Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753899AbaBDM0R (ORCPT ); Tue, 4 Feb 2014 07:26:17 -0500 Received: from mail-pb0-f50.google.com ([209.85.160.50]:58584 "EHLO mail-pb0-f50.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751079AbaBDM0N (ORCPT ); Tue, 4 Feb 2014 07:26:13 -0500 Date: Tue, 4 Feb 2014 04:25:24 -0800 (PST) From: Hugh Dickins X-X-Sender: hugh@eggly.anvils To: Kent Overstreet cc: Hugh Dickins , "Martin K. Petersen" , Jens Axboe , Shaohua Li , Andrew Morton , linux-kernel@vger.kernel.org Subject: Re: [PATCH] block: Explicitly handle discard/write same segments In-Reply-To: <20140204101748.GA12440@kmo-pixel> Message-ID: References: <20140114023346.GN9037@kmo> <20140114044841.GO9037@kmo> <20140114222424.GA7333@moria.home.lan> <20140117012105.GS9037@kmo> <20140204101748.GA12440@kmo-pixel> User-Agent: Alpine 2.11 (LSU 23 2013-08-11) MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Tue, 4 Feb 2014, Kent Overstreet wrote: > Immutable biovecs changed the way biovecs are interpreted - drivers no > longer use bi_vcnt, they have to go by bi_iter.bi_size (to allow for > using part of an existing segment without modifying it). > > This breaks with discards and write_same bios, since for those bi_size > has nothing to do with segments in the biovec. So for now, we need a > fairly gross hack - we fortunately know that there will never be more > than one segment for the entire request, so we can special case > discard/write_same. > > Signed-off-by: Kent Overstreet > --- > On Fri, Jan 31, 2014 at 09:17:25AM -0800, Hugh Dickins wrote: > > On Thu, 16 Jan 2014, Kent Overstreet wrote: > > > > > > Ok, I reread the code and figured it out - the analagous change also has to be > > > made in __blk_segment_map_sg(). I'll mail out a patch for this tomorrow after > > > I've stared at the code more and had less beer. > > > > I'd been hoping for a patch to try, but now your changes have hit Linus's > > tree: so today we have discard broken there too, crashing as originally > > reported on the NULL struct page pointer in __blk_recalc_rq_segments()'s > > page_to_pfn(bv.bv_page). > > > > How to reproduce it? I hope you'll find easier ways, but I get it with > > swapping to SSD (remember "swapon -d" to enable discard). I'm just doing > > what I've done for years, running a pair of make -j20 kbuilds to tmpfs in > > limited RAM (I use mem=700M with 1.5G of swap: but that would be far too > > little RAM for a general config of current tree), to get plenty of fairly > > chaotic swapping but good forward progress nonetheless (if the sizes are > > too small, then it'll just thrash abysmally or be OOM-killed). > > > > But please do send me a patch and I'll give it a try - thanks. > > Hugh - can you give this patch a try? Passes my tests but I was never > able to reproduce your crash, unfortunately. Thanks a lot, Kent: I'm glad to report, this is working fine for me. Hugh > > block/blk-merge.c | 91 +++++++++++++++++++++++++++++++++++++------------------ > 1 file changed, 62 insertions(+), 29 deletions(-) > > diff --git a/block/blk-merge.c b/block/blk-merge.c > index 8f8adaa954..6c583f9c5b 100644 > --- a/block/blk-merge.c > +++ b/block/blk-merge.c > @@ -21,6 +21,16 @@ static unsigned int __blk_recalc_rq_segments(struct request_queue *q, > if (!bio) > return 0; > > + /* > + * This should probably be returning 0, but blk_add_request_payload() > + * (Christoph!!!!) > + */ > + if (bio->bi_rw & REQ_DISCARD) > + return 1; > + > + if (bio->bi_rw & REQ_WRITE_SAME) > + return 1; > + > fbio = bio; > cluster = blk_queue_cluster(q); > seg_size = 0; > @@ -161,30 +171,60 @@ new_segment: > *bvprv = *bvec; > } > > -/* > - * map a request to scatterlist, return number of sg entries setup. Caller > - * must make sure sg can hold rq->nr_phys_segments entries > - */ > -int blk_rq_map_sg(struct request_queue *q, struct request *rq, > - struct scatterlist *sglist) > +static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, > + struct scatterlist *sglist, > + struct scatterlist **sg) > { > struct bio_vec bvec, bvprv = { NULL }; > - struct req_iterator iter; > - struct scatterlist *sg; > + struct bvec_iter iter; > int nsegs, cluster; > > nsegs = 0; > cluster = blk_queue_cluster(q); > > - /* > - * for each bio in rq > - */ > - sg = NULL; > - rq_for_each_segment(bvec, rq, iter) { > - __blk_segment_map_sg(q, &bvec, sglist, &bvprv, &sg, > - &nsegs, &cluster); > - } /* segments in rq */ > + if (bio->bi_rw & REQ_DISCARD) { > + /* > + * This is a hack - drivers should be neither modifying the > + * biovec, nor relying on bi_vcnt - but because of > + * blk_add_request_payload(), a discard bio may or may not have > + * a payload we need to set up here (thank you Christoph) and > + * bi_vcnt is really the only way of telling if we need to. > + */ > + > + if (bio->bi_vcnt) > + goto single_segment; > + > + return 0; > + } > + > + if (bio->bi_rw & REQ_WRITE_SAME) { > +single_segment: > + *sg = sglist; > + bvec = bio_iovec(bio); > + sg_set_page(*sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset); > + return 1; > + } > + > + for_each_bio(bio) > + bio_for_each_segment(bvec, bio, iter) > + __blk_segment_map_sg(q, &bvec, sglist, &bvprv, sg, > + &nsegs, &cluster); > > + return nsegs; > +} > + > +/* > + * map a request to scatterlist, return number of sg entries setup. Caller > + * must make sure sg can hold rq->nr_phys_segments entries > + */ > +int blk_rq_map_sg(struct request_queue *q, struct request *rq, > + struct scatterlist *sglist) > +{ > + struct scatterlist *sg = NULL; > + int nsegs = 0; > + > + if (rq->bio) > + nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg); > > if (unlikely(rq->cmd_flags & REQ_COPY_USER) && > (blk_rq_bytes(rq) & q->dma_pad_mask)) { > @@ -230,20 +270,13 @@ EXPORT_SYMBOL(blk_rq_map_sg); > int blk_bio_map_sg(struct request_queue *q, struct bio *bio, > struct scatterlist *sglist) > { > - struct bio_vec bvec, bvprv = { NULL }; > - struct scatterlist *sg; > - int nsegs, cluster; > - struct bvec_iter iter; > - > - nsegs = 0; > - cluster = blk_queue_cluster(q); > - > - sg = NULL; > - bio_for_each_segment(bvec, bio, iter) { > - __blk_segment_map_sg(q, &bvec, sglist, &bvprv, &sg, > - &nsegs, &cluster); > - } /* segments in bio */ > + struct scatterlist *sg = NULL; > + int nsegs; > + struct bio *next = bio->bi_next; > + bio->bi_next = NULL; > > + nsegs = __blk_bios_map_sg(q, bio, sglist, &sg); > + bio->bi_next = next; > if (sg) > sg_mark_end(sg); > > -- > 1.9.rc1 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/