Re: [Qemu-devel] [PATCH] fix qemu-system-unicore32 crashing when calling without -kernel

[Philippe Mathieu-Daudé]

On 07/26/2017 01:05 PM, Thomas Huth wrote:
> On 26.07.2017 18:00, Philippe Mathieu-Daudé wrote:
> > On 07/26/2017 12:04 PM, Eduardo Otubo wrote:
> > > Starting qemu-system-unicore32 without the -kernel parameter results in
> > > an assert() returns false and aborts qemu. This patch replaces it with a
> > > proper error message followed by exit(1).
> > >
> > > Signed-off-by: Eduardo Otubo <address@hidden>
> > > ---
> > >    hw/unicore32/puv3.c | 5 ++++-
> > >    1 file changed, 4 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/hw/unicore32/puv3.c b/hw/unicore32/puv3.c
> > > index e9d1a60b6f..ff62efb4df 100644
> > > --- a/hw/unicore32/puv3.c
> > > +++ b/hw/unicore32/puv3.c
> > > @@ -92,7 +92,10 @@ static void puv3_load_kernel(const char
> > > *kernel_filename)
> > >        if (kernel_filename == NULL && qtest_enabled()) {
> > >            return;
> > >        }
> > > -    assert(kernel_filename != NULL);
> > > +    if (kernel_filename == NULL) {
> > > +        error_report("kernel parameter cannot be empty");
> > > +        exit(1);
> > > +    }
> >
> > This seems a temporary kludge for 2.10 but not the correct long-term fix.
> >
> > As commented in another thread [1] where I got a bit discomfited, it'd
> > be nice if during the 2.11 schedule the industry provide some help to
> > clean/unify the loader.c code, ideally remove all duplicated code not
> > arch-specific and use hw/core/loader*.
> >
> > [1] http://lists.nongnu.org/archive/html/qemu-devel/2017-07/msg06760.html
>
> Sorry, but this here *is* a machine-specific problem. I fail to see how
> this should be related to the hw/core/loader.c code?

Yes, this assert is a problem, same as the one in xen_load_linux() and 
this is a surgical fix, indeed.

I probably misused the "kludge" word, sorry. My comment was about 
long-term use of -kernel.

Maybe I didn't point to the correct file, I was thinking about the 
various hw/$arch/boot.c. My feeling is the -kernel parameter started as 
something simple years ago then grow and now his usage is somehow 
abused. I'd rather see a common hw/core/boot.c with each arch just 
adding arch-related fixes.

For example in hw/arm/boot.c, see arm_load_kernel_notify(), except 
features parsing, initrd_start and the fixupcontext part, most of the 
code is multiarch and could benefit other archs less lov^H^H^Hsupported. 
arm_load_kernel() is generic. In load_dtb() only the acells/scells is 
arm-specific and could replace dup microblaze code, work on mips, etc.

Regards,

Phil.