On 29/10/25 4:04 am, Ackerley Tng wrote:
> Hi!
>
> I'm trying to use multi-index xarrays and I was experimenting with
> test_xarray.c.
>
> I'm trying to use xa_erase() on every index after splitting the entry in the
> xarray. (and I commented out every other test case just to focus on this test)
>
> Should erasing every index within the xarray after splitting be a supported use
> case?
>
> Here's the diff:
>
>    diff --git i/lib/test_xarray.c w/lib/test_xarray.c
>    index 5ca0aefee9aa5..fe74f44bbbd92 100644
>    --- i/lib/test_xarray.c
>    +++ w/lib/test_xarray.c
>    @@ -1868,6 +1868,9 @@ static void check_split_1(struct xarray *xa, unsigned long index,
>     	rcu_read_unlock();
>     	XA_BUG_ON(xa, found != 1 << (order - new_order));
>
>    +	for (i = 0; i < (1 << order); i++)
>    +		xa_erase(xa, index + i);
>    +
>     	xa_destroy(xa);
>    }
>
> And made a call to
>
>    check_split_1(xa, 0, 3, 2);
>
> Here's the assertion I hit:
>
>    node 0x7c4de89e01c0x offset 0 parent 0x7c4de89e0100x shift 0 count 4 values 254 array 0x55edd2dd8940x list 0x7c4de89e01d8x 0x7c4de89e01d8x marks 0 10 0
>    xarray: ../shared/../../../lib/xarray.c:764: update_node: Assertion `!(1)' failed.

I changed that function to the following:

staticvoidcheck_split_1(structxarray *xa, unsignedlongindex,

unsignedintorder, unsignedintnew_order)

{

XA_STATE_ORDER(xas, xa, index, new_order);

unsignedinti;

for(i = 0; i < (1<< order); i += (1<< new_order))

xa_store_order(xa, i, new_order, xa, GFP_KERNEL);

xas_lock(&xas);

for(i = 0; i < (1<< order); i += (1<< new_order))

__xa_store(xa, index + i, xa_mk_index(index + i), 0);

xas_unlock(&xas);

xa_erase(xa, index);

xa_destroy(xa);

} and I still hit the assertion. I think the new function is still a valid code sequence -
Store multi-indices of new_order, for length = 1 << order.  Then, replace those entries
with xa_mk_index(index + i). Then erase the first index.

>
>
> I think I've narrowed down the issue to the for (;;) loop in xas_store(), which
> I believe isn't counting the `values` to be updated in update_node() correctly.
>
> Is `values += !xa_is_value(first) - !value;` intended to compute the increase in
> number of values with replacement of every slot being iterated by the new entry?
>
> Why does the computation of `count` involve next and entry, and why does the
> computation for `values` only statically depend on the initial value of entry,
> and on first instead of next?

Yeah there is some issue here only - but the code seems to work, tried a lot of
different combinations in my head like throwing out a non-value and putting in
a value, then vice-versa, then putting a NULL entry - can't figure out the problem :(

>
> Thanks,
>
>
> Ackerley
>