From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 51EB0C3DA4A for ; Mon, 19 Aug 2024 19:56:29 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 1E53510E35B; Mon, 19 Aug 2024 19:56:29 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=intel.com header.i=@intel.com header.b="fELjEvv+"; dkim-atps=neutral Received: from mgamail.intel.com (mgamail.intel.com [198.175.65.12]) by gabe.freedesktop.org (Postfix) with ESMTPS id 53AB510E35B for ; Mon, 19 Aug 2024 19:56:27 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1724097387; x=1755633387; h=message-id:date:mime-version:subject:to:cc:references: from:in-reply-to:content-transfer-encoding; bh=qdvkJkOpn7dJCgyWaeEmvLH/JGP4M/kxjf43uv9lXtM=; b=fELjEvv+74BuLsrRzHULlPQcuV1hK4RAPVOp32IyVCtZkTeY49O0ss6t Wdz1kIB/Nbgf8pFrmogXe+i2bBV310Sz1dEhLSmcI+7js4XMdcG5JB92Q ZAbGRM/x+r6ulhb2qYoAXUOzt2bOy0d9H6QkL58L2hK/0L2kIymi3iYka PB1QkAtKBXmwes2v261PEdxyUBeJb5075wuP3Yiis79fHcP4PeDlEFxrG cMTGq1ho8lVNqXQmKrpYTF6vQYf7UkXRNZ04xCzFHb25mTAljKmddHgjy EfagLvRKU8LGYBGQI+wauVouBKg0p7/r7OintBAe+IMCeuzjePXlj6E+1 w==; X-CSE-ConnectionGUID: Ku/5K+bNSzyVz2X1wi+nYg== X-CSE-MsgGUID: 6EeGiYubQkynDldBDy7rsQ== X-IronPort-AV: E=McAfee;i="6700,10204,11169"; a="33749832" X-IronPort-AV: E=Sophos;i="6.10,160,1719903600"; d="scan'208";a="33749832" Received: from fmviesa007.fm.intel.com ([10.60.135.147]) by orvoesa104.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 19 Aug 2024 12:56:27 -0700 X-CSE-ConnectionGUID: 9v+2JTj2Sgyd9pnSQS13Zg== X-CSE-MsgGUID: CB7FSWagSTywzx92wF6CNA== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="6.10,160,1719903600"; d="scan'208";a="60179837" Received: from irvmail002.ir.intel.com ([10.43.11.120]) by fmviesa007.fm.intel.com with ESMTP; 19 Aug 2024 12:56:24 -0700 Received: from [10.246.1.253] (mwajdecz-MOBL.ger.corp.intel.com [10.246.1.253]) by irvmail002.ir.intel.com (Postfix) with ESMTP id DD00A2FC5F; Mon, 19 Aug 2024 20:56:22 +0100 (IST) Message-ID: <5ac53aec-d79b-44a3-977d-650962929ccd@intel.com> Date: Mon, 19 Aug 2024 21:56:22 +0200 MIME-Version: 1.0 User-Agent: Mozilla Thunderbird Subject: Re: [PATCH 11/12] drm/xe: Make xe_ggtt_node struct independent To: Rodrigo Vivi , intel-xe@lists.freedesktop.org Cc: lucas.demarchi@intel.com, jose.souza@intel.com, Matthew Auld , Matthew Brost References: <20240817103556.163783-1-rodrigo.vivi@intel.com> <20240817103556.163783-11-rodrigo.vivi@intel.com> Content-Language: en-US From: Michal Wajdeczko In-Reply-To: <20240817103556.163783-11-rodrigo.vivi@intel.com> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit X-BeenThere: intel-xe@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Intel Xe graphics driver List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-xe-bounces@lists.freedesktop.org Sender: "Intel-xe" On 17.08.2024 12:35, Rodrigo Vivi wrote: > In some rare cases, the drm_mm node cannot be removed synchronously > due to runtime PM conditions. In this situation, the node removal will > be delegated to a workqueue that will be able to wake up the device > before removing the node. > > However, in this situation, the lifetime of the xe_ggtt_node cannot > be restricted to the lifetime of the parent object. So, this patch > introduces the infrastructure so the xe_ggtt_node struct can be > allocated in advance and freed when needed. > > By having the ggtt backpointer, it also ensure that the init function > is always called before any attempt to insert or reserve the node > in the GGTT. > > v2: s/xe_ggtt_node_force_fini/xe_ggtt_node_fini and use it > internaly (Brost) > v3: - Use GF_NOFS for node allocation (CI) > - Avoid ggtt argument, now that we have it inside the node (Lucas) > - Fix some missed fini cases (CI) > > Cc: Matthew Auld > Cc: Michal Wajdeczko > Cc: Matthew Brost > Reviewed-by: Matthew Brost #2 > Signed-off-by: Rodrigo Vivi > --- > .../gpu/drm/xe/compat-i915-headers/i915_vma.h | 4 +- > drivers/gpu/drm/xe/display/xe_fb_pin.c | 36 +++-- > drivers/gpu/drm/xe/xe_bo.c | 2 +- > drivers/gpu/drm/xe/xe_bo.h | 9 +- > drivers/gpu/drm/xe/xe_bo_types.h | 2 +- > drivers/gpu/drm/xe/xe_device_types.h | 2 +- > drivers/gpu/drm/xe/xe_ggtt.c | 145 ++++++++++++++---- > drivers/gpu/drm/xe/xe_ggtt.h | 14 +- > drivers/gpu/drm/xe/xe_ggtt_types.h | 8 +- > drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c | 41 +++-- > .../gpu/drm/xe/xe_gt_sriov_pf_config_types.h | 2 +- > drivers/gpu/drm/xe/xe_gt_sriov_vf.c | 27 +++- > 12 files changed, 209 insertions(+), 83 deletions(-) > > diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h > index 3028ac1ba72f..bdae8392e125 100644 > --- a/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h > +++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h > @@ -20,7 +20,7 @@ struct xe_bo; > > struct i915_vma { > struct xe_bo *bo, *dpt; > - struct xe_ggtt_node node; > + struct xe_ggtt_node *node; > }; > > #define i915_ggtt_clear_scanout(bo) do { } while (0) > @@ -29,7 +29,7 @@ struct i915_vma { > > static inline u32 i915_ggtt_offset(const struct i915_vma *vma) > { > - return vma->node.base.start; > + return vma->node->base.start; > } > > #endif > diff --git a/drivers/gpu/drm/xe/display/xe_fb_pin.c b/drivers/gpu/drm/xe/display/xe_fb_pin.c > index de4930b67a29..f7a9649745c2 100644 > --- a/drivers/gpu/drm/xe/display/xe_fb_pin.c > +++ b/drivers/gpu/drm/xe/display/xe_fb_pin.c > @@ -204,20 +204,28 @@ static int __xe_pin_fb_vma_ggtt(const struct intel_framebuffer *fb, > if (xe_bo_is_vram(bo) && ggtt->flags & XE_GGTT_FLAGS_64K) > align = max_t(u32, align, SZ_64K); > > - if (bo->ggtt_node.base.size && view->type == I915_GTT_VIEW_NORMAL) { > + if (bo->ggtt_node && view->type == I915_GTT_VIEW_NORMAL) { > vma->node = bo->ggtt_node; > } else if (view->type == I915_GTT_VIEW_NORMAL) { > u32 x, size = bo->ttm.base.size; > > - ret = xe_ggtt_node_insert_locked(ggtt, &vma->node, size, align, 0); > - if (ret) > + vma->node = xe_ggtt_node_init(ggtt); > + if (IS_ERR(vma->node)) { > + ret = PTR_ERR(vma->node); > goto out_unlock; > + } > + > + ret = xe_ggtt_node_insert_locked(vma->node, size, align, 0); > + if (ret) { > + xe_ggtt_node_fini(vma->node); > + goto out_unlock; > + } > > for (x = 0; x < size; x += XE_PAGE_SIZE) { > u64 pte = ggtt->pt_ops->pte_encode_bo(bo, x, > xe->pat.idx[XE_CACHE_NONE]); > > - ggtt->pt_ops->ggtt_set_pte(ggtt, vma->node.base.start + x, pte); > + ggtt->pt_ops->ggtt_set_pte(ggtt, vma->node->base.start + x, pte); > } > } else { > u32 i, ggtt_ofs; > @@ -226,11 +234,19 @@ static int __xe_pin_fb_vma_ggtt(const struct intel_framebuffer *fb, > /* display seems to use tiles instead of bytes here, so convert it back.. */ > u32 size = intel_rotation_info_size(rot_info) * XE_PAGE_SIZE; > > - ret = xe_ggtt_node_insert_locked(ggtt, &vma->node, size, align, 0); > - if (ret) > + vma->node = xe_ggtt_node_init(ggtt); > + if (IS_ERR(vma->node)) { > + ret = PTR_ERR(vma->node); > goto out_unlock; > + } > + > + ret = xe_ggtt_node_insert_locked(vma->node, size, align, 0); > + if (ret) { > + xe_ggtt_node_fini(vma->node); > + goto out_unlock; > + } > > - ggtt_ofs = vma->node.base.start; > + ggtt_ofs = vma->node->base.start; > > for (i = 0; i < ARRAY_SIZE(rot_info->plane); i++) > write_ggtt_rotated(bo, ggtt, &ggtt_ofs, > @@ -323,9 +339,9 @@ static void __xe_unpin_fb_vma(struct i915_vma *vma) > > if (vma->dpt) > xe_bo_unpin_map_no_vm(vma->dpt); > - else if (!xe_ggtt_node_allocated(&vma->bo->ggtt_node) || > - vma->bo->ggtt_node.base.start != vma->node.base.start) > - xe_ggtt_node_remove(ggtt, &vma->node, false); > + else if (!xe_ggtt_node_allocated(vma->bo->ggtt_node) || > + vma->bo->ggtt_node->base.start != vma->node->base.start) > + xe_ggtt_node_remove(ggtt, vma->node, false); > > ttm_bo_reserve(&vma->bo->ttm, false, false, NULL); > ttm_bo_unpin(&vma->bo->ttm); > diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c > index ae8a786f5d65..f89a382124ca 100644 > --- a/drivers/gpu/drm/xe/xe_bo.c > +++ b/drivers/gpu/drm/xe/xe_bo.c > @@ -1098,7 +1098,7 @@ static void xe_ttm_bo_destroy(struct ttm_buffer_object *ttm_bo) > > xe_assert(xe, list_empty(&ttm_bo->base.gpuva.list)); > > - if (bo->ggtt_node.base.size) > + if (bo->ggtt_node && bo->ggtt_node->base.size) if (xe_ggtt_node_allocated(bo->ggtt_node)) ? > xe_ggtt_remove_bo(bo->tile->mem.ggtt, bo); > > #ifdef CONFIG_PROC_FS > diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h > index faffbda55517..f588e3a98105 100644 > --- a/drivers/gpu/drm/xe/xe_bo.h > +++ b/drivers/gpu/drm/xe/xe_bo.h > @@ -195,9 +195,12 @@ xe_bo_main_addr(struct xe_bo *bo, size_t page_size) > static inline u32 > xe_bo_ggtt_addr(struct xe_bo *bo) > { > - XE_WARN_ON(bo->ggtt_node.base.size > bo->size); > - XE_WARN_ON(bo->ggtt_node.base.start + bo->ggtt_node.base.size > (1ull << 32)); > - return bo->ggtt_node.base.start; > + if (XE_WARN_ON(!bo->ggtt_node)) > + return -ENOENT; -ENOENT is still a non zero u32, maybe returning 0 is a better choice ? > + > + XE_WARN_ON(bo->ggtt_node->base.size > bo->size); > + XE_WARN_ON(bo->ggtt_node->base.start + bo->ggtt_node->base.size > (1ull << 32)); the bo has pointer to a tile, so maybe instead of XE_WARN_ON we can use here our assert: xe_tile_assert(bo->tile, ....) ? > + return bo->ggtt_node->base.start; > } > > int xe_bo_vmap(struct xe_bo *bo); > diff --git a/drivers/gpu/drm/xe/xe_bo_types.h b/drivers/gpu/drm/xe/xe_bo_types.h > index 4b1de9f5be00..2ed558ac2264 100644 > --- a/drivers/gpu/drm/xe/xe_bo_types.h > +++ b/drivers/gpu/drm/xe/xe_bo_types.h > @@ -40,7 +40,7 @@ struct xe_bo { > /** @placement: current placement for this BO */ > struct ttm_placement placement; > /** @ggtt_node: GGTT node if this BO is mapped in the GGTT */ > - struct xe_ggtt_node ggtt_node; > + struct xe_ggtt_node *ggtt_node; > /** @vmap: iosys map of this buffer */ > struct iosys_map vmap; > /** @ttm_kmap: TTM bo kmap object for internal use only. Keep off. */ > diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h > index d2b3d8a0c1bd..06fcb92568c7 100644 > --- a/drivers/gpu/drm/xe/xe_device_types.h > +++ b/drivers/gpu/drm/xe/xe_device_types.h > @@ -204,7 +204,7 @@ struct xe_tile { > struct xe_memirq memirq; > > /** @sriov.vf.ggtt_balloon: GGTT regions excluded from use. */ > - struct xe_ggtt_node ggtt_balloon[2]; > + struct xe_ggtt_node *ggtt_balloon[2]; > } vf; > } sriov; > > diff --git a/drivers/gpu/drm/xe/xe_ggtt.c b/drivers/gpu/drm/xe/xe_ggtt.c > index a359d94a979e..8ea223301ab9 100644 > --- a/drivers/gpu/drm/xe/xe_ggtt.c > +++ b/drivers/gpu/drm/xe/xe_ggtt.c > @@ -348,20 +348,25 @@ static void xe_ggtt_dump_node(struct xe_ggtt *ggtt, > > /** > * xe_ggtt_node_balloon - prevent allocation of specified GGTT addresses > - * @ggtt: the &xe_ggtt where we want to make reservation > - > * @node: the &xe_ggtt_node to hold reserved GGTT node > * @start: the starting GGTT address of the reserved region > * @end: then end GGTT address of the reserved region > * > + * It cannot be called without first having called xe_ggtt_init(). > * Use xe_ggtt_node_deballoon() to release a reserved GGTT node. > * > * Return: 0 on success or a negative error code on failure. > */ > -int xe_ggtt_node_balloon(struct xe_ggtt *ggtt, struct xe_ggtt_node *node, u64 start, u64 end) > +int xe_ggtt_node_balloon(struct xe_ggtt_node *node, u64 start, u64 end) hmm, we want to 'balloon' the GGTT, not the 'node' (which