Skip to content

virtio_net: apply DMA cache maintenance for a non-coherent rproc peer #15

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
UlasSertan wants to merge 1 commit into
base: v6.12.24-ti-arm64-r43-t3-gem-o1
Choose a base branch
from
Open

virtio_net: apply DMA cache maintenance for a non-coherent rproc peer #15

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
43 changes: 43 additions & 0 deletions drivers/net/virtio_net.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,7 @@
#include <net/netdev_rx_queue.h>
#include <net/netdev_queues.h>
#include <net/xdp_sock_drv.h>
#include <linux/dma-mapping.h>

static int napi_weight = NAPI_POLL_WEIGHT;
module_param(napi_weight, int, 0444);
Expand Down Expand Up @@ -875,6 +876,25 @@ static struct sk_buff *page_to_skb(struct virtnet_info *vi,
return skb;
}

/* T3-GEM-O1 LOCAL PATCH: the R5F remoteproc virtio peer is not cache-
* coherent with the A53, and the legacy rproc vdev cannot negotiate
* VIRTIO_F_ACCESS_PLATFORM (32-bit resource-table feature field), so the
* core vring never applies DMA sync to payload buffers. Do explicit
* cache maintenance against the vdev's parent (the rvdev platform
* device, non-coherent dma-direct) around every payload access.
*/
static void virtnet_cache_sync(struct virtio_device *vdev, void *buf,
unsigned int len, enum dma_data_direction dir,
bool for_cpu)
{
struct device *dev = vdev->dev.parent;

if (for_cpu)
dma_sync_single_for_cpu(dev, virt_to_phys(buf), len, dir);
else
dma_sync_single_for_device(dev, virt_to_phys(buf), len, dir);
}

static void virtnet_rq_unmap(struct receive_queue *rq, void *buf, u32 len)
{
struct page *page = virt_to_head_page(buf);
Expand Down Expand Up @@ -911,6 +931,9 @@ static void *virtnet_rq_get_buf(struct receive_queue *rq, u32 *len, void **ctx)
buf = virtqueue_get_buf_ctx(rq->vq, len, ctx);
if (buf && rq->do_dma)
virtnet_rq_unmap(rq, buf, *len);
else if (buf) /* T3-GEM-O1: invalidate before reading peer data */
virtnet_cache_sync(rq->vq->vdev, buf, *len,
DMA_FROM_DEVICE, true);

return buf;
}
Expand All @@ -924,6 +947,13 @@ static void virtnet_rq_init_one_sg(struct receive_queue *rq, void *buf, u32 len)

if (!rq->do_dma) {
sg_init_one(rq->sg, buf, len);

/* T3-GEM-O1: write back any dirty lines so later evictions
* cannot overwrite data the peer writes into this buffer.
*/

virtnet_cache_sync(rq->vq->vdev, buf, len,
DMA_FROM_DEVICE, false);
return;
}

Expand Down Expand Up @@ -3072,6 +3102,19 @@ static int xmit_skb(struct send_queue *sq, struct sk_buff *skb, bool orphan)
return num_sg;
num_sg++;
}
{
/* T3-GEM-O1: push hdr+payload to memory for the
* non-coherent R5F peer before publishing.
*/

struct scatterlist *sg;
int i;

for_each_sg(sq->sg, sg, num_sg, i)
virtnet_cache_sync(sq->vq->vdev, sg_virt(sg),
sg->length, DMA_TO_DEVICE, false);
}

return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg,
skb_to_ptr(skb, orphan), GFP_ATOMIC);
}
Expand Down