From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from mgamail.intel.com (mgamail.intel.com [198.175.65.12]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id F074036CE1E; Thu, 11 Jun 2026 19:41:36 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=198.175.65.12 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1781206901; cv=none; b=XVAynPn1X6rx7sDhNNlPkVEyVZNjnaEmKzWEQnZRNL02LmonDoDbn2qoet6SfWsOYbuAkbY/h3Hgvr0YW8MfIa+VNkK4bz0R+6slbvLqPmgMlWakrC6q+pUlb/AysybXDor1NDleXRB8eaVUviDq3PRRCcTq7q8R5conrqJGa3s= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1781206901; c=relaxed/simple; bh=XGA8XvGhar3Z1r6Oh+DGdi0DMjCB5+1+8ldPsokTHIE=; h=Message-ID:Subject:From:To:Cc:Date:In-Reply-To:References: Content-Type:MIME-Version; b=rLUkubp1JfPsPLQBd4O8RcdZ2oT1j7axa6J9UW+98UsubXwskd7W7psEkOVd+SIrXfdN7Kl/JLkxKkG/LZZesbnD/3iRLzmHFPvVLYs6WTxCjmxnbomBZqJyZiNs1hgJBICdVOh5qk6QpunKFisqbQH+bTbwxFpW7/4YRSRaPFw= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=linux.intel.com; spf=pass smtp.mailfrom=linux.intel.com; dkim=pass (2048-bit key) header.d=intel.com header.i=@intel.com header.b=KO5G2tCE; arc=none smtp.client-ip=198.175.65.12 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=linux.intel.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=linux.intel.com Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=intel.com header.i=@intel.com header.b="KO5G2tCE" DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1781206897; x=1812742897; h=message-id:subject:from:to:cc:date:in-reply-to: references:content-transfer-encoding:mime-version; bh=XGA8XvGhar3Z1r6Oh+DGdi0DMjCB5+1+8ldPsokTHIE=; b=KO5G2tCE96w4Na6uToqh1+gUtdxWoPHX1fjFG2wThAnn/9swpN4kweCI XoSCSJX3hT7irJ14MZGZuD272QLlADhJnzDd1lJ/O3/FRo++DT2cqj4ir jSBl+aTzAyxuhk+KRvwaizMMoi059H8KnLX+aCVHEyqE655W+K2YRNLym f7M5Sp/qGy7uD8yhST7KAM1VZuw5WAfHpBwqCEPXSuRL1rM+Ih35Yl6ji MCh3WOOvpx7dOlArcF8XKn1Qg5+NAin/V9J1++73uzidfcw3PcdsGvIIE CRrOD4Iy/CgW0plUaSJw2P/HvGvK3h3HDkBlwKp0u24m7oHhIeeyA+inA w==; X-CSE-ConnectionGUID: y5bBTs6QQfegVI1DH+9I+w== X-CSE-MsgGUID: E60/WcdxSPyrM6v6iL1/rw== X-IronPort-AV: E=McAfee;i="6800,10657,11813"; a="93520571" X-IronPort-AV: E=Sophos;i="6.24,199,1774335600"; d="scan'208";a="93520571" Received: from fmviesa003.fm.intel.com ([10.60.135.143]) by orvoesa104.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 11 Jun 2026 12:41:36 -0700 X-CSE-ConnectionGUID: vvwdwSeZRd228JgWLwt3IQ== X-CSE-MsgGUID: NS0KcAVJTj2pqMYozOfh6Q== X-ExtLoop1: 1 Received: from amilburn-desk.amilburn-desk (HELO [10.245.244.169]) ([10.245.244.169]) by fmviesa003-auth.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 11 Jun 2026 12:41:31 -0700 Message-ID: Subject: Re: [PATCH v6 6/6] drm/amdgpu: Wire up dmem cgroup reclaim for VRAM manager From: Thomas =?ISO-8859-1?Q?Hellstr=F6m?= To: intel-xe@lists.freedesktop.org Cc: Natalie Vock , Johannes Weiner , Tejun Heo , Michal =?ISO-8859-1?Q?Koutn=FD?= , cgroups@vger.kernel.org, Huang Rui , Matthew Brost , Matthew Auld , Maarten Lankhorst , Maxime Ripard , Thomas Zimmermann , Simona Vetter , David Airlie , Christian =?ISO-8859-1?Q?K=F6nig?= , Alex Deucher , Rodrigo Vivi , dri-devel@lists.freedesktop.org, amd-gfx@lists.freedesktop.org, linux-kernel@vger.kernel.org Date: Thu, 11 Jun 2026 21:41:28 +0200 In-Reply-To: <20260611173301.17473-7-thomas.hellstrom@linux.intel.com> References: <20260611173301.17473-1-thomas.hellstrom@linux.intel.com> <20260611173301.17473-7-thomas.hellstrom@linux.intel.com> Organization: Intel Sweden AB, Registration Number: 556189-6027 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable User-Agent: Evolution 3.58.3 (3.58.3-1.fc43) Precedence: bulk X-Mailing-List: cgroups@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 On Thu, 2026-06-11 at 19:33 +0200, Thomas Hellstr=C3=B6m wrote: > Register the VRAM manager with the dmem cgroup reclaim infrastructure > so that lowering dmem.max below current VRAM usage triggers TTM > eviction rather than failing with -EBUSY. >=20 > Guard place->flags in amdgpu_ttm_bo_eviction_valuable() against NULL, > as the TTM reclaim path passes a NULL place in cgroup drain mode. >=20 > v3: > - Rebased on fix for uninitialized list and buddy allocator on the > =C2=A0 drmm_cgroup_register_region() error path. >=20 > v5: > - Rebased on the introduction of struct dmem_cgroup_init. > - Clear the reclaim callback in amdgpu_vram_mgr_fini() to prevent > =C2=A0 use-after-free if cgroup reclaim is triggered after driver unbind > =C2=A0 while userspace holds an open DRM file descriptor. (Sashiko-bot) > - Switch from drmm_cgroup_register_region() to the raw > =C2=A0 dmem_cgroup_register_region() and store the region in > =C2=A0 amdgpu_vram_mgr.cg_region. Call dmem_cgroup_unregister_region() > =C2=A0 in amdgpu_vram_mgr_fini() after ttm_resource_manager_evict_all() > =C2=A0 to drain in-flight reclaim callbacks, and clear man->cg afterwards= . > =C2=A0 This is required because amdgpu's vram manager fini is called > =C2=A0 explicitly during driver unbind, which may precede the DRM device > =C2=A0 release and thus precede any drmm-based cleanup. (Sashiko-bot) >=20 > v6: > - Fix mgr->cg_region never being assigned, so > =C2=A0 dmem_cgroup_unregister_region() in fini silently no-ops on NULL > =C2=A0 and leaks the region. (Sashiko-bot) > - Reorder fini to call set_used(false) and evict_all() before > =C2=A0 dmem_cgroup_unregister_region(), so ttm_resource_free() can > =C2=A0 uncharge via man->cg during eviction; clear man->cg after > =C2=A0 unregister. (Sashiko-bot) >=20 > Assisted-by: GitHub_Copilot:claude-sonnet-4.6 > Signed-off-by: Thomas Hellstr=C3=B6m > --- > =C2=A0drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c=C2=A0=C2=A0=C2=A0=C2=A0=C2= =A0 |=C2=A0 2 +- > =C2=A0drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 31 ++++++++++++++++-= - > -- > =C2=A0drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h |=C2=A0 2 ++ > =C2=A03 files changed, 28 insertions(+), 7 deletions(-) >=20 > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c > b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c > index 2740de94e93c..8cbcd33f51a5 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c > @@ -1488,7 +1488,7 @@ static bool > amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo, > =C2=A0 dma_resv_for_each_fence(&resv_cursor, bo->base.resv, > =C2=A0 DMA_RESV_USAGE_BOOKKEEP, f) { > =C2=A0 if (amdkfd_fence_check_mm(f, current->mm) && > - =C2=A0=C2=A0=C2=A0 !(place->flags & TTM_PL_FLAG_CONTIGUOUS)) > + =C2=A0=C2=A0=C2=A0 !