is just an implementation detail) so maybe moving the balloon nodes to the xe_ggtt should be done in this series before this patch ? > { > + struct xe_ggtt *ggtt; > int err; > > + if (!node || !node->ggtt) > + return -ENOENT; hmm, NULL or uninitialized node is a bug on our side, so IMO we shouldn't do runtime checks for it > + > + ggtt = node->ggtt; > + > xe_tile_assert(ggtt->tile, start < end); > xe_tile_assert(ggtt->tile, IS_ALIGNED(start, XE_PAGE_SIZE)); > xe_tile_assert(ggtt->tile, IS_ALIGNED(end, XE_PAGE_SIZE)); > @@ -386,64 +391,112 @@ int xe_ggtt_node_balloon(struct xe_ggtt *ggtt, struct xe_ggtt_node *node, u64 st > > /** > * xe_ggtt_node_deballoon - release a reserved GGTT region > - * @ggtt: the &xe_ggtt where reserved node belongs > * @node: the &xe_ggtt_node with reserved GGTT region > * > * See xe_ggtt_node_balloon() for details. > */ > -void xe_ggtt_node_deballoon(struct xe_ggtt *ggtt, struct xe_ggtt_node *node) > +void xe_ggtt_node_deballoon(struct xe_ggtt_node *node) > { > - if (!drm_mm_node_allocated(&node->base)) > + if (!node || !node->ggtt) > return; > > - xe_ggtt_dump_node(ggtt, &node->base, "deballoon"); > + if (!drm_mm_node_allocated(&node->base)) > + goto free_node; > > - mutex_lock(&ggtt->lock); > + xe_ggtt_dump_node(node->ggtt, &node->base, "deballoon"); > + > + mutex_lock(&node->ggtt->lock); > drm_mm_remove_node(&node->base); > - mutex_unlock(&ggtt->lock); > + mutex_unlock(&node->ggtt->lock); > + > +free_node: > + xe_ggtt_node_fini(node); > } > > /** > * xe_ggtt_node_insert_locked - Locked version to insert a &xe_ggtt_node into the GGTT > - * @ggtt: the &xe_ggtt where node will be inserted > * @node: the &xe_ggtt_node to be inserted > * @size: size of the node > * @align: alignment constrain of the node > * @mm_flags: flags to control the node behavior > * > + * It cannot be called without first having called xe_ggtt_init() once. > * To be used in cases where ggtt->lock is already taken. > * > * Return: 0 on success or a negative error code on failure. > */ > -int xe_ggtt_node_insert_locked(struct xe_ggtt *ggtt, struct xe_ggtt_node *node, > +int xe_ggtt_node_insert_locked(struct xe_ggtt_node *node, > u32 size, u32 align, u32 mm_flags) > { > - return drm_mm_insert_node_generic(&ggtt->mm, &node->base, size, align, 0, > + if (!node || !node->ggtt) > + return -ENOENT; ditto > + > + return drm_mm_insert_node_generic(&node->ggtt->mm, &node->base, size, align, 0, > mm_flags); > } > > /** > * xe_ggtt_node_insert - Insert a &xe_ggtt_node into the GGTT > - * @ggtt: the &xe_ggtt where node will be inserted > * @node: the &xe_ggtt_node to be inserted > * @size: size of the node > * @align: alignment constrain of the node > * > + * It cannot be called without first having called xe_ggtt_init() once. > + * > * Return: 0 on success or a negative error code on failure. > */ > -int xe_ggtt_node_insert(struct xe_ggtt *ggtt, struct xe_ggtt_node *node, > - u32 size, u32 align) > +int xe_ggtt_node_insert(struct xe_ggtt_node *node, u32 size, u32 align) > { > int ret; > > - mutex_lock(&ggtt->lock); > - ret = xe_ggtt_node_insert_locked(ggtt, node, size, > - align, DRM_MM_INSERT_HIGH); > - mutex_unlock(&ggtt->lock); > + if (!node || !node->ggtt) > + return -ENOENT; > + > + mutex_lock(&node->ggtt->lock); > + ret = xe_ggtt_node_insert_locked(node, size, align, > + DRM_MM_INSERT_HIGH); > + mutex_unlock(&node->ggtt->lock); > > return ret; > } > > +/** > + * xe_ggtt_node_init - Initialize %xe_ggtt_node struct > + * @ggtt: the &xe_ggtt where the new node will later be inserted/reserved. > + * > + * This function will allocated the struct %xe_ggtt_node and return it's pointer. > + * This struct will then be freed after the node removal upon xe_ggtt_node_remove() > + * or