Re: [PATCH v2] drm/amdgpu/fence: Fix oops due to non-matching drm_sched init/fini

From: "Guilherme G. Piccoli" <gpiccoli@igalia.com>
To: "Christian König" <christian.koenig@amd.com>,
	"Guchun Chen" <guchun.chen@amd.com>,
	"Luben Tuikov" <luben.tuikov@amd.com>
Cc: kernel@gpiccoli.net, Xinhui.Pan@amd.com,
	amd-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org,
	kernel-dev@igalia.com, alexander.deucher@amd.com,
	Mario Limonciello <mario.limonciello@amd.com>
Subject: Re: [PATCH v2] drm/amdgpu/fence: Fix oops due to non-matching drm_sched init/fini
Date: Thu, 2 Feb 2023 10:52:03 -0300	[thread overview]
Message-ID: <9f35c852-18c4-2f3f-d03b-ff76def8516a@igalia.com> (raw)
In-Reply-To: <cd0e2994-f85f-d837-609f-7056d5fb7231@amd.com>

On 02/02/2023 08:58, Christian König wrote:
> [...]
>> +		if (!ring->no_scheduler && ring->sched.ops)
>>   			drm_sched_fini(&ring->sched);
> 
> I think we should drop the check for no_scheduler here and just call 
> drm_sched_fini() when the scheduler instance was initialized before.
> 
> Background is that I've found a couple of places where we potentially 
> set no_scheduler to false after the scheduler was already initialized.
> 
> This is then probably leading to a memory leak or worth.
> 
> Regards,
> Christian.

Thanks Christian, nice finding! And thanks for the reviews Guchun / Luben =)

I just submitted a V3 [0], but didn't add the review tags from Guchun /
Luben, since the patch changed.

Cheers,

Guilherme

[0]
https://lore.kernel.org/dri-devel/20230202134856.1382169-1-gpiccoli@igalia.com/