<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=us-ascii">
<style type="text/css" style="display:none;"> P {margin-top:0;margin-bottom:0;} </style>
</head>
<body dir="ltr">
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
Hi Felix,</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
I reworded this patch in the next series. Please review the first two patches in this series.</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
Regards,</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
<span style="color: rgb(0, 0, 0); font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt;">Yong </span></div>
<div id="appendonsend"></div>
<hr style="display:inline-block;width:98%" tabindex="-1">
<div id="divRplyFwdMsg" dir="ltr"><font face="Calibri, sans-serif" style="font-size:11pt" color="#000000"><b>From:</b> amd-gfx <amd-gfx-bounces@lists.freedesktop.org> on behalf of Zhao, Yong <Yong.Zhao@amd.com><br>
<b>Sent:</b> Wednesday, September 25, 2019 2:34 PM<br>
<b>To:</b> Kuehling, Felix <Felix.Kuehling@amd.com>; amd-gfx@lists.freedesktop.org <amd-gfx@lists.freedesktop.org><br>
<b>Subject:</b> Re: [PATCH 3/3] drm/amdkfd: Remove the control stack workaround for GFX10</font>
<div> </div>
</div>
<div class="BodyFragment"><font size="2"><span style="font-size:11pt;">
<div class="PlainText">Yes. I confirmed with CP guys and they said the behavior on GFX10 is the
<br>
same as GFX8 now. I remember that the workaround on GFX9 was to help <br>
with a HW bug, but not too sure.<br>
<br>
Regards,<br>
<br>
Yong<br>
<br>
On 2019-09-25 2:25 p.m., Kuehling, Felix wrote:<br>
> On 2019-09-25 2:15 p.m., Zhao, Yong wrote:<br>
>> The GFX10 does not have this hardware bug any more, so remove it.<br>
> I wouldn't call this a bug and a workaround. More like a change in the<br>
> HW or FW behaviour and a corresponding driver change. I.e. in GFXv8 the<br>
> control stack was in the user mode CWSR allocation. In GFXv9 it moved<br>
> into a kernel mode buffer next to the MQD. So in GFXv10 the control<br>
> stack moved back into the user mode CWSR buffer?<br>
><br>
> Regards,<br>
> Felix<br>
><br>
>> Change-Id: I446c9685549a09ac8846a42ee22d86cfb93fd98c<br>
>> Signed-off-by: Yong Zhao <Yong.Zhao@amd.com><br>
>> ---<br>
>> .../gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c | 37 ++-----------------<br>
>> 1 file changed, 4 insertions(+), 33 deletions(-)<br>
>><br>
>> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c<br>
>> index 9cd3eb2d90bd..4a236b2c2354 100644<br>
>> --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c<br>
>> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c<br>
>> @@ -69,35 +69,13 @@ static void update_cu_mask(struct mqd_manager *mm, void *mqd,<br>
>> static struct kfd_mem_obj *allocate_mqd(struct kfd_dev *kfd,<br>
>> struct queue_properties *q)<br>
>> {<br>
>> - int retval;<br>
>> - struct kfd_mem_obj *mqd_mem_obj = NULL;<br>
>> + struct kfd_mem_obj *mqd_mem_obj;<br>
>> <br>
>> - /* From V9, for CWSR, the control stack is located on the next page<br>
>> - * boundary after the mqd, we will use the gtt allocation function<br>
>> - * instead of sub-allocation function.<br>
>> - */<br>
>> - if (kfd->cwsr_enabled && (q->type == KFD_QUEUE_TYPE_COMPUTE)) {<br>
>> - mqd_mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_NOIO);<br>
>> - if (!mqd_mem_obj)<br>
>> - return NULL;<br>
>> - retval = amdgpu_amdkfd_alloc_gtt_mem(kfd->kgd,<br>
>> - ALIGN(q->ctl_stack_size, PAGE_SIZE) +<br>
>> - ALIGN(sizeof(struct v10_compute_mqd), PAGE_SIZE),<br>
>> - &(mqd_mem_obj->gtt_mem),<br>
>> - &(mqd_mem_obj->gpu_addr),<br>
>> - (void *)&(mqd_mem_obj->cpu_ptr), true);<br>
>> - } else {<br>
>> - retval = kfd_gtt_sa_allocate(kfd, sizeof(struct v10_compute_mqd),<br>
>> - &mqd_mem_obj);<br>
>> - }<br>
>> -<br>
>> - if (retval) {<br>
>> - kfree(mqd_mem_obj);<br>
>> + if (kfd_gtt_sa_allocate(kfd, sizeof(struct v10_compute_mqd),<br>
>> + &mqd_mem_obj))<br>
>> return NULL;<br>
>> - }<br>
>> <br>
>> return mqd_mem_obj;<br>
>> -<br>
>> }<br>
>> <br>
>> static void init_mqd(struct mqd_manager *mm, void **mqd,<br>
>> @@ -250,14 +228,7 @@ static int destroy_mqd(struct mqd_manager *mm, void *mqd,<br>
>> static void free_mqd(struct mqd_manager *mm, void *mqd,<br>
>> struct kfd_mem_obj *mqd_mem_obj)<br>
>> {<br>
>> - struct kfd_dev *kfd = mm->dev;<br>
>> -<br>
>> - if (mqd_mem_obj->gtt_mem) {<br>
>> - amdgpu_amdkfd_free_gtt_mem(kfd->kgd, mqd_mem_obj->gtt_mem);<br>
>> - kfree(mqd_mem_obj);<br>
>> - } else {<br>
>> - kfd_gtt_sa_free(mm->dev, mqd_mem_obj);<br>
>> - }<br>
>> + kfd_gtt_sa_free(mm->dev, mqd_mem_obj);<br>
>> }<br>
>> <br>
>> static bool is_occupied(struct mqd_manager *mm, void *mqd,<br>
_______________________________________________<br>
amd-gfx mailing list<br>
amd-gfx@lists.freedesktop.org<br>
<a href="https://lists.freedesktop.org/mailman/listinfo/amd-gfx">https://lists.freedesktop.org/mailman/listinfo/amd-gfx</a></div>
</span></font></div>
</body>
</html>