[PATCH] Add a "no-split-stack" function attribute.

[Luqman Aden]

On Sun, Oct 13, 2013 at 1:05 PM, Luqman Aden <me+llvm at luqman.ca> wrote:
> On Sun, Oct 13, 2013 at 7:13 AM, Joerg Sonnenberger
> <joerg at britannica.bec.de> wrote:
>> On Sat, Oct 12, 2013 at 12:23:31PM -0700, Luqman Aden wrote:
>>> On Sat, Oct 12, 2013 at 5:26 AM, Joerg Sonnenberger
>>> <joerg at britannica.bec.de> wrote:
>>> > On Fri, Oct 11, 2013 at 06:17:35PM -0700, Luqman Aden wrote:
>>> >> This disable emission of segmented-stack prologues on a per-function
>>> >> basis, even if segmented stacks are turned on globally.
>>> >
>>> > This sounds quite wrong without any further comment. What ensurse that
>>> > the function will find enough stack space for example?
>>> >
>>> > Joerg
>>>
>>> It's not supposed to guarantee any function tagged with it will have
>>> enough stack space. In terms of use case (from #rust-internals on
>>> Mozilla):
>>>
>>> acrichto> I've been using it a lot where we're managing the stack
>>> limits ourselves
>>> acrichto> so during a context switch I need no_split_stack
>>> acrichto> also the starting function for a thread
>>> acrichto> basically the edge cases around setting up the state of the
>>> world for functions with __morestack
>>
>> I would expect either case to be implemented in assembler for natural
>> reasons, so I don't really like this attribute in the current form.
>>
>
> We do actually currently have a branch wherein we do a lot of that with inline
> assembly from rust and in that case we wouldn't want the _morestack stuff
> to be emitted in the prologue.
>

Also, gcc supports a no_split_stack attribute in its split stack implementation.

_______________________________________________
llvm-commits mailing list
llvm-commits at cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits