sleeping in dev->tx_timeout?

sleeping in dev->tx_timeout?

[Abraham van der Merwe]

Hi!

Is it safe to sleep in tx_timeout (in the networking code), i.e. to call
schedule_timeout and friends from that routine?

-- 

Regards
 Abraham

: How would you disambiguate these situations?

By shooting the person who did the latter.
             -- Larry Wall in <199710290235.SAA02444@wall.org>

___________________________________________________
 Abraham vd Merwe - Frogfoot Networks CC
 9 Kinnaird Court, 33 Main Street, Newlands, 7700
 Phone: +27 21 686 1665 Cell: +27 82 565 4451
 Http: http://www.frogfoot.net/ Email: abz@frogfoot.net

Re: sleeping in dev->tx_timeout?

[Zwane Mwaikambo]

On Sun, 3 Aug 2003, Abraham van der Merwe wrote:

> Hi!
> 
> Is it safe to sleep in tx_timeout (in the networking code), i.e. to call
> schedule_timeout and friends from that routine?

No it's called from softirq context and with the dev->xmit_lock held in 
places.

-- 
function.linuxpower.ca

Re: sleeping in dev->tx_timeout?

[Abraham van der Merwe]

Hi Zwane                                         >@2003.08.03_21:10:56_+0200

> > Is it safe to sleep in tx_timeout (in the networking code), i.e. to call
> > schedule_timeout and friends from that routine?
> 
> No it's called from softirq context and with the dev->xmit_lock held in 
> places.

That's what I thought. How are you supposed to wait for long periods from
tx_timeout? My problem is that I have a chip reset which takes around 10ms
(i.e. too long to use mdelay())

-- 

Regards
 Abraham

No problem is insoluble in all conceivable circumstances.

___________________________________________________
 Abraham vd Merwe - Frogfoot Networks CC
 9 Kinnaird Court, 33 Main Street, Newlands, 7700
 Phone: +27 21 686 1665 Cell: +27 82 565 4451
 Http: http://www.frogfoot.net/ Email: abz@frogfoot.net

Re: sleeping in dev->tx_timeout?

[Jeff Garzik]

Abraham van der Merwe wrote:
> Hi Zwane                                         >@2003.08.03_21:10:56_+0200
> 
> 
>>>Is it safe to sleep in tx_timeout (in the networking code), i.e. to call
>>>schedule_timeout and friends from that routine?
>>
>>No it's called from softirq context and with the dev->xmit_lock held in 
>>places.
> 
> 
> That's what I thought. How are you supposed to wait for long periods from
> tx_timeout? My problem is that I have a chip reset which takes around 10ms
> (i.e. too long to use mdelay())

Simple answer, don't wait in tx_timeout :)

These days drivers often need quite a while for hardware reset.  I am 
pushing to move this code, long term, into process context.  So, in 
tx_timeout:
* disable NIC and interrupts as best you can, quickly
* schedule_task/schedule_work to schedule the full hardware reset

	Jeff

Re: sleeping in dev->tx_timeout?

[Alan Cox]

On Sul, 2003-08-03 at 20:48, Jeff Garzik wrote:
> These days drivers often need quite a while for hardware reset.  I am 
> pushing to move this code, long term, into process context.  So, in 
> tx_timeout:
> * disable NIC and interrupts as best you can, quickly
> * schedule_task/schedule_work to schedule the full hardware reset

And if the hardware is hard to recover or needs messy recovery code take
a look at the PCI layer tricks in -ac