- 28 Aug, 2024 8 commits
-
-
Veronika Molnarova authored
Shellcheck is becoming a standard when building perf to prevent any unnecessary mistakes. Fix shellcheck warnings in perf testsuite. Signed-off-by:
Veronika Molnarova <vmolnaro@redhat.com> Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com> Cc: Ian Rogers <irogers@google.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/20240702110849.31904-5-vmolnaro@redhat.comSigned-off-by:
Michael Petlan <mpetlan@redhat.com> Signed-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Veronika Molnarova authored
Merge perf testsuite setting files into common settings to reduce duplicates and prevent errors. Signed-off-by:
-
Michael Petlan authored
The test scripts in base_* directories currently have their own drivers that run them. Before this patch, the shell test-suite generator causes them to run twice. Fix that by skipping them in the generator. A cleaner solution (for future) will be to use the directory structure idea (introduced by Carsten Haitzler in 7391db64 ("perf test: Refactor shell tests allowing subdirs")) to generate test entries with subtests, like: $ perf test list [...] 97: perf probe shell tests 97:1: perf probe basic functionality 97:2: perf probe tests with arguments 97:3: perf probe invalid options handling [...] There is already a lot of shell test scripts and many are about to come, so there is a need for some hierarchy. Signed-off-by:
-
Arnaldo Carvalho de Melo authored
The getname_flags() routine changed recently and thus the place where we were getting the pathname is not probeable anymore, albeit still present, so use the next line for that, before: root@number:/home/acme/git/perf-tools-next# perf test vfs_getname 91: Add vfs_getname probe to get syscall args filenames : FAILED! 93: Use vfs_getname probe to get syscall args filenames : FAILED! 126: Check open filename arg using perf trace + vfs_getname : FAILED! root@number:/home/acme/git/perf-tools-next# Now tests 91 and 126 are passing, some more investigation is needed for test 93, that continues to fail. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Namhyung Kim <namhyung@kernel.org> Signed-off-by: -
Namhyung Kim authored
Add Multiple bpf-filter test for two or more events with filters. It uses task-clock and page-faults events with different filter expressions and check the perf script output $ sudo ./perf test filtering -vv 96: perf record sample filtering (by BPF) tests: --- start --- test child forked, pid 2804025 Checking BPF-filter privilege Basic bpf-filter test Basic bpf-filter test [Success] Failing bpf-filter test Error: task-clock event does not have PERF_SAMPLE_CPU Failing bpf-filter test [Success] Group bpf-filter test Error: task-clock event does not have PERF_SAMPLE_CPU Error: task-clock event does not have PERF_SAMPLE_CODE_PAGE_SIZE Group bpf-filter test [Success] Multiple bpf-filter test Multiple bpf-filter test [Success] ---- end(0) ---- 96: perf record sample filtering (by BPF) tests : Ok Signed-off-by:
Namhyung Kim <namhyung@kernel.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@kernel.org> Cc: KP Singh <kpsingh@kernel.org> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Song Liu <song@kernel.org> Link: https://lore.kernel.org/r/20240820154504.128923-3-namhyung@kernel.orgSigned-off-by:
-
Namhyung Kim authored
Print the actual dropped sample count in the event stat. $ sudo perf record -o- -e cycles --filter 'period < 10000' \ -e instructions --filter 'ip > 0x8000000000000000' perf test -w noploop | \ perf report --stat -i- [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.058 MB - ] Aggregated stats: TOTAL events: 469 MMAP events: 268 (57.1%) COMM events: 2 ( 0.4%) EXIT events: 1 ( 0.2%) SAMPLE events: 16 ( 3.4%) MMAP2 events: 22 ( 4.7%) LOST_SAMPLES events: 2 ( 0.4%) KSYMBOL events: 89 (19.0%) BPF_EVENT events: 39 ( 8.3%) ATTR events: 2 ( 0.4%) FINISHED_ROUND events: 1 ( 0.2%) ID_INDEX events: 1 ( 0.2%) THREAD_MAP events: 1 ( 0.2%) CPU_MAP events: 1 ( 0.2%) EVENT_UPDATE events: 2 ( 0.4%) TIME_CONV events: 1 ( 0.2%) FEATURE events: 20 ( 4.3%) FINISHED_INIT events: 1 ( 0.2%) cycles stats: SAMPLE events: 2 LOST_SAMPLES (BPF) events: 4010 instructions stats: SAMPLE events: 14 LOST_SAMPLES (BPF) events: 3990 Signed-off-by: -
Namhyung Kim authored
So far it used tgid as a key to get the filter expressions in the pinned filters map for regular users but it won't work well if the has more than one filters at the same time. Let's add the event id to the key of the filter hash map so that it can identify the right filter expression in the BPF program. As the event can be inherited to child tasks, it should use the primary id which belongs to the parent (original) event. Since evsel opens the event for multiple CPUs and tasks, it needs to maintain a separate hash map for the event id. In the user space, it keeps a list for the multiple evsel and release the entries in the both hash map when it closes the event. Signed-off-by:
-
Kan Liang authored
Perf crashes as below when applying --no-group # perf record -e "{cache-misses,branches"} -b sleep 1 # perf report --stdio --no-group free(): invalid next size (fast) Aborted (core dumped) # In the __hpp__fmt(), only 1 hpp_fmt_value is allocated for the current event when --no-group is applied. However, the current implementation tries to assign the hists from all members to the hpp_fmt_value, which exceeds the allocated memory. Fixes: 8f6071a3 ("perf hist: Simplify __hpp_fmt() using hpp_fmt_data") Signed-off-by:Kan Liang <kan.liang@linux.intel.com> Acked-by:
-
- 26 Aug, 2024 1 commit
-
-
Andi Kleen authored
Extend the searching for the test files so that it works when running perf from a separate objdir, and also when the perf executable is symlinked. Signed-off-by:
Andi Kleen <ak@linux.intel.com> Acked-by:
-
- 22 Aug, 2024 6 commits
-
-
Arnaldo Carvalho de Melo authored
The -Wcast-function-type-mismatch option was introduced in clang 19 and its enabled by default, since we use -Werror, and python bindings do casts that are valid but trips this warning, disable it if present. Closes: https://lore.kernel.org/all/CA+icZUXoJ6BS3GMhJHV3aZWyb5Cz2haFneX0C5pUMUUhG-UVKQ@mail.gmail.comReported-by:
Sedat Dilek <sedat.dilek@gmail.com> Tested-by:
-
Arnaldo Carvalho de Melo authored
We'll need to check if an warning option introduced in clang 19 is available on the clang version being used, so cover the error message emitted when testing for a -W option. Tested-by:
-
Namhyung Kim authored
In some cases, compilers don't set the location expression in DWARF precisely. For instance, it may assign a variable to a register after copying it from a different register. Then it should use the register for the new type but still uses the old register. This makes hard to track the type information properly. This is an example I found in __tcp_transmit_skb(). The first argument (sk) of this function is a pointer to sock and there's a variable (tp) for tcp_sock. static int __tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it, gfp_t gfp_mask, u32 rcv_nxt) { ... struct tcp_sock *tp; BUG_ON(!skb || !tcp_skb_pcount(skb)); tp = tcp_sk(sk); prior_wstamp = tp->tcp_wstamp_ns; tp->tcp_wstamp_ns = max(tp->tcp_wstamp_ns, tp->tcp_clock_cache); ... So it basically calls tcp_sk(sk) to get the tcp_sock pointer from sk. But it turned out to be the same value because tcp_sock embeds sock as the first member. The sk is located in reg5 (RDI) and tp is in reg3 (RBX). The offset of tcp_wstamp_ns is 0x748 and tcp_clock_cache is 0x750. So you need to use RBX (reg3) to access the fields in the tcp_sock. But the code used RDI (reg5) as it has the same value. $ pahole --hex -C tcp_sock vmlinux | grep -e 748 -e 750 u64 tcp_wstamp_ns; /* 0x748 0x8 */ u64 tcp_clock_cache; /* 0x750 0x8 */ And this is the disassembly of the part of the function. <__tcp_transmit_skb>: ... 44: mov %rdi, %rbx 47: mov 0x748(%rdi), %rsi 4e: mov 0x750(%rdi), %rax 55: cmp %rax, %rsi Because compiler put the debug info to RBX, it only knows RDI is a pointer to sock and accessing those two fields resulted in error due to offset being beyond the type size. ----------------------------------------------------------- find data type for 0x748(reg5) at __tcp_transmit_skb+0x63 CU for net/ipv4/tcp_output.c (die:0x817f543) frame base: cfa=0 fbreg=6 scope: [1/1] (die:81aac3e) bb: [0 - 30] var [0] -0x98(stack) type='struct tcp_out_options' size=0x28 (die:0x81af3df) var [5] reg8 type='unsigned int' size=0x4 (die:0x8180ed6) var [5] reg2 type='unsigned int' size=0x4 (die:0x8180ed6) var [5] reg1 type='int' size=0x4 (die:0x818059e) var [5] reg4 type='struct sk_buff*' size=0x8 (die:0x8181360) var [5] reg5 type='struct sock*' size=0x8 (die:0x8181a0c) <<<--- the first argument ('sk' at %RDI) mov [19] reg8 -> -0xa8(stack) type='unsigned int' size=0x4 (die:0x8180ed6) mov [20] stack canary -> reg0 mov [29] reg0 -> -0x30(stack) stack canary bb: [36 - 3e] mov [36] reg4 -> reg15 type='struct sk_buff*' size=0x8 (die:0x8181360) bb: [44 - 63] mov [44] reg5 -> reg3 type='struct sock*' size=0x8 (die:0x8181a0c) <<<--- calling tcp_sk() var [47] reg3 type='struct tcp_sock*' size=0x8 (die:0x819eead) <<<--- new variable ('tp' at %RBX) var [4e] reg4 type='unsigned long long' size=0x8 (die:0x8180edd) mov [58] reg4 -> -0xc0(stack) type='unsigned long long' size=0x8 (die:0x8180edd) chk [63] reg5 offset=0x748 ok=1 kind=1 (struct sock*) : offset bigger than size <<<--- access with old variable final result: offset bigger than size While it's a fault in the compiler, we could work around this issue by using the type of new variable when it's copied directly. So I've added copied_from field in the register state to track those direct register to register copies. After that new register gets a new type and the old register still has the same type, it'll update (copy it back) the type of the old register. For example, if we can update type of reg5 at __tcp_transmit_skb+0x47, we can find the target type of the instruction at 0x63 like below: ----------------------------------------------------------- find data type for 0x748(reg5) at __tcp_transmit_skb+0x63 ... bb: [44 - 63] mov [44] reg5 -> reg3 type='struct sock*' size=0x8 (die:0x8181a0c) var [47] reg3 type='struct tcp_sock*' size=0x8 (die:0x819eead) var [47] copyback reg5 type='struct tcp_sock*' size=0x8 (die:0x819eead) <<<--- here mov [47] 0x748(reg5) -> reg4 type='unsigned long long' size=0x8 (die:0x8180edd) mov [4e] 0x750(reg5) -> reg0 type='unsigned long long' size=0x8 (die:0x8180edd) mov [58] reg4 -> -0xc0(stack) type='unsigned long long' size=0x8 (die:0x8180edd) chk [63] reg5 offset=0x748 ok=1 kind=1 (struct tcp_sock*) : Good! <<<--- new type found by insn track: 0x748(reg5) type-offset=0x748 final result: type='struct tcp_sock' size=0xa98 (die:0x819eeb2) Signed-off-by: -
Namhyung Kim authored
When checking the match variable at the target instruction, it might not have any information if it's a first write to a stack slot. In this case it could spill a register value into the stack so the type info is in the source operand. But currently it's hard to get the operand from the checking function. Let's process the instruction and retry to get the type info from the stack if there's no information already. This is an example of __tcp_transmit_skb(). The instructions are <__tcp_transmit_skb>: 0: nopl 0x0(%rax, %rax, 1) 5: push %rbp 6: mov %rsp, %rbp 9: push %r15 b: push %r14 d: push %r13 f: push %r12 11: push %rbx 12: sub $0x98, %rsp 19: mov %r8d, -0xa8(%rbp) ... It cannot find any variable at -0xa8(%rbp) at this point. ----------------------------------------------------------- find data type for -0xa8(reg6) at __tcp_transmit_skb+0x19 CU for net/ipv4/tcp_output.c (die:0x817f543) frame base: cfa=0 fbreg=6 scope: [1/1] (die:81aac3e) bb: [0 - 19] var [0] -0x98(stack) type='struct tcp_out_options' size=0x28 (die:0x81af3df) var [5] reg8 type='unsigned int' size=0x4 (die:0x8180ed6) var [5] reg2 type='unsigned int' size=0x4 (die:0x8180ed6) var [5] reg1 type='int' size=0x4 (die:0x818059e) var [5] reg4 type='struct sk_buff*' size=0x8 (die:0x8181360) var [5] reg5 type='struct sock*' size=0x8 (die:0x8181a0c) chk [19] reg6 offset=-0xa8 ok=0 kind=0 fbreg : no type information no type information And it was able to find the type after processing the 'mov' instruction. ----------------------------------------------------------- find data type for -0xa8(reg6) at __tcp_transmit_skb+0x19 CU for net/ipv4/tcp_output.c (die:0x817f543) frame base: cfa=0 fbreg=6 scope: [1/1] (die:81aac3e) bb: [0 - 19] var [0] -0x98(stack) type='struct tcp_out_options' size=0x28 (die:0x81af3df) var [5] reg8 type='unsigned int' size=0x4 (die:0x8180ed6) var [5] reg2 type='unsigned int' size=0x4 (die:0x8180ed6) var [5] reg1 type='int' size=0x4 (die:0x818059e) var [5] reg4 type='struct sk_buff*' size=0x8 (die:0x8181360) var [5] reg5 type='struct sock*' size=0x8 (die:0x8181a0c) chk [19] reg6 offset=-0xa8 ok=0 kind=0 fbreg : retry <<<--- here mov [19] reg8 -> -0xa8(stack) type='unsigned int' size=0x4 (die:0x8180ed6) chk [19] reg6 offset=-0xa8 ok=0 kind=0 fbreg : Good! found by insn track: -0xa8(reg6) type-offset=0 final result: type='unsigned int' size=0x4 (die:0x8180ed6) Signed-off-by:
-
Namhyung Kim authored
In check_matching_type(), it'd be easier to display the typename in question if it's available. For example, check out the line starts with 'chk'. ----------------------------------------------------------- find data type for 0x10(reg0) at cpuacct_charge+0x13 CU for kernel/sched/build_utility.c (die:0x137ee0b) frame base: cfa=1 fbreg=7 scope: [3/3] (die:13d9632) bb: [c - 13] var [c] reg5 type='struct task_struct*' size=0x8 (die:0x1381230) mov [c] 0xdf8(reg5) -> reg0 type='struct css_set*' size=0x8 (die:0x1385c56) chk [13] reg0 offset=0x10 ok=1 kind=1 (struct css_set*) : Good! <<<--- here found by insn track: 0x10(reg0) type-offset=0x10 final result: type='struct css_set' size=0x250 (die:0x1385b0e) Another example: ----------------------------------------------------------- find data type for 0x8(reg0) at menu_select+0x279 CU for drivers/cpuidle/governors/menu.c (die:0x7b0fe79) frame base: cfa=1 fbreg=7 scope: [2/2] (die:7b11010) bb: [273 - 277] bb: [279 - 279] chk [279] reg0 offset=0x8 ok=0 kind=0 cfa : no type information scope: [1/2] (die:7b10cbc) bb: [0 - 64] ... mov [26a] imm=0xffffffff -> reg15 bb: [273 - 277] bb: [279 - 279] chk [279] reg0 offset=0x8 ok=1 kind=1 (long long unsigned int) : no/void pointer <<<--- here final result: no/void pointer Also change some places to print negative offsets properly. Before: ----------------------------------------------------------- find data type for 0xffffff40(reg6) at __tcp_transmit_skb+0x58 After: ----------------------------------------------------------- find data type for -0xc0(reg6) at __tcp_transmit_skb+0x58 Signed-off-by:
-
Namhyung Kim authored
The __die_find_member_offset_cb() missed to handle bitfield members which don't have DW_AT_data_member_location. Like in adding member types in __add_member_cb() it should fallback to check the bit offset when it resolves the member type for an offset. Fixes: 437683a9 ("perf dwarf-aux: Handle type transfer for memory access") Signed-off-by:
-
- 21 Aug, 2024 6 commits
-
-
Namhyung Kim authored
Sometimes it's useful to organize member fields in cache-line boundary. The 'typecln' sort key is short for type-cacheline and to show samples in each cacheline. The cacheline size is fixed to 64 for now, but it can read the actual size once it saves the value from sysfs. For example, you maybe want to which cacheline in a target is hot or cold. The following shows members in the cfs_rq's first cache line. $ perf report -s type,typecln,typeoff -H ... - 2.67% struct cfs_rq + 1.23% struct cfs_rq: cache-line 2 + 0.57% struct cfs_rq: cache-line 4 + 0.46% struct cfs_rq: cache-line 6 - 0.41% struct cfs_rq: cache-line 0 0.39% struct cfs_rq +0x14 (h_nr_running) 0.02% struct cfs_rq +0x38 (tasks_timeline.rb_leftmost) ... Committer testing: # root@number:~# perf report -s type,typecln,typeoff -H --stdio # Total Lost Samples: 0 # # Samples: 5K of event 'cpu_atom/mem-loads,ldlat=5/P' # Event count (approx.): 312251 # # Overhead Data Type / Data Type Cacheline / Data Type Offset # .............. .................................................. # <SNIP> 0.07% struct sigaction 0.05% struct sigaction: cache-line 1 0.02% struct sigaction +0x58 (sa_mask) 0.02% struct sigaction +0x78 (sa_mask) 0.03% struct sigaction: cache-line 0 0.02% struct sigaction +0x38 (sa_mask) 0.01% struct sigaction +0x8 (sa_mask) <SNIP> Signed-off-by: -
Namhyung Kim authored
It'd be better to have them in hex to check cacheline alignment. Percent offset size field 100.00 0 0x1c0 struct cfs_rq { 0.00 0 0x10 struct load_weight load { 0.00 0 0x8 long unsigned int weight; 0.00 0x8 0x4 u32 inv_weight; }; 0.00 0x10 0x4 unsigned int nr_running; 14.56 0x14 0x4 unsigned int h_nr_running; 0.00 0x18 0x4 unsigned int idle_nr_running; 0.00 0x1c 0x4 unsigned int idle_h_nr_running; ... Committer notes: Justification from Namhyung when asked about why it would be "better": Cache line sizes are power of 2 so it'd be natural to use hex and check whether an offset is in the same boundary. Also 'perf annotate' shows instruction offsets in hex. > > Maybe this should be selectable? I can add an option and/or a config if you want. Signed-off-by: -
Yang Ruibin authored
The check that map is NULL is already done in the bpf_map__fd(map) and returns an errno, which does not run further checks. In addition, even if the check for map is run, the return is a pointer, which is not consistent with the err_number returned by bpf_map__fd(map). Signed-off-by:
Yang Ruibin <11162571@vivo.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephen Rothwell <sfr@canb.auug.org.au> Cc: opensource.kernel@vivo.com Link: https://lore.kernel.org/r/20240821101500.4568-1-11162571@vivo.comSigned-off-by:
-
Namhyung Kim authored
In check_matching_type(), it checks the type state of the register in a wrong order. When it's the percpu pointer, it should check the type for the pointer, but it checks the CFA bit first and thought it has no type in the stack slot. This resulted in no type info. ----------------------------------------------------------- find data type for 0x28(reg1) at hrtimer_reprogram+0x88 CU for kernel/time/hrtimer.c (die:0x18f219f) frame base: cfa=1 fbreg=7 ... add [72] percpu 0x24500 -> reg1 pointer type='struct hrtimer_cpu_base' size=0x240 (die:0x18f6d46) bb: [7a - 7e] bb: [80 - 86] (here) bb: [88 - 88] vvv chk [88] reg1 offset=0x28 ok=1 kind=4 cfa : no type information no type information Here, instruction at 0x72 found reg1 has a (percpu) pointer and got the correct type. But when it checks the final result, it wrongly thought it was stack variable because it checks the cfa bit first. After changing the order of state check: ----------------------------------------------------------- find data type for 0x28(reg1) at hrtimer_reprogram+0x88 CU for kernel/time/hrtimer.c (die:0x18f219f) frame base: cfa=1 fbreg=7 ... (here) vvvvvvvvvv chk [88] reg1 offset=0x28 ok=1 kind=4 percpu ptr : Good! found by insn track: 0x28(reg1) type-offset=0x28 final type: type='struct hrtimer_cpu_base' size=0x240 (die:0x18f6d46) Signed-off-by: -
Namhyung Kim authored
Sometimes a compound type can have a single field and the size is the same as the base type. But it's still preferred as struct or union could carry more information than the base type. Also put a slight priority on the typedef for the same reason. Signed-off-by:
-
Namhyung Kim authored
I found it missed to copy the immediate constant when it moves the register value. This could result in a wrong type inference since the address for the per-cpu variable would be 0 always. Fixes: eb9190af ("perf annotate-data: Handle ADD instructions") Signed-off-by:
-
- 20 Aug, 2024 3 commits
-
-
Ian Rogers authored
Remove dependence on libcap. libcap is only used to query whether a capability is supported, which is just 1 capget system call. If the capget system call fails, fall back on root permission checking. Previously if libcap fails then the permission is assumed not present which may be pessimistic/wrong. Add a used_root out argument to perf_cap__capable to say whether the fall back root check was used. This allows the correct error message, "root" vs "users with the CAP_PERFMON or CAP_SYS_ADMIN capability", to be selected. Tidy uses of perf_cap__capable so that tests aren't repeated if capget isn't supported. Signed-off-by:
Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com> Cc: Changbin Du <changbin.du@huawei.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linux.dev> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Oliver Upton <oliver.upton@linux.dev> Cc: Peter Zijlstra <peterz@infradead.org> Link: https://lore.kernel.org/r/20240806220614.831914-1-irogers@google.comSigned-off-by:
-
Namhyung Kim authored
The bitfield members might not have DW_AT_data_member_location. Let's use DW_AT_data_bit_offset to set the member offset correct. Also use DW_AT_bit_size for the name like in a C program. Before: Annotate type: 'struct sk_buff' (1 samples) Percent Offset Size Field - 100.00 0 232 struct sk_buff { + 0.00 0 24 union ; + 0.00 24 8 union ; + 0.00 32 8 union ; 0.00 40 48 char[] cb; + 0.00 88 16 union ; 0.00 104 8 long unsigned int _nfct; 100.00 112 4 unsigned int len; 0.00 116 4 unsigned int data_len; 0.00 120 2 __u16 mac_len; 0.00 122 2 __u16 hdr_len; 0.00 124 2 __u16 queue_mapping; 0.00 126 0 __u8[] __cloned_offset; 0.00 0 1 __u8 cloned; 0.00 0 1 __u8 nohdr; 0.00 0 1 __u8 fclone; 0.00 0 1 __u8 peeked; 0.00 0 1 __u8 head_frag; 0.00 0 1 __u8 pfmemalloc; 0.00 0 1 __u8 pp_recycle; 0.00 127 1 __u8 active_extensions; + 0.00 128 60 union ; 0.00 188 4 sk_buff_data_t tail; 0.00 192 4 sk_buff_data_t end; 0.00 200 8 unsigned char* head; After: Annotate type: 'struct sk_buff' (1 samples) Percent Offset Size Field - 100.00 0 232 struct sk_buff { + 0.00 0 24 union ; + 0.00 24 8 union ; + 0.00 32 8 union ; 0.00 40 48 char[] cb + 0.00 88 16 union ; 0.00 104 8 long unsigned int _nfct; 100.00 112 4 unsigned int len; 0.00 116 4 unsigned int data_len; 0.00 120 2 __u16 mac_len; 0.00 122 2 __u16 hdr_len; 0.00 124 2 __u16 queue_mapping; 0.00 126 0 __u8[] __cloned_offset; 0.00 126 1 __u8 cloned:1; 0.00 126 1 __u8 nohdr:1; 0.00 126 1 __u8 fclone:2; 0.00 126 1 __u8 peeked:1; 0.00 126 1 __u8 head_frag:1; 0.00 126 1 __u8 pfmemalloc:1; 0.00 126 1 __u8 pp_recycle:1; 0.00 127 1 __u8 active_extensions; + 0.00 128 60 union ; 0.00 188 4 sk_buff_data_t tail; 0.00 192 4 sk_buff_data_t end; 0.00 200 8 unsigned char* head; Commiter notes: Collect some data: root@number:~# perf mem record -a --ldlat 5 -- ping -s 8193 -f 192.168.86.1 Memory events are enabled on a subset of CPUs: 16-27 PING 192.168.86.1 (192.168.86.1) 8193(8221) bytes of data. .^C --- 192.168.86.1 ping statistics --- 13881 packets transmitted, 13880 received, 0.00720409% packet loss, time 8664ms rtt min/avg/max/mdev = 0.510/0.599/7.768/0.115 ms, ipg/ewma 0.624/0.593 ms [ perf record: Woken up 8 times to write data ] [ perf record: Captured and wrote 14.877 MB perf.data (46785 samples) ] root@number:~# root@number:~# perf evlist cpu_atom/mem-loads,ldlat=5/P cpu_atom/mem-stores/P dummy:u root@number:~# perf evlist -v cpu_atom/mem-loads,ldlat=5/P: type: 10 (cpu_atom), size: 136, config: 0x5d0 (mem-loads), { sample_period, sample_freq }: 4000, sample_type: IP|TID|TIME|ADDR|CPU|PERIOD|IDENTIFIER|DATA_SRC|WEIGHT_STRUCT, read_format: ID|LOST, disabled: 1, inherit: 1, freq: 1, precise_ip: 3, sample_id_all: 1, { bp_addr, config1 }: 0x7 cpu_atom/mem-stores/P: type: 10 (cpu_atom), size: 136, config: 0x6d0 (mem-stores), { sample_period, sample_freq }: 4000, sample_type: IP|TID|TIME|ADDR|CPU|PERIOD|IDENTIFIER|DATA_SRC|WEIGHT_STRUCT, read_format: ID|LOST, disabled: 1, inherit: 1, freq: 1, precise_ip: 3, sample_id_all: 1 dummy:u: type: 1 (software), size: 136, config: 0x9 (PERF_COUNT_SW_DUMMY), { sample_period, sample_freq }: 1, sample_type: IP|TID|TIME|ADDR|CPU|IDENTIFIER|DATA_SRC|WEIGHT_STRUCT, read_format: ID|LOST, inherit: 1, exclude_kernel: 1, exclude_hv: 1, mmap: 1, comm: 1, task: 1, mmap_data: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1 root@number:~# Ok, now lets see what changes from before this patch to after it: root@number:~# perf annotate --data-type > /tmp/before Apply the patch, build: root@number:~# perf annotate --data-type > /tmp/after The first hunk of the diff, for a glib data structure, in userspace, look at those bitfields: root@number:~# diff -u10 /tmp/before /tmp/after | head -20 --- /tmp/before 2024-08-20 17:29:58.306765780 -0300 +++ /tmp/after 2024-08-20 17:33:13.210582596 -0300 @@ -163,22 +163,22 @@ Annotate type: 'GHashTable' in /usr/lib64/libglib-2.0.so.0.8000.3 (1 samples): ============================================================================ Percent offset size field 100.00 0 96 GHashTable { 0.00 0 8 gsize size; 0.00 8 4 gint mod; 100.00 12 4 guint mask; 0.00 16 4 guint nnodes; 0.00 20 4 guint noccupied; - 0.00 0 4 guint have_big_keys; - 0.00 0 4 guint have_big_values; + 0.00 24 1 guint have_big_keys:1; + 0.00 24 1 guint have_big_values:1; 0.00 32 8 gpointer keys; 0.00 40 8 guint* hashes; 0.00 48 8 gpointer values; root@number:~# As advertised :-) Signed-off-by: -
Arnaldo Carvalho de Melo authored
The previous attempt fixed the build on debian:experimental-x-mipsel, but when building on a larger set of containers I noticed it broke the build on some other 32-bit architectures such as: 42 7.87 ubuntu:18.04-x-arm : FAIL gcc version 7.5.0 (Ubuntu/Linaro 7.5.0-3ubuntu1~18.04) builtin-daemon.c: In function 'cmd_session_list': builtin-daemon.c:692:16: error: format '%llu' expects argument of type 'long long unsigned int', but argument 4 has type 'long int' [-Werror=format=] fprintf(out, "%c%" PRIu64, ^~~~~ builtin-daemon.c:694:13: csv_sep, (curr - daemon->start) / 60); ~~~~~~~~~~~~~~~~~~~~~~~~~~~ In file included from builtin-daemon.c:3:0: /usr/arm-linux-gnueabihf/include/inttypes.h:105:34: note: format string is defined here # define PRIu64 __PRI64_PREFIX "u" So lets cast that time_t (32-bit/64-bit) to uint64_t to make sure it builds everywhere. Fixes: 4bbe6002 ("perf daemon: Fix the build on 32-bit architectures") Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/lkml/ZsPmldtJ0D9Cua9_@x1Signed-off-by:
-
- 19 Aug, 2024 16 commits
-
-
Namhyung Kim authored
Add it to the record.sh shell test to verify if it tracks cgroup information correctly. It records with --all-cgroups option can check if it has PERF_RECORD_CGROUP and the names are not "unknown". $ sudo ./perf test -vv 95 95: perf record tests: --- start --- test child forked, pid 2871922 169c90-169cd0 g test_loop perf does have symbol 'test_loop' Basic --per-thread mode test Basic --per-thread mode test [Success] Register capture test Register capture test [Success] Basic --system-wide mode test Basic --system-wide mode test [Success] Basic target workload test Basic target workload test [Success] Branch counter test branch counter feature not supported on all core PMUs (/sys/bus/event_source/devices/cpu) [Skipped] Cgroup sampling test Cgroup sampling test [Success] ---- end(0) ---- 95: perf record tests : Ok Reviewed-by:
-
Namhyung Kim authored
The recent change in 'struct perf_tool' constification broke the cgroup and/or namespace tracking by resetting tool fields. It should set the values after perf_tool__init(). Fixes: cecb1cf1 ("perf record: Use perf_tool__init()") Reviewed-by:
-
Ian Rogers authored
The handling of mmap and mmap2 events is near identical. Add a common helper function and call that by the two event handling functions. Signed-off-by:
-
Ian Rogers authored
There are repipe, build ID and JIT dump variants of the mmap and mmap2 repipe functions. The organization doesn't allow JIT dump to work with build ID injection and the structure is less than clear. Combine the function and enable the different behaviors based on ifs. Signed-off-by:
-
Ian Rogers authored
It is clearer to have a single enum that determines how build ids are injected, it also allows for future extension. Set the header build ID feature whether lazy or all are generated, previously only the lazy case would set it. Allow parsing of known build IDs for either the lazy or all cases. Signed-off-by:
-
Ian Rogers authored
Test recording of call-graphs and injecting --build-all. Add/expand trap handler. Signed-off-by:
-
Ian Rogers authored
Allows evsel__id_hdr_size() to be used when the evsel is const. Signed-off-by:
-
Ian Rogers authored
The passed dso_id is copied and so is never an out argument. Remove its mutability. Signed-off-by:
-
Ian Rogers authored
Make it clearer the argument is just being used as a string. Signed-off-by:
-
Ian Rogers authored
map__init() is only used internally so make it static. Assume memory is zero initialized, which will better support adding fields to struct map in the future and was already the case for map__new2. To reduce complexity, change set_priv and set_erange_warned to not take a value to assign as they always assign true. Signed-off-by:
-
Ian Rogers authored
Make sure the memset of a synthesized event only zeros the necessary tracing data part of the event, as a full event can be over 4kb in size. Signed-off-by:
-
Xu Yang authored
The 32-bit arm build system will complain: tools/perf/util/python.c:75:28: error: field ‘sample’ has incomplete type 75 | struct perf_sample sample; However, arm64 build system doesn't complain this. The root cause is arm64 define "HAVE_KVM_STAT_SUPPORT := 1" in tools/perf/arch/arm64/Makefile, but arm arch doesn't define this. This will lead to kvm-stat.h include other header files on arm64 build system, especially "util/sample.h" for util/python.c. This will try to directly include "util/sample.h" for "util/python.c" to avoid such build issue on arm platform. Signed-off-by:Xu Yang <xu.yang_2@nxp.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: imx@lists.linux.dev Link: https://lore.kernel.org/r/20240819023403.201324-1-xu.yang_2@nxp.comSigned-off-by:
-
Namhyung Kim authored
After trying all possibilities with DWARF and instruction tracking. Signed-off-by:
-
Namhyung Kim authored
Sometimes it matches a variable in the inner scope but it fails because the actual access can be on a different type. Let's try variables in every scope and choose the best one using is_better_type(). I have an example with update_blocked_averages(), at first it found a variable (__mptr) but it's a void pointer. So it moved on to the upper scope and found another variable (cfs_rq). $ perf --debug type-profile annotate --data-type --stdio ... ----------------------------------------------------------- find data type for 0x140(reg14) at update_blocked_averages+0x2db CU for kernel/sched/fair.c (die:0x12dd892) frame base: cfa=1 fbreg=7 found "__mptr" (die: 0x13022f1) in scope=4/4 (die: 0x13022e8) failed: no/void pointer variable location: base=reg14, offset=0x140 type='void*' size=0x8 (die:0x12dd8f9) found "cfs_rq" (die: 0x1301721) in scope=3/4 (die: 0x130171c) type_offset=0x140 variable location: reg14 type='struct cfs_rq' size=0x1c0 (die:0x12e37e5) final type: type='struct cfs_rq' size=0x1c0 (die:0x12e37e5) IIUC the scope is like below: 1: update_blocked_averages 2: __update_blocked_fair 3: for_each_leaf_cfs_rq_safe 4: list_entry -> (container_of) The container_of is implemented like: #define container_of(ptr, type, member) ({ \ void *__mptr = (void *)(ptr); \ static_assert(__same_type(*(ptr), ((type *)0)->member) || \ __same_type(*(ptr), void), \ "pointer type mismatch in container_of()"); \ ((type *)(__mptr - offsetof(type, member))); }) That's why we see the __mptr variable first but it failed since it has no type information. Then for_each_leaf_cfs_rq_safe() is defined as #define for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos) \ list_for_each_entry_safe(cfs_rq, pos, &rq->leaf_cfs_rq_list, \ leaf_cfs_rq_list) Note that the access was 0x140(r14). And the cfs_rq has leaf_cfs_rq_list at the 0x140. So it converts the list_head pointer to a pointer to struct cfs_rq here. $ pahole --hex -C cfs_rq vmlinux | grep 140 struct cfs_rq struct list_head leaf_cfs_rq_list; /* 0x140 0x10 */ Signed-off-by: -
Namhyung Kim authored
Sometimes more than one variables are located in the same register or a stack slot. Or it can overwrite existing information with others. I found this is not helpful in some cases so it needs to update the type information from the variable only if it's better. But it's hard to know which one is better, so we needs heuristics. :) As it deals with memory accesses, the location should have a pointer or something similar (like array or reference). So if it had an integer type and a variable is a pointer, we can take the variable's type to resolve the target of the access. If it has a pointer type and a variable with the same location has a different pointer type, it'll take one with bigger target type. This can be useful when the target type embeds a smaller type (like list header or RB-tree node) at the beginning so their location is same. Signed-off-by:
-
Namhyung Kim authored
It treats pointers and arrays in the same way. Let's add the helper and use it when it checks if it needs a pointer. Signed-off-by:
-
