From: Bob Pearson <rpearsonhpe@gmail.com>
To: jgg@nvidia.com, zyjzyj2000@gmail.com, linux-rdma@vger.kernel.org
Subject: Re: [PATCH 7/7] RDMA/rxe: Extend ICRC to support nonlinear skbs
Date: Tue, 29 Jun 2021 15:16:16 -0500 [thread overview]
Message-ID: <b2736253-e066-4148-66e5-2ae9601231b6@gmail.com> (raw)
In-Reply-To: <20210629201412.28306-8-rpearsonhpe@gmail.com>
On 6/29/21 3:14 PM, Bob Pearson wrote:
> Make ICRC calculations aware of potential non-linear skbs.
> This is a step towards getting rid of skb_linearize() and its
> extra data copy.
>
> Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
> ---
> drivers/infiniband/sw/rxe/rxe_icrc.c | 150 +++++++++++++++++----------
> drivers/infiniband/sw/rxe/rxe_loc.h | 4 +-
> drivers/infiniband/sw/rxe/rxe_net.c | 7 +-
> drivers/infiniband/sw/rxe/rxe_recv.c | 2 +-
> 4 files changed, 103 insertions(+), 60 deletions(-)
>
> diff --git a/drivers/infiniband/sw/rxe/rxe_icrc.c b/drivers/infiniband/sw/rxe/rxe_icrc.c
> index f5ebd9d23d12..d730c76bbeae 100644
> --- a/drivers/infiniband/sw/rxe/rxe_icrc.c
> +++ b/drivers/infiniband/sw/rxe/rxe_icrc.c
> @@ -63,97 +63,134 @@ static __be32 rxe_crc32(struct rxe_dev *rxe, __be32 crc, void *addr,
> }
>
> /**
> - * rxe_icrc_hdr - Compute a partial ICRC for the IB transport headers.
> + * rxe_icrc_packet - Compute the ICRC for a packet
> * @skb: packet buffer
> * @pkt: packet information
> + * @icrcp: pointer to returned ICRC
> *
> - * Returns the partial ICRC
> + * Support linear or nonlinear skbs with frags
> + *
> + * Returns ICRC in *icrcp and 0 if no error occurs
> + * else returns an error.
> * For details see the InfiniBand Architecture spec and Annex 17
> * the RoCE v2 spec.
> */
> -static __be32 rxe_icrc_hdr(struct sk_buff *skb, struct rxe_pkt_info *pkt)
> +static int rxe_icrc_packet(struct sk_buff *skb, struct rxe_pkt_info *pkt,
> + __be32 *icrcp)
> {
> + struct skb_shared_info *info = skb_shinfo(skb);
> + struct rxe_dev *rxe = pkt->rxe;
> + struct iphdr *ip4h;
> + struct ipv6hdr *ip6h;
> struct udphdr *udph;
> struct rxe_bth *bth;
> - __be32 crc;
> - int length;
> - int hdr_size = sizeof(struct udphdr) +
> + __be32 icrc;
> + int hdr_size;
> + u8 pseudo_hdr[128];
> + int resid;
> + int bytes;
> + int nfrag;
> + skb_frag_t *frag;
> + u8 *addr;
> + int page_offset;
> + int start;
> + int len;
> + int ret;
> +
> + hdr_size = rxe_opcode[pkt->opcode].length + sizeof(struct udphdr) +
> (skb->protocol == htons(ETH_P_IP) ?
> - sizeof(struct iphdr) : sizeof(struct ipv6hdr));
> - /* pseudo header buffer size is calculate using ipv6 header size since
> - * it is bigger than ipv4
> - */
> - u8 pshdr[sizeof(struct udphdr) +
> - sizeof(struct ipv6hdr) +
> - RXE_BTH_BYTES];
> -
> - /* This seed is the result of computing a CRC with a seed of
> - * 0xfffffff and 8 bytes of 0xff representing a masked LRH.
