4.9-stable review patch. If anyone has any objections, please let me know.
------------------ From: Baoquan He <[email protected]> commit fc5f9d5f151c9fff21d3d1d2907b888a5aec3ff7 upstream. Jeff Moyer reported that on his system with two memory regions 0~64G and 1T~1T+192G, and kernel option "memmap=192G!1024G" added, enabling KASLR will make the system hang intermittently during boot. While adding 'nokaslr' won't. The back trace is: Oops: 0000 [#1] SMP RIP: memcpy_erms() [ .... ] Call Trace: pmem_rw_page() bdev_read_page() do_mpage_readpage() mpage_readpages() blkdev_readpages() __do_page_cache_readahead() force_page_cache_readahead() page_cache_sync_readahead() generic_file_read_iter() blkdev_read_iter() __vfs_read() vfs_read() SyS_read() entry_SYSCALL_64_fastpath() This crash happens because the for loop count calculation in sync_global_pgds() is not correct. When a mapping area crosses PGD entries, we should calculate the starting address of region which next PGD covers and assign it to next for loop count, but not add PGDIR_SIZE directly. The old code works right only if the mapping area is an exact multiple of PGDIR_SIZE, otherwize the end region could be skipped so that it can't be synchronized to all other processes from kernel PGD init_mm.pgd. In Jeff's system, emulated pmem area [1024G, 1216G) is smaller than PGDIR_SIZE. While 'nokaslr' works because PAGE_OFFSET is 1T aligned, it makes this area be mapped inside one PGD entry. With KASLR enabled, this area could cross two PGD entries, then the next PGD entry won't be synced to all other processes. That is why we saw empty PGD. Fix it. Reported-by: Jeff Moyer <[email protected]> Signed-off-by: Baoquan He <[email protected]> Cc: Andrew Morton <[email protected]> Cc: Andy Lutomirski <[email protected]> Cc: Borislav Petkov <[email protected]> Cc: Brian Gerst <[email protected]> Cc: Dan Williams <[email protected]> Cc: Dave Hansen <[email protected]> Cc: Dave Young <[email protected]> Cc: Denys Vlasenko <[email protected]> Cc: H. Peter Anvin <[email protected]> Cc: Jinbum Park <[email protected]> Cc: Josh Poimboeuf <[email protected]> Cc: Kees Cook <[email protected]> Cc: Kirill A. Shutemov <[email protected]> Cc: Linus Torvalds <[email protected]> Cc: Peter Zijlstra <[email protected]> Cc: Thomas Garnier <[email protected]> Cc: Thomas Gleixner <[email protected]> Cc: Yasuaki Ishimatsu <[email protected]> Cc: Yinghai Lu <[email protected]> Link: http://lkml.kernel.org/r/[email protected] Signed-off-by: Ingo Molnar <[email protected]> Signed-off-by: Dan Williams <[email protected]> Signed-off-by: Greg Kroah-Hartman <[email protected]> --- arch/x86/mm/init_64.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) --- a/arch/x86/mm/init_64.c +++ b/arch/x86/mm/init_64.c @@ -94,10 +94,10 @@ __setup("noexec32=", nonx32_setup); */ void sync_global_pgds(unsigned long start, unsigned long end, int removed) { - unsigned long address; + unsigned long addr; - for (address = start; address <= end; address += PGDIR_SIZE) { - const pgd_t *pgd_ref = pgd_offset_k(address); + for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) { + const pgd_t *pgd_ref = pgd_offset_k(addr); struct page *page; /* @@ -113,7 +113,7 @@ void sync_global_pgds(unsigned long star pgd_t *pgd; spinlock_t *pgt_lock; - pgd = (pgd_t *)page_address(page) + pgd_index(address); + pgd = (pgd_t *)page_address(page) + pgd_index(addr); /* the pgt_lock only for Xen */ pgt_lock = &pgd_page_get_mm(page)->page_table_lock; spin_lock(pgt_lock);
