Please ignore, sent to wrong list.
> -----Original Message-----
> From: Dylan Yip <dylan....@xilinx.com>
> Sent: Tuesday, September 17, 2019 1:21 PM
> To: linux-media@vger.kernel.org; Satish Kumar Nagireddy
> <satis...@xilinx.com>
> Cc: Dylan Yip <dyl...@xilinx.com>
> Subject: [LINUX PATCH] dma-mapping: Control memset operation using gfp
> flags
>
> In case of 4k video buffer, the allocation from a reserved memory is taking a
> long time, ~500ms. This is root caused to the memset() operations on the
> allocated memory which is consuming more cpu cycles.
> Due to this delay, we see that initial frames are being dropped.
>
> To fix this, we have wrapped the default memset, done when allocating
> coherent memory, under the __GFP_ZERO flag. So, we only clear allocated
> memory if __GFP_ZERO flag is enabled. We believe this should be safe as the
> video decoder always writes before reading.
> This optimizes decoder initialization as we do not set the __GFP_ZERO flag
> when allocating memory for decoder. With this optimization, we don't see
> initial frame drops and decoder initialization time is ~100ms.
>
> This patch adds plumbing through dma_alloc functions to pass gfp flag set by
> user to __dma_alloc_from_coherent(). Here gfp flag is checked for
> __GFP_ZERO. If present, we memset the buffer to 0 otherwise we skip
> memset.
>
> Signed-off-by: Dylan Yip <dylan....@xilinx.com>
> ---
> arch/arm/mm/dma-mapping-nommu.c | 2 +-
> include/linux/dma-mapping.h | 11 +++++++----
> kernel/dma/coherent.c | 15 +++++++++------
> kernel/dma/mapping.c | 2 +-
> 4 files changed, 18 insertions(+), 12 deletions(-)
>
> diff --git a/arch/arm/mm/dma-mapping-nommu.c b/arch/arm/mm/dma-
> mapping-nommu.c index 52b8255..242b2c3 100644
> --- a/arch/arm/mm/dma-mapping-nommu.c
> +++ b/arch/arm/mm/dma-mapping-nommu.c
> @@ -35,7 +35,7 @@ static void *arm_nommu_dma_alloc(struct device *dev,
> size_t size,
> unsigned long attrs)
>
> {
> - void *ret = dma_alloc_from_global_coherent(size, dma_handle);
> + void *ret = dma_alloc_from_global_coherent(size, dma_handle,
> gfp);
>
> /*
> * dma_alloc_from_global_coherent() may fail because:
> diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
> index f7d1eea..b715c9f 100644
> --- a/include/linux/dma-mapping.h
> +++ b/include/linux/dma-mapping.h
> @@ -160,24 +160,27 @@ static inline int is_device_dma_capable(struct
> device *dev)
> * Don't use them in device drivers.
> */
> int dma_alloc_from_dev_coherent(struct device *dev, ssize_t size,
> - dma_addr_t *dma_handle, void **ret);
> + dma_addr_t *dma_handle, void **ret,
> + gfp_t flag);
> int dma_release_from_dev_coherent(struct device *dev, int order, void
> *vaddr);
>
> int dma_mmap_from_dev_coherent(struct device *dev, struct
> vm_area_struct *vma,
> void *cpu_addr, size_t size, int *ret);
>
> -void *dma_alloc_from_global_coherent(ssize_t size, dma_addr_t
> *dma_handle);
> +void *dma_alloc_from_global_coherent(ssize_t size, dma_addr_t
> *dma_handle,
> + gfp_t flag);
> int dma_release_from_global_coherent(int order, void *vaddr); int
> dma_mmap_from_global_coherent(struct vm_area_struct *vma, void
> *cpu_addr,
> size_t size, int *ret);
>
> #else
> -#define dma_alloc_from_dev_coherent(dev, size, handle, ret) (0)
> +#define dma_alloc_from_dev_coherent(dev, size, handle, ret, flag) (0)
> #define dma_release_from_dev_coherent(dev, order, vaddr) (0) #define
> dma_mmap_from_dev_coherent(dev, vma, vaddr, order, ret) (0)
>
> static inline void *dma_alloc_from_global_coherent(ssize_t size,
> - dma_addr_t *dma_handle)
> + dma_addr_t *dma_handle,
> + gfp_t flag)
> {
> return NULL;
> }
> diff --git a/kernel/dma/coherent.c b/kernel/dma/coherent.c index
> 29fd659..d85fab5 100644
> --- a/kernel/dma/coherent.c
> +++ b/kernel/dma/coherent.c
> @@ -136,7 +136,7 @@ void dma_release_declared_memory(struct device
> *dev) EXPORT_SYMBOL(dma_release_declared_memory);
>
> static void *__dma_alloc_from_coherent(struct dma_coherent_mem
> *mem,
> - ssize_t size, dma_addr_t *dma_handle)
> + ssize_t size, dma_addr_t *dma_handle, gfp_t gfp_flag)
> {
> int order = get_order(size);
> unsigned long flags;
> @@ -158,7 +158,8 @@ static void *__dma_alloc_from_coherent(struct
> dma_coherent_mem *mem,
> *dma_handle = mem->device_base + (pageno << PAGE_SHIFT);
> ret = mem->virt_base + (pageno << PAGE_SHIFT);
> spin_unlock_irqrestore(&mem->spinlock, flags);
> - memset(ret, 0, size);
> + if (gfp_flag & __GFP_ZERO)
> + memset(ret, 0, size);
> return ret;
> err:
> spin_unlock_irqrestore(&mem->spinlock, flags); @@ -172,6 +173,7
> @@ static void *__dma_alloc_from_coherent(struct dma_coherent_mem
> *mem,
> * @dma_handle: This will be filled with the correct dma handle
> * @ret: This pointer will be filled with the virtual address
> * to allocated area.
> + * @flag: gfp flag set by user
> *
> * This function should be only called from per-arch dma_alloc_coherent()
> * to support allocation from per-device coherent memory pools.
> @@ -180,24 +182,25 @@ static void *__dma_alloc_from_coherent(struct
> dma_coherent_mem *mem,
> * generic memory areas, or !0 if dma_alloc_coherent should return @ret.
> */
> int dma_alloc_from_dev_coherent(struct device *dev, ssize_t size,
> - dma_addr_t *dma_handle, void **ret)
> + dma_addr_t *dma_handle, void **ret, gfp_t flag)
> {
> struct dma_coherent_mem *mem =
> dev_get_coherent_memory(dev);
>
> if (!mem)
> return 0;
>
> - *ret = __dma_alloc_from_coherent(mem, size, dma_handle);
> + *ret = __dma_alloc_from_coherent(mem, size, dma_handle, flag);
> return 1;
> }
>
> -void *dma_alloc_from_global_coherent(ssize_t size, dma_addr_t
> *dma_handle)
> +void *dma_alloc_from_global_coherent(ssize_t size, dma_addr_t
> *dma_handle,
> + gfp_t flag)
> {
> if (!dma_coherent_default_memory)
> return NULL;
>
> return
> __dma_alloc_from_coherent(dma_coherent_default_memory, size,
> - dma_handle);
> + dma_handle, flag);
> }
>
> static int __dma_release_from_coherent(struct dma_coherent_mem
> *mem, diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c index
> b0038ca..bfea1d2 100644
> --- a/kernel/dma/mapping.c
> +++ b/kernel/dma/mapping.c
> @@ -272,7 +272,7 @@ void *dma_alloc_attrs(struct device *dev, size_t size,
> dma_addr_t *dma_handle,
>
> WARN_ON_ONCE(!dev->coherent_dma_mask);
>
> - if (dma_alloc_from_dev_coherent(dev, size, dma_handle,
> &cpu_addr))
> + if (dma_alloc_from_dev_coherent(dev, size, dma_handle,
> &cpu_addr,
> +flag))
> return cpu_addr;
>
> /* let the implementation decide on the zone to allocate from: */
> --
> 2.7.4
