| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Tue, 15 Feb 2022 19:21:23 -0800
From: Dan Williams <dan.j.williams@...el.com>
To: Pankaj Gupta <pankaj.gupta.linux@...il.com>
Cc: Linux NVDIMM <nvdimm@...ts.linux.dev>,
virtualization@...ts.linux-foundation.org,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
jmoyer <jmoyer@...hat.com>,
Stefan Hajnoczi <stefanha@...hat.com>,
David Hildenbrand <david@...hat.com>,
"Michael S. Tsirkin" <mst@...hat.com>,
Cornelia Huck <cohuck@...hat.com>,
Vishal L Verma <vishal.l.verma@...el.com>,
Dave Jiang <dave.jiang@...el.com>,
"Weiny, Ira" <ira.weiny@...el.com>,
Pankaj Gupta <pankaj.gupta@...os.com>
Subject: Re: [RFC v3 1/2] virtio-pmem: Async virtio-pmem flush
On Tue, Jan 11, 2022 at 8:23 AM Pankaj Gupta
<pankaj.gupta.linux@...il.com> wrote:
>
> Enable asynchronous flush for virtio pmem using work queue. Also,
> coalesce the flush requests when a flush is already in process.
> This functionality is copied from md/RAID code.
>
> When a flush is already in process, new flush requests wait till
> previous flush completes in another context (work queue). For all
> the requests come between ongoing flush and new flush start time, only
> single flush executes, thus adhers to flush coalscing logic. This is
s/adhers/adheres/
s/coalscing/coalescing/
> important for maintaining the flush request order with request coalscing.
s/coalscing/coalescing/
>
> Signed-off-by: Pankaj Gupta <pankaj.gupta.linux@...il.com>
> ---
> drivers/nvdimm/nd_virtio.c | 74 +++++++++++++++++++++++++++---------
> drivers/nvdimm/virtio_pmem.c | 10 +++++
> drivers/nvdimm/virtio_pmem.h | 16 ++++++++
> 3 files changed, 83 insertions(+), 17 deletions(-)
>
> diff --git a/drivers/nvdimm/nd_virtio.c b/drivers/nvdimm/nd_virtio.c
> index 10351d5b49fa..179ea7a73338 100644
> --- a/drivers/nvdimm/nd_virtio.c
> +++ b/drivers/nvdimm/nd_virtio.c
> @@ -100,26 +100,66 @@ static int virtio_pmem_flush(struct nd_region *nd_region)
> /* The asynchronous flush callback function */
> int async_pmem_flush(struct nd_region *nd_region, struct bio *bio)
> {
> - /*
> - * Create child bio for asynchronous flush and chain with
> - * parent bio. Otherwise directly call nd_region flush.
> + /* queue asynchronous flush and coalesce the flush requests */
> + struct virtio_device *vdev = nd_region->provider_data;
> + struct virtio_pmem *vpmem = vdev->priv;
> + ktime_t req_start = ktime_get_boottime();
> + int ret = -EINPROGRESS;
> +
> + spin_lock_irq(&vpmem->lock);
Why a new lock and not continue to use ->pmem_lock?
Have you tested this with CONFIG_PROVE_LOCKING?
Along those lines do you have a selftest that can be added to the
kernel as well so that 0day or other bots could offer early warnings
on regressions?
> + /* flush requests wait until ongoing flush completes,
> + * hence coalescing all the pending requests.
> */
> - if (bio && bio->bi_iter.bi_sector != -1) {
> - struct bio *child = bio_alloc(GFP_ATOMIC, 0);
> -
> - if (!child)
> - return -ENOMEM;
> - bio_copy_dev(child, bio);
> - child->bi_opf = REQ_PREFLUSH;
> - child->bi_iter.bi_sector = -1;
> - bio_chain(child, bio);
> - submit_bio(child);
> - return 0;
> + wait_event_lock_irq(vpmem->sb_wait,
> + !vpmem->flush_bio ||
> + ktime_before(req_start, vpmem->prev_flush_start),
> + vpmem->lock);
> + /* new request after previous flush is completed */
> + if (ktime_after(req_start, vpmem->prev_flush_start)) {
> + WARN_ON(vpmem->flush_bio);
> + vpmem->flush_bio = bio;
> + bio = NULL;
> + }
> + spin_unlock_irq(&vpmem->lock);
> +
> + if (!bio)
> + queue_work(vpmem->pmem_wq, &vpmem->flush_work);
> + else {
> + /* flush completed in other context while we waited */
> + if (bio && (bio->bi_opf & REQ_PREFLUSH))
> + bio->bi_opf &= ~REQ_PREFLUSH;
> + else if (bio && (bio->bi_opf & REQ_FUA))
> + bio->bi_opf &= ~REQ_FUA;
> +
> + ret = vpmem->prev_flush_err;
> }
> - if (virtio_pmem_flush(nd_region))
> - return -EIO;
>
> - return 0;
> + return ret;
> };
> EXPORT_SYMBOL_GPL(async_pmem_flush);
> +
> +void submit_async_flush(struct work_struct *ws)
This name is too generic to be exported from drivers/nvdimm/nd_virtio.c
...it strikes me that there is little reason for nd_virtio and
virtio_pmem to be separate modules. They are both enabled by the same
Kconfig, so why not combine them into one module and drop the exports?
