Hi Finn,
thanks - tests (including __clear_user) with my 030 did show one Oops in
the same place that __generic_copy_to_user did last fault, that is, the
btst after the movesl instruction. Adding that instruction to the
exception table avoids the Oops.
A range of path lengths as well as offsets tested. Still some tests
pending... Couldn't find any evidence that a NOP is needed either.
I'll post the final version as soon as I've completed testing.
Cheers,
Michael
Am 27.04.2024 um 13:44 schrieb Finn Thain:
On Fri, 26 Apr 2024, I wrote:
I will test it tomorrow. I expect that a NOP is needed at the end of
__clear_user.
Everything seems to work now.
I tested this series together with my own RFC patch on a Quadra (68040)
and IIci (68030). My tests covered getcwd and llseek syscalls. I tried a
build with the extra call to __clear_user() and a build without it. And I
covered a range of working directory names.
Tested-by: Finn Thain <fthain@xxxxxxxxxxxxxx>
So apparently there'e no need for a NOP at the end of __clear_user() -- on
'040 and '030 at least.