Re: [kvm-unit-tests PATCH 3/4] arm/arm64: Track whether thread_info has been initialized

[Andrew Jones <drjones@redhat.com>]

On Mon, Mar 22, 2021 at 10:59:31AM +0000, Nikos Nikoleris wrote:
> Hi Alex,
> 
> On 22/03/2021 10:34, Alexandru Elisei wrote:
> > Hi Nikos,
> > 
> > On 3/19/21 12:24 PM, Nikos Nikoleris wrote:
> > > Introduce a new flag in the thread_info to track whether a thread_info
> > > struct is initialized yet or not.
> > 
> > There's no explanation why this is needed. The flag checked only by is_user(), and
> > before thread_info is initialized, flags is zero, so is_user() would return false,
> > right? Or am I missing something?
> > 
> 
> I am still not sure what's the right approach here. I didn't like and I
> still don't like the fact that we rely on implicit 0 initialization to get
> the right behavior. This will break once we add support for EFI. I think we
> should explicitly initialize thread_info to 0.

I just sent a patch doing this. Let me know what you think.


> I was thinking of adding a
> thread_info_alloc() function to do this.

I'm not sure how this would look. We want the thread-info to live on the
bottom of the stack and the bootcpu's stack is allocated in the linker
script.

> 
> By having this flag I think we can guard accesses to the thread_info in
> general. I didn't want to turn the define smp_processor_id to a function
> here but I think we should and assert that the thread_info is valid and
> avoid reading current_thread_info()->cpu.

Hmm, yeah, hopefully we can avoid this flag and adding an assert to
smp_processor_id(). Let's take another look at this after we ensure
that the thread-info is explicitly zeroed at startup.

Thanks,
drew