summaryrefslogtreecommitdiffstats
path: root/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm')
-rw-r--r--drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm274
1 files changed, 13 insertions, 261 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm
index 997a383dcb8b7..a2a04bb64096f 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm
@@ -20,9 +20,12 @@
* OTHER DEALINGS IN THE SOFTWARE.
*/
-#if 0
-HW (VI) source code for CWSR trap handler
-#Version 18 + multiple trap handler
+/* To compile this assembly code:
+ * PROJECT=vi ./sp3 cwsr_trap_handler_gfx8.asm -hex tmp.hex
+ */
+
+/* HW (VI) source code for CWSR trap handler */
+/* Version 18 + multiple trap handler */
// this performance-optimal version was originally from Seven Xu at SRDC
@@ -98,6 +101,7 @@ var SWIZZLE_EN = 0 //whether we use swi
/**************************************************************************/
var SQ_WAVE_STATUS_INST_ATC_SHIFT = 23
var SQ_WAVE_STATUS_INST_ATC_MASK = 0x00800000
+var SQ_WAVE_STATUS_SPI_PRIO_SHIFT = 1
var SQ_WAVE_STATUS_SPI_PRIO_MASK = 0x00000006
var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT = 12
@@ -149,7 +153,7 @@ var s_save_spi_init_lo = exec_lo
var s_save_spi_init_hi = exec_hi
//tba_lo and tba_hi need to be saved/restored
-var s_save_pc_lo = ttmp0 //{TTMP1, TTMP0} = {3??h0,pc_rewind[3:0], HT[0],trapID[7:0], PC[47:0]}
+var s_save_pc_lo = ttmp0 //{TTMP1, TTMP0} = {3'h0,pc_rewind[3:0], HT[0],trapID[7:0], PC[47:0]}
var s_save_pc_hi = ttmp1
var s_save_exec_lo = ttmp2
var s_save_exec_hi = ttmp3
@@ -319,6 +323,10 @@ end
s_sendmsg sendmsg(MSG_SAVEWAVE) //send SPI a message and wait for SPI's write to EXEC
end
+ // Set SPI_PRIO=2 to avoid starving instruction fetch in the waves we're waiting for.
+ s_or_b32 s_save_tmp, s_save_status, (2 << SQ_WAVE_STATUS_SPI_PRIO_SHIFT)
+ s_setreg_b32 hwreg(HW_REG_STATUS), s_save_tmp
+
L_SLEEP:
s_sleep 0x2 // sleep 1 (64clk) is not enough for 8 waves per SIMD, which will cause SQ hang, since the 7,8th wave could not get arbit to exec inst, while other waves are stuck into the sleep-loop and waiting for wrexec!=0
@@ -1007,8 +1015,6 @@ end
s_waitcnt lgkmcnt(0) //from now on, it is safe to restore STATUS and IB_STS
- s_and_b32 s_restore_pc_hi, s_restore_pc_hi, 0x0000ffff //pc[47:32] //Do it here in order not to affect STATUS
-
