2017-12-07 23:43:01

by John Stultz

[permalink] [raw]
Subject: [RFC][PATCH] staging: ion: Fix ion_cma_heap allocations

In trying to add support for drm_hwcomposer to HiKey,
I've needed to utilize the ION CMA heap, and I've noticed
problems with allocations on newer kernels failing.

It seems back with 204f672255c22 ("ion: Use CMA APIs directly"),
the ion_cma_heap code was modified to use the CMA API, but
kept the arguments as buffer lengths rather then number of pages.

This results in errors as we don't have enough pages in CMA to
satisfy the exaggerated requests.

This patch converts the ion_cma_heap CMA API usage to properly
request pages.

It also fixes a minor issue in the allocation where in the error
path, the cma_release is called with the buffer->size value which
hasn't yet been set.

I'm no memory expert so close review would be appreciated!

Cc: Laura Abbott <[email protected]>
Cc: Sumit Semwal <[email protected]>
Cc: Benjamin Gaignard <[email protected]>
Cc: Archit Taneja <[email protected]>
Cc: Greg KH <[email protected]>
Cc: Daniel Vetter <[email protected]>
Cc: Dmitry Shmidt <[email protected]>
Cc: Todd Kjos <[email protected]>
Cc: Amit Pundir <[email protected]>
Signed-off-by: John Stultz <[email protected]>
---
drivers/staging/android/ion/ion_cma_heap.c | 15 +++++++++++----
1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/drivers/staging/android/ion/ion_cma_heap.c b/drivers/staging/android/ion/ion_cma_heap.c
index dd5545d..b0fb895 100644
--- a/drivers/staging/android/ion/ion_cma_heap.c
+++ b/drivers/staging/android/ion/ion_cma_heap.c
@@ -39,9 +39,15 @@ static int ion_cma_allocate(struct ion_heap *heap, struct ion_buffer *buffer,
struct ion_cma_heap *cma_heap = to_cma_heap(heap);
struct sg_table *table;
struct page *pages;
+ unsigned long size = PAGE_ALIGN(len);
+ unsigned long nr_pages = size >> PAGE_SHIFT;
+ unsigned long align = get_order(size);
int ret;

- pages = cma_alloc(cma_heap->cma, len, 0, GFP_KERNEL);
+ if (align > CONFIG_CMA_ALIGNMENT)
+ align = CONFIG_CMA_ALIGNMENT;
+
+ pages = cma_alloc(cma_heap->cma, nr_pages, align, GFP_KERNEL);
if (!pages)
return -ENOMEM;

@@ -53,7 +59,7 @@ static int ion_cma_allocate(struct ion_heap *heap, struct ion_buffer *buffer,
if (ret)
goto free_mem;

- sg_set_page(table->sgl, pages, len, 0);
+ sg_set_page(table->sgl, pages, size, 0);

buffer->priv_virt = pages;
buffer->sg_table = table;
@@ -62,7 +68,7 @@ static int ion_cma_allocate(struct ion_heap *heap, struct ion_buffer *buffer,
free_mem:
kfree(table);
err:
- cma_release(cma_heap->cma, pages, buffer->size);
+ cma_release(cma_heap->cma, pages, nr_pages);
return -ENOMEM;
}

@@ -70,9 +76,10 @@ static void ion_cma_free(struct ion_buffer *buffer)
{
struct ion_cma_heap *cma_heap = to_cma_heap(buffer->heap);
struct page *pages = buffer->priv_virt;
+ unsigned long nr_pages = PAGE_ALIGN(buffer->size) >> PAGE_SHIFT;

/* release memory */
- cma_release(cma_heap->cma, pages, buffer->size);
+ cma_release(cma_heap->cma, pages, nr_pages);
/* release sg table */
sg_free_table(buffer->sg_table);
kfree(buffer->sg_table);
--
2.7.4


2017-12-08 00:55:37

by Laura Abbott

[permalink] [raw]
Subject: Re: [RFC][PATCH] staging: ion: Fix ion_cma_heap allocations

On 12/07/2017 03:42 PM, John Stultz wrote:
> In trying to add support for drm_hwcomposer to HiKey,
> I've needed to utilize the ION CMA heap, and I've noticed
> problems with allocations on newer kernels failing.
>
> It seems back with 204f672255c22 ("ion: Use CMA APIs directly"),
> the ion_cma_heap code was modified to use the CMA API, but
> kept the arguments as buffer lengths rather then number of pages.
>
> This results in errors as we don't have enough pages in CMA to
> satisfy the exaggerated requests.
>
> This patch converts the ion_cma_heap CMA API usage to properly
> request pages.
>
> It also fixes a minor issue in the allocation where in the error
> path, the cma_release is called with the buffer->size value which
> hasn't yet been set.
>
> I'm no memory expert so close review would be appreciated!
>

Yup.

Acked-by: Laura Abbott <[email protected]>

> Cc: Laura Abbott <[email protected]>
> Cc: Sumit Semwal <[email protected]>
> Cc: Benjamin Gaignard <[email protected]>
> Cc: Archit Taneja <[email protected]>
> Cc: Greg KH <[email protected]>
> Cc: Daniel Vetter <[email protected]>
> Cc: Dmitry Shmidt <[email protected]>
> Cc: Todd Kjos <[email protected]>
> Cc: Amit Pundir <[email protected]>
> Signed-off-by: John Stultz <[email protected]>
> ---
> drivers/staging/android/ion/ion_cma_heap.c | 15 +++++++++++----
> 1 file changed, 11 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/staging/android/ion/ion_cma_heap.c b/drivers/staging/android/ion/ion_cma_heap.c
> index dd5545d..b0fb895 100644
> --- a/drivers/staging/android/ion/ion_cma_heap.c
> +++ b/drivers/staging/android/ion/ion_cma_heap.c
> @@ -39,9 +39,15 @@ static int ion_cma_allocate(struct ion_heap *heap, struct ion_buffer *buffer,
> struct ion_cma_heap *cma_heap = to_cma_heap(heap);
> struct sg_table *table;
> struct page *pages;
> + unsigned long size = PAGE_ALIGN(len);
> + unsigned long nr_pages = size >> PAGE_SHIFT;
> + unsigned long align = get_order(size);
> int ret;
>
> - pages = cma_alloc(cma_heap->cma, len, 0, GFP_KERNEL);
> + if (align > CONFIG_CMA_ALIGNMENT)
> + align = CONFIG_CMA_ALIGNMENT;
> +
> + pages = cma_alloc(cma_heap->cma, nr_pages, align, GFP_KERNEL);
> if (!pages)
> return -ENOMEM;
>
> @@ -53,7 +59,7 @@ static int ion_cma_allocate(struct ion_heap *heap, struct ion_buffer *buffer,
> if (ret)
> goto free_mem;
>
> - sg_set_page(table->sgl, pages, len, 0);
> + sg_set_page(table->sgl, pages, size, 0);
>
> buffer->priv_virt = pages;
> buffer->sg_table = table;
> @@ -62,7 +68,7 @@ static int ion_cma_allocate(struct ion_heap *heap, struct ion_buffer *buffer,
> free_mem:
> kfree(table);
> err:
> - cma_release(cma_heap->cma, pages, buffer->size);
> + cma_release(cma_heap->cma, pages, nr_pages);
> return -ENOMEM;
> }
>
> @@ -70,9 +76,10 @@ static void ion_cma_free(struct ion_buffer *buffer)
> {
> struct ion_cma_heap *cma_heap = to_cma_heap(buffer->heap);
> struct page *pages = buffer->priv_virt;
> + unsigned long nr_pages = PAGE_ALIGN(buffer->size) >> PAGE_SHIFT;
>
> /* release memory */
> - cma_release(cma_heap->cma, pages, buffer->size);
> + cma_release(cma_heap->cma, pages, nr_pages);
> /* release sg table */
> sg_free_table(buffer->sg_table);
> kfree(buffer->sg_table);
>

2017-12-08 01:21:40

by John Stultz

[permalink] [raw]
Subject: Re: [RFC][PATCH] staging: ion: Fix ion_cma_heap allocations

On Thu, Dec 7, 2017 at 4:55 PM, Laura Abbott <[email protected]> wrote:
> On 12/07/2017 03:42 PM, John Stultz wrote:
>>
>> In trying to add support for drm_hwcomposer to HiKey,
>> I've needed to utilize the ION CMA heap, and I've noticed
>> problems with allocations on newer kernels failing.
>>
>> It seems back with 204f672255c22 ("ion: Use CMA APIs directly"),
>> the ion_cma_heap code was modified to use the CMA API, but
>> kept the arguments as buffer lengths rather then number of pages.
>>
>> This results in errors as we don't have enough pages in CMA to
>> satisfy the exaggerated requests.
>>
>> This patch converts the ion_cma_heap CMA API usage to properly
>> request pages.
>>
>> It also fixes a minor issue in the allocation where in the error
>> path, the cma_release is called with the buffer->size value which
>> hasn't yet been set.
>>
>> I'm no memory expert so close review would be appreciated!
>>
>
> Yup.
>
> Acked-by: Laura Abbott <[email protected]>

Thanks for that!

I realized right after sending that there's a checkpatch issue to be
fixed up, so I'll resend tomorrow (including your ack) if there aren't
any other objections.

thanks
-john