kernel/tools/perf/util/bpf-event.c, branch linux-5.2.y

perf bpf: Show more BPF program info in print_bpf_prog_info()

2019-03-21T14:27:04Z

This patch enables showing bpf program name, address, and size in the header. Before the patch: perf report --header-only ... # bpf_prog_info of id 9 # bpf_prog_info of id 10 # bpf_prog_info of id 13 After the patch: # bpf_prog_info 9: bpf_prog_7be49e3934a125ba addr 0xffffffffa0024947 size 229 # bpf_prog_info 10: bpf_prog_2a142ef67aaad174 addr 0xffffffffa007c94d size 229 # bpf_prog_info 13: bpf_prog_47368425825d7384_task__task_newt addr 0xffffffffa0251137 size 369 Committer notes: Fix the fallback definition when HAVE_LIBBPF_SUPPORT is not defined, i.e. add the missing 'static inline' and add the __maybe_unused to the args. Also add stdio.h since we now use FILE * in bpf-event.h. Signed-off-by: Song Liu Tested-by: Arnaldo Carvalho de Melo Cc: Alexei Starovoitov Cc: Daniel Borkmann Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stanislav Fomichev Link: http://lkml.kernel.org/r/20190319165454.1298742-3-songliubraving@fb.com Signed-off-by: Arnaldo Carvalho de Melo

perf bpf: Extract logic to create program names from perf_event__synthesize_one_bpf_prog()

2019-03-21T14:27:04Z

Extract logic to create program names to synthesize_bpf_prog_name(), so that it can be reused in header.c:print_bpf_prog_info(). This commit doesn't change the behavior. Signed-off-by: Song Liu Cc: Alexei Starovoitov Cc: Daniel Borkmann Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stanislav Fomichev Link: http://lkml.kernel.org/r/20190319165454.1298742-2-songliubraving@fb.com Signed-off-by: Arnaldo Carvalho de Melo

perf tools: Save bpf_prog_info and BTF of new BPF programs

2019-03-21T14:27:04Z

To fully annotate BPF programs with source code mapping, 4 different information are needed: 1) PERF_RECORD_KSYMBOL 2) PERF_RECORD_BPF_EVENT 3) bpf_prog_info 4) btf This patch handles 3) and 4) for BPF programs loaded after 'perf record|top'. For timely process of these information, a dedicated event is added to the side band evlist. When PERF_RECORD_BPF_EVENT is received via the side band event, the polling thread gathers 3) and 4) vis sys_bpf and store them in perf_env. This information is saved to perf.data at the end of 'perf record'. Committer testing: The 'wakeup_watermark' member in 'struct perf_event_attr' is inside a unnamed union, so can't be used in a struct designated initialization with older gccs, get it out of that, isolating as 'attr.wakeup_watermark = 1;' to work with all gcc versions. We also need to add '--no-bpf-event' to the 'perf record' perf_event_attr tests in 'perf test', as the way that that test goes is to intercept the events being setup and looking if they match the fields described in the control files, since now it finds first the side band event used to catch the PERF_RECORD_BPF_EVENT, they all fail. With these issues fixed: Same scenario as for testing BPF programs loaded before 'perf record' or 'perf top' starts, only start the BPF programs after 'perf record|top', so that its information get collected by the sideband threads, the rest works as for the programs loaded before start monitoring. Add missing 'inline' to the bpf_event__add_sb_event() when HAVE_LIBBPF_SUPPORT is not defined, fixing the build in systems without binutils devel files installed. Signed-off-by: Song Liu Reviewed-by: Jiri Olsa Tested-by: Arnaldo Carvalho de Melo Cc: Alexei Starovoitov Cc: Daniel Borkmann Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stanislav Fomichev Link: http://lkml.kernel.org/r/20190312053051.2690567-16-songliubraving@fb.com Signed-off-by: Arnaldo Carvalho de Melo

