[Lldb-commits] [PATCH] D24629: Allow for tests to be disabled at runtime

[Pavel Labath via lldb-commits]

labath added a comment.

In https://reviews.llvm.org/D24629#550841, @fjricci wrote:

> In https://reviews.llvm.org/D24629#550823, @tfiala wrote:
>
> > > > There is no reasonable thing we can base the expectation as the exact same device with a different cpu revision could support watchpoints just fine, so we could just define the list of these tests externally (in this case, I would probably annotate them with the watchpoint category and then do the skips based on categories instead).
> >
> > > 
> >
> >
> > Tangential: most chips I've worked on that had hardware watchpoint support had an instruction that could be called to find out if such a feature exists.  I think ARM does this.  I would think we could expose an API that says whether watchpoints are supported or not, and use that info in LLDB and the test suite to enable or disable them.
>
>
> I believe that PTRACE_GETHBPREGS with a value of 0 returns the hardware stoppoint info on arm, and the byte representing the number of available hardware watchpoints will be 0 if they aren't supported. Not sure if there's a simpler way.


It's a bit trickier than that. In some cases that call will still return non-zero as the number of supported watchpoints, but the "watchpoint size" field will be zero, and it will still mean that watchpoints don't work. This is probably a kernel bug, though it is pretty easy to work around. The more boring part would be plumbing that information all the way to the test suite - Nothing that can't be done, it's just a bit laborious, so I haven't done that yet.


Repository:
  rL LLVM

https://reviews.llvm.org/D24629