[1/2] drm/tegra: Return an error code if fails

  Return -ENOMEM if tegra_bo_mmap() fails.

Signed-off-by: Sui Jingfeng <suijingfeng@loongson.cn>
---
 drivers/gpu/drm/tegra/gem.c | 2 ++
 1 file changed, 2 insertions(+)
  

Hi,

Gentle ping for this series.


On 2023/6/26 22:33, Sui Jingfeng wrote:
> Return -ENOMEM if tegra_bo_mmap() fails.
>
> Signed-off-by: Sui Jingfeng <suijingfeng@loongson.cn>
> ---
>   drivers/gpu/drm/tegra/gem.c | 2 ++
>   1 file changed, 2 insertions(+)
>
> diff --git a/drivers/gpu/drm/tegra/gem.c b/drivers/gpu/drm/tegra/gem.c
> index dea38892d6e6..0ce22935fbd3 100644
> --- a/drivers/gpu/drm/tegra/gem.c
> +++ b/drivers/gpu/drm/tegra/gem.c
> @@ -710,6 +710,8 @@ static int tegra_gem_prime_vmap(struct dma_buf *buf, struct iosys_map *map)
>   	void *vaddr;
>   
>   	vaddr = tegra_bo_mmap(&bo->base);
> +	if (!vaddr)
> +		return -ENOMEM;
>   	if (IS_ERR(vaddr))
>   		return PTR_ERR(vaddr);
>
  

On Mon, Jun 26, 2023 at 10:33:30PM +0800, Sui Jingfeng wrote:
> Return -ENOMEM if tegra_bo_mmap() fails.
> 
> Signed-off-by: Sui Jingfeng <suijingfeng@loongson.cn>
> ---
>  drivers/gpu/drm/tegra/gem.c | 2 ++
>  1 file changed, 2 insertions(+)

Sorry, this fell through the cracks. I think it'd be better if
tegra_bo_mmap() were to be improved to always return either an ERR_PTR()
encoded error code or a valid pointer. Throwing NULL into the mix isn't
useful because it typically means something like -ENOMEM anyway. Error
codes are more explicit, so since we're already using them for some
cases, might as well return them for all.

Actually, looks like tegra_bo_mmap() never actually returns an ERR_PTR()
encoded error code. It's either obj->vaddr, the return value of vmap()
(which is either NULL or the address of the mapping), or the address
obtained from dma_buf_vmap_unlocked() (i.e. map.vaddr) or NULL on
failure. So I think it would equally make sense to keep your patch and
to remove the IS_ERR() check below it.

I would slightly prefer the first option, but either is fine.

Thierry

> diff --git a/drivers/gpu/drm/tegra/gem.c b/drivers/gpu/drm/tegra/gem.c
> index dea38892d6e6..0ce22935fbd3 100644
> --- a/drivers/gpu/drm/tegra/gem.c
> +++ b/drivers/gpu/drm/tegra/gem.c
> @@ -710,6 +710,8 @@ static int tegra_gem_prime_vmap(struct dma_buf *buf, struct iosys_map *map)
>  	void *vaddr;
>  
>  	vaddr = tegra_bo_mmap(&bo->base);
> +	if (!vaddr)
> +		return -ENOMEM;
>  	if (IS_ERR(vaddr))
>  		return PTR_ERR(vaddr);
>  
> -- 
> 2.25.1
>
  

On Tue, Oct 10, 2023 at 03:22:56PM +0200, Thierry Reding wrote:
> On Mon, Jun 26, 2023 at 10:33:30PM +0800, Sui Jingfeng wrote:
> > Return -ENOMEM if tegra_bo_mmap() fails.
> > 
> > Signed-off-by: Sui Jingfeng <suijingfeng@loongson.cn>
> > ---
> >  drivers/gpu/drm/tegra/gem.c | 2 ++
> >  1 file changed, 2 insertions(+)
> 
> Sorry, this fell through the cracks. I think it'd be better if
> tegra_bo_mmap() were to be improved to always return either an ERR_PTR()
> encoded error code or a valid pointer. Throwing NULL into the mix isn't
> useful because it typically means something like -ENOMEM anyway. Error
> codes are more explicit, so since we're already using them for some
> cases, might as well return them for all.
> 
> Actually, looks like tegra_bo_mmap() never actually returns an ERR_PTR()
> encoded error code. It's either obj->vaddr, the return value of vmap()
> (which is either NULL or the address of the mapping), or the address
> obtained from dma_buf_vmap_unlocked() (i.e. map.vaddr) or NULL on
> failure. So I think it would equally make sense to keep your patch and
> to remove the IS_ERR() check below it.
> 
> I would slightly prefer the first option, but either is fine.