> - */
> - crc = 0xdebb20e3;
> + sizeof(struct iphdr) : sizeof(struct ipv6hdr));
>
> - if (skb->protocol == htons(ETH_P_IP)) { /* IPv4 */
> - struct iphdr *ip4h;
> + start = skb->network_header + skb->head - skb->data;
> + ret = skb_copy_bits(skb, start, pseudo_hdr, hdr_size);
> + if (unlikely(ret)) {
> + pr_warn_ratelimited("Malformed skb\n");
> + return ret;
> + }
>
> - memcpy(pshdr, ip_hdr(skb), hdr_size);
> - ip4h = (struct iphdr *)pshdr;
> + if (skb->protocol == htons(ETH_P_IP)) { /* IPv4 */
> + ip4h = (struct iphdr *)pseudo_hdr;
> udph = (struct udphdr *)(ip4h + 1);
> + bth = (struct rxe_bth *)(udph + 1);
>
> ip4h->ttl = 0xff;
> ip4h->check = CSUM_MANGLED_0;
> ip4h->tos = 0xff;
> } else { /* IPv6 */
> - struct ipv6hdr *ip6h;
> -
> - memcpy(pshdr, ipv6_hdr(skb), hdr_size);
> - ip6h = (struct ipv6hdr *)pshdr;
> + ip6h = (struct ipv6hdr *)pseudo_hdr;
> udph = (struct udphdr *)(ip6h + 1);
> + bth = (struct rxe_bth *)(udph + 1);
>
> - memset(ip6h->flow_lbl, 0xff, sizeof(ip6h->flow_lbl));
> ip6h->priority = 0xf;
> ip6h->hop_limit = 0xff;
> }
>
> udph->check = CSUM_MANGLED_0;
> -
> - bth = (struct rxe_bth *)(udph + 1);
> - memcpy(bth, pkt->hdr, RXE_BTH_BYTES);
> -
> - /* exclude bth.resv8a */
> bth->qpn |= cpu_to_be32(~BTH_QPN_MASK);
>
> - length = hdr_size + RXE_BTH_BYTES;
> - crc = rxe_crc32(pkt->rxe, crc, pshdr, length);
> + icrc = 0xdebb20e3;
> + icrc = rxe_crc32(pkt->rxe, icrc, pseudo_hdr, hdr_size);
> +
> + resid = (payload_size(pkt) + 0x3) & ~0x3;
> + nfrag = -1;
> +
> + while (resid) {
> + if (nfrag < 0) {
> + addr = skb_network_header(skb) + hdr_size;
> + len = skb_tail_pointer(skb) - skb_network_header(skb);
> + } else if (nfrag < info->nr_frags) {
> + frag = &info->frags[nfrag];
> + page_offset = frag->bv_offset + hdr_size;
> + addr = kmap_atomic(frag->bv_page) + page_offset;
> + len = frag->bv_len;
> + } else {
> + pr_warn_ratelimited("Malformed skb\n");
> + return -EINVAL;
> + }
> +
> + bytes = len - hdr_size;
> + if (bytes > 0) {
> + if (bytes > resid)
> + bytes = resid;
> + icrc = rxe_crc32(rxe, icrc, addr, bytes);
> + resid -= bytes;
> + hdr_size = 0;
> + } else {
> + hdr_size -= len;
> + }
> +
> + if (nfrag++ >= 0)
> + kunmap_atomic(addr);
> + }
> +
> + *icrcp = ~icrc;
>
> - /* And finish to compute the CRC on the remainder of the headers. */
> - crc = rxe_crc32(pkt->rxe, crc, pkt->hdr + RXE_BTH_BYTES,
> - rxe_opcode[pkt->opcode].length - RXE_BTH_BYTES);
> - return crc;
> + return 0;
> }
>
> /**
> * rxe_check_icrc - Compute ICRC for a packet and compare to the ICRC
> - * delivered in the packet.
> - * @skb: packet buffer with packet info in cb[] (receive path)
> + * in the packet.
> + * @skb: packet buffer
> + * @pkt: packet information
> *
> * Returns 0 if the ICRCs match or an error on failure
> */
> -int rxe_icrc_check(struct sk_buff *skb)
> +int rxe_icrc_check(struct sk_buff *skb, struct rxe_pkt_info *pkt)
> {
> - struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
> __be32 *icrcp;
> __be32 packet_icrc;
> - __be32 computed_icrc;
> + __be32 icrc;
> + int ret;
>
> icrcp = (__be32 *)(pkt->hdr + pkt->paylen - RXE_ICRC_SIZE);
> packet_icrc = *icrcp;
>
> - computed_icrc = rxe_icrc_hdr(skb, pkt);
> - computed_icrc = rxe_crc32(pkt->rxe, computed_icrc,
> - (u8 *)payload_addr(pkt), payload_size(pkt) + bth_pad(pkt));
> - computed_icrc = ~computed_icrc;
> + ret = rxe_icrc_packet(skb, pkt, &icrc);
> + if (unlikely(ret))
> + return ret;
>
> - if (unlikely(computed_icrc != packet_icrc)) {
> + if (unlikely(icrc != packet_icrc)) {
> if (skb->protocol == htons(ETH_P_IPV6))
> pr_warn_ratelimited("bad ICRC from %pI6c\n",
> &ipv6_hdr(skb)->saddr);
> @@ -162,7 +199,6 @@ int rxe_icrc_check(struct sk_buff *skb)
> &ip_hdr(skb)->saddr);
> else
> pr_warn_ratelimited("bad ICRC from unknown\n");
> -
> return -EINVAL;
> }
>
> @@ -174,15 +210,19 @@ int rxe_icrc_check(struct sk_buff *skb)
> * correct position after the payload and pad.
