Re: lmc(4) mutex locking

[Andrew Doran <ad%netbsd.org@localhost>]

On Wed, Jun 25, 2008 at 07:05:11PM -0700, Gregory McGarry wrote:

> To remove gcc extensions from the lmc(4) driver, I converted the driver to 
> mutexes.  Does it look correct?  Can anybody test it?
>  -- Gregory McGarry <g.mcgarry%ieee.org@localhost>
> 
> 
> 
>       Get the name you always wanted with the new y7mail email address.
> www.yahoo7.com.au/mail
> Index: if_lmc.c
> ===================================================================
> RCS file: /cvsroot/src/sys/dev/pci/if_lmc.c,v
> retrieving revision 1.42
> diff -u -r1.42 if_lmc.c
> --- if_lmc.c  12 Jun 2008 22:44:47 -0000      1.42
> +++ if_lmc.c  26 Jun 2008 02:04:08 -0000
> @@ -7167,7 +7167,7 @@
>      }
>  
>    /* Initialize the top-half and bottom-half locks. */
> -  __cpu_simple_lock_init(&sc->top_lock);
> +  mutex_init(&sc->top_lock, MUTEX_DEFAULT, IPL_NET);

IPL_VM

> @@ -1055,10 +1055,9 @@
>  # define WRITE_CSR(sc, csr, val) bus_space_write_4((sc)->csr_tag, 
> (sc)->csr_handle, csr, val)
>  # define NAME_UNIT           device_xname(&sc->dev)
>  # define BOOT_VERBOSE                (boothowto & AB_VERBOSE)
> -# define TOP_LOCK(sc)                ({ while 
> (__cpu_simple_lock_try(&(sc)->top_lock)==0) \
> -                              tsleep((sc), PCATCH|PZERO, DEVICE_NAME, 1); 0; 
> })
> -# define TOP_TRYLOCK(sc)     __cpu_simple_lock_try(&(sc)->top_lock)
> -# define TOP_UNLOCK(sc)              __cpu_simple_unlock  (&(sc)->top_lock)
> +# define TOP_LOCK(sc)                (mutex_enter(&(sc)->top_lock), 0)
> +# define TOP_TRYLOCK(sc)     mutex_tryenter(&(sc)->top_lock)
> +# define TOP_UNLOCK(sc)              mutex_exit(&(sc)->top_lock)
>  # define BOTTOM_TRYLOCK(sc)  __cpu_simple_lock_try(&(sc)->bottom_lock)
>  # define BOTTOM_UNLOCK(sc)   __cpu_simple_unlock  (&(sc)->bottom_lock)
>  # define CHECK_CAP           kauth_authorize_generic(curlwp->l_cred, 
> KAUTH_GENERIC_ISSUSER, NULL)

mutex_spin_enter + mutex_spin_exit

> @@ -1290,7 +1289,11 @@
>    void *irq_cookie;
>    void *sdh_cookie;
>    struct mbuf *tx_mbuf;                      /* hang mbuf here while 
> building dma descs */
> +#if defined(__NetBSD__)
> +  kmutex_t top_lock;                 /* lock card->watchdog vs ioctls        
>    */
> +#else
>    __cpu_simple_lock_t top_lock;              /* lock card->watchdog vs 
> ioctls           */
> +#endif
>    __cpu_simple_lock_t bottom_lock;   /* lock buf queues & descriptor rings   
> */
>  #endif                                       /* __NetBSD__ || __OpenBSD__ */

The #ifdefs are probably not worth it. Otherwise it looks fine to me.

Andrew