+	0xb97302dc, 0x8f739973,
+	0x8873737f, 0xbefe03c1,
 	0x907c9973, 0x877c817c,
 	0xbf06817c, 0xbf850002,
-	0x8f7a897a, 0xbf820001,
-	0x8f7a8a7a, 0xb9761e06,
-	0x8f768a76, 0x807a767a,
-	0x807aff7a, 0x00000200,
+	0xbeff0380, 0xbf820002,
+	0xbeff03c1, 0xbf82000b,
 	0xbef603ff, 0x01000000,
-	0x7e000280, 0x7e020280,
-	0x7e040280, 0xbefc0380,
-	0xd7610002, 0x0000f87b,
+	0xe0704000, 0x705d0000,
+	0xe0704080, 0x705d0100,
+	0xe0704100, 0x705d0200,
+	0xe0704180, 0x705d0300,
+	0xbf82000a, 0xbef603ff,
+	0x01000000, 0xe0704000,
+	0x705d0000, 0xe0704100,
+	0x705d0100, 0xe0704200,
+	0x705d0200, 0xe0704300,
+	0x705d0300, 0xb9702a05,
+	0x80708170, 0xbf0d9973,
+	0xbf850002, 0x8f708970,
+	0xbf820001, 0x8f708a70,
+	0xb97a1e06, 0x8f7a8a7a,
+	0x80707a70, 0x8070ff70,
+	0x00000200, 0xbef603ff,
+	0x01000000, 0x7e000280,
+	0x7e020280, 0x7e040280,
+	0xbefc0380, 0xd7610002,
+	0x0000f871, 0x807c817c,
+	0xd7610002, 0x0000f86c,
 	0x807c817c, 0xd7610002,
-	0x0000f86c, 0x807c817c,
-	0xd7610002, 0x0000f86d,
+	0x0000f86d, 0x807c817c,
+	0xd7610002, 0x0000f86e,
 	0x807c817c, 0xd7610002,
-	0x0000f86e, 0x807c817c,
-	0xd7610002, 0x0000f86f,
+	0x0000f86f, 0x807c817c,
+	0xd7610002, 0x0000f878,
+	0x807c817c, 0xb97af803,
+	0xd7610002, 0x0000f87a,
 	0x807c817c, 0xd7610002,
-	0x0000f878, 0x807c817c,
-	0xb971f803, 0xd7610002,
+	0x0000f87b, 0x807c817c,
+	0xb971f801, 0xd7610002,
 	0x0000f871, 0x807c817c,
-	0xd7610002, 0x0000f872,
-	0x807c817c, 0xb97bf801,
-	0xd7610002, 0x0000f87b,
-	0x807c817c, 0xb97bf814,
-	0xd7610002, 0x0000f87b,
-	0x807c817c, 0xb97bf815,
-	0xd7610002, 0x0000f87b,
-	0x807c817c, 0xbeff0380,
-	0xe0704000, 0x7a5d0200,
-	0xb97a2a05, 0x807a817a,
-	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbf850002,
-	0x8f7a897a, 0xbf820001,
-	0x8f7a8a7a, 0xb9761e06,
-	0x8f768a76, 0x807a767a,
-	0xbef603ff, 0x01000000,
-	0xbef90380, 0xbefc0380,
-	0xbf800000, 0xbe802f00,
-	0xbe822f02, 0xbe842f04,
-	0xbe862f06, 0xbe882f08,
-	0xbe8a2f0a, 0xbe8c2f0c,
-	0xbe8e2f0e, 0xd7610002,
-	0x0000f200, 0x80798179,
-	0xd7610002, 0x0000f201,
+	0xb971f814, 0xd7610002,
+	0x0000f871, 0x807c817c,
+	0xb971f815, 0xd7610002,
+	0x0000f871, 0x807c817c,
+	0xbeff0380, 0xe0704000,
+	0x705d0200, 0xb9702a05,
+	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,
-	0x7a5d0200, 0x807aff7a,
-	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,
-	0x7a5d0200, 0xbefe03c1,
-	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbf850002,
-	0xbeff0380, 0xbf820001,
-	0xbeff03c1, 0xb9714306,
-	0x8771c171, 0xbf840046,
-	0xbf8a0000, 0x8776ff73,
-	0x04000000, 0xbf840042,
-	0x8f718671, 0x8f718271,
-	0xbef60371, 0xb97a2a05,
-	0x807a817a, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850002, 0x8f7a897a,
-	0xbf820001, 0x8f7a8a7a,
-	0xb9761e06, 0x8f768a76,
-	0x807a767a, 0x807aff7a,
-	0x00000200, 0x807aff7a,
-	0x00000080, 0xbef603ff,
-	0x01000000, 0xd7650000,
-	0x000100c1, 0xd7660000,
-	0x000200c1, 0x16000084,
-	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbefc0380,
-	0xbf850012, 0xbe8303ff,
-	0x00000080, 0xbf800000,
-	0xbf800000, 0xbf800000,
-	0xd8d80000, 0x01000000,
-	0xbf8c0000, 0xe0704000,
-	0x7a5d0100, 0x807c037c,
-	0x807a037a, 0xd5250000,
-	0x0001ff00, 0x00000080,
-	0xbf0a717c, 0xbf85fff4,
-	0xbf820011, 0xbe8303ff,
-	0x00000100, 0xbf800000,
-	0xbf800000, 0xbf800000,
-	0xd8d80000, 0x01000000,
-	0xbf8c0000, 0xe0704000,
-	0x7a5d0100, 0x807c037c,
-	0x807a037a, 0xd5250000,
-	0x0001ff00, 0x00000100,
-	0xbf0a717c, 0xbf85fff4,
-	0xbefe03c1, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850004, 0xbefa03ff,
-	0x00000200, 0xbeff0380,
-	0xbf820003, 0xbefa03ff,
-	0x00000400, 0xbeff03c1,
-	0xb9712a05, 0x80718171,
-	0x8f718271, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850017, 0xbef603ff,
-	0x01000000, 0xbefc0384,
-	0xbf0a717c, 0xbf840037,
-	0x7e008700, 0x7e028701,
-	0x7e048702, 0x7e068703,
-	0xe0704000, 0x7a5d0000,
-	0xe0704080, 0x7a5d0100,
-	0xe0704100, 0x7a5d0200,
-	0xe0704180, 0x7a5d0300,
-	0x807c847c, 0x807aff7a,
-	0x00000200, 0xbf0a717c,
-	0xbf85ffef, 0xbf820025,
-	0xbef603ff, 0x01000000,
-	0xbefc0384, 0xbf0a717c,
-	0xbf840020, 0x7e008700,
-	0x7e028701, 0x7e048702,
-	0x7e068703, 0xe0704000,
-	0x7a5d0000, 0xe0704100,
-	0x7a5d0100, 0xe0704200,
-	0x7a5d0200, 0xe0704300,
-	0x7a5d0300, 0x807c847c,
-	0x807aff7a, 0x00000400,
-	0xbf0a717c, 0xbf85ffef,
-	0xb9711e06, 0x8771c171,
-	0xbf84000c, 0x8f718371,
-	0x80717c71, 0xbefe03c1,
-	0xbeff0380, 0x7e008700,
-	0xe0704000, 0x7a5d0000,
-	0x807c817c, 0x807aff7a,
-	0x00000080, 0xbf0a717c,
-	0xbf85fff8, 0xbf82012d,
-	0xbef4037e, 0x8775ff7f,
-	0x0000ffff, 0x8875ff75,
-	0x00040000, 0xbef60380,
-	0xbef703ff, 0x10807fac,
-	0x8772ff7f, 0x08000000,
-	0x90728372, 0x88777277,
-	0x8772ff7f, 0x70000000,
-	0x90728172, 0x88777277,
-	0xb97302dc, 0x8f739973,
-	0x8873737f, 0x8772ff7f,
-	0x04000000, 0xbf840036,
+	0x0000f207, 0x80798179,
+	0xd7610002, 0x0000f208,
+	0x80798179, 0xd7610002,
+	0x0000f209, 0x80798179,
+	0xd7610002, 0x0000f20a,
+	0x80798179, 0xd7610002,
+	0x0000f20b, 0x80798179,
+	0xe0704000, 0x705d0200,
 	0xbefe03c1, 0x907c9973,
 	0x877c817c, 0xbf06817c,
 	0xbf850002, 0xbeff0380,
 	0xbf820001, 0xbeff03c1,
-	0xb96f4306, 0x876fc16f,
-	0xbf84002b, 0x8f6f866f,
-	0x8f6f826f, 0xbef6036f,
-	0xb9782a05, 0x80788178,
-	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbf850002,
-	0x8f788978, 0xbf820001,
-	0x8f788a78, 0xb9721e06,
-	0x8f728a72, 0x80787278,
-	0x8078ff78, 0x00000200,
-	0x8078ff78, 0x00000080,
+	0xb97b4306, 0x877bc17b,
+	0xbf840044, 0xbf8a0000,
+	0x877aff73, 0x04000000,
+	0xbf840040, 0x8f7b867b,
+	0x8f7b827b, 0xbef6037b,
