- 03 Apr, 2024 1 commit
-
-
Puranjay Mohan authored
Add support for [LDX | STX | ST], PROBE_MEM32, [B | H | W | DW] instructions. They are similar to PROBE_MEM instructions with the following differences: - PROBE_MEM32 supports store. - PROBE_MEM32 relies on the verifier to clear upper 32-bit of the src/dst register - PROBE_MEM32 adds 64-bit kern_vm_start address (which is stored in R28 in the prologue). Due to bpf_arena constructions such R28 + reg + off16 access is guaranteed to be within arena virtual range, so no address check at run-time. - PROBE_MEM32 allows STX and ST. If they fault the store is a nop. When LDX faults the destination register is zeroed. To support these on arm64, we do tmp2 = R28 + src/dst reg and then use tmp2 as the new src/dst register. This allows us to reuse most of the code for normal [LDX | STX | ST]. Signed-off-by:
Puranjay Mohan <puranjay12@gmail.com> Link: https://lore.kernel.org/r/20240325150716.4387-2-puranjay12@gmail.comSigned-off-by:
Alexei Starovoitov <ast@kernel.org>
-
- 02 Apr, 2024 11 commits
-
-
Geliang Tang authored
In order to prevent mptcpify prog from affecting the running results of other BPF tests, a pid limit was added to restrict it from only modifying its own program. Suggested-by:
Martin KaFai Lau <martin.lau@kernel.org> Signed-off-by:
Geliang Tang <tanggeliang@kylinos.cn> Signed-off-by:
Andrii Nakryiko <andrii@kernel.org> Link: https://lore.kernel.org/bpf/8987e2938e15e8ec390b85b5dcbee704751359dc.1712054986.git.tanggeliang@kylinos.cn
-
Tobias Böhm authored
Commit 20d59ee5 ("libbpf: add bpf_core_cast() macro") added a bpf_helpers include in bpf_core_read.h as a system include. Usually, the includes are local, though, like in bpf_tracing.h. This commit adjusts the include to be local as well. Signed-off-by:
Tobias Böhm <tobias@aibor.de> Signed-off-by:
-
Jose Fernandez authored
This patch improves the run-time calculation for program stats by capturing the duration as soon as possible after the program returns. Previously, the duration included u64_stats_t operations. While the instrumentation overhead is part of the total time spent when stats are enabled, distinguishing between the program's native execution time and the time spent due to instrumentation is crucial for accurate performance analysis. By making this change, the patch facilitates more precise optimization of BPF programs, enabling users to understand their performance in environments without stats enabled. I used a virtualized environment to measure the run-time over one minute for a basic raw_tracepoint/sys_enter program, which just increments a local counter. Although the virtualization introduced some performance degradation that could affect the results, I observed approximately a 16% decrease in average run-time reported by stats with this change (310 -> 260 nsec). Signed-off-by:
Jose Fernandez <josef@netflix.com> Signed-off-by:
Daniel Borkmann <daniel@iogearbox.net> Acked-by:
-
Pu Lehui authored
When testing send_signal and stacktrace_build_id_nmi using the riscv sbi pmu driver without the sscofpmf extension or the riscv legacy pmu driver, then failures as follows are encountered: test_send_signal_common:FAIL:perf_event_open unexpected perf_event_open: actual -1 < expected 0 #272/3 send_signal/send_signal_nmi:FAIL test_stacktrace_build_id_nmi:FAIL:perf_event_open err -1 errno 95 #304 stacktrace_build_id_nmi:FAIL The reason is that the above pmu driver or hardware does not support sampling events, that is, PERF_PMU_CAP_NO_INTERRUPT is set to pmu capabilities, and then perf_event_open returns EOPNOTSUPP. Since PERF_PMU_CAP_NO_INTERRUPT is not only set in the riscv-related pmu driver, it is better to skip testing when this capability is set. Signed-off-by:Pu Lehui <pulehui@huawei.com> Signed-off-by:
-
Andrii Nakryiko authored
When generated BPF skeleton header is included in C++ code base, some compiler setups will emit warning about using language extensions due to typeof() usage, resulting in something like: error: extension used [-Werror,-Wlanguage-extension-token] obj->struct_ops.empty_tcp_ca = (typeof(obj->struct_ops.empty_tcp_ca)) ^ It looks like __typeof__() is a preferred way to do typeof() with better C++ compatibility behavior, so switch to that. With __typeof__() we get no such warning. Fixes: c2a0257c ("bpftool: Cast pointers for shadow types explicitly.") Fixes: 00389c58 ("bpftool: Add support for subskeletons") Signed-off-by:Kui-Feng Lee <thinker.li@gmail.com> Acked-by:
Quentin Monnet <qmo@kernel.org> Acked-by:
John Fastabend <john.fastabend@gmail.com> Link: https://lore.kernel.org/bpf/20240401170713.2081368-1-andrii@kernel.org
-
Yonghong Song authored
Currently, cond_break macro uses bytes to encode the may_goto insn. Patch [1] in llvm implemented may_goto insn in BPF backend. Replace byte-level encoding with llvm inline asm for better usability. Using llvm may_goto insn is controlled by macro __BPF_FEATURE_MAY_GOTO. [1] https://github.com/llvm/llvm-project/commit/0e0bfacff71859d1f9212205f8f873d47029d3fbSigned-off-by:
Yonghong Song <yonghong.song@linux.dev> Signed-off-by:
-
Anton Protopopov authored
When more than 64 maps are used by a program and its subprograms the verifier returns -E2BIG. Add a verbose message which highlights the source of the error and also print the actual limit. Signed-off-by:
Anton Protopopov <aspsk@isovalent.com> Signed-off-by:
-
David Lechner authored
In a few places in the bpf uapi headers, EOPNOTSUPP is missing a "P" in the doc comments. This adds the missing "P". Signed-off-by:
David Lechner <dlechner@baylibre.com> Signed-off-by:
-
Rameez Rehman authored
Improve the formatting of the attach flags for cgroup programs in the relevant man page, and fix typos ("can be on of", "an userspace inet socket") when introducing that list. Also fix a couple of other trivial issues in docs. [ Quentin: Fixed trival issues in bpftool-gen.rst and bpftool-iter.rst ] Signed-off-by:Rameez Rehman <rameezrehman408@hotmail.com> Signed-off-by:
-
Rameez Rehman authored
As it turns out, the terms in definition lists in the rST file are already rendered with bold-ish formatting when generating the man pages; all double-star sequences we have in the commands for the command description are unnecessary, and can be removed to make the documentation easier to read. The rST files were automatically processed with: sed -i '/DESCRIPTION/,/OPTIONS/ { /^\*/ s/\*\*//g }' b*.rst Signed-off-by: -
Rameez Rehman authored
The rST manual pages for bpftool would use a mix of tabs and spaces for indentation. While this is the norm in C code, this is rather unusual for rST documents, and over time we've seen many contributors use a wrong level of indentation for documentation update. Let's fix bpftool's indentation in docs once and for all: - Let's use spaces, that are more common in rST files. - Remove one level of indentation for the synopsis, the command description, and the "see also" section. As a result, all sections start with the same indentation level in the generated man page. - Rewrap the paragraphs after the changes. There is no content change in this patch, only indentation and rewrapping changes. The wrapping in the generated source files for the manual pages is changed, but the pages displayed with "man" remain the same, apart from the adjusted indentation level on relevant sections. [ Quentin: rebased on bpf-next, removed indent level for command description and options, updated synopsis, command summary, and "see also" sections. ] Signed-off-by:
-
- 30 Mar, 2024 1 commit
-
-
Andrii Nakryiko authored
When BPF selftests are built in RELEASE=1 mode with -O2 optimization level, uprobe_multi binary, called from multi-uprobe tests is optimized to the point that all the thousands of target uprobe_multi_func_XXX functions are eliminated, breaking tests. So ensure they are preserved by using weak attribute. But, actually, compiling uprobe_multi binary with -O2 takes a really long time, and is quite useless (it's not a benchmark). So in addition to ensuring that uprobe_multi_func_XXX functions are preserved, opt-out of -O2 explicitly in Makefile and stick to -O0. This saves a lot of compilation time. With -O2, just recompiling uprobe_multi: $ touch uprobe_multi.c $ time make RELEASE=1 -j90 make RELEASE=1 -j90 291.66s user 2.54s system 99% cpu 4:55.52 total With -O0: $ touch uprobe_multi.c $ time make RELEASE=1 -j90 make RELEASE=1 -j90 22.40s user 1.91s system 99% cpu 24.355 total 5 minutes vs (still slow, but...) 24 seconds. Signed-off-by:
-
- 29 Mar, 2024 27 commits
-
-
Alexei Starovoitov authored
syzbot reported the following lock sequence: cpu 2: grabs timer_base lock spins on bpf_lpm lock cpu 1: grab rcu krcp lock spins on timer_base lock cpu 0: grab bpf_lpm lock spins on rcu krcp lock bpf_lpm lock can be the same. timer_base lock can also be the same due to timer migration. but rcu krcp lock is always per-cpu, so it cannot be the same lock. Hence it's a false positive. To avoid lockdep complaining move kfree_rcu() after spin_unlock. Reported-by: syzbot+1fa663a2100308ab6eab@syzkaller.appspotmail.com Signed-off-by: -
Martin KaFai Lau authored
Geliang Tang says: ==================== Simplify bpf_tcp_ca test by using connect_fd_to_fd and start_server helpers. v4: - Matt reminded me that I shouldn't send a square-to patch to BPF (thanks), so I update them into two patches in v4. v3: - split v2 as two patches as Daniel suggested. - The patch "selftests/bpf: Use start_server in bpf_tcp_ca" is merged by Daniel (thanks), but I forgot to drop 'settimeo(lfd, 0)' in it, so I send a squash-to patch to fix this. ==================== Signed-off-by: -
Geliang Tang authored
settimeo is invoked in start_server() and in connect_fd_to_fd() already, no need to invoke settimeo(lfd, 0) and settimeo(fd, 0) in do_test() anymore. This patch drops them. Signed-off-by:
-
Geliang Tang authored
To simplify the code, use BPF selftests helper connect_fd_to_fd() in bpf_tcp_ca.c instead of open-coding it. This helper is defined in network_helpers.c, and exported in network_helpers.h, which is already included in bpf_tcp_ca.c. Signed-off-by:
-
Martin KaFai Lau authored
syzbot reported uninit memory usages during map_{lookup,delete}_elem. ========== BUG: KMSAN: uninit-value in __dev_map_lookup_elem kernel/bpf/devmap.c:441 [inline] BUG: KMSAN: uninit-value in dev_map_lookup_elem+0xf3/0x170 kernel/bpf/devmap.c:796 __dev_map_lookup_elem kernel/bpf/devmap.c:441 [inline] dev_map_lookup_elem+0xf3/0x170 kernel/bpf/devmap.c:796 ____bpf_map_lookup_elem kernel/bpf/helpers.c:42 [inline] bpf_map_lookup_elem+0x5c/0x80 kernel/bpf/helpers.c:38 ___bpf_prog_run+0x13fe/0xe0f0 kernel/bpf/core.c:1997 __bpf_prog_run256+0xb5/0xe0 kernel/bpf/core.c:2237 ========== The reproducer should be in the interpreter mode. The C reproducer is trying to run the following bpf prog: 0: (18) r0 = 0x0 2: (18) r1 = map[id:49] 4: (b7) r8 = 16777216 5: (7b) *(u64 *)(r10 -8) = r8 6: (bf) r2 = r10 7: (07) r2 += -229 ^^^^^^^^^^ 8: (b7) r3 = 8 9: (b7) r4 = 0 10: (85) call dev_map_lookup_elem#1543472 11: (95) exit It is due to the "void *key" (r2) passed to the helper. bpf allows uninit stack memory access for bpf prog with the right privileges. This patch uses kmsan_unpoison_memory() to mark the stack as initialized. This should address different syzbot reports on the uninit "void *key" argument during map_{lookup,delete}_elem. Reported-by: syzbot+603bcd9b0bf1d94dbb9b@syzkaller.appspotmail.com Closes: https://lore.kernel.org/bpf/000000000000f9ce6d061494e694@google.com/ Reported-by: syzbot+eb02dc7f03dce0ef39f3@syzkaller.appspotmail.com Closes: https://lore.kernel.org/bpf/000000000000a5c69c06147c2238@google.com/ Reported-by: syzbot+b4e65ca24fd4d0c734c3@syzkaller.appspotmail.com Closes: https://lore.kernel.org/bpf/000000000000ac56fb06143b6cfa@google.com/ Reported-by: syzbot+d2b113dc9fea5e1d2848@syzkaller.appspotmail.com Closes: https://lore.kernel.org/bpf/0000000000000d69b206142d1ff7@google.com/ Reported-by: syzbot+1a3cf6f08d68868f9db3@syzkaller.appspotmail.com Closes: https://lore.kernel.org/bpf/0000000000006f876b061478e878@google.com/ Tested-by: syzbot+1a3cf6f08d68868f9db3@syzkaller.appspotmail.com Suggested-by: -
Alexei Starovoitov authored
Yonghong Song says: ==================== bpf: Fix a couple of test failures with LTO kernel With a LTO kernel built with clang, with one of earlier version of kernel, I encountered two test failures, ksyms and kprobe_multi_bench_attach/kernel. Now with latest bpf-next, only kprobe_multi_bench_attach/kernel failed. But it is possible in the future ksyms selftest may fail again. Both test failures are due to static variable/function renaming due to cross-file inlining. For Ksyms failure, the solution is to strip .llvm.<hash> suffixes for symbols in /proc/kallsyms before comparing against the ksym in bpf program. For kprobe_multi_bench_attach/kernel failure, the solution is to either provide names in /proc/kallsyms to the kernel or ignore those names who have .llvm.<hash> suffix since the kernel sym name comparison is against /proc/kallsyms. Please see each individual patches for details. Changelogs: v2 -> v3: - no need to check config file, directly so strstr with '.llvm.'. - for kprobe_multi_bench with syms, instead of skipping the syms, consult /proc/kallyms to find corresponding names. - add a test with populating addrs to the kernel for kprobe multi attach. v1 -> v2: - Let libbpf handle .llvm.<hash suffixes since it may impact bpf program ksym. ==================== Link: https://lore.kernel.org/r/20240326041443.1197498-1-yonghong.song@linux.devSigned-off-by: -
Yonghong Song authored
Get addrs directly from available_filter_functions_addrs and send to the kernel during kprobe_multi_attach. This avoids consultation of /proc/kallsyms. But available_filter_functions_addrs is introduced in 6.5, i.e., it is introduced recently, so I skip the test if the kernel does not support it. Signed-off-by:
-
Yonghong Song authored
In my locally build clang LTO kernel (enabling CONFIG_LTO and CONFIG_LTO_CLANG_THIN), kprobe_multi_bench_attach/kernel subtest failed like: test_kprobe_multi_bench_attach:PASS:get_syms 0 nsec test_kprobe_multi_bench_attach:PASS:kprobe_multi_empty__open_and_load 0 nsec libbpf: prog 'test_kprobe_empty': failed to attach: No such process test_kprobe_multi_bench_attach:FAIL:bpf_program__attach_kprobe_multi_opts unexpected error: -3 #117/1 kprobe_multi_bench_attach/kernel:FAIL There are multiple symbols in /sys/kernel/debug/tracing/available_filter_functions are renamed in /proc/kallsyms due to cross file inlining. One example is for static function __access_remote_vm in mm/memory.c. In a non-LTO kernel, we have the following call stack: ptrace_access_vm (global, kernel/ptrace.c) access_remote_vm (global, mm/memory.c) __access_remote_vm (static, mm/memory.c) With LTO kernel, it is possible that access_remote_vm() is inlined by ptrace_access_vm(). So we end up with the following call stack: ptrace_access_vm (global, kernel/ptrace.c) __access_remote_vm (static, mm/memory.c) The compiler renames __access_remote_vm to __access_remote_vm.llvm.<hash> to prevent potential name collision. The kernel bpf_kprobe_multi_link_attach() and ftrace_lookup_symbols() try to find addresses based on /proc/kallsyms, hence the current test failed with LTO kenrel. This patch consulted /proc/kallsyms to find the corresponding entries for the ksym and this solved the issue. Signed-off-by: -
Yonghong Song authored
These two functions allow selftests to do loading/searching kallsyms based on their specific compare functions. Signed-off-by:
-
Yonghong Song authored
Refactor trace helper function load_kallsyms_local() such that it invokes a common function with a compare function as input. The common function will be used later for other local functions. Signed-off-by:
-
Yonghong Song authored
Refactor some functions in kprobe_multi_test.c to extract some helper functions who will be used in later patches to avoid code duplication. Signed-off-by:
-
Yonghong Song authored
With CONFIG_LTO_CLANG_THIN enabled, with some of previous version of kernel code base ([1]), I hit the following error: test_ksyms:PASS:kallsyms_fopen 0 nsec test_ksyms:FAIL:ksym_find symbol 'bpf_link_fops' not found #118 ksyms:FAIL The reason is that 'bpf_link_fops' is renamed to bpf_link_fops.llvm.8325593422554671469 Due to cross-file inlining, the static variable 'bpf_link_fops' in syscall.c is used by a function in another file. To avoid potential duplicated names, the llvm added suffix '.llvm.<hash>' ([2]) to 'bpf_link_fops' variable. Such renaming caused a problem in libbpf if 'bpf_link_fops' is used in bpf prog as a ksym but 'bpf_link_fops' does not match any symbol in /proc/kallsyms. To fix this issue, libbpf needs to understand that suffix '.llvm.<hash>' is caused by clang lto kernel and to process such symbols properly. With latest bpf-next code base built with CONFIG_LTO_CLANG_THIN, I cannot reproduce the above failure any more. But such an issue could happen with other symbols or in the future for bpf_link_fops symbol. For example, with my current kernel, I got the following from /proc/kallsyms: ffffffff84782154 d __func__.net_ratelimit.llvm.6135436931166841955 ffffffff85f0a500 d tk_core.llvm.726630847145216431 ffffffff85fdb960 d __fs_reclaim_map.llvm.10487989720912350772 ffffffff864c7300 d fake_dst_ops.llvm.54750082607048300 I could not easily create a selftest to test newly-added libbpf functionality with a static C test since I do not know which symbol is cross-file inlined. But based on my particular kernel, the following test change can run successfully. > diff --git a/tools/testing/selftests/bpf/prog_tests/ksyms.c b/tools/testing/selftests/bpf/prog_tests/ksyms.c > index 6a86d1f07800..904a103f7b1d 100644 > --- a/tools/testing/selftests/bpf/prog_tests/ksyms.c > +++ b/tools/testing/selftests/bpf/prog_tests/ksyms.c > @@ -42,6 +42,7 @@ void test_ksyms(void) > ASSERT_EQ(data->out__bpf_link_fops, link_fops_addr, "bpf_link_fops"); > ASSERT_EQ(data->out__bpf_link_fops1, 0, "bpf_link_fops1"); > ASSERT_EQ(data->out__btf_size, btf_size, "btf_size"); > + ASSERT_NEQ(data->out__fake_dst_ops, 0, "fake_dst_ops"); > ASSERT_EQ(data->out__per_cpu_start, per_cpu_start_addr, "__per_cpu_start"); > > cleanup: > diff --git a/tools/testing/selftests/bpf/progs/test_ksyms.c b/tools/testing/selftests/bpf/progs/test_ksyms.c > index 6c9cbb5a3bdf..fe91eef54b66 100644 > --- a/tools/testing/selftests/bpf/progs/test_ksyms.c > +++ b/tools/testing/selftests/bpf/progs/test_ksyms.c > @@ -9,11 +9,13 @@ __u64 out__bpf_link_fops = -1; > __u64 out__bpf_link_fops1 = -1; > __u64 out__btf_size = -1; > __u64 out__per_cpu_start = -1; > +__u64 out__fake_dst_ops = -1; > > extern const void bpf_link_fops __ksym; > extern const void __start_BTF __ksym; > extern const void __stop_BTF __ksym; > extern const void __per_cpu_start __ksym; > +extern const void fake_dst_ops __ksym; > /* non-existing symbol, weak, default to zero */ > extern const void bpf_link_fops1 __ksym __weak; > > @@ -23,6 +25,7 @@ int handler(const void *ctx) > out__bpf_link_fops = (__u64)&bpf_link_fops; > out__btf_size = (__u64)(&__stop_BTF - &__start_BTF); > out__per_cpu_start = (__u64)&__per_cpu_start; > + out__fake_dst_ops = (__u64)&fake_dst_ops; > > out__bpf_link_fops1 = (__u64)&bpf_link_fops1; This patch fixed the issue in libbpf such that the suffix '.llvm.<hash>' will be ignored during comparison of bpf prog ksym vs. symbols in /proc/kallsyms, this resolved the issue. Currently, only static variables in /proc/kallsyms are checked with '.llvm.<hash>' suffix since in bpf programs function ksyms with '.llvm.<hash>' suffix are most likely kfunc's and unlikely to be cross-file inlined. Note that currently kernel does not support gcc build with lto. [1] https://lore.kernel.org/bpf/20240302165017.1627295-1-yonghong.song@linux.dev/ [2] https://github.com/llvm/llvm-project/blob/release/18.x/llvm/include/llvm/IR/ModuleSummaryIndex.h#L1714-L1719Signed-off-by:
-
Yonghong Song authored
Currently libbpf_kallsyms_parse() function is declared as a global function but actually it is not a API and there is no external users in bpftool/bpf-selftests. So let us mark the function as static. Signed-off-by:
-
Yonghong Song authored
Replace CHECK with ASSERT macros for ksyms tests. This test failed earlier with clang lto kernel, but the issue is gone with latest code base. But replacing CHECK with ASSERT still improves code as ASSERT is preferred in selftests. Signed-off-by:
-
Martin KaFai Lau authored
This patch adds a test to ensure all static tcp-cc kfuncs is visible to the struct_ops bpf programs. It is checked by successfully loading the struct_ops programs calling these tcp-cc kfuncs. This patch needs to enable the CONFIG_TCP_CONG_DCTCP and the CONFIG_TCP_CONG_BBR. Signed-off-by:
-
Martin KaFai Lau authored
The commit 7aae231a ("bpf: tcp: Limit calling some tcp cc functions to CONFIG_DYNAMIC_FTRACE") added CONFIG_DYNAMIC_FTRACE guard because pahole was only generating btf for ftrace-able functions. The ftrace filter had already been removed from pahole, so the CONFIG_DYNAMIC_FTRACE guard can be removed. The commit 569c484f ("bpf: Limit static tcp-cc functions in the .BTF_ids list to x86") has added CONFIG_X86 guard because it failed the powerpc arch which prepended a "." to the local static function, so "cubictcp_init" becomes ".cubictcp_init". "__bpf_kfunc" has been added to kfunc since then and it uses the __unused compiler attribute. There is an existing "__bpf_kfunc static u32 bpf_kfunc_call_test_static_unused_arg(u32 arg, u32 unused)" test in bpf_testmod.c to cover the static kfunc case. cross compile on ppc64 with CONFIG_DYNAMIC_FTRACE disabled: > readelf -s vmlinux | grep cubictcp_ 56938: c00000000144fd00 184 FUNC LOCAL DEFAULT 2 cubictcp_cwnd_event [<localentry>: 8] 56939: c00000000144fdb8 200 FUNC LOCAL DEFAULT 2 cubictcp_recalc_[...] [<localentry>: 8] 56940: c00000000144fe80 296 FUNC LOCAL DEFAULT 2 cubictcp_init [<localentry>: 8] 56941: c00000000144ffa8 228 FUNC LOCAL DEFAULT 2 cubictcp_state [<localentry>: 8] 56942: c00000000145008c 1908 FUNC LOCAL DEFAULT 2 cubictcp_cong_avoid [<localentry>: 8] 56943: c000000001450800 1644 FUNC LOCAL DEFAULT 2 cubictcp_acked [<localentry>: 8] > bpftool btf dump file vmlinux | grep cubictcp_ [51540] FUNC 'cubictcp_acked' type_id=38137 linkage=static [51541] FUNC 'cubictcp_cong_avoid' type_id=38122 linkage=static [51543] FUNC 'cubictcp_cwnd_event' type_id=51542 linkage=static [51544] FUNC 'cubictcp_init' type_id=9186 linkage=static [51545] FUNC 'cubictcp_recalc_ssthresh' type_id=35021 linkage=static [51547] FUNC 'cubictcp_state' type_id=38141 linkage=static The patch removed both config guards. Cc: Jiri Olsa <jolsa@kernel.org> Signed-off-by:
-
Yafang Shao authored
Following the recent upgrade of one of our BPF programs, we encountered significant latency spikes affecting other applications running on the same host. After thorough investigation, we identified that these spikes were primarily caused by the prolonged duration required to free a non-preallocated htab with approximately 2 million keys. Notably, our kernel configuration lacks the presence of CONFIG_PREEMPT. In scenarios where kernel execution extends excessively, other threads might be starved of CPU time, resulting in latency issues across the system. To mitigate this, we've adopted a proactive approach by incorporating cond_resched() calls within the kernel code. This ensures that during lengthy kernel operations, the scheduler is invoked periodically to provide opportunities for other threads to execute. Signed-off-by:
Yafang Shao <laoar.shao@gmail.com> Acked-by:
-
Alexei Starovoitov authored
Andrii Nakryiko says: ==================== bench: fast in-kernel triggering benchmarks Remove "legacy" triggering benchmarks which rely on syscalls (and thus syscall overhead is a noticeable part of benchmark, unfortunately). Replace them with faster versions that rely on triggering BPF programs in-kernel through another simple "driver" BPF program. See patch #2 with comparison results. raw_tp/tp/fmodret benchmarks required adding a simple kfunc in kernel to be able to trigger a simple tracepoint from BPF program (plus it is also allowed to be replaced by fmod_ret programs). This limits raw_tp/tp/fmodret benchmarks to new kernels only, but it keeps bench tool itself very portable and most of other benchmarks will still work on wide variety of kernels without the need to worry about building and deploying custom kernel module. See patches #5 and #6 for details. v1->v2: - move new TP closer to BPF test run code; - rename/move kfunc and register it for fmod_rets (Alexei); - limit --trig-batch-iters param to [1, 1000] (Alexei). ==================== Link: https://lore.kernel.org/r/20240326162151.3981687-1-andrii@kernel.orgSigned-off-by:
-
Andrii Nakryiko authored
Utilize bpf_modify_return_test_tp() kfunc to have a fast way to trigger tp/raw_tp/fmodret programs from another BPF program, which gives us comparable batched benchmarks to (batched) kprobe/fentry benchmarks. We don't switch kprobe/fentry batched benchmarks to this kfunc to make bench tool usable on older kernels as well. Signed-off-by:
-
Andrii Nakryiko authored
Add a simple bpf_modify_return_test_tp() kfunc, available to all program types, that is useful for various testing and benchmarking scenarios, as it allows to trigger most tracing BPF program types from BPF side, allowing to do complex testing and benchmarking scenarios. It is also attachable to for fmod_ret programs, making it a good and simple way to trigger fmod_ret program under test/benchmark. Signed-off-by:
-
Andrii Nakryiko authored
Instead of front-loading all possible benchmarking BPF programs for trigger benchmarks, explicitly specify which BPF programs are used by specific benchmark and load only it. This allows to be more flexible in supporting older kernels, where some program types might not be possible to load (e.g., those that rely on newly added kfunc). Signed-off-by:
-
Andrii Nakryiko authored
Remove "legacy" benchmarks triggered by syscalls in favor of newly added in-kernel/batched benchmarks. Drop -batched suffix now as well. Next patch will restore "feature parity" by adding back tp/raw_tp/fmodret benchmarks based on in-kernel kfunc approach. Signed-off-by:
-
Andrii Nakryiko authored
Existing kprobe/fentry triggering benchmarks have 1-to-1 mapping between one syscall execution and BPF program run. While we use a fast get_pgid() syscall, syscall overhead can still be non-trivial. This patch adds kprobe/fentry set of benchmarks significantly amortizing the cost of syscall vs actual BPF triggering overhead. We do this by employing BPF_PROG_TEST_RUN command to trigger "driver" raw_tp program which does a tight parameterized loop calling cheap BPF helper (bpf_get_numa_node_id()), to which kprobe/fentry programs are attached for benchmarking. This way 1 bpf() syscall causes N executions of BPF program being benchmarked. N defaults to 100, but can be adjusted with --trig-batch-iters CLI argument. For comparison we also implement a new baseline program that instead of triggering another BPF program just does N atomic per-CPU counter increments, establishing the limit for all other types of program within this batched benchmarking setup. Taking the final set of benchmarks added in this patch set (including tp/raw_tp/fmodret, added in later patch), and keeping for now "legacy" syscall-driven benchmarks, we can capture all triggering benchmarks in one place for comparison, before we remove the legacy ones (and rename xxx-batched into just xxx). $ benchs/run_bench_trigger.sh usermode-count : 79.500 ± 0.024M/s kernel-count : 49.949 ± 0.081M/s syscall-count : 9.009 ± 0.007M/s fentry-batch : 31.002 ± 0.015M/s fexit-batch : 20.372 ± 0.028M/s fmodret-batch : 21.651 ± 0.659M/s rawtp-batch : 36.775 ± 0.264M/s tp-batch : 19.411 ± 0.248M/s kprobe-batch : 12.949 ± 0.220M/s kprobe-multi-batch : 15.400 ± 0.007M/s kretprobe-batch : 5.559 ± 0.011M/s kretprobe-multi-batch: 5.861 ± 0.003M/s fentry-legacy : 8.329 ± 0.004M/s fexit-legacy : 6.239 ± 0.003M/s fmodret-legacy : 6.595 ± 0.001M/s rawtp-legacy : 8.305 ± 0.004M/s tp-legacy : 6.382 ± 0.001M/s kprobe-legacy : 5.528 ± 0.003M/s kprobe-multi-legacy : 5.864 ± 0.022M/s kretprobe-legacy : 3.081 ± 0.001M/s kretprobe-multi-legacy: 3.193 ± 0.001M/s Note how xxx-batch variants are measured with significantly higher throughput, even though it's exactly the same in-kernel overhead. As such, results can be compared only between benchmarks of the same kind (syscall vs batched): fentry-legacy : 8.329 ± 0.004M/s fentry-batch : 31.002 ± 0.015M/s kprobe-multi-legacy : 5.864 ± 0.022M/s kprobe-multi-batch : 15.400 ± 0.007M/s Note also that syscall-count is setting a theoretical limit for syscall-triggered benchmarks, while kernel-count is setting similar limits for batch variants. usermode-count is a happy and unachievable case of user space counting without doing any syscalls, and is mostly the measure of CPU speed for such a trivial benchmark. As was mentioned, tp/raw_tp/fmodret require kernel-side kfunc to produce similar benchmark, which we address in a separate patch. Note that run_bench_trigger.sh allows to override a list of benchmarks to run, which is very useful for performance work. Cc: Jiri Olsa <jolsa@kernel.org> Signed-off-by:
-
Andrii Nakryiko authored
Rename uprobe-base to more precise usermode-count (it will match other baseline-like benchmarks, kernel-count and syscall-count). Also use BENCH_TRIG_USERMODE() macro to define all usermode-based triggering benchmarks, which include usermode-count and uprobe/uretprobe benchmarks. Signed-off-by:
-
Haiyue Wang authored
Use the well defined helper sizeof_field() to calculate the size of a struct member, instead of doing custom calculations. Signed-off-by:
Haiyue Wang <haiyue.wang@intel.com> Link: https://lore.kernel.org/r/20240327065334.8140-1-haiyue.wang@intel.comSigned-off-by:
-
Mykyta Yatsenko authored
BPF verifier emits "unknown func" message when given BPF program type does not support BPF helper. This message may be confusing for users, as important context that helper is unknown only to current program type is not provided. This patch changes message to "program of this type cannot use helper " and aligns dependent code in libbpf and tests. Any suggestions on improving/changing this message are welcome. Signed-off-by:
Mykyta Yatsenko <yatsenko@meta.com> Acked-by:
-
Anton Protopopov authored
The struct bpf_fib_lookup should not grow outside of its 64 bytes. Add a static assert to validate this. Suggested-by:
David Ahern <dsahern@kernel.org> Signed-off-by:
-
