[Qemu-devel] ARM hosts: code_gen_alloc() maps code buffer on top of libc heap

[Qemu-devel] ARM hosts: code_gen_alloc() maps code buffer on top of libc heap

[Peter Maydell]

I've been tracking down a bug where qemu run on ARM hosts will
(about half of the time) abort early in execution with:

qemu-system-i386: malloc.c:3096: sYSMALLOc: Assertion `(old_top ==
(((mbinptr) (((char *) &((av)->bins[((1) - 1) * 2])) -
__builtin_offsetof (struct malloc_chunk, fd)))) && old_size == 0) ||
((unsigned long) (old_size) >= (unsigned long)((((__builtin_offsetof
(struct malloc_chunk, fd_nextsize))+((2 * (sizeof(size_t))) - 1)) &
~((2 * (sizeof(size_t))) - 1))) && ((old_top)->size & 0x1) &&
((unsigned long)old_end & pagemask) == 0)' failed.

This turns out to be because code_gen_alloc() is using
mmap(MAP_FIXED) to map the code buffer at address
0x01000000UL, which is in the area glibc happens to be using
for its heap. This tends to make the next malloc() abort,
although occasionally the stars align and we pass that
and fail weirdly later on.

I suspect we need to drop the MAP_FIXED requirement and
fix the TCG code to cope with emitting code for longer-range
branches for calls to host fns etc (calls/branches within the
generated code should be ok to keep using the short-range
branch insn I think). There is already no guarantee that
the generated code and the host C code are within short
branch range of each other...

-- PMM

Re: [Qemu-devel] ARM hosts: code_gen_alloc() maps code buffer on top of libc heap

[Paolo Bonzini]

On 10/28/2011 04:32 PM, Peter Maydell wrote:
>
> I suspect we need to drop the MAP_FIXED requirement and
> fix the TCG code to cope with emitting code for longer-range
> branches for calls to host fns etc (calls/branches within the
> generated code should be ok to keep using the short-range
> branch insn I think). There is already no guarantee that
> the generated code and the host C code are within short
> branch range of each other...

Does USE_STATIC_CODE_GEN_BUFFER fix it?  Do you know why "Currently it 
is not recommended to allocate big chunks of data in user mode"?

Paolo

Re: [Qemu-devel] ARM hosts: code_gen_alloc() maps code buffer on top of libc heap

[Peter Maydell]

On 28 October 2011 17:14, Paolo Bonzini <pbonzini@redhat.com> wrote:
> On 10/28/2011 04:32 PM, Peter Maydell wrote:
>>
>> I suspect we need to drop the MAP_FIXED requirement and
>> fix the TCG code to cope with emitting code for longer-range
>> branches for calls to host fns etc (calls/branches within the
>> generated code should be ok to keep using the short-range
>> branch insn I think). There is already no guarantee that
>> the generated code and the host C code are within short
>> branch range of each other...
>
> Does USE_STATIC_CODE_GEN_BUFFER fix it?

Well, it avoids the malloc abort. We seem to get stuck in an
infinite cycle of the guest trying to start grub and rebooting
before it manages to print the blue-background grub menu screen,
though, so something's still not right. Likely a different bug,
though.

> Do you know why "Currently it is
> not recommended to allocate big chunks of data in user mode"?

Don't know, but my guess is that it is or was likely to clash
with attempts to put the guest binary in the right place in
the address space.

-- PMM