perf annotate: Enable annotation of BPF programs

2019-03-20T19:43:15Z

In symbol__disassemble(), DSO_BINARY_TYPE__BPF_PROG_INFO dso calls into a new function symbol__disassemble_bpf(), where annotation line information is filled based on the bpf_prog_info and btf data saved in given perf_env. symbol__disassemble_bpf() uses binutils's libopcodes to disassemble bpf programs. Committer testing: After fixing this: - u64 *addrs = (u64 *)(info_linear->info.jited_ksyms); + u64 *addrs = (u64 *)(uintptr_t)(info_linear->info.jited_ksyms); Detected when crossbuilding to a 32-bit arch. And making all this dependent on HAVE_LIBBFD_SUPPORT and HAVE_LIBBPF_SUPPORT: 1) Have a BPF program running, one that has BTF info, etc, I used the tools/perf/examples/bpf/augmented_raw_syscalls.c put in place by 'perf trace'. # grep -B1 augmented_raw ~/.perfconfig [trace] add_events = /home/acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.c # # perf trace -e *mmsg dnf/6245 sendmmsg(20, 0x7f5485a88030, 2, MSG_NOSIGNAL) = 2 NetworkManager/10055 sendmmsg(22, 0x7f8126ad1bb0, 2, MSG_NOSIGNAL) = 2 2) Then do a 'perf record' system wide for a while: # perf record -a ^C[ perf record: Woken up 68 times to write data ] [ perf record: Captured and wrote 19.427 MB perf.data (366891 samples) ] # 3) Check that we captured BPF and BTF info in the perf.data file: # perf report --header-only | grep 'b[pt]f' # event : name = cycles:ppp, , id = { 294789, 294790, 294791, 294792, 294793, 294794, 294795, 294796 }, size = 112, { sample_period, sample_freq } = 4000, sample_type = IP|TID|TIME|CPU|PERIOD, read_format = ID, disabled = 1, inherit = 1, mmap = 1, comm = 1, freq = 1, task = 1, precise_ip = 3, sample_id_all = 1, exclude_guest = 1, mmap2 = 1, comm_exec = 1, ksymbol = 1, bpf_event = 1 # bpf_prog_info of id 13 # bpf_prog_info of id 14 # bpf_prog_info of id 15 # bpf_prog_info of id 16 # bpf_prog_info of id 17 # bpf_prog_info of id 18 # bpf_prog_info of id 21 # bpf_prog_info of id 22 # bpf_prog_info of id 41 # bpf_prog_info of id 42 # btf info of id 2 # 4) Check which programs got recorded: # perf report | grep bpf_prog | head 0.16% exe bpf_prog_819967866022f1e1_sys_enter [k] bpf_prog_819967866022f1e1_sys_enter 0.14% exe bpf_prog_c1bd85c092d6e4aa_sys_exit [k] bpf_prog_c1bd85c092d6e4aa_sys_exit 0.08% fuse-overlayfs bpf_prog_819967866022f1e1_sys_enter [k] bpf_prog_819967866022f1e1_sys_enter 0.07% fuse-overlayfs bpf_prog_c1bd85c092d6e4aa_sys_exit [k] bpf_prog_c1bd85c092d6e4aa_sys_exit 0.01% clang-4.0 bpf_prog_c1bd85c092d6e4aa_sys_exit [k] bpf_prog_c1bd85c092d6e4aa_sys_exit 0.01% clang-4.0 bpf_prog_819967866022f1e1_sys_enter [k] bpf_prog_819967866022f1e1_sys_enter 0.00% clang bpf_prog_c1bd85c092d6e4aa_sys_exit [k] bpf_prog_c1bd85c092d6e4aa_sys_exit 0.00% runc bpf_prog_819967866022f1e1_sys_enter [k] bpf_prog_819967866022f1e1_sys_enter 0.00% clang bpf_prog_819967866022f1e1_sys_enter [k] bpf_prog_819967866022f1e1_sys_enter 0.00% sh bpf_prog_c1bd85c092d6e4aa_sys_exit [k] bpf_prog_c1bd85c092d6e4aa_sys_exit # This was with the default --sort order for 'perf report', which is: --sort comm,dso,symbol If we just look for the symbol, for instance: # perf report --sort symbol | grep bpf_prog | head 0.26% [k] bpf_prog_819967866022f1e1_sys_enter - - 0.24% [k] bpf_prog_c1bd85c092d6e4aa_sys_exit - - # or the DSO: # perf report --sort dso | grep bpf_prog | head 0.26% bpf_prog_819967866022f1e1_sys_enter 0.24% bpf_prog_c1bd85c092d6e4aa_sys_exit # We'll see the two BPF programs that augmented_raw_syscalls.o puts in place, one attached to the raw_syscalls:sys_enter and another to the raw_syscalls:sys_exit tracepoints, as expected. Now we can finally do, from the command line, annotation for one of those two symbols, with the original BPF program source coude intermixed with the disassembled JITed code: # perf annotate --stdio2 bpf_prog_819967866022f1e1_sys_enter Samples: 950 of event 'cycles:ppp', 4000 Hz, Event count (approx.): 553756947, [percent: local period] bpf_prog_819967866022f1e1_sys_enter() bpf_prog_819967866022f1e1_sys_enter Percent int sys_enter(struct syscall_enter_args *args) 53.41 push %rbp 0.63 mov %rsp,%rbp 0.31 sub $0x170,%rsp 1.93 sub $0x28,%rbp 7.02 mov %rbx,0x0(%rbp) 3.20 mov %r13,0x8(%rbp) 1.07 mov %r14,0x10(%rbp) 0.61 mov %r15,0x18(%rbp) 0.11 xor %eax,%eax 1.29 mov %rax,0x20(%rbp) 0.11 mov %rdi,%rbx return bpf_get_current_pid_tgid(); 2.02 → callq *ffffffffda6776d9 2.76 mov %eax,-0x148(%rbp) mov %rbp,%rsi int sys_enter(struct syscall_enter_args *args) add $0xfffffffffffffeb8,%rsi return bpf_map_lookup_elem(pids, &pid) != NULL; movabs $0xffff975ac2607800,%rdi 1.26 → callq *ffffffffda6789e9 cmp $0x0,%rax 2.43 → je 0 add $0x38,%rax 0.21 xor %r13d,%r13d if (pid_filter__has(&pids_filtered, getpid())) 0.81 cmp $0x0,%rax → jne 0 mov %rbp,%rdi probe_read(&augmented_args.args, sizeof(augmented_args.args), args); 2.22 add $0xfffffffffffffeb8,%rdi 0.11 mov $0x40,%esi 0.32 mov %rbx,%rdx 2.74 → callq *ffffffffda658409 syscall = bpf_map_lookup_elem(&syscalls, &augmented_args.args.syscall_nr); 0.22 mov %rbp,%rsi 1.69 add $0xfffffffffffffec0,%rsi syscall = bpf_map_lookup_elem(&syscalls, &augmented_args.args.syscall_nr); movabs $0xffff975bfcd36000,%rdi add $0xd0,%rdi 0.21 mov 0x0(%rsi),%eax 0.93 cmp $0x200,%rax → jae 0 0.10 shl $0x3,%rax 0.11 add %rdi,%rax 0.11 → jmp 0 xor %eax,%eax if (syscall == NULL || !syscall->enabled) 1.07 cmp $0x0,%rax → je 0 if (syscall == NULL || !syscall->enabled) 6.57 movzbq 0x0(%rax),%rdi if (syscall == NULL || !syscall->enabled) cmp $0x0,%rdi 0.95 → je 0 mov $0x40,%r8d switch (augmented_args.args.syscall_nr) { mov -0x140(%rbp),%rdi switch (augmented_args.args.syscall_nr) { cmp $0x2,%rdi → je 0 cmp $0x101,%rdi → je 0 cmp $0x15,%rdi → jne 0 case SYS_OPEN: filename_arg = (const void *)args->args[0]; mov 0x10(%rbx),%rdx → jmp 0 case SYS_OPENAT: filename_arg = (const void *)args->args[1]; mov 0x18(%rbx),%rdx if (filename_arg != NULL) { cmp $0x0,%rdx → je 0 xor %edi,%edi augmented_args.filename.reserved = 0; mov %edi,-0x104(%rbp) augmented_args.filename.size = probe_read_str(&augmented_args.filename.value, mov %rbp,%rdi add $0xffffffffffffff00,%rdi augmented_args.filename.size = probe_read_str(&augmented_args.filename.value, mov $0x100,%esi → callq *ffffffffda658499 mov $0x148,%r8d augmented_args.filename.size = probe_read_str(&augmented_args.filename.value, mov %eax,-0x108(%rbp) augmented_args.filename.size = probe_read_str(&augmented_args.filename.value, mov %rax,%rdi shl $0x20,%rdi shr $0x20,%rdi if (augmented_args.filename.size < sizeof(augmented_args.filename.value)) { cmp $0xff,%rdi → ja 0 len -= sizeof(augmented_args.filename.value) - augmented_args.filename.size; add $0x48,%rax len &= sizeof(augmented_args.filename.value) - 1; and $0xff,%rax mov %rax,%r8 mov %rbp,%rcx return perf_event_output(args, &__augmented_syscalls__, BPF_F_CURRENT_CPU, &augmented_args, len); add $0xfffffffffffffeb8,%rcx mov %rbx,%rdi movabs $0xffff975fbd72d800,%rsi mov $0xffffffff,%edx → callq *ffffffffda658ad9 mov %rax,%r13 } mov %r13,%rax 0.72 mov 0x0(%rbp),%rbx mov 0x8(%rbp),%r13 1.16 mov 0x10(%rbp),%r14 0.10 mov 0x18(%rbp),%r15 0.42 add $0x28,%rbp 0.54 leaveq 0.54 ← retq # Please see 'man perf-config' to see how to control what should be seen, via ~/.perfconfig [annotate] section, for instance, one can suppress the source code and see just the disassembly, etc. Alternatively, use the TUI bu just using 'perf annotate', press '/bpf_prog' to see the bpf symbols, press enter and do the interactive annotation, which allows for dumping to a file after selecting the the various output tunables, for instance, the above without source code intermixed, plus showing all the instruction offsets: # perf annotate bpf_prog_819967866022f1e1_sys_enter Then press: 's' to hide the source code + 'O' twice to show all instruction offsets, then 'P' to print to the bpf_prog_819967866022f1e1_sys_enter.annotation file, which