Hi Kemeng,
On Wed, 18 Dec 2024 at 07:58, Kemeng Shi <shikemeng@xxxxxxxxxxxxxxx> wrote:
> After xas_load(), xas->index could point to mid of found multi-index entry
> and xas->index's bits under node->shift maybe non-zero. The afterward
> xas_pause() will move forward xas->index with xa->node->shift with bits
> under node->shift un-masked and thus skip some index unexpectedly.
>
> Consider following case:
> Assume XA_CHUNK_SHIFT is 4.
> xa_store_range(xa, 16, 31, ...)
> xa_store(xa, 32, ...)
> XA_STATE(xas, xa, 17);
> xas_for_each(&xas,...)
> xas_load(&xas)
> /* xas->index = 17, xas->xa_offset = 1, xas->xa_node->xa_shift = 4 */
> xas_pause()
> /* xas->index = 33, xas->xa_offset = 2, xas->xa_node->xa_shift = 4 */
> As we can see, index of 32 is skipped unexpectedly.
>
> Fix this by mask bit under node->xa_shift when move forward index in
> xas_pause().
>
> For now, this will not cause serious problems. Only minor problem
> like cachestat return less number of page status could happen.
>
> Signed-off-by: Kemeng Shi <shikemeng@xxxxxxxxxxxxxxx>
Thanks for your patch, which is now commit c9ba5249ef8b080c ("Xarray:
move forward index correctly in xas_pause()") upstream.
> --- a/lib/test_xarray.c
> +++ b/lib/test_xarray.c
> @@ -1448,6 +1448,41 @@ static noinline void check_pause(struct xarray *xa)
> XA_BUG_ON(xa, count != order_limit);
>
> xa_destroy(xa);
> +
> + index = 0;
> + for (order = XA_CHUNK_SHIFT; order > 0; order--) {
> + XA_BUG_ON(xa, xa_store_order(xa, index, order,
> + xa_mk_index(index), GFP_KERNEL));
> + index += 1UL << order;
> + }
> +
> + index = 0;
> + count = 0;
> + xas_set(&xas, 0);
> + rcu_read_lock();
> + xas_for_each(&xas, entry, ULONG_MAX) {
> + XA_BUG_ON(xa, entry != xa_mk_index(index));
> + index += 1UL << (XA_CHUNK_SHIFT - count);
> + count++;
> + }
> + rcu_read_unlock();
> + XA_BUG_ON(xa, count != XA_CHUNK_SHIFT);
> +
> + index = 0;
> + count = 0;
> + xas_set(&xas, XA_CHUNK_SIZE / 2 + 1);
> + rcu_read_lock();
> + xas_for_each(&xas, entry, ULONG_MAX) {
> + XA_BUG_ON(xa, entry != xa_mk_index(index));
> + index += 1UL << (XA_CHUNK_SHIFT - count);
> + count++;
> + xas_pause(&xas);
> + }
> + rcu_read_unlock();
> + XA_BUG_ON(xa, count != XA_CHUNK_SHIFT);
> +
> + xa_destroy(xa);
> +
> }
On m68k, the last four XA_BUG_ON() checks above are triggered when
running the test. With extra debug prints added:
entry = 00000002 xa_mk_index(index) = 000000c1
entry = 00000002 xa_mk_index(index) = 000000e1
entry = 00000002 xa_mk_index(index) = 000000f1
...
entry = 000000e2 xa_mk_index(index) = fffff0ff
entry = 000000f9 xa_mk_index(index) = fffff8ff
entry = 000000f2 xa_mk_index(index) = fffffcff
count = 63 XA_CHUNK_SHIFT = 6
entry = 00000081 xa_mk_index(index) = 00000001
entry = 00000002 xa_mk_index(index) = 00000081
entry = 00000002 xa_mk_index(index) = 000000c1
...
entry = 000000e2 xa_mk_index(index) = ffffe0ff
entry = 000000f9 xa_mk_index(index) = fffff0ff
entry = 000000f2 xa_mk_index(index) = fffff8ff
count = 62 XA_CHUNK_SHIFT = 6
On arm32, the test succeeds, so it's probably not a 32-vs-64-bit issue.
Perhaps a big-endian or alignment issue (alignof(int/long) = 2)?
> --- a/lib/xarray.c
> +++ b/lib/xarray.c
> @@ -1147,6 +1147,7 @@ void xas_pause(struct xa_state *xas)
> if (!xa_is_sibling(xa_entry(xas->xa, node, offset)))
> break;
> }
> + xas->xa_index &= ~0UL << node->shift;
> xas->xa_index += (offset - xas->xa_offset) << node->shift;
> if (xas->xa_index == 0)
> xas->xa_node = XAS_BOUNDS;
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
