Em Tue, Jun 08, 2021 at 12:50:13AM +0530, Deepak Kumar Mishra escreveu:
> CMakeLists.txt does not allow creation of static library and link applications
> accordingly.
>
> Creation of SHARED and STATIC should be allowed using -DBUILD_SHARED_LIBS
> If -DBUILD_SHARED_LIBS option is not supplied, CMakeLists.txt sets it to ON.
>
> Ex:
> cmake -D__LIB=lib -DBUILD_SHARED_LIBS=OFF ..
> cmake -D__LIB=lib -DBUILD_SHARED_LIBS=ON ..
Had to do some fixups due to a previous patch touching CMakeLists.txt,
please check below.
I tested it and added some performance notes.
Thanks!
- Arnaldo
commit aa2027708659f172780f85698f14303c7de6a1d2
Author: Deepak Kumar Mishra <deepakkumar.mishra@xxxxxxx>
Date: Tue Jun 8 00:50:13 2021 +0530
CMakeLists.txt: Enable SHARED and STATIC lib creation
CMakeLists.txt does not allow creation of static library and link applications
accordingly.
Creation of SHARED and STATIC should be allowed using -DBUILD_SHARED_LIBS
If -DBUILD_SHARED_LIBS option is not supplied, CMakeLists.txt sets it to ON.
Ex:
$ cmake -D__LIB=lib -DBUILD_SHARED_LIBS=OFF ..
$ cmake -D__LIB=lib -DBUILD_SHARED_LIBS=ON ..
Committer testing:
I had to fixup it a bit due to changes related to allowing building with
libbpf-devel, test resuts:
With the default, i.e. creating libdwarves.so and using it:
⬢[acme@toolbox pahole]$ rm -f vmlinux.btf ; perf stat -r5 pahole -j vmlinux.btf vmlinux && perf stat -r5 btfdiff vmlinux vmlinux.btf
Performance counter stats for 'pahole -j vmlinux.btf vmlinux' (5 runs):
8,612.69 msec task-clock:u # 1.014 CPUs utilized ( +- 0.68% )
0 context-switches:u # 0.000 /sec
0 cpu-migrations:u # 0.000 /sec
775,702 page-faults:u # 91.330 K/sec ( +- 0.00% )
33,720,048,514 cycles:u # 3.970 GHz ( +- 0.38% ) (83.33%)
689,752,139 stalled-cycles-frontend:u # 2.05% frontend cycles idle ( +- 2.57% ) (83.33%)
5,070,723,369 stalled-cycles-backend:u # 15.04% backend cycles idle ( +- 1.20% ) (83.34%)
77,270,640,084 instructions:u # 2.29 insn per cycle
# 0.07 stalled cycles per insn ( +- 0.01% ) (83.34%)
18,164,028,242 branches:u # 2.139 G/sec ( +- 0.04% ) (83.33%)
150,194,338 branch-misses:u # 0.83% of all branches ( +- 0.17% ) (83.32%)
8.4979 +- 0.0590 seconds time elapsed ( +- 0.69% )
Performance counter stats for 'btfdiff vmlinux vmlinux.btf' (5 runs):
7,001.79 msec task-clock:u # 1.006 CPUs utilized ( +- 1.02% )
0 context-switches:u # 0.000 /sec
0 cpu-migrations:u # 0.000 /sec
727,980 page-faults:u # 104.736 K/sec ( +- 0.00% )
26,912,307,448 cycles:u # 3.872 GHz ( +- 0.31% ) (83.33%)
509,467,882 stalled-cycles-frontend:u # 1.90% frontend cycles idle ( +- 2.44% ) (83.33%)
3,602,777,263 stalled-cycles-backend:u # 13.43% backend cycles idle ( +- 1.34% ) (83.32%)
66,192,815,701 instructions:u # 2.47 insn per cycle
# 0.06 stalled cycles per insn ( +- 0.01% ) (83.35%)
15,753,663,095 branches:u # 2.267 G/sec ( +- 0.02% ) (83.35%)
98,345,696 branch-misses:u # 0.62% of all branches ( +- 0.14% ) (83.34%)
6.9586 +- 0.0720 seconds time elapsed ( +- 1.03% )
⬢[acme@toolbox pahole]$
Then building with:
⬢[acme@toolbox pahole]$ rm -rf build ; mkdir build ; cd build ; cmake -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE=Release .. ; cd .. ; make -C build
⬢[acme@toolbox pahole]$ ldd build/pahole
linux-vdso.so.1 (0x00007ffff0172000)
libdw.so.1 => /lib64/libdw.so.1 (0x00007fdd3905d000)
libelf.so.1 => /lib64/libelf.so.1 (0x00007fdd39042000)
libz.so.1 => /lib64/libz.so.1 (0x00007fdd39028000)
libc.so.6 => /lib64/libc.so.6 (0x00007fdd38e59000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fdd38e52000)
libzstd.so.1 => /lib64/libzstd.so.1 (0x00007fdd38d5c000)
liblzma.so.5 => /lib64/liblzma.so.5 (0x00007fdd38d2e000)
libbz2.so.1 => /lib64/libbz2.so.1 (0x00007fdd38d1b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fdd38cfa000)
/lib64/ld-linux-x86-64.so.2 (0x00007fdd390ff000)
⬢[acme@toolbox pahole]$
We get some performance improvement:
- First is encoding detached BTF, i.e. read DWARF, generate BTF.
- Second is about loading both DWARF and BTF, producing output for both, that must match.
⬢[acme@toolbox pahole]$ rm -f vmlinux.btf ; perf stat -r5 pahole -j vmlinux.btf vmlinux && perf stat -r5 btfdiff vmlinux vmlinux.btf
Performance counter stats for 'pahole -j vmlinux.btf vmlinux' (5 runs):
8,566.34 msec task-clock:u # 1.025 CPUs utilized ( +- 0.85% )
0 context-switches:u # 0.000 /sec
0 cpu-migrations:u # 0.000 /sec
775,685 page-faults:u # 92.865 K/sec ( +- 0.00% )
33,333,991,512 cycles:u # 3.991 GHz ( +- 0.23% ) (83.34%)
799,187,919 stalled-cycles-frontend:u # 2.42% frontend cycles idle ( +- 1.95% ) (83.33%)
5,157,722,792 stalled-cycles-backend:u # 15.61% backend cycles idle ( +- 1.25% ) (83.33%)
76,273,972,066 instructions:u # 2.31 insn per cycle
# 0.07 stalled cycles per insn ( +- 0.03% ) (83.33%)
17,843,388,470 branches:u # 2.136 G/sec ( +- 0.05% ) (83.33%)
150,507,690 branch-misses:u # 0.84% of all branches ( +- 0.46% ) (83.33%)
8.3561 +- 0.0729 seconds time elapsed ( +- 0.87% )
Performance counter stats for 'btfdiff vmlinux vmlinux.btf' (5 runs):
6,685.39 msec task-clock:u # 0.986 CPUs utilized ( +- 0.54% )
0 context-switches:u # 0.000 /sec
0 cpu-migrations:u # 0.000 /sec
727,931 page-faults:u # 107.475 K/sec ( +- 0.00% )
26,149,371,139 cycles:u # 3.861 GHz ( +- 0.15% ) (83.36%)
525,918,808 stalled-cycles-frontend:u # 2.00% frontend cycles idle ( +- 2.96% ) (83.33%)
3,547,589,546 stalled-cycles-backend:u # 13.51% backend cycles idle ( +- 1.28% ) (83.35%)
65,389,507,702 instructions:u # 2.49 insn per cycle
# 0.06 stalled cycles per insn ( +- 0.03% ) (83.29%)
15,553,175,605 branches:u # 2.296 G/sec ( +- 0.02% ) (83.34%)
97,812,971 branch-misses:u # 0.63% of all branches ( +- 0.23% ) (83.34%)
6.7784 +- 0.0359 seconds time elapsed ( +- 0.53% )
⬢[acme@toolbox pahole]$
Signed-off-by: Deepak Kumar Mishra <deepakkumar.mishra@xxxxxxx>
Tested-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
Cc: Qais Yousef <qais.yousef@xxxxxxx>
Signed-off-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
diff --git a/CMakeLists.txt b/CMakeLists.txt
index df16b8ec3031f3e6..669ea59167c7ef65 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -47,6 +47,11 @@ endif (NOT CMAKE_BUILD_TYPE)
set(CMAKE_C_FLAGS_DEBUG "-Wall -Werror -ggdb -O0")
set(CMAKE_C_FLAGS_RELEASE "-Wall -O2")
+if (NOT DEFINED BUILD_SHARED_LIBS)
+ set (BUILD_SHARED_LIBS ON)
+ message(STATUS "Setting BUILD_SHARED_LIBS = ${BUILD_SHARED_LIBS}")
+endif (NOT DEFINED BUILD_SHARED_LIBS)
+
# Just for grepping, DWARVES_VERSION isn't used anywhere anymore
# add_definitions(-D_GNU_SOURCE -DDWARVES_VERSION="v1.21")
add_definitions(-D_GNU_SOURCE -DDWARVES_MAJOR_VERSION=1)
@@ -111,18 +116,18 @@ set(dwarves_LIB_SRCS dwarves.c dwarves_fprintf.c gobuffer.c strings.c
if (NOT LIBBPF_FOUND)
list(APPEND dwarves_LIB_SRCS $<TARGET_OBJECTS:bