Re: [Xen-devel] [PATCH RFC] xen: add undefined behaviour sanitizer

[Julien Grall <julien.grall@xxxxxxx>]

Hi,

On 03/10/17 17:19, Andrew Cooper wrote:
> On 03/10/17 17:03, Wei Liu wrote:
> > diff --git a/xen/common/ubsan/ubsan.c b/xen/common/ubsan/ubsan.c
> > new file mode 100644
> > index 0000000000..aa359e712c
> > --- /dev/null
> > +++ b/xen/common/ubsan/ubsan.c
> > @@ -0,0 +1,486 @@
> > +/*
> > + * UBSAN error reporting functions
> > + *
> > + * Copyright (c) 2014 Samsung Electronics Co., Ltd.
> > + * Author: Andrey Ryabinin <ryabinin.a.a@xxxxxxxxx>
> > + *
> > + * This program is free software; you can redistribute it and/or modify
> > + * it under the terms of the GNU General Public License version 2 as
> > + * published by the Free Software Foundation.
> > + *
> > + * Adapt to Xen by:
> > + *  Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> > + *  Wei Liu <wei.liu2@xxxxxxxxxx>
> > + */
> > +
> > +#include <xen/bitops.h>
> > +#include <xen/kernel.h>
> > +#include <xen/lib.h>
> > +#include <xen/types.h>
> > +#include <xen/sched.h>
> > +#include <xen/spinlock.h>
> > +#include <xen/percpu.h>
> > +
> > +#define pr_err(...) printk(XENLOG_ERR __VA_ARGS__)
> > +#define dump_stack dump_execution_state
> > +#define __noreturn noreturn
> > +#define CONFIG_ARCH_SUPPORTS_INT128
> > +#define u64 long long unsigned int
> > +#define s64 long long int
> These last three should be in asm-$ARCH/config.h or similar.  I'm fairly
> sure arm32 will struggle with INT128.
AFAICT, Linux doesn't enable this option for both arm64 and arm32.

> Also, I think the current (re)engineering should stay.  There is nothing
> vcpu specific about the in_ubsan flag.  (Especially, given your
> tbutils.o observation above.)
>
> As for commits introducing the code, I had a vague plan to introduce the
> unmodified code from upstream Linux, then my forward ported patch (as
> that isn't in Linux, even today), and then a change adding the Kconfig
> option, and these defines, to get it to compile.
That would be handy to see the changes.

> Given that this is clearly a debugging option, and its basically ready,
> I wouldn't necessarily hold off from putting it into 4.10.
I thought about the same when I saw the patch. I guess we can get UBSAN 
support for x86 into Xen 4.10. I don't think it will be possible for Arm 
given the necessary rework in the memory layout (we need to expand the 
region size for Xen mapping) and assembly.
Cheers,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel