On Tue, Mar 30, 2010 at 10:09:47PM +0200, Juergen Lock wrote:
> In article <4bb2540b.90...@twiddle.net> you write:
> >On 03/30/2010 12:16 PM, Juergen Lock wrote:
> >> I first tried to replace the endaddr in the !h2g_valid(endaddr) case with
> >> ((abi_ulong)1 << L1_MAP_ADDR_SPACE_BITS) - 1
> >> if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS (which comes from the condition
> >> of the assert in page_set_flags() that was triggered on the ~0ul value),
> >> but that caused the qemu process to grow into swap and made the box
> >> usuable when that code was reached and I had to kill qemu. (The box has
> >> 8 GB RAM.) And so I thought just leaving that page range unprotected
> >> if only the start address is valid was the lesser evil...
> >
> >What's are the real arguments to the page_set_flags that causes things
> >to go into swap? I can't imagine the range really being so large that
> >it causes massive allocation within that function...
>
> Oh sorry if that was not clear, things go into swap if I _replace_ the
> endaddr ~0ul (which caused the assert) with the max value the assert
> still tolerates i.e.
> ((abi_ulong)1 << L1_MAP_ADDR_SPACE_BITS) - 1
> which in this case seems to be 0x7fffffffffff:
>
> #3 0x0000000060012731 in page_set_flags (start=140737488224256,
> end=18446744073709551615, flags=32)
> at
> /usr/ports/emulators/qemu-devel-20100323a/work/qemu-snapshot-20100323_20/exec.c:2426
> 2426 assert(end < ((abi_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
> (gdb) i li 2426
> Line 2426 of
> "/usr/ports/emulators/qemu-devel-20100323a/work/qemu-snapshot-20100323_20/exec.c"
> starts at address 0x60012662 <page_set_flags+34>
> and ends at 0x60012675 <page_set_flags+53>.
> (gdb) disassemble 0x60012662 0x60012675
> Dump of assembler code from 0x60012662 to 0x60012675:
> 0x0000000060012662 <page_set_flags+34>: mov $0x7fffffffffff,%rax
> ^^^^^^^^^^^^^^
> 0x000000006001266c <page_set_flags+44>: cmp %rax,%rsi
> 0x000000006001266f <page_set_flags+47>: ja 0x60012718
> <page_set_flags+216>
> End of assembler dump.
> (gdb) q
Ok sorry about the confusion, this is a different problem, I just looked
at the value of start, it seems to be:
(gdb) p start
$2 = 0x7ffffffe0000
So I'd say the real problem is page_set_flags() has a bug that makes
it allocate too much if the range is the last allowed page...
Cheers,
Juergen
