A cursory grep indicates the architectures which may be affected by this bug are: arm, avr32, m32r, microblaze, mips, powerpc, s390, sparc, x86 The problem is the assumption that if you have a piece of contiguous physical memory, you can iterate over it by starting with the first struct page *page and doing page++. The page++ trick works with FLATMEM and DISCONTIGMEM, but not SPARSEMEM. The reason is that the SPARSEMEM section arrays have arbitrary power of 2 boundaries, so you can end up with physically adjacent pieces of memory which are in different section arrays, meaning that page++ doesn't work, because it doesn't move from the end of the old section array to the beginning of the new one. The fix is to iterate by means of the pfn instead (the pfn_to_page() functions do the right thing with respect to the section arrays) or simply replace page++ with page = pfn_to_page(page_to_pfn(page) + 1); A large number of the architectures seem to pick up the problem through huge pages, so I note that as long as your SECTION_SIZE_BITS is large enough (i.e. huge page size < 1 << (SECTION_SIZE_BITS + PAGE_SIZE)) you should be OK. James -- To unsubscribe from this list: send the line "unsubscribe linux-arch" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
