Re: [PATCH v2 16/16] iomap: Make readpage synchronous

[Matthew Wilcox <willy@xxxxxxxxxxxxx>]

On Thu, Oct 15, 2020 at 10:42:03AM +0100, Christoph Hellwig wrote:
> > +static void iomap_read_page_end_io(struct bio_vec *bvec,
> > +		struct completion *done, bool error)
> 
> I really don't like the parameters here.  Part of the problem is
> that ctx is only assigned to bi_private conditionally, which can
> easily be fixed.  The other part is the strange bool error when
> we can just pass on bi_stats.  See the patch at the end of what
> I'd do intead.

I prefer assigning ctx conditionally to propagating the knowledge
that !rac means synchronous.  I've gone with this:

 static void iomap_read_page_end_io(struct bio_vec *bvec,
-               struct completion *done, bool error)
+               struct iomap_readpage_ctx *ctx, blk_status_t status)
 {
        struct page *page = bvec->bv_page;
        struct iomap_page *iop = to_iomap_page(page);
 
-       if (!error)
+       if (status == BLK_STS_OK) {
                iomap_set_range_uptodate(page, bvec->bv_offset, bvec->bv_len);
+       } else if (ctx && ctx->status == BLK_STS_OK) {
+               ctx->status = status;
+       }
 
        if (!iop ||
            atomic_sub_and_test(bvec->bv_len, &iop->read_bytes_pending)) {
-               if (done)
-                       complete(done);
+               if (ctx)
+                       complete(&ctx->done);
                else
                        unlock_page(page);
        }

> >  	} else {
> >  		WARN_ON_ONCE(ctx.cur_page_in_bio);
> > -		unlock_page(page);
> > +		complete(&ctx.done);
> >  	}
> >  
> > +	wait_for_completion(&ctx.done);
> 
> I don't think we need the complete / wait_for_completion dance in
> this case.
> 
> > +	if (ret >= 0)
> > +		ret = blk_status_to_errno(ctx.status);
> > +	if (ret == 0)
> > +		return AOP_UPDATED_PAGE;
> > +	unlock_page(page);
> > +	return ret;
> 
> Nipick, but I'd rather have a goto out_unlock for both error case
> and have the AOP_UPDATED_PAGE for the normal path straight in line.
> 
> Here is an untested patch with my suggestions:

I think we can go a little further here:

@@ -340,16 +335,12 @@ iomap_readpage(struct page *page, const struct iomap_ops *
ops)
 
        if (ctx.bio) {
                submit_bio(ctx.bio);
-               WARN_ON_ONCE(!ctx.cur_page_in_bio);
-       } else {
-               WARN_ON_ONCE(ctx.cur_page_in_bio);
-               complete(&ctx.done);
+               wait_for_completion(&ctx.done);
+               if (ret > 0)
+                       ret = blk_status_to_errno(ctx.status);
        }
 
-       wait_for_completion(&ctx.done);
        if (ret >= 0)
-               ret = blk_status_to_errno(ctx.status);
-       if (ret == 0)
                return AOP_UPDATED_PAGE;
        unlock_page(page);
        return ret;


... there's no need to call blk_status_to_errno if we never submitted a bio.