Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
drm/amdkfd: update the cwsr area size for gfx950
Update cwsr area size for gfx950 to fit the new user queue buffer validation.
The size of LDS calculation is referred from gfx950 thunk implementation.
Signed-off-by: Le Ma <le.ma@amd.com>
Acked-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
authored by
Le Ma
and committed by
Alex Deucher
5a7c8c57
5690011a
+6
-4
+6
-4
drivers/gpu/drm/amd/amdkfd/kfd_queue.c
+6
-4
drivers/gpu/drm/amd/amdkfd/kfd_queue.c
···
394
394
395
395
if ((gfxv / 100 * 100) == 90400 || /* GFX_VERSION_AQUA_VANJARAM */
396
396
gfxv == 90010 || /* GFX_VERSION_ALDEBARAN */
397
-
gfxv == 90008) /* GFX_VERSION_ARCTURUS */
397
+
gfxv == 90008 || /* GFX_VERSION_ARCTURUS */
398
+
gfxv == 90500)
398
399
vgpr_size = 0x80000;
399
400
else if (gfxv == 110000 || /* GFX_VERSION_PLUM_BONITO */
400
401
gfxv == 110001 || /* GFX_VERSION_WHEAT_NAS */
···
406
405
return vgpr_size;
407
406
}
408
407
409
-
#define WG_CONTEXT_DATA_SIZE_PER_CU(gfxv) \
408
+
#define WG_CONTEXT_DATA_SIZE_PER_CU(gfxv, props) \
410
409
(kfd_get_vgpr_size_per_cu(gfxv) + SGPR_SIZE_PER_CU +\
411
-
LDS_SIZE_PER_CU + HWREG_SIZE_PER_CU)
410
+
(((gfxv) == 90500) ? (props->lds_size_in_kb << 10) : LDS_SIZE_PER_CU) +\
411
+
HWREG_SIZE_PER_CU)
412
412
413
413
#define CNTL_STACK_BYTES_PER_WAVE(gfxv) \
414
414
((gfxv) >= 100100 ? 12 : 8) /* GFX_VERSION_NAVI10*/
···
433
431
min(cu_num * 40, props->array_count / props->simd_arrays_per_engine * 512)
434
432
: cu_num * 32;
435
433
436
-
wg_data_size = ALIGN(cu_num * WG_CONTEXT_DATA_SIZE_PER_CU(gfxv), PAGE_SIZE);
434
+
wg_data_size = ALIGN(cu_num * WG_CONTEXT_DATA_SIZE_PER_CU(gfxv, props), PAGE_SIZE);
437
435
ctl_stack_size = wave_num * CNTL_STACK_BYTES_PER_WAVE(gfxv) + 8;
438
436
ctl_stack_size = ALIGN(SIZEOF_HSA_USER_CONTEXT_SAVE_AREA_HEADER + ctl_stack_size,
439
437
PAGE_SIZE);