| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAL9ddJcdS1mUx2DW-T5Sje=JGFfJceHdoDnLKgpaHotRpgJQcg@mail.gmail.com>
Date: Fri, 6 Nov 2020 14:27:28 -0800
From: David Awogbemila <awogbemila@...gle.com>
To: Saeed Mahameed <saeed@...nel.org>
Cc: netdev@...r.kernel.org, Catherine Sullivan <csully@...gle.com>,
Yangchun Fu <yangchun@...gle.com>
Subject: Re: [PATCH 4/4] gve: Add support for raw addressing in the tx path
On Tue, Nov 3, 2020 at 4:35 PM Saeed Mahameed <saeed@...nel.org> wrote:
>
> On Tue, 2020-11-03 at 09:46 -0800, David Awogbemila wrote:
> > From: Catherine Sullivan <csully@...gle.com>
> >
> > During TX, skbs' data addresses are dma_map'ed and passed to the NIC.
> > This means that the device can perform DMA directly from these
> > addresses
> > and the driver does not have to copy the buffer content into
> > pre-allocated buffers/qpls (as in qpl mode).
> >
> > Reviewed-by: Yangchun Fu <yangchun@...gle.com>
> > Signed-off-by: Catherine Sullivan <csully@...gle.com>
> > Signed-off-by: David Awogbemila <awogbemila@...gle.com>
> > ---
> > drivers/net/ethernet/google/gve/gve.h | 18 +-
> > drivers/net/ethernet/google/gve/gve_adminq.c | 4 +-
> > drivers/net/ethernet/google/gve/gve_desc.h | 8 +-
> > drivers/net/ethernet/google/gve/gve_tx.c | 207 +++++++++++++++
> > ----
> > 4 files changed, 194 insertions(+), 43 deletions(-)
> >
>
> > static inline u32 gve_num_tx_qpls(struct gve_priv *priv)
> > {
> > - return priv->tx_cfg.num_queues;
> > + if (priv->raw_addressing)
> > + return 0;
> > + else
> > + return priv->tx_cfg.num_queues;
>
> redundant else statement.
Ok, I'll remove the else.
>
> >
> > -static void gve_dma_sync_for_device(struct device *dev, dma_addr_t
> > *page_buses,
> > +static void gve_dma_sync_for_device(struct gve_priv *priv,
> > + dma_addr_t *page_buses,
> > u64 iov_offset, u64 iov_len)
> > {
> > u64 last_page = (iov_offset + iov_len - 1) / PAGE_SIZE;
> > u64 first_page = iov_offset / PAGE_SIZE;
> > - dma_addr_t dma;
> > u64 page;
> >
> > for (page = first_page; page <= last_page; page++) {
> > - dma = page_buses[page];
> > - dma_sync_single_for_device(dev, dma, PAGE_SIZE,
> > DMA_TO_DEVICE);
> > + dma_addr_t dma = page_buses[page];
> > +
> > + dma_sync_single_for_device(&priv->pdev->dev, dma,
> > PAGE_SIZE, DMA_TO_DEVICE);
>
> Why did you change the function params to pass priv here ?
> I don't see any valid reason.
Good point, I will revert this.
>
> ...
>
> >
> > - gve_dma_sync_for_device(dev, tx->tx_fifo.qpl->page_buses,
> > + gve_dma_sync_for_device(priv, tx->tx_fifo.qpl->page_buses,
> > info->iov[hdr_nfrags - 1].iov_offset,
> > info->iov[hdr_nfrags - 1].iov_len);
> >
> ...
>
> > - gve_dma_sync_for_device(dev, tx->tx_fifo.qpl-
> > >page_buses,
> > + gve_dma_sync_for_device(priv, tx->tx_fifo.qpl-
> > >page_buses,
> > info->iov[i].iov_offset,
> > info->iov[i].iov_len);
> > copy_offset += info->iov[i].iov_len;
> > @@ -472,6 +499,98 @@ static int gve_tx_add_skb(struct gve_tx_ring
> > *tx, struct sk_buff *skb,
> > return 1 + payload_nfrags;
> > }
> >
> > +static int gve_tx_add_skb_no_copy(struct gve_priv *priv, struct
> > gve_tx_ring *tx,
> > + struct sk_buff *skb)
> > +{
> > + const struct skb_shared_info *shinfo = skb_shinfo(skb);
> > + int hlen, payload_nfrags, l4_hdr_offset, seg_idx_bias;
> > + union gve_tx_desc *pkt_desc, *seg_desc;
> > + struct gve_tx_buffer_state *info;
> > + bool is_gso = skb_is_gso(skb);
> > + u32 idx = tx->req & tx->mask;
> > + struct gve_tx_dma_buf *buf;
> > + int last_mapped = 0;
> > + u64 addr;
> > + u32 len;
> > + int i;
> > +
> > + info = &tx->info[idx];
> > + pkt_desc = &tx->desc[idx];
> > +
> > + l4_hdr_offset = skb_checksum_start_offset(skb);
> > + /* If the skb is gso, then we want only up to the tcp header in
> > the first segment
> > + * to efficiently replicate on each segment otherwise we want
> > the linear portion
> > + * of the skb (which will contain the checksum because skb-
> > >csum_start and
> > + * skb->csum_offset are given relative to skb->head) in the
> > first segment.
> > + */
> > + hlen = is_gso ? l4_hdr_offset + tcp_hdrlen(skb) :
> > + skb_headlen(skb);
> > + len = skb_headlen(skb);
> > +
> > + info->skb = skb;
> > +
> > + addr = dma_map_single(tx->dev, skb->data, len, DMA_TO_DEVICE);
> > + if (unlikely(dma_mapping_error(tx->dev, addr))) {
> > + priv->dma_mapping_error++;
> > + goto drop;
> > + }
> > + buf = &info->buf;
> > + dma_unmap_len_set(buf, len, len);
> > + dma_unmap_addr_set(buf, dma, addr);
> > +
> > + payload_nfrags = shinfo->nr_frags;
> > + if (hlen < len) {
> > + /* For gso the rest of the linear portion of the skb
> > needs to
> > + * be in its own descriptor.
> > + */
> > + payload_nfrags++;
> > + gve_tx_fill_pkt_desc(pkt_desc, skb, is_gso,
> > l4_hdr_offset,
> > + 1 + payload_nfrags, hlen, addr);
> > +
> > + len -= hlen;
> > + addr += hlen;
> > + seg_desc = &tx->desc[(tx->req + 1) & tx->mask];
> > + seg_idx_bias = 2;
> > + gve_tx_fill_seg_desc(seg_desc, skb, is_gso, len, addr);
> > + } else {
> > + seg_idx_bias = 1;
> > + gve_tx_fill_pkt_desc(pkt_desc, skb, is_gso,
> > l4_hdr_offset,
> > + 1 + payload_nfrags, hlen, addr);
> > + }
> > + idx = (tx->req + seg_idx_bias) & tx->mask;
> > +
> > + for (i = 0; i < payload_nfrags - (seg_idx_bias - 1); i++) {
> > + const skb_frag_t *frag = &shinfo->frags[i];
> > +
> > + seg_desc = &tx->desc[idx];
> > + len = skb_frag_size(frag);
> > + addr = skb_frag_dma_map(tx->dev, frag, 0, len,
> > DMA_TO_DEVICE);
> > + if (unlikely(dma_mapping_error(tx->dev, addr))) {
> > + priv->dma_mapping_error++;
>
> don't you need to protect this from parallel access ?
I think you're right. I'll protect it with rtnl_lock/unlock/end.
> > + goto unmap_drop;
> > + }
> > + buf = &tx->info[idx].buf;
> > + tx->info[idx].skb = NULL;
> > + dma_unmap_len_set(buf, len, len);
> > + dma_unmap_addr_set(buf, dma, addr);
> > +
> > + gve_tx_fill_seg_desc(seg_desc, skb, is_gso, len, addr);
> > + idx = (idx + 1) & tx->mask;
> > + }
> > +
> > + return 1 + payload_nfrags;
> > +
> > +unmap_drop:
> > + i--;
> > + for (last_mapped = i + seg_idx_bias; last_mapped >= 0;
> > last_mapped--) {
> > + idx = (tx->req + last_mapped) & tx->mask;
> > + gve_tx_unmap_buf(tx->dev, &tx->info[idx]);
> > + }
> > +drop:
> > + tx->dropped_pkt++;
> > + return 0;
> > +}
> > +
> ...
>
>
Powered by blists - more mailing lists