On Fri, Aug 27, 2021 at 4:53 PM John Garry [off-list ref] wrote:

On 18/08/2021 13:06, Xie Yongji wrote:

quoted

+
+static dma_addr_t
+vduse_domain_alloc_iova(struct iova_domain *iovad,
+                     unsigned long size, unsigned long limit)
+{
+     unsigned long shift = iova_shift(iovad);
+     unsigned long iova_len = iova_align(iovad, size) >> shift;
+     unsigned long iova_pfn;
+
+     /*
+      * Freeing non-power-of-two-sized allocations back into the IOVA caches
+      * will come back to bite us badly, so we have to waste a bit of space
+      * rounding up anything cacheable to make sure that can't happen. The
+      * order of the unadjusted size will still match upon freeing.
+      */
+     if (iova_len < (1 << (IOVA_RANGE_CACHE_MAX_SIZE - 1)))
+             iova_len = roundup_pow_of_two(iova_len);

Whether it's proper to use this "fast" API or not here, this seems to be
copied verbatim from dma-iommu.c, which tells me that something should
be factored out.

Agreed.

Indeed, this rounding up seems a requirement of the rcache, so not sure
why this is not done there.

Me too. I guess it is to let users know that space is wasted.

Thanks,
Yongji