Use netif_queue_set_napi to map NAPIs to queue IDs so that the mapping
can be accessed by user apps. Note that the netif_queue_set_napi
currently requires RTNL, so care must be taken to ensure RTNL is held on
paths where this API might be reached.
The paths in the driver where this API can be reached appear to be:
- ndo_open, ndo_close, which hold RTNL so no driver change is needed.
- rx_pause, rx_resume, tx_pause, tx_resume are reached either via
an ethtool ioctl or via XSK - neither path requires a driver change.
- power management paths (which call open and close), which have been
updated to hold/release RTNL.
$ ethtool -i ens4 | grep driver
driver: virtio_net
$ sudo ethtool -L ens4 combined 4
$ ./tools/net/ynl/pyynl/cli.py \
--spec Documentation/netlink/specs/netdev.yaml \
--dump queue-get --json='{"ifindex": 2}'
[{'id': 0, 'ifindex': 2, 'napi-id': 8289, 'type': 'rx'},
{'id': 1, 'ifindex': 2, 'napi-id': 8290, 'type': 'rx'},
{'id': 2, 'ifindex': 2, 'napi-id': 8291, 'type': 'rx'},
{'id': 3, 'ifindex': 2, 'napi-id': 8292, 'type': 'rx'},
{'id': 0, 'ifindex': 2, 'type': 'tx'},
{'id': 1, 'ifindex': 2, 'type': 'tx'},
{'id': 2, 'ifindex': 2, 'type': 'tx'},
{'id': 3, 'ifindex': 2, 'type': 'tx'}]
Note that virtio_net has TX-only NAPIs which do not have NAPI IDs, so
the lack of 'napi-id' in the above output is expected.
Signed-off-by: Joe Damato <jdamato@fastly.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Link: https://patch.msgid.link/20250307011215.266806-4-jdamato@fastly.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Create virtnet_napi_disable helper and refactor virtnet_napi_tx_disable
to take a struct send_queue.
Signed-off-by: Joe Damato <jdamato@fastly.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Tested-by: Lei Yang <leiyang@redhat.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Link: https://patch.msgid.link/20250307011215.266806-3-jdamato@fastly.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Refactor virtnet_napi_enable and virtnet_napi_tx_enable to take a struct
receive_queue. Create a helper, virtnet_napi_do_enable, which contains
the logic to enable a NAPI.
Signed-off-by: Joe Damato <jdamato@fastly.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Tested-by: Lei Yang <leiyang@redhat.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Link: https://patch.msgid.link/20250307011215.266806-2-jdamato@fastly.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
There are several issues existed in start_xmit():
- Transmitted packets need to be freed before sending a packet, this
introduces delay and increases the average packets transmit
time. This also increase the time that spent in holding the TX lock.
- Notification is enabled after free_old_xmit_skbs() which will
introduce unnecessary interrupts if TX notification happens on the
same CPU that is doing the transmission now (actually, virtio-net
driver are optimized for this case).
So this patch tries to avoid those issues by not cleaning transmitted
packets in start_xmit() when TX NAPI is enabled and disable
notifications even more aggressively. Notification will be since the
beginning of the start_xmit(). But we can't enable delayed
notification after TX is stopped as we will lose the
notifications. Instead, the delayed notification needs is enabled
after the virtqueue is kicked for best performance.
Performance numbers:
1) single queue 2 vcpus guest with pktgen_sample03_burst_single_flow.sh
(burst 256) + testpmd (rxonly) on the host:
- When pinning TX IRQ to pktgen VCPU: split virtqueue PPS were
increased 55% from 6.89 Mpps to 10.7 Mpps and 32% TX interrupts were
eliminated. Packed virtqueue PPS were increased 50% from 7.09 Mpps to
10.7 Mpps, 99% TX interrupts were eliminated.
- When pinning TX IRQ to VCPU other than pktgen: split virtqueue PPS
were increased 96% from 5.29 Mpps to 10.4 Mpps and 45% TX interrupts
were eliminated; Packed virtqueue PPS were increased 78% from 6.12
Mpps to 10.9 Mpps and 99% TX interrupts were eliminated.
2) single queue 1 vcpu guest + vhost-net/TAP on the host: single
session netperf from guest to host shows 82% improvement from
31Gb/s to 58Gb/s, %stddev were reduced from 34.5% to 1.9% and 88%
of TX interrupts were eliminated.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
virtnet_sq_bind_xsk_pool() flushes tx skbs and then resets tx queue, so
DQL counters need to be reset when flushing has actually occurred, Add
virtnet_sq_free_unused_buf_done() as a callback for virtqueue_resize()
to handle this.
Fixes: 21a4e3ce6d ("virtio_net: xsk: bind/unbind xsk for tx")
Signed-off-by: Koichiro Den <koichiro.den@canonical.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When virtqueue_reset() has actually recycled all unused buffers,
additional work may be required in some cases. Relying solely on its
return status is fragile, so introduce a new function argument
'recycle_done', which is invoked when it really occurs.
Signed-off-by: Koichiro Den <koichiro.den@canonical.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
virtnet_tx_resize() flushes remaining tx skbs, requiring DQL counters to
be reset when flushing has actually occurred. Add
virtnet_sq_free_unused_buf_done() as a callback for virtqueue_reset() to
handle this.
Fixes: c8bd1f7f3e ("virtio_net: add support for Byte Queue Limits")
Cc: <stable@vger.kernel.org> # v6.11+
Signed-off-by: Koichiro Den <koichiro.den@canonical.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When virtqueue_resize() has actually recycled all unused buffers,
additional work may be required in some cases. Relying solely on its
return status is fragile, so introduce a new function argument
'recycle_done', which is invoked when the recycle really occurs.
Cc: <stable@vger.kernel.org> # v6.11+
Signed-off-by: Koichiro Den <koichiro.den@canonical.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
While not harmful, using vq2rxq where it's always sq appears odd.
Replace it with the more appropriate vq2txq for clarity and correctness.
Fixes: 89f86675cb ("virtio_net: xsk: tx: support xmit xsk buffer")
Signed-off-by: Koichiro Den <koichiro.den@canonical.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When virtnet_close is followed by virtnet_open, some TX completions can
possibly remain unconsumed, until they are finally processed during the
first NAPI poll after the netdev_tx_reset_queue(), resulting in a crash
[1]. Commit b96ed2c97c ("virtio_net: move netdev_tx_reset_queue() call
before RX napi enable") was not sufficient to eliminate all BQL crash
cases for virtio-net.
This issue can be reproduced with the latest net-next master by running:
`while :; do ip l set DEV down; ip l set DEV up; done` under heavy network
TX load from inside the machine.
netdev_tx_reset_queue() can actually be dropped from virtnet_open path;
the device is not stopped in any case. For BQL core part, it's just like
traffic nearly ceases to exist for some period. For stall detector added
to BQL, even if virtnet_close could somehow lead to some TX completions
delayed for long, followed by virtnet_open, we can just take it as stall
as mentioned in commit 6025b9135f ("net: dqs: add NIC stall detector
based on BQL"). Note also that users can still reset stall_max via sysfs.
So, drop netdev_tx_reset_queue() from virtnet_enable_queue_pair(). This
eliminates the BQL crashes. As a result, netdev_tx_reset_queue() is now
explicitly required in freeze/restore path. This patch adds it to
immediately after free_unused_bufs(), following the rule of thumb:
netdev_tx_reset_queue() should follow any SKB freeing not followed by
netdev_tx_completed_queue(). This seems the most consistent and
streamlined approach, and now netdev_tx_reset_queue() runs whenever
free_unused_bufs() is done.
[1]:
------------[ cut here ]------------
kernel BUG at lib/dynamic_queue_limits.c:99!