+	0xb9702a05, 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, 0xbefe03c1,
 	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbefc0380,
-	0xbf850009, 0xe0310000,
-	0x781d0000, 0x807cff7c,
-	0x00000080, 0x8078ff78,
-	0x00000080, 0xbf0a6f7c,
-	0xbf85fff8, 0xbf820008,
-	0xe0310000, 0x781d0000,
-	0x807cff7c, 0x00000100,
-	0x8078ff78, 0x00000100,
-	0xbf0a6f7c, 0xbf85fff8,
-	0xbef80380, 0xbefe03c1,
+	0xbf06817c, 0xbf850004,
+	0xbef003ff, 0x00000200,
+	0xbeff0380, 0xbf820003,
+	0xbef003ff, 0x00000400,
+	0xbeff03c1, 0xb97b2a05,
+	0x807b817b, 0x8f7b827b,
 	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, 0xbef603ff,
+	0x01000000, 0xbefc0384,
+	0xbf0a7b7c, 0xbf840020,
+	0x7e008700, 0x7e028701,
+	0x7e048702, 0x7e068703,
+	0xe0704000, 0x705d0000,
+	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,
+	0xbf82013a, 0xbef4037e,
+	0x8775ff7f, 0x0000ffff,
+	0x8875ff75, 0x00040000,
+	0xbef60380, 0xbef703ff,
+	0x10807fac, 0x876eff7f,
+	0x08000000, 0x906e836e,
+	0x88776e77, 0x876eff7f,
+	0x70000000, 0x906e816e,
+	0x88776e77, 0xb97202dc,
+	0x8f729972, 0x8872727f,
+	0x876eff7f, 0x04000000,
+	0xbf840034, 0xbefe03c1,
+	0x907c9972, 0x877c817c,
 	0xbf06817c, 0xbf850002,
 	0xbeff0380, 0xbf820001,
-	0xbeff03c1, 0xb96f2a05,
-	0x806f816f, 0x8f6f826f,
-	0x907c9973, 0x877c817c,
-	0xbf06817c, 0xbf850021,
-	0xbef603ff, 0x01000000,
-	0xbef20378, 0x8078ff78,
-	0x00000200, 0xbefc0384,
-	0xe0304000, 0x785d0000,
-	0xe0304080, 0x785d0100,
-	0xe0304100, 0x785d0200,
-	0xe0304180, 0x785d0300,
-	0xbf8c3f70, 0x7e008500,
-	0x7e028501, 0x7e048502,
-	0x7e068503, 0x807c847c,
+	0xbeff03c1, 0xb96f4306,
+	0x876fc16f, 0xbf840029,
+	0x8f6f866f, 0x8f6f826f,
+	0xbef6036f, 0xb9782a05,
+	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,
+	0xbefe03c1, 0x907c9972,
+	0x877c817c, 0xbf06817c,
+	0xbf850002, 0xbeff0380,
+	0xbf820001, 0xbeff03c1,
+	0xb96f2a05, 0x806f816f,
+	0x8f6f826f, 0x907c9972,
+	0x877c817c, 0xbf06817c,
+	0xbf850021, 0xbef603ff,
+	0x01000000, 0xbeee0378,
 	0x8078ff78, 0x00000200,
-	0xbf0a6f7c, 0xbf85ffee,
-	0xe0304000, 0x725d0000,
-	0xe0304080, 0x725d0100,
-	0xe0304100, 0x725d0200,
-	0xe0304180, 0x725d0300,
-	0xbf820032, 0xbef603ff,
-	0x01000000, 0xbef20378,
-	0x8078ff78, 0x00000400,
 	0xbefc0384, 0xe0304000,
-	0x785d0000, 0xe0304100,
-	0x785d0100, 0xe0304200,
-	0x785d0200, 0xe0304300,
+	0x785d0000, 0xe0304080,
+	0x785d0100, 0xe0304100,
+	0x785d0200, 0xe0304180,
 	0x785d0300, 0xbf8c3f70,
 	0x7e008500, 0x7e028501,
 	0x7e048502, 0x7e068503,
 	0x807c847c, 0x8078ff78,
-	0x00000400, 0xbf0a6f7c,
-	0xbf85ffee, 0xb96f1e06,
-	0x876fc16f, 0xbf84000e,
-	0x8f6f836f, 0x806f7c6f,
-	0xbefe03c1, 0xbeff0380,
+	0x00000200, 0xbf0a6f7c,
+	0xbf85ffee, 0xe0304000,
+	0x6e5d0000, 0xe0304080,
+	0x6e5d0100, 0xe0304100,
+	0x6e5d0200, 0xe0304180,
+	0x6e5d0300, 0xbf820032,
+	0xbef603ff, 0x01000000,
