Re: usb-serial ipaq kernel problem

["Luiz Fernando N. Capitulino" <lcapitulino@mandriva.com.br>]

On Tue, 30 May 2006 17:06:27 +0200
Frank Gevaerts <frank.gevaerts@fks.be> wrote:

| On Tue, May 30, 2006 at 11:38:01AM -0300, Luiz Fernando N. Capitulino wrote:
| > On Tue, 30 May 2006 10:21:41 +0200
| > Frank Gevaerts <frank.gevaerts@fks.be> wrote:
| > 
| > | On Mon, May 29, 2006 at 07:33:30PM -0300, Luiz Fernando N. Capitulino wrote:
| > | > On Mon, 29 May 2006 22:47:24 +0200
| > | >  I see.
| > | > 
| > | >  Did you try to just kill the read urb in the ipaq_open's error path?
| > | 
| > | Yes, that's what I did at first. It works, but with the long waits (we see
| > | waits up to 80-90 seconds right now) I was afraid that the urb might timeout
| > | before the control message succeeds.
| > 
| >  Hmmm, I see.
| > 
| >  My only (obvious) question is that if it's really safe to send the read
| > urb after the control message..
| 
| Since it is bulk, it is not guaranteed to start before the control
| message anyway, so it should be safe.
| 
| The patch looks correct to me, but I would still like to increase
| KP_RETRIES a bit. If I read the code correctly, the current setup allows
| for 10 seconds between usb connect and acknowledging the control
| message. This is enough if the device is only connected when booted
| (which is the normal use case). However, in our case, we do
| software-initiated reboots of the ipaq while it is attached to the usb
| bus, which can take much longer, so for us KP_RETRIES should be at least
| 1000, maybe 2000. Of course, we can always run a patched kernel for this.

 Hmmm, what do you think about keeping the current default value and
adding a module parameter to change it?

| I'll test the patch later today.
| 
| Anyway, we have not seen the usb_serial_disconnect bug since applying
| your patch, so that bug is also probably gone (we have had nearly 1000
| successfull connects/disconnects since then)

 Nice to know.

-- 
Luiz Fernando N. Capitulino