Messages in this thread | | | From | Marco Elver <> | Date | Fri, 17 Jan 2020 23:51:48 +0100 | Subject | Re: [PATCH -rcu] kcsan: Make KCSAN compatible with lockdep |
| |
On Fri, 17 Jan 2020 at 17:59, Qian Cai <cai@lca.pw> wrote: > > > > > On Jan 17, 2020, at 11:40 AM, Paul E. McKenney <paulmck@kernel.org> wrote: > > > > True enough, but even if we reach the nirvana state where there is general > > agreement on what constitutes a data race in need of fixing and KCSAN > > faithfully checks based on that data-race definition, we need to handle > > the case where someone introduces a bug that results in a destructive > > off-CPU access to a per-CPU variable, which is exactly the sort of thing > > that KCSAN is supposed to detect. But suppose that this variable is > > frequently referenced from functions that are inlined all over the place. > > > > Then that one bug might result in huge numbers of data-race reports in > > a very short period of time, especially on a large system. > > It sounds like the case with debug_pagealloc where it prints a spam of those, and then the system is just dead. > > [ 28.992752][ T394] Reported by Kernel Concurrency Sanitizer on: > [ 28.992752][ T394] CPU: 0 PID: 394 Comm: pgdatinit0 Not tainted 5.5.0-rc6-next-20200115+ #3 > [ 28.992752][ T394] Hardware name: HP ProLiant XL230a Gen9/ProLiant XL230a Gen9, BIOS U13 01/22/2018 > [ 28.992752][ T394] =============================================================== > [ 28.992752][ T394] ================================================================== > [ 28.992752][ T394] BUG: KCSAN: data-race in __change_page_attr / __change_page_attr > [ 28.992752][ T394] > [ 28.992752][ T394] read to 0xffffffffa01a6de0 of 8 bytes by task 395 on cpu 16: > [ 28.992752][ T394] __change_page_attr+0xe81/0x1620 > [ 28.992752][ T394] __change_page_attr_set_clr+0xde/0x4c0 > [ 28.992752][ T394] __set_pages_np+0xcc/0x100 > [ 28.992752][ T394] __kernel_map_pages+0xd6/0xdb > [ 28.992752][ T394] __free_pages_ok+0x1a8/0x730 > [ 28.992752][ T394] __free_pages+0x51/0x90 > [ 28.992752][ T394] __free_pages_core+0x1c7/0x2c0 > [ 28.992752][ T394] deferred_free_range+0x59/0x8f > [ 28.992752][ T394] deferred_init_max21d > [ 28.992752][ T394] deferred_init_memmap+0x14a/0x1c1 > [ 28.992752][ T394] kthread+0x1e0/0x200 > [ 28.992752][ T394] ret_from_fork+0x3a/0x50 > [ 28.992752][ T394] > [ 28.992752][ T394] write to 0xffffffffa01a6de0 of 8 bytes by task 394 on cpu 0: > [ 28.992752][ T394] __change_page_attr+0xe9c/0x1620 > [ 28.992752][ T394] __change_page_attr_set_clr+0xde/0x4c0 > [ 28.992752][ T394] __set_pages_np+0xcc/0x100 > [ 28.992752][ T394] __kernel_map_pages+0xd6/0xdb > [ 28.992752][ T394] __free_pages_ok+0x1a8/0x730 > [ 28.992752][ T394] __free_pages+0x51/0x90 > [ 28.992752][ T394] __free_pages_core+0x1c7/0x2c0 > [ 28.992752][ T394] deferred_free_range+0x59/0x8f > [ 28.992752][ T394] deferred_init_maxorder+0x1d6/0x21d > [ 28.992752][ T394] deferred_init_memmap+0x14a/0x1c1 > [ 28.992752][ T394] kthread+0x1e0/0x200 > [ 28.992752][ T394] ret_from_fork+0x3a/0x50 > > It point out to this, > > pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr); > pgprot_val(new_prot) |= pgprot_val(cpa->mask_set); > > cpa_inc_4k_install(); > /* Hand in lpsize = 0 to enforce the protection mechanism */ > new_prot = static_protections(new_prot, address, pfn, 1, 0, > CPA_PROTECT); > > In static_protections(), > > /* > * There is no point in checking RW/NX conflicts when the requested > * mapping is setting the page !PRESENT. > */ > if (!(pgprot_val(prot) & _PAGE_PRESENT)) > return prot; > > Is there a data race there?
Yes. I was finally able to reproduce this data race on linux-next (my system doesn't crash though, maybe not enough cores?). Here is a trace with line numbers:
read to 0xffffffffaa59a000 of 8 bytes by interrupt on cpu 7: cpa_inc_4k_install arch/x86/mm/pat/set_memory.c:131 [inline] __change_page_attr+0x10cf/0x1840 arch/x86/mm/pat/set_memory.c:1514 __change_page_attr_set_clr+0xce/0x490 arch/x86/mm/pat/set_memory.c:1636 __set_pages_np+0xc4/0xf0 arch/x86/mm/pat/set_memory.c:2148 __kernel_map_pages+0xb0/0xc8 arch/x86/mm/pat/set_memory.c:2178 kernel_map_pages include/linux/mm.h:2719 [inline] <snip>
write to 0xffffffffaa59a000 of 8 bytes by task 1 on cpu 6: cpa_inc_4k_install arch/x86/mm/pat/set_memory.c:131 [inline] __change_page_attr+0x10ea/0x1840 arch/x86/mm/pat/set_memory.c:1514 __change_page_attr_set_clr+0xce/0x490 arch/x86/mm/pat/set_memory.c:1636 __set_pages_p+0xc4/0xf0 arch/x86/mm/pat/set_memory.c:2129 __kernel_map_pages+0x2e/0xc8 arch/x86/mm/pat/set_memory.c:2176 kernel_map_pages include/linux/mm.h:2719 [inline] <snip>
Both accesses are due to the same "cpa_4k_install++" in cpa_inc_4k_install. Now you can see that a data race here could be potentially undesirable: depending on compiler optimizations or how x86 executes a non-LOCK'd increment, you may lose increments, corrupt the counter, etc. Since this counter only seems to be used for printing some stats, this data race itself is unlikely to cause harm to the system though.
Thanks, -- Marco
| |