On Mon, Nov 9, 2020 at 6:29 PM Thomas Monjalon <tho...@monjalon.net> wrote:
>
> 09/11/2020 13:01, Jerin Jacob:
> > Hi @Thomas Monjalon
> >
> > Any specific reason why you removed the static assert from octeontx2.
>
> I have a build failure when cross-compiling for octeontx2.
I am trying the below command, I am not able to see any issue
meson build --cross-file config/arm/arm64_octeontx2_linux_gcc
Are you facing the issue with 32bit? Could you share the steps to
reproduce and gcc version?
>
> > I am not able to compilation issue with that static assert.
>
> There is no issue when compiling for x86.
>
> > The current vector driver assumes pool and tx offload needs to 2 DWORDS
> > apart,
> > Which is the case before and after your change.
>
> You're right, pool and tx_offload are moved together.
> The only difference is passing the cache line frontier.
>
> > Please remove that static assert change, No issue from my side on this
> > patch.
>
> I cannot remove it without fixing something else,
> maybe an issue in cache line alignment?
Can try the below fix. If the issue is seen with 32bit.
diff --git a/drivers/net/octeontx2/otx2_ethdev.c
b/drivers/net/octeontx2/otx2_ethdev.c
index 6cebbe677..66a4d429d 100644
--- a/drivers/net/octeontx2/otx2_ethdev.c
+++ b/drivers/net/octeontx2/otx2_ethdev.c
@@ -749,7 +749,7 @@ nix_tx_offload_flags(struct rte_eth_dev *eth_dev)
RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, pkt_len) !=
offsetof(struct rte_mbuf, ol_flags) + 12);
RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, tx_offload) !=
- offsetof(struct rte_mbuf, pool) + 2 * sizeof(void *));
+ offsetof(struct rte_mbuf, pool) + 2 *
sizeof(uint64_t));
>
> > In general, it is too much effort to re-verify and measure performance
> > impact with
> > all the cases after RC2, I hope this will last mbuf change in this release.
>
> Yes it is the last change to mbuf layout,
> sorry for pushing all these changes so late.
>
>
>
