[PATCH 03/11] drm/amdgpu/gfx: add generic handling for disable_kq

Khatri, Sunil sukhatri at amd.com
Thu Mar 6 09:57:33 UTC 2025


On 3/6/2025 6:36 AM, Felix Kuehling wrote:
>
> On 2025-03-05 15:47, Alex Deucher wrote:
>> Add proper checks for disable_kq functionality in
>> gfx helper functions.  Add special logic for families
>> that require the clear state setup.
>>
>> Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 92 +++++++++++++++++--------
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h |  2 +
>>   2 files changed, 67 insertions(+), 27 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
>> index a194bf3347cbc..af3f8b62f6fd5 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
>> @@ -371,6 +371,18 @@ int amdgpu_gfx_kiq_init(struct amdgpu_device *adev,
>>       return 0;
>>   }
>>   +static bool amdgpu_gfx_disable_gfx_kq(struct amdgpu_device *adev)
>> +{
>> +    if (adev->gfx.disable_kq) {
>> +        /* GFX11 needs the GFX ring for clear buffer */
>> +        if (amdgpu_ip_version(adev, GC_HWIP, 0) <= IP_VERSION(12, 0, 
>> 0))

Yes the check has to be  < as gfx12 do not need the clear buffer based 
on our discussions.

Regards
Sunil

>
> Should this be < instead of <=?

>
> Regards,
>   Felix
>
>> +            return false;
>> +        else
>> +            return true;
>> +    }
>> +    return false;
>> +}
>> +
>>   /* create MQD for each compute/gfx queue */
>>   int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
>>                  unsigned int mqd_size, int xcc_id)
>> @@ -379,6 +391,7 @@ int amdgpu_gfx_mqd_sw_init(struct amdgpu_device 
>> *adev,
>>       struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
>>       struct amdgpu_ring *ring = &kiq->ring;
>>       u32 domain = AMDGPU_GEM_DOMAIN_GTT;
>> +    bool disable_kq_gfx = amdgpu_gfx_disable_gfx_kq(adev);

name of variable and function could be in sync. disable_gfx_kq and 
amdgpu_gfx_disable_gfx_kq or change function name according to variable.

Also another suggestion here is better to have one more variable in the 
gfx struct or ring and read this amdgpu_gfx_disable_gfx_kq once and use 
it in all the places. It does looks confusing
so many similar sounding names.

Regards
Sunil
>>     #if !defined(CONFIG_ARM) && !defined(CONFIG_ARM64)
>>       /* Only enable on gfx10 and 11 for now to avoid changing 
>> behavior on older chips */
>> @@ -413,7 +426,8 @@ int amdgpu_gfx_mqd_sw_init(struct amdgpu_device 
>> *adev,
>>           }
>>       }
>>   -    if (adev->asic_type >= CHIP_NAVI10 && amdgpu_async_gfx_ring) {
>> +    if (adev->asic_type >= CHIP_NAVI10 && amdgpu_async_gfx_ring &&
>> +        !disable_kq_gfx) {
>>           /* create MQD for each KGQ */
>>           for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
>>               ring = &adev->gfx.gfx_ring[i];
>> @@ -437,25 +451,28 @@ int amdgpu_gfx_mqd_sw_init(struct amdgpu_device 
>> *adev,
>>           }
>>       }
>>   -    /* create MQD for each KCQ */
>> -    for (i = 0; i < adev->gfx.num_compute_rings; i++) {
>> -        j = i + xcc_id * adev->gfx.num_compute_rings;
>> -        ring = &adev->gfx.compute_ring[j];
>> -        if (!ring->mqd_obj) {
>> -            r = amdgpu_bo_create_kernel(adev, mqd_size, PAGE_SIZE,
>> -                            domain, &ring->mqd_obj,
>> -                            &ring->mqd_gpu_addr, &ring->mqd_ptr);
>> -            if (r) {
>> -                dev_warn(adev->dev, "failed to create ring mqd bo 
>> (%d)", r);
>> -                return r;
>> -            }
>> +    if (!adev->gfx.disable_kq) {
>
> Maybe just set adev->gfx.num_compute_rings to 0 somewhere, then you 
> don't need this condition.
>
>
>> +        /* create MQD for each KCQ */
>> +        for (i = 0; i < adev->gfx.num_compute_rings; i++) {
>> +            j = i + xcc_id * adev->gfx.num_compute_rings;
>> +            ring = &adev->gfx.compute_ring[j];
>> +            if (!ring->mqd_obj) {
>> +                r = amdgpu_bo_create_kernel(adev, mqd_size, PAGE_SIZE,
>> +                                domain, &ring->mqd_obj,
>> +                                &ring->mqd_gpu_addr, &ring->mqd_ptr);
>> +                if (r) {
>> +                    dev_warn(adev->dev, "failed to create ring mqd 
>> bo (%d)", r);
>> +                    return r;
>> +                }
>>   -            ring->mqd_size = mqd_size;
>> -            /* prepare MQD backup */
>> -            adev->gfx.mec.mqd_backup[j] = kzalloc(mqd_size, 
>> GFP_KERNEL);
>> -            if (!adev->gfx.mec.mqd_backup[j]) {
>> -                dev_warn(adev->dev, "no memory to create MQD backup 
>> for ring %s\n", ring->name);
>> -                return -ENOMEM;
>> +                ring->mqd_size = mqd_size;
>> +                /* prepare MQD backup */
>> +                adev->gfx.mec.mqd_backup[j] = kzalloc(mqd_size, 
>> GFP_KERNEL);
>> +                if (!adev->gfx.mec.mqd_backup[j]) {
>> +                    dev_warn(adev->dev, "no memory to create MQD 
>> backup for ring %s\n",
>> +                         ring->name);
>> +                    return -ENOMEM;
>> +                }
>>               }
>>           }
>>       }
>> @@ -468,8 +485,10 @@ void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device 
>> *adev, int xcc_id)
>>       struct amdgpu_ring *ring = NULL;
>>       int i, j;
>>       struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
>> +    bool disable_kq_gfx = amdgpu_gfx_disable_gfx_kq(adev);
>>   -    if (adev->asic_type >= CHIP_NAVI10 && amdgpu_async_gfx_ring) {
>> +    if (adev->asic_type >= CHIP_NAVI10 && amdgpu_async_gfx_ring &&
>> +        !disable_kq_gfx) {
>>           for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
>>               ring = &adev->gfx.gfx_ring[i];
>>               kfree(adev->gfx.me.mqd_backup[i]);
>> @@ -479,13 +498,15 @@ void amdgpu_gfx_mqd_sw_fini(struct 
>> amdgpu_device *adev, int xcc_id)
>>           }
>>       }
>>   -    for (i = 0; i < adev->gfx.num_compute_rings; i++) {
>> -        j = i + xcc_id * adev->gfx.num_compute_rings;
>> -        ring = &adev->gfx.compute_ring[j];
>> -        kfree(adev->gfx.mec.mqd_backup[j]);
>> -        amdgpu_bo_free_kernel(&ring->mqd_obj,
>> -                      &ring->mqd_gpu_addr,
>> -                      &ring->mqd_ptr);
>> +    if (!adev->gfx.disable_kq) {
>
> Same as above.
>
>
>> +        for (i = 0; i < adev->gfx.num_compute_rings; i++) {
>> +            j = i + xcc_id * adev->gfx.num_compute_rings;
>> +            ring = &adev->gfx.compute_ring[j];
>> +            kfree(adev->gfx.mec.mqd_backup[j]);
>> +            amdgpu_bo_free_kernel(&ring->mqd_obj,
>> +                          &ring->mqd_gpu_addr,
>> +                          &ring->mqd_ptr);
>> +        }
>>       }
>>         ring = &kiq->ring;
>> @@ -502,6 +523,9 @@ int amdgpu_gfx_disable_kcq(struct amdgpu_device 
>> *adev, int xcc_id)
>>       int i, r = 0;
>>       int j;
>>   +    if (adev->gfx.disable_kq)
>
> Same as above.
>
>
>> +        return 0;
>> +
>>       if (adev->enable_mes) {
>>           for (i = 0; i < adev->gfx.num_compute_rings; i++) {
>>               j = i + xcc_id * adev->gfx.num_compute_rings;
>> @@ -547,11 +571,15 @@ int amdgpu_gfx_disable_kcq(struct amdgpu_device 
>> *adev, int xcc_id)
>>     int amdgpu_gfx_disable_kgq(struct amdgpu_device *adev, int xcc_id)
>>   {
>> +    bool disable_kq_gfx = amdgpu_gfx_disable_gfx_kq(adev);
>>       struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
>>       struct amdgpu_ring *kiq_ring = &kiq->ring;
>>       int i, r = 0;
>>       int j;
>>   +    if (disable_kq_gfx)
>> +        return 0;
> Maybe just set adev->gfx.num_gfx_rings to 0 somewhere, then you don't 
> need this condition.
>
> Regards,
>   Felix
>
>
>> +
>>       if (adev->enable_mes) {
>>           if (amdgpu_gfx_is_master_xcc(adev, xcc_id)) {
>>               for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
>> @@ -657,6 +685,9 @@ int amdgpu_gfx_enable_kcq(struct amdgpu_device 
>> *adev, int xcc_id)
>>       uint64_t queue_mask = 0;
>>       int r, i, j;
>>   +    if (adev->gfx.disable_kq)
>> +        return 0;
>> +
>>       if (adev->mes.enable_legacy_queue_map)
>>           return amdgpu_gfx_mes_enable_kcq(adev, xcc_id);
>>   @@ -716,10 +747,14 @@ int amdgpu_gfx_enable_kcq(struct 
>> amdgpu_device *adev, int xcc_id)
>>     int amdgpu_gfx_enable_kgq(struct amdgpu_device *adev, int xcc_id)
>>   {
>> +    bool disable_kq_gfx = amdgpu_gfx_disable_gfx_kq(adev);
>>       struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
>>       struct amdgpu_ring *kiq_ring = &kiq->ring;
>>       int r, i, j;
>>   +    if (disable_kq_gfx)
>> +        return 0;
>> +
>>       if (!kiq->pmf || !kiq->pmf->kiq_map_queues)
>>           return -EINVAL;
>>   @@ -1544,6 +1579,9 @@ static ssize_t 
>> amdgpu_gfx_set_run_cleaner_shader(struct device *dev,
>>       if (adev->in_suspend && !adev->in_runpm)
>>           return -EPERM;
>>   +    if (adev->gfx.disable_kq)
>> +        return -ENOTSUPP;
>> +
>>       ret = kstrtol(buf, 0, &value);
>>         if (ret)
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
>> index ddf4533614bac..8fa68a4ac34f1 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
>> @@ -483,6 +483,8 @@ struct amdgpu_gfx {
>>         atomic_t            total_submission_cnt;
>>       struct delayed_work        idle_work;
>> +
>> +    bool                disable_kq;
>>   };
>>     struct amdgpu_gfx_ras_reg_entry {


More information about the amd-gfx mailing list