Date: Tue, 07 Mar 2017 12:13:30 +0000 From: Andrew Gierth <andrew@tao11.riddles.org.uk> To: Mark Millard <markmi@dsl-only.net> Cc: Warner Losh <imp@bsdimp.com>, "freebsd-arm\@freebsd.org" <freebsd-arm@freebsd.org> Subject: Re: Is CPUTYPE=cortex-A7 supposed to work? Message-ID: <87fuipe460.fsf@news-spur.riddles.org.uk> In-Reply-To: <3056BE0A-6441-4878-905D-741C61BDC47C@dsl-only.net> (Mark Millard's message of "Sun, 5 Mar 2017 20:26:49 -0800") References: <871suc3nv8.fsf@news-spur.riddles.org.uk> <CANCZdfq4EwH%2B_9FVNai8s6Y-gdTjHJ8dNkJwSrnF%2BSAkdwvYdg@mail.gmail.com> <87tw7820fc.fsf@news-spur.riddles.org.uk> <644D1F49-BF5D-409D-BFC4-4F7E6E73085B@dsl-only.net> <87lgsk1udz.fsf@news-spur.riddles.org.uk> <9677298B-5A5E-44BF-928E-28DDDADB310A@dsl-only.net> <87h93814rb.fsf@news-spur.riddles.org.uk> <70AE704A-C9FF-4742-88E9-147CD5B77BE8@dsl-only.net> <87tw77iwx6.fsf@news-spur.riddles.org.uk> <8760jnfvwp.fsf@news-spur.riddles.org.uk> <3056BE0A-6441-4878-905D-741C61BDC47C@dsl-only.net>
index | next in thread | previous in thread | raw e-mail
Bingo. Signal delivery isn't preserving the NEON registers (at all, as far as I can tell); if the signal handler changes any of them, those changes are visible in the interrupted code. If libthr is compiled with -mcpu=cortex-a7, then it uses NEON registers for data copying in thr_sighandler, so obviously things go south at that point. Equally bad things happen if the signal handler does any floating point or vector work itself whether linked with libthr or not. So it looks like this is a kernel bug after all, no? Looking at arm/arm/machdep.c reveals a noticable absence of any attempt to preserve the VFP state in sendsig(), contrary to what happens on other platforms. -- Andrew.help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?87fuipe460.fsf>