2023-04-15 01:26:18

by John Ogness

Subject: [PATCH v2] drm/nouveau: fix incorrect conversion to dma_resv_wait_timeout()

Commit 41d351f29528 ("drm/nouveau: stop using ttm_bo_wait")
converted from ttm_bo_wait_ctx() to dma_resv_wait_timeout().
However, dma_resv_wait_timeout() returns greater than zero on
success as opposed to ttm_bo_wait_ctx(). As a result, relocs
will fail and log errors even when it was a success.

Change the return code handling to match that of
nouveau_gem_ioctl_cpu_prep(), which was already using
dma_resv_wait_timeout() correctly.

Fixes: 41d351f29528 ("drm/nouveau: stop using ttm_bo_wait")
Reported-by: Tanmay Bhushan <[email protected]>
Link: https://lore.kernel.org/lkml/[email protected]
Signed-off-by: John Ogness <[email protected]>
---
The original report was actually a patch that needed fixing.
Since nobody has stepped up to fix this regression correctly,
I'm posting the v2.

This is a real regression introduced in 6.3-rc1.

drivers/gpu/drm/nouveau/nouveau_gem.c | 18 ++++++++++++------
1 file changed, 12 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index f77e44958037..346839c24273 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -645,8 +645,9 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
struct drm_nouveau_gem_pushbuf_reloc *reloc,
struct drm_nouveau_gem_pushbuf_bo *bo)
{
- long ret = 0;
+ int ret = 0;
unsigned i;
+ long lret;

for (i = 0; i < req->nr_relocs; i++) {
struct drm_nouveau_gem_pushbuf_reloc *r = &reloc[i];
@@ -703,13 +704,18 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
data |= r->vor;
}

- ret = dma_resv_wait_timeout(nvbo->bo.base.resv,
- DMA_RESV_USAGE_BOOKKEEP,
- false, 15 * HZ);
- if (ret == 0)
+ lret = dma_resv_wait_timeout(nvbo->bo.base.resv,
+ DMA_RESV_USAGE_BOOKKEEP,
+ false, 15 * HZ);
+ if (!lret)
ret = -EBUSY;
+ else if (lret > 0)
+ ret = 0;
+ else
+ ret = lret;
+
if (ret) {
- NV_PRINTK(err, cli, "reloc wait_idle failed: %ld\n",
+ NV_PRINTK(err, cli, "reloc wait_idle failed: %d\n",
ret);
break;
}

base-commit: 09a9639e56c01c7a00d6c0ca63f4c7c41abe075d
--
2.30.2

2023-04-15 02:05:15

by John Ogness

Subject: [PATCH v3] drm/nouveau: fix incorrect conversion to dma_resv_wait_timeout()

Commit 41d351f29528 ("drm/nouveau: stop using ttm_bo_wait")
converted from ttm_bo_wait_ctx() to dma_resv_wait_timeout().
However, dma_resv_wait_timeout() returns greater than zero on
success as opposed to ttm_bo_wait_ctx(). As a result, relocs
will fail and log errors even when it was a success.

Change the return code handling to match that of
nouveau_gem_ioctl_cpu_prep(), which was already using
dma_resv_wait_timeout() correctly.

Fixes: 41d351f29528 ("drm/nouveau: stop using ttm_bo_wait")
Reported-by: Tanmay Bhushan <[email protected]>
Link: https://lore.kernel.org/lkml/[email protected]
Signed-off-by: John Ogness <[email protected]>
---
I just realized that the nouveau driver style prefers to scope
variables used only in loops.

v3: Define @lret within the for-loop.

drivers/gpu/drm/nouveau/nouveau_gem.c | 18 ++++++++++++------
1 file changed, 12 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index f77e44958037..ab9062e50977 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -645,7 +645,7 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
struct drm_nouveau_gem_pushbuf_reloc *reloc,
struct drm_nouveau_gem_pushbuf_bo *bo)
{
- long ret = 0;
+ int ret = 0;
unsigned i;

for (i = 0; i < req->nr_relocs; i++) {
@@ -653,6 +653,7 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
struct drm_nouveau_gem_pushbuf_bo *b;
struct nouveau_bo *nvbo;
uint32_t data;
+ long lret;

if (unlikely(r->bo_index >= req->nr_buffers)) {
NV_PRINTK(err, cli, "reloc bo index invalid\n");
@@ -703,13 +704,18 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
data |= r->vor;
}

- ret = dma_resv_wait_timeout(nvbo->bo.base.resv,
- DMA_RESV_USAGE_BOOKKEEP,
- false, 15 * HZ);
- if (ret == 0)
+ lret = dma_resv_wait_timeout(nvbo->bo.base.resv,
+ DMA_RESV_USAGE_BOOKKEEP,
+ false, 15 * HZ);
+ if (!lret)
ret = -EBUSY;
+ else if (lret > 0)
+ ret = 0;
+ else
+ ret = lret;
+
if (ret) {
- NV_PRINTK(err, cli, "reloc wait_idle failed: %ld\n",
+ NV_PRINTK(err, cli, "reloc wait_idle failed: %d\n",
ret);
break;
}

base-commit: 09a9639e56c01c7a00d6c0ca63f4c7c41abe075d
--
2.30.2

2023-04-17 06:00:30

by Christian König

Subject: Re: [PATCH v3] drm/nouveau: fix incorrect conversion to dma_resv_wait_timeout()

Am 15.04.23 um 04:02 schrieb John Ogness:
> Commit 41d351f29528 ("drm/nouveau: stop using ttm_bo_wait")
> converted from ttm_bo_wait_ctx() to dma_resv_wait_timeout().
> However, dma_resv_wait_timeout() returns greater than zero on
> success as opposed to ttm_bo_wait_ctx(). As a result, relocs
> will fail and log errors even when it was a success.
>
> Change the return code handling to match that of
> nouveau_gem_ioctl_cpu_prep(), which was already using
> dma_resv_wait_timeout() correctly.
>
> Fixes: 41d351f29528 ("drm/nouveau: stop using ttm_bo_wait")
> Reported-by: Tanmay Bhushan <[email protected]>
> Link: https://lore.kernel.org/lkml/[email protected]
> Signed-off-by: John Ogness <[email protected]>

Reviewed-by: Christian König <[email protected]>

> ---
> I just realized that the nouveau driver style prefers to scope
> variables used only in loops.
>
> v3: Define @lret within the for-loop.
>
> drivers/gpu/drm/nouveau/nouveau_gem.c | 18 ++++++++++++------
> 1 file changed, 12 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
> index f77e44958037..ab9062e50977 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_gem.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
> @@ -645,7 +645,7 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
> struct drm_nouveau_gem_pushbuf_reloc *reloc,
> struct drm_nouveau_gem_pushbuf_bo *bo)
> {
> - long ret = 0;
> + int ret = 0;
> unsigned i;
>
> for (i = 0; i < req->nr_relocs; i++) {
> @@ -653,6 +653,7 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
> struct drm_nouveau_gem_pushbuf_bo *b;
> struct nouveau_bo *nvbo;
> uint32_t data;
> + long lret;
>
> if (unlikely(r->bo_index >= req->nr_buffers)) {
> NV_PRINTK(err, cli, "reloc bo index invalid\n");
> @@ -703,13 +704,18 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
> data |= r->vor;
> }
>
> - ret = dma_resv_wait_timeout(nvbo->bo.base.resv,
> - DMA_RESV_USAGE_BOOKKEEP,
> - false, 15 * HZ);
> - if (ret == 0)
> + lret = dma_resv_wait_timeout(nvbo->bo.base.resv,
> + DMA_RESV_USAGE_BOOKKEEP,
> + false, 15 * HZ);
> + if (!lret)
> ret = -EBUSY;
> + else if (lret > 0)
> + ret = 0;
> + else
> + ret = lret;
> +
> if (ret) {
> - NV_PRINTK(err, cli, "reloc wait_idle failed: %ld\n",
> + NV_PRINTK(err, cli, "reloc wait_idle failed: %d\n",
> ret);
> break;
> }
>
> base-commit: 09a9639e56c01c7a00d6c0ca63f4c7c41abe075d

2023-04-17 12:58:09

by Karol Herbst

Subject: Re: [PATCH v3] drm/nouveau: fix incorrect conversion to dma_resv_wait_timeout()

On Mon, Apr 17, 2023 at 7:55 AM Christian König
<[email protected]> wrote:
>
> Am 15.04.23 um 04:02 schrieb John Ogness:
> > Commit 41d351f29528 ("drm/nouveau: stop using ttm_bo_wait")
> > converted from ttm_bo_wait_ctx() to dma_resv_wait_timeout().
> > However, dma_resv_wait_timeout() returns greater than zero on
> > success as opposed to ttm_bo_wait_ctx(). As a result, relocs
> > will fail and log errors even when it was a success.
> >
> > Change the return code handling to match that of
> > nouveau_gem_ioctl_cpu_prep(), which was already using
> > dma_resv_wait_timeout() correctly.
> >
> > Fixes: 41d351f29528 ("drm/nouveau: stop using ttm_bo_wait")
> > Reported-by: Tanmay Bhushan <[email protected]>
> > Link: https://lore.kernel.org/lkml/[email protected]
> > Signed-off-by: John Ogness <[email protected]>
>
> Reviewed-by: Christian König <[email protected]>
>

Reviewed-by: Karol Herbst <[email protected]>

will push shortly to drm-misc-fixes

> > ---
> > I just realized that the nouveau driver style prefers to scope
> > variables used only in loops.
> >
> > v3: Define @lret within the for-loop.
> >
> > drivers/gpu/drm/nouveau/nouveau_gem.c | 18 ++++++++++++------
> > 1 file changed, 12 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
> > index f77e44958037..ab9062e50977 100644
> > --- a/drivers/gpu/drm/nouveau/nouveau_gem.c
> > +++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
> > @@ -645,7 +645,7 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
> > struct drm_nouveau_gem_pushbuf_reloc *reloc,
> > struct drm_nouveau_gem_pushbuf_bo *bo)
> > {
> > - long ret = 0;
> > + int ret = 0;
> > unsigned i;
> >
> > for (i = 0; i < req->nr_relocs; i++) {
> > @@ -653,6 +653,7 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
> > struct drm_nouveau_gem_pushbuf_bo *b;
> > struct nouveau_bo *nvbo;
> > uint32_t data;
> > + long lret;
> >
> > if (unlikely(r->bo_index >= req->nr_buffers)) {
> > NV_PRINTK(err, cli, "reloc bo index invalid\n");
> > @@ -703,13 +704,18 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
> > data |= r->vor;
> > }
> >
> > - ret = dma_resv_wait_timeout(nvbo->bo.base.resv,
> > - DMA_RESV_USAGE_BOOKKEEP,
> > - false, 15 * HZ);
> > - if (ret == 0)
> > + lret = dma_resv_wait_timeout(nvbo->bo.base.resv,
> > + DMA_RESV_USAGE_BOOKKEEP,
> > + false, 15 * HZ);
> > + if (!lret)
> > ret = -EBUSY;
> > + else if (lret > 0)
> > + ret = 0;
> > + else
> > + ret = lret;
> > +
> > if (ret) {
> > - NV_PRINTK(err, cli, "reloc wait_idle failed: %ld\n",
> > + NV_PRINTK(err, cli, "reloc wait_idle failed: %d\n",
> > ret);
> > break;
> > }
> >
> > base-commit: 09a9639e56c01c7a00d6c0ca63f4c7c41abe075d
>