On 2015/3/30 8:43, Hugh Dickins wrote: > On Sat, 28 Mar 2015, Xishi Qiu wrote: >> On 2015/3/26 21:23, Xishi Qiu wrote: >> >>> Here are two panic logs from smart phone test, and the kernel version is >>> v3.10. >>> >>> log1 is "Unable to handle kernel paging request at virtual address >>> c0704da020", it should be ffffffc0704da020, right? > > That one was an oops at get_ksm_page+0x34/0x150: I'm pretty sure that > comes from the "kpfn = ACCESS_ONCE(stable_node->kpfn)" line, that the > stable_node pointer (in x21 or x22) has upper bits cleared; which > suggests corruption of the rmap_item supposed to point to it. > > get_ksm_page() is tricky with ACCESS_ONCEs against page migration, > and the structures tricky with unions; but pointers overlay pointers > in those unions, I don't see any way we might pick up an address with > the upper 24 or 32 bits cleared due to that. > >>> and log2 is "Unable to handle kernel paging request at virtual address >>> 1e000796", it should be ffffffc01e000796, right? > > And this one was an oops at ksm_scan_thread+0x4ac/0xce0; as is the oops > you posted below. Which contains lots of hex numbers, but very little > info I can work from. > > Please make a CONFIG_DEBUG_INFO=y build of one of the kernels you're > hitting this with, then use the disassembler (objdump -ld perhaps) to > identify precisely which line of ksm.c that is oopsing on: the compiler > will have inlined more interesting functions into ksm_scan_thread, so > I haven't a clue where it's actually oopsing. > > Maybe we'll find that it's also oopsing on a kernel virtual address > from an rmap_item, maybe we won't. > > And I don't read arm64 assembler at all, so I shall be rather limited > in what I can tell you, I'm afraid. > >>> >>> I cann't repeat the panic by test, so could anyone tell me this is the >>> bug of ksm or other reason? > > I've not heard of any problem like this with KSM on other architectures. > Maybe it is making some assumption which is invalid on arm64, but I'd > have thought we'd have heard about that before now. My guess is that > something in your kernel is stamping on KSM's structures. > > A relevant experiment (after identifying the oops line in your current > kernel) might be to switch from CONFIG_SLAB=y to CONFIG_SLUB=y or vice > versa. I doubt SLAB or SLUB is to blame, but changing allocator might > shake things up in a way that either hides the problem, or shifts it > elsewhere. > > Hugh >
Hi Hugh, Thanks for your reply. There are 3 cases as follows, at first I think maybe something causes the oops, but all of the cases are relevant to "rmap_item", so I have no idea. 1. ksm_scan_thread+0xa88/0xce0 -> unstable_tree_search_insert() -> tree_rmap_item = rb_entry(*new, struct rmap_item, node); 2. ksm_scan_thread+0x4ac/0xce0 -> get_next_rmap_item() -> if ((rmap_item->address & PAGE_MASK) == addr) 3. get_ksm_page+0x34/0x150 -> get_ksm_page() -> kpfn = ACCESS_ONCE(stable_node->kpfn); Thanks, Xishi Qiu -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
