[Prism54-devel] handle mgmt timeouts by resetting hardware

[Denis Vlasenko]

On Wednesday 04 August 2004 11:34, Margit Schubert-While wrote:
> Could you change your coding style ?
> eg :
> if(rate_enum<0) return rate_enum;
>
> Please a space after the the "if" and put the command(s) on a seperate
> line.
> Also, I don't agree with another source file (setrate.c).

There is a comment on top on it which says that
this code is not prism54-specific at all.
It can be used with other 802.11 drivers
and is, in fact, already used by acx100 driver.

The plan is to move it into wireless.c at a later stage.

> If it has to be done this way, then setrate.h :-)
> However, why not roll it into one of the *mgt.h files ?
>
> Actually, there is something else wrong.
> I'll try to be brief.
> By accident, I discovered I could trigger the "NETDEV WATCHDOG"
> on one of my lappies due to :
> a) a slightly misconfigured kernel   :-)
> and
> b) Prism54 sharing an IRQ with the X vga driver (and usb)
>
> On a restart of X, I got :
> --------------------------------
> irq 11: nobody cared!
> Disabling IRQ #11
> --------------------------------
> and eventually :
> -------------------------------
> NETDEV WATCHDOG: eth1: transmit timed out
> -------------------------------
>
> The watchdog kicks and prism54 tries a reset
> (which, of course, will never succeed)
> (Q, How can we pick up the fact that the IRQ is disabled?)

We simply must be prepared to the fact that
'reset done' IRQ might never happen,
as part of usual "don't trust hardware"
attitude.

> Fine, I thought, nice way to check paths that don't usually get run.
> I put in Denis's fixes and after the "Disabling IRQ" message,
> the box was hard locked.

Can you try splitted patches? I'll send them shortly

> Not sure what is going on here, as the circumstances are not completely
> valid. (The kernel thinks it's operating with AGP, the vga driver doesn't
> think it can operate
>   with AGP and has dropped into PCI mode).
> However, interesting.
--
vda