similar to: [RFC] virtio: Support releasing lock during kick

Remove wrapper functions. This makes the allocation type explicit in all callers; I used GPF_KERNEL where it seemed obvious, left it at GFP_ATOMIC otherwise. Signed-off-by: Rusty Russell <rusty at rustcorp.com.au> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c --- a/drivers/block/virtio_blk.c +++ b/drivers/block/virtio_blk.c @@ -171,7 +171,7 @@ static bool

Remove wrapper functions. This makes the allocation type explicit in all callers; I used GPF_KERNEL where it seemed obvious, left it at GFP_ATOMIC otherwise. Signed-off-by: Rusty Russell <rusty at rustcorp.com.au> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c --- a/drivers/block/virtio_blk.c +++ b/drivers/block/virtio_blk.c @@ -171,7 +171,7 @@ static bool

virtqueue ops were introduced in the hope that we'll have multiple implementations besides virtio_ring, but none have surfaced so far, and given that existing virtio ring is deployed in production we are likely stuck with it now, so this layer just adds complexity and overhead. Further, the need to pass vq twice to each call (as in dev->vq->vq_ops->kick(dev->vq) ) adds potential

virtqueue ops were introduced in the hope that we'll have multiple implementations besides virtio_ring, but none have surfaced so far, and given that existing virtio ring is deployed in production we are likely stuck with it now, so this layer just adds complexity and overhead. Further, the need to pass vq twice to each call (as in dev->vq->vq_ops->kick(dev->vq) ) adds potential

Rather than handing a scatterlist[] and out and in numbers to virtqueue_add_buf(), hand two separate ones which can be chained. I shall refrain from ranting about what a disgusting hack chained scatterlists are. I'll just note that this doesn't make things simpler (see diff). The scatterlists we use can be too large for the stack, so we put them in our device struct and reuse them. But

Rather than handing a scatterlist[] and out and in numbers to virtqueue_add_buf(), hand two separate ones which can be chained. I shall refrain from ranting about what a disgusting hack chained scatterlists are. I'll just note that this doesn't make things simpler (see diff). The scatterlists we use can be too large for the stack, so we put them in our device struct and reuse them. But

OK, this is (ab)uses some of Paolo's patches. The first 7 are candidates for this merge window (maybe), the rest I'm not so sure about. Thanks, Rusty. Paolo Bonzini (3): scatterlist: introduce sg_unmark_end virtio-blk: reorganize virtblk_add_req virtio-blk: use virtqueue_add_sgs on req path Rusty Russell (13): virtio_ring: virtqueue_add_sgs, to add multiple sgs. virtio-blk:

OK, this is (ab)uses some of Paolo's patches. The first 7 are candidates for this merge window (maybe), the rest I'm not so sure about. Thanks, Rusty. Paolo Bonzini (3): scatterlist: introduce sg_unmark_end virtio-blk: reorganize virtblk_add_req virtio-blk: use virtqueue_add_sgs on req path Rusty Russell (13): virtio_ring: virtqueue_add_sgs, to add multiple sgs. virtio-blk:

Add virtqueue_add_sgs which is more general than virtqueue_add_buf, which makes virtio-scsi and virtio-blk nicer, then add virtqueue_add_inbuf and virtqueue_add_outbuf which handle the more general case, and finally delete virtqueue_add_buf(). I'm hoping this will be the final post of the whole series, and it can move from my pending-rebases tree into virtio-next. Thanks! Rusty. Paolo

Add virtqueue_add_sgs which is more general than virtqueue_add_buf, which makes virtio-scsi and virtio-blk nicer, then add virtqueue_add_inbuf and virtqueue_add_outbuf which handle the more general case, and finally delete virtqueue_add_buf(). I'm hoping this will be the final post of the whole series, and it can move from my pending-rebases tree into virtio-next. Thanks! Rusty. Paolo

Most device drivers do not need to perform any postprocessing on the scatterlists they receive from higher-level drivers (e.g. the block or SCSI layer), because they translate the request metadata directly from the various C structs into the data that is required by the device. virtio devices however do this translation in two steps: a device-specific step in the device driver, and generic

Most device drivers do not need to perform any postprocessing on the scatterlists they receive from higher-level drivers (e.g. the block or SCSI layer), because they translate the request metadata directly from the various C structs into the data that is required by the device. virtio devices however do this translation in two steps: a device-specific step in the device driver, and generic

Hi Rusty, The linux-next kernel was failing my virtio-console test suite for a while. I looked into it today, and it's due to the virtqueue_add_buf() change that doesn't return > 0 values anymore. I found your commit that adjusts virtio_console.c, but you missed one instance where the return value mattered, and as a result not enough buffers were queued for the host to send in data.

Hi Rusty, The linux-next kernel was failing my virtio-console test suite for a while. I looked into it today, and it's due to the virtqueue_add_buf() change that doesn't return > 0 values anymore. I found your commit that adjusts virtio_console.c, but you missed one instance where the return value mattered, and as a result not enough buffers were queued for the host to send in data.

A virtio driver does virtqueue_add_buf() multiple times before finally calling virtqueue_kick(); previously we only exposed the added buffers in the virtqueue_kick() call. This means we don't need a memory barrier in virtqueue_add_buf(), but it reduces concurrency as the device (ie. host) can't see the buffers until the kick. Signed-off-by: Rusty Russell <rusty at rustcorp.com.au>

A virtio driver does virtqueue_add_buf() multiple times before finally calling virtqueue_kick(); previously we only exposed the added buffers in the virtqueue_kick() call. This means we don't need a memory barrier in virtqueue_add_buf(), but it reduces concurrency as the device (ie. host) can't see the buffers until the kick. Signed-off-by: Rusty Russell <rusty at rustcorp.com.au>

The virtqueue_add_buf function has two limitations: 1) it requires the caller to provide all the buffers in a single call; 2) it does not support chained scatterlists: the buffers must be provided as an array of struct scatterlist. Because of these limitations, virtio-scsi has to copy each request into a scatterlist internal to the driver. It cannot just use the one that was prepared by the

The virtqueue_add_buf function has two limitations: 1) it requires the caller to provide all the buffers in a single call; 2) it does not support chained scatterlists: the buffers must be provided as an array of struct scatterlist. Because of these limitations, virtio-scsi has to copy each request into a scatterlist internal to the driver. It cannot just use the one that was prepared by the

Currently if VIRTIO_RING_F_INDIRECT_DESC is enabled we will use indirect descriptors even if we have plenty of space in the ring. This means that we take a performance hit at all times due to the overhead of creating indirect descriptors. Instead, use it only after we're below a configurable offset. Signed-off-by: Sasha Levin <levinsasha928 at gmail.com> --- drivers/block/virtio_blk.c

Currently if VIRTIO_RING_F_INDIRECT_DESC is enabled we will use indirect descriptors even if we have plenty of space in the ring. This means that we take a performance hit at all times due to the overhead of creating indirect descriptors. Instead, use it only after we're below a configurable offset. Signed-off-by: Sasha Levin <levinsasha928 at gmail.com> --- drivers/block/virtio_blk.c