How about the attached patch?

Thierry
From b34a09efcf7b1d2c25d3baf8c6d91c5ca09b4e0f Mon Sep 17 00:00:00 2001
From: Thierry Reding <treding@nvidia.com>
Date: Tue, 10 Oct 2023 17:26:14 +0200
Subject: [PATCH] drm/tegra: gem: Do not return NULL in tegra_bo_mmap()

It's confusing for a function to return NULL and ERR_PTR()-encoded error
codes on failure. Make sure we only ever return the latter since that's
what callers already expect.

Signed-off-by: Thierry Reding <treding@nvidia.com>
---
 drivers/gpu/drm/tegra/gem.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/tegra/gem.c b/drivers/gpu/drm/tegra/gem.c
index 11296de59c5a..679460e05c05 100644
--- a/drivers/gpu/drm/tegra/gem.c
+++ b/drivers/gpu/drm/tegra/gem.c
@@ -178,6 +178,7 @@ static void *tegra_bo_mmap(struct host1x_bo *bo)
 {
 	struct tegra_bo *obj = host1x_to_tegra_bo(bo);
 	struct iosys_map map;
+	void *vaddr;
 	int ret;
 
 	if (obj->vaddr)
@@ -185,10 +186,18 @@ static void *tegra_bo_mmap(struct host1x_bo *bo)
 
 	if (obj->gem.import_attach) {
 		ret = dma_buf_vmap_unlocked(obj->gem.import_attach->dmabuf, &map);
-		return ret ? NULL : map.vaddr;
+		if (ret < 0)
+			return ERR_PTR(ret);
+
+		return map.vaddr;
 	}
 
-	return vmap(obj->pages, obj->num_pages, VM_MAP, pgprot_writecombine(PAGE_KERNEL));
+	vaddr = vmap(obj->pages, obj->num_pages, VM_MAP,
+		     pgprot_writecombine(PAGE_KERNEL));
+	if (!vaddr)
+		return -ENOMEM;
+
+	return vaddr;
 }
 
 static void tegra_bo_munmap(struct host1x_bo *bo, void *addr)
  

Hi,


On 2023/10/10 23:31, Thierry Reding wrote:
> On Tue, Oct 10, 2023 at 03:22:56PM +0200, Thierry Reding wrote:
>> On Mon, Jun 26, 2023 at 10:33:30PM +0800, Sui Jingfeng wrote:
>>> Return -ENOMEM if tegra_bo_mmap() fails.
>>>
>>> Signed-off-by: Sui Jingfeng <suijingfeng@loongson.cn>
>>> ---
>>>   drivers/gpu/drm/tegra/gem.c | 2 ++
>>>   1 file changed, 2 insertions(+)
>> Sorry, this fell through the cracks. I think it'd be better if
>> tegra_bo_mmap() were to be improved to always return either an ERR_PTR()
>> encoded error code or a valid pointer. Throwing NULL into the mix isn't
>> useful because it typically means something like -ENOMEM anyway. Error
>> codes are more explicit, so since we're already using them for some
>> cases, might as well return them for all.
>>
>> Actually, looks like tegra_bo_mmap() never actually returns an ERR_PTR()
>> encoded error code. It's either obj->vaddr, the return value of vmap()
>> (which is either NULL or the address of the mapping), or the address
>> obtained from dma_buf_vmap_unlocked() (i.e. map.vaddr) or NULL on
>> failure. So I think it would equally make sense to keep your patch and
>> to remove the IS_ERR() check below it.
>>
>> I would slightly prefer the first option, but either is fine.
> How about the attached patch?


I also prefer the prefer the first option.
The attached patch is more better, because it solve the problem at lower level.

Reviewed-by: Sui Jingfeng <suijingfeng@loongson.cn>

> Thierry