From mboxrd@z Thu Jan 1 00:00:00 1970 From: Konstantin Khorenko Date: Tue, 25 Aug 2026 13:03:53 +0200 Subject: [Devel] [PATCH RHEL10 COMMIT] drivers/vhost/blk: fix vhost_blk_req_done() softirq handling In-Reply-To: <20260817000234.337030-1-andrey.zhadchenko@virtuozzo.com> Message-ID: <202608251103.67PB3r4A778191@f0.sw.ru> List-Id: The commit is pushed to "branch-rh10-6.12.0-211.39.1.16.x.vz10-ovz" and will appear at git at bitbucket.org:openvz/vzkernel.git after rh10-6.12.0-211.39.1.16.9.vz10 ------> commit e85a71f5859f94dc45079eb3b028cfcb28ea6eeb Author: Andrey Zhadchenko Date: Mon Aug 17 03:02:34 2026 +0300 drivers/vhost/blk: fix vhost_blk_req_done() softirq handling vhost_blk_req_done() does bio_release_pages(), which may sleep on folio_lock() when we need to dirty pages. So we get this: [ 9120.053244] BUG: scheduling while atomic: swapper/1/0/0x00000102 ... [ 9120.079613] Hardware name: Acronis OpenStack Compute/Virtuozzo, BIOS 1.16.1-1.vz9.2 04/01/2014 [ 9120.079629] Call Trace: [ 9120.079632] [ 9120.079638] dump_stack_lvl+0x4e/0x70 [ 9120.079646] __schedule_bug.cold+0x3e/0x4a [ 9120.079650] schedule_debug.isra.0+0x93/0xc0 [ 9120.079655] __schedule+0x7a/0x630 [ 9120.079659] ? enqueue_task_fair+0x150/0x710 [ 9120.079664] schedule+0x27/0x80 [ 9120.079666] io_schedule+0x46/0x70 [ 9120.079669] folio_wait_bit_common+0x13a/0x340 [ 9120.079686] ? __pfx_wake_page_function+0x10/0x10 [ 9120.079690] __bio_release_pages+0x25b/0x280 [ 9120.079698] vhost_blk_req_done+0x98/0xa0 [vhost_blk] [ 9120.079702] blk_update_request+0x17c/0x420 [ 9120.079707] blk_mq_end_request+0x1c/0x30 [ 9120.079711] dm_softirq_done+0x158/0x280 [dm_mod] [ 9120.079737] blk_complete_reqs+0x40/0x50 [ 9120.079740] handle_softirqs+0xe5/0x2a0 [ 9120.079744] __irq_exit_rcu+0xbd/0xe0 [ 9120.079746] sysvec_call_function_single+0x71/0x90 [ 9120.079750] [ 9120.079751] [ 9120.079752] asm_sysvec_call_function_single+0x1a/0x20 Kernel already has a suitable helper: bio_check_pages_dirty(). It takes bio ownership and dirties pages/puts bio later in some workqueue (see bio_set_pages_dirty() annotation). Also we should manually mark the pages dirty from vhost_blk_bio_send(), as they can be modified through DMA. While at here, move request posting below bio_put. I don't expect the worker could process and reuse the request before we access req->bb and req->bi_opf, but better be safe than sorry. https://virtuozzo.atlassian.net/browse/VSTOR-141284 Signed-off-by: Andrey Zhadchenko Reviewed-by: Vasileios Almpanis --- drivers/vhost/blk.c | 24 +++++++++++++++--------- 1 file changed, 15 insertions(+), 9 deletions(-) diff --git a/drivers/vhost/blk.c b/drivers/vhost/blk.c index c03945ac04234..dac03566bfca5 100644 --- a/drivers/vhost/blk.c +++ b/drivers/vhost/blk.c @@ -156,19 +156,22 @@ static void vhost_blk_req_done(struct bio *bio) if (err) req->bio_err = err; - if (atomic_dec_and_test(&req->bio_nr)) { - llist_add(&req->llnode, &req->blk_vq->llhead); - vhost_vq_work_queue(&req->blk_vq->vq, &req->blk_vq->work); - } - /* * Bounce buffer adds kvec to bio (instead of user backed memory), * so there is no reference/pin to bio pages in this case. */ - if (!req->bb) - bio_release_pages(bio, !req->bi_opf); + if (!req->bb && req->bi_opf == REQ_OP_READ) { + bio_check_pages_dirty(bio); + } else { + if (!req->bb) + bio_release_pages(bio, false); + bio_put(bio); + } - bio_put(bio); + if (atomic_dec_and_test(&req->bio_nr)) { + llist_add(&req->llnode, &req->blk_vq->llhead); + vhost_vq_work_queue(&req->blk_vq->vq, &req->blk_vq->work); + } } static void vhost_blk_req_cleanup(struct vhost_blk_req *req) @@ -387,8 +390,11 @@ static inline void vhost_blk_bio_send(struct vhost_blk_req *req) bio_nr = atomic_read(&req->bio_nr); blk_start_plug(&plug); - for (i = 0; i < bio_nr; i++) + for (i = 0; i < bio_nr; i++) { + if (!req->bb && req->bi_opf == REQ_OP_READ) + bio_set_pages_dirty(req->bio[i]); submit_bio(req->bio[i]); + } blk_finish_plug(&plug); }