Messages in this thread | | | From | "Rafael J. Wysocki" <> | Subject | Re: S4 resume broken since 2.6.39 (3.1, too) | Date | Tue, 27 Sep 2011 19:03:14 +0200 |
| |
On Tuesday, September 27, 2011, Yinghai Lu wrote: > On 09/26/2011 03:24 PM, Rafael J. Wysocki wrote: > > > On Thursday, September 22, 2011, Yinghai Lu wrote: > >> On Wed, Sep 21, 2011 at 11:48 AM, Rafael J. Wysocki <rjw@sisk.pl> wrote: > >>> It looks like init_memory_mapping() is sometimes called with "end" > >>> beyond the last mapped PFN and it explodes when we try to write stuff to > >>> that address during image restoration. > >>> > >>> IOW, the Yinghai's assumption that init_memory_mapping() would always be > >>> called with a "good end" on x86_64 was overomptimistic. > >> > >> for 64bit x86, kernel_physical_mapping_init() will use > >> map_low_page()/call early_memmap() to access ram for page_table that is above > >> rather last mapped PFN. > >> > >> the point is: > >> on system with 64g, usable ram will be [0,2048m), [4g, 64g) > >> init_memory_mapping will be called two times for them. > >> before putting page_table high, > >> page table will be two parts: one is just below 512M, and one below 2048m. > >> after putting page_table high, > >> page table will be two parts: one is just below 2048M, and one below 64G. > >> > >> one of the purposes is finding biggest continuous big range under > >> 1024m for kdump. > > > > This is all fine so long as we can ensure that the "end" value we're > > passing to init_memory_mapping() will always be a valid address, which > > evidently is not the case sometimes. > > > I don't understand why end is not valid could happen. > > end should be always valid address. one is max_low_pfn under 4g, and another one is max_pfn... > > > > > > So, in my opinion we should simply apply the Takashi's patch at this > > point and revisit the kdump issue later, when we actually know how to do > > the right thing. > > > Takashi said: 2.6.37 with that commit is ok, only 2.6.39 somehow has the 1/20 chance has the reset problem. > > so that commit should not the cause. could be some hidden assumption from > restore code ?
Quite frankly, I doubt it. The only remotely related change between 2.6.37 and 2.6.37 seems to be commit d1ee433 (x86, trampoline: Use the unified trampoline setup for ACPI wakeup).
Thanks, Rafael
| |