zhong jiang <zhongjiang@xxxxxxxxxx> writes: > On 2016/7/12 23:19, Eric W. Biederman wrote: >> zhongjiang <zhongjiang@xxxxxxxxxx> writes: >> >>> From: zhong jiang <zhongjiang@xxxxxxxxxx> >>> >>> In general, kexec alloc pages from buddy system, it cannot exceed >>> the physical address in the system. >>> >>> The patch just remove this unnecessary code, no functional change. >> On 32bit systems with highmem support kexec can very easily receive a >> page from the buddy allocator that can exceed 4GiB. This doesn't show >> up on 64bit systems as typically the memory limits are less than the >> address space. But this code is very necessary on some systems and >> removing it is not ok. >> >> Nacked-by: "Eric W. Biederman" <ebiederm@xxxxxxxxxxxx> >> > This viewpoint is as opposed to me, 32bit systems architectural decide it can not > access exceed 4GiB whether the highmem or not. but there is one exception, > when PAE enable, its physical address should be extended to 36, new paging mechanism > established for it. therefore, the page from the buddy allocator > can exceed 4GiB. Exactly. And I was dealing with PAE systems in 2001 or so with > 4GiB of RAM. Which is where the unusable_pages work comes from. Other architectures such as ARM also followed a similar path, so it isn't just x86 that has 32bit systems with > 32 address lines. > moreover, on 32bit systems I can not understand why KEXEC_SOURCE_MEMORY_LIMIT > is defined to -1UL. therefore, kimge_aloc_page allocate page will always add to unusable_pages. -1UL is a short way of writing 0xffffffffUL Which is as close as you can get to writing 0x100000000UL in 32bits. kimage_alloc_page won't always add to unusable_pages as there is memory below 4GiB but it isn't easily found so there may temporarily be a memory shortage, as it allocates it's way there. Unfortunately whenever I have looked there are memory zones that line up with the memory the kexec is looking for. So it does a little bit of a weird dance to get the memory it needs and to discard the memory it can't use. Eric -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>