Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <1552561932-23776-1-git-send-email-huangzhaoyang@gmail.com>
In-Reply-To: <1552561932-23776-1-git-send-email-huangzhaoyang@gmail.com>
From:   Zhaoyang Huang <huangzhaoyang@gmail.com>
Date:   Fri, 15 Mar 2019 09:18:46 +0800
Message-ID: <CAGWkznF3T-2ZdFex1ZYpwZx8MB_d4+b7fOBxqVfrnYJ6bx3tuw@mail.gmail.com>
Subject: Re: [PATCH] driver : staging : ion: optimization for decreasing
 memory fragmentaion
To:     "open list:ANDROID ION DRIVER" <devel@driverdev.osuosl.org>,
        "open list:ANDROID ION DRIVER" <dri-devel@lists.freedesktop.org>,
        "moderated list:ANDROID ION DRIVER" <linaro-mm-sig@lists.linaro.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi All,
I would like to explain more for the purpose of this patch. During our
daily test on android system, we found that there is always showing
the memory fragmentation after some test cases(etc camera snapshot),
which lead to huge amount of order0 pages which other page blocks
remain none(200M order0 page blocks in a 1G RAM system). By analysing
the DDR dump file, we can find that it is the ION SYSTEM HEAP which
cut the whole memory into pieces. That is to say, Burst huge order0
allocation is alike a knife that cut higher order blocks into small
one and can not back to original buddy somehow.


On Thu, Mar 14, 2019 at 7:12 PM Zhaoyang Huang <huangzhaoyang@gmail.com> wrote:
>
> From: Zhaoyang Huang <zhaoyang.huang@unisoc.com>
>
> Two action for this patch:
> 1. set a batch size for system heap's shrinker, which can have it buffer
> reasonable page blocks in pool for future allocation.
> 2. reverse the order sequence when free page blocks, the purpose is also
> to have system heap keep as more big blocks as it can.
>
> By testing on an android system with 2G RAM, the changes with setting
> batch = 48MB can help reduce the fragmentation obviously and improve
> big block allocation speed for 15%.
>
> Signed-off-by: Zhaoyang Huang <zhaoyang.huang@unisoc.com>
> ---
>  drivers/staging/android/ion/ion_heap.c        | 12 +++++++++++-
>  drivers/staging/android/ion/ion_system_heap.c |  2 +-
>  2 files changed, 12 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/staging/android/ion/ion_heap.c b/drivers/staging/android/ion/ion_heap.c
> index 31db510..68aa970 100644
> --- a/drivers/staging/android/ion/ion_heap.c
> +++ b/drivers/staging/android/ion/ion_heap.c
> @@ -16,6 +16,8 @@
>  #include <linux/vmalloc.h>
>  #include "ion.h"
>
> +unsigned long ion_heap_batch;
> +
>  void *ion_heap_map_kernel(struct ion_heap *heap,
>                           struct ion_buffer *buffer)
>  {
> @@ -303,7 +305,15 @@ int ion_heap_init_shrinker(struct ion_heap *heap)
>         heap->shrinker.count_objects = ion_heap_shrink_count;
>         heap->shrinker.scan_objects = ion_heap_shrink_scan;
>         heap->shrinker.seeks = DEFAULT_SEEKS;
> -       heap->shrinker.batch = 0;
> +       heap->shrinker.batch = ion_heap_batch;
>
>         return register_shrinker(&heap->shrinker);
>  }
> +
> +static int __init ion_system_heap_batch_init(char *arg)
> +{
> +        ion_heap_batch = memparse(arg, NULL);
> +
> +       return 0;
> +}
> +early_param("ion_batch", ion_system_heap_batch_init);
> diff --git a/drivers/staging/android/ion/ion_system_heap.c b/drivers/staging/android/ion/ion_system_heap.c
> index 701eb9f..d249f8d 100644
> --- a/drivers/staging/android/ion/ion_system_heap.c
> +++ b/drivers/staging/android/ion/ion_system_heap.c
> @@ -182,7 +182,7 @@ static int ion_system_heap_shrink(struct ion_heap *heap, gfp_t gfp_mask,
>         if (!nr_to_scan)
>                 only_scan = 1;
>
> -       for (i = 0; i < NUM_ORDERS; i++) {
> +       for (i = NUM_ORDERS - 1; i >= 0; i--) {
>                 pool = sys_heap->pools[i];
>
>                 if (only_scan) {
> --
> 1.9.1
>