Re: [parisc-linux] rmap: parisc __flush_dcache_page

[James Bottomley <James.Bottomley@steeleye.com>]

On Thu, 2004-04-08 at 10:34, Andrea Arcangeli wrote:
> yes, the spinlock in struct address_space would be enough, and that's
> what 2.4 does too, Andrew changed it to a semaphore in 2.6 but it can be
> made a spinlock again. Then you can fix it (as far as you never call it
> from an irq and as far as you don't generate exceptions inside the
> critical section, but I'm sure you don't).

Well, yes, of course we do.  We're a sofware tlb arch, so we generate
exceptions on tlb misses, which can occur anywhere (even in critical
sections).  However, the exceptions are carefully crafted not to take
spinlocks, so everything should be safe.

I'm not sure about the no in irq assertion.  The biggest use of
flush_dcache_page is on the I/O return path ... that looks like a good
candidate for being in interrupt (even though most drivers should be
offloading to softirq/tasklets).

James