AArnaldo Carvalho de Meloperf tools powerpc: Add support for extended regs in power10
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
tools, perf: add and use optimized ring_buffer_{read_head, write_tail} helpers Currently, on x86-64, perf uses LFENCE and MFENCE (rmb() and mb(), respectively) when processing events from the perf ring buffer which is unnecessarily expensive as we can do more lightweight in particular given this is critical fast-path in perf. According to Peter rmb()/mb() were added back then via a94d342b9cb0 ("tools/perf: Add required memory barriers") at a time where kernel still supported chips that needed it, but nowadays support for these has been ditched completely, therefore we can fix them up as well. While for x86-64, replacing rmb() and mb() with smp_*() variants would result in just a compiler barrier for the former and LOCK + ADD for the latter (__sync_synchronize() uses slower MFENCE by the way), Peter suggested we can use smp_{load_acquire,store_release}() instead for architectures where its implementation doesn't resolve in slower smp_mb(). Thus, e.g. in x86-64 we would be able to avoid CPU barrier entirely due to TSO. For architectures where the latter needs to use smp_mb() e.g. on arm, we stick to cheaper smp_rmb() variant for fetching the head. This work adds helpers ring_buffer_read_head() and ring_buffer_write_tail() for tools infrastructure that either switches to smp_load_acquire() for architectures where it is cheaper or uses READ_ONCE() + smp_rmb() barrier for those where it's not in order to fetch the data_head from the perf control page, and it uses smp_store_release() to write the data_tail. Latter is smp_mb() + WRITE_ONCE() combination or a cheaper variant if architecture allows for it. Those that rely on smp_rmb() and smp_mb() can further improve performance in a follow up step by implementing the two under tools/arch/*/include/asm/barrier.h such that they don't have to fallback to rmb() and mb() in tools/include/asm/barrier.h. Switch perf to use ring_buffer_read_head() and ring_buffer_write_tail() so it can make use of the optimizations. Later, we convert libbpf as well to use the same helpers. Side note [0]: the topic has been raised of whether one could simply use the C11 gcc builtins [1] for the smp_load_acquire() and smp_store_release() instead: __atomic_load_n(ptr, __ATOMIC_ACQUIRE); __atomic_store_n(ptr, val, __ATOMIC_RELEASE); Kernel and (presumably) tooling shipped along with the kernel has a minimum requirement of being able to build with gcc-4.6 and the latter does not have C11 builtins. While generally the C11 memory models don't align with the kernel's, the C11 load-acquire and store-release alone /could/ suffice, however. Issue is that this is implementation dependent on how the load-acquire and store-release is done by the compiler and the mapping of supported compilers must align to be compatible with the kernel's implementation, and thus needs to be verified/tracked on a case by case basis whether they match (unless an architecture uses them also from kernel side). The implementations for smp_load_acquire() and smp_store_release() in this patch have been adapted from the kernel side ones to have a concrete and compatible mapping in place. [0] http://patchwork.ozlabs.org/patch/985422/ [1] https://gcc.gnu.org/onlinedocs/gcc/_005f_005fatomic-Builtins.html Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> | 7 年前 | |
Merge tag 'perf-tools-2020-08-10' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux Pull perf tools updates from Arnaldo Carvalho de Melo: "New features: - Introduce controlling how 'perf stat' and 'perf record' works via a control file descriptor, allowing starting with events configured but disabled until commands are received via the control file descriptor. This allows, for instance for tools such as Intel VTune to make further use of perf as its Linux platform driver. - Improve 'perf record' to to register in a perf.data file header the clockid used to help later correlate things like syslog files and perf events recorded. - Add basic syscall and find_next_bit benchmarks to 'perf bench'. - Allow using computed metrics in calculating other metrics. For instance: { .metric_expr = "l2_rqsts.demand_data_rd_hit + l2_rqsts.pf_hit + l2_rqsts.rfo_hit", .metric_name = "DCache_L2_All_Hits", }, { .metric_expr = "max(l2_rqsts.all_demand_data_rd - l2_rqsts.demand_data_rd_hit, 0) + l2_rqsts.pf_miss + l2_rqsts.rfo_miss", .metric_name = "DCache_L2_All_Miss", }, { .metric_expr = "dcache_l2_all_hits + dcache_l2_all_miss", .metric_name = "DCache_L2_All", } - Add suport for 'd_ratio', '>' and '<' operators to the expression resolver used in calculating metrics in 'perf stat'. Support for new kernel features: - Support TEXT_POKE and KSYMBOL_TYPE_OOL perf metadata events to cope with things like ftrace, trampolines, i.e. changes in the kernel text that gets in the way of properly decoding Intel PT hardware traces, for instance. Intel PT: - Add various knobs to reduce the volume of Intel PT traces by reducing the level of details such as decoding just some types of packets (e.g., FUP/TIP, PSB+), also filtering by time range. - Add new itrace options (log flags to the 'd' option, error flags to the 'e' one, etc), controlling how Intel PT is transformed into perf events, document some missing options (e.g., how to synthesize callchains). BPF: - Properly report BPF errors when parsing events. - Do not setup side-band events if LIBBPF is not linked, fixing a segfault. Libraries: - Improvements to the libtraceevent plugin mechanism. - Improve libtracevent support for KVM trace events SVM exit reasons. - Add a libtracevent plugins for decoding syscalls/sys_enter_futex and for tlb_flush. - Ensure sample_period is set libpfm4 events in 'perf test'. - Fixup libperf namespacing, to make sure what is in libperf has the perf_ namespace while what is now only in tools/perf/ doesn't use that prefix. Arch specific: - Improve the testing of vendor events and metrics in 'perf test'. - Allow no ARM CoreSight hardware tracer sink to be specified on command line. - Fix arm_spe_x recording when mixed with other perf events. - Add s390 idle functions 'psw_idle' and 'psw_idle_exit' to list of idle symbols. - List kernel supplied event aliases for arm64 in 'perf list'. - Add support for extended register capability in PowerPC 9 and 10. - Added nest IMC power9 metric events. Miscellaneous: - No need to setup sample_regs_intr/sample_regs_user for dummy events. - Update various copies of kernel headers, some causing perf to handle new syscalls, MSRs, etc. - Improve usage of flex and yacc, enabling warnings and addressing the fallout. - Add missing '--output' option to 'perf kmem' so that it can pass it along to 'perf record'. - 'perf probe' fixes related to adding multiple probes on the same address for the same event. - Make 'perf probe' warn if the target function is a GNU indirect function. - Remove //anon mmap events from 'perf inject jit' to fix supporting both using ELF files for generated functions and the perf-PID.map approaches" * tag 'perf-tools-2020-08-10' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux: (144 commits) perf record: Skip side-band event setup if HAVE_LIBBPF_SUPPORT is not set perf tools powerpc: Add support for extended regs in power10 perf tools powerpc: Add support for extended register capability tools headers UAPI: Sync drm/i915_drm.h with the kernel sources tools arch x86: Sync asm/cpufeatures.h with the kernel sources tools arch x86: Sync the msr-index.h copy with the kernel sources tools headers UAPI: update linux/in.h copy tools headers API: Update close_range affected files perf script: Add 'tod' field to display time of day perf script: Change the 'enum perf_output_field' enumerators to be 64 bits perf data: Add support to store time of day in CTF data conversion perf tools: Move clockid_res_ns under clock struct perf header: Store clock references for -k/--clockid option perf tools: Add clockid_name function perf clockid: Move parse_clockid() to new clockid object tools lib traceevent: Handle possible strdup() error in tep_add_plugin_path() API libtraceevent: Fixed description of tep_add_plugin_path() API libtraceevent: Fixed type in PRINT_FMT_STING libtraceevent: Fixed broken indentation in parse_ip4_print_args() libtraceevent: Improve error handling of tep_plugin_add_option() API ... | 5 年前 |