| Message ID | 20231115004932.650702-1-ak@linux.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] kbuild: Add inline-account tool | expand |
On Wed, Nov 15, 2023 at 9:49 AM Andi Kleen <ak@linux.intel.com> wrote: This is a standalone script, which is unrelated to kbuild. Please change the subject to "scripts: Add inline-account tool" > > A common cause of binary code bloat is excessive inlining. Traditional > tools (like nm --size-sort -t d) don't address that directly because > they only see the final functions, but don't know about inlines. > > This patch adds inline-account that makes it easy to track that down > by accounting code bytes to all functions visible in the debug information, > as well as code lines. > > Here are some examples: > > Show all inlines that increase code size by >1K in the core scheduler: > > $ inline-account.py --min-bytes=1000 kernel/sched/core.o > Total code bytes seen 75690 > > Code bytes by functions: > Function Total Avg Num > rq_pin_lock 1401 (0.02%) 35 39 > __sched_setscheduler 1277 (0.02%) 41 31 > perf_fetch_caller_regs 1012 (0.01%) 17 58 > > Code bytes by nearby source line blocks: > prefix /home/ak/lsrc/git/linux/ > Line Total > kernel/sched/sched.h:1610 1387 (0.02%) > include/trace/events/sched.h:16 1172 (0.02%) > include/trace/events/sched.h:222 1058 (0.01%) > > This indicates that rq_pin_lock should likely be not inline, > and perhaps perf_fetch_caller_regs not either. > > Note that not all large inlines are necessary bloat. If there is only necessarily bloat? > a single call site it isn't bloat (the tool currently cannot distinguish > that case). For example it is commonly seen with syscall definitions > that use single large inlines with only a single caller. In the example > above I think it's the case with __sched_setscheduler. > > Show the >1K inlines in lib/maple_tree.o, which for some reason > comes in at a incredible 73k of code size: > > $ inline-account.py --min-bytes 1000 lib/maple_tree.o > Total code bytes seen 73578 > > Code bytes by functions: > Function Total Avg Num > mas_mab_cp 5537 (0.08%) 37 149 > mas_pop_node 3798 (0.05%) 28 131 > ma_slots 2368 (0.03%) 14 162 > ma_pivots 2353 (0.03%) 10 222 > mas_destroy_rebalance 2056 (0.03%) 42 48 > mas_start 1661 (0.02%) 13 125 > mas_set_parent 1454 (0.02%) 20 72 > mas_set_alloc_req 1410 (0.02%) 17 80 > mte_node_type 1360 (0.02%) 5 228 > mas_data_end 1189 (0.02%) 16 74 > mte_to_node 1085 (0.01%) 3 276 > mas_split 1053 (0.01%) 65 16 > mas_topiary_replace 1033 (0.01%) 38 27 > mas_root_expand 1001 (0.01%) 35 28 > > Code bytes by nearby source line blocks: > prefix /home/ak/lsrc/git/linux/ > Line Total > lib/maple_tree.c:210 1360 (0.02%) > include/trace/events/maple_tree.h:80 1283 (0.02%) > lib/maple_tree.c:649 1193 (0.02%) > lib/maple_tree.c:288 1097 (0.01%) > > It's clear there is a lot of potential for shrinking here, as a quick > experiment shows: > > $ size lib/maple_tree.o > text data bss dec hex filename > 72257 5312 8 77577 12f09 lib/maple_tree.o > $ sed -i -e s/__always_inline// -e 's/ inline/ /' lib/maple_tree.c > $ make -s lib/maple_tree.o > $ size lib/maple_tree.o > text data bss dec hex filename > 47774 4720 8 52502 cd16 lib/maple_tree.o > > 34% reduction just from trusting the compiler. Most of it seems > to come from abuse of __always_inline. I suppose a large scale > tree purge of that would give some decent binary size results. > > $ inline-account.py --show=5 kernel/workqueue.o > Total code bytes seen 40403 > > Code bytes by functions: > Function Total Avg Num > bitmap_copy 1477 (0.04%) 26 56 > show_pwq 912 (0.02%) 76 12 > workqueue_init_early 846 (0.02%) 29 29 > __flush_workqueue 753 (0.02%) 31 24 > alloc_and_link_pwqs 558 (0.01%) 69 8 > > Code bytes by nearby source line blocks: > prefix /home/ak/lsrc/git/linux/ > Line Total > include/linux/bitmap.h:268 1336 (0.03%) > include/trace/events/workqueue.h:23 1038 (0.03%) > include/trace/events/workqueue.h:108 732 (0.02%) > include/trace/events/workqueue.h:59 694 (0.02%) > include/trace/events/workqueue.h:82 670 (0.02%) > $ > > This is an interesting case because bitmap_copy is just > > static inline void bitmap_copy(unsigned long *dst, const unsigned long *src, > unsigned int nbits) > { > unsigned int len = BITS_TO_LONGS(nbits) * sizeof(unsigned long); > > if (small_const_nbits(nbits)) > *dst = *src; > else > memcpy(dst, src, len); > } > > memcpy (which is a macro) must sometimes generate a lot of > code. The small_const_nbits case definitely should be inlined though > because it's likely even smaller than a call. Would need > more investigation. > > The other large inlines are trace points. Perhaps there is something > there that could be done to shrink that a bit. > > Finally we can do a global accounting (currently with multiple runs): > > (ignore the percentage numbers since they are just for the local file) > > $ find -name '*.o' | xargs -n1 inline-account.py > a > $ sort -n -r -k 2 a | head -30 > ZSTD_count 81799 (0.19%) 32 2514 > ZSTD_count 52233 (0.25%) 33 1544 > kmalloc 43324 (0.00%) 12 3334 > pv_queued_spin_unlock 42027 (0.00%) 9 4580 > constant_test_bit 41667 (0.00%) 5 8005 > arch/x86/include/asm/paravirt.h:591 41044 (0.00%) > arch/x86/include/asm/bitops.h:207 40153 (0.00%) > __refcount_add 37968 (0.00%) 24 1532 > page_fixed_fake_head 36368 (0.00%) 19 1832 > include/linux/slab.h:599 35654 (0.00%) > arch/x86/include/asm/jump_label.h:27 35156 (0.00%) > spin_lock 32170 (0.00%) 10 3007 > __refcount_sub_and_test 32068 (0.00%) 17 1842 > include/linux/spinlock.h:351 31102 (0.00%) > arch_static_branch 30874 (0.00%) 4 7022 > get_current 30714 (0.00%) 9 3351 > arch/x86/include/asm/current.h:41 29912 (0.00%) > trace_trigger_soft_disabled 29814 (0.00%) 21 1368 > perf_fetch_caller_regs 27504 (0.00%) 16 1634 > ZSTD_storeSeq 26060 (0.06%) 30 862 > hid_map_usage 25582 (0.00%) 88 288 > ZSTD_compressBlock_lazy_generic 24953 (0.12%) 46 535 > ZSTD_compressBlock_lazy_generic 24953 (0.06%) 46 535 > paravirt_ret0 24152 (0.00%) 24152 1 > spin_unlock_irqrestore 23253 (0.00%) 10 2281 > include/linux/spinlock.h:406 22526 (0.00%) > ZSTD_RowFindBestMatch 21527 (0.10%) 23 922 > ZSTD_RowFindBestMatch 21527 (0.05%) 23 922 > __list_add 21209 (0.00%) 11 1851 > include/linux/refcount.h:283 20642 (0.00%) > > - So my kernel is spending around ~30K just for getting task_structs in > current. > - I'm sure ZSTD is great, but is it >200K in duplicated code worth great? > - page_fixed_fake_head probably shouldn't be inlined > - There might be some potential in out lining reference counts > (although that one might be truly performance critical) > - There's maybe some potential in shrinking trace point bloat? > > ... and more similar insights. > > Also of course there can be critical inlines that really need > to be inline in many sites for best performance. But that's rarely the case > if they are big because it's unlikely the small call overhead is making > a significant difference for a large chunk of code. > > In any case the tool is useful, so I think it deserves its place > in scripts/ > > Signed-off-by: Andi Kleen <ak@linux.intel.com> > > --- > > v2: Address review comments. Change balancing to be address based. > Add option to set objdump binary. > --- > scripts/inline-account.py | 201 ++++++++++++++++++++++++++++++++++++++ > 1 file changed, 201 insertions(+) > create mode 100755 scripts/inline-account.py > > diff --git a/scripts/inline-account.py b/scripts/inline-account.py > new file mode 100755 > index 000000000000..c32cb0547172 > --- /dev/null > +++ b/scripts/inline-account.py > @@ -0,0 +1,201 @@ > +#!/usr/bin/env python3 > +# account code bytes per source code / functions from objdump -Sl output > +# useful to find inline bloat > +# > +# SPDX-License-Identifier: GPL-2.0-only Please move this above. The first line is the shebang. The second line is SPDX. > +# Author: Andi Kleen > + > +import os > +import sys > +import re > +import argparse > +import bisect > +import multiprocessing Please sort imports alphabetically. > +from collections import Counter > +from functools import reduce, partial > + > +def get_args(): > + p = argparse.ArgumentParser( > + description=""" > +Account code bytes per source code / functions from objdump. > +Useful to find inline bloat. > + > +The line numbers are the beginning of a block, so the actual code can be later. > +Line numbers can be a also little off due to objdump bugs > +also some misaccounting can happen due to inexact gcc debug information. > +The number output for functions may account a single large function multiple > +times. program/object files need to be built with -g. > + > +This is somewhat slow due to objdump -S being slow. It helps to have > +plenty of cores.""") > + p.add_argument('--min-bytes', type=int, help='minimum bytes to report', default=100) > + p.add_argument('--threads', '-t', type=int, default=multiprocessing.cpu_count(), Please check that args.threads >= 1. If not, please show an error message. > + help='Number of objdump processes to run') > + p.add_argument('--verbose', '-v', action='store_true', help="Print more") > + p.add_argument('--show', type=int, help='Number of results to show') > + p.add_argument('--objdump', default="objdump", help="Set objdump binary to run") > + p.add_argument('file', help='object file/program as input') > + return p.parse_args() > + > +def get_syms(fn): > + syms = [] > + pc = None > + with os.popen("nm -n --print-size " + fn) as f: > + for l in f: > + n = l.split() > + if len(n) > 2 and n[2].upper() == "T": > + pc = int(n[0], 16) > + syms.append(pc) > + ln = int(n[1], 16) > + if not pc: > + sys.exit(fn + " has no symbols") "if not pc" becomes true in two cases: - pc is zero - pc is None Only the latter case is "no symbols" because a symbol address may become zero in relocatable ELF. if len(syms) == 0: ... will work correctly. And, "pc = None" is unneeded. > + syms.append(pc + ln) > + return syms > + > +class Account: > + def __init__(self): > + self.funcbytes = Counter() > + self.linebytes = Counter() > + self.funccount = Counter() > + self.nolinebytes = 0 > + self.nofuncbytes = 0 > + self.total = 0 > + > + def add(self, b): > + self.funcbytes += b.funcbytes > + self.linebytes += b.linebytes > + self.funccount += b.funccount > + self.nolinebytes += b.nolinebytes > + self.nofuncbytes += b.nofuncbytes > + self.total += b.total > + return self > + Since this script implements a class, please follow this blank line rule. https://peps.python.org/pep-0008/#blank-lines > +# dont add sys.exit here, causes deadlocks > +def account_range(args, r): > + a = Account() > + line = None > + func = None > + codefunc = None > + > + cmd = ("%s -Sl %s --start-address=%#x --stop-address=%#x" % > + (args.objdump, args.file, r[0], r[1])) > + if args.verbose: > + print(cmd) > + with os.popen(cmd) as f: > + for l in f: > + # 250: e8 00 00 00 00 callq 255 <proc_skip_spaces+0x5> > + m = re.match(r'\s*([0-9a-fA-F]+):\s+(.*)', l) > + if m: > + bytes = len(re.findall(r'[0-9a-f][0-9a-f] ', m.group(2))) > + if not func: > + a.nofuncbytes += bytes > + continue > + if not line: > + a.nolinebytes += bytes > + continue > + a.total += bytes > + a.funcbytes[func] += bytes > + a.linebytes[(file, line)] += bytes > + codefunc = func > + continue > + > + # sysctl_init(): > + m = re.match(r'([a-zA-Z_][a-zA-Z0-9_]*)\(\):$', l) > + if m: > + if codefunc and m.group(1) != codefunc: > + a.funccount[codefunc] += 1 > + codefunc = None > + func = m.group(1) > + continue > + > + # /sysctl.c:1666 > + m = re.match(r'^([^:]+):(\d+)$', l) > + if m: > + file, line = m.group(1), int(m.group(2)) > + continue > + > + if codefunc: > + a.funccount[codefunc] += 1 > + return a > + > +def get_boundaries(syms, sym_sizes, chunk): > + run = 0 > + boundaries = [syms[0]] > + for i, x in enumerate(sym_sizes): > + run += x > + if run >= chunk: > + boundaries.append(syms[i]) > + run = 0 > + boundaries.append(syms[-1]) > + return boundaries > + > + > +def process(args): > + # objdump -S is slow, so we parallelize > + > + # split symbol table into chunks for parallelization > + # we split on functions boundaries to avoid mis-accounting > + syms = get_syms(args.file) > + if len(syms) < 2: > + print("not enough symbols") > + return This message should go to stderr instead of stdout. sys.exit("not enough symbols") > + sym_sizes = [syms[x + 1] - syms[x] for x, _ in enumerate(syms[:-1])] > + sym_total = sum(sym_sizes) This is equivalent to sym_total = syms[-1] - syms[0] isn't it? > + chunk = max(int(sym_total / args.threads), 1) Do you know the "//" operator ? int(sym_total / args.threads) should be equivalent to sym_total // args.threads > + boundaries = get_boundaries(syms, sym_sizes, chunk) > + ranges = [(boundaries[x], boundaries[x+1]) for x in range(0, len(boundaries) - 1)] Very tedious code. And, your get_boundaries() helper uses less threads than args.threads. How about this to compute the ranges directly? def get_ranges(syms, threads): ranges = [] prev = syms[0] i = 1 while threads > 0: boundary = prev + 1 + (syms[-1] - prev - 1) // threads while syms[i] < boundary: i += 1 ranges.append((prev, syms[i])) prev = syms[i] threads -= 1 if (prev == syms[-1]): break return ranges ranges = get_ranges(syms, args.threads) > + assert ranges[0][0] == syms[0] > + assert ranges[-1][1] == syms[-1] > + > + # map-reduce > + account_func = partial(account_range, args) > + if args.threads == 1: > + al = list(map(account_func, ranges)) > + else: > + al = multiprocessing.Pool(args.threads).map(account_func, ranges) > + a = reduce(lambda a, b: a.add(b), al) > + > + print("Total code bytes seen", a.total) > + if args.verbose: > + print("Bytes with no function %d (%.2f%%)" % (a.nofuncbytes, 100.0*(float(a.nofuncbytes)/a.total))) > + print("Bytes with no lines %d (%.2f%%)" % (a.nolinebytes, 100.0*(float(a.nolinebytes)/a.total))) > + > + def sort_map(m): > + return sorted(list(m.keys()), key=lambda x: m[x], reverse=True) > + > + print("\nCode bytes by functions:") > + print("%-50s %-5s %-5s %-5s %-5s" % ("Function", "Total", "", "Avg", "Num")) > + for i, j in enumerate(sort_map(a.funcbytes)): > + if a.funcbytes[j] < args.min_bytes: > + break > + if args.show and i >= args.show: > + break > + print("%-50s %-5d (%.2f%%) %-5d %-5d" % ( > + j, > + a.funcbytes[j], > + a.funcbytes[j] / float(a.total), > + a.funcbytes[j] / a.funccount[j], > + a.funccount[j])) > + > + for j in list(a.linebytes.keys()): > + if a.linebytes[j] < args.min_bytes: > + del a.linebytes[j] > + > + prefix = os.path.commonprefix([x[0] for x in list(a.linebytes.keys())]) > + > + print("\nCode bytes by nearby source line blocks:") > + print("prefix", prefix) > + > + print("%-50s %-5s" % ("Line", "Total")) > + for i, j in enumerate(sort_map(a.linebytes)): > + if args.show and i >= args.show: > + break > + print("%-50s %-5d (%.2f%%)" % ( > + "%s:%d" % (j[0][len(prefix):], j[1]), > + a.linebytes[j], > + a.linebytes[j] / float(a.total))) > + if len(a.linebytes) == 0: > + print("Nothing found. enable CONFIG_DEBUG_INFO / -g?") I thnik this should be checked much earlier. If CONFIG_DEBUG_INFO is disabled, there is nothing to print. > + > +if __name__ == '__main__': > + process(get_args()) > -- > 2.41.0 > -- Best Regards Masahiro Yamada
On Mon, Nov 20, 2023 at 1:32 AM Masahiro Yamada <masahiroy@kernel.org> wrote: > > On Wed, Nov 15, 2023 at 9:49 AM Andi Kleen <ak@linux.intel.com> wrote: > > > This is a standalone script, which is unrelated to kbuild. > > Please change the subject to > > "scripts: Add inline-account tool" > > > > > > > > > A common cause of binary code bloat is excessive inlining. Traditional > > tools (like nm --size-sort -t d) don't address that directly because > > they only see the final functions, but don't know about inlines. > > > > This patch adds inline-account that makes it easy to track that down > > by accounting code bytes to all functions visible in the debug information, > > as well as code lines. > > > > Here are some examples: > > > > Show all inlines that increase code size by >1K in the core scheduler: > > > > $ inline-account.py --min-bytes=1000 kernel/sched/core.o > > Total code bytes seen 75690 > > > > Code bytes by functions: > > Function Total Avg Num > > rq_pin_lock 1401 (0.02%) 35 39 > > __sched_setscheduler 1277 (0.02%) 41 31 > > perf_fetch_caller_regs 1012 (0.01%) 17 58 > > > > Code bytes by nearby source line blocks: > > prefix /home/ak/lsrc/git/linux/ > > Line Total > > kernel/sched/sched.h:1610 1387 (0.02%) > > include/trace/events/sched.h:16 1172 (0.02%) > > include/trace/events/sched.h:222 1058 (0.01%) > > > > This indicates that rq_pin_lock should likely be not inline, > > and perhaps perf_fetch_caller_regs not either. > > > > Note that not all large inlines are necessary bloat. If there is only > > > necessarily bloat? > > > > > > > > > > > a single call site it isn't bloat (the tool currently cannot distinguish > > that case). For example it is commonly seen with syscall definitions > > that use single large inlines with only a single caller. In the example > > above I think it's the case with __sched_setscheduler. > > > > Show the >1K inlines in lib/maple_tree.o, which for some reason > > comes in at a incredible 73k of code size: > > > > $ inline-account.py --min-bytes 1000 lib/maple_tree.o > > Total code bytes seen 73578 > > > > Code bytes by functions: > > Function Total Avg Num > > mas_mab_cp 5537 (0.08%) 37 149 > > mas_pop_node 3798 (0.05%) 28 131 > > ma_slots 2368 (0.03%) 14 162 > > ma_pivots 2353 (0.03%) 10 222 > > mas_destroy_rebalance 2056 (0.03%) 42 48 > > mas_start 1661 (0.02%) 13 125 > > mas_set_parent 1454 (0.02%) 20 72 > > mas_set_alloc_req 1410 (0.02%) 17 80 > > mte_node_type 1360 (0.02%) 5 228 > > mas_data_end 1189 (0.02%) 16 74 > > mte_to_node 1085 (0.01%) 3 276 > > mas_split 1053 (0.01%) 65 16 > > mas_topiary_replace 1033 (0.01%) 38 27 > > mas_root_expand 1001 (0.01%) 35 28 > > > > Code bytes by nearby source line blocks: > > prefix /home/ak/lsrc/git/linux/ > > Line Total > > lib/maple_tree.c:210 1360 (0.02%) > > include/trace/events/maple_tree.h:80 1283 (0.02%) > > lib/maple_tree.c:649 1193 (0.02%) > > lib/maple_tree.c:288 1097 (0.01%) > > > > It's clear there is a lot of potential for shrinking here, as a quick > > experiment shows: > > > > $ size lib/maple_tree.o > > text data bss dec hex filename > > 72257 5312 8 77577 12f09 lib/maple_tree.o > > $ sed -i -e s/__always_inline// -e 's/ inline/ /' lib/maple_tree.c > > $ make -s lib/maple_tree.o > > $ size lib/maple_tree.o > > text data bss dec hex filename > > 47774 4720 8 52502 cd16 lib/maple_tree.o > > > > 34% reduction just from trusting the compiler. Most of it seems > > to come from abuse of __always_inline. I suppose a large scale > > tree purge of that would give some decent binary size results. > > > > $ inline-account.py --show=5 kernel/workqueue.o > > Total code bytes seen 40403 > > > > Code bytes by functions: > > Function Total Avg Num > > bitmap_copy 1477 (0.04%) 26 56 > > show_pwq 912 (0.02%) 76 12 > > workqueue_init_early 846 (0.02%) 29 29 > > __flush_workqueue 753 (0.02%) 31 24 > > alloc_and_link_pwqs 558 (0.01%) 69 8 > > > > Code bytes by nearby source line blocks: > > prefix /home/ak/lsrc/git/linux/ > > Line Total > > include/linux/bitmap.h:268 1336 (0.03%) > > include/trace/events/workqueue.h:23 1038 (0.03%) > > include/trace/events/workqueue.h:108 732 (0.02%) > > include/trace/events/workqueue.h:59 694 (0.02%) > > include/trace/events/workqueue.h:82 670 (0.02%) > > $ > > > > This is an interesting case because bitmap_copy is just > > > > static inline void bitmap_copy(unsigned long *dst, const unsigned long *src, > > unsigned int nbits) > > { > > unsigned int len = BITS_TO_LONGS(nbits) * sizeof(unsigned long); > > > > if (small_const_nbits(nbits)) > > *dst = *src; > > else > > memcpy(dst, src, len); > > } > > > > memcpy (which is a macro) must sometimes generate a lot of > > code. The small_const_nbits case definitely should be inlined though > > because it's likely even smaller than a call. Would need > > more investigation. > > > > The other large inlines are trace points. Perhaps there is something > > there that could be done to shrink that a bit. > > > > Finally we can do a global accounting (currently with multiple runs): > > > > (ignore the percentage numbers since they are just for the local file) > > > > $ find -name '*.o' | xargs -n1 inline-account.py > a > > $ sort -n -r -k 2 a | head -30 > > ZSTD_count 81799 (0.19%) 32 2514 > > ZSTD_count 52233 (0.25%) 33 1544 > > kmalloc 43324 (0.00%) 12 3334 > > pv_queued_spin_unlock 42027 (0.00%) 9 4580 > > constant_test_bit 41667 (0.00%) 5 8005 > > arch/x86/include/asm/paravirt.h:591 41044 (0.00%) > > arch/x86/include/asm/bitops.h:207 40153 (0.00%) > > __refcount_add 37968 (0.00%) 24 1532 > > page_fixed_fake_head 36368 (0.00%) 19 1832 > > include/linux/slab.h:599 35654 (0.00%) > > arch/x86/include/asm/jump_label.h:27 35156 (0.00%) > > spin_lock 32170 (0.00%) 10 3007 > > __refcount_sub_and_test 32068 (0.00%) 17 1842 > > include/linux/spinlock.h:351 31102 (0.00%) > > arch_static_branch 30874 (0.00%) 4 7022 > > get_current 30714 (0.00%) 9 3351 > > arch/x86/include/asm/current.h:41 29912 (0.00%) > > trace_trigger_soft_disabled 29814 (0.00%) 21 1368 > > perf_fetch_caller_regs 27504 (0.00%) 16 1634 > > ZSTD_storeSeq 26060 (0.06%) 30 862 > > hid_map_usage 25582 (0.00%) 88 288 > > ZSTD_compressBlock_lazy_generic 24953 (0.12%) 46 535 > > ZSTD_compressBlock_lazy_generic 24953 (0.06%) 46 535 > > paravirt_ret0 24152 (0.00%) 24152 1 > > spin_unlock_irqrestore 23253 (0.00%) 10 2281 > > include/linux/spinlock.h:406 22526 (0.00%) > > ZSTD_RowFindBestMatch 21527 (0.10%) 23 922 > > ZSTD_RowFindBestMatch 21527 (0.05%) 23 922 > > __list_add 21209 (0.00%) 11 1851 > > include/linux/refcount.h:283 20642 (0.00%) > > > > - So my kernel is spending around ~30K just for getting task_structs in > > current. > > - I'm sure ZSTD is great, but is it >200K in duplicated code worth great? > > - page_fixed_fake_head probably shouldn't be inlined > > - There might be some potential in out lining reference counts > > (although that one might be truly performance critical) > > - There's maybe some potential in shrinking trace point bloat? > > > > ... and more similar insights. > > > > Also of course there can be critical inlines that really need > > to be inline in many sites for best performance. But that's rarely the case > > if they are big because it's unlikely the small call overhead is making > > a significant difference for a large chunk of code. > > > > In any case the tool is useful, so I think it deserves its place > > in scripts/ > > > > Signed-off-by: Andi Kleen <ak@linux.intel.com> > > > > --- > > > > v2: Address review comments. Change balancing to be address based. > > Add option to set objdump binary. > > --- > > scripts/inline-account.py | 201 ++++++++++++++++++++++++++++++++++++++ > > 1 file changed, 201 insertions(+) > > create mode 100755 scripts/inline-account.py > > > > diff --git a/scripts/inline-account.py b/scripts/inline-account.py > > new file mode 100755 > > index 000000000000..c32cb0547172 > > --- /dev/null > > +++ b/scripts/inline-account.py > > @@ -0,0 +1,201 @@ > > +#!/usr/bin/env python3 > > +# account code bytes per source code / functions from objdump -Sl output > > +# useful to find inline bloat > > +# > > +# SPDX-License-Identifier: GPL-2.0-only > > > Please move this above. > > The first line is the shebang. > The second line is SPDX. > > > > > > > +# Author: Andi Kleen > > + > > +import os > > +import sys > > +import re > > +import argparse > > +import bisect > > +import multiprocessing > > > Please sort imports alphabetically. > > > > > +from collections import Counter > > +from functools import reduce, partial > > + > > +def get_args(): > > + p = argparse.ArgumentParser( > > + description=""" > > +Account code bytes per source code / functions from objdump. > > +Useful to find inline bloat. > > + > > +The line numbers are the beginning of a block, so the actual code can be later. > > +Line numbers can be a also little off due to objdump bugs > > +also some misaccounting can happen due to inexact gcc debug information. > > +The number output for functions may account a single large function multiple > > +times. program/object files need to be built with -g. > > + > > +This is somewhat slow due to objdump -S being slow. It helps to have > > +plenty of cores.""") > > + p.add_argument('--min-bytes', type=int, help='minimum bytes to report', default=100) > > + p.add_argument('--threads', '-t', type=int, default=multiprocessing.cpu_count(), > > > Please check that args.threads >= 1. > > If not, please show an error message. > > > > > > > + help='Number of objdump processes to run') > > + p.add_argument('--verbose', '-v', action='store_true', help="Print more") > > + p.add_argument('--show', type=int, help='Number of results to show') > > + p.add_argument('--objdump', default="objdump", help="Set objdump binary to run") > > + p.add_argument('file', help='object file/program as input') > > + return p.parse_args() > > + > > +def get_syms(fn): > > + syms = [] > > + pc = None > > + with os.popen("nm -n --print-size " + fn) as f: > > + for l in f: > > + n = l.split() > > + if len(n) > 2 and n[2].upper() == "T": > > + pc = int(n[0], 16) > > + syms.append(pc) > > + ln = int(n[1], 16) > > + if not pc: > > + sys.exit(fn + " has no symbols") > > > "if not pc" becomes true in two cases: > - pc is zero > - pc is None > > > Only the latter case is "no symbols" because > a symbol address may become zero in relocatable ELF. > > > > if len(syms) == 0: > ... > > will work correctly. > > And, "pc = None" is unneeded. > > > > > > > > > > + syms.append(pc + ln) > > + return syms > > + > > +class Account: > > + def __init__(self): > > + self.funcbytes = Counter() > > + self.linebytes = Counter() > > + self.funccount = Counter() > > + self.nolinebytes = 0 > > + self.nofuncbytes = 0 > > + self.total = 0 > > + > > + def add(self, b): > > + self.funcbytes += b.funcbytes > > + self.linebytes += b.linebytes > > + self.funccount += b.funccount > > + self.nolinebytes += b.nolinebytes > > + self.nofuncbytes += b.nofuncbytes > > + self.total += b.total > > + return self > > + > > > Since this script implements a class, > please follow this blank line rule. > > > https://peps.python.org/pep-0008/#blank-lines > > > > > > > > > > > +# dont add sys.exit here, causes deadlocks > > +def account_range(args, r): > > + a = Account() > > + line = None > > + func = None > > + codefunc = None > > + > > + cmd = ("%s -Sl %s --start-address=%#x --stop-address=%#x" % > > + (args.objdump, args.file, r[0], r[1])) > > + if args.verbose: > > + print(cmd) > > + with os.popen(cmd) as f: > > + for l in f: > > + # 250: e8 00 00 00 00 callq 255 <proc_skip_spaces+0x5> > > + m = re.match(r'\s*([0-9a-fA-F]+):\s+(.*)', l) > > + if m: > > + bytes = len(re.findall(r'[0-9a-f][0-9a-f] ', m.group(2))) > > + if not func: > > + a.nofuncbytes += bytes > > + continue > > + if not line: > > + a.nolinebytes += bytes > > + continue > > + a.total += bytes > > + a.funcbytes[func] += bytes > > + a.linebytes[(file, line)] += bytes > > + codefunc = func > > + continue > > + > > + # sysctl_init(): > > + m = re.match(r'([a-zA-Z_][a-zA-Z0-9_]*)\(\):$', l) > > + if m: > > + if codefunc and m.group(1) != codefunc: > > + a.funccount[codefunc] += 1 > > + codefunc = None > > + func = m.group(1) > > + continue > > + > > + # /sysctl.c:1666 > > + m = re.match(r'^([^:]+):(\d+)$', l) > > + if m: > > + file, line = m.group(1), int(m.group(2)) > > + continue > > + > > + if codefunc: > > + a.funccount[codefunc] += 1 > > + return a > > + > > +def get_boundaries(syms, sym_sizes, chunk): > > + run = 0 > > + boundaries = [syms[0]] > > + for i, x in enumerate(sym_sizes): > > + run += x > > + if run >= chunk: > > + boundaries.append(syms[i]) > > + run = 0 > > + boundaries.append(syms[-1]) > > + return boundaries > > + > > + > > +def process(args): > > + # objdump -S is slow, so we parallelize > > + > > + # split symbol table into chunks for parallelization > > + # we split on functions boundaries to avoid mis-accounting > > + syms = get_syms(args.file) > > + if len(syms) < 2: > > + print("not enough symbols") > > + return > > This message should go to stderr instead of stdout. > > > sys.exit("not enough symbols") > Or, you can remove this check entirely if you check the length in get_syms(). get_syms() requires at least one element from the loop, and append() one more at the end. So, get_syms() always has two or more elements.
> > > + sym_sizes = [syms[x + 1] - syms[x] for x, _ in enumerate(syms[:-1])] > > + sym_total = sum(sym_sizes) > > > This is equivalent to > > sym_total = syms[-1] - syms[0] > > isn't it? No it's not when there are gaps in the symbols, which always happens in the kernel. The previous version used the range but that led to inbalances. > def get_ranges(syms, threads): > > ranges = [] > prev = syms[0] > i = 1 > > while threads > 0: > > boundary = prev + 1 + (syms[-1] - prev - 1) // threads > > while syms[i] < boundary: > i += 1 Note the boundary is not the address, but the sum of symbol sizes -Andi
On Tue, Nov 21, 2023 at 5:08 AM Andi Kleen <ak@linux.intel.com> wrote: > > > > > > + sym_sizes = [syms[x + 1] - syms[x] for x, _ in enumerate(syms[:-1])] > > > + sym_total = sum(sym_sizes) > > > > > > This is equivalent to > > > > sym_total = syms[-1] - syms[0] > > > > isn't it? > > No it's not when there are gaps in the symbols, which always happens in > the kernel. The previous version used the range but that led to inbalances. I do not understand because there is nothing to handle gaps in your code. get_syms() appends only 'pc' into the list. Then, sym_sizes = [syms[x + 1] - syms[x] for x, _ in enumerate(syms[:-1])] computes each symbol size by: (address of the next symbol) - (address of the current symbol). So, your code is equivalent to: sym_total = syms[-1] - syms[0] > > def get_ranges(syms, threads): > > > > ranges = [] > > prev = syms[0] > > i = 1 > > > > while threads > 0: > > > > boundary = prev + 1 + (syms[-1] - prev - 1) // threads > > > > while syms[i] < boundary: > > i += 1 > > > Note the boundary is not the address, but the sum of symbol sizes > > > -Andi -- Best Regards Masahiro Yamada
diff --git a/scripts/inline-account.py b/scripts/inline-account.py new file mode 100755 index 000000000000..c32cb0547172 --- /dev/null +++ b/scripts/inline-account.py @@ -0,0 +1,201 @@ +#!/usr/bin/env python3 +# account code bytes per source code / functions from objdump -Sl output +# useful to find inline bloat +# +# SPDX-License-Identifier: GPL-2.0-only +# Author: Andi Kleen + +import os +import sys +import re +import argparse +import bisect +import multiprocessing +from collections import Counter +from functools import reduce, partial + +def get_args(): + p = argparse.ArgumentParser( + description=""" +Account code bytes per source code / functions from objdump. +Useful to find inline bloat. + +The line numbers are the beginning of a block, so the actual code can be later. +Line numbers can be a also little off due to objdump bugs +also some misaccounting can happen due to inexact gcc debug information. +The number output for functions may account a single large function multiple +times. program/object files need to be built with -g. + +This is somewhat slow due to objdump -S being slow. It helps to have +plenty of cores.""") + p.add_argument('--min-bytes', type=int, help='minimum bytes to report', default=100) + p.add_argument('--threads', '-t', type=int, default=multiprocessing.cpu_count(), + help='Number of objdump processes to run') + p.add_argument('--verbose', '-v', action='store_true', help="Print more") + p.add_argument('--show', type=int, help='Number of results to show') + p.add_argument('--objdump', default="objdump", help="Set objdump binary to run") + p.add_argument('file', help='object file/program as input') + return p.parse_args() + +def get_syms(fn): + syms = [] + pc = None + with os.popen("nm -n --print-size " + fn) as f: + for l in f: + n = l.split() + if len(n) > 2 and n[2].upper() == "T": + pc = int(n[0], 16) + syms.append(pc) + ln = int(n[1], 16) + if not pc: + sys.exit(fn + " has no symbols") + syms.append(pc + ln) + return syms + +class Account: + def __init__(self): + self.funcbytes = Counter() + self.linebytes = Counter() + self.funccount = Counter() + self.nolinebytes = 0 + self.nofuncbytes = 0 + self.total = 0 + + def add(self, b): + self.funcbytes += b.funcbytes + self.linebytes += b.linebytes + self.funccount += b.funccount + self.nolinebytes += b.nolinebytes + self.nofuncbytes += b.nofuncbytes + self.total += b.total + return self + +# dont add sys.exit here, causes deadlocks +def account_range(args, r): + a = Account() + line = None + func = None + codefunc = None + + cmd = ("%s -Sl %s --start-address=%#x --stop-address=%#x" % + (args.objdump, args.file, r[0], r[1])) + if args.verbose: + print(cmd) + with os.popen(cmd) as f: + for l in f: + # 250: e8 00 00 00 00 callq 255 <proc_skip_spaces+0x5> + m = re.match(r'\s*([0-9a-fA-F]+):\s+(.*)', l) + if m: + bytes = len(re.findall(r'[0-9a-f][0-9a-f] ', m.group(2))) + if not func: + a.nofuncbytes += bytes + continue + if not line: + a.nolinebytes += bytes + continue + a.total += bytes + a.funcbytes[func] += bytes + a.linebytes[(file, line)] += bytes + codefunc = func + continue + + # sysctl_init(): + m = re.match(r'([a-zA-Z_][a-zA-Z0-9_]*)\(\):$', l) + if m: + if codefunc and m.group(1) != codefunc: + a.funccount[codefunc] += 1 + codefunc = None + func = m.group(1) + continue + + # /sysctl.c:1666 + m = re.match(r'^([^:]+):(\d+)$', l) + if m: + file, line = m.group(1), int(m.group(2)) + continue + + if codefunc: + a.funccount[codefunc] += 1 + return a + +def get_boundaries(syms, sym_sizes, chunk): + run = 0 + boundaries = [syms[0]] + for i, x in enumerate(sym_sizes): + run += x + if run >= chunk: + boundaries.append(syms[i]) + run = 0 + boundaries.append(syms[-1]) + return boundaries + + +def process(args): + # objdump -S is slow, so we parallelize + + # split symbol table into chunks for parallelization + # we split on functions boundaries to avoid mis-accounting + syms = get_syms(args.file) + if len(syms) < 2: + print("not enough symbols") + return + sym_sizes = [syms[x + 1] - syms[x] for x, _ in enumerate(syms[:-1])] + sym_total = sum(sym_sizes) + chunk = max(int(sym_total / args.threads), 1) + boundaries = get_boundaries(syms, sym_sizes, chunk) + ranges = [(boundaries[x], boundaries[x+1]) for x in range(0, len(boundaries) - 1)] + assert ranges[0][0] == syms[0] + assert ranges[-1][1] == syms[-1] + + # map-reduce + account_func = partial(account_range, args) + if args.threads == 1: + al = list(map(account_func, ranges)) + else: + al = multiprocessing.Pool(args.threads).map(account_func, ranges) + a = reduce(lambda a, b: a.add(b), al) + + print("Total code bytes seen", a.total) + if args.verbose: + print("Bytes with no function %d (%.2f%%)" % (a.nofuncbytes, 100.0*(float(a.nofuncbytes)/a.total))) + print("Bytes with no lines %d (%.2f%%)" % (a.nolinebytes, 100.0*(float(a.nolinebytes)/a.total))) + + def sort_map(m): + return sorted(list(m.keys()), key=lambda x: m[x], reverse=True) + + print("\nCode bytes by functions:") + print("%-50s %-5s %-5s %-5s %-5s" % ("Function", "Total", "", "Avg", "Num")) + for i, j in enumerate(sort_map(a.funcbytes)): + if a.funcbytes[j] < args.min_bytes: + break + if args.show and i >= args.show: + break + print("%-50s %-5d (%.2f%%) %-5d %-5d" % ( + j, + a.funcbytes[j], + a.funcbytes[j] / float(a.total), + a.funcbytes[j] / a.funccount[j], + a.funccount[j])) + + for j in list(a.linebytes.keys()): + if a.linebytes[j] < args.min_bytes: + del a.linebytes[j] + + prefix = os.path.commonprefix([x[0] for x in list(a.linebytes.keys())]) + + print("\nCode bytes by nearby source line blocks:") + print("prefix", prefix) + + print("%-50s %-5s" % ("Line", "Total")) + for i, j in enumerate(sort_map(a.linebytes)): + if args.show and i >= args.show: + break + print("%-50s %-5d (%.2f%%)" % ( + "%s:%d" % (j[0][len(prefix):], j[1]), + a.linebytes[j], + a.linebytes[j] / float(a.total))) + if len(a.linebytes) == 0: + print("Nothing found. enable CONFIG_DEBUG_INFO / -g?") + +if __name__ == '__main__': + process(get_args())
A common cause of binary code bloat is excessive inlining. Traditional tools (like nm --size-sort -t d) don't address that directly because they only see the final functions, but don't know about inlines. This patch adds inline-account that makes it easy to track that down by accounting code bytes to all functions visible in the debug information, as well as code lines. Here are some examples: Show all inlines that increase code size by >1K in the core scheduler: $ inline-account.py --min-bytes=1000 kernel/sched/core.o Total code bytes seen 75690 Code bytes by functions: Function Total Avg Num rq_pin_lock 1401 (0.02%) 35 39 __sched_setscheduler 1277 (0.02%) 41 31 perf_fetch_caller_regs 1012 (0.01%) 17 58 Code bytes by nearby source line blocks: prefix /home/ak/lsrc/git/linux/ Line Total kernel/sched/sched.h:1610 1387 (0.02%) include/trace/events/sched.h:16 1172 (0.02%) include/trace/events/sched.h:222 1058 (0.01%) This indicates that rq_pin_lock should likely be not inline, and perhaps perf_fetch_caller_regs not either. Note that not all large inlines are necessary bloat. If there is only a single call site it isn't bloat (the tool currently cannot distinguish that case). For example it is commonly seen with syscall definitions that use single large