xe_ggtt_node_deballoon(). > + * Having %xe_ggtt_node struct allocated doesn't mean that the node is already allocated > + * in GGTT. Only the xe_ggtt_node_insert(), xe_ggtt_node_insert_locked(), > + * xe_ggtt_node_balloon() will ensure the node is inserted or reserved in GGTT. > + * > + * Return: A pointer to %xe_ggtt_node struct on success. An ERR_PTR otherwise. > + **/ > +struct xe_ggtt_node *xe_ggtt_node_init(struct xe_ggtt *ggtt) > +{ > + struct xe_ggtt_node *node = kzalloc(sizeof(*node), GFP_NOFS); > + > + if (!node) > + return ERR_PTR(-ENOMEM); > + > + node->ggtt = ggtt; > + return node; > +} > + > +/** > + * xe_ggtt_node_fini - Forcebly finalize %xe_ggtt_node struct > + * @node: the &xe_ggtt_node to be freed > + * > + * If anything went wrong with either xe_ggtt_node_insert(), xe_ggtt_node_insert_locked(), > + * or xe_ggtt_node_balloon(); and this @node is not going to be reused, then, > + * this function needs to be called to free the %xe_ggtt_node struct > + **/ > +void xe_ggtt_node_fini(struct xe_ggtt_node *node) > +{ maybe we should assert that node is not allocated ? > + kfree(node); > +} > + > /** > * xe_ggtt_node_remove - Remove a &xe_ggtt_node from the GGTT > * @ggtt: the &xe_ggtt where node will be removed > @@ -457,6 +510,9 @@ void xe_ggtt_node_remove(struct xe_ggtt *ggtt, struct xe_ggtt_node *node, ggtt is redundant here as it is the same as node->ggtt > bool bound; > int idx; > > + if (!node || !node->ggtt) > + return; > + > bound = drm_dev_enter(&xe->drm, &idx); > if (bound) > xe_pm_runtime_get_noresume(xe); > @@ -469,23 +525,29 @@ void xe_ggtt_node_remove(struct xe_ggtt *ggtt, struct xe_ggtt_node *node, > mutex_unlock(&ggtt->lock); > > if (!bound) > - return; > + goto free_node; > > if (invalidate) > xe_ggtt_invalidate(ggtt); > > xe_pm_runtime_put(xe); > drm_dev_exit(idx); > + > +free_node: > + xe_ggtt_node_fini(node); > } > > /** > - * xe_ggtt_node_allocated - Check if node is allocated > + * xe_ggtt_node_allocated - Check if node is allocated in GGTT > * @node: the &xe_ggtt_node to be inspected > * > * Return: True if allocated, False otherwise. > */ > bool xe_ggtt_node_allocated(const struct xe_ggtt_node *node) > { > + if (!node || !node->ggtt) > + return false; > + > return drm_mm_node_allocated(&node->base); > } > > @@ -498,9 +560,14 @@ void xe_ggtt_map_bo(struct xe_ggtt *ggtt, struct xe_bo *bo) > { > u16 cache_mode = bo->flags & XE_BO_FLAG_NEEDS_UC ? XE_CACHE_NONE : XE_CACHE_WB; > u16 pat_index = tile_to_xe(ggtt->tile)->pat.idx[cache_mode]; > - u64 start = bo->ggtt_node.base.start; > + u64 start; > u64 offset, pte; > > + if (XE_WARN_ON(!bo->ggtt_node)) > + return; > + > + start = bo->ggtt_node->base.start; > + > for (offset = 0; offset < bo->size; offset += XE_PAGE_SIZE) { > pte = ggtt->pt_ops->pte_encode_bo(bo, offset, pat_index); > ggtt->pt_ops->ggtt_set_pte(ggtt, start + offset, pte); > @@ -516,9 +583,9 @@ static int __xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo, > if (xe_bo_is_vram(bo) && ggtt->flags & XE_GGTT_FLAGS_64K) > alignment = SZ_64K; > > - if (XE_WARN_ON(bo->ggtt_node.base.size)) { > + if (XE_WARN_ON(bo->ggtt_node)) { > /* Someone's already inserted this BO in the GGTT */ > - xe_tile_assert(ggtt->tile, bo->ggtt_node.base.size == bo->size); > + xe_tile_assert(ggtt->tile, bo->ggtt_node->base.size == bo->size); > return 0; > } > > @@ -527,15 +594,26 @@ static int __xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo, > return err; > > xe_pm_runtime_get_noresume(tile_to_xe(ggtt->tile)); > + > + bo->ggtt_node = xe_ggtt_node_init(ggtt); > + if (IS_ERR(bo->ggtt_node)) { > + err = PTR_ERR(bo->ggtt_node); > + goto out; > + } > + > mutex_lock(&ggtt->lock); > - err = drm_mm_insert_node_in_range(&ggtt->mm, &bo->ggtt_node.base, bo->size, > + err = drm_mm_insert_node_in_range(&ggtt->mm, &bo->ggtt_node->base, bo->size, > alignment, 0, start, end, 0); > - if (!err) > + if (err) > + xe_ggtt_node_fini(bo->ggtt_node); > + else > xe_ggtt_map_bo(ggtt, bo); > mutex_unlock(&ggtt->lock); > > if (!err && bo->flags & XE_BO_FLAG_GGTT_INVALIDATE) > xe_ggtt_invalidate(ggtt); > + > +out: > xe_pm_runtime_put(tile_to_xe(ggtt->tile)); > > return err; > @@ -575,13 +653,13 @@ int xe_ggtt_insert_bo(struct xe_ggtt *ggtt, struct xe_bo *bo) > */ > void xe_ggtt_remove_bo(struct xe_ggtt *ggtt, struct xe_bo *bo) > { > - if (XE_WARN_ON(!bo->ggtt_node.base.size)) > + if (XE_WARN_ON(!bo->ggtt_node)) > return; > > /* This BO is not currently in the GGTT */ > - xe_tile_assert(ggtt->tile, bo->ggtt_node.base.size == bo->size); > + xe_tile_assert(ggtt->tile, bo->ggtt_node->base.size == bo->size); > > - xe_ggtt_node_remove(ggtt, &bo->ggtt_node, > + xe_ggtt_node_remove(ggtt, bo->ggtt_node, > bo->flags & XE_BO_FLAG_GGTT_INVALIDATE); > } > > @@ -648,7 +726,6 @@ static void xe_ggtt_assign_locked(struct xe_ggtt *ggtt, const struct drm_mm_node > > /** > * xe_ggtt_assign - assign a GGTT region to the VF > - * @ggtt: the &xe_ggtt where the node belongs > * @node: the &xe_ggtt_node to update > * @vfid: the VF identifier > * > @@ -656,11 +733,11 @@ static void xe_ggtt_assign_locked(struct xe_ggtt *ggtt, const struct drm_mm_node > * In addition to PTE's VFID bits 11:2 also PRESENT bit 0 is set as on some > * platforms VFs can't modify that either. > */ > -void xe_ggtt_assign(struct xe_ggtt *ggtt, const struct xe_ggtt_node *node, u16 vfid) > +void xe_ggtt_assign(const struct xe_ggtt_node *node, u16 vfid) it probable should be named as: xe_ggtt_node_assign() > { > - mutex_lock(&ggtt->lock); > - xe_ggtt_assign_locked(ggtt, &node->base, vfid); > - mutex_unlock(&ggtt->lock); > + mutex_lock(&node->ggtt->lock); > + xe_ggtt_assign_locked(node->ggtt, &node->base, vfid); > + mutex_unlock(&node->ggtt->lock); > } > #endif > > diff --git a/drivers/gpu/drm/xe/xe_ggtt.h b/drivers/gpu/drm/xe/xe_ggtt.h > index d61f4f976a5a..96f79a3c8aa6 100644 > --- a/drivers/gpu/drm/xe/xe_ggtt.h > +++ b/drivers/gpu/drm/xe/xe_ggtt.h > @@ -13,13 +13,13 @@ struct drm_printer; > int xe_ggtt_init_early(struct xe_ggtt *ggtt); > int xe_ggtt_init(struct xe_ggtt *ggtt); > > -int xe_ggtt_node_balloon(struct xe_ggtt *ggtt, struct xe_ggtt_node *node, u64 start, u64 size); > -void xe_ggtt_node_deballoon(struct xe_ggtt *ggtt, struct xe_ggtt_node *node); > +struct xe_ggtt_node *xe_ggtt_node_init(struct xe_ggtt *ggtt); > +void xe_ggtt_node_fini(struct xe_ggtt_node *node); > +int xe_ggtt_node_balloon(struct xe_ggtt_node *node, u64 start, u64 size); > +void xe_ggtt_node_deballoon(struct xe_ggtt_node *node); > > -int xe_ggtt_node_insert(struct xe_ggtt *ggtt, struct xe_ggtt_node *node, > - u32 size, u32 align); > -int xe_ggtt_node_insert_locked(struct xe_ggtt *ggtt, > - struct xe_ggtt_node *node, > +int xe_ggtt_node_insert(struct xe_ggtt_node *node, u32 size, u32 align); > +int xe_ggtt_node_insert_locked(struct xe_ggtt_node *node, > u32 size, u32 align, u32 mm_flags); > void xe_ggtt_node_remove(struct xe_ggtt *ggtt, struct xe_ggtt_node *node, > bool invalidate); > @@ -35,7 +35,7 @@ int xe_ggtt_dump(struct xe_ggtt *ggtt, struct drm_printer *p); > u64 xe_ggtt_print_holes(struct xe_ggtt *ggtt, u64 alignment, struct drm_printer *p); > > #ifdef CONFIG_PCI_IOV > -void xe_ggtt_assign(struct xe_ggtt *ggtt, const struct xe_ggtt_node *node, u16 vfid); > +void xe_ggtt_assign(const struct xe_ggtt_node *node, u16 vfid); > #endif > > #endif > diff --git a/drivers/gpu/drm/xe/xe_ggtt_types.h b/drivers/gpu/drm/xe/xe_ggtt_types.h > index af312a7d1031..0e8822ae13fc 100644 > --- a/drivers/gpu/drm/xe/xe_ggtt_types.h > +++ b/drivers/gpu/drm/xe/xe_ggtt_types.h > @@ -50,9 +50,15 @@ struct xe_ggtt { > }; > > /** > - * struct xe_ggtt_node - A node in GGTT > + * struct xe_ggtt_node - A node in GGTT. > + * > + * This struct needs to be initialized (only-once) with xe_ggtt_node_init() before any node > + * insertion, reservation, or 'ballooning'. > + * It will, then, be finalized by either xe_ggtt_node_remove() or xe_ggtt_node_deballoon(). > */ > struct xe_ggtt_node { > + /** @ggtt: Back pointer to xe_ggtt where this region will be inserted at */ > + struct xe_ggtt *ggtt; > /** @base: A drm_mm_node */ > struct drm_mm_node base; > }; > diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c > index e133594cc6bd..c0313d13c27b 100644 > --- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c > +++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c > @@ -232,14 +232,14 @@ static u32 encode_config_ggtt(u32 *cfg, const struct xe_gt_sriov_config *config) > { > u32 n = 0; > > - if (xe_ggtt_node_allocated(&config->ggtt_region)) { > + if (xe_ggtt_node_allocated(config->ggtt_region)) { > cfg[n++] = PREP_GUC_KLV_TAG(VF_CFG_GGTT_START); > - cfg[n++] = lower_32_bits(config->ggtt_region.base.start); > - cfg[n++] = upper_32_bits(config->ggtt_region.base.start); > + cfg[n++] = lower_32_bits(config->ggtt_region->base.start); > + cfg[n++] = upper_32_bits(config->ggtt_region->base.start); > > cfg[n++] = PREP_GUC_KLV_TAG(VF_CFG_GGTT_SIZE); > - cfg[n++] = lower_32_bits(config->ggtt_region.base.size); > - cfg[n++] = upper_32_bits(config->ggtt_region.base.size); > + cfg[n++] = lower_32_bits(config->ggtt_region->base.size); > + cfg[n++] = upper_32_bits(config->ggtt_region->base.size); > } > > return n; > @@ -385,13 +385,13 @@ static void pf_release_ggtt(struct xe_tile *tile, struct xe_ggtt_node *node) > > static void pf_release_vf_config_ggtt(struct xe_gt *gt, struct xe_gt_sriov_config *config) > { > - pf_release_ggtt(gt_to_tile(gt), &config->ggtt_region); > + pf_release_ggtt(gt_to_tile(gt), config->ggtt_region); in config->ggtt_region here we may still keep stale pointer to the node that will be already or soon released > } > > static int pf_provision_vf_ggtt(struct xe_gt *gt, unsigned int vfid, u64 size) > { > struct xe_gt_sriov_config *config = pf_pick_vf_config(gt, vfid); > - struct xe_ggtt_node *node = &config->ggtt_region; > + struct xe_ggtt_node *node = config->ggtt_region; here 'node' a copy of 'config->ggtt_region' > struct xe_tile *tile = gt_to_tile(gt); > struct xe_ggtt *ggtt = tile->mem.ggtt; > u64 alignment = pf_get_ggtt_alignment(gt); > @@ -415,25 +415,32 @@ static int pf_provision_vf_ggtt(struct xe_gt *gt, unsigned int vfid, u64 size) > if (!size) > return 0; > > - err = xe_ggtt_node_insert(ggtt, node, size, alignment); > + node = xe_ggtt_node_init(ggtt); but now 'node' is overwritten with new object > + if (IS_ERR(node)) > + return PTR_ERR(node); > + > + err = xe_ggtt_node_insert(node, size, alignment); > if (unlikely(err)) > - return err; > + goto err; > > - xe_ggtt_assign(ggtt, node, vfid); > + xe_ggtt_assign(node, vfid); > xe_gt_sriov_dbg_verbose(gt, "VF%u assigned GGTT %llx-%llx\n", > vfid, node->base.start, node->base.start + node->base.size - 1); > > err = pf_distribute_config_ggtt(gt->tile, vfid, node->base.start, node->base.size); > if (unlikely(err)) > - return err; > + goto err; and this new 'node' is never stored anywhere and thus released, which leads to ENOSPC when trying to enable/disable VFs few times > > return 0; > +err: > + xe_ggtt_node_fini(node); > + return err; > } > > static u64 pf_get_vf_config_ggtt(struct xe_gt *gt, unsigned int vfid) > { > struct xe_gt_sriov_config *config = pf_pick_vf_config(gt, vfid); > - struct xe_ggtt_node *node = &config->ggtt_region; > + struct xe_ggtt_node *node = config->ggtt_region; > > xe_gt_assert(gt, !xe_gt_is_media_type(gt)); > return xe_ggtt_node_allocated(node) ? node->base.size : 0; > @@ -2006,13 +2013,15 @@ int xe_gt_sriov_pf_config_print_ggtt(struct xe_gt *gt, struct drm_printer *p) > > for (n = 1; n <= total_vfs; n++) { > config = >->sriov.pf.vfs[n].config; > - if (!xe_ggtt_node_allocated(&config->ggtt_region)) > + if (!xe_ggtt_node_allocated(config->ggtt_region)) > continue; > > - string_get_size(config->ggtt_region.base.size, 1, STRING_UNITS_2, buf, sizeof(buf)); > + string_get_size(config->ggtt_region->base.size, 1, STRING_UNITS_2, > + buf, sizeof(buf)); > drm_printf(p, "VF%u:\t%#0llx-%#llx\t(%s)\n", > - n, config->ggtt_region.base.start, > - config->ggtt_region.base.start + config->ggtt_region.base.size - 1, buf); > + n, config->ggtt_region->base.start, > + config->ggtt_region->base.start + config->ggtt_region->base.size - 1, > + buf); > } > > return 0; > diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h > index a73d9a4b9e64..2d3b73d78f14 100644 > --- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h > +++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h > @@ -18,7 +18,7 @@ struct xe_bo; > */ > struct xe_gt_sriov_config { > /** @ggtt_region: GGTT region assigned to the VF. */ > - struct xe_ggtt_node ggtt_region; > + struct xe_ggtt_node *ggtt_region; > /** @lmem_obj: LMEM allocation for use by the VF. */ > struct xe_bo *lmem_obj; > /** @num_ctxs: number of GuC contexts IDs. */ > diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_vf.c b/drivers/gpu/drm/xe/xe_gt_sriov_vf.c > index db9761a68ce0..6ea5fa016a82 100644 > --- a/drivers/gpu/drm/xe/xe_gt_sriov_vf.c > +++ b/drivers/gpu/drm/xe/xe_gt_sriov_vf.c > @@ -495,6 +495,22 @@ u64 xe_gt_sriov_vf_lmem(struct xe_gt *gt) > return gt->sriov.vf.self_config.lmem_size; > } > > +static int vf_balloon_ggtt_node(struct xe_ggtt *ggtt, struct xe_ggtt_node *node, > + u64 start, u64 end) > +{ > + int err; > + > + node = xe_ggtt_node_init(ggtt); > + if (IS_ERR(node)) > + return PTR_ERR(node); > + > + err = xe_ggtt_node_balloon(node, start, end); > + if (err) > + xe_ggtt_node_fini(node); > + > + return err; > +} > + > static int vf_balloon_ggtt(struct xe_gt *gt) > { > struct xe_gt_sriov_vf_selfconfig *config = >->sriov.vf.self_config; > @@ -528,7 +544,7 @@ static int vf_balloon_ggtt(struct xe_gt *gt) > start = xe_wopcm_size(xe); > end = config->ggtt_base; > if (end != start) { > - err = xe_ggtt_node_balloon(ggtt, &tile->sriov.vf.ggtt_balloon[0], start, end); > + err = vf_balloon_ggtt_node(ggtt, tile->sriov.vf.ggtt_balloon[0], start, end); > if (err) > goto failed; > } > @@ -536,7 +552,7 @@ static int vf_balloon_ggtt(struct xe_gt *gt) > start = config->ggtt_base + config->ggtt_size; > end = GUC_GGTT_TOP; > if (end != start) { > - err = xe_ggtt_node_balloon(ggtt, &tile->sriov.vf.ggtt_balloon[1], start, end); > + err = vf_balloon_ggtt_node(ggtt, tile->sriov.vf.ggtt_balloon[1], start, end); > if (err) > goto deballoon; > } > @@ -544,7 +560,7 @@ static int vf_balloon_ggtt(struct xe_gt *gt) > return 0; > > deballoon: > - xe_ggtt_node_deballoon(ggtt, &tile->sriov.vf.ggtt_balloon[0]); > + xe_ggtt_node_deballoon(tile->sriov.vf.ggtt_balloon[0]); > failed: > return err; > } > @@ -552,11 +568,10 @@ static int vf_balloon_ggtt(struct xe_gt *gt) > static void deballoon_ggtt(struct drm_device *drm, void *arg) > { > struct xe_tile *tile = arg; > - struct xe_ggtt *ggtt = tile->mem.ggtt; > > xe_tile_assert(tile, IS_SRIOV_VF(tile_to_xe(tile))); > - xe_ggtt_node_deballoon(ggtt, &tile->sriov.vf.ggtt_balloon[1]); > - xe_ggtt_node_deballoon(ggtt, &tile->sriov.vf.ggtt_balloon[0]); > + xe_ggtt_node_deballoon(tile->sriov.vf.ggtt_balloon[1]); > + xe_ggtt_node_deballoon(tile->sriov.vf.ggtt_balloon[0]); > } > > /**