Re: Thread safety of coroutine-sigaltstack

From: Peter Maydell <peter.maydell@linaro.org>
To: "Daniel P. Berrangé" <berrange@redhat.com>
Cc: Kevin Wolf <kwolf@redhat.com>,
	"qemu-devel@nongnu.org" <qemu-devel@nongnu.org>,
	Max Reitz <mreitz@redhat.com>,
	Stefan Hajnoczi <stefanha@redhat.com>,
	Paolo Bonzini <pbonzini@redhat.com>,
	Laszlo Ersek <lersek@redhat.com>
Subject: Re: Thread safety of coroutine-sigaltstack
Date: Thu, 21 Jan 2021 16:44:40 +0000	[thread overview]
Message-ID: <CAFEAcA-zVvu+16xxjCvXqeZxZMqppUUb7aQkL-5nBTNY+JeyGg@mail.gmail.com> (raw)
In-Reply-To: <20210121160733.GP3125227@redhat.com>

On Thu, 21 Jan 2021 at 16:10, Daniel P. Berrangé <berrange@redhat.com> wrote:
> FWIW The libucontext impl is all ASM based and has coverage for all the
> arches we care about:
>
>    https://github.com/kaniini/libucontext
>
> so doesn't seem like there's a need for  coroutine-asm if we can rely
> on libucontext for portability where neede.

The README for that notes a couple of pretty major omissions:
 * it doesn't handle floating point registers
 * it doesn't do anything about the signal mask
I'm pretty sure that not handling the fp regs could cause breakage
for Arm at least (depending on what the compiler chooses to do
in the functions that work with the ucontext functions). The
signal mask stuff might be OK for us because of the carefully
limited use we make of the ucontext functions, but it would be
a bit of a pain to have to analyse that code for both sets of semantics.

thanks
-- PMM