On Mon, Jan 16, 2023 at 07:01:33PM +0000, Matthew Wilcox wrote:
> On Fri, Jan 13, 2023 at 11:19:17AM +0800, Baoquan He wrote:
> > + spin_lock(&vb->lock);
> > + if (bitmap_empty(vb->used_map, VMAP_BBMAP_BITS)) {
> > + spin_unlock(&vb->lock);
> > + memset(buf, 0, count);
> > + return;
> > + }
> > + for_each_set_bitrange(rs, re, vb->used_map, VMAP_BBMAP_BITS) {
> > + if (!count)
> > + break;
> > + start = vmap_block_vaddr(vb->va->va_start, rs);
> > + while (addr < start) {
> > + if (count == 0)
> > + break;
> > + *buf = '\0';
> > + buf++;
> > + addr++;
> > + count--;
> > + }
> > + /*it could start reading from the middle of used region*/
> > + offset = offset_in_page(addr);
> > + n = ((re - rs + 1) << PAGE_SHIFT) - offset;
> > + if (n > count)
> > + n = count;
> > + aligned_vread(buf, start+offset, n);
>
> The whole vread() interface is rather suboptimal. The only user is proc,
> which is trying to copy to userspace. But the vread() interface copies
> to a kernel address, so kcore has to copy to a bounce buffer. That makes
> this spinlock work, but the price is that we can't copy to a user address
> in the future. Ideally, read_kcore() would be kcore_read_iter() and
> we'd pass an iov_iter into vread(). vread() would then need to use a
> mutex rather than a spinlock.
>
> I don't think this needs to be done now, but if someone's looking for
> a project ...
Interesting! I may take a look at this if I get the time :)
