gve: support unreadable netmem
Subsystems (3)
Harshitha Ramamurthy
maintainerPATCH v2 1/1 gve: support unreadable netmem
From: Mina Almasry <[email protected]>
Declare PP_FLAG_ALLOW_UNREADABLE_NETMEM to turn on unreadable netmem support in GVE.
We also drop any net_iov packets where header split is not enabled. We're unable to process packets where the header landed in unreadable netmem.
Use page_pool_dma_sync_netmem_for_cpu in lieu of dma_sync_single_range_for_cpu to correctly handle unreadable netmem that should not be dma-sync'd.
Disable rx_copybreak optimization if payload is unreadable netmem as that needs access to the payload.
---
Changes in v2:
* Change to handle error paths using existing 'goto error' in gve_rx_dqo (Simon Horman)
--- .../ethernet/google/gve/gve_buffer_mgmt_dqo.c | 5 +++ drivers/net/ethernet/google/gve/gve_rx_dqo.c | 35 ++++++++++++++++--- 2 files changed, 35 insertions(+), 5 deletions(-)
drivers/net/ethernet/google/gve/gve_buffer_mgmt_dqo.c +5 -0
@@ -260,6 +260,11 @@ struct page_pool *gve_rx_create_page_pool(struct gve_priv *priv, .offset = xdp ? XDP_PACKET_HEADROOM : 0, }; +if (priv->header_split_enabled) { + pp.flags |=PP_FLAG_ALLOW_UNREADABLE_NETMEM ; + pp.queue_idx = rx->q_num; + } +return page_pool_create (&pp); }
drivers/net/ethernet/google/gve/gve_rx_dqo.c +30 -5
@@ -718,6 +718,24 @@ static int gve_rx_xsk_dqo(struct napi_struct *napi, struct gve_rx_ring *rx,return 0 ; } +static void gve_dma_sync (struct gve_priv *priv,struct gve_rx_ring *rx, +struct gve_rx_buf_state_dqo *buf_state,u16 buf_len) +{ +struct gve_rx_slot_page_info *page_info = &buf_state->page_info; + +if (rx->dqo.page_pool) { +page_pool_dma_sync_netmem_for_cpu (rx->dqo.page_pool, + page_info->netmem, + page_info->page_offset, + buf_len); + }else { +dma_sync_single_range_for_cpu (&priv->pdev->dev, buf_state->addr, + page_info->page_offset + + page_info->pad, + buf_len,DMA_FROM_DEVICE ); + } +} +/* Returns 0 if descriptor is completed successfully. * Returns -EINVAL if descriptor is invalid. * Returns -ENOMEM if data cannot be copied to skb.
@@ -793,13 +811,18 @@ static int gve_rx_dqo(struct napi_struct *napi, struct gve_rx_ring *rx, rx->rx_hsplit_unsplit_pkt += unsplit; rx->rx_hsplit_bytes += hdr_len;u64_stats_update_end (&rx->statss); + }else if (!rx->ctx.skb_head && rx->dqo.page_pool && +netmem_is_net_iov (buf_state->page_info.netmem)) { +/* when header split is disabled, the header went to the packet +* buffer. If the packet buffer is a net_iov, those can't be +* easily mapped into the kernel space to access the header +* required to process the packet. +*/ + goto error; }/* Sync the portion of dma buffer for CPU to read. */ -dma_sync_single_range_for_cpu (&priv->pdev->dev, buf_state->addr, - buf_state->page_info.page_offset + - buf_state->page_info.pad, - buf_len,DMA_FROM_DEVICE ); +gve_dma_sync (priv, rx, buf_state, buf_len);/* Append to current skb if one exists. */ if (rx->ctx.skb_head) {
@@ -837,7 +860,9 @@ static int gve_rx_dqo(struct napi_struct *napi, struct gve_rx_ring *rx,u64_stats_update_end (&rx->statss); } -if (eop && buf_len <= priv->rx_copybreak) { +if (eop && buf_len <= priv->rx_copybreak && + !(rx->dqo.page_pool && +netmem_is_net_iov (buf_state->page_info.netmem))) { rx->ctx.skb_head =gve_rx_copy (priv->dev, napi, &buf_state->page_info, buf_len);if (unlikely (!rx->ctx.skb_head))
Signature · 1 line
2.51.0.rc0.155.g4a0f42376b-goog
This patch was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This patch was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Quoted 10 lines · On Mon, 18 Aug 2025 21:05:07 +0000 you wrote:
From: Mina Almasry <[email protected]>
Declare
PP_FLAG_ALLOW_UNREADABLE_NETMEMto turn on unreadable netmem support in GVE.We also drop any
net_iovpackets where header split is not enabled. We're unable to process packets where the header landed in unreadable netmem.[...]
Here is the summary with links:
- [net-next,v2] gve: support unreadable netmem
https://git.kernel.org/netdev/net-next/c/62d7f40503bc
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html