Re: MT & SINGELMAP_MEMORY model

[Vladimir T. <vtz...@gm...>]

I already tried disabling only SINGLEMAP_MEMORY_STACK and got immediate
segfault (single threaded build as well) on startup in init_object_tab()
. Didn't investigate much but seems STACK is bogus.

It looks to me there are two ways to enable MT with SINGLEMAP_MEMORY:
* disable SINGLEMAP_MEMORY_STACK and fix the above problem.
* allocate new threads stacks in designated memory region instead of
malloc-ing. In this way make_framepointer/topofframe/uTheFramepointer will
behave as expected in the context of new threads.

I tend towards the latter.




On Tue, May 23, 2017 at 11:35 AM, Bruno Haible <br...@cl...> wrote:

> Hi Vladimir,
>
> > I disabled SINGLEMAP_MEMORY model in MT builds. The problem is that this
> > model assumes single fixed stack address range while every thread
> requires
> > it's own stack which is currently malloc-ed. What is the good way to
> > workaround this? Allocate new threads stacks in the above range?
>
> You mean, the only problem with MT and SINGLEMAP_MEMORY was that in spvw.d
> lines 3230..3245 we allocate one STACK range only, not multiple ones?
> If yes, I would try to keep SINGLEMAP_MEMORY turned on and disable only
> SINGLEMAP_MEMORY_STACK.
>
> Bruno
>
>