Re: [PATCH] block: respect virtual boundary mask in bvecs

[Bart Van Assche <bvanassche@acm.org>]

On Tue, 2018-11-06 at 19:30 -0800, Sagi Grimberg wrote:
> > > How about changing that expression into the following to make it easier
> > > for the compiler to optimize this code?
> > > 
> > > (offset | (bprv->bv_offset + bprv->bv_len)) & queue_virt_boundary(q)
> > 
> > Uhm I have to admit I'm not really able to parse the above expression.
> > Sure GCC will do it but I think it's less readable (at least for me).
> > Let's see what other's think.
> 
> I personally not a huge fan of decoding complicated expressions. But if
> others are fine with it then I am too...

What I proposed is not a new pattern. It is a pattern that is already used
elsewhere in the Linux kernel. A few examples:

From dmabounce.c:

		/* Figure out if we need to bounce from the DMA mask. */
		if ((dma_addr | (dma_addr + size - 1)) & ~mask)
			return 1;

From dma-direct.h:

	if ((addr | (addr + size - 1)) & ~mask)
		return 0;

Bart.