will have: # cat bpf_prog_819967866022f1e1_sys_enter.annotation bpf_prog_819967866022f1e1_sys_enter() bpf_prog_819967866022f1e1_sys_enter Event: cycles:ppp 53.41 0: push %rbp 0.63 1: mov %rsp,%rbp 0.31 4: sub $0x170,%rsp 1.93 b: sub $0x28,%rbp 7.02 f: mov %rbx,0x0(%rbp) 3.20 13: mov %r13,0x8(%rbp) 1.07 17: mov %r14,0x10(%rbp) 0.61 1b: mov %r15,0x18(%rbp) 0.11 1f: xor %eax,%eax 1.29 21: mov %rax,0x20(%rbp) 0.11 25: mov %rdi,%rbx 2.02 28: → callq *ffffffffda6776d9 2.76 2d: mov %eax,-0x148(%rbp) 33: mov %rbp,%rsi 36: add $0xfffffffffffffeb8,%rsi 3d: movabs $0xffff975ac2607800,%rdi 1.26 47: → callq *ffffffffda6789e9 4c: cmp $0x0,%rax 2.43 50: → je 0 52: add $0x38,%rax 0.21 56: xor %r13d,%r13d 0.81 59: cmp $0x0,%rax 5d: → jne 0 63: mov %rbp,%rdi 2.22 66: add $0xfffffffffffffeb8,%rdi 0.11 6d: mov $0x40,%esi 0.32 72: mov %rbx,%rdx 2.74 75: → callq *ffffffffda658409 0.22 7a: mov %rbp,%rsi 1.69 7d: add $0xfffffffffffffec0,%rsi 84: movabs $0xffff975bfcd36000,%rdi 8e: add $0xd0,%rdi 0.21 95: mov 0x0(%rsi),%eax 0.93 98: cmp $0x200,%rax 9f: → jae 0 0.10 a1: shl $0x3,%rax 0.11 a5: add %rdi,%rax 0.11 a8: → jmp 0 aa: xor %eax,%eax 1.07 ac: cmp $0x0,%rax b0: → je 0 6.57 b6: movzbq 0x0(%rax),%rdi bb: cmp $0x0,%rdi 0.95 bf: → je 0 c5: mov $0x40,%r8d cb: mov -0x140(%rbp),%rdi d2: cmp $0x2,%rdi d6: → je 0 d8: cmp $0x101,%rdi df: → je 0 e1: cmp $0x15,%rdi e5: → jne 0 e7: mov 0x10(%rbx),%rdx eb: → jmp 0 ed: mov 0x18(%rbx),%rdx f1: cmp $0x0,%rdx f5: → je 0 f7: xor %edi,%edi f9: mov %edi,-0x104(%rbp) ff: mov %rbp,%rdi 102: add $0xffffffffffffff00,%rdi 109: mov $0x100,%esi 10e: → callq *ffffffffda658499 113: mov $0x148,%r8d 119: mov %eax,-0x108(%rbp) 11f: mov %rax,%rdi 122: shl $0x20,%rdi 126: shr $0x20,%rdi 12a: cmp $0xff,%rdi 131: → ja 0 133: add $0x48,%rax 137: and $0xff,%rax 13d: mov %rax,%r8 140: mov %rbp,%rcx 143: add $0xfffffffffffffeb8,%rcx 14a: mov %rbx,%rdi 14d: movabs $0xffff975fbd72d800,%rsi 157: mov $0xffffffff,%edx 15c: → callq *ffffffffda658ad9 161: mov %rax,%r13 164: mov %r13,%rax 0.72 167: mov 0x0(%rbp),%rbx 16b: mov 0x8(%rbp),%r13 1.16 16f: mov 0x10(%rbp),%r14 0.10 173: mov 0x18(%rbp),%r15 0.42 177: add $0x28,%rbp 0.54 17b: leaveq 0.54 17c: ← retq Another cool way to test all this is to symple use 'perf top' look for those symbols, go there and press enter, annotate it live :-) Signed-off-by: Song Liu Reviewed-by: Jiri Olsa Tested-by: Arnaldo Carvalho de Melo Cc: Alexei Starovoitov Cc: Daniel Borkmann Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stanislav Fomichev Link: http://lkml.kernel.org/r/20190312053051.2690567-13-songliubraving@fb.com Signed-off-by: Arnaldo Carvalho de Melo

perf bpf: Process PERF_BPF_EVENT_PROG_LOAD for annotation

2019-03-19T19:52:07Z

This patch adds processing of PERF_BPF_EVENT_PROG_LOAD, which sets proper DSO type/id/etc of memory regions mapped to BPF programs to DSO_BINARY_TYPE__BPF_PROG_INFO. Signed-off-by: Song Liu Reviewed-by: Jiri Olsa Cc: Alexei Starovoitov Cc: Daniel Borkmann Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stanislav Fomichev Cc: kernel-team@fb.com Link: http://lkml.kernel.org/r/20190312053051.2690567-14-songliubraving@fb.com Signed-off-by: Arnaldo Carvalho de Melo

perf bpf: Save BTF in a rbtree in perf_env

2019-03-19T19:52:07Z

BTF contains information necessary to annotate BPF programs. This patch saves BTF for BPF programs loaded in the system. Signed-off-by: Song Liu Reviewed-by: Jiri Olsa Cc: Alexei Starovoitov Cc: Daniel Borkmann Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stanislav Fomichev Cc: kernel-team@fb.com Link: http://lkml.kernel.org/r/20190312053051.2690567-9-songliubraving@fb.com Signed-off-by: Arnaldo Carvalho de Melo

perf bpf: Save bpf_prog_info in a rbtree in perf_env

2019-03-19T19:52:06Z

bpf_prog_info contains information necessary to annotate bpf programs. This patch saves bpf_prog_info for bpf programs loaded in the system. Some big picture of the next few patches: To fully annotate BPF programs with source code mapping, 4 different informations are needed: 1) PERF_RECORD_KSYMBOL 2) PERF_RECORD_BPF_EVENT 3) bpf_prog_info 4) btf Before this set, 1) and 2) in the list are already saved to perf.data file. For BPF programs that are already loaded before perf run, 1) and 2) are synthesized by perf_event__synthesize_bpf_events(). For short living BPF programs, 1) and 2) are generated by kernel. This set handles 3) and 4) from the list. Again, it is necessary to handle existing BPF program and short living program separately. This patch handles 3) for exising BPF programs while synthesizing 1) and 2) in perf_event__synthesize_bpf_events(). These data are stored in perf_env. The next patch saves these data from perf_env to perf.data as headers. Similarly, the two patches after the next saves 4) of existing BPF programs to perf_env and perf.data. Another patch later will handle 3) and 4) for short living BPF programs by monitoring 1) and 2) in a dedicate thread. Signed-off-by: Song Liu Reviewed-by: Jiri Olsa Cc: Alexei Starovoitov Cc: Daniel Borkmann Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stanislav Fomichev Cc: kernel-team@fb.com Link: http://lkml.kernel.org/r/20190312053051.2690567-7-songliubraving@fb.com [ set env->bpf_progs.infos_cnt to zero in perf_env__purge_bpf() as noted by jolsa ] Signed-off-by: Arnaldo Carvalho de Melo

perf bpf: Make synthesize_bpf_events() receive perf_session pointer instead of perf_tool

2019-03-19T19:52:06Z

This patch changes the arguments of perf_event__synthesize_bpf_events() to include perf_session* instead of perf_tool*. perf_session will be used in the next patch. Signed-off-by: Song Liu Reviewed-by: Jiri Olsa Cc: Alexei Starovoitov Cc: Daniel Borkmann Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stanislav Fomichev Cc: kernel-team@fb.com Link: http://lkml.kernel.org/r/20190312053051.2690567-6-songliubraving@fb.com Signed-off-by: Arnaldo Carvalho de Melo

perf bpf: Synthesize bpf events with bpf_program__get_prog_info_linear()

2019-03-19T19:52:06Z

With bpf_program__get_prog_info_linear, we can simplify the logic that synthesizes bpf events. This patch doesn't change the behavior of the code. Commiter notes: Needed this (for all four variables), suggested by Song, to overcome build failure on debian experimental cross building to MIPS 32-bit: - u8 (*prog_tags)[BPF_TAG_SIZE] = (void *)(info->prog_tags); + u8 (*prog_tags)[BPF_TAG_SIZE] = (void *)(uintptr_t)(info->prog_tags); util/bpf-event.c: In function 'perf_event__synthesize_one_bpf_prog': util/bpf-event.c:143:35: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast] u8 (*prog_tags)[BPF_TAG_SIZE] = (void *)(info->prog_tags); ^ util/bpf-event.c:144:22: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast] __u32 *prog_lens = (__u32 *)(info->jited_func_lens); ^ util/bpf-event.c:145:23: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast] __u64 *prog_addrs = (__u64 *)(info->jited_ksyms); ^ util/bpf-event.c:146:22: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast] void *func_infos = (void *)(info->func_info); ^ cc1: all warnings being treated as errors Signed-off-by: Song Liu Reviewed-by: Jiri Olsa Cc: Alexei Starovoitov Cc: Daniel Borkmann Cc: kernel-team@fb.com Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stanislav Fomichev Link: http://lkml.kernel.org/r/20190312053051.2690567-5-songliubraving@fb.com Signed-off-by: Arnaldo Carvalho de Melo

perf record: Replace option --bpf-event with --no-bpf-event

2019-03-19T19:52:06Z

Currently, monitoring of BPF programs through bpf_event is off by default for 'perf record'. To turn it on, the user need to use option "--bpf-event". As BPF gets wider adoption in different subsystems, this option becomes inconvenient. This patch makes bpf_event on by default, and adds option "--no-bpf-event" to turn it off. Since option --bpf-event is not released yet, it is safe to remove it. Signed-off-by: Song Liu Reviewed-by: Jiri Olsa Cc: Alexei Starovoitov Cc: Daniel Borkmann Cc: kernel-team@fb.com Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stanislav Fomichev Link: http://lkml.kernel.org/r/20190312053051.2690567-2-songliubraving@fb.com Signed-off-by: Arnaldo Carvalho de Melo