DPDK patches and discussions
 help / color / mirror / Atom feed
From: Ilya Maximets <i.maximets@samsung.com>
To: Tiwei Bie <tiwei.bie@intel.com>,
	maxime.coquelin@redhat.com, zhihong.wang@intel.com, dev@dpdk.org
Subject: Re: [dpdk-dev] [1/4] net/virtio: fix the control vq support
Date: Wed, 23 Jan 2019 19:33:21 +0300	[thread overview]
Message-ID: <9799a0cf-52a5-0ea4-03d8-b812b338ce59@samsung.com> (raw)
In-Reply-To: <b58bc7de-ec34-e3a3-4d2c-d4b49836c4de@samsung.com>

Hmm. Nevermind.
Please, ignore my previous comments to this patch.
Patch seems compliant to spec, but the spec is not very clear.

Best regards, Ilya Maximets.

On 23.01.2019 16:09, Ilya Maximets wrote:
> On 22.01.2019 20:01, Tiwei Bie wrote:
>> This patch mainly fixed below issues in the packed ring based
>> control vq support in virtio driver:
>>
>> 1. When parsing the used descriptors, we have to track the
>>    number of descs that we need to skip;
>> 2. vq->vq_free_cnt was decreased twice for a same desc;
>>
>> Meanwhile, make the function name consistent with other parts.
>>
>> Fixes: ec194c2f1895 ("net/virtio: support packed queue in send command")
>> Fixes: a4270ea4ff79 ("net/virtio: check head desc with correct wrap counter")
>>
>> Signed-off-by: Tiwei Bie <tiwei.bie@intel.com>
>> ---
>>  drivers/net/virtio/virtio_ethdev.c | 62 ++++++++++++++----------------
>>  drivers/net/virtio/virtqueue.h     | 12 +-----
>>  2 files changed, 31 insertions(+), 43 deletions(-)
>>
>> diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
>> index ee5a98b7c..a3fe65599 100644
>> --- a/drivers/net/virtio/virtio_ethdev.c
>> +++ b/drivers/net/virtio/virtio_ethdev.c
>> @@ -142,16 +142,17 @@ static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
>>  struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
>>  
>>  static struct virtio_pmd_ctrl *
>> -virtio_pq_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
>> -		       int *dlen, int pkt_num)
>> +virtio_send_command_packed(struct virtnet_ctl *cvq,
>> +			   struct virtio_pmd_ctrl *ctrl,
>> +			   int *dlen, int pkt_num)
>>  {
>>  	struct virtqueue *vq = cvq->vq;
>>  	int head;
>>  	struct vring_packed_desc *desc = vq->ring_packed.desc_packed;
>>  	struct virtio_pmd_ctrl *result;
>> -	bool avail_wrap_counter, used_wrap_counter;
>> -	uint16_t flags;
>> +	bool avail_wrap_counter;
>>  	int sum = 0;
>> +	int nb_descs = 0;
>>  	int k;
>>  
>>  	/*
>> @@ -162,11 +163,10 @@ virtio_pq_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
>>  	 */
>>  	head = vq->vq_avail_idx;
>>  	avail_wrap_counter = vq->avail_wrap_counter;
>> -	used_wrap_counter = vq->used_wrap_counter;
>> -	desc[head].flags = VRING_DESC_F_NEXT;
>>  	desc[head].addr = cvq->virtio_net_hdr_mem;
>>  	desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
>>  	vq->vq_free_cnt--;
>> +	nb_descs++;
>>  	if (++vq->vq_avail_idx >= vq->vq_nentries) {
>>  		vq->vq_avail_idx -= vq->vq_nentries;
>>  		vq->avail_wrap_counter ^= 1;
>> @@ -177,55 +177,51 @@ virtio_pq_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
>>  			+ sizeof(struct virtio_net_ctrl_hdr)
>>  			+ sizeof(ctrl->status) + sizeof(uint8_t) * sum;
>>  		desc[vq->vq_avail_idx].len = dlen[k];
>> -		flags = VRING_DESC_F_NEXT;
> 
> Looks like barriers was badly placed here before this patch.
> Anyway, you need a write barrier here between {addr, len} and flags updates.
> 
>> +		desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
>> +			VRING_DESC_F_AVAIL(vq->avail_wrap_counter) |
>> +			VRING_DESC_F_USED(!vq->avail_wrap_counter);
>>  		sum += dlen[k];
>>  		vq->vq_free_cnt--;
>> -		flags |= VRING_DESC_F_AVAIL(vq->avail_wrap_counter) |
>> -			 VRING_DESC_F_USED(!vq->avail_wrap_counter);
>> -		desc[vq->vq_avail_idx].flags = flags;
>> -		rte_smp_wmb();
>> -		vq->vq_free_cnt--;
>> +		nb_descs++;
>>  		if (++vq->vq_avail_idx >= vq->vq_nentries) {
>>  			vq->vq_avail_idx -= vq->vq_nentries;
>>  			vq->avail_wrap_counter ^= 1;
