From: Tiwei Bie <tiwei.bie@intel.com>
To: Maxime Coquelin <maxime.coquelin@redhat.com>
Cc: yliu@fridaylinux.org, jianfeng.tan@intel.com,
jfreimann@redhat.com, ferruh.yigit@intel.com, dev@dpdk.org,
stable@dpdk.org
Subject: Re: [dpdk-stable] [PATCH] vhost: fix iotlb pool out-of-memory handling
Date: Fri, 26 Jan 2018 16:03:16 +0800 [thread overview]
Message-ID: <20180126080316.xm7gqvozuaasqvfl@debian-xvivbkq.sh.intel.com> (raw)
In-Reply-To: <20180125150653.32284-1-maxime.coquelin@redhat.com>
On Thu, Jan 25, 2018 at 04:06:53PM +0100, Maxime Coquelin wrote:
> In the unlikely case the IOTLB memory pool runs out of memory,
> an issue may happen if all entries are used by the IOTLB cache,
> and an IOTLB miss happen. If the iotlb pending list is empty,
> then no memory is freed and allocation fails a second time.
>
> This patch fixes this by doing an IOTLB cache random evict if
> the IOTLB pending list is empty, ensuring the second allocation
> try will succeed.
>
> In the same spirit, the opposite is done when inserting an
> IOTLB entry in the IOTLB cache fails due to out of memory. In
> this case, the IOTLB pending is flushed if the IOTLB cache is
> empty to ensure the new entry can be inserted.
>
> Fixes: d012d1f293f4 ("vhost: add IOTLB helper functions")
> Fixes: f72c2ad63aeb ("vhost: add pending IOTLB miss request list and helpers")
>
[...]
> @@ -95,9 +98,11 @@ vhost_user_iotlb_pending_insert(struct vhost_virtqueue *vq,
>
> ret = rte_mempool_get(vq->iotlb_pool, (void **)&node);
> if (ret) {
> - RTE_LOG(INFO, VHOST_CONFIG,
> - "IOTLB pool empty, clear pending misses\n");
> - vhost_user_iotlb_pending_remove_all(vq);
> + RTE_LOG(INFO, VHOST_CONFIG, "IOTLB pool empty, clear entries\n");
> + if (!TAILQ_EMPTY(&vq->iotlb_pending_list))
> + vhost_user_iotlb_pending_remove_all(vq);
> + else
> + vhost_user_iotlb_cache_random_evict(vq);
> ret = rte_mempool_get(vq->iotlb_pool, (void **)&node);
> if (ret) {
> RTE_LOG(ERR, VHOST_CONFIG, "IOTLB pool still empty, failure\n");
> @@ -186,8 +191,11 @@ vhost_user_iotlb_cache_insert(struct vhost_virtqueue *vq, uint64_t iova,
>
> ret = rte_mempool_get(vq->iotlb_pool, (void **)&new_node);
> if (ret) {
> - RTE_LOG(DEBUG, VHOST_CONFIG, "IOTLB pool empty, evict one entry\n");
> - vhost_user_iotlb_cache_random_evict(vq);
> + RTE_LOG(DEBUG, VHOST_CONFIG, "IOTLB pool empty, clear entries\n");
Maybe we should use the same log level for both cases?
Currently, the first one is INFO, and this one is DEBUG.
> + if (!TAILQ_EMPTY(&vq->iotlb_list))
> + vhost_user_iotlb_cache_random_evict(vq);
> + else
> + vhost_user_iotlb_pending_remove_all(vq);
The IOTLB entry will be invalidated explicitly if it's
unmapped in the frontend. So all the IOTLB entries in
IOTLB cache are supposed to be valid. So I think maybe
it's better to always prefer to free memory from IOTLB
pending list. Something like:
if (TAILQ_EMPTY(&vq->iotlb_list))
vhost_user_iotlb_pending_remove_all(vq);
else if (TAILQ_EMPTY(&vq->iotlb_pending_list))
vhost_user_iotlb_cache_random_evict(vq);
else
vhost_user_iotlb_pending_random_evict(vq);
Besides, in __vhost_iova_to_vva():
uint64_t
__vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
uint64_t iova, uint64_t size, uint8_t perm)
{
uint64_t vva, tmp_size;
if (unlikely(!size))
return 0;
tmp_size = size;
vva = vhost_user_iotlb_cache_find(vq, iova, &tmp_size, perm);
if (tmp_size == size)
return vva;
if (!vhost_user_iotlb_pending_miss(vq, iova + tmp_size, perm)) {
/*
* iotlb_lock is read-locked for a full burst,
* but it only protects the iotlb cache.
* In case of IOTLB miss, we might block on the socket,
* which could cause a deadlock with QEMU if an IOTLB update
* is being handled. We can safely unlock here to avoid it.
*/
vhost_user_iotlb_rd_unlock(vq);
vhost_user_iotlb_pending_insert(vq, iova + tmp_size, perm);
vhost_user_iotlb_miss(dev, iova + tmp_size, perm);
The vhost_user_iotlb_miss() may fail. If it fails,
the inserted pending entry should be removed from
the pending list.
Best regards,
Tiwei Bie
next prev parent reply other threads:[~2018-01-26 8:03 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-01-25 15:06 Maxime Coquelin
2018-01-26 8:03 ` Tiwei Bie [this message]
2018-01-26 15:23 ` Maxime Coquelin
2018-01-26 12:40 ` [dpdk-stable] [dpdk-dev] " Jens Freimann
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20180126080316.xm7gqvozuaasqvfl@debian-xvivbkq.sh.intel.com \
--to=tiwei.bie@intel.com \
--cc=dev@dpdk.org \
--cc=ferruh.yigit@intel.com \
--cc=jfreimann@redhat.com \
--cc=jianfeng.tan@intel.com \
--cc=maxime.coquelin@redhat.com \
--cc=stable@dpdk.org \
--cc=yliu@fridaylinux.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).