Re: [RFC 1/1] swiotlb: Reduce calls to swiotlb_find_pool()

[Petr Tesařík <petr@xxxxxxxxxxx>]

On Thu, 27 Jun 2024 08:02:51 +0200
"hch@xxxxxx" <hch@xxxxxx> wrote:

> On Wed, Jun 26, 2024 at 11:58:13PM +0000, Michael Kelley wrote:
> > > This patch trades off making many of the core swiotlb APIs take
> > > an additional argument in order to avoid duplicating calls to
> > > swiotlb_find_pool(). The current code seems rather wasteful in
> > > making 6 calls per round-trip, but I'm happy to accept others'
> > > judgment as to whether getting rid of the waste is worth the
> > > additional code complexity.  
> > 
> > Quick ping on this RFC.  Is there any interest in moving forward?
> > Quite a few lines of code are affected because of adding the
> > additional "pool" argument to several functions, but the change
> > is conceptually pretty simple.  
> 
> Yes, this looks sensible to me.  I'm tempted to apply it.

Oh, right. The idea is good, but I was not able to reply immediately
and then forgot about it.

For the record, I considered an alternative: Call swiotlb_* functions
unconditionally and bail out early if the pool is NULL. But it's no
good, because is_swiotlb_buffer() can be inlined, so this approach
would replace a quick check with a function call. And then there's also
swiotlb_tbl_unmap_single()...

I have only a very minor suggestion: Could is_swiotlb_buffer() be
renamed now that it no longer returns a bool? OTOH I have no good
immediate idea myself.

Petr T