>>  		}
>>  	}
>>  
>> -
>>  	desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
>>  		+ sizeof(struct virtio_net_ctrl_hdr);
>>  	desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
>> -	flags = VRING_DESC_F_WRITE;
>> -	flags |= VRING_DESC_F_AVAIL(vq->avail_wrap_counter) |
>> -		 VRING_DESC_F_USED(!vq->avail_wrap_counter);
>> -	desc[vq->vq_avail_idx].flags = flags;
>> -	flags = VRING_DESC_F_NEXT;
>> -	flags |= VRING_DESC_F_AVAIL(avail_wrap_counter) |
>> -		 VRING_DESC_F_USED(!avail_wrap_counter);
>> -	desc[head].flags = flags;
>> -	rte_smp_wmb();
>> -
> 
> Same here. We need a write barrier to be sure that {addr, len} written
> before updating flags.
> 
> Another way to avoid most of barriers is to work similar to
> 'flush_shadow_used_ring_packed',
> i.e. update all the data in a loop - write barrier - update all the flags.
> 
>> +	desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
>> +		VRING_DESC_F_AVAIL(vq->avail_wrap_counter) |
>> +		VRING_DESC_F_USED(!vq->avail_wrap_counter);
>>  	vq->vq_free_cnt--;
>> +	nb_descs++;
>>  	if (++vq->vq_avail_idx >= vq->vq_nentries) {
>>  		vq->vq_avail_idx -= vq->vq_nentries;
>>  		vq->avail_wrap_counter ^= 1;
>>  	}
>>  
>> +	virtio_wmb(vq->hw->weak_barriers);
>> +	desc[head].flags = VRING_DESC_F_NEXT |
>> +		VRING_DESC_F_AVAIL(avail_wrap_counter) |
>> +		VRING_DESC_F_USED(!avail_wrap_counter);
>> +
>> +	virtio_wmb(vq->hw->weak_barriers);
>>  	virtqueue_notify(vq);
>>  
>>  	/* wait for used descriptors in virtqueue */
>> -	do {
>> -		rte_rmb();
>> +	while (!desc_is_used(&desc[head], vq))
>>  		usleep(100);
>> -	} while (!__desc_is_used(&desc[head], used_wrap_counter));
>> +
>> +	virtio_rmb(vq->hw->weak_barriers);
>>  
>>  	/* now get used descriptors */
>> -	while (desc_is_used(&desc[vq->vq_used_cons_idx], vq)) {
>> -		vq->vq_free_cnt++;
>> -		if (++vq->vq_used_cons_idx >= vq->vq_nentries) {
>> -			vq->vq_used_cons_idx -= vq->vq_nentries;
>> -			vq->used_wrap_counter ^= 1;
>> -		}
>> +	vq->vq_free_cnt += nb_descs;
>> +	vq->vq_used_cons_idx += nb_descs;
>> +	if (vq->vq_used_cons_idx >= vq->vq_nentries) {
>> +		vq->vq_used_cons_idx -= vq->vq_nentries;
>> +		vq->used_wrap_counter ^= 1;
>>  	}
>>  
>>  	result = cvq->virtio_net_hdr_mz->addr;
>> @@ -266,7 +262,7 @@ virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
>>  		sizeof(struct virtio_pmd_ctrl));
>>  
>>  	if (vtpci_packed_queue(vq->hw)) {
>> -		result = virtio_pq_send_command(cvq, ctrl, dlen, pkt_num);
>> +		result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
>>  		goto out_unlock;
>>  	}
>>  
>> diff --git a/drivers/net/virtio/virtqueue.h b/drivers/net/virtio/virtqueue.h
>> index 7fcde5643..ca9d8e6e3 100644
>> --- a/drivers/net/virtio/virtqueue.h
>> +++ b/drivers/net/virtio/virtqueue.h
>> @@ -281,7 +281,7 @@ struct virtio_tx_region {
>>  };
>>  
>>  static inline int
>> -__desc_is_used(struct vring_packed_desc *desc, bool wrap_counter)
>> +desc_is_used(struct vring_packed_desc *desc, struct virtqueue *vq)
>>  {
>>  	uint16_t used, avail, flags;
>>  
>> @@ -289,16 +289,9 @@ __desc_is_used(struct vring_packed_desc *desc, bool wrap_counter)
>>  	used = !!(flags & VRING_DESC_F_USED(1));
>>  	avail = !!(flags & VRING_DESC_F_AVAIL(1));
>>  
>> -	return avail == used && used == wrap_counter;
>> +	return avail == used && used == vq->used_wrap_counter;
>>  }
>>  
>> -static inline int
>> -desc_is_used(struct vring_packed_desc *desc, struct virtqueue *vq)
>> -{
>> -	return __desc_is_used(desc, vq->used_wrap_counter);
>> -}
>> -
>> -
>>  static inline void
>>  vring_desc_init_packed(struct virtqueue *vq, int n)
>>  {
>> @@ -354,7 +347,6 @@ virtqueue_enable_intr_packed(struct virtqueue *vq)
>>  {
>>  	uint16_t *event_flags = &vq->ring_packed.driver_event->desc_event_flags;
>>  
>> -
>>  	if (vq->event_flags_shadow == RING_EVENT_FLAGS_DISABLE) {
>>  		virtio_wmb(vq->hw->weak_barriers);
>>  		vq->event_flags_shadow = RING_EVENT_FLAGS_ENABLE;
>>