inlines with only a single caller. In the example above I think it's the case with __sched_setscheduler. Show the >1K inlines in lib/maple_tree.o, which for some reason comes in at a incredible 73k of code size: $ inline-account.py --min-bytes 1000 lib/maple_tree.o Total code bytes seen 73578 Code bytes by functions: Function Total Avg Num mas_mab_cp 5537 (0.08%) 37 149 mas_pop_node 3798 (0.05%) 28 131 ma_slots 2368 (0.03%) 14 162 ma_pivots 2353 (0.03%) 10 222 mas_destroy_rebalance 2056 (0.03%) 42 48 mas_start 1661 (0.02%) 13 125 mas_set_parent 1454 (0.02%) 20 72 mas_set_alloc_req 1410 (0.02%) 17 80 mte_node_type 1360 (0.02%) 5 228 mas_data_end 1189 (0.02%) 16 74 mte_to_node 1085 (0.01%) 3 276 mas_split 1053 (0.01%) 65 16 mas_topiary_replace 1033 (0.01%) 38 27 mas_root_expand 1001 (0.01%) 35 28 Code bytes by nearby source line blocks: prefix /home/ak/lsrc/git/linux/ Line Total lib/maple_tree.c:210 1360 (0.02%) include/trace/events/maple_tree.h:80 1283 (0.02%) lib/maple_tree.c:649 1193 (0.02%) lib/maple_tree.c:288 1097 (0.01%) It's clear there is a lot of potential for shrinking here, as a quick experiment shows: $ size lib/maple_tree.o text data bss dec hex filename 72257 5312 8 77577 12f09 lib/maple_tree.o $ sed -i -e s/__always_inline// -e 's/ inline/ /' lib/maple_tree.c $ make -s lib/maple_tree.o $ size lib/maple_tree.o text data bss dec hex filename 47774 4720 8 52502 cd16 lib/maple_tree.o 34% reduction just from trusting the compiler. Most of it seems to come from abuse of __always_inline. I suppose a large scale tree purge of that would give some decent binary size results. $ inline-account.py --show=5 kernel/workqueue.o Total code bytes seen 40403 Code bytes by functions: Function Total Avg Num bitmap_copy 1477 (0.04%) 26 56 show_pwq 912 (0.02%) 76 12 workqueue_init_early 846 (0.02%) 29 29 __flush_workqueue 753 (0.02%) 31 24 alloc_and_link_pwqs 558 (0.01%) 69 8 Code bytes by nearby source line blocks: prefix /home/ak/lsrc/git/linux/ Line Total include/linux/bitmap.h:268 1336 (0.03%) include/trace/events/workqueue.h:23 1038 (0.03%) include/trace/events/workqueue.h:108 732 (0.02%) include/trace/events/workqueue.h:59 694 (0.02%) include/trace/events/workqueue.h:82 670 (0.02%) $ This is an interesting case because bitmap_copy is just static inline void bitmap_copy(unsigned long *dst, const unsigned long *src, unsigned int nbits) { unsigned int len = BITS_TO_LONGS(nbits) * sizeof(unsigned long); if (small_const_nbits(nbits)) *dst = *src; else memcpy(dst, src, len); } memcpy (which is a macro) must sometimes generate a lot of code. The small_const_nbits case definitely should be inlined though because it's likely even smaller than a call. Would need more investigation. The other large inlines are trace points. Perhaps there is something there that could be done to shrink that a bit. Finally we can do a global accounting (currently with multiple runs): (ignore the percentage numbers since they are just for the local file) $ find -name '*.o' | xargs -n1 inline-account.py > a $ sort -n -r -k 2 a | head -30 ZSTD_count 81799 (0.19%) 32 2514 ZSTD_count 52233 (0.25%) 33 1544 kmalloc 43324 (0.00%) 12 3334 pv_queued_spin_unlock 42027 (0.00%) 9 4580 constant_test_bit 41667 (0.00%) 5 8005 arch/x86/include/asm/paravirt.h:591 41044 (0.00%) arch/x86/include/asm/bitops.h:207 40153 (0.00%) __refcount_add 37968 (0.00%) 24 1532 page_fixed_fake_head 36368 (0.00%) 19 1832 include/linux/slab.h:599 35654 (0.00%) arch/x86/include/asm/jump_label.h:27 35156 (0.00%) spin_lock 32170 (0.00%) 10 3007 __refcount_sub_and_test 32068 (0.00%) 17 1842 include/linux/spinlock.h:351 31102 (0.00%) arch_static_branch 30874 (0.00%) 4 7022 get_current 30714 (0.00%) 9 3351 arch/x86/include/asm/current.h:41 29912 (0.00%) trace_trigger_soft_disabled 29814 (0.00%) 21 1368 perf_fetch_caller_regs 27504 (0.00%) 16 1634 ZSTD_storeSeq 26060 (0.06%) 30 862 hid_map_usage 25582 (0.00%) 88 288 ZSTD_compressBlock_lazy_generic 24953 (0.12%) 46 535 ZSTD_compressBlock_lazy_generic 24953 (0.06%) 46 535 paravirt_ret0 24152 (0.00%) 24152 1 spin_unlock_irqrestore 23253 (0.00%) 10 2281 include/linux/spinlock.h:406 22526 (0.00%) ZSTD_RowFindBestMatch 21527 (0.10%) 23 922 ZSTD_RowFindBestMatch 21527 (0.05%) 23 922 __list_add 21209 (0.00%) 11 1851 include/linux/refcount.h:283 20642 (0.00%) - So my kernel is spending around ~30K just for getting task_structs in current. - I'm sure ZSTD is great, but is it >200K in duplicated code worth great? - page_fixed_fake_head probably shouldn't be inlined - There might be some potential in out lining reference counts (although that one might be truly performance critical) - There's maybe some potential in shrinking trace point bloat? ... and more similar insights. Also of course there can be critical inlines that really need to be inline in many sites for best performance. But that's rarely the case if they are big because it's unlikely the small call overhead is making a significant difference for a large chunk of code. In any case the tool is useful, so I think it deserves its place in scripts/ Signed-off-by: Andi Kleen <ak@linux.intel.com> --- v2: Address review comments. Change balancing to be address based. Add option to set objdump binary. --- scripts/inline-account.py | 201 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 201 insertions(+) create mode 100755 scripts/inline-account.py