Re: [linux-usb-devel] Re: [PATCH 00/10] usb-serial: Switches from spin lock to atomic_t.

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

On Wed, 7 Dec 2005 14:01:25 +0100
Oliver Neukum <oliver@neukum.org> wrote:

| Am Mittwoch, 7. Dezember 2005 13:25 schrieb Luiz Fernando Capitulino:
| > On Tue, 6 Dec 2005 23:36:47 +0100
| > Oliver Neukum <oliver@neukum.org> wrote:
| > 
| > | Am Dienstag, 6. Dezember 2005 22:18 schrieb Luiz Fernando Capitulino:
| > | > 
| > | >  Hi Pete,
| > | > 
| > | > On Tue, 6 Dec 2005 13:02:07 -0800
| > | > Pete Zaitcev <zaitcev@redhat.com> wrote:
| > | > 
| > | > | On Tue, 6 Dec 2005 18:14:49 -0200, Luiz Fernando Capitulino <lcapitulino@mandriva.com.br> wrote:
| > | > | 
| > | > | >  The spinlock makes the code less clear, error prone, and we already a
| > | > | > semaphore in the struct usb_serial_port.
| > | > | > 
| > | > | >  The spinlocks _seems_ useless to me.
| > | > | 
| > | > | Dude, semaphores are not compatible with interrupts. Surely you
| > | > | understand that?
| > | > 
| > | >  Sure thing man, take a look at this thread:
| > | > 
| > | > http://marc.theaimsgroup.com/?l=linux-kernel&m=113216151918308&w=2
| > | > 
| > | >  My comment 'we already have a semaphore in struct usb_serial_port'
| > | > was about what we've discussed in that thread, where question like
| > | > 'why should we have yet another lock here?' have been made.
| > | > 
| > | >  And *not* 'let's use the semaphore instead'.
| > | > 
| > | >  If _speed_ does not make difference, the spinlock seems useless,
| > | > because we could use atomic_t instead.
| > | 
| > | You can atomically set _one_ value using atomic_t. A spinlock allows
| > | that and other more complex schemes.
| > 
| >  We only need to set 'write_urb_busy', nothing more.
| 
| So go hence and encapsulate that using the existent infrastructure. Thus
| you get the most efficient solution.

 Yes, I was speaking about it with Eduardo some minutes ago.

 My only question is: currently the spin lock is not acquired for unlock
operations (ie, setting 'write_urb_busy' to 0), and to check
'write_usb_busy' value. I don't know if it's safe.

  But, If I add the spin_lock()/spin_unlock() functions in my 'unlock'
and 'locked' methods, I could increase the latency for SMP systems.
 
 Suggestions? Eduardo? Greg?

-- 
Luiz Fernando N. Capitulino
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/