+	0xbeee0378, 0x8078ff78,
+	0x00000400, 0xbefc0384,
 	0xe0304000, 0x785d0000,
+	0xe0304100, 0x785d0100,
+	0xe0304200, 0x785d0200,
+	0xe0304300, 0x785d0300,
 	0xbf8c3f70, 0x7e008500,
-	0x807c817c, 0x8078ff78,
-	0x00000080, 0xbf0a6f7c,
-	0xbf85fff7, 0xbeff03c1,
-	0xe0304000, 0x725d0000,
-	0xe0304100, 0x725d0100,
-	0xe0304200, 0x725d0200,
-	0xe0304300, 0x725d0300,
-	0xbf8c3f70, 0xb9782a05,
-	0x80788178, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850002, 0x8f788978,
-	0xbf820001, 0x8f788a78,
-	0xb9721e06, 0x8f728a72,
-	0x80787278, 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, 0xb9782a05,
-	0x80788178, 0x907c9973,
-	0x877c817c, 0xbf06817c,
-	0xbf850002, 0x8f788978,
-	0xbf820001, 0x8f788a78,
-	0xb9721e06, 0x8f728a72,
-	0x80787278, 0x8078ff78,
-	0x00000200, 0xbef603ff,
-	0x01000000, 0xf4211bfa,
+	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,
+	0xb9782a05, 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,
+	0xb9782a05, 0x80788178,
+	0xbf0d9972, 0xbf850002,
+	0x8f788978, 0xbf820001,
+	0x8f788a78, 0xb96e1e06,
+	0x8f6e8a6e, 0x80786e78,
+	0x8078ff78, 0x00000200,
+	0xbef603ff, 0x01000000,
+	0xf4211bfa, 0xf0000000,
+	0x80788478, 0xf4211b3a,
 	0xf0000000, 0x80788478,
-	0xf4211b3a, 0xf0000000,
-	0x80788478, 0xf4211b7a,
+	0xf4211b7a, 0xf0000000,
+	0x80788478, 0xf4211c3a,
 	0xf0000000, 0x80788478,
-	0xf4211eba, 0xf0000000,
-	0x80788478, 0xf4211efa,
+	0xf4211c7a, 0xf0000000,
+	0x80788478, 0xf4211eba,
 	0xf0000000, 0x80788478,
-	0xf4211c3a, 0xf0000000,
-	0x80788478, 0xf4211c7a,
+	0xf4211efa, 0xf0000000,
+	0x80788478, 0xf4211e7a,
 	0xf0000000, 0x80788478,
-	0xf4211e7a, 0xf0000000,
-	0x80788478, 0xf4211cfa,
+	0xf4211cfa, 0xf0000000,
+	0x80788478, 0xf4211bba,
 	0xf0000000, 0x80788478,
+	0xbf8cc07f, 0xb9eef814,
 	0xf4211bba, 0xf0000000,
 	0x80788478, 0xbf8cc07f,
-	0xb9eef814, 0xf4211bba,
-	0xf0000000, 0x80788478,
-	0xbf8cc07f, 0xb9eef815,
-	0xbef2036d, 0x876dff72,
-	0x0000ffff, 0xbefc036f,
-	0xbefe037a, 0xbeff037b,
-	0x876f71ff, 0x000003ff,
-	0xb9ef4803, 0x876f71ff,
+	0xb9eef815, 0xbefc036f,
+	0xbefe0370, 0xbeff0371,
+	0x876f7bff, 0x000003ff,
+	0xb9ef4803, 0x876f7bff,
 	0xfffff800, 0x906f8b6f,
 	0xb9efa2c3, 0xb9f3f801,
+	0xb96e2a05, 0x806e816e,
+	0xbf0d9972, 0xbf850002,
+	0x8f6e896e, 0xbf820001,
+	0x8f6e8a6e, 0x806eff6e,
+	0x00000200, 0x806e746e,
+	0x826f8075, 0x876fff6f,
+	0x0000ffff, 0xf4091c37,
+	0xfa000050, 0xf4091d37,
+	0xfa000060, 0xf4011e77,
+	0xfa000074, 0xbf8cc07f,
+	0x876dff6d, 0x0000ffff,
 	0x87fe7e7e, 0x87ea6a6a,
-	0xb9f0f802, 0xbf8a0000,
+	0xb9faf802, 0xbf8a0000,
 	0xbe80226c, 0xbf810000,
 	0xbf9f0000, 0xbf9f0000,
 	0xbf9f0000, 0xbf9f0000,
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 05dd29035772..5b220f2a7501 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm
@@ -33,6 +33,8 @@
 
 #define NO_SQC_STORE !ASIC_TARGET_NAVI1X
 
+var SINGLE_STEP_MISSED_WORKAROUND		= 1	//workaround for lost MODE.DEBUG_EN exception when SAVECTX raised
+
 var SQ_WAVE_STATUS_INST_ATC_SHIFT		= 23
 var SQ_WAVE_STATUS_INST_ATC_MASK		= 0x00800000
 var SQ_WAVE_STATUS_SPI_PRIO_MASK		= 0x00000006
@@ -72,6 +74,8 @@ var SQ_WAVE_IB_STS_RCNT_SIZE			= 6
 var SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK	= 0x003F8000
 var SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK_NEG	= 0x00007FFF
 
+var SQ_WAVE_MODE_DEBUG_EN_MASK			= 0x800
+
 var SQ_BUF_RSRC_WORD1_ATC_SHIFT			= 24
 var SQ_BUF_RSRC_WORD3_MTYPE_SHIFT		= 27
 
@@ -109,17 +113,19 @@ var s_save_pc_hi				= ttmp1
 var s_save_exec_lo				= ttmp2
 var s_save_exec_hi				= ttmp3
 var s_save_status				= ttmp12
-var s_save_trapsts				= ttmp5
-var s_save_xnack_mask				= ttmp6
+var s_save_trapsts				= ttmp15
+var s_save_xnack_mask				= s_save_trapsts
 var s_wave_size					= ttmp7
 var s_save_buf_rsrc0				= ttmp8
 var s_save_buf_rsrc1				= ttmp9
 var s_save_buf_rsrc2				= ttmp10
 var s_save_buf_rsrc3				= ttmp11
-var s_save_mem_offset				= ttmp14
+var s_save_mem_offset				= ttmp4
 var s_save_alloc_size				= s_save_trapsts
-var s_save_tmp					= s_save_buf_rsrc2
-var s_save_m0					= ttmp15
+var s_save_tmp					= ttmp14
+var s_save_m0					= ttmp5
+var s_save_ttmps_lo				= s_save_tmp
+var s_save_ttmps_hi				= s_save_trapsts
 
 var S_RESTORE_BUF_RSRC_WORD1_STRIDE		= S_SAVE_BUF_RSRC_WORD1_STRIDE
 var S_RESTORE_BUF_RSRC_WORD3_MISC		= S_SAVE_BUF_RSRC_WORD3_MISC
@@ -141,23 +147,25 @@ var s_restore_spi_init_lo			= exec_lo
 var s_restore_spi_init_hi			= exec_hi
 var s_restore_mem_offset			= ttmp12
 var s_restore_alloc_size			= ttmp3
-var s_restore_tmp				= ttmp6
+var s_restore_tmp				= ttmp2
 var s_restore_mem_offset_save			= s_restore_tmp
 var s_restore_m0				= s_restore_alloc_size
 var s_restore_mode				= ttmp7
-var s_restore_flat_scratch			= ttmp2
+var s_restore_flat_scratch			= s_restore_tmp
 var s_restore_pc_lo				= ttmp0
 var s_restore_pc_hi				= ttmp1
-var s_restore_exec_lo				= ttmp14
-var s_restore_exec_hi				= ttmp15
-var s_restore_status				= ttmp4
-var s_restore_trapsts				= ttmp5
+var s_restore_exec_lo				= ttmp4
+var s_restore_exec_hi				= ttmp5
+var s_restore_status				= ttmp14
+var s_restore_trapsts				= ttmp15
 var s_restore_xnack_mask			= ttmp13
 var s_restore_buf_rsrc0				= ttmp8
 var s_restore_buf_rsrc1				= ttmp9
 var s_restore_buf_rsrc2				= ttmp10
 var s_restore_buf_rsrc3				= ttmp11
-var s_restore_size				= ttmp7
+var s_restore_size				= ttmp6
+var s_restore_ttmps_lo				= s_restore_tmp
+var s_restore_ttmps_hi				= s_restore_alloc_size
 
