Re: Tor Browser on -current dies with SIGILL

On Tue, Jul 02, 2024 at 01:00:34PM +0100, Stuart Henderson wrote:
> On 2024/07/01 22:09, Matthias Schmidt wrote:
> > Hi,
> >
> > the recent update to tor-browser-13.5 on amd64 -current broke it for me.
> > It aborts right after start with SIGILL. Doesn't matter if I use an existing
> > config or a fresh one.
> >
> > $ egdb /usr/local/lib/tor-browser/tor-browser
> > Reading symbols from /usr/local/lib/tor-browser/tor-browser...
> > (No debugging symbols found in /usr/local/lib/tor-browser/tor-browser)
> > (gdb) r
> > Starting program: /usr/local/lib/tor-browser/tor-browser
> > [New thread 600821]
> > [New thread 465054]
> > [New thread 362995]
> > [New thread 178360]
> > [New thread 381574]
> > [New thread 412107]
> > [New thread 509418]
> > [New thread 420661]
> >
> > Thread 1 received signal SIGILL, Illegal instruction.
> > 0x00000f503c91d3a0 in ?? ()
> > (gdb) bt
> > #0 0x00000f503c91d3a0 in ?? ()
> > #1 0x00000f4ff04c0c3f in js::jit::MaybeEnterJit(JSContext*, js::RunState&) () from /usr/local/lib/tor-browser/libxul.so.9.0
> >
> > I have a 11th gen Intel CPU with BTI so it might be related to that...
>
> Based on the function name, SIGILL, and cpu that is quite likely.
> The way to be sure is to run under ktrace and check kdump, on a
> 12th gen Intel machine I have
>
> 67324 tor-browser PSIG SIGILL caught handler=0x3d79d66aa40 mask=0<> code=ILL_BTCFI addr=0x3d80b3fa3a0 trapno=21
>
> so yes.
>
> mozilla.port.mk is setting USE_NOBTCFI so it's supposed to be built
> with the relevant linker flags to turn this off, but perhaps there's
> something in tor-browser build which is overriding that..
>
> firefox-esr (also 115-based) *is* working correctly on this machine.

Useful to know, thanks.

I'm building both ports and I'll compare the logs to see if I can spot
a relevant difference (for example, to see whether $WRKDIR/bin/ld
is used properly). I've looked at the port Makefiles but I could not
immediately spot the reason for this.

Caspar