[Bug breakpoints/23131] XFAIL watchpoint tracker for powerpc

[Bug breakpoints/23131] XFAIL watchpoint tracker for powerpc

[luis.machado at linaro dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=23131

Luis Machado <luis.machado at linaro dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |luis.machado at linaro dot org

--- Comment #2 from Luis Machado <luis.machado at linaro dot org> ---
Is there more background to this? I see some tests consider the outlier
architectures to be broken (as in we get FAIL's/XFAIL's in the testsuite).

I see many failures in gdb.base/watchpoint-unaligned.exp that are due to read
watchpoints not being supported on a particular target.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug breakpoints/23131] XFAIL watchpoint tracker for powerpc

[jan.kratochvil at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=23131

--- Comment #3 from Jan Kratochvil <jan.kratochvil at redhat dot com> ---
(In reply to Luis Machado from comment #2)
> Is there more background to this?

IIRC it tests there are no false positives (catching accesses outside of the
watched area but still inside an align-extended area - that was probably the
case of PowerPC) and no false negatives (unreported=missed access inside the
watched area). Not sure what more background you mean. I also no longer
remember it much, I would have to read it again myself.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug breakpoints/23131] XFAIL watchpoint tracker for powerpc

[luis.machado at linaro dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=23131

--- Comment #4 from Luis Machado <luis.machado at linaro dot org> ---
(In reply to Jan Kratochvil from comment #3)
> (In reply to Luis Machado from comment #2)
> > Is there more background to this?
> 
> IIRC it tests there are no false positives (catching accesses outside of the
> watched area but still inside an align-extended area - that was probably the
> case of PowerPC) and no false negatives (unreported=missed access inside the
> watched area). Not sure what more background you mean. I also no longer
> remember it much, I would have to read it again myself.

Thanks. That's useful. I have a particular ARM32 target whose kernel advertises
hardware watchpoints but can't use those in practice. Therefore the static
"skip_hw_watchpoint_tests" is not enough, and we need to test support
dynamically right before insertion.

That particular test seems to rely on hardware read watchpoints, so I'm
guessing it would make sense for this particular ARM32 target to skip those
sections completely.

-- 
You are receiving this mail because:
You are on the CC list for the bug.