Re: [Qemu-devel] [PATCH 1/2] softfloat: Enable run-time-configurable meaning of signaling NaN bit

From: Eduardo Habkost <ehabkost@redhat.com>
To: Peter Maydell <peter.maydell@linaro.org>
Cc: "proljc@gmail.com" <proljc@gmail.com>,
	"kbastian@mail.uni-paderborn.de" <kbastian@mail.uni-paderborn.de>,
	"mark.cave-ayland@ilande.co.uk" <mark.cave-ayland@ilande.co.uk>,
	"qemu-devel@nongnu.org" <qemu-devel@nongnu.org>,
	"agraf@suse.de" <agraf@suse.de>,
	"blauwirbel@gmail.com" <blauwirbel@gmail.com>,
	"jcmvbkbc@gmail.com" <jcmvbkbc@gmail.com>,
	Aleksandar Markovic <Aleksandar.Markovic@imgtec.com>,
	"qemu-arm@nongnu.org" <qemu-arm@nongnu.org>,
	"qemu-ppc@nongnu.org" <qemu-ppc@nongnu.org>,
	Petar Jovanovic <Petar.Jovanovic@imgtec.com>,
	"pbonzini@redhat.com" <pbonzini@redhat.com>,
	Miodrag Dinic <Miodrag.Dinic@imgtec.com>,
	"edgar.iglesias@gmail.com" <edgar.iglesias@gmail.com>,
	"gxt@mprc.pku.edu.cn" <gxt@mprc.pku.edu.cn>,
	Leon Alrae <Leon.Alrae@imgtec.com>,
	"afaerber@suse.de" <afaerber@suse.de>,
	"aurelien@aurel32.net" <aurelien@aurel32.net>,
	Richard Henderson <rth@twiddle.net>
Subject: Re: [Qemu-devel] [PATCH 1/2] softfloat: Enable run-time-configurable meaning of signaling NaN bit
Date: Mon, 4 Apr 2016 16:42:36 -0300	[thread overview]
Message-ID: <20160404194236.GE22665@thinpad.lan.raisama.net> (raw)
In-Reply-To: <CAFEAcA9axhkDC0tDT1ujNZUpHbCnMqSxaL-KTD2=TV=XOyEbuw@mail.gmail.com>

On Mon, Apr 04, 2016 at 08:38:54PM +0100, Peter Maydell wrote:
> On 4 April 2016 at 20:37, Eduardo Habkost <ehabkost@redhat.com> wrote:
> > On Mon, Apr 04, 2016 at 02:31:47PM +0100, Peter Maydell wrote:
> >> On 4 April 2016 at 14:21, Aleksandar Markovic
> >> <Aleksandar.Markovic@imgtec.com> wrote:
> >> > B. arm - explicitely sets other fields of float_status,
> >> >       explicit invocation of set_snan_bit_is_one(0) added
> >>
> >> We zero the float_status structs on reset, because they are earlier
> >> in the CPUARMState structure than the 'features' field (and so the
> >> memset() in arm_cpu_reset() will clear them). So you don't
> >> need to explicitly zero a field like this. I expect the other
> >> architectures are the same.
> >
> > Even if it is not zeroed on reset, it is zeroed on object_new().
> > Isn't that enough?
> 
> It must be zeroed on reset, otherwise we won't get the right
> behaviour if you reset the CPU after running it for a bit.
> object_new() zeroing is not sufficient.

The only calls to set_snan_bit_is_one() with non-zero arguments I
see on this patch are during CPU init or reset. How exactly would
the snan_bit_is_one field change to non-zero during runtime, to
require zeroing it again on reset?

-- 
Eduardo