Re: [Qemu-devel] [PATCH] exec-obsolete: fix length handling

[Blue Swirl]

On Sun, Jan 29, 2012 at 13:20, Avi Kivity <address@hidden> wrote:
> On 01/29/2012 03:16 PM, Blue Swirl wrote:
>> On Sun, Jan 29, 2012 at 12:08, Avi Kivity <address@hidden> wrote:
>> > On 01/28/2012 08:13 PM, Blue Swirl wrote:
>> >> Fix suspend/resume broken by off-by-one error in
>> >> 59abb06198ee9471e29c970f294eae80c0b39be1.
>> >>
>> >> Adjust the loop so that it handles correctly the case
>> >> start = (ram_addr_t)-TARGET_PAGE_SIZE, length = TARGET_PAGE_SIZE.
>> >>
>> >> Reported-by: Stefan Berger <address@hidden>
>> >> Signed-off-by: Blue Swirl <address@hidden>
>> >> ---
>> >>  exec-obsolete.h |   10 ++++------
>> >>  1 files changed, 4 insertions(+), 6 deletions(-)
>> >>
>> >> diff --git a/exec-obsolete.h b/exec-obsolete.h
>> >> index 03cf35e..1bba970 100644
>> >> --- a/exec-obsolete.h
>> >> +++ b/exec-obsolete.h
>> >> @@ -81,11 +81,10 @@ static inline void
>> >> cpu_physical_memory_set_dirty_range(ram_addr_t start,
>> >>                                                         int dirty_flags)
>> >>  {
>> >>      uint8_t *p;
>> >> -    ram_addr_t addr, end;
>> >> +    ram_addr_t cur;
>> >>
>> >> -    end = start + length;
>> >>      p = ram_list.phys_dirty + (start >> TARGET_PAGE_BITS);
>> >> -    for (addr = start; addr <= end; addr += TARGET_PAGE_SIZE) {
>> >> +    for (cur = 0; cur < length; cur += TARGET_PAGE_SIZE) {
>> >>          *p++ |= dirty_flags;
>> >>      }
>> >
>> > I think this is still wrong - if length == 2 it will iterate once, but
>> > we need two iterations if start == 0xfff.
>>
>> Yes, tricky. We could do something like
>> for (cur = start & TARGET_PAGE_MASK; cur < length; cur += TARGET_PAGE_SIZE) {
>> but I'll send a new patch with just s/<=/</.
>
> That's broken too.

Because length should be adjusted by -1?

> I have:
>
>     uint8_t *p;
>     ram_addr_t addr, end;
>
> -    end = start + length;
> +    end = (start + length - 1) | (TARGET_PAGE_SIZE - 1);

Why  | (TARGET_PAGE_SIZE - 1), for length == 1? TARGET_PAGE_ALIGN()
could be useful here.

> +    start &= TARGET_PAGE_MASK;
>     p = ram_list.phys_dirty + (start >> TARGET_PAGE_BITS);
>     for (addr = start; addr <= end; addr += TARGET_PAGE_SIZE) {
>         *p++ |= dirty_flags;
> @@ -98,7 +99,8 @@ static inline void
> cpu_physical_memory_mask_dirty_range(ram_addr_t start,
>     uint8_t *p;
>     ram_addr_t addr, end;
>
> -    end = start + length;
> +    end = (start + length - 1) | (TARGET_PAGE_SIZE - 1);
> +    start &= TARGET_PAGE_MASK;
>     mask = ~dirty_flags;
>     p = ram_list.phys_dirty + (start >> TARGET_PAGE_BITS);
>     for (addr = start; addr <= end; addr += TARGET_PAGE_SIZE) {
>
>
> And a non-terminating migration - not sure if this is the cause.
>
> --
> error compiling committee.c: too many arguments to function
>