Re: BUG: sleeping function called from invalid context at ./include/linux/uaccess.h:LINE

From: Dmitry Vyukov <dvyukov@google.com>
To: Paolo Bonzini <pbonzini@redhat.com>
Cc: "David Hildenbrand" <david@redhat.com>,
	"Nick Desaulniers" <nick.desaulniers@gmail.com>,
	syzbot
	<bot+eb276410a47365e399ee50d68e73650cac178e0f@syzkaller.appspotmail.com>,
	"Christian Borntraeger" <borntraeger@de.ibm.com>,
	"H. Peter Anvin" <hpa@zytor.com>,
	"KVM list" <kvm@vger.kernel.org>,
	LKML <linux-kernel@vger.kernel.org>,
	"Lorenzo Stoakes" <lstoakes@gmail.com>,
	"Michal Hocko" <mhocko@suse.com>,
	"Ingo Molnar" <mingo@redhat.com>,
	"Radim Krčmář" <rkrcmar@redhat.com>,
	syzkaller-bugs@googlegroups.com,
	"Thomas Gleixner" <tglx@linutronix.de>,
	"the arch/x86 maintainers" <x86@kernel.org>
Subject: Re: BUG: sleeping function called from invalid context at ./include/linux/uaccess.h:LINE
Date: Tue, 7 Nov 2017 13:44:43 +0100	[thread overview]
Message-ID: <CACT4Y+aQEkT2TnGbs3VZu42e1EKcM8M_Q=VGZveZH-dLkOXSrg@mail.gmail.com> (raw)
In-Reply-To: <890dcad0-800a-e840-be7f-90c709831741@redhat.com>

On Tue, Nov 7, 2017 at 1:29 PM, Paolo Bonzini <pbonzini@redhat.com> wrote:
>>>>> Does it have to be stack allocated?
>>>>
>>>> We can't use kmalloc and friends in emulate.c. We would have to
>>>> introduce new emulator callbacks.
>>>>
>>>> a) for malloc and free. hmmm.
>>>> b) for carrying out the fxrstr/fixup.
>>>>
>>>> Paolo, what do you suggest?
>>>
>>> You can use kmalloc.  Any userspace user of emulate.c would have to
>>> write a wrapper.
>>
>>
>> Can you please tell me more about this? Is it used for testing? Is
>> there an example code that builds and tests this in user-space?
>
> Not quite, there's no user outside KVM yet.  But the emulator code is
> designed to be independent from KVM's memory access primitives; with
> "nm" you can see how there are very few undefined symbols:
>
>                  U ex_handler_default
>                  U find_first_bit
>                  U find_next_bit
>                  U memcpy
>                  U printk
>
> Exceptions are only used for div/idiv, if it gets in the way it's okay
> to just revert commit b8c0b6ae498f ("KVM: x86 emulator: convert DIV/IDIV
> to fastop", 2013-05-21).
>
> On the other hand, dependencies on Linux headers have sneaked in more
> and more, but refactoring those away should not be too hard.

Thanks for the info. We may be potentially interested in fuzzing this
in user-space.