Thread exit deadlock #5

jattine · 2020-05-21T14:42:29Z

No description provided.

When CPUs are overcommitted in an LWK partition, the additional threads were being assigned to the CPU of the thread that created the new threads. For example, if 10 CPUs were in the reservation for the process and 100 threads were created by the main thread of the process, there would be 1 thread running on each of the upper 9 CPUs and 91 threads all trying to all run on the first CPU of the reservation. This mOS problem was introduced during a recent rebase in which Linux changed the clone system call code flow. Change-Id: I068dd29ebdb8dc195798a68ef268f5719247e59d Signed-off-by: John Attinella <john.e.attinella@intel.com>

The mOS schduler has a spin lock in its CPU-scoped run-queue object. This spin lock is obtained when we are committing and un-comitting a thread to a specific CPU. All calls to the spin lock must occur when interrupts are disabled. However, in the thread exit path interrupts are enabled. In this exit path we are un-commtting the exiting thread. While we held the spin lock, a scheduler timer tick fired. This fired because we have over-committed threads on this CPU (we disable the timer tick if we are not over-comitted). The timer tick processing drove us through the mOS code to wake up and dispatch another thread. This flow attempts to obtain the spin lock to commit the thread to this run-queue, resulting in a deadlock. The fix is to use a more robust spin-lock interface to lock and unlock which guarantees that interrupts are always disabled while the spin-lock is held. Change-Id: I6b296abc4a78c2b143973a27ae1e0e12b996904f Signed-off-by: John Attinella <john.e.attinella@intel.com>

[ Upstream commit 99d4850 ] Found by leak sanitizer: ``` ==1632594==ERROR: LeakSanitizer: detected memory leaks Direct leak of 21 byte(s) in 1 object(s) allocated from: #0 0x7f2953a7077b in __interceptor_strdup ../../../../src/libsanitizer/asan/asan_interceptors.cpp:439 #1 0x556701d6fbbf in perf_env__read_cpuid util/env.c:369 #2 0x556701d70589 in perf_env__cpuid util/env.c:465 intel#3 0x55670204bba2 in x86__is_amd_cpu arch/x86/util/env.c:14 intel#4 0x5567020487a2 in arch__post_evsel_config arch/x86/util/evsel.c:83 intel#5 0x556701d8f78b in evsel__config util/evsel.c:1366 intel#6 0x556701ef5872 in evlist__config util/record.c:108 intel#7 0x556701cd6bcd in test__PERF_RECORD tests/perf-record.c:112 intel#8 0x556701cacd07 in run_test tests/builtin-test.c:236 intel#9 0x556701cacfac in test_and_print tests/builtin-test.c:265 intel#10 0x556701cadddb in __cmd_test tests/builtin-test.c:402 intel#11 0x556701caf2aa in cmd_test tests/builtin-test.c:559 intel#12 0x556701d3b557 in run_builtin tools/perf/perf.c:323 intel#13 0x556701d3bac8 in handle_internal_command tools/perf/perf.c:377 #14 0x556701d3be90 in run_argv tools/perf/perf.c:421 #15 0x556701d3c3f8 in main tools/perf/perf.c:537 #16 0x7f2952a46189 in __libc_start_call_main ../sysdeps/nptl/libc_start_call_main.h:58 SUMMARY: AddressSanitizer: 21 byte(s) leaked in 1 allocation(s). ``` Fixes: f7b58cb ("perf mem/c2c: Add load store event mappings for AMD") Signed-off-by: Ian Rogers <irogers@google.com> Acked-by: Ravi Bangoria <ravi.bangoria@amd.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ravi Bangoria <ravi.bangoria@amd.com> Link: https://lore.kernel.org/r/20230613235416.1650755-1-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>

[ Upstream commit b684c09 ] ppc_save_regs() skips one stack frame while saving the CPU register states. Instead of saving current R1, it pulls the previous stack frame pointer. When vmcores caused by direct panic call (such as `echo c > /proc/sysrq-trigger`), are debugged with gdb, gdb fails to show the backtrace correctly. On further analysis, it was found that it was because of mismatch between r1 and NIP. GDB uses NIP to get current function symbol and uses corresponding debug info of that function to unwind previous frames, but due to the mismatching r1 and NIP, the unwinding does not work, and it fails to unwind to the 2nd frame and hence does not show the backtrace. GDB backtrace with vmcore of kernel without this patch: --------- (gdb) bt #0 0xc0000000002a53e8 in crash_setup_regs (oldregs=<optimized out>, newregs=0xc000000004f8f8d8) at ./arch/powerpc/include/asm/kexec.h:69 #1 __crash_kexec (regs=<optimized out>) at kernel/kexec_core.c:974 #2 0x0000000000000063 in ?? () intel#3 0xc000000003579320 in ?? () --------- Further analysis revealed that the mismatch occurred because "ppc_save_regs" was saving the previous stack's SP instead of the current r1. This patch fixes this by storing current r1 in the saved pt_regs. GDB backtrace with vmcore of patched kernel: -------- (gdb) bt #0 0xc0000000002a53e8 in crash_setup_regs (oldregs=0x0, newregs=0xc00000000670b8d8) at ./arch/powerpc/include/asm/kexec.h:69 #1 __crash_kexec (regs=regs@entry=0x0) at kernel/kexec_core.c:974 #2 0xc000000000168918 in panic (fmt=fmt@entry=0xc000000001654a60 "sysrq triggered crash\n") at kernel/panic.c:358 intel#3 0xc000000000b735f8 in sysrq_handle_crash (key=<optimized out>) at drivers/tty/sysrq.c:155 intel#4 0xc000000000b742cc in __handle_sysrq (key=key@entry=99, check_mask=check_mask@entry=false) at drivers/tty/sysrq.c:602 intel#5 0xc000000000b7506c in write_sysrq_trigger (file=<optimized out>, buf=<optimized out>, count=2, ppos=<optimized out>) at drivers/tty/sysrq.c:1163 intel#6 0xc00000000069a7bc in pde_write (ppos=<optimized out>, count=<optimized out>, buf=<optimized out>, file=<optimized out>, pde=0xc00000000362cb40) at fs/proc/inode.c:340 intel#7 proc_reg_write (file=<optimized out>, buf=<optimized out>, count=<optimized out>, ppos=<optimized out>) at fs/proc/inode.c:352 intel#8 0xc0000000005b3bbc in vfs_write (file=file@entry=0xc000000006aa6b00, buf=buf@entry=0x61f498b4f60 <error: Cannot access memory at address 0x61f498b4f60>, count=count@entry=2, pos=pos@entry=0xc00000000670bda0) at fs/read_write.c:582 intel#9 0xc0000000005b4264 in ksys_write (fd=<optimized out>, buf=0x61f498b4f60 <error: Cannot access memory at address 0x61f498b4f60>, count=2) at fs/read_write.c:637 intel#10 0xc00000000002ea2c in system_call_exception (regs=0xc00000000670be80, r0=<optimized out>) at arch/powerpc/kernel/syscall.c:171 intel#11 0xc00000000000c270 in system_call_vectored_common () at arch/powerpc/kernel/interrupt_64.S:192 -------- Nick adds: So this now saves regs as though it was an interrupt taken in the caller, at the instruction after the call to ppc_save_regs, whereas previously the NIP was there, but R1 came from the caller's caller and that mismatch is what causes gdb's dwarf unwinder to go haywire. Signed-off-by: Aditya Gupta <adityag@linux.ibm.com> Fixes: d16a58f ("powerpc: Improve ppc_save_regs()") Reivewed-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://msgid.link/20230615091047.90433-1-adityag@linux.ibm.com Signed-off-by: Sasha Levin <sashal@kernel.org>

[ Upstream commit 5fc8cbe ] pr_info() is called with rtp->cbs_gbl_lock spin lock locked. Because pr_info() calls printk() that might sleep, this will result in BUG like below: [ 0.206455] cblist_init_generic: Setting adjustable number of callback queues. [ 0.206463] [ 0.206464] ============================= [ 0.206464] [ BUG: Invalid wait context ] [ 0.206465] 5.19.0-00428-g9de1f9c8ca51 intel#5 Not tainted [ 0.206466] ----------------------------- [ 0.206466] swapper/0/1 is trying to lock: [ 0.206467] ffffffffa0167a58 (&port_lock_key){....}-{3:3}, at: serial8250_console_write+0x327/0x4a0 [ 0.206473] other info that might help us debug this: [ 0.206473] context-{5:5} [ 0.206474] 3 locks held by swapper/0/1: [ 0.206474] #0: ffffffff9eb597e0 (rcu_tasks.cbs_gbl_lock){....}-{2:2}, at: cblist_init_generic.constprop.0+0x14/0x1f0 [ 0.206478] #1: ffffffff9eb579c0 (console_lock){+.+.}-{0:0}, at: _printk+0x63/0x7e [ 0.206482] #2: ffffffff9ea77780 (console_owner){....}-{0:0}, at: console_emit_next_record.constprop.0+0x111/0x330 [ 0.206485] stack backtrace: [ 0.206486] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.19.0-00428-g9de1f9c8ca51 intel#5 [ 0.206488] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.0-1.fc36 04/01/2014 [ 0.206489] Call Trace: [ 0.206490] <TASK> [ 0.206491] dump_stack_lvl+0x6a/0x9f [ 0.206493] __lock_acquire.cold+0x2d7/0x2fe [ 0.206496] ? stack_trace_save+0x46/0x70 [ 0.206497] lock_acquire+0xd1/0x2f0 [ 0.206499] ? serial8250_console_write+0x327/0x4a0 [ 0.206500] ? __lock_acquire+0x5c7/0x2720 [ 0.206502] _raw_spin_lock_irqsave+0x3d/0x90 [ 0.206504] ? serial8250_console_write+0x327/0x4a0 [ 0.206506] serial8250_console_write+0x327/0x4a0 [ 0.206508] console_emit_next_record.constprop.0+0x180/0x330 [ 0.206511] console_unlock+0xf7/0x1f0 [ 0.206512] vprintk_emit+0xf7/0x330 [ 0.206514] _printk+0x63/0x7e [ 0.206516] cblist_init_generic.constprop.0.cold+0x24/0x32 [ 0.206518] rcu_init_tasks_generic+0x5/0xd9 [ 0.206522] kernel_init_freeable+0x15b/0x2a2 [ 0.206523] ? rest_init+0x160/0x160 [ 0.206526] kernel_init+0x11/0x120 [ 0.206527] ret_from_fork+0x1f/0x30 [ 0.206530] </TASK> [ 0.207018] cblist_init_generic: Setting shift to 1 and lim to 1. This patch moves pr_info() so that it is called without rtp->cbs_gbl_lock locked. Signed-off-by: Shigeru Yoshida <syoshida@redhat.com> Tested-by: "Zhang, Qiang1" <qiang1.zhang@intel.com> Signed-off-by: Paul E. McKenney <paulmck@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>

[ Upstream commit 93a3319 ] The cited commit holds encap tbl lock unconditionally when setting up dests. But it may cause the following deadlock: PID: 1063722 TASK: ffffa062ca5d0000 CPU: 13 COMMAND: "handler8" #0 [ffffb14de05b7368] __schedule at ffffffffa1d5aa91 #1 [ffffb14de05b7410] schedule at ffffffffa1d5afdb #2 [ffffb14de05b7430] schedule_preempt_disabled at ffffffffa1d5b528 intel#3 [ffffb14de05b7440] __mutex_lock at ffffffffa1d5d6cb intel#4 [ffffb14de05b74e8] mutex_lock_nested at ffffffffa1d5ddeb intel#5 [ffffb14de05b74f8] mlx5e_tc_tun_encap_dests_set at ffffffffc12f2096 [mlx5_core] intel#6 [ffffb14de05b7568] post_process_attr at ffffffffc12d9fc5 [mlx5_core] intel#7 [ffffb14de05b75a0] mlx5e_tc_add_fdb_flow at ffffffffc12de877 [mlx5_core] intel#8 [ffffb14de05b75f0] __mlx5e_add_fdb_flow at ffffffffc12e0eef [mlx5_core] intel#9 [ffffb14de05b7660] mlx5e_tc_add_flow at ffffffffc12e12f7 [mlx5_core] intel#10 [ffffb14de05b76b8] mlx5e_configure_flower at ffffffffc12e1686 [mlx5_core] intel#11 [ffffb14de05b7720] mlx5e_rep_indr_offload at ffffffffc12e3817 [mlx5_core] intel#12 [ffffb14de05b7730] mlx5e_rep_indr_setup_tc_cb at ffffffffc12e388a [mlx5_core] intel#13 [ffffb14de05b7740] tc_setup_cb_add at ffffffffa1ab2ba8 #14 [ffffb14de05b77a0] fl_hw_replace_filter at ffffffffc0bdec2f [cls_flower] #15 [ffffb14de05b7868] fl_change at ffffffffc0be6caa [cls_flower] #16 [ffffb14de05b7908] tc_new_tfilter at ffffffffa1ab71f0 [1031218.028143] wait_for_completion+0x24/0x30 [1031218.028589] mlx5e_update_route_decap_flows+0x9a/0x1e0 [mlx5_core] [1031218.029256] mlx5e_tc_fib_event_work+0x1ad/0x300 [mlx5_core] [1031218.029885] process_one_work+0x24e/0x510 Actually no need to hold encap tbl lock if there is no encap action. Fix it by checking if encap action exists or not before holding encap tbl lock. Fixes: 37c3b9f ("net/mlx5e: Prevent encap offload when neigh update is running") Signed-off-by: Chris Mi <cmi@nvidia.com> Reviewed-by: Vlad Buslov <vladbu@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com> Signed-off-by: Sasha Levin <sashal@kernel.org>

jattine added 2 commits May 8, 2020 14:58

attaufer merged commit 32e3295 into 5.4.18_0.8.mos Jun 1, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Thread exit deadlock #5

Thread exit deadlock #5

jattine commented May 21, 2020

Thread exit deadlock #5

Thread exit deadlock #5

Conversation

jattine commented May 21, 2020