  parent reply	other threads:[~2019-01-23 16:33 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-01-22 17:01 [dpdk-dev] [PATCH 0/4] Virtio fixes Tiwei Bie
2019-01-22 17:01 ` [dpdk-dev] [PATCH 1/4] net/virtio: fix the control vq support Tiwei Bie
     [not found]   ` <CGME20190123130903eucas1p2730776e71039a79024dd7602b5dcad7d@eucas1p2.samsung.com>
2019-01-23 13:09     ` [dpdk-dev] [1/4] " Ilya Maximets
     [not found]       ` <CGME20190123163323eucas1p1baaec2c637cdc656ab9b26dbfd455bae@eucas1p1.samsung.com>
2019-01-23 16:33         ` Ilya Maximets [this message]
2019-01-23 22:02           ` Maxime Coquelin
2019-01-23 22:03   ` [dpdk-dev] [PATCH 1/4] " Maxime Coquelin
2019-01-22 17:01 ` [dpdk-dev] [PATCH 2/4] net/virtio-user: " Tiwei Bie
2019-01-23 22:07   ` Maxime Coquelin
2019-01-22 17:01 ` [dpdk-dev] [PATCH 3/4] net/virtio: use virtio barrier in packed ring Tiwei Bie
     [not found]   ` <CGME20190123155232eucas1p28bdd3a5c220452b81e21e48c19f3e5a7@eucas1p2.samsung.com>
2019-01-23 15:52     ` [dpdk-dev] [3/4] " Ilya Maximets
2019-01-23 22:09   ` [dpdk-dev] [PATCH 3/4] " Maxime Coquelin
2019-01-22 17:01 ` [dpdk-dev] [PATCH 4/4] net/virtio-user: fix used ring update in cvq handling Tiwei Bie
2019-01-23 22:08   ` Maxime Coquelin
2019-01-23 22:25 ` [dpdk-dev] [PATCH 0/4] Virtio fixes Maxime Coquelin

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=9799a0cf-52a5-0ea4-03d8-b812b338ce59@samsung.com \
    --to=i.maximets@samsung.com \
    --cc=dev@dpdk.org \
    --cc=maxime.coquelin@redhat.com \
    --cc=tiwei.bie@intel.com \
    --cc=zhihong.wang@intel.com \
    /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).