drm/amdkfd: Save/restore flat_scratch_lo/hi on gfx10

These moved from SGPRs in gfx9 to HWREG in gfx10.

Cc: Shaoyun Liu <shaoyun.liu@amd.com>
Signed-off-by: Jay Cornwall <jay.cornwall@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: shaoyunl <shaoyun.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
index c10e424dd1f5878758cd0390ee6f904194235df6..8089bb37f3930c2d64dab33edc6b4792fa78cb3b 100644 (file)
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
@@ -680,7 +680,7 @@ static const uint32_t cwsr_trap_gfx9_hex[] = {
 };
 
 static const uint32_t cwsr_trap_gfx10_hex[] = {
-       0xbf820001, 0xbf8201b2,
+       0xbf820001, 0xbf8201c0,
        0xb0804004, 0xb978f802,
        0x8a788678, 0xb971f803,
        0x876eff71, 0x00000400,
@@ -772,6 +772,13 @@ static const uint32_t cwsr_trap_gfx10_hex[] = {
        0xb97bf801, 0xbefe037c,
        0xbefc037a, 0xf4611efa,
        0xf8000000, 0x807a847a,
+       0xbefc037e, 0xb97bf814,
+       0xbefe037c, 0xbefc037a,
+       0xf4611efa, 0xf8000000,
+       0x807a847a, 0xbefc037e,
+       0xb97bf815, 0xbefe037c,
+       0xbefc037a, 0xf4611efa,
+       0xf8000000, 0x807a847a,
        0xbefc037e, 0x8776ff7f,
        0x04000000, 0xbeef0380,
        0x886f6f76, 0xb97a2a05,
@@ -897,7 +904,7 @@ static const uint32_t cwsr_trap_gfx10_hex[] = {
        0xe0704000, 0x7a5d0000,
        0x807c817c, 0x807aff7a,
        0x00000080, 0xbf0a717c,
-       0xbf85fff8, 0xbf820138,
+       0xbf85fff8, 0xbf820141,
        0xbef4037e, 0x8775ff7f,
        0x0000ffff, 0x8875ff75,
        0x00040000, 0xbef60380,
@@ -1033,30 +1040,35 @@ static const uint32_t cwsr_trap_gfx10_hex[] = {
        0x80788478, 0xf4211e7a,
        0xf0000000, 0x80788478,
        0xf4211cfa, 0xf0000000,
+       0x80788478, 0xf4211bba,
+       0xf0000000, 0x80788478,
+       0xbf8cc07f, 0xb9eef814,
+       0xf4211bba, 0xf0000000,
        0x80788478, 0xbf8cc07f,
-       0xbef2036d, 0x876dff72,
-       0x0000ffff, 0xbefc036f,
-       0xbefe037a, 0xbeff037b,
-       0x876f71ff, 0x000003ff,
-       0xb9ef4803, 0xb9f9f816,
-       0x876f71ff, 0xfffff800,
-       0x906f8b6f, 0xb9efa2c3,
-       0xb9f3f801, 0x876fff72,
-       0xfc000000, 0x906f9a6f,
-       0x8f6f906f, 0xbef30380,
+       0xb9eef815, 0xbef2036d,
+       0x876dff72, 0x0000ffff,
+       0xbefc036f, 0xbefe037a,
+       0xbeff037b, 0x876f71ff,
+       0x000003ff, 0xb9ef4803,
+       0xb9f9f816, 0x876f71ff,
+       0xfffff800, 0x906f8b6f,
+       0xb9efa2c3, 0xb9f3f801,
+       0x876fff72, 0xfc000000,
+       0x906f9a6f, 0x8f6f906f,
+       0xbef30380, 0x88736f73,
+       0x876fff72, 0x02000000,
+       0x906f996f, 0x8f6f8f6f,
        0x88736f73, 0x876fff72,
-       0x02000000, 0x906f996f,
-       0x8f6f8f6f, 0x88736f73,
-       0x876fff72, 0x01000000,
-       0x906f986f, 0x8f6f996f,
-       0x88736f73, 0x876fff70,
-       0x00800000, 0x906f976f,
-       0xb9f3f807, 0x87fe7e7e,
-       0x87ea6a6a, 0xb9f0f802,
-       0xbf8a0000, 0xbe80226c,
-       0xbf810000, 0xbf9f0000,
+       0x01000000, 0x906f986f,
+       0x8f6f996f, 0x88736f73,
+       0x876fff70, 0x00800000,
+       0x906f976f, 0xb9f3f807,
+       0x87fe7e7e, 0x87ea6a6a,
+       0xb9f0f802, 0xbf8a0000,
+       0xbe80226c, 0xbf810000,
        0xbf9f0000, 0xbf9f0000,
        0xbf9f0000, 0xbf9f0000,
+       0xbf9f0000, 0x00000000,
 };
 static const uint32_t cwsr_trap_arcturus_hex[] = {
        0xbf820001, 0xbf8202c4,

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 be6f7d1847fa553c6bb249f1a82c9b2a2f6f3a2b..fafdfd2ac610835ad08e67d2ffd354527f3c54de 100644 (file)
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm
@@ -132,6 +132,7 @@ var s_restore_tmp                           = ttmp6
 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_pc_lo                            = ttmp0
 var s_restore_pc_hi                            = ttmp1
 var s_restore_exec_lo                          = ttmp14
@@ -313,6 +314,12 @@ L_SAVE_HWREG:
        s_getreg_b32    s_save_m0, hwreg(HW_REG_MODE)
        write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)
 
+       s_getreg_b32    s_save_m0, hwreg(HW_REG_SHADER_FLAT_SCRATCH_LO)
+       write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)
+
+       s_getreg_b32    s_save_m0, hwreg(HW_REG_SHADER_FLAT_SCRATCH_HI)
+       write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)
+
        /* the first wave in the threadgroup */
        s_and_b32       s_save_tmp, s_save_spi_init_hi, S_SAVE_SPI_INIT_FIRST_WAVE_MASK
        s_mov_b32       s_save_exec_hi, 0x0
@@ -824,9 +831,16 @@ L_RESTORE_HWREG:
        read_hwreg_from_mem(s_restore_trapsts, s_restore_buf_rsrc0, s_restore_mem_offset)
        read_hwreg_from_mem(s_restore_xnack_mask, s_restore_buf_rsrc0, s_restore_mem_offset)
        read_hwreg_from_mem(s_restore_mode, s_restore_buf_rsrc0, s_restore_mem_offset)
+       read_hwreg_from_mem(s_restore_flat_scratch, s_restore_buf_rsrc0, s_restore_mem_offset)
+       s_waitcnt       lgkmcnt(0)
+
+       s_setreg_b32    hwreg(HW_REG_SHADER_FLAT_SCRATCH_LO), s_restore_flat_scratch
 
+       read_hwreg_from_mem(s_restore_flat_scratch, s_restore_buf_rsrc0, s_restore_mem_offset)
        s_waitcnt       lgkmcnt(0)                                              //from now on, it is safe to restore STATUS and IB_STS
 
+       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