 shader main
 	asic(DEFAULT)
@@ -172,6 +180,24 @@ L_JUMP_TO_RESTORE:
 L_SKIP_RESTORE:
 	s_getreg_b32	s_save_status, hwreg(HW_REG_STATUS)			//save STATUS since we will change SCC
 	s_andn2_b32	s_save_status, s_save_status, SQ_WAVE_STATUS_SPI_PRIO_MASK
+
+if SINGLE_STEP_MISSED_WORKAROUND
+	// No single step exceptions if MODE.DEBUG_EN=0.
+	s_getreg_b32    ttmp2, hwreg(HW_REG_MODE)
+	s_and_b32       ttmp2, ttmp2, SQ_WAVE_MODE_DEBUG_EN_MASK
+	s_cbranch_scc0  L_NO_SINGLE_STEP_WORKAROUND
+
+	// Second-level trap already handled exception if STATUS.HALT=1.
+	s_and_b32       ttmp2, s_save_status, SQ_WAVE_STATUS_HALT_MASK
+
+	// Prioritize single step exception over context save.
+	// Second-level trap will halt wave and RFE, re-entering for SAVECTX.
+	s_cbranch_scc0  L_FETCH_2ND_TRAP
+
+L_NO_SINGLE_STEP_WORKAROUND:
+end
+
+
 	s_getreg_b32	s_save_trapsts, hwreg(HW_REG_TRAPSTS)
 	s_and_b32	ttmp2, s_save_trapsts, SQ_WAVE_TRAPSTS_SAVECTX_MASK	//check whether this is for save
 	s_cbranch_scc1	L_SAVE
@@ -241,25 +267,11 @@ L_EXCP_CASE:
 	s_rfe_b64	[ttmp0, ttmp1]
 
 L_SAVE:
-	//check whether there is mem_viol
-	s_getreg_b32	s_save_trapsts, hwreg(HW_REG_TRAPSTS)
-	s_and_b32	s_save_trapsts, s_save_trapsts, SQ_WAVE_TRAPSTS_MEM_VIOL_MASK
-	s_cbranch_scc0	L_NO_PC_REWIND
-
-	//if so, need rewind PC assuming GDS operation gets NACKed
-	s_mov_b32	s_save_tmp, 0
-	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_MEM_VIOL_SHIFT, 1), s_save_tmp	//clear mem_viol bit
 	s_and_b32	s_save_pc_hi, s_save_pc_hi, 0x0000ffff			//pc[47:32]
-	s_sub_u32	s_save_pc_lo, s_save_pc_lo, 8				//pc[31:0]-8
-	s_subb_u32	s_save_pc_hi, s_save_pc_hi, 0x0
-
-L_NO_PC_REWIND:
 	s_mov_b32	s_save_tmp, 0
 	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_SAVECTX_SHIFT, 1), s_save_tmp	//clear saveCtx bit
 
 #if ASIC_TARGET_NAVI1X
-	s_getreg_b32	s_save_xnack_mask, hwreg(HW_REG_SHADER_XNACK_MASK)
-
 	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_RCNT_SHIFT, SQ_WAVE_IB_STS_RCNT_SIZE)
 	s_lshl_b32	s_save_tmp, s_save_tmp, S_SAVE_PC_HI_RCNT_SHIFT
 	s_or_b32	s_save_pc_hi, s_save_pc_hi, s_save_tmp
@@ -293,6 +305,21 @@ L_SLEEP:
 	s_waitcnt	lgkmcnt(0)
 #endif
 
+	// Save trap temporaries 4-11, 13 initialized by SPI debug dispatch logic
+	// ttmp SR memory offset : size(VGPR)+size(SGPR)+0x40
+	get_wave_size(s_save_ttmps_hi)
+	get_vgpr_size_bytes(s_save_ttmps_lo, s_save_ttmps_hi)
+	s_and_b32	s_save_ttmps_hi, s_save_spi_init_hi, 0xFFFF
+	s_add_u32	s_save_ttmps_lo, s_save_ttmps_lo, get_sgpr_size_bytes()
+	s_add_u32	s_save_ttmps_lo, s_save_ttmps_lo, s_save_spi_init_lo
+	s_addc_u32	s_save_ttmps_hi, s_save_ttmps_hi, 0x0
+
+#if ASIC_TARGET_NAVI1X
+	s_store_dwordx4	[ttmp4, ttmp5, ttmp6, ttmp7], [s_save_ttmps_lo, s_save_ttmps_hi], 0x50 glc:1
+	s_store_dwordx4	[ttmp8, ttmp9, ttmp10, ttmp11], [s_save_ttmps_lo, s_save_ttmps_hi], 0x60 glc:1
+	s_store_dword   ttmp13, [s_save_ttmps_lo, s_save_ttmps_hi], 0x74 glc:1
+#endif
+
 	/* setup Resource Contants */
 	s_mov_b32	s_save_buf_rsrc0, s_save_spi_init_lo			//base_addr_lo
 	s_and_b32	s_save_buf_rsrc1, s_save_spi_init_hi, 0x0000FFFF	//base_addr_hi
@@ -310,9 +337,13 @@ L_SLEEP:
 
