Re: [linux-usb-devel] device struct bloat

[David Brownell <david-b@pacbell.net>]

On Monday 05 November 2007, Greg KH wrote:
> --- linux-2.6-2.orig/drivers/spi/spi.c
> +++ linux-2.6-2/drivers/spi/spi.c

It'd be quicker to end up in the right hands if you had
split this big and random patch according to subsystem...

There's already a patch in the MM queue that removes
the SPI-private semaphore.  Except that it's missing
the bug noted below.

The class semaphore removal would be a different issue.


> @@ -613,7 +613,7 @@ int spi_write_then_read(struct spi_devic
>         }
>  
>         /* ... unless someone else is using the pre-allocated buffer */
> -       if (down_trylock(&lock)) {
> +       if (mutex_trylock(&lock)) {

According to its kerneldoc, mutex_trylock() follows the
spinlock model not the semaphore model.  So the sense of
this test is incorrect ... as will be any similar changes
in other parts of this patch:

 * NOTE: this function follows the spin_trylock() convention, so
 * it is negated to the down_trylock() return values! Be careful
 * about this when converting semaphore users to mutexes.

So the patch in the MM queue says "if (!mutex_trylock(...)) {

>                 local_buf = kmalloc(SPI_BUFSIZ, GFP_KERNEL);
>                 if (!local_buf)
>                         return -ENOMEM;