Re: ipv4/tcp.c:4234:1: error: the frame size of 1152 bytes is larger than 1024 bytes [-Werror=frame-larger-than=]

[Wei Liu <wei.liu@kernel.org>]

On Tue, Sep 07, 2021 at 04:14:24PM -0700, Linus Torvalds wrote:
> [ Added maintainers for various bits and pieces, since I spent the
> time trying to look at why those bits and pieces wasted stack-space
> and caused problems ]
> 
> On Tue, Sep 7, 2021 at 3:16 PM Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
[...]
> 
> There are many more of these cases. I've seen Hyper-V allocate 'struct
> cpumask' on the stack, which is once again an absolute no-no that
> people have apparently just ignored the warning for. When you have
> NR_CPUS being the maximum of 8k, those bits add up, and a single
> cpumask is 1kB in size. Which is why you should never do that on
> stack, and instead use '
> 
>        cpumask_var_t mask;
>        alloc_cpumask_var(&mask,..)
> 
> which will do a much more reasonable job. But the reason I call out
> hyperv is that as far as I know, hyperv itself doesn't actually
> support 8192 CPU's. So all that apic noise with 'struct cpumask' that
> uses 1kB of data when NR_CPUS is set to 8192 is just wasted. Maybe I'm
> wrong. Adding hyperv people to the cc too.
> 
> A lot of the stack frame size warnings are hidden by the fact that our
> default value for warning about stack usage is 2kB for 64-bit builds.
> 
> Probably exactly because people did things like that cpumask thing,
> and have these arrays of structures that are often even bigger in the
> 64-bit world.
> 

Thanks for the heads-up. I found one instance of this bad practice in
hv_apic.c. Presumably that's the one you were referring to.

However calling into the allocator from that IPI path seems very heavy
weight. I will discuss with fellow engineers on how to fix it properly.

Wei.

>                 Linus