Oops: invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
CPU: 7 UID: 0 PID: 1598 Comm: ip Tainted: G N 6.12.0net-next_main+ #2
Tainted: [N]=TEST
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), \
BIOS rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014
RIP: 0010:dql_completed+0x26b/0x290
Code: b7 c2 49 89 e9 44 89 da 89 c6 4c 89 d7 e8 ed 17 47 00 58 65 ff 0d
4d 27 90 7e 0f 85 fd fe ff ff e8 ea 53 8d ff e9 f3 fe ff ff <0f> 0b 01
d2 44 89 d1 29 d1 ba 00 00 00 00 0f 48 ca e9 28 ff ff ff
RSP: 0018:ffffc900002b0d08 EFLAGS: 00010297
RAX: 0000000000000000 RBX: ffff888102398c80 RCX: 0000000080190009
RDX: 0000000000000000 RSI: 000000000000006a RDI: 0000000000000000
RBP: ffff888102398c00 R08: 0000000000000000 R09: 0000000000000000
R10: 00000000000000ca R11: 0000000000015681 R12: 0000000000000001
R13: ffffc900002b0d68 R14: ffff88811115e000 R15: ffff8881107aca40
FS: 00007f41ded69500(0000) GS:ffff888667dc0000(0000)
knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000556ccc2dc1a0 CR3: 0000000104fd8003 CR4: 0000000000772ef0
PKRU: 55555554
Call Trace:
<IRQ>
? die+0x32/0x80
? do_trap+0xd9/0x100
? dql_completed+0x26b/0x290
? dql_completed+0x26b/0x290
? do_error_trap+0x6d/0xb0
? dql_completed+0x26b/0x290
? exc_invalid_op+0x4c/0x60
? dql_completed+0x26b/0x290
? asm_exc_invalid_op+0x16/0x20
? dql_completed+0x26b/0x290
__free_old_xmit+0xff/0x170 [virtio_net]
free_old_xmit+0x54/0xc0 [virtio_net]
virtnet_poll+0xf4/0xe30 [virtio_net]
? __update_load_avg_cfs_rq+0x264/0x2d0
? update_curr+0x35/0x260
? reweight_entity+0x1be/0x260
__napi_poll.constprop.0+0x28/0x1c0
net_rx_action+0x329/0x420
? enqueue_hrtimer+0x35/0x90
? trace_hardirqs_on+0x1d/0x80
? kvm_sched_clock_read+0xd/0x20
? sched_clock+0xc/0x30
? kvm_sched_clock_read+0xd/0x20
? sched_clock+0xc/0x30
? sched_clock_cpu+0xd/0x1a0
handle_softirqs+0x138/0x3e0
do_softirq.part.0+0x89/0xc0
</IRQ>
<TASK>
__local_bh_enable_ip+0xa7/0xb0
virtnet_open+0xc8/0x310 [virtio_net]
__dev_open+0xfa/0x1b0
__dev_change_flags+0x1de/0x250
dev_change_flags+0x22/0x60
do_setlink.isra.0+0x2df/0x10b0
? rtnetlink_rcv_msg+0x34f/0x3f0
? netlink_rcv_skb+0x54/0x100
? netlink_unicast+0x23e/0x390
? netlink_sendmsg+0x21e/0x490
? ____sys_sendmsg+0x31b/0x350
? avc_has_perm_noaudit+0x67/0xf0
? cred_has_capability.isra.0+0x75/0x110
? __nla_validate_parse+0x5f/0xee0
? __pfx___probestub_irq_enable+0x3/0x10
? __create_object+0x5e/0x90
? security_capable+0x3b/0x70
rtnl_newlink+0x784/0xaf0
? avc_has_perm_noaudit+0x67/0xf0
? cred_has_capability.isra.0+0x75/0x110
? stack_depot_save_flags+0x24/0x6d0
? __pfx_rtnl_newlink+0x10/0x10
rtnetlink_rcv_msg+0x34f/0x3f0
? do_syscall_64+0x6c/0x180
? entry_SYSCALL_64_after_hwframe+0x76/0x7e
? __pfx_rtnetlink_rcv_msg+0x10/0x10
netlink_rcv_skb+0x54/0x100
netlink_unicast+0x23e/0x390
netlink_sendmsg+0x21e/0x490
____sys_sendmsg+0x31b/0x350
? copy_msghdr_from_user+0x6d/0xa0
___sys_sendmsg+0x86/0xd0
? __pte_offset_map+0x17/0x160
? preempt_count_add+0x69/0xa0
? __call_rcu_common.constprop.0+0x147/0x610
? preempt_count_add+0x69/0xa0
? preempt_count_add+0x69/0xa0
? _raw_spin_trylock+0x13/0x60
? trace_hardirqs_on+0x1d/0x80
__sys_sendmsg+0x66/0xc0
do_syscall_64+0x6c/0x180
entry_SYSCALL_64_after_hwframe+0x76/0x7e
RIP: 0033:0x7f41defe5b34
Code: 15 e1 12 0f 00 f7 d8 64 89 02 b8 ff ff ff ff eb bf 0f 1f 44 00 00
f3 0f 1e fa 80 3d 35 95 0f 00 00 74 13 b8 2e 00 00 00 0f 05 <48> 3d 00
f0 ff ff 77 4c c3 0f 1f 00 55 48 89 e5 48 83 ec 20 89 55
RSP: 002b:00007ffe5336ecc8 EFLAGS: 00000202 ORIG_RAX: 000000000000002e
RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 00007f41defe5b34
RDX: 0000000000000000 RSI: 00007ffe5336ed30 RDI: 0000000000000003
RBP: 00007ffe5336eda0 R08: 0000000000000010 R09: 0000000000000001
R10: 00007ffe5336f6f9 R11: 0000000000000202 R12: 0000000000000003
R13: 0000000067452259 R14: 0000556ccc28b040 R15: 0000000000000000
</TASK>
[...]
