From: "Ananyev, Konstantin" <konstantin.ananyev@intel.com>
To: "Ouyang, Changchun" <changchun.ouyang@intel.com>,
"dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [dpdk-dev] [PATCH 2/3] pmd: RX function need keep EXTERNAL_MBUF flag
Date: Fri, 24 Oct 2014 10:46:06 +0000 [thread overview]
Message-ID: <2601191342CEEE43887BDE71AB9772582139EBAF@IRSMSX105.ger.corp.intel.com> (raw)
In-Reply-To: <1414138209-24431-3-git-send-email-changchun.ouyang@intel.com>
Hi Changchun,
> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Ouyang Changchun
> Sent: Friday, October 24, 2014 9:10 AM
> To: dev@dpdk.org
> Subject: [dpdk-dev] [PATCH 2/3] pmd: RX function need keep EXTERNAL_MBUF flag
>
> Every pmd RX function need keep the EXTERNAL_MBUF flag
> in mbuf.ol_flags, and can't overwrite it when filling ol_flags from
> descriptor to mbuf, otherwise, it probably cause to crash when freeing a mbuf
> and trying to freeing its attached external buffer, say, from guest space.
>
Don't really like the idea to put:
mb->ol_flags = pkt_flags | (mb->ol_flags & EXTERNAL_MBUF);
in each and every PMD from now on...
>From other side, it is probably not very good that RX functions update whole ol_flags, not only RX related part.
Wonder can we reserve low 32bits of ol_flags for RX, and high 32bits for TX and generic stuff.
So our ol_flags will look something like that:
union {
uint64_t ol_raw_flags;
struct {
uint32_t rx;
uint32_t gen_tx;
} ol_flags
};
And make all PMD RX functions to operate on rx part of the flags only:
mb->ol_flags.rx = pkt_flags;
?
Konstantin
> Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
> ---
> lib/librte_pmd_e1000/igb_rxtx.c | 5 +++--
> lib/librte_pmd_i40e/i40e_rxtx.c | 8 +++++---
> lib/librte_pmd_ixgbe/ixgbe_rxtx.c | 8 +++++---
> lib/librte_pmd_ixgbe/ixgbe_rxtx_vec.c | 12 ++++++++----
> 4 files changed, 21 insertions(+), 12 deletions(-)
>
> diff --git a/lib/librte_pmd_e1000/igb_rxtx.c b/lib/librte_pmd_e1000/igb_rxtx.c
> index f09c525..4123310 100644
> --- a/lib/librte_pmd_e1000/igb_rxtx.c
> +++ b/lib/librte_pmd_e1000/igb_rxtx.c
> @@ -786,7 +786,7 @@ eth_igb_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
> pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
> pkt_flags = pkt_flags | rx_desc_status_to_pkt_flags(staterr);
> pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
> - rxm->ol_flags = pkt_flags;
> + rxm->ol_flags = pkt_flags | (rxm->ol_flags & EXTERNAL_MBUF);
>
> /*
> * Store the mbuf address into the next entry of the array
> @@ -1020,7 +1020,8 @@ eth_igb_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
> pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
> pkt_flags = pkt_flags | rx_desc_status_to_pkt_flags(staterr);
> pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
> - first_seg->ol_flags = pkt_flags;
> + first_seg->ol_flags = pkt_flags |
> + (first_seg->ol_flags & EXTERNAL_MBUF);
>
> /* Prefetch data of first segment, if configured to do so. */
> rte_packet_prefetch((char *)first_seg->buf_addr +
> diff --git a/lib/librte_pmd_i40e/i40e_rxtx.c b/lib/librte_pmd_i40e/i40e_rxtx.c
> index 2b53677..68c3695 100644
> --- a/lib/librte_pmd_i40e/i40e_rxtx.c
> +++ b/lib/librte_pmd_i40e/i40e_rxtx.c
> @@ -637,7 +637,8 @@ i40e_rx_scan_hw_ring(struct i40e_rx_queue *rxq)
> pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
> pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
> pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
> - mb->ol_flags = pkt_flags;
> + mb->ol_flags = pkt_flags |
> + (mb->ol_flags & EXTERNAL_MBUF);
> if (pkt_flags & PKT_RX_RSS_HASH)
> mb->hash.rss = rte_le_to_cpu_32(\
> rxdp->wb.qword0.hi_dword.rss);
> @@ -873,7 +874,7 @@ i40e_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
> pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
> pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
> pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
> - rxm->ol_flags = pkt_flags;
> + rxm->ol_flags = pkt_flags | (rxm->ol_flags & EXTERNAL_MBUF);
> if (pkt_flags & PKT_RX_RSS_HASH)
> rxm->hash.rss =
> rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
> @@ -1027,7 +1028,8 @@ i40e_recv_scattered_pkts(void *rx_queue,
> pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
> pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
> pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
> - first_seg->ol_flags = pkt_flags;
> + first_seg->ol_flags = pkt_flags |
> + (first_seg->ol_flags & EXTERNAL_MBUF);
> if (pkt_flags & PKT_RX_RSS_HASH)
> rxm->hash.rss =
> rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
> diff --git a/lib/librte_pmd_ixgbe/ixgbe_rxtx.c b/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
> index 1aefe5c..77e8689 100644
> --- a/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
> +++ b/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
> @@ -949,7 +949,8 @@ ixgbe_rx_scan_hw_ring(struct igb_rx_queue *rxq)
> /* reuse status field from scan list */
> pkt_flags |= rx_desc_status_to_pkt_flags(s[j]);
> pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
> - mb->ol_flags = pkt_flags;
> + mb->ol_flags = pkt_flags |
> + (mb->ol_flags & EXTERNAL_MBUF);
>
> if (likely(pkt_flags & PKT_RX_RSS_HASH))
> mb->hash.rss = rxdp[j].wb.lower.hi_dword.rss;
> @@ -1271,7 +1272,7 @@ ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
> pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
> pkt_flags = pkt_flags | rx_desc_status_to_pkt_flags(staterr);
> pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
> - rxm->ol_flags = pkt_flags;
> + rxm->ol_flags = pkt_flags | (rxm->ol_flags & EXTERNAL_MBUF);
>
> if (likely(pkt_flags & PKT_RX_RSS_HASH))
> rxm->hash.rss = rxd.wb.lower.hi_dword.rss;
> @@ -1515,7 +1516,8 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
> rx_desc_status_to_pkt_flags(staterr));
> pkt_flags = (uint16_t)(pkt_flags |
> rx_desc_error_to_pkt_flags(staterr));
> - first_seg->ol_flags = pkt_flags;
> + first_seg->ol_flags = pkt_flags |
> + (first_seg->ol_flags & EXTERNAL_MBUF);
>
> if (likely(pkt_flags & PKT_RX_RSS_HASH))
> first_seg->hash.rss = rxd.wb.lower.hi_dword.rss;
> diff --git a/lib/librte_pmd_ixgbe/ixgbe_rxtx_vec.c b/lib/librte_pmd_ixgbe/ixgbe_rxtx_vec.c
> index e813e43..af7b1cd 100644
> --- a/lib/librte_pmd_ixgbe/ixgbe_rxtx_vec.c
> +++ b/lib/librte_pmd_ixgbe/ixgbe_rxtx_vec.c
> @@ -156,10 +156,14 @@ desc_to_olflags_v(__m128i descs[4], struct rte_mbuf **rx_pkts)
> ptype1 = _mm_or_si128(ptype1, vtag1);
> vol.dword = _mm_cvtsi128_si64(ptype1) & OLFLAGS_MASK_V;
>
> - rx_pkts[0]->ol_flags = vol.e[0];
> - rx_pkts[1]->ol_flags = vol.e[1];
> - rx_pkts[2]->ol_flags = vol.e[2];
> - rx_pkts[3]->ol_flags = vol.e[3];
> + rx_pkts[0]->ol_flags = vol.e[0] |
> + (rx_pkts[0]->ol_flags & EXTERNAL_MBUF);
> + rx_pkts[1]->ol_flags = vol.e[1] |
> + (rx_pkts[1]->ol_flags & EXTERNAL_MBUF);
> + rx_pkts[2]->ol_flags = vol.e[2] |
> + (rx_pkts[2]->ol_flags & EXTERNAL_MBUF);
> + rx_pkts[3]->ol_flags = vol.e[3] |
> + (rx_pkts[3]->ol_flags & EXTERNAL_MBUF);
> }
> #else
> #define desc_to_olflags_v(desc, rx_pkts) do {} while (0)
> --
> 1.8.4.2
next prev parent reply other threads:[~2014-10-24 10:37 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-10-24 8:10 [dpdk-dev] [PATCH 0/3] Vhost app removes dependency of REFCNT Ouyang Changchun
2014-10-24 8:10 ` [dpdk-dev] [PATCH 1/3] mbuf: Use EXTERNAL_MBUF to indicate external buffer Ouyang Changchun
2014-10-24 8:10 ` [dpdk-dev] [PATCH 2/3] pmd: RX function need keep EXTERNAL_MBUF flag Ouyang Changchun
2014-10-24 10:46 ` Ananyev, Konstantin [this message]
2014-10-24 12:34 ` Bruce Richardson
2014-10-24 15:43 ` Bruce Richardson
2014-10-24 15:58 ` Ananyev, Konstantin
2014-10-25 2:08 ` Ouyang, Changchun
2014-10-24 8:10 ` [dpdk-dev] [PATCH 3/3] vhost: Removes dependency on REFCNT for zero copy Ouyang Changchun
2014-10-24 9:47 ` [dpdk-dev] [PATCH 0/3] Vhost app removes dependency of REFCNT Thomas Monjalon
2014-10-24 10:47 ` Bruce Richardson
2014-10-25 1:01 ` Ouyang, Changchun
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=2601191342CEEE43887BDE71AB9772582139EBAF@IRSMSX105.ger.corp.intel.com \
--to=konstantin.ananyev@intel.com \
--cc=changchun.ouyang@intel.com \
--cc=dev@dpdk.org \
/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;
as well as URLs for NNTP newsgroup(s).