 	/* global mem offset */
 	s_mov_b32	s_save_mem_offset, 0x0
-	s_getreg_b32	s_wave_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE)
-	s_lshl_b32	s_wave_size, s_wave_size, S_WAVE_SIZE
-	s_or_b32	s_wave_size, s_save_spi_init_hi, s_wave_size		//share s_wave_size with exec_hi, it's at bit25
+	get_wave_size(s_wave_size)
+
+#if ASIC_TARGET_NAVI1X
+	// Save and clear vector XNACK state late to free up SGPRs.
+	s_getreg_b32	s_save_xnack_mask, hwreg(HW_REG_SHADER_XNACK_MASK)
+	s_setreg_imm32_b32	hwreg(HW_REG_SHADER_XNACK_MASK), 0x0
+#endif
 
 	/* save first 4 VGPRs, needed for SGPR save */
 	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
@@ -371,8 +402,8 @@ L_SAVE_HWREG:
 	write_hwreg_to_mem(s_save_exec_hi, s_save_buf_rsrc0, s_save_mem_offset)
 	write_hwreg_to_mem(s_save_status, s_save_buf_rsrc0, s_save_mem_offset)
 
-	s_getreg_b32	s_save_trapsts, hwreg(HW_REG_TRAPSTS)
-	write_hwreg_to_mem(s_save_trapsts, s_save_buf_rsrc0, s_save_mem_offset)
+	s_getreg_b32	s_save_tmp, hwreg(HW_REG_TRAPSTS)
+	write_hwreg_to_mem(s_save_tmp, s_save_buf_rsrc0, s_save_mem_offset)
 
 	// Not used on Sienna_Cichlid but keep layout same for debugger.
 	write_hwreg_to_mem(s_save_xnack_mask, s_save_buf_rsrc0, s_save_mem_offset)
@@ -650,9 +681,7 @@ L_RESTORE:
 	s_lshr_b32	s_restore_tmp, s_restore_tmp, (S_RESTORE_SPI_INIT_MTYPE_SHIFT-SQ_BUF_RSRC_WORD3_MTYPE_SHIFT)
 	s_or_b32	s_restore_buf_rsrc3, s_restore_buf_rsrc3, s_restore_tmp	//or MTYPE
 	//determine it is wave32 or wave64
-	s_getreg_b32	s_restore_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE)
-	s_lshl_b32	s_restore_size, s_restore_size, S_WAVE_SIZE
-	s_or_b32	s_restore_size, s_restore_spi_init_hi, s_restore_size
+	get_wave_size(s_restore_size)
 
 	s_and_b32	s_restore_tmp, s_restore_spi_init_hi, S_RESTORE_SPI_INIT_FIRST_WAVE_MASK
 	s_cbranch_scc0	L_RESTORE_VGPR
@@ -686,7 +715,7 @@ L_RESTORE_LDS_NORMAL:
 
 	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
 
-	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
+	s_lshr_b32	m0, s_restore_size, S_WAVE_SIZE
 	s_and_b32	m0, m0, 1
 	s_cmp_eq_u32	m0, 1
 	s_mov_b32	m0, 0x0
@@ -894,9 +923,6 @@ L_RESTORE_HWREG:
 
 	s_setreg_b32	hwreg(HW_REG_SHADER_FLAT_SCRATCH_HI), s_restore_flat_scratch
 
-	s_mov_b32	s_restore_tmp, s_restore_pc_hi
-	s_and_b32	s_restore_pc_hi, s_restore_tmp, 0x0000ffff		//pc[47:32] //Do it here in order not to affect STATUS
-
 	s_mov_b32	m0, s_restore_m0
 	s_mov_b32	exec_lo, s_restore_exec_lo
 	s_mov_b32	exec_hi, s_restore_exec_hi
@@ -913,26 +939,39 @@ L_RESTORE_HWREG:
 	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT, SQ_WAVE_TRAPSTS_POST_SAVECTX_SIZE), s_restore_m0
 	s_setreg_b32	hwreg(HW_REG_MODE), s_restore_mode
 