> * @skb: packet buffer
> * @pkt: packet information
> + *
> + * Returns 0 on success or an error
> */
> -void rxe_icrc_generate(struct sk_buff *skb, struct rxe_pkt_info *pkt)
> +int rxe_icrc_generate(struct sk_buff *skb, struct rxe_pkt_info *pkt)
> {
> __be32 *icrcp;
> - __be32 icrc;
> + int ret;
>
> icrcp = (__be32 *)(pkt->hdr + pkt->paylen - RXE_ICRC_SIZE);
> - icrc = rxe_icrc_hdr(skb, pkt);
> - icrc = rxe_crc32(pkt->rxe, icrc, (u8 *)payload_addr(pkt),
> - payload_size(pkt) + bth_pad(pkt));
> - *icrcp = ~icrc;
> +
> + ret = rxe_icrc_packet(skb, pkt, icrcp);
> + if (unlikely(ret))
> + return ret;
> +
> + return 0;
> }
> diff --git a/drivers/infiniband/sw/rxe/rxe_loc.h b/drivers/infiniband/sw/rxe/rxe_loc.h
> index e8e87336469b..09836cdb1e89 100644
> --- a/drivers/infiniband/sw/rxe/rxe_loc.h
> +++ b/drivers/infiniband/sw/rxe/rxe_loc.h
> @@ -194,8 +194,8 @@ int rxe_responder(void *arg);
>
> /* rxe_icrc.c */
> int rxe_icrc_init(struct rxe_dev *rxe);
> -int rxe_icrc_check(struct sk_buff *skb);
> -void rxe_icrc_generate(struct sk_buff *skb, struct rxe_pkt_info *pkt);
> +int rxe_icrc_check(struct sk_buff *skb, struct rxe_pkt_info *pkt);
> +int rxe_icrc_generate(struct sk_buff *skb, struct rxe_pkt_info *pkt);
>
> void rxe_resp_queue_pkt(struct rxe_qp *qp, struct sk_buff *skb);
>
> diff --git a/drivers/infiniband/sw/rxe/rxe_net.c b/drivers/infiniband/sw/rxe/rxe_net.c
> index 4d109e5b33ff..d708ff19e774 100644
> --- a/drivers/infiniband/sw/rxe/rxe_net.c
> +++ b/drivers/infiniband/sw/rxe/rxe_net.c
> @@ -434,8 +434,11 @@ int rxe_xmit_packet(struct rxe_qp *qp, struct rxe_pkt_info *pkt,
> goto drop;
> }
>
> - if (rxe_must_generate_icrc)
> - rxe_icrc_generate(skb, pkt);
> + if (rxe_must_generate_icrc) {
> + err = rxe_icrc_generate(skb, pkt);
> + if (unlikely(err))
> + goto drop;
> + }
>
> if (pkt->mask & RXE_LOOPBACK_MASK) {
> memcpy(SKB_TO_PKT(skb), pkt, sizeof(*pkt));
> diff --git a/drivers/infiniband/sw/rxe/rxe_recv.c b/drivers/infiniband/sw/rxe/rxe_recv.c
> index 01d425b3991e..7f51b9e92437 100644
> --- a/drivers/infiniband/sw/rxe/rxe_recv.c
> +++ b/drivers/infiniband/sw/rxe/rxe_recv.c
> @@ -383,7 +383,7 @@ void rxe_rcv(struct sk_buff *skb)
> goto drop;
>
> if (rxe_must_check_icrc) {
> - err = rxe_icrc_check(skb);
> + err = rxe_icrc_check(skb, pkt);
> if (unlikely(err))
> goto drop;
> }
>
Please ignore. This was sent in error. Only patches 0-5 belong.
next prev parent reply other threads:[~2021-06-29 20:16 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-06-29 20:14 [PATCH 0/7] RDMA/rxe: Cleanup ICRC Bob Pearson
2021-06-29 20:14 ` [PATCH 1/5] RDMA/rxe: Move ICRC checking to a subroutine Bob Pearson
2021-06-29 20:14 ` [PATCH 2/5] RDMA/rxe: Move rxe_xmit_packet " Bob Pearson
2021-06-29 20:14 ` [PATCH 3/5] RDMA/rxe: Move ICRC generation " Bob Pearson
2021-06-29 20:14 ` [PATCH 4/5] RDMA/rxe: Move rxe_crc32 " Bob Pearson
2021-06-29 20:14 ` [PATCH 5/5] RDMA/rxe: Move crc32 init code to rxe_icrc.c Bob Pearson
2021-06-30 4:09 ` Zhu Yanjun
2021-06-30 4:13 ` Pearson, Robert B
2021-07-01 3:38 ` kernel test robot
2021-06-29 20:14 ` [PATCH 6/7] RDMA/rxe: Add parameters to control checking/generating ICRC Bob Pearson
2021-06-29 20:14 ` [PATCH 7/7] RDMA/rxe: Extend ICRC to support nonlinear skbs Bob Pearson
2021-06-29 20:16 ` Bob Pearson [this message]
2021-06-29 20:17 ` Bob Pearson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=b2736253-e066-4148-66e5-2ae9601231b6@gmail.com \
--to=rpearsonhpe@gmail.com \
--cc=jgg@nvidia.com \
--cc=linux-rdma@vger.kernel.org \
--cc=zyjzyj2000@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox