On 29/01/2026 18:04, Michael Kelly wrote:
On 29/01/2026 17:14, Mike Kelly wrote:
Here is a patch for review to address the xmm register value
corruption occurring in user code that is heavily bombarded by signals.
It's a small patch to address what I found to be a very difficult
problem. The xstate_size of the interrupted thread's xstate is not
saved in any structure. I therefore had to implement the assumption
that xstate_size returns the same value for all calls to
__i386_get_xstate_size() which is indeed currently the case. There is
no actual issue here but it seemed wrong to me to do so. Comments
welcome.
Also note that this patch has been tested successfully on hurd-i386
and hurd-amd64 using the test case 'sigtest_xmm0.c' attached to
message https://lists.gnu.org/archive/html/bug-hurd/2026-01/msg00241.html
There is a limitation with this patch in the case where the xstate is
not supported by the system and managed state should default to the
basic sc_i386_float_state. I will rework the patch accordingly and also
attempt to address the issue of the xstate_size mentioned above.
Regards,
Mike.
