linux-kernel - Re: [RFC PATCH] drivers: base: dma-coherent: find free region without alignment

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-id: <94f35456-b661-908a-bdc0-ddd74cbaf9ec@samsung.com>
Date:   Tue, 14 Nov 2017 12:07:25 +0100
From:   Marek Szyprowski <m.szyprowski@...sung.com>
To:     Jaewon Kim <jaewon31.kim@...sung.com>, hch@....de,
        robin.murphy@....com, gregkh@...uxfoundation.org,
        iommu@...ts.linux-foundation.org
Cc:     akpm@...ux-foundation.org, mhocko@...e.com, vbabka@...e.cz,
        linux-mm@...ck.org, linux-kernel@...r.kernel.org,
        jaewon31.kim@...il.com
Subject: Re: [RFC PATCH] drivers: base: dma-coherent: find free region
 without alignment

Hi Jaewon,

On 2017-11-14 09:42, Jaewon Kim wrote:
> dma-coherent uses bitmap API which internally consider align based on the
> requested size. Depending on some usage pattern, using align, I think, may
> be good for fast search and anti-fragmentation. But with the align, an
> allocation may be failed.
>
> This is a example, total size is 30MB, only few memory at front is being
> used, and 9MB is being requsted. Then 9MB will be aligned to 16MB. The
> first try on offset 0MB will be failed because of others already using. The
> second try on offset 16MB will be failed because of ouf of bound.
>
> So if the align is not necessary on dma-coherent, this patch removes the
> align policy to allow allocation without increasing the total size.

You are right that keeping strict alignment is waste of memory for large
allocations. However for the smaller ones, typically under 1MiB, it helps
to reduce memory fragmentation. The alignment of the allocated buffers is
de-facto guaranteed by the memory management framework in Linux kernel
and there are drivers that depends on this feature.

Maybe it would make sense to keep alignment for buffers smaller than some
predefined value (like 1MiB), something similar to config
ARM_DMA_IOMMU_ALIGNMENT in arch/arm/Kconfig. Otherwise I would expect that
some drivers will be broken by this patch.

> Signed-off-by: Jaewon Kim <jaewon31.kim@...sung.com>
> ---
>   drivers/base/dma-coherent.c | 8 +++++---
>   1 file changed, 5 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/base/dma-coherent.c b/drivers/base/dma-coherent.c
> index 744f64f43454..b86a96d0cd07 100644
> --- a/drivers/base/dma-coherent.c
> +++ b/drivers/base/dma-coherent.c
> @@ -162,7 +162,7 @@ EXPORT_SYMBOL(dma_mark_declared_memory_occupied);
>   static void *__dma_alloc_from_coherent(struct dma_coherent_mem *mem,
>   		ssize_t size, dma_addr_t *dma_handle)
>   {
> -	int order = get_order(size);
> +	int nr_page = PAGE_ALIGN(size) >> PAGE_SHIFT;
>   	unsigned long flags;
>   	int pageno;
>   	void *ret;
> @@ -172,9 +172,11 @@ static void *__dma_alloc_from_coherent(struct dma_coherent_mem *mem,
>   	if (unlikely(size > (mem->size << PAGE_SHIFT)))
>   		goto err;
>   
> -	pageno = bitmap_find_free_region(mem->bitmap, mem->size, order);
> -	if (unlikely(pageno < 0))
> +	pageno = bitmap_find_next_zero_area(mem->bitmap, mem->size, 0,
> +					    nr_page, 0);
> +	if (unlikely(pageno >= mem->size)) {
>   		goto err;
> +	bitmap_set(mem->bitmap, pageno, nr_page);
>   
>   	/*
>   	 * Memory was found in the coherent area.

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland