Question about STEP_DECOMPRESS_NOWQ

Question about STEP_DECOMPRESS_NOWQ

[Daeho Jeong]

Hi Chao,

I have a question about the below flag for decompression.

STEP_DECOMPRESS_NOWQ,           /* handle normal cluster data inplace */

According to the comment, you added this for using inplace
decompression but inplace decompression mode is not being activated
now, since we are setting STEP_DECOMPRESS right after bio_add_page().

                if (bio_add_page(bio, page, blocksize, 0) < blocksize)
                        goto submit_and_realloc;

                /* tag STEP_DECOMPRESS to handle IO in wq */
                ctx = bio->bi_private;
                if (!(ctx->enabled_steps & (1 << STEP_DECOMPRESS)))
                        ctx->enabled_steps |= 1 << STEP_DECOMPRESS;

Did you mean to deactivate inplace decompression mode on purpose,
since it had an issue? I am a little bit confused about this.

Re: [f2fs-dev] Question about STEP_DECOMPRESS_NOWQ

[Chao Yu]

Hi Daeho,

On 2020/9/9 12:43, Daeho Jeong wrote:
> Hi Chao,
> 
> I have a question about the below flag for decompression.
> 
> STEP_DECOMPRESS_NOWQ,           /* handle normal cluster data inplace */
> 
> According to the comment, you added this for using inplace

I added this for the condition that: in compressed inode, partial clusters are
written as normal (non-compressed) one due to bad compress ratio, during reading
data in normal cluster, we don't need to queue them in workqueue, instead, we
can just handle them in IRQ context (end_io()).

That says, if all pages in bio are non-compressed, we tag step w/ STEP_DECOMPRESS_NOWQ,
once there is at least one page is compressed, we change step to
STEP_DECOMPRESS.

> decompression but inplace decompression mode is not being activated

'inplace decompression' means decompress in
a. reader thread or
b. IRQ context ?

Thanks

> now, since we are setting STEP_DECOMPRESS right after bio_add_page().
> 
>                  if (bio_add_page(bio, page, blocksize, 0) < blocksize)
>                          goto submit_and_realloc;
> 
>                  /* tag STEP_DECOMPRESS to handle IO in wq */
>                  ctx = bio->bi_private;
>                  if (!(ctx->enabled_steps & (1 << STEP_DECOMPRESS)))
>                          ctx->enabled_steps |= 1 << STEP_DECOMPRESS;
> 
> Did you mean to deactivate inplace decompression mode on purpose,
> since it had an issue? I am a little bit confused about this.
> 
> 
> _______________________________________________
> Linux-f2fs-devel mailing list
> Linux-f2fs-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
> .
> 

Re: [f2fs-dev] Question about STEP_DECOMPRESS_NOWQ

[Daeho Jeong]

Got it. I misunderstood that is some residue of trying to implement
softirq context decompression. :)

2020년 9월 9일 (수) 오후 4:15, Chao Yu <yuchao0@huawei.com>님이 작성:
>
> Hi Daeho,
>
> On 2020/9/9 12:43, Daeho Jeong wrote:
> > Hi Chao,
> >
> > I have a question about the below flag for decompression.
> >
> > STEP_DECOMPRESS_NOWQ,           /* handle normal cluster data inplace */
> >
> > According to the comment, you added this for using inplace
>
> I added this for the condition that: in compressed inode, partial clusters are
> written as normal (non-compressed) one due to bad compress ratio, during reading
> data in normal cluster, we don't need to queue them in workqueue, instead, we
> can just handle them in IRQ context (end_io()).
>
> That says, if all pages in bio are non-compressed, we tag step w/ STEP_DECOMPRESS_NOWQ,
> once there is at least one page is compressed, we change step to
> STEP_DECOMPRESS.
>
> > decompression but inplace decompression mode is not being activated
>
> 'inplace decompression' means decompress in
> a. reader thread or
> b. IRQ context ?
>
> Thanks
>
> > now, since we are setting STEP_DECOMPRESS right after bio_add_page().
> >
> >                  if (bio_add_page(bio, page, blocksize, 0) < blocksize)
> >                          goto submit_and_realloc;
> >
> >                  /* tag STEP_DECOMPRESS to handle IO in wq */
> >                  ctx = bio->bi_private;
> >                  if (!(ctx->enabled_steps & (1 << STEP_DECOMPRESS)))
> >                          ctx->enabled_steps |= 1 << STEP_DECOMPRESS;
> >
> > Did you mean to deactivate inplace decompression mode on purpose,
> > since it had an issue? I am a little bit confused about this.
> >
> >
> > _______________________________________________
> > Linux-f2fs-devel mailing list
> > Linux-f2fs-devel@lists.sourceforge.net
> > https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
> > .
> >