//for normal save & restore, the saved PC points to the next inst to execute, no adjustment needs to be made, otherwise:
if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))
s_add_u32 s_restore_pc_lo, s_restore_pc_lo, 8 //pc[31:0]+8 //two back-to-back s_trap are used (first for save and second for restore)
@@ -1044,6 +1050,7 @@ end
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_tmp
+ 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
@@ -1127,258 +1134,3 @@ end
function get_hwreg_size_bytes
return 128 //HWREG size 128 bytes
end
-
-
-#endif
-
-static const uint32_t cwsr_trap_gfx8_hex[] = {
- 0xbf820001, 0xbf820123,
- 0xb8f4f802, 0x89748674,
- 0xb8f5f803, 0x8675ff75,
- 0x00000400, 0xbf850011,
- 0xc00a1e37, 0x00000000,
- 0xbf8c007f, 0x87777978,
- 0xbf840002, 0xb974f802,
- 0xbe801d78, 0xb8f5f803,
- 0x8675ff75, 0x000001ff,
- 0xbf850002, 0x80708470,
- 0x82718071, 0x8671ff71,
- 0x0000ffff, 0xb974f802,
- 0xbe801f70, 0xb8f5f803,
- 0x8675ff75, 0x00000100,
- 0xbf840006, 0xbefa0080,
- 0xb97a0203, 0x8671ff71,
- 0x0000ffff, 0x80f08870,
- 0x82f18071, 0xbefa0080,
- 0xb97a0283, 0xbef60068,
- 0xbef70069, 0xb8fa1c07,
- 0x8e7a9c7a, 0x87717a71,
- 0xb8fa03c7, 0x8e7a9b7a,
- 0x87717a71, 0xb8faf807,
- 0x867aff7a, 0x00007fff,
- 0xb97af807, 0xbef2007e,
- 0xbef3007f, 0xbefe0180,
- 0xbf900004, 0xbf8e0002,
- 0xbf88fffe, 0xbef8007e,
- 0x8679ff7f, 0x0000ffff,
- 0x8779ff79, 0x00040000,
- 0xbefa0080, 0xbefb00ff,
- 0x00807fac, 0x867aff7f,
- 0x08000000, 0x8f7a837a,
- 0x877b7a7b, 0x867aff7f,
- 0x70000000, 0x8f7a817a,
- 0x877b7a7b, 0xbeef007c,
- 0xbeee0080, 0xb8ee2a05,
- 0x806e816e, 0x8e6e8a6e,
- 0xb8fa1605, 0x807a817a,
- 0x8e7a867a, 0x806e7a6e,
- 0xbefa0084, 0xbefa00ff,
- 0x01000000, 0xbefe007c,
- 0xbefc006e, 0xc0611bfc,
- 0x0000007c, 0x806e846e,
- 0xbefc007e, 0xbefe007c,
- 0xbefc006e, 0xc0611c3c,
- 0x0000007c, 0x806e846e,
- 0xbefc007e, 0xbefe007c,
- 0xbefc006e, 0xc0611c7c,
- 0x0000007c, 0x806e846e,
- 0xbefc007e, 0xbefe007c,
- 0xbefc006e, 0xc0611cbc,
- 0x0000007c, 0x806e846e,
- 0xbefc007e, 0xbefe007c,
- 0xbefc006e, 0xc0611cfc,
- 0x0000007c, 0x806e846e,
- 0xbefc007e, 0xbefe007c,
- 0xbefc006e, 0xc0611d3c,
- 0x0000007c, 0x806e846e,
- 0xbefc007e, 0xb8f5f803,
- 0xbefe007c, 0xbefc006e,
- 0xc0611d7c, 0x0000007c,
- 0x806e846e, 0xbefc007e,
- 0xbefe007c, 0xbefc006e,
- 0xc0611dbc, 0x0000007c,
- 0x806e846e, 0xbefc007e,
- 0xbefe007c, 0xbefc006e,
- 0xc0611dfc, 0x0000007c,
- 0x806e846e, 0xbefc007e,
- 0xb8eff801, 0xbefe007c,
- 0xbefc006e, 0xc0611bfc,
- 0x0000007c, 0x806e846e,
- 0xbefc007e, 0xbefe007c,
- 0xbefc006e, 0xc0611b3c,
- 0x0000007c, 0x806e846e,
- 0xbefc007e, 0xbefe007c,
- 0xbefc006e, 0xc0611b7c,
- 0x0000007c, 0x806e846e,
- 0xbefc007e, 0x867aff7f,
- 0x04000000, 0xbef30080,
- 0x8773737a, 0xb8ee2a05,
- 0x806e816e, 0x8e6e8a6e,
- 0xb8f51605, 0x80758175,
- 0x8e758475, 0x8e7a8275,
- 0xbefa00ff, 0x01000000,
- 0xbef60178, 0x80786e78,
- 0x82798079, 0xbefc0080,
- 0xbe802b00, 0xbe822b02,
- 0xbe842b04, 0xbe862b06,
- 0xbe882b08, 0xbe8a2b0a,
- 0xbe8c2b0c, 0xbe8e2b0e,
- 0xc06b003c, 0x00000000,
- 0xc06b013c, 0x00000010,
- 0xc06b023c, 0x00000020,
- 0xc06b033c, 0x00000030,
- 0x8078c078, 0x82798079,
- 0x807c907c, 0xbf0a757c,
- 0xbf85ffeb, 0xbef80176,
- 0xbeee0080, 0xbefe00c1,
- 0xbeff00c1, 0xbefa00ff,
- 0x01000000, 0xe0724000,
- 0x6e1e0000, 0xe0724100,
- 0x6e1e0100, 0xe0724200,
- 0x6e1e0200, 0xe0724300,
- 0x6e1e0300, 0xbefe00c1,
- 0xbeff00c1, 0xb8f54306,
- 0x8675c175, 0xbf84002c,
- 0xbf8a0000, 0x867aff73,
- 0x04000000, 0xbf840028,
- 0x8e758675, 0x8e758275,
- 0xbefa0075, 0xb8ee2a05,
- 0x806e816e, 0x8e6e8a6e,
- 0xb8fa1605, 0x807a817a,
- 0x8e7a867a, 0x806e7a6e,
- 0x806eff6e, 0x00000080,
- 0xbefa00ff, 0x01000000,
- 0xbefc0080, 0xd28c0002,
- 0x000100c1, 0xd28d0003,
- 0x000204c1, 0xd1060002,
- 0x00011103, 0x7e0602ff,
- 0x00000200, 0xbefc00ff,
- 0x00010000, 0xbe80007b,
- 0x867bff7b, 0xff7fffff,
- 0x877bff7b, 0x00058000,
- 0xd8ec0000, 0x00000002,
- 0xbf8c007f, 0xe0765000,
- 0x6e1e0002, 0x32040702,
- 0xd0c9006a, 0x0000eb02,
- 0xbf87fff7, 0xbefb0000,
- 0xbeee00ff, 0x00000400,
- 0xbefe00c1, 0xbeff00c1,
- 0xb8f52a05, 0x80758175,
- 0x8e758275, 0x8e7a8875,
- 0xbefa00ff, 0x01000000,
- 0xbefc0084, 0xbf0a757c,
- 0xbf840015, 0xbf11017c,
- 0x8075ff75, 0x00001000,
- 0x7e000300, 0x7e020301,
- 0x7e040302, 0x7e060303,
- 0xe0724000, 0x6e1e0000,
- 0xe0724100, 0x6e1e0100,
- 0xe0724200, 0x6e1e0200,
- 0xe0724300, 0x6e1e0300,
- 0x807c847c, 0x806eff6e,
- 0x00000400, 0xbf0a757c,
- 0xbf85ffef, 0xbf9c0000,
- 0xbf8200ca, 0xbef8007e,
- 0x8679ff7f, 0x0000ffff,
- 0x8779ff79, 0x00040000,
- 0xbefa0080, 0xbefb00ff,
- 0x00807fac, 0x8676ff7f,
- 0x08000000, 0x8f768376,
- 0x877b767b, 0x8676ff7f,
- 0x70000000, 0x8f768176,
- 0x877b767b, 0x8676ff7f,
- 0x04000000, 0xbf84001e,
- 0xbefe00c1, 0xbeff00c1,
- 0xb8f34306, 0x8673c173,
- 0xbf840019, 0x8e738673,
- 0x8e738273, 0xbefa0073,
- 0xb8f22a05, 0x80728172,
- 0x8e728a72, 0xb8f61605,
- 0x80768176, 0x8e768676,
- 0x80727672, 0x8072ff72,
- 0x00000080, 0xbefa00ff,
- 0x01000000, 0xbefc0080,
- 0xe0510000, 0x721e0000,
- 0xe0510100, 0x721e0000,
- 0x807cff7c, 0x00000200,
- 0x8072ff72, 0x00000200,
- 0xbf0a737c, 0xbf85fff6,
- 0xbef20080, 0xbefe00c1,
- 0xbeff00c1, 0xb8f32a05,
- 0x80738173, 0x8e738273,
- 0x8e7a8873, 0xbefa00ff,
- 0x01000000, 0xbef60072,
- 0x8072ff72, 0x00000400,
- 0xbefc0084, 0xbf11087c,
- 0x8073ff73, 0x00008000,
- 0xe0524000, 0x721e0000,
- 0xe0524100, 0x721e0100,
- 0xe0524200, 0x721e0200,
- 0xe0524300, 0x721e0300,
- 0xbf8c0f70, 0x7e000300,
- 0x7e020301, 0x7e040302,
- 0x7e060303, 0x807c847c,
- 0x8072ff72, 0x00000400,
- 0xbf0a737c, 0xbf85ffee,
- 0xbf9c0000, 0xe0524000,
- 0x761e0000, 0xe0524100,
- 0x761e0100, 0xe0524200,
- 0x761e0200, 0xe0524300,
- 0x761e0300, 0xb8f22a05,
- 0x80728172, 0x8e728a72,
- 0xb8f61605, 0x80768176,
- 0x8e768676, 0x80727672,
- 0x80f2c072, 0xb8f31605,
- 0x80738173, 0x8e738473,
- 0x8e7a8273, 0xbefa00ff,
- 0x01000000, 0xbefc0073,
- 0xc031003c, 0x00000072,
- 0x80f2c072, 0xbf8c007f,
- 0x80fc907c, 0xbe802d00,
- 0xbe822d02, 0xbe842d04,
- 0xbe862d06, 0xbe882d08,
- 0xbe8a2d0a, 0xbe8c2d0c,
- 0xbe8e2d0e, 0xbf06807c,
- 0xbf84fff1, 0xb8f22a05,
- 0x80728172, 0x8e728a72,
- 0xb8f61605, 0x80768176,
- 0x8e768676, 0x80727672,
- 0xbefa0084, 0xbefa00ff,
- 0x01000000, 0xc0211cfc,
- 0x00000072, 0x80728472,
- 0xc0211c3c, 0x00000072,
- 0x80728472, 0xc0211c7c,
- 0x00000072, 0x80728472,
- 0xc0211bbc, 0x00000072,
- 0x80728472, 0xc0211bfc,
- 0x00000072, 0x80728472,
- 0xc0211d3c, 0x00000072,
- 0x80728472, 0xc0211d7c,
- 0x00000072, 0x80728472,
- 0xc0211a3c, 0x00000072,
- 0x80728472, 0xc0211a7c,
- 0x00000072, 0x80728472,
- 0xc0211dfc, 0x00000072,
- 0x80728472, 0xc0211b3c,
- 0x00000072, 0x80728472,
- 0xc0211b7c, 0x00000072,
- 0x80728472, 0xbf8c007f,
- 0x8671ff71, 0x0000ffff,
- 0xbefc0073, 0xbefe006e,
- 0xbeff006f, 0x867375ff,
- 0x000003ff, 0xb9734803,
- 0x867375ff, 0xfffff800,
- 0x8f738b73, 0xb973a2c3,
- 0xb977f801, 0x8673ff71,
- 0xf0000000, 0x8f739c73,
- 0x8e739073, 0xbef60080,
- 0x87767376, 0x8673ff71,
- 0x08000000, 0x8f739b73,
- 0x8e738f73, 0x87767376,
- 0x8673ff74, 0x00800000,
- 0x8f739773, 0xb976f807,
- 0x86fe7e7e, 0x86ea6a6a,
- 0xb974f802, 0xbf8a0000,
- 0x95807370, 0xbf810000,
-};
-
generated by cgit v1.2.3 (git 2.39.1) at 2024-04-19 16:35:33 +0000