Re: [RFC 0/2] tls: add macros for coroutine-safe TLS variables

[Stefan Hajnoczi]

On Tue, Oct 26, 2021 at 10:10:44AM -0700, Richard Henderson wrote:
> On 10/26/21 9:34 AM, Stefan Hajnoczi wrote:
> > On Tue, Oct 26, 2021 at 08:10:16AM -0700, Richard Henderson wrote:
> > > On 10/26/21 6:22 AM, Stefan Hajnoczi wrote:
> > > > If "safe" TLS variables are opt-in then we'll likely have obscure bugs
> > > > when code changes to access a TLS variable that was previously never
> > > > accessed from a coroutine. There is no compiler error and no way to
> > > > detect this. When it happens debugging it is painful.
> > > 
> > > Co-routines are never used in user-only builds.
> > 
> > If developers have the choice of using __thread then bugs can slip
> > through.
> 
> Huh?  How.  No, really.

If there is no checkpatch.pl error then more instances of __thread will
slip in. Not everyone in the QEMU community will be aware of this issue,
so it's likely that code with __thread will get merged.

Subsystems that use coroutines today include block, 9p, mpqemu, io
channels, migration, colo, and monitor commands.

I understand that qemu-user is particularly unlikely to use coroutines.
Thomas' suggestion sounds good to me. Let's allow __thread only in
subsystems where it's safe.

Stefan

signature.asc
Description: PGP signature