user/sven/linux.git/drivers/virtio/virtio_ring.c, branch v4.4

virtio_ring: shadow available ring flags & index

2015-12-07T15:28:11Z

Improves cacheline transfer flow of available ring header. Virtqueues are implemented as a pair of rings, one producer->consumer avail ring and one consumer->producer used ring; preceding the avail ring in memory are two contiguous u16 fields -- avail->flags and avail->idx. A producer posts work by writing to avail->idx and a consumer reads avail->idx. The flags and idx fields only need to be written by a producer CPU and only read by a consumer CPU; when the producer and consumer are running on different CPUs and the virtio_ring code is structured to only have source writes/sink reads, we can continuously transfer the avail header cacheline between 'M' states between cores. This flow optimizes core -> core bandwidth on certain CPUs. (see: "Software Optimization Guide for AMD Family 15h Processors", Section 11.6; similar language appears in the 10h guide and should apply to CPUs w/ exclusive caches, using LLC as a transfer cache) Unfortunately the existing virtio_ring code issued reads to the avail->idx and read-modify-writes to avail->flags on the producer. This change shadows the flags and index fields in producer memory; the vring code now reads from the shadows and only ever writes to avail->flags and avail->idx, allowing the cacheline to transfer core -> core optimally. In a concurrent version of vring_bench, the time required for 10,000,000 buffer checkout/returns was reduced by ~2% (average across many runs) on an AMD Piledriver (15h) CPU: (w/o shadowing): Performance counter stats for './vring_bench': 5,451,082,016 L1-dcache-loads ... 2.221477739 seconds time elapsed (w/ shadowing): Performance counter stats for './vring_bench': 5,405,701,361 L1-dcache-loads ... 2.168405376 seconds time elapsed The further away (in a NUMA sense) virtio producers and consumers are from each other, the more we expect to benefit. Physical implementations of virtio devices and implementations of virtio where the consumer polls vring avail indexes (vhost) should also benefit. Signed-off-by: Venkatesh Srinivas Signed-off-by: Michael S. Tsirkin

virtio: Do not drop __GFP_HIGH in alloc_indirect

2015-12-07T15:28:11Z

b92b1b89a33c ("virtio: force vring descriptors to be allocated from lowmem") tried to exclude highmem pages for descriptors so it cleared __GFP_HIGHMEM from a given gfp mask. The patch also cleared __GFP_HIGH which doesn't make much sense for this fix because __GFP_HIGH only controls access to memory reserves and it doesn't have any influence on the zone selection. Some of the call paths use GFP_ATOMIC and dropping __GFP_HIGH will reduce their changes for success because the lack of access to memory reserves. Signed-off-by: Michal Hocko Signed-off-by: Michael S. Tsirkin Acked-by: Will Deacon Reviewed-by: Mel Gorman

virtio: Avoid possible kernel panic if DEBUG is enabled.

2015-02-11T04:33:14Z

The virtqueue_add() calls START_USE() upon entry. The virtqueue_kick() is called if vq->num_added == (1 << 16) - 1 before calling END_USE(). The virtqueue_kick_prepare() called via virtqueue_kick() calls START_USE() upon entry, and will call panic() if DEBUG is enabled. Move this virtqueue_kick() call to after END_USE() call. Signed-off-by: Tetsuo Handa Signed-off-by: Rusty Russell

virtio_ring: coding style fix

2015-01-21T05:58:57Z

Most of our code has struct foo { } Fix one instances where ring is inconsistent. Signed-off-by: Michael S. Tsirkin Signed-off-by: Rusty Russell

virtio: make VIRTIO_F_VERSION_1 a transport bit

2014-12-09T10:06:32Z

Activate VIRTIO_F_VERSION_1 automatically unless legacy_only is set. Signed-off-by: Michael S. Tsirkin

virtio: allow transports to get avail/used addresses

2014-12-09T10:05:25Z

For virtio-1, we can theoretically have a more complex virtqueue layout with avail and used buffers not on a contiguous memory area with the descriptor table. For now, it's fine for a transport driver to stay with the old layout: It needs, however, a way to access the locations of the avail/used rings so it can register them with the host. Reviewed-by: David Hildenbrand Signed-off-by: Cornelia Huck Signed-off-by: Michael S. Tsirkin

virtio_ring: switch to new memory access APIs

2014-12-09T10:05:25Z

Use virtioXX_to_cpu and friends for access to all multibyte structures in memory. Note: this is intentionally mechanical. A follow-up patch will split long lines etc. Signed-off-by: Michael S. Tsirkin Reviewed-by: Cornelia Huck

virtio: use u32, not bitmap for features

2014-12-09T10:05:23Z

It seemed like a good idea to use bitmap for features in struct virtio_device, but it's actually a pain, and seems to become even more painful when we get more than 32 feature bits. Just change it to a u32 for now. Based on patch by Rusty. Suggested-by: David Hildenbrand Signed-off-by: Rusty Russell Signed-off-by: Cornelia Huck Signed-off-by: Michael S. Tsirkin Reviewed-by: Cornelia Huck

virtio_ring: unify direct/indirect code paths.

2014-09-13T16:52:35Z

virtqueue_add() populates the virtqueue descriptor table from the sgs given. If it uses an indirect descriptor table, then it puts a single descriptor in the descriptor table pointing to the kmalloc'ed indirect table where the sg is populated. Previously vring_add_indirect() did the allocation and the simple linear layout. We replace that with alloc_indirect() which allocates the indirect table then chains it like the normal descriptor table so we can reuse the core logic. This slows down pktgen by less than 1/2 a percent (which uses direct descriptors), as well as vring_bench, but it's far neater. vring_bench before: 1061485790-1104800648(1.08254e+09+/-6.6e+06)ns vring_bench after: 1125610268-1183528965(1.14172e+09+/-8e+06)ns pktgen before: 787781-796334(793165+/-2.4e+03)pps 365-369(367.5+/-1.2)Mb/sec (365530384-369498976(3.68028e+08+/-1.1e+06)bps) errors: 0 pktgen after: 779988-790404(786391+/-2.5e+03)pps 361-366(364.35+/-1.3)Mb/sec (361914432-366747456(3.64885e+08+/-1.2e+06)bps) errors: 0 Now, if we make force indirect descriptors by turning off any_header_sg in virtio_net.c: pktgen before: 713773-721062(718374+/-2.1e+03)pps 331-334(332.95+/-0.92)Mb/sec (331190672-334572768(3.33325e+08+/-9.6e+05)bps) errors: 0 pktgen after: 710542-719195(714898+/-2.4e+03)pps 329-333(331.15+/-1.1)Mb/sec (329691488-333706480(3.31713e+08+/-1.1e+06)bps) errors: 0 Signed-off-by: Rusty Russell Signed-off-by: David S. Miller

virtio_ring: assume sgs are always well-formed.

2014-09-13T16:50:46Z

We used to have several callers which just used arrays. They're gone, so we can use sg_next() everywhere, simplifying the code. On my laptop, this slowed down vring_bench by 15%: vring_bench before: 936153354-967745359(9.44739e+08+/-6.1e+06)ns vring_bench after: 1061485790-1104800648(1.08254e+09+/-6.6e+06)ns However, a more realistic test using pktgen on a AMD FX(tm)-8320 saw a few percent improvement: pktgen before: 767390-792966(785159+/-6.5e+03)pps 356-367(363.75+/-2.9)Mb/sec (356068960-367936224(3.64314e+08+/-3e+06)bps) errors: 0 pktgen after: 787781-796334(793165+/-2.4e+03)pps 365-369(367.5+/-1.2)Mb/sec (365530384-369498976(3.68028e+08+/-1.1e+06)bps) errors: 0 Signed-off-by: Rusty Russell Signed-off-by: David S. Miller