Re: [PATCH RFC] yet more struct scsi_lun

[Jens Axboe <axboe@suse.de>]

On Thu, Nov 03 2005, Mike Christie wrote:
> Jens Axboe wrote:
> >
> >On the SCSI side, I would suggest just making shost->max_sectors set
> >q->max_hw_sectors and leave q->max_sectors to some generic kernel-wide
> >block layer define (of course making sure that ->max_sectors <=
> 
> If the value for this block layer define was around 16,000 sectors, 
> would that be ok? The reason I ask is becuase when I get .....

Nope :)

> >->max_hw_sectors). That's the easy part.
> >
> >The bio_add_page() stuff is a little trickier, since it wants to know if
> >this is fs or 'generic' io. For fs io, we would like to cap the building
> >of the bio to ->max_sectors, but for eg SG_IO issued io it should go as
> >high as ->max_hw_sectors. Perhaps the easiest is just to have
> >bio_fs_add_page() and bio_pc_add_page(), each just passing in the max
> >value as an integer to bio_add_page(). But it's not exactly pretty.
> >
> >The ll_rw_blk.c merging is easy, since you don't need to do anything
> >there. It should test against ->max_sectors as it already does, since
> >this (sadly) is still the primary way we build large ios.
> >
> 
> .... here, I am running into a problem. Basically, as you know the 
> largest BIO we can make is 1 MB due to BIO_MAX_PAGES, and for st and sg 
> we need to support commands around 6 MB, so we would have a request with 
>  6 BIOs. To make this monster request I wanted to use the block layer 
> functions and do something like this:
> 
> +       for (i = 0; i < nsegs; i++) {
> +               bio = bio_map_pages(q, sg[i].page, sg[i].length, 
> sg[i].offset, gfp);
> +               if (IS_ERR(bio)) {
> +                       err = PTR_ERR(bio);
> +                       goto free_bios;
> +               }
> +               len += sg[i].length;
> +
> +               bio->bi_flags &= ~(1 << BIO_SEG_VALID);
> +               if (rq_data_dir(rq) == WRITE)
> +                       bio->bi_rw |= (1 << BIO_RW);
> +               blk_queue_bounce(q, &bio);
> +
> +               if (i == 0)
> +                       blk_rq_bio_prep(q, rq, bio);
> 
> /* hope to carve out the __make_request code that does the below 
> operations and make a fucntion that can be shared */
> 
> +               else if (!q->back_merge_fn(q, rq, bio)) {
> +                       err = -EINVAL;
> +                       bio_endio(bio, bio->bi_size, 0);
> +                       goto free_bios;
> +               } else {
> +                       rq->biotail->bi_next = bio;
> +                       rq->biotail = bio;
> +                       rq->hard_nr_sectors += bio_sectors(bio);
> +                       rq->nr_sectors = rq->hard_nr_sectors;
> 
> ........
> 
> But since q->back_merge_fn() tests against q->max_sectors, it must be a 
> high value so that we can merge in those BIOs. I mean if q->max_sectors 
> is some reasonable number like only 1024 sectors, q->back_merge_fn will 
> return a failure. Should I instead seperate ll_back_merge_fn into two 
> functions, one that checks the sectors and one that checks the segments 
> or if ll_back_merge_fn tested for max_hw_sectors we would be ok too?

It will take a whole lot of factoring out to make this happen I fear,
the results will not be easier for the eyes.

How about just keeping it simple - add a bio and request flag that
basically just says "don't honor soft max size" and whenever that is
set, you check for ->max_hw_sectors instead of ->max_sectors? Might even
be enough to just do this for the request and just require stuffing more
bio's into the request. But the bio has plenty of flag room left, so...

-- 
Jens Axboe