|
Hi Felix,
I reworded this patch in the next series. Please review the first two patches in this series.
Regards,
Yong
From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> on behalf of Zhao, Yong <Yong.Zhao@xxxxxxx>
Sent: Wednesday, September 25, 2019 2:34 PM To: Kuehling, Felix <Felix.Kuehling@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx <amd-gfx@xxxxxxxxxxxxxxxxxxxxx> Subject: Re: [PATCH 3/3] drm/amdkfd: Remove the control stack workaround for GFX10 Yes. I confirmed with CP guys and they said the behavior on GFX10 is the
same as GFX8 now. I remember that the workaround on GFX9 was to help with a HW bug, but not too sure. Regards, Yong On 2019-09-25 2:25 p.m., Kuehling, Felix wrote: > On 2019-09-25 2:15 p.m., Zhao, Yong wrote: >> The GFX10 does not have this hardware bug any more, so remove it. > I wouldn't call this a bug and a workaround. More like a change in the > HW or FW behaviour and a corresponding driver change. I.e. in GFXv8 the > control stack was in the user mode CWSR allocation. In GFXv9 it moved > into a kernel mode buffer next to the MQD. So in GFXv10 the control > stack moved back into the user mode CWSR buffer? > > Regards, > Felix > >> Change-Id: I446c9685549a09ac8846a42ee22d86cfb93fd98c >> Signed-off-by: Yong Zhao <Yong.Zhao@xxxxxxx> >> --- >> .../gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c | 37 ++----------------- >> 1 file changed, 4 insertions(+), 33 deletions(-) >> >> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c >> index 9cd3eb2d90bd..4a236b2c2354 100644 >> --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c >> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c >> @@ -69,35 +69,13 @@ static void update_cu_mask(struct mqd_manager *mm, void *mqd, >> static struct kfd_mem_obj *allocate_mqd(struct kfd_dev *kfd, >> struct queue_properties *q) >> { >> - int retval; >> - struct kfd_mem_obj *mqd_mem_obj = NULL; >> + struct kfd_mem_obj *mqd_mem_obj; >> >> - /* From V9, for CWSR, the control stack is located on the next page >> - * boundary after the mqd, we will use the gtt allocation function >> - * instead of sub-allocation function. >> - */ >> - if (kfd->cwsr_enabled && (q->type == KFD_QUEUE_TYPE_COMPUTE)) { >> - mqd_mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_NOIO); >> - if (!mqd_mem_obj) >> - return NULL; >> - retval = amdgpu_amdkfd_alloc_gtt_mem(kfd->kgd, >> - ALIGN(q->ctl_stack_size, PAGE_SIZE) + >> - ALIGN(sizeof(struct v10_compute_mqd), PAGE_SIZE), >> - &(mqd_mem_obj->gtt_mem), >> - &(mqd_mem_obj->gpu_addr), >> - (void *)&(mqd_mem_obj->cpu_ptr), true); >> - } else { >> - retval = kfd_gtt_sa_allocate(kfd, sizeof(struct v10_compute_mqd), >> - &mqd_mem_obj); >> - } >> - >> - if (retval) { >> - kfree(mqd_mem_obj); >> + if (kfd_gtt_sa_allocate(kfd, sizeof(struct v10_compute_mqd), >> + &mqd_mem_obj)) >> return NULL; >> - } >> >> return mqd_mem_obj; >> - >> } >> >> static void init_mqd(struct mqd_manager *mm, void **mqd, >> @@ -250,14 +228,7 @@ static int destroy_mqd(struct mqd_manager *mm, void *mqd, >> static void free_mqd(struct mqd_manager *mm, void *mqd, >> struct kfd_mem_obj *mqd_mem_obj) >> { >> - struct kfd_dev *kfd = mm->dev; >> - >> - if (mqd_mem_obj->gtt_mem) { >> - amdgpu_amdkfd_free_gtt_mem(kfd->kgd, mqd_mem_obj->gtt_mem); >> - kfree(mqd_mem_obj); >> - } else { >> - kfd_gtt_sa_free(mm->dev, mqd_mem_obj); >> - } >> + kfd_gtt_sa_free(mm->dev, mqd_mem_obj); >> } >> >> static bool is_occupied(struct mqd_manager *mm, void *mqd, _______________________________________________ amd-gfx mailing list amd-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/amd-gfx |
_______________________________________________ amd-gfx mailing list amd-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/amd-gfx
