On Thu, Nov 07, 2024 at 05:34:45PM +0100, Michal Wajdeczko wrote:
> +size_t copy_iomem_to_iter(const void __iomem *from, size_t offset,
> + size_t bytes, struct iov_iter *i)
> +{
> + unsigned char buf[SMP_CACHE_BYTES];
> + size_t progress = 0, copied, len;
> +
> + from += offset;
> + while (bytes) {
> + len = min(bytes, sizeof(buf));
> + memcpy_fromio(buf, from + progress, len);
> + copied = _copy_to_iter(buf, len, i);
> + if (!copied)
> + break;
> + bytes -= copied;
> + progress += copied;
> + }
This seems like a rather sad implementation. Why not:
if (WARN_ON_ONCE(i->data_source))
return 0;
if (user_backed_iter(i))
might_fault();
return iterate_and_advance(i, bytes, (void *)addr,
copy_iomem_to_iter, memcpy_iomem_to_iter);
along with
size_t memcpy_iomem_to_iter()
{
memcpy_fromio(iter_to, from + progress, len);
return 0;
}`