Fixes: c8bd1f7f3e ("virtio_net: add support for Byte Queue Limits")
Cc: <stable@vger.kernel.org> # v6.11+
Signed-off-by: Koichiro Den <koichiro.den@canonical.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
[ pabeni: trimmed possibly troublesome separator ]
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Now, we support AF_XDP(xsk). Add NETDEV_XDP_ACT_XSK_ZEROCOPY to
xdp_features.
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20241112012928.102478-14-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If send queue sent some packets, we update the tx timeout
record to prevent the tx timeout.
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20241112012928.102478-13-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The driver's tx napi is very important for XSK. It is responsible for
obtaining data from the XSK queue and sending it out.
At the beginning, we need to trigger tx napi.
virtnet_free_old_xmit distinguishes three type ptr(skb, xdp frame, xsk
buffer) by the last bits of the pointer.
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20241112012928.102478-12-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since xsk's TX queue is consumed by TX NAPI, if sq is bound to xsk, then
we must stop tx napi from being disabled.
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20241112012928.102478-11-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Because the af-xdp will introduce a new xmit type, so I refactor the
xmit type mechanism first.
We know both xdp_frame and sk_buff are at least 4 bytes aligned.
For the xdp tx, we do not pass any pointer to virtio core as data,
we just need to pass the len of the packet. So we will push len
to the void pointer. We can make sure the pointer is 4 bytes aligned.
And the data structure of AF_XDP also is at least 4 bytes aligned.
So the last two bits of the pointers are free, we can't use these to
distinguish them.
00 for skb
01 for SKB_ORPHAN
10 for XDP
11 for AF-XDP tx
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20241112012928.102478-9-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
RSS configuration should be updated with queue number. In particular, it
should be updated when (1) rss enabled and (2) default rss configuration
is used without user modification.
During rss command processing, device updates queue_pairs using
rss.max_tx_vq. That is, the device updates queue_pairs together with
rss, so we can skip the sperate queue_pairs update
(VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET below) and return directly.
Also remove the `vi->has_rss ?` check when setting vi->rss.max_tx_vq,
because this is not used in the other hash_report case.
Fixes: c7114b1249 ("drivers/net/virtio_net: Added basic RSS support.")
Signed-off-by: Philo Lu <lulie@linux.alibaba.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
During virtnet_probe, default rss configuration is initialized, but was
not committed to the device. This patch fix this by sending rss command
after device ready in virtnet_probe. Otherwise, the actual rss
configuration used by device can be different with that read by user
from driver, which may confuse the user.
If the command committing fails, driver rss will be disabled.
Fixes: c7114b1249 ("drivers/net/virtio_net: Added basic RSS support.")
Signed-off-by: Philo Lu <lulie@linux.alibaba.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Joe Damato <jdamato@fastly.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Add hash_key_length check in virtnet_probe() to avoid possible out of
bound errors when setting/reading the hash key.
Fixes: c7114b1249 ("drivers/net/virtio_net: Added basic RSS support.")
Signed-off-by: Philo Lu <lulie@linux.alibaba.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Joe Damato <jdamato@fastly.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When reading/writing virtio_net_ctrl_rss, we get the indirection table
size from vi->rss_indir_table_size, which is initialized in
virtnet_probe(). However, the actual size of indirection_table was set
as VIRTIO_NET_RSS_MAX_TABLE_LEN=128. This collision may cause issues if
the vi->rss_indir_table_size exceeds 128.
This patch instead uses dynamic indirection table, allocated with
vi->rss after vi->rss_indir_table_size initialized. And free it in
virtnet_remove().
In virtnet_commit_rss_command(), sgs for rss is initialized differently
with hash_report. So indirection_table is not used if !vi->has_rss, and
then we don't need to alloc indirection_table for hash_report only uses.
Fixes: c7114b1249 ("drivers/net/virtio_net: Added basic RSS support.")
Signed-off-by: Philo Lu <lulie@linux.alibaba.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Joe Damato <jdamato@fastly.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Now, the premapped mode can be enabled unconditionally.
So we can remove the failover code for merge and small mode.
The virtnet_rq_xxx() helper would be only used if the mode is using pre
mapping. A check is added to prevent misusing of these API.
Tested-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently, the virtio core will perform a dma operation for each
buffer. Although, the same page may be operated multiple times.
In premapped mod, we can perform only one dma operation for the pages of
the alloc frag. This is beneficial for the iommu device.
kernel command line: intel_iommu=on iommu.passthrough=0
| strict=0 | strict=1
Before | 775496pps | 428614pps
After | 1109316pps | 742853pps
In the 6.11, we disabled this feature because a regress [1].
Now, we fix the problem and re-enable it.
[1]: http://lore.kernel.org/all/8b20cc28-45a9-4643-8e87-ba164a540c0a@oracle.com
Tested-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The virtio-net big mode did not enable premapped mode,
so we did not need to check the unmap. And the subsequent
commit will remove the failover code for failing enable
premapped for merge and small mode. So we need to remove
the checking do_dma code in the big mode path.
Tested-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When the frag just got a page, then may lead to regression on VM.
Specially if the sysctl net.core.high_order_alloc_disable value is 1,
then the frag always get a page when do refill.
Which could see reliable crashes or scp failure (scp a file 100M in size
to VM).
The issue is that the virtnet_rq_dma takes up 16 bytes at the beginning
of a new frag. When the frag size is larger than PAGE_SIZE,
everything is fine. However, if the frag is only one page and the
total size of the buffer and virtnet_rq_dma is larger than one page, an
overflow may occur.
The commit f9dac92ba9 ("virtio_ring: enable premapped mode whatever
use_dma_api") introduced this problem. And we reverted some commits to
fix this in last linux version. Now we try to enable it and fix this
bug directly.
Here, when the frag size is not enough, we reduce the buffer len to fix
this problem.
Reported-by: "Si-Wei Liu" <si-wei.liu@oracle.com>
Tested-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
net_dim() is currently passed a struct dim_sample argument by value.
struct dim_sample is 24 bytes. Since this is greater 16 bytes, x86-64
passes it on the stack. All callers have already initialized dim_sample
on the stack, so passing it by value requires pushing a duplicated copy
to the stack. Either witing to the stack and immediately reading it, or
perhaps dereferencing addresses relative to the stack pointer in a chain
of push instructions, seems to perform quite poorly.
In a heavy TCP workload, mlx5e_handle_rx_dim() consumes 3% of CPU time,
94% of which is attributed to the first push instruction to copy
dim_sample on the stack for the call to net_dim():
// Call ktime_get()
0.26 |4ead2: call 4ead7 <mlx5e_handle_rx_dim+0x47>
// Pass the address of struct dim in %rdi
|4ead7: lea 0x3d0(%rbx),%rdi
// Set dim_sample.pkt_ctr
|4eade: mov %r13d,0x8(%rsp)
// Set dim_sample.byte_ctr
|4eae3: mov %r12d,0xc(%rsp)
// Set dim_sample.event_ctr
0.15 |4eae8: mov %bp,0x10(%rsp)
// Duplicate dim_sample on the stack
94.16 |4eaed: push 0x10(%rsp)
2.79 |4eaf1: push 0x10(%rsp)
0.07 |4eaf5: push %rax
// Call net_dim()
0.21 |4eaf6: call 4eafb <mlx5e_handle_rx_dim+0x6b>
To allow the caller to reuse the struct dim_sample already on the stack,
pass the struct dim_sample by reference to net_dim().
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Arthur Kiyanovski <akiyano@amazon.com>
Reviewed-by: Louis Peens <louis.peens@corigine.com>
Link: https://patch.msgid.link/20241031002326.3426181-2-csander@purestorage.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Static analysis on linux-next has detected the following issue
in function virtnet_stats_ctx_init, in drivers/net/virtio_net.c :
if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_CVQ) {
queue_type = VIRTNET_Q_TYPE_CQ;
ctx->bitmap[queue_type] |= VIRTIO_NET_STATS_TYPE_CVQ;
ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_cvq_desc);
ctx->size[queue_type] += sizeof(struct virtio_net_stats_cvq);
}
ctx->bitmap is declared as a u32 however it is being bit-wise or'd with
VIRTIO_NET_STATS_TYPE_CVQ and this is defined as 1 << 32:
include/uapi/linux/virtio_net.h:#define VIRTIO_NET_STATS_TYPE_CVQ (1ULL << 32)
..and hence the bit-wise or operation won't set any bits in ctx->bitmap
because 1ULL < 32 is too wide for a u32.
In fact, the field is read into a u64:
u64 offset, bitmap;
....
bitmap = ctx->bitmap[queue_type];
so to fix, it is enough to make bitmap an array of u64.
Fixes: 941168f8b4 ("virtio_net: support device stats")
Reported-by: "Colin King (gmail)" <colin.i.king@gmail.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Link: https://patch.msgid.link/53e2bd6728136d5916e384a7840e5dc7eebff832.1729099611.git.mst@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently, the virtio-net driver will perform a pre-dma-mapping for
small or mergeable RX buffer. But for small packets, a mismatched address
without VIRTNET_RX_PAD and xdp_headroom is used for unmapping.
That will result in unsynchronized buffers when SWIOTLB is enabled, for
example, when running as a TDX guest.
This patch unifies the address passed to the virtio core as the address of
the virtnet header and fixes the mismatched buffer address.
Changes from v2: unify the buf that passed to the virtio core in small
and merge mode.
Changes from v1: Use ctx to get xdp_headroom.
Fixes: 295525e29a ("virtio_net: merge dma operations when filling mergeable buffers")
Signed-off-by: Wenbo Li <liwenbo.martin@bytedance.com>
Signed-off-by: Jiahui Cen <cenjiahui@bytedance.com>
Signed-off-by: Ying Fang <fangying.tommy@bytedance.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Link: https://patch.msgid.link/20240919081351.51772-1-liwenbo.martin@bytedance.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Cross-merge networking fixes after downstream PR.
No conflicts (sort of) and no adjacent changes.
This merge reverts commit b3c9e65eb2 ("net: hsr: remove seqnr_lock")
from net, as it was superseded by
commit 430d67bdcb ("net: hsr: Use the seqnr lock for frames received via interlink port.")
in net-next.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This reverts commit defd28aa5a.
Recover the code to disable premapped mode.
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Takero Funaki <flintglass@gmail.com>
Link: https://patch.msgid.link/20240906123137.108741-2-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Cross-merge networking fixes after downstream PR.
No conflicts.
Adjacent changes:
drivers/net/ethernet/broadcom/bnxt/bnxt.h
c948c0973d ("bnxt_en: Don't clear ntuple filters and rss contexts during ethtool ops")
f2878cdeb7 ("bnxt_en: Add support to call FW to update a VNIC")
Link: https://patch.msgid.link/20240822210125.1542769-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We calculate guest offloads during probe without the protection of
rtnl_lock. This lead to race between probe and ndo_set_features. Fix
this by moving the calculation under the rtnl_lock.
Fixes: 3f93522ffa ("virtio-net: switch off offloads on demand if possible on XDP set")
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20240814052228.4654-5-jasowang@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch synchronizes operstate with admin state per RFC2863.
This is done by trying to toggle the carrier upon open/close and
synchronize with the config change work. This allows to propagate
status correctly to stacked devices like:
ip link add link enp0s3 macvlan0 type macvlan
ip link set link enp0s3 down
ip link show
Before this patch:
3: enp0s3: <BROADCAST,MULTICAST> mtu 1500 qdisc pfifo_fast state DOWN mode DEFAULT group default qlen 1000
link/ether 00:00:05:00:00:09 brd ff:ff:ff:ff:ff:ff
......
5: macvlan0@enp0s3: <BROADCAST,MULTICAST,UP,LOWER_UP,M-DOWN> mtu 1500 qdisc noqueue state UP mode DEFAULT group default qlen 1000
link/ether b2:a9:c5:04:da:53 brd ff:ff:ff:ff:ff:ff
After this patch:
3: enp0s3: <BROADCAST,MULTICAST> mtu 1500 qdisc pfifo_fast state DOWN mode DEFAULT group default qlen 1000
link/ether 00:00:05:00:00:09 brd ff:ff:ff:ff:ff:ff
...
5: macvlan0@enp0s3: <NO-CARRIER,BROADCAST,MULTICAST,UP,M-DOWN> mtu 1500 qdisc noqueue state LOWERLAYERDOWN mode DEFAULT group default qlen 1000
link/ether b2:a9:c5:04:da:53 brd ff:ff:ff:ff:ff:ff
Cc: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com>
Cc: Gia-Khanh Nguyen <gia-khanh.nguyen@oracle.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20240814052228.4654-4-jasowang@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
During suspend/resume the following BUG was hit:
------------[ cut here ]------------
kernel BUG at lib/dynamic_queue_limits.c:99!
Internal error: Oops - BUG: 0 [#1] SMP ARM
Modules linked in: bluetooth ecdh_generic ecc libaes
CPU: 1 PID: 1282 Comm: rtcwake Not tainted
6.10.0-rc3-00732-gc8bd1f7f3e61 #15240
Hardware name: Generic DT based system
PC is at dql_completed+0x270/0x2cc
LR is at __free_old_xmit+0x120/0x198
pc : [<c07ffa54>] lr : [<c0c42bf4>] psr: 80000013
...
Flags: Nzcv IRQs on FIQs on Mode SVC_32 ISA ARM Segment none
Control: 10c5387d Table: 43a4406a DAC: 00000051
...
Process rtcwake (pid: 1282, stack limit = 0xfbc21278)
Stack: (0xe0805e80 to 0xe0806000)
...
Call trace:
dql_completed from __free_old_xmit+0x120/0x198
__free_old_xmit from free_old_xmit+0x44/0xe4
free_old_xmit from virtnet_poll_tx+0x88/0x1b4
virtnet_poll_tx from __napi_poll+0x2c/0x1d4
__napi_poll from net_rx_action+0x140/0x2b4
net_rx_action from handle_softirqs+0x11c/0x350
handle_softirqs from call_with_stack+0x18/0x20
call_with_stack from do_softirq+0x48/0x50
do_softirq from __local_bh_enable_ip+0xa0/0xa4
__local_bh_enable_ip from virtnet_open+0xd4/0x21c
virtnet_open from virtnet_restore+0x94/0x120
virtnet_restore from virtio_device_restore+0x110/0x1f4
virtio_device_restore from dpm_run_callback+0x3c/0x100
dpm_run_callback from device_resume+0x12c/0x2a8
device_resume from dpm_resume+0x12c/0x1e0
dpm_resume from dpm_resume_end+0xc/0x18
dpm_resume_end from suspend_devices_and_enter+0x1f0/0x72c
suspend_devices_and_enter from pm_suspend+0x270/0x2a0
pm_suspend from state_store+0x68/0xc8
state_store from kernfs_fop_write_iter+0x10c/0x1cc
kernfs_fop_write_iter from vfs_write+0x2b0/0x3dc
vfs_write from ksys_write+0x5c/0xd4
ksys_write from ret_fast_syscall+0x0/0x54
Exception stack(0xe8bf1fa8 to 0xe8bf1ff0)
...
---[ end trace 0000000000000000 ]---
After virtnet_napi_enable() is called, the following path is hit:
__napi_poll()
-> virtnet_poll()
-> virtnet_poll_cleantx()
-> netif_tx_wake_queue()
That wakes the TX queue and allows skbs to be submitted and accounted by
BQL counters.
Then netdev_tx_reset_queue() is called that resets BQL counters and
eventually leads to the BUG in dql_completed().
Move virtnet_napi_tx_enable() what does BQL counters reset before RX
napi enable to avoid the issue.
Reported-by: Marek Szyprowski <m.szyprowski@samsung.com>
Closes: https://lore.kernel.org/netdev/e632e378-d019-4de7-8f13-07c572ab37a9@samsung.com/
Fixes: c8bd1f7f3e ("virtio_net: add support for Byte Queue Limits")
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20240814122500.1710279-1-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Don't break the resize action if the vq coalescing feature
named VIRTIO_NET_F_VQ_NOTF_COAL is not negotiated.
Fixes: f61fe5f081 ("virtio-net: fix the vq coalescing setting for vq resize")
Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Eugenio Pé rez <eperezma@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Virtio spec says:
The driver MUST have negotiated the VIRTIO_NET_F_VQ_NOTF_COAL
feature when issuing commands VIRTIO_NET_CTRL_NOTF_COAL_VQ_SET
and VIRTIO_NET_CTRL_NOTF_COAL_VQ_GET.
So we add the feature negotiation check to
virtnet_send_{r,t}x_ctrl_coal_vq_cmd as a basis for the next bugfix patch.
Suggested-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Several new features here:
- Virtio find vqs API has been reworked
(required to fix the scalability issue we have with
adminq, which I hope to merge later in the cycle)
- vDPA driver for Marvell OCTEON
- virtio fs performance improvement
- mlx5 migration speedups
Fixes, cleanups all over the place.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
-----BEGIN PGP SIGNATURE-----
iQFDBAABCAAtFiEEXQn9CHHI+FuUyooNKB8NuNKNVGkFAmaXjQQPHG1zdEByZWRo
YXQuY29tAAoJECgfDbjSjVRpnIsH/jVNqAQbe/vaBQdNMdnsA+P9A9unLbYRxYCQ
tN73mQRIXKtnZHBRAEbMGq52HPYg8HlN2HJSgyNo6I6t8VD+PiOco7m+3GpmqEcW
aXPOPl0BAbVoDgyutxRuuodP8Z61lBx0mG6iOxpzTXOPGlpQqtPCFHO8YnodqnPf
tMix/5uAqgZKV2siCbw5DtzwEc0gDHU8qsD0/nyoS5nBDF9yh/ardr5P/qiyFDQH
atCNYTOhIFU83pLAaw0fpCGbkt7gxf+5RpWVx3wkYww+/MwvYhsveRvQyaGbBz3n
WDtET3SOtVTta98OAGIKCq/2z8f6mYXBP7vXapBgnJG3vwS/poQ=
=LYua
-----END PGP SIGNATURE-----
Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
Pull virtio updates from Michael Tsirkin:
"Several new features here:
- Virtio find vqs API has been reworked (required to fix the
scalability issue we have with adminq, which I hope to merge later
in the cycle)
- vDPA driver for Marvell OCTEON
- virtio fs performance improvement
- mlx5 migration speedups
Fixes, cleanups all over the place"
* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost: (56 commits)
virtio: rename virtio_find_vqs_info() to virtio_find_vqs()
virtio: remove unused virtio_find_vqs() and virtio_find_vqs_ctx() helpers
virtio: convert the rest virtio_find_vqs() users to virtio_find_vqs_info()
virtio_balloon: convert to use virtio_find_vqs_info()
virtiofs: convert to use virtio_find_vqs_info()
scsi: virtio_scsi: convert to use virtio_find_vqs_info()
virtio_net: convert to use virtio_find_vqs_info()
virtio_crypto: convert to use virtio_find_vqs_info()
virtio_console: convert to use virtio_find_vqs_info()
virtio_blk: convert to use virtio_find_vqs_info()
virtio: rename find_vqs_info() op to find_vqs()
virtio: remove the original find_vqs() op
virtio: call virtio_find_vqs_info() from virtio_find_single_vq() directly
virtio: convert find_vqs() op implementations to find_vqs_info()
virtio_pci: convert vp_*find_vqs() ops to find_vqs_info()
virtio: introduce virtio_queue_info struct and find_vqs_info() config op
virtio: make virtio_find_single_vq() call virtio_find_vqs()
virtio: make virtio_find_vqs() call virtio_find_vqs_ctx()
caif_virtio: use virtio_find_single_vq() for single virtqueue finding
vdpa/mlx5: Don't enable non-active VQs in .set_vq_ready()
...
Since the original virtio_find_vqs() is no longer present, rename
virtio_find_vqs_info() back to virtio_find_vqs().
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Message-Id: <20240708074814.1739223-20-jiri@resnulli.us>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Instead of passing separate names and callbacks arrays
to virtio_find_vqs_ctx(), allocate one of virtual_queue_info structs
and pass it to virtio_find_vqs_info().
Suggested-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Message-Id: <20240708074814.1739223-14-jiri@resnulli.us>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
After the commit bdacf3e349 ("net: Use nested-BH locking for
napi_alloc_cache.") was merged, the following warning began to appear:
WARNING: CPU: 5 PID: 1 at net/core/skbuff.c:1451 napi_skb_cache_put+0x82/0x4b0
__warn+0x12f/0x340
napi_skb_cache_put+0x82/0x4b0
napi_skb_cache_put+0x82/0x4b0
report_bug+0x165/0x370
handle_bug+0x3d/0x80
exc_invalid_op+0x1a/0x50
asm_exc_invalid_op+0x1a/0x20
__free_old_xmit+0x1c8/0x510
napi_skb_cache_put+0x82/0x4b0
__free_old_xmit+0x1c8/0x510
__free_old_xmit+0x1c8/0x510
__pfx___free_old_xmit+0x10/0x10
The issue arises because virtio is assuming it's running in NAPI context
even when it's not, such as in the netpoll case.
To resolve this, modify virtnet_poll_tx() to only set NAPI when budget
is available. Same for virtnet_poll_cleantx(), which always assumed that
it was in a NAPI context.
Fixes: df133f3f96 ("virtio_net: bulk free tx skbs")
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Heng Qi <hengqi@linux.alibaba.com>
Link: https://patch.msgid.link/20240712115325.54175-1-leitao@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Support AF-XDP for merge mode.
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20240708112537.96291-11-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In the process:
1. We may need to copy data to create skb for XDP_PASS.
2. We may need to call xsk_buff_free() to release the buffer.
3. The handle for xdp_buff is difference from the buffer.
If we pushed this logic into existing receive handle(merge and small),
we would have to maintain code scattered inside merge and small (and big).
So I think it is a good choice for us to put the xsk code into an
independent function.
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20240708112537.96291-10-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Implement the logic of filling rq with XSK buffers.
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20240708112537.96291-9-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
xsk wakeup is used to trigger the logic for xsk xmit by xsk framework or
user.
Virtio-net does not support to actively generate an interruption, so it
tries to trigger tx NAPI on the local cpu.
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20240708112537.96291-8-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch implement the logic of bind/unbind xsk pool to rq.
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20240708112537.96291-7-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