> +{
> + struct virtio_pmem *vpmem = container_of(ws, struct virtio_pmem, flush_work);
> + struct bio *bio = vpmem->flush_bio;
> +
> + vpmem->start_flush = ktime_get_boottime();
> + vpmem->prev_flush_err = virtio_pmem_flush(vpmem->nd_region);
> + vpmem->prev_flush_start = vpmem->start_flush;
> + vpmem->flush_bio = NULL;
> + wake_up(&vpmem->sb_wait);
> +
> + if (vpmem->prev_flush_err)
> + bio->bi_status = errno_to_blk_status(-EIO);
> +
> + /* Submit parent bio only for PREFLUSH */
> + if (bio && (bio->bi_opf & REQ_PREFLUSH)) {
> + bio->bi_opf &= ~REQ_PREFLUSH;
> + submit_bio(bio);
> + } else if (bio && (bio->bi_opf & REQ_FUA)) {
> + bio->bi_opf &= ~REQ_FUA;
> + bio_endio(bio);
> + }
> +}
> +EXPORT_SYMBOL_GPL(submit_async_flush);
> MODULE_LICENSE("GPL");
> diff --git a/drivers/nvdimm/virtio_pmem.c b/drivers/nvdimm/virtio_pmem.c
> index 726c7354d465..75ed9b7ddea1 100644
> --- a/drivers/nvdimm/virtio_pmem.c
> +++ b/drivers/nvdimm/virtio_pmem.c
> @@ -24,6 +24,7 @@ static int init_vq(struct virtio_pmem *vpmem)
> return PTR_ERR(vpmem->req_vq);
>
> spin_lock_init(&vpmem->pmem_lock);
> + spin_lock_init(&vpmem->lock);
> INIT_LIST_HEAD(&vpmem->req_list);
>
> return 0;
> @@ -57,7 +58,14 @@ static int virtio_pmem_probe(struct virtio_device *vdev)
> dev_err(&vdev->dev, "failed to initialize virtio pmem vq's\n");
> goto out_err;
> }
> + vpmem->pmem_wq = alloc_workqueue("vpmem_wq", WQ_MEM_RECLAIM, 0);
> + if (!vpmem->pmem_wq) {
> + err = -ENOMEM;
> + goto out_err;
> + }
>
> + INIT_WORK(&vpmem->flush_work, submit_async_flush);
> + init_waitqueue_head(&vpmem->sb_wait);
> virtio_cread_le(vpmem->vdev, struct virtio_pmem_config,
> start, &vpmem->start);
> virtio_cread_le(vpmem->vdev, struct virtio_pmem_config,
> @@ -90,10 +98,12 @@ static int virtio_pmem_probe(struct virtio_device *vdev)
> goto out_nd;
> }
> nd_region->provider_data = dev_to_virtio(nd_region->dev.parent->parent);
> + vpmem->nd_region = nd_region;
> return 0;
> out_nd:
> nvdimm_bus_unregister(vpmem->nvdimm_bus);
> out_vq:
> + destroy_workqueue(vpmem->pmem_wq);
> vdev->config->del_vqs(vdev);
> out_err:
> return err;
> diff --git a/drivers/nvdimm/virtio_pmem.h b/drivers/nvdimm/virtio_pmem.h
> index 0dddefe594c4..495dc20e1cdb 100644
> --- a/drivers/nvdimm/virtio_pmem.h
> +++ b/drivers/nvdimm/virtio_pmem.h
> @@ -35,9 +35,24 @@ struct virtio_pmem {
> /* Virtio pmem request queue */
> struct virtqueue *req_vq;
>
> + struct bio *flush_bio;
> + /* last_flush is when the last completed flush was started */
> + ktime_t prev_flush_start, start_flush;
> + int prev_flush_err;
> +
> + /* work queue for deferred flush */
> + struct work_struct flush_work;
> + struct workqueue_struct *pmem_wq;
> +
> + /* Synchronize flush wait queue data */
> + spinlock_t lock;
> + /* for waiting for previous flush to complete */
> + wait_queue_head_t sb_wait;
> +
> /* nvdimm bus registers virtio pmem device */
> struct nvdimm_bus *nvdimm_bus;
> struct nvdimm_bus_descriptor nd_desc;
> + struct nd_region *nd_region;
>
> /* List to store deferred work if virtqueue is full */
> struct list_head req_list;
> @@ -52,4 +67,5 @@ struct virtio_pmem {
>
> void virtio_pmem_host_ack(struct virtqueue *vq);
> int async_pmem_flush(struct nd_region *nd_region, struct bio *bio);
> +void submit_async_flush(struct work_struct *ws);
> #endif
> --
> 2.25.1
>
>
Powered by blists - more mailing lists