On 2019/8/9 23:39, Jaegeuk Kim wrote:
> On 08/10, Ju Hyung Park wrote:
>> Hi Jaegeuk,
>>
>> Just out of curiosity, what's the point of this?
>>
>> I thought flash chips skip erasing blocks if it's already erased to
>> preserve P/E cycles as much as possible.
>> All Android devices I had(various versions of eMMC and UFS) ran full range
>> block-level discards pretty fast too.
> 
> Unfortunately, some of them are giving long delays on a bunch of unmap 
> commands
> resulting in user janky issue.

Agreed, the performance completely depends on FTL implementation, some of them
perform very bad...

Thanks,

> 
>>
>> Thanks.
>>
>> 2019년 8월 10일 (토) 오전 12:13, Jaegeuk Kim <jaeg...@kernel.org>님이 작성:
>>
>>> We actually don't need to issue trim on entire disk by checking first
>>> blocks having zeros.
>>>
>>> Signed-off-by: Jaegeuk Kim <jaeg...@kernel.org>
>>> ---
>>> v2 from v1:
>>>  - clean up
>>>
>>>  mkfs/f2fs_format_utils.c | 53 ++++++++++++++++++++++++++++++++++++++--
>>>  1 file changed, 51 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/mkfs/f2fs_format_utils.c b/mkfs/f2fs_format_utils.c
>>> index 8bf128c..f2d55ad 100644
>>> --- a/mkfs/f2fs_format_utils.c
>>> +++ b/mkfs/f2fs_format_utils.c
>>> @@ -25,6 +25,7 @@
>>>  #include <stdio.h>
>>>  #include <unistd.h>
>>>  #include <stdlib.h>
>>> +#include <stdbool.h>
>>>  #ifndef ANDROID_WINDOWS_HOST
>>>  #include <sys/ioctl.h>
>>>  #endif
>>> @@ -110,13 +111,61 @@ static int trim_device(int i)
>>>         return 0;
>>>  }
>>>
>>> +static bool is_wiped_device(int i)
>>> +{
>>> +#ifdef WITH_ANDROID
>>> +       struct device_info *dev = c.devices + i;
>>> +       int fd = dev->fd;
>>> +       char *buf, *zero_buf;
>>> +       bool wiped = true;
>>> +       int nblocks = 4096;     /* 16MB size */
>>> +       int j;
>>> +
>>> +       buf = malloc(F2FS_BLKSIZE);
>>> +       if (buf == NULL) {
>>> +               MSG(1, "\tError: Malloc Failed for buf!!!\n");
>>> +               return false;
>>> +       }
>>> +       zero_buf = calloc(1, F2FS_BLKSIZE);
>>> +       if (zero_buf == NULL) {
>>> +               MSG(1, "\tError: Calloc Failed for zero buf!!!\n");
>>> +               free(buf);
>>> +               return false;
>>> +       }
>>> +
>>> +       if (lseek(fd, 0, SEEK_SET) < 0) {
>>> +               free(zero_buf);
>>> +               free(buf);
>>> +               return false;
>>> +       }
>>> +
>>> +       /* check first n blocks */
>>> +       for (j = 0; j < nblocks; j++) {
>>> +               if (read(fd, buf, F2FS_BLKSIZE) != F2FS_BLKSIZE ||
>>> +                               memcmp(buf, zero_buf, F2FS_BLKSIZE)) {
>>> +                       wiped = false;
>>> +                       break;
>>> +               }
>>> +       }
>>> +       free(zero_buf);
>>> +       free(buf);
>>> +
>>> +       if (wiped)
>>> +               MSG(0, "Info: Found all zeros in first %d blocks\n",
>>> nblocks);
>>> +       return wiped;
>>> +#else
>>> +       return false;
>>> +#endif
>>> +}
>>> +
>>>  int f2fs_trim_devices(void)
>>>  {
>>>         int i;
>>>
>>> -       for (i = 0; i < c.ndevs; i++)
>>> -               if (trim_device(i))
>>> +       for (i = 0; i < c.ndevs; i++) {
>>> +               if (!is_wiped_device(i) && trim_device(i))
>>>                         return -1;
>>> +       }
>>>         c.trimmed = 1;
>>>         return 0;
>>>  }
>>> --
>>> 2.19.0.605.g01d371f741-goog
>>>
>>>
>>>
>>> _______________________________________________
>>> Linux-f2fs-devel mailing list
>>> Linux-f2fs-devel@lists.sourceforge.net
>>> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
>>>
> 
> 
> _______________________________________________
> Linux-f2fs-devel mailing list
> Linux-f2fs-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
> 


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Reply via email to