+	// Restore trap temporaries 4-11, 13 initialized by SPI debug dispatch logic
+	// ttmp SR memory offset : size(VGPR)+size(SGPR)+0x40
+	get_vgpr_size_bytes(s_restore_ttmps_lo, s_restore_size)
+	s_add_u32	s_restore_ttmps_lo, s_restore_ttmps_lo, get_sgpr_size_bytes()
+	s_add_u32	s_restore_ttmps_lo, s_restore_ttmps_lo, s_restore_buf_rsrc0
+	s_addc_u32	s_restore_ttmps_hi, s_restore_buf_rsrc1, 0x0
+	s_and_b32	s_restore_ttmps_hi, s_restore_ttmps_hi, 0xFFFF
+	s_load_dwordx4	[ttmp4, ttmp5, ttmp6, ttmp7], [s_restore_ttmps_lo, s_restore_ttmps_hi], 0x50 glc:1
+	s_load_dwordx4	[ttmp8, ttmp9, ttmp10, ttmp11], [s_restore_ttmps_lo, s_restore_ttmps_hi], 0x60 glc:1
+	s_load_dword	ttmp13, [s_restore_ttmps_lo, s_restore_ttmps_hi], 0x74 glc:1
+	s_waitcnt	lgkmcnt(0)
+
 #if ASIC_TARGET_NAVI1X
-	s_and_b32	s_restore_m0, s_restore_tmp, S_SAVE_PC_HI_RCNT_MASK
+	s_and_b32	s_restore_m0, s_restore_pc_hi, S_SAVE_PC_HI_RCNT_MASK
 	s_lshr_b32	s_restore_m0, s_restore_m0, S_SAVE_PC_HI_RCNT_SHIFT
 	s_lshl_b32	s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_RCNT_SHIFT
-	s_mov_b32	s_restore_mode, 0x0
-	s_or_b32	s_restore_mode, s_restore_mode, s_restore_m0
-	s_and_b32	s_restore_m0, s_restore_tmp, S_SAVE_PC_HI_FIRST_REPLAY_MASK
+	s_mov_b32	s_restore_tmp, 0x0
+	s_or_b32	s_restore_tmp, s_restore_tmp, s_restore_m0
+	s_and_b32	s_restore_m0, s_restore_pc_hi, S_SAVE_PC_HI_FIRST_REPLAY_MASK
 	s_lshr_b32	s_restore_m0, s_restore_m0, S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
 	s_lshl_b32	s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT
-	s_or_b32	s_restore_mode, s_restore_mode, s_restore_m0
-	s_and_b32	s_restore_m0, s_restore_tmp, S_SAVE_PC_HI_REPLAY_W64H_MASK
+	s_or_b32	s_restore_tmp, s_restore_tmp, s_restore_m0
+	s_and_b32	s_restore_m0, s_restore_pc_hi, S_SAVE_PC_HI_REPLAY_W64H_MASK
 	s_lshr_b32	s_restore_m0, s_restore_m0, S_SAVE_PC_HI_REPLAY_W64H_SHIFT
 	s_lshl_b32	s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT
-	s_or_b32	s_restore_mode, s_restore_mode, s_restore_m0
+	s_or_b32	s_restore_tmp, s_restore_tmp, s_restore_m0
 
 	s_and_b32	s_restore_m0, s_restore_status, SQ_WAVE_STATUS_INST_ATC_MASK
 	s_lshr_b32	s_restore_m0, s_restore_m0, SQ_WAVE_STATUS_INST_ATC_SHIFT
-	s_setreg_b32 	hwreg(HW_REG_IB_STS), s_restore_mode
+	s_setreg_b32 	hwreg(HW_REG_IB_STS), s_restore_tmp
 #endif
 
+	s_and_b32	s_restore_pc_hi, s_restore_pc_hi, 0x0000ffff		//pc[47:32] //Do it here in order not to affect STATUS
 	s_and_b64	exec, exec, exec					// Restore STATUS.EXECZ, not writable by s_setreg_b32
 	s_and_b64	vcc, vcc, vcc						// Restore STATUS.VCCZ, not writable by s_setreg_b32
 	s_setreg_b32	hwreg(HW_REG_STATUS), s_restore_status			// SCC is included, which is changed by previous salu
@@ -1022,9 +1061,7 @@ end
 function get_vgpr_size_bytes(s_vgpr_size_byte, s_size)
 	s_getreg_b32	s_vgpr_size_byte, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE)
 	s_add_u32	s_vgpr_size_byte, s_vgpr_size_byte, 1
-	s_lshr_b32	m0, s_size, S_WAVE_SIZE
-	s_and_b32	m0, m0, 1
-	s_cmp_eq_u32	m0, 1
+	s_bitcmp1_b32	s_size, S_WAVE_SIZE
 	s_cbranch_scc1	L_ENABLE_SHIFT_W64
 	s_lshl_b32	s_vgpr_size_byte, s_vgpr_size_byte, (2+7)		//Number of VGPRs = (vgpr_size + 1) * 4 * 32 * 4   (non-zero value)
 	s_branch	L_SHIFT_DONE
@@ -1045,3 +1082,9 @@ end
 function get_hwreg_size_bytes
 	return 128
 end
+
+function get_wave_size(s_reg)
+	s_getreg_b32	s_reg, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE)
+	s_lshl_b32	s_reg, s_reg, S_WAVE_SIZE
+	s_or_b32	s_reg, s_save_spi_init_hi, s_reg			//share with exec_hi, it's at bit25
+end
-- 
2.25.4



More information about the amd-gfx mailing list