[Intel-gfx,v6,08/17] drm/ttm: use gem vma_node

Submitted by Thierry Reding on Aug. 14, 2019, 9:35 a.m.

Details

Message ID 20190814093524.GA31345@ulmo
State New
Headers show
Series "Series without cover letter" ( rev: 3 2 ) in Nouveau

Not browsing as part of any series.

Commit Message

Thierry Reding Aug. 14, 2019, 9:35 a.m.
On Wed, Aug 14, 2019 at 07:58:27AM +0200, Gerd Hoffmann wrote:
> > Hi Gerd,
> > 
> > I've been seeing a regression on Nouveau with recent linux-next releases
> > and git bisect points at this commit as the first bad one. If I revert
> > it (there's a tiny conflict with a patch that was merged subsequently),
> > things are back to normal.
> > 
> > I think the reason for this issue is that Nouveau doesn't use GEM
> > objects for all buffer objects,
> 
> That shouldn't be a problem ...
> 
> > and even when it uses GEM objects, the
> > code will not initialize the GEM object until after the buffer objects
> > and the backing TTM objects have been created.
> 
> ... but the initialization order is.
> 
> ttm_bo_uses_embedded_gem_object() assumes gem gets initialized first.
> 
> drm_gem_object_init() init calling drm_vma_node_reset() again is
> probably the root cause for the breakage.
> 
> > I tried to fix that by making sure drm_gem_object_init() gets called by
> > Nouveau before ttm_bo_init(), but the changes are fairly involved and I
> > was unable to get the GEM reference counting right. I can look into the
> > proper fix some more, but it might be worth reverting this patch for
> > now to get Nouveau working again.
> 
> Changing the order doesn't look hard.  Patch attached (untested, have no
> test hardware).  But maybe I missed some detail ...
> 
> The other patch attached works around the issue with a flag, to avoid
> drm_vma_node_reset() being called twice.

I came up with something very similar by splitting up nouveau_bo_new()
into allocation and initialization steps, so that when necessary the GEM
object can be initialized in between. I think that's slightly more
flexible and easier to understand than a boolean flag.

Thierry
From a1130a6affcb7c00133e89f3e498cb6757f5bb51 Mon Sep 17 00:00:00 2001
From: Thierry Reding <treding@nvidia.com>
Date: Wed, 14 Aug 2019 11:00:48 +0200
Subject: [PATCH] drm/nouveau: Initialize GEM object before TTM object

TTM assumes that drivers initialize the embedded GEM object before
calling the ttm_bo_init() function. This is not currently the case
in the Nouveau driver. Fix this by splitting up nouveau_bo_new()
into nouveau_bo_alloc() and nouveau_bo_init() so that the GEM can
be initialized before TTM BO initialization when necessary.

Fixes: b96f3e7c8069 ("drm/ttm: use gem vma_node")
Signed-off-by: Thierry Reding <treding@nvidia.com>
---
 drivers/gpu/drm/nouveau/nouveau_bo.c    | 69 ++++++++++++++++---------
 drivers/gpu/drm/nouveau/nouveau_bo.h    |  4 ++
 drivers/gpu/drm/nouveau/nouveau_gem.c   | 29 ++++++-----
 drivers/gpu/drm/nouveau/nouveau_prime.c | 16 ++++--
 4 files changed, 77 insertions(+), 41 deletions(-)

Patch hide | download patch | download mbox

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 99e391be9370..b3d3e07de1af 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -185,31 +185,24 @@  nouveau_bo_fixup_align(struct nouveau_bo *nvbo, u32 flags,
 	*size = roundup_64(*size, PAGE_SIZE);
 }
 
-int
-nouveau_bo_new(struct nouveau_cli *cli, u64 size, int align,
-	       uint32_t flags, uint32_t tile_mode, uint32_t tile_flags,
-	       struct sg_table *sg, struct reservation_object *robj,
-	       struct nouveau_bo **pnvbo)
+struct nouveau_bo *
+nouveau_bo_alloc(struct nouveau_cli *cli, u64 size, u32 flags, u32 tile_mode,
+		 u32 tile_flags)
 {
 	struct nouveau_drm *drm = cli->drm;
 	struct nouveau_bo *nvbo;
 	struct nvif_mmu *mmu = &cli->mmu;
 	struct nvif_vmm *vmm = cli->svm.cli ? &cli->svm.vmm : &cli->vmm.vmm;
-	size_t acc_size;
-	int type = ttm_bo_type_device;
-	int ret, i, pi = -1;
+	int i, pi = -1;
 
 	if (!size) {
 		NV_WARN(drm, "skipped size %016llx\n", size);
-		return -EINVAL;
+		return ERR_PTR(-EINVAL);
 	}
 
-	if (sg)
-		type = ttm_bo_type_sg;
-
 	nvbo = kzalloc(sizeof(struct nouveau_bo), GFP_KERNEL);
 	if (!nvbo)
-		return -ENOMEM;
+		return ERR_PTR(-ENOMEM);
 	INIT_LIST_HEAD(&nvbo->head);
 	INIT_LIST_HEAD(&nvbo->entry);
 	INIT_LIST_HEAD(&nvbo->vma_list);
@@ -231,7 +224,7 @@  nouveau_bo_new(struct nouveau_cli *cli, u64 size, int align,
 		nvbo->kind = (tile_flags & 0x0000ff00) >> 8;
 		if (!nvif_mmu_kind_valid(mmu, nvbo->kind)) {
 			kfree(nvbo);
-			return -EINVAL;
+			return ERR_PTR(-EINVAL);
 		}
 
 		nvbo->comp = mmu->kind[nvbo->kind] != nvbo->kind;
@@ -241,7 +234,7 @@  nouveau_bo_new(struct nouveau_cli *cli, u64 size, int align,
 		nvbo->comp = (tile_flags & 0x00030000) >> 16;
 		if (!nvif_mmu_kind_valid(mmu, nvbo->kind)) {
 			kfree(nvbo);
-			return -EINVAL;
+			return ERR_PTR(-EINVAL);
 		}
 	} else {
 		nvbo->zeta = (tile_flags & 0x00000007);
@@ -278,7 +271,7 @@  nouveau_bo_new(struct nouveau_cli *cli, u64 size, int align,
 	}
 
 	if (WARN_ON(pi < 0))
-		return -EINVAL;
+		return ERR_PTR(-EINVAL);
 
 	/* Disable compression if suitable settings couldn't be found. */
 	if (nvbo->comp && !vmm->page[pi].comp) {
@@ -288,23 +281,51 @@  nouveau_bo_new(struct nouveau_cli *cli, u64 size, int align,
 	}
 	nvbo->page = vmm->page[pi].shift;
 
+	return nvbo;
+}
+
+int
+nouveau_bo_init(struct nouveau_bo *nvbo, u64 size, int align, u32 flags,
+		struct sg_table *sg, struct reservation_object *robj)
+{
+	int type = sg ? ttm_bo_type_sg : ttm_bo_type_device;
+	size_t acc_size;
+	int ret;
+
+	acc_size = ttm_bo_dma_acc_size(nvbo->bo.bdev, size, sizeof(*nvbo));
+
 	nouveau_bo_fixup_align(nvbo, flags, &align, &size);
 	nvbo->bo.mem.num_pages = size >> PAGE_SHIFT;
 	nouveau_bo_placement_set(nvbo, flags, 0);
 
-	acc_size = ttm_bo_dma_acc_size(&drm->ttm.bdev, size,
-				       sizeof(struct nouveau_bo));
-
-	ret = ttm_bo_init(&drm->ttm.bdev, &nvbo->bo, size,
-			  type, &nvbo->placement,
-			  align >> PAGE_SHIFT, false, acc_size, sg,
-			  robj, nouveau_bo_del_ttm);
-
+	ret = ttm_bo_init(nvbo->bo.bdev, &nvbo->bo, size, type,
+			  &nvbo->placement, align >> PAGE_SHIFT, false,
+			  acc_size, sg, robj, nouveau_bo_del_ttm);
 	if (ret) {
 		/* ttm will call nouveau_bo_del_ttm if it fails.. */
 		return ret;
 	}
 
+	return 0;
+}
+
+int
+nouveau_bo_new(struct nouveau_cli *cli, u64 size, int align,
+	       uint32_t flags, uint32_t tile_mode, uint32_t tile_flags,
+	       struct sg_table *sg, struct reservation_object *robj,
+	       struct nouveau_bo **pnvbo)
+{
+	struct nouveau_bo *nvbo;
+	int ret;
+
+	nvbo = nouveau_bo_alloc(cli, size, flags, tile_mode, tile_flags);
+	if (IS_ERR(nvbo))
+		return PTR_ERR(nvbo);
+
+	ret = nouveau_bo_init(nvbo, size, align, flags, sg, robj);
+	if (ret)
+		return ret;
+
 	*pnvbo = nvbo;
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
index d675efe8e7f9..7529035b971f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.h
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.h
@@ -71,6 +71,10 @@  nouveau_bo_ref(struct nouveau_bo *ref, struct nouveau_bo **pnvbo)
 extern struct ttm_bo_driver nouveau_bo_driver;
 
 void nouveau_bo_move_init(struct nouveau_drm *);
+struct nouveau_bo *nouveau_bo_alloc(struct nouveau_cli *, u64 size, u32 flags,
+				    u32 tile_mode, u32 tile_flags);
+int  nouveau_bo_init(struct nouveau_bo *, u64 size, int align, u32 flags,
+		     struct sg_table *sg, struct reservation_object *robj);
 int  nouveau_bo_new(struct nouveau_cli *, u64 size, int align, u32 flags,
 		    u32 tile_mode, u32 tile_flags, struct sg_table *sg,
 		    struct reservation_object *robj,
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index c7368aa0bdec..e9c772e07789 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -188,11 +188,23 @@  nouveau_gem_new(struct nouveau_cli *cli, u64 size, int align, uint32_t domain,
 	if (domain & NOUVEAU_GEM_DOMAIN_COHERENT)
 		flags |= TTM_PL_FLAG_UNCACHED;
 
-	ret = nouveau_bo_new(cli, size, align, flags, tile_mode,
-			     tile_flags, NULL, NULL, pnvbo);
-	if (ret)
+	nvbo = nouveau_bo_alloc(cli, size, flags, tile_mode, tile_flags);
+	if (IS_ERR(nvbo))
+		return PTR_ERR(nvbo);
+
+	/* Initialize the embedded gem-object. We return a single gem-reference
+	 * to the caller, instead of a normal nouveau_bo ttm reference. */
+	ret = drm_gem_object_init(drm->dev, &nvbo->bo.base, size);
+	if (ret) {
+		nouveau_bo_ref(NULL, &nvbo);
+		return ret;
+	}
+
+	ret = nouveau_bo_init(nvbo, size, align, flags, NULL, NULL);
+	if (ret) {
+		nouveau_bo_ref(NULL, &nvbo);
 		return ret;
-	nvbo = *pnvbo;
+	}
 
 	/* we restrict allowed domains on nv50+ to only the types
 	 * that were requested at creation time.  not possibly on
@@ -203,15 +215,8 @@  nouveau_gem_new(struct nouveau_cli *cli, u64 size, int align, uint32_t domain,
 	if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA)
 		nvbo->valid_domains &= domain;
 
-	/* Initialize the embedded gem-object. We return a single gem-reference
-	 * to the caller, instead of a normal nouveau_bo ttm reference. */
-	ret = drm_gem_object_init(drm->dev, &nvbo->bo.base, nvbo->bo.mem.size);
-	if (ret) {
-		nouveau_bo_ref(NULL, pnvbo);
-		return -ENOMEM;
-	}
-
 	nvbo->bo.persistent_swap_storage = nvbo->bo.base.filp;
+	*pnvbo = nvbo;
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_prime.c b/drivers/gpu/drm/nouveau/nouveau_prime.c
index e86ad7ae622b..0ca71a84e23a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_prime.c
+++ b/drivers/gpu/drm/nouveau/nouveau_prime.c
@@ -63,28 +63,34 @@  struct drm_gem_object *nouveau_gem_prime_import_sg_table(struct drm_device *dev,
 	struct nouveau_drm *drm = nouveau_drm(dev);
 	struct nouveau_bo *nvbo;
 	struct reservation_object *robj = attach->dmabuf->resv;
+	size_t size = attach->dmabuf->size;
 	u32 flags = 0;
 	int ret;
 
 	flags = TTM_PL_FLAG_TT;
 
 	reservation_object_lock(robj, NULL);
-	ret = nouveau_bo_new(&drm->client, attach->dmabuf->size, 0, flags, 0, 0,
-			     sg, robj, &nvbo);
+	nvbo = nouveau_bo_alloc(&drm->client, size, flags, 0, 0);
 	reservation_object_unlock(robj);
-	if (ret)
-		return ERR_PTR(ret);
+	if (IS_ERR(nvbo))
+		return ERR_CAST(nvbo);
 
 	nvbo->valid_domains = NOUVEAU_GEM_DOMAIN_GART;
 
 	/* Initialize the embedded gem-object. We return a single gem-reference
 	 * to the caller, instead of a normal nouveau_bo ttm reference. */
-	ret = drm_gem_object_init(dev, &nvbo->bo.base, nvbo->bo.mem.size);
+	ret = drm_gem_object_init(dev, &nvbo->bo.base, size);
 	if (ret) {
 		nouveau_bo_ref(NULL, &nvbo);
 		return ERR_PTR(-ENOMEM);
 	}
 
+	ret = nouveau_bo_init(nvbo, size, 0, flags, sg, robj);
+	if (ret) {
+		nouveau_bo_ref(NULL, &nvbo);
+		return ERR_PTR(ret);
+	}
+
 	return &nvbo->bo.base;
 }
 

Comments

Hi,

> > Changing the order doesn't look hard.  Patch attached (untested, have no
> > test hardware).  But maybe I missed some detail ...
> 
> I came up with something very similar by splitting up nouveau_bo_new()
> into allocation and initialization steps, so that when necessary the GEM
> object can be initialized in between. I think that's slightly more
> flexible and easier to understand than a boolean flag.

Yes, that should work too.

Acked-by: Gerd Hoffmann <kraxel@redhat.com>

cheers,
  Gerd
On Wed, 14 Aug 2019 at 20:14, Gerd Hoffmann <kraxel@redhat.com> wrote:
>
>   Hi,
>
> > > Changing the order doesn't look hard.  Patch attached (untested, have no
> > > test hardware).  But maybe I missed some detail ...
> >
> > I came up with something very similar by splitting up nouveau_bo_new()
> > into allocation and initialization steps, so that when necessary the GEM
> > object can be initialized in between. I think that's slightly more
> > flexible and easier to understand than a boolean flag.
>
> Yes, that should work too.
>
> Acked-by: Gerd Hoffmann <kraxel@redhat.com>
Acked-by: Ben Skeggs <bskeggs@redhat.com>

>
> cheers,
>   Gerd
>
> _______________________________________________
> Nouveau mailing list
> Nouveau@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/nouveau

On Wed, Aug 21, 2019 at 7:55 AM Thierry Reding <thierry.reding@gmail.com> wrote:
>
> On Wed, Aug 21, 2019 at 04:33:58PM +1000, Ben Skeggs wrote:
> > On Wed, 14 Aug 2019 at 20:14, Gerd Hoffmann <kraxel@redhat.com> wrote:
> > >
> > >   Hi,
> > >
> > > > > Changing the order doesn't look hard.  Patch attached (untested, have no
> > > > > test hardware).  But maybe I missed some detail ...
> > > >
> > > > I came up with something very similar by splitting up nouveau_bo_new()
> > > > into allocation and initialization steps, so that when necessary the GEM
> > > > object can be initialized in between. I think that's slightly more
> > > > flexible and easier to understand than a boolean flag.
> > >
> > > Yes, that should work too.
> > >
> > > Acked-by: Gerd Hoffmann <kraxel@redhat.com>
> > Acked-by: Ben Skeggs <bskeggs@redhat.com>
>
> Thanks guys, applied to drm-misc-next.

Hi Thierry,

Initial investigations suggest that this commit currently in drm-next

commit 019cbd4a4feb3aa3a917d78e7110e3011bbff6d5
Author: Thierry Reding <treding@nvidia.com>
Date:   Wed Aug 14 11:00:48 2019 +0200

    drm/nouveau: Initialize GEM object before TTM object

breaks nouveau userspace which tries to allocate GEM objects with a
non-page-aligned size. Previously nouveau_gem_new would just call
nouveau_bo_init which would call nouveau_bo_fixup_align before
initializing the GEM object. With this change, it is done after. What
do you think -- OK to just move that bit of logic into the new
nouveau_bo_alloc() (and make size/align be pointers so that they can
be fixed up?)

Cheers,

  -ilia