usb/107642: [patch]Ralink Technology RT2501USB/RT2601USB chipset driver

[Valery V.Chikalov]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Kevin Lo пишет:
> Valery V.Chikalov wrote:
>> Hans Petter Selasky wrote:
>>> On Sunday 29 April 2007 15:02, Valery V.Chikalov wrote:
>>>> Kevin Lo wrote:
>>>>> Valery V.Chikalov wrote:
>>>>>> The following reply was made to PR usb/107642; it has been noted by
>>>>>> GNATS.
>>>>>>
>>>>>> From: "Valery V.Chikalov" <valera at chikalov.dp.ua>
>>>>>> To: bug-followup at FreeBSD.org,  valera at chikalov.dp.ua
>>>>>> Cc:
>>>>>> Subject: Re: usb/107642: [patch]Ralink Technology RT2501USB/RT2601USB
>>>>>> chipset driver
>>>>>> Date: Sun, 22 Apr 2007 11:32:18 +0300
>>>>>>
>>>>>>  This is a multi-part message in MIME format.
>>>>>>  --------------030900090303000507070905
>>>>>>  Content-Type: text/plain; charset=UTF-8
>>>>>>  Content-Transfer-Encoding: 7bit
>>>>>>
> if_rum(4) for 7.0-CURRENT
> 
> replaced amrr_* functions by "standard" ones already existed in
> net80211/ieee80211_amrr.c
> 
>>>>> Hi Valery,
>>>>>
>>>>> I guess you wasn't aware that I've already ported rum(4) to FreeBSD.
>>>>> The patch is available at: http://people.freebsd.org/~kevlo/patch-rum
>>>>> Maybe you can test my patch? Thanks.
>>>>>
>>>>> 	Kevin
>>>> Hi, Kevin,
>>>>
>>>> Your driver not working for me. Fortunately, the errors that I see
>>>> exactly the same which i fight when I made my driver.
>>>>
>>>> $ uname -a
>>>> FreeBSD tiger.novakom.dp.ua 7.0-CURRENT FreeBSD 7.0-CURRENT #6: Sun Apr
>>>> 29 13:58:48 EEST 2007
>>>> root at tiger.novakom.dp.ua:/usr/obj/usr/src/sys/TIGER64  amd64
>>>>
>>>> $ sysctl kern.osreldate
>>>> kern.osreldate: 700037
>>>>
>>>> cvsup'ed 29.04.2007
>>>>
>>>> kernel with:
>>>>
>>>> makeoptions     DEBUG=-g
>>>>
>>>> options         KDB
>>>>
>>>> options         DDB
>>>>
>>>>
>>>>
>>>> options         INVARIANTS
>>>>
>>>> options         INVARIANT_SUPPORT
>>>>
>>>> options         WITNESS
>>>>
>>>> At first, when I make kldload if_rum, I get kernel panic.
>>>> But I cant get saved core, as ddb just hangs during "call doadump"
>>>>
>>> I have a solution for all of this locking stuff!
>>>
>>>> So I add
>>>>
>>>> #define RUM_LOCK(sc)    do { ((sc) = (sc)); mtx_lock(&Giant); } while
>>>> (0)
>>>> #define RUM_UNLOCK(sc)  mtx_unlock(&Giant)
>>>>
>>>> in  if_rumvar.h
>>>>
>>>> I spend a lot of time in attempts get rid of Giant ant always got only
>>>> panics.
>>> You _cannot_ do that with the old USB stack, because you must lock Giant 
>>> before calling into the usbxxx functions. Then in the USB callback, Giant is 
>>> locked, and then you cannot lock RUM_LOCK()! That means you will most likely 
>>> end up with a deadlock pretty soon, if you see that.
>>>
>> Thanks, for explanations. I suspected that thing are like that, and I 
>> have tried make porting by analogue with other drivers which I can find 
>> in dev/usb, but I was not can find the description of doing "right way" 
>> locking before.

> Firstly, thanks for taking the time to test my patch.
> I think your issue is related to watchdog timers. The revised patch is
> at http://people.freebsd.org/~kevlo/patch-rum

> Besides, I don't get a panic during high load when getting a file about
> 400MB via ftp.

Sorry, but the error that I see the same, just after inserting dongle,
or if it was inserted before - after kldload if_rum:

panic: sleeping without a lock

the top of stack:
at usbd_transfer +0x1fe
....
rum_ioctl

And I cant get saved kernel coredump at the end.

I have not serial console configured so it was just subscribed by hand.
Of course I can reproduce it again and subscribe what you will ask.

I have SMP kernel and use AMD64 architecture, maybe this are the reasons
why we are getting different results.

Can you please, explain me in two words(or give a reference) why you are
dropping mtx_lock(&Giant). Yes I have seen the same in if_ural.c and as
i have wrote before unsuccessfully have tried repeat it in if_rum. I
know that current USB stack is not Giant free, and Hans noted that this
is impossible (if I understand him right). Maybe exists some tunable or
kernel option or another trick that allows drivers working without the
Giant under current USB stack, which I must use.

Thank you for your time and patience.
Sorry for my english.

Valery.

>>>> After that I get hangs,
>>>> which i resolved by modifying rum_ioctl:
>>> I'm almost finished converting "if_rum.c()" to the new USB stack.
>>>
>>> In some hours I will update it with support for "if_rum".
>>>
>>> If you can test that and forget about the old USB stack, I will be very 
>>> happy :-)
>>>
>> I will do it with pleasure. I was almost ready to do it (converting to 
>> new USB stack) by myself, but I was stopped by the fact that I cant make 
>> it compiled under CURRENT. I have seen your mail that your are working 
>> on this.  Is new the USB stack ready for CURRENT now?
>>
>> Valery.
>>
>>> --HPS
>>>
>>> http://www.turbocat.net/~hselasky/usb4bsd

> 	Kevin


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFGNeJv760S3kDvEC4RAgrVAJ4m/iXooa1Yt2txMVz0LOwmC+vqAwCgicKb
xPMZsEyzAodOccQSjq6KL0A=
=JVix
-----END PGP SIGNATURE-----