Fixes for ENA driver
Subsystems (3)
v2 cover Fixes for ENA driver
Hi all,
This series fixes some issues in the ENA driver:
- fix wrong data offset on machines that support rx offset
- work-around Intel iommu issue
- fix out of bound access when request id is wrong
- return error code if XDP TX xmit fails
Changes from previous version:v1->v2: switched to using dma_set_mask_and_coherent() function
in second patch
Shay Agroskin (4):
net: ena: handle bad request id in ena_netdev
net: ena: set initial DMA width to avoid intel iommu issue
net: ena: fix packet's addresses for rx_offset feature
net: ena: return error code from ena_xdp_xmit_buff
drivers/net/ethernet/amazon/ena/ena_eth_com.c | 3 + drivers/net/ethernet/amazon/ena/ena_netdev.c | 86 +++++++------------ 2 files changed, 36 insertions(+), 53 deletions(-)
Signature · 1 line
2.17.1
PATCH v2 1/4 net: ena: handle bad request id in ena_netdev
After request id is checked in validate_rx_req_id() its value is still used in the line
rx_ring->free_ids[next_to_clean] = rx_ring->ena_bufs[i].req_id;
even if it was found to be out-of-bound for the array free_ids.
The patch moves the request id to an earlier stage in the napi routine and makes sure its value isn't used if it's found out-of-bounds.
--- drivers/net/ethernet/amazon/ena/ena_eth_com.c | 3 ++ drivers/net/ethernet/amazon/ena/ena_netdev.c | 43 +++++-------------- 2 files changed, 14 insertions(+), 32 deletions(-)
drivers/net/ethernet/amazon/ena/ena_eth_com.c +3 -0
@@ -516,6 +516,7 @@ int ena_com_rx_pkt(struct ena_com_io_cq *io_cq, {struct ena_com_rx_buf_info *ena_buf = &ena_rx_ctx->ena_bufs[0 ];struct ena_eth_io_rx_cdesc_base *cdesc =NULL ; +u16 q_depth = io_cq->q_depth;u16 cdesc_idx =0 ;u16 nb_hw_desc;u16 i =0 ;
@@ -543,6 +544,8 @@ int ena_com_rx_pkt(struct ena_com_io_cq *io_cq, do { ena_buf[i].len = cdesc->length; ena_buf[i].req_id = cdesc->req_id; + if (unlikely(ena_buf[i].req_id >= q_depth)) + return -EIO; if (++i >= nb_hw_desc) break;
drivers/net/ethernet/amazon/ena/ena_netdev.c +11 -32
72 lines not on this page · raw message
Signature · 1 line
2.17.1
PATCH v2 2/4 net: ena: set initial DMA width to avoid intel iommu issue
In reply to Shay AgroskinThe ENA driver uses the readless mechanism, which uses DMA, to find out what the DMA mask is supposed to be.
If DMA is used without setting the dma_mask first, it causes the Intel IOMMU driver to think that ENA is a 32-bit device and therefore disables IOMMU passthrough permanently.
This patch sets the dma_mask to be ENA_MAX_PHYS_ADDR_SIZE_BITS=48 before readless initialization in ena_device_init()->ena_com_mmio_reg_read_request_init(), which is large enough to workaround the intel_iommu issue.
DMA mask is set again to the correct value after it's received from the device after readless is initialized.
The patch also changes the driver to use dma_set_mask_and_coherent() function instead of the two pci_set_dma_mask() and pci_set_consistent_dma_mask() ones. Both methods achieve the same effect.
--- drivers/net/ethernet/amazon/ena/ena_netdev.c | 17 ++++++++--------- 1 file changed, 8 insertions(+), 9 deletions(-)
drivers/net/ethernet/amazon/ena/ena_netdev.c +8 -9
@@ -3367,16 +3367,9 @@ static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev, goto err_mmio_read_less; } - rc =pci_set_dma_mask (pdev,DMA_BIT_MASK (dma_width)); + rc =dma_set_mask_and_coherent (dev,DMA_BIT_MASK (dma_width));if (rc) { -dev_err (dev,"pci_set_dma_mask failed 0x%x\n" , rc); - goto err_mmio_read_less; - } - - rc =pci_set_consistent_dma_mask (pdev,DMA_BIT_MASK (dma_width)); -if (rc) { -dev_err (dev,"err_pci_set_consistent_dma_mask failed 0x%x\n" , - rc); +dev_err (dev,"dma_set_mask_and_coherent failed %d\n" , rc); goto err_mmio_read_less; }
@@ -4146,6 +4139,12 @@ static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)return rc; } + rc =dma_set_mask_and_coherent (&pdev->dev,DMA_BIT_MASK (ENA_MAX_PHYS_ADDR_SIZE_BITS )); +if (rc) { +dev_err (&pdev->dev,"dma_set_mask_and_coherent failed %d\n" , rc); + goto err_disable_device; + } +pci_set_master (pdev); ena_dev =vzalloc (sizeof (*ena_dev));
Signature · 1 line
2.17.1
PATCH v2 3/4 net: ena: fix packet's addresses for rx_offset feature
In reply to Shay AgroskinThis patch fixes two lines in which the rx_offset received by the device wasn't taken into account:
- prefetch function:
In our driver the copied data would reside in rx_info->page + rx_headroom + rx_offset
so the prefetch function is changed accordingly.
- setting page_offset to zero for descriptors > 1:
for every descriptor but the first, the rx_offset is zero. Hence the page_offset value should be set to rx_headroom.
The previous implementation changed the value of rx_info after the descriptor was added to the SKB (essentially providing wrong page offset).
--- drivers/net/ethernet/amazon/ena/ena_netdev.c | 20 +++++++++++--------- 1 file changed, 11 insertions(+), 9 deletions(-)
drivers/net/ethernet/amazon/ena/ena_netdev.c +11 -9
@@ -908,10 +908,14 @@ static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)static int ena_alloc_rx_page (struct ena_ring *rx_ring,struct ena_rx_buffer *rx_info,gfp_t gfp) { +int headroom = rx_ring->rx_headroom;struct ena_com_buf *ena_buf;struct page *page;dma_addr_t dma; +/* restore page offset value in case it has been changed by device */ + rx_info->page_offset = headroom; +/* if previous allocated page is not used */ if (unlikely (rx_info->page))return 0 ;
@@ -941,10 +945,9 @@ static int ena_alloc_rx_page(struct ena_ring *rx_ring, "Allocate page %p, rx_info %p\n", page, rx_info); rx_info->page = page; - rx_info->page_offset =0 ; ena_buf = &rx_info->ena_buf; - ena_buf->paddr = dma + rx_ring->rx_headroom; - ena_buf->len =ENA_PAGE_SIZE - rx_ring->rx_headroom; + ena_buf->paddr = dma + headroom; + ena_buf->len =ENA_PAGE_SIZE - headroom;return 0 ; }
@@ -1356,7 +1359,8 @@ static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,/* save virt address of first buffer */ va =page_address (rx_info->page) + rx_info->page_offset; -prefetch (va +NET_IP_ALIGN ); + +prefetch (va);if (len <= rx_ring->rx_copybreak) { skb =ena_alloc_skb (rx_ring, false);
@@ -1397,8 +1401,6 @@ static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,skb_add_rx_frag (skb,skb_shinfo (skb)->nr_frags, rx_info->page, rx_info->page_offset, len,ENA_PAGE_SIZE ); -/* The offset is non zero only for the first buffer */ - rx_info->page_offset =0 ; netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev, "RX skb updated. len %d. data_len %d\n",
@@ -1517,8 +1519,7 @@ static int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)int ret; rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0 ].req_id]; - xdp->data =page_address (rx_info->page) + - rx_info->page_offset + rx_ring->rx_headroom; + xdp->data =page_address (rx_info->page) + rx_info->page_offset;xdp_set_data_meta_invalid (xdp); xdp->data_hard_start =page_address (rx_info->page); xdp->data_end = xdp->data + rx_ring->ena_bufs[0 ].len;
@@ -1585,8 +1586,9 @@ static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,if (unlikely (ena_rx_ctx.descs ==0 ))break ; +/* First descriptor might have an offset set by the device */ rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0 ].req_id]; - rx_info->page_offset = ena_rx_ctx.pkt_offset; + rx_info->page_offset += ena_rx_ctx.pkt_offset; netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev, "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
Signature · 1 line
2.17.1
PATCH v2 4/4 net: ena: return error code from ena_xdp_xmit_buff
In reply to Shay AgroskinThe function mistakenly returns NETDEV_TX_OK regardless of the transmission success. This patch fixes this behavior by returning the error code from the function.
--- drivers/net/ethernet/amazon/ena/ena_netdev.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/net/ethernet/amazon/ena/ena_netdev.c +3 -3
@@ -284,9 +284,9 @@ static int ena_xdp_xmit_buff(struct net_device *dev,struct ena_tx_buffer *tx_info;struct ena_ring *xdp_ring;u16 next_to_use, req_id; -int rc;void *push_hdr;u32 push_len; +int rc; xdp_ring = &adapter->tx_ring[qid]; next_to_use = xdp_ring->next_to_use;
@@ -322,14 +322,14 @@ static int ena_xdp_xmit_buff(struct net_device *dev, xdp_ring->tx_stats.doorbells++;u64_stats_update_end (&xdp_ring->syncp); -return NETDEV_TX_OK ; +return rc; error_unmap_dma:ena_unmap_tx_buff (xdp_ring, tx_info); tx_info->xdpf =NULL ; error_drop_packet:__free_page (tx_info->xdp_rx_page); -return NETDEV_TX_OK ; +return rc; } static int ena_xdp_execute(struct ena_ring *rx_ring,
Signature · 1 line
2.17.1
Jakub Kicinski
maintainerRe: [PATCH V2 net 4/4] net: ena: return error code from ena_xdp_xmit_buff
Quoted 6 lines · On Thu, 19 Nov 2020 22:28:51 +0200 Shay Agroskin wrote:
The function mistakenly returns
NETDEV_TX_OKregardless of the transmission success. This patch fixes this behavior by returning the error code from the function.Fixes 548c4940b9f1 ("net: ena: Implement XDP_TX action")Signed-off-by Shay Agroskin <[email protected]>
Doesn't seem like a legitimate bug fix, since the only caller of this function ignores its return value.
Re: [PATCH V2 net 4/4] net: ena: return error code from ena_xdp_xmit_buff
Quoted 10 lines · Jakub Kicinski
Quoted 6 lines · On Thu, 19 Nov 2020 22:28:51 +0200 Shay Agroskin wrote:
The function mistakenly returns
NETDEV_TX_OKregardless of the transmission success. This patch fixes this behavior by returning the error code from the function.Fixes 548c4940b9f1 ("net: ena: Implement XDP_TX action")Signed-off-by Shay Agroskin <[email protected]>Doesn't seem like a legitimate bug fix, since the only caller of this function ignores its return value.
Hi,
I plan to use the return value from this function in future patch (next-net series), do you think we better send this fix with this future patch?
Shay
Jakub Kicinski
maintainerRe: [PATCH V2 net 4/4] net: ena: return error code from ena_xdp_xmit_buff
Quoted 11 lines · On Sun, 22 Nov 2020 09:19:05 +0200 Shay Agroskin wrote:
Quoted 9 lines · Jakub Kicinski
Quoted 7 lines · On Thu, 19 Nov 2020 22:28:51 +0200 Shay Agroskin wrote:
The function mistakenly returns
NETDEV_TX_OKregardless of the transmission success. This patch fixes this behavior by returning the
error code from the function.Fixes 548c4940b9f1 ("net: ena: Implement XDP_TX action")Signed-off-by Shay Agroskin <[email protected]>
Quoted 8 lines
Quoted 3 lines
Doesn't seem like a legitimate bug fix, since the only caller of this
function ignores its return value.Hi,
I plan to use the return value from this function in future patch (next-net series), do you think we better send this fix with this future patch?
Yes, it's fine to include this in a net-next series. It doesn't fix any bug that could cause trouble to the users. If you had a fix that depended on it then maybe, but if you only need it in net-next we can put it there with clear conscience.