[PATCH] Link state change with BCM5709

[Atte Peltomäki-3]

Hi,

Link state change interrupt was not generated due to a missing bit in
the MAC event register. This fixes at least carp and ifstated for 5709
chip (eg. in Dell R610).

Note that Broadcom Linux driver also sets STATUS_ATTN_BITS_TIMER_ABORT
on the same go which I is detection for a hung pci bus. Code for this
interrupt is missing in the bnx_phy_intr() handler, so I left it out
for now. If you have any ideas how to test it, I can try porting the
handler code.


--- sys/dev/pci/if_bnx.c   Fri Oct 30 21:04:04 2009
+++ sys/dev/pci/if_bnx.c        Fri Oct 30 21:05:56 2009
@@ -3461,6 +3461,7 @@

        /* Set up link change interrupt generation. */
        REG_WR(sc, BNX_EMAC_ATTENTION_ENA, BNX_EMAC_ATTENTION_ENA_LINK);
+       REG_WR(sc, BNX_HC_ATTN_BITS_ENABLE, STATUS_ATTN_BITS_LINK_STATE);

        /* Program the physical address of the status block. */
        REG_WR(sc, BNX_HC_STATUS_ADDR_L, (u_int32_t)(sc->status_block_paddr));

--
Atte Peltomdki
     atte.peltomaki@... <> http://kameli.org
"Your effort to remain what you are is what limits you"

[David Gwynne]

On Fri, Oct 30, 2009 at 09:50:34PM +0200, Atte Peltom?ki wrote:
> Hi,
>
> Link state change interrupt was not generated due to a missing bit in
> the MAC event register. This fixes at least carp and ifstated for 5709
> chip (eg. in Dell R610).

i have verified this on the 5709. i dont have a box with older
versions of bnx in it to test on them though. have you been able
to try this on a 2950 or r200?

> Note that Broadcom Linux driver also sets STATUS_ATTN_BITS_TIMER_ABORT
> on the same go which I is detection for a hung pci bus. Code for this
> interrupt is missing in the bnx_phy_intr() handler, so I left it out
> for now. If you have any ideas how to test it, I can try porting the
> handler code.
>
>
> --- sys/dev/pci/if_bnx.c   Fri Oct 30 21:04:04 2009
> +++ sys/dev/pci/if_bnx.c        Fri Oct 30 21:05:56 2009
> @@ -3461,6 +3461,7 @@
>
>         /* Set up link change interrupt generation. */
>         REG_WR(sc, BNX_EMAC_ATTENTION_ENA, BNX_EMAC_ATTENTION_ENA_LINK);
> +       REG_WR(sc, BNX_HC_ATTN_BITS_ENABLE, STATUS_ATTN_BITS_LINK_STATE);
>
>         /* Program the physical address of the status block. */
>         REG_WR(sc, BNX_HC_STATUS_ADDR_L, (u_int32_t)(sc->status_block_paddr));
>
> --
> Atte Peltomdki
>      atte.peltomaki@... <> http://kameli.org
> "Your effort to remain what you are is what limits you"

[Atte Peltomäki-3]

On Mon, Nov 09, 2009 at 10:12:48PM +1000, David Gwynne wrote:
> On Fri, Oct 30, 2009 at 09:50:34PM +0200, Atte Peltom?ki wrote:
> > Link state change interrupt was not generated due to a missing bit in
> > the MAC event register. This fixes at least carp and ifstated for 5709
> > chip (eg. in Dell R610).
>
> i have verified this on the 5709. i dont have a box with older
> versions of bnx in it to test on them though. have you been able
> to try this on a 2950 or r200?

Unfortunately not, these R610 are only OpenBSD boxes we have so far.
However, judging from both the Broadcom tech docs for opensource
developers and their own linux driver source, this patch is needed for
all chips using the bnx(4) driver.

Since carp and ifstated use routing socket to determine link status, it
is possible that they work without this patch if something else updates
the socket. For example with 5709, using ifconfig that talks via ioctl
interface in the driver that polls link status from MII. On the same go
the routing socket status is updated and this is when carp and ifstated
notice it, too.

--
Atte Peltomdki
     atte.peltomaki@... <> http://kameli.org
"Your effort to remain what you are is what limits you"

[Stuart Henderson]

On 2009/11/09 15:14, Atte Peltomdki wrote:
> On Mon, Nov 09, 2009 at 10:12:48PM +1000, David Gwynne wrote:
> > On Fri, Oct 30, 2009 at 09:50:34PM +0200, Atte Peltom?ki wrote:
> > > Link state change interrupt was not generated due to a missing bit in
> > > the MAC event register. This fixes at least carp and ifstated for 5709
> > > chip (eg. in Dell R610).

it quite possibly hits trunk(4) too.

> > i have verified this on the 5709. i dont have a box with older
> > versions of bnx in it to test on them though. have you been able
> > to try this on a 2950 or r200?

R200 has bge(4). Dell 1950 and HP DL360 G5 have BCM5708.