(place && (place->flags & > TTM_PL_FLAG_CONTIGUOUS))) > =C2=A0 return false; > =C2=A0 } > =C2=A0 > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > index 08f05c3aed1d..2250bab0970d 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > @@ -906,6 +906,10 @@ static const struct ttm_resource_manager_func > amdgpu_vram_mgr_func =3D { > =C2=A0 .debug =3D amdgpu_vram_mgr_debug > =C2=A0}; > =C2=A0 > +static const struct dmem_cgroup_ops amdgpu_vram_mgr_dmem_ops =3D { > + .reclaim =3D ttm_resource_manager_dmem_reclaim, > +}; Probably might want to block reclaim after device unbind, just like xe. I'll look at that for v7. > + > =C2=A0/** > =C2=A0 * amdgpu_vram_mgr_init - init VRAM manager and DRM MM > =C2=A0 * > @@ -917,6 +921,7 @@ int amdgpu_vram_mgr_init(struct amdgpu_device > *adev) > =C2=A0{ > =C2=A0 struct amdgpu_vram_mgr *mgr =3D &adev->mman.vram_mgr; > =C2=A0 struct ttm_resource_manager *man =3D &mgr->manager; > + struct dmem_cgroup_region *cg; > =C2=A0 int err; > =C2=A0 > =C2=A0 ttm_resource_manager_init(man, &adev->mman.bdev, > @@ -933,12 +938,16 @@ int amdgpu_vram_mgr_init(struct amdgpu_device > *adev) > =C2=A0 if (err) > =C2=A0 return err; > =C2=A0 > - man->cg =3D drmm_cgroup_register_region(adev_to_drm(adev), > "vram", > - =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 &(struct > dmem_cgroup_init){ > - .size =3D adev- > >gmc.real_vram_size, > - =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 }); > - if (IS_ERR(man->cg)) > - return PTR_ERR(man->cg); > + cg =3D dmem_cgroup_register_region(&(struct dmem_cgroup_init){ > + =C2=A0=C2=A0=C2=A0=C2=A0 .size =3D adev- > >gmc.real_vram_size, > + =C2=A0=C2=A0=C2=A0=C2=A0 .ops =3D > &amdgpu_vram_mgr_dmem_ops, > + =C2=A0=C2=A0=C2=A0=C2=A0 .reclaim_priv =3D man, > + }, "vram"); > + if (IS_ERR(cg)) > + return PTR_ERR(cg); > + > + mgr->cg_region =3D cg; > + ttm_resource_manager_set_dmem_region(man, cg); > =C2=A0 > =C2=A0 ttm_set_driver_manager(&adev->mman.bdev, TTM_PL_VRAM, &mgr- > >manager); > =C2=A0 ttm_resource_manager_set_used(man, true); > @@ -966,6 +975,16 @@ void amdgpu_vram_mgr_fini(struct amdgpu_device > *adev) > =C2=A0 if (ret) > =C2=A0 return; > =C2=A0 > + /* > + * Drain any in-flight dmem cgroup reclaim callbacks and > remove the > + * region from the global list.=C2=A0 This must happen after > evict_all() > + * so that ttm_resource_free() can still uncharge via man- > >cg while > + * BOs are being evicted. > + */ > + dmem_cgroup_unregister_region(mgr->cg_region); > + mgr->cg_region =3D NULL; > + man->cg =3D NULL; > + > =C2=A0 mutex_lock(&mgr->lock); > =C2=A0 list_for_each_entry_safe(rsv, temp, &mgr- > >reservations_pending, blocks) > =C2=A0 kfree(rsv); > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h > b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h > index 429a21a2e9b2..07103cddb335 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h > @@ -36,6 +36,8 @@ struct amdgpu_vram_mgr { > =C2=A0 atomic64_t vis_usage; > =C2=A0 u64 default_page_size; > =C2=A0 struct list_head allocated_vres_list; > + /** @cg_region: dmem cgroup region for VRAM; unregistered in > fini. */ > + struct dmem_cgroup_region *cg_region; > =C2=A0}; > =C2=A0 > =C2=A0struct amdgpu_vres_task {