AMD-GFX Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: "Das, Nirmoy" <nirmoy.das@amd.com>
To: "Christian König" <christian.koenig@amd.com>,
	"Lazar, Lijo" <lijo.lazar@amd.com>,
	amd-gfx@lists.freedesktop.org
Cc: andrey.grodzovsky@amd.com
Subject: Re: [PATCH 1/3] drm/amdgpu: do not pass ttm_resource_manager to gtt_mgr
Date: Wed, 20 Oct 2021 13:12:14 +0200	[thread overview]
Message-ID: <80997b49-77bb-ea45-24ef-04a1d1ca44c7@amd.com> (raw)
In-Reply-To: <42e31246-5b59-b806-9eac-b2a95a821fa2@amd.com>


On 10/20/2021 12:49 PM, Christian König wrote:
> Am 20.10.21 um 11:19 schrieb Lazar, Lijo:
>>
>>
>> On 10/20/2021 2:18 PM, Das, Nirmoy wrote:
>>>
>>> On 10/20/2021 8:49 AM, Christian König wrote:
>>>> Am 19.10.21 um 20:14 schrieb Nirmoy Das:
>>>>> Do not allow exported amdgpu_gtt_mgr_*() to accept
>>>>> any ttm_resource_manager pointer. Also there is no need
>>>>> to force other module to call a ttm function just to
>>>>> eventually call gtt_mgr functions.
>>>>
>>>> That's a rather bad idea I think.
>>>>
>>>> The GTT and VRAM manager work on their respective objects and not 
>>>> on the adev directly.
>>>
>>>
>>> What is bothering me is : it is obvious that  the 
>>> amdgpu_gtt_mgr_usage() for example should only calculate
>>>
>>> usages for TTM_PL_TT type resource manager, why to pass that 
>>> explicitly. I am trying to leverage the fact that
>>>
>>> we only have one gtt/vram manager for a adev and the functions that 
>>> I changed  work on whole gtt/vram manager
>>>
>>> as a unit.
>>>
>>
>> Don't know about the functional aspects. From a sofware perspective, 
>> amdgpu_gtt_mgr_*() operating on struct amdgpu_gtt_mgr *mgr seems more 
>> logical.
>
> What we could do is to pass in amdgpu_gtt_mgr instead of 
> ttm_resource_manager and then use &adev->mman.gtt_mgr.


Sounds good, I will try this way then.


Regards,

Nirmoy

>
> Regards,
> Christian.
>
>>
>>
>> Thanks,
>> Lijo
>>
>>>
>>> Regards,
>>>
>>> Nirmoy
>>>
>>>
>>>>
>>>> Christian.
>>>>
>>>>>
>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>> ---
>>>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  |  4 +--
>>>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 31 
>>>>> ++++++++++++---------
>>>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c     |  4 +--
>>>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  4 +--
>>>>>   4 files changed, 24 insertions(+), 19 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
>>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>>>> index 41ce86244144..5807df52031c 100644
>>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>>>> @@ -4287,7 +4287,7 @@ static int amdgpu_device_reset_sriov(struct 
>>>>> amdgpu_device *adev,
>>>>>         amdgpu_virt_init_data_exchange(adev);
>>>>>       /* we need recover gart prior to run SMC/CP/SDMA resume */
>>>>> - amdgpu_gtt_mgr_recover(ttm_manager_type(&adev->mman.bdev, 
>>>>> TTM_PL_TT));
>>>>> +    amdgpu_gtt_mgr_recover(adev);
>>>>>         r = amdgpu_device_fw_loading(adev);
>>>>>       if (r)
>>>>> @@ -4604,7 +4604,7 @@ int amdgpu_do_asic_reset(struct list_head 
>>>>> *device_list_handle,
>>>>>                       amdgpu_inc_vram_lost(tmp_adev);
>>>>>                   }
>>>>>   -                r = 
>>>>> amdgpu_gtt_mgr_recover(ttm_manager_type(&tmp_adev->mman.bdev, 
>>>>> TTM_PL_TT));
>>>>> +                r = amdgpu_gtt_mgr_recover(tmp_adev);
>>>>>                   if (r)
>>>>>                       goto out;
>>>>>   diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
>>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
>>>>> index c18f16b3be9c..5e41f8ef743a 100644
>>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
>>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
>>>>> @@ -77,10 +77,8 @@ static ssize_t 
>>>>> amdgpu_mem_info_gtt_used_show(struct device *dev,
>>>>>   {
>>>>>       struct drm_device *ddev = dev_get_drvdata(dev);
>>>>>       struct amdgpu_device *adev = drm_to_adev(ddev);
>>>>> -    struct ttm_resource_manager *man;
>>>>>   -    man = ttm_manager_type(&adev->mman.bdev, TTM_PL_TT);
>>>>> -    return sysfs_emit(buf, "%llu\n", amdgpu_gtt_mgr_usage(man));
>>>>> +    return sysfs_emit(buf, "%llu\n", amdgpu_gtt_mgr_usage(adev));
>>>>>   }
>>>>>     static DEVICE_ATTR(mem_info_gtt_total, S_IRUGO,
>>>>> @@ -206,14 +204,19 @@ static void amdgpu_gtt_mgr_del(struct 
>>>>> ttm_resource_manager *man,
>>>>>   /**
>>>>>    * amdgpu_gtt_mgr_usage - return usage of GTT domain
>>>>>    *
>>>>> - * @man: TTM memory type manager
>>>>> + * @adev: amdgpu_device pointer
>>>>>    *
>>>>>    * Return how many bytes are used in the GTT domain
>>>>>    */
>>>>> -uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager *man)
>>>>> +uint64_t amdgpu_gtt_mgr_usage(struct amdgpu_device *adev)
>>>>>   {
>>>>> -    struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
>>>>> -    s64 result = man->size - atomic64_read(&mgr->available);
>>>>> +    struct ttm_resource_manager *man;
>>>>> +    struct amdgpu_gtt_mgr *mgr;
>>>>> +    s64 result;
>>>>> +
>>>>> +    man = ttm_manager_type(&adev->mman.bdev, TTM_PL_TT);
>>>>> +    mgr = to_gtt_mgr(man);
>>>>> +    result = man->size - atomic64_read(&mgr->available);
>>>>>         return (result > 0 ? result : 0) * PAGE_SIZE;
>>>>>   }
>>>>> @@ -221,19 +224,20 @@ uint64_t amdgpu_gtt_mgr_usage(struct 
>>>>> ttm_resource_manager *man)
>>>>>   /**
>>>>>    * amdgpu_gtt_mgr_recover - re-init gart
>>>>>    *
>>>>> - * @man: TTM memory type manager
>>>>> + * @adev: amdgpu_device pointer
>>>>>    *
>>>>>    * Re-init the gart for each known BO in the GTT.
>>>>>    */
>>>>> -int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
>>>>> +int amdgpu_gtt_mgr_recover(struct amdgpu_device *adev)
>>>>>   {
>>>>> -    struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
>>>>> -    struct amdgpu_device *adev;
>>>>> +    struct ttm_resource_manager *man;
>>>>> +    struct amdgpu_gtt_mgr *mgr;
>>>>>       struct amdgpu_gtt_node *node;
>>>>>       struct drm_mm_node *mm_node;
>>>>>       int r = 0;
>>>>>   -    adev = container_of(mgr, typeof(*adev), mman.gtt_mgr);
>>>>> +    man = ttm_manager_type(&adev->mman.bdev, TTM_PL_TT);
>>>>> +    mgr = to_gtt_mgr(man);
>>>>>       spin_lock(&mgr->lock);
>>>>>       drm_mm_for_each_node(mm_node, &mgr->mm) {
>>>>>           node = container_of(mm_node, typeof(*node), 
>>>>> base.mm_nodes[0]);
>>>>> @@ -260,6 +264,7 @@ static void amdgpu_gtt_mgr_debug(struct 
>>>>> ttm_resource_manager *man,
>>>>>                    struct drm_printer *printer)
>>>>>   {
>>>>>       struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
>>>>> +    struct amdgpu_device *adev = container_of(mgr, typeof(*adev), 
>>>>> mman.gtt_mgr);
>>>>>         spin_lock(&mgr->lock);
>>>>>       drm_mm_print(&mgr->mm, printer);
>>>>> @@ -267,7 +272,7 @@ static void amdgpu_gtt_mgr_debug(struct 
>>>>> ttm_resource_manager *man,
>>>>>         drm_printf(printer, "man size:%llu pages, gtt 
>>>>> available:%lld pages, usage:%lluMB\n",
>>>>>              man->size, (u64)atomic64_read(&mgr->available),
>>>>> -           amdgpu_gtt_mgr_usage(man) >> 20);
>>>>> +           amdgpu_gtt_mgr_usage(adev) >> 20);
>>>>>   }
>>>>>     static const struct ttm_resource_manager_func 
>>>>> amdgpu_gtt_mgr_func = {
>>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
>>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
>>>>> index d2955ea4a62b..b9b38f70e416 100644
>>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
>>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
>>>>> @@ -678,7 +678,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, 
>>>>> void *data, struct drm_file *filp)
>>>>>           ui64 = 
>>>>> amdgpu_vram_mgr_vis_usage(ttm_manager_type(&adev->mman.bdev, 
>>>>> TTM_PL_VRAM));
>>>>>           return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT 
>>>>> : 0;
>>>>>       case AMDGPU_INFO_GTT_USAGE:
>>>>> -        ui64 = 
>>>>> amdgpu_gtt_mgr_usage(ttm_manager_type(&adev->mman.bdev, TTM_PL_TT));
>>>>> +        ui64 = amdgpu_gtt_mgr_usage(adev);
>>>>>           return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT 
>>>>> : 0;
>>>>>       case AMDGPU_INFO_GDS_CONFIG: {
>>>>>           struct drm_amdgpu_info_gds gds_info;
>>>>> @@ -738,7 +738,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, 
>>>>> void *data, struct drm_file *filp)
>>>>>           mem.gtt.usable_heap_size = mem.gtt.total_heap_size -
>>>>>               atomic64_read(&adev->gart_pin_size);
>>>>>           mem.gtt.heap_usage =
>>>>> -            amdgpu_gtt_mgr_usage(gtt_man);
>>>>> +            amdgpu_gtt_mgr_usage(adev);
>>>>>           mem.gtt.max_allocation = mem.gtt.usable_heap_size * 3 / 4;
>>>>>             return copy_to_user(out, &mem,
>>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h 
>>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>>>>> index 91a087f9dc7c..6e337cacef51 100644
>>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>>>>> @@ -114,8 +114,8 @@ int amdgpu_vram_mgr_init(struct amdgpu_device 
>>>>> *adev);
>>>>>   void amdgpu_vram_mgr_fini(struct amdgpu_device *adev);
>>>>>     bool amdgpu_gtt_mgr_has_gart_addr(struct ttm_resource *mem);
>>>>> -uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager *man);
>>>>> -int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man);
>>>>> +uint64_t amdgpu_gtt_mgr_usage(struct amdgpu_device *adev);
>>>>> +int amdgpu_gtt_mgr_recover(struct amdgpu_device *adev);
>>>>>     uint64_t amdgpu_preempt_mgr_usage(struct ttm_resource_manager 
>>>>> *man);
>>>>
>

      reply	other threads:[~2021-10-20 11:12 UTC|newest]

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-10-19 18:14 [PATCH 1/3] drm/amdgpu: do not pass ttm_resource_manager to gtt_mgr Nirmoy Das
2021-10-19 18:14 ` [PATCH 2/3] drm/amdgpu: do not pass ttm_resource_manager to vram_mgr Nirmoy Das
2021-10-19 18:14 ` [PATCH v2 3/3] drm/amdgpu: recover gart table at resume Nirmoy Das
2021-10-20  6:52   ` Christian König
2021-10-20  8:41     ` Das, Nirmoy
2021-10-20  9:11   ` Lazar, Lijo
2021-10-20  9:53     ` Das, Nirmoy
2021-10-20 10:03       ` Lazar, Lijo
2021-10-20 10:12         ` Das, Nirmoy
2021-10-20 10:15           ` Lazar, Lijo
2021-10-20 10:21             ` Das, Nirmoy
2021-10-20 10:51               ` Christian König
2021-10-20 11:10                 ` Das, Nirmoy
2021-10-20  6:49 ` [PATCH 1/3] drm/amdgpu: do not pass ttm_resource_manager to gtt_mgr Christian König
2021-10-20  8:48   ` Das, Nirmoy
2021-10-20  9:19     ` Lazar, Lijo
2021-10-20 10:49       ` Christian König
2021-10-20 11:12         ` Das, Nirmoy [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=80997b49-77bb-ea45-24ef-04a1d1ca44c7@amd.com \
    --to=nirmoy.das@amd.com \
    --cc=amd-gfx@lists.freedesktop.org \
    --cc=andrey.grodzovsky@amd.com \
    --cc=christian.koenig@amd.com \
    --cc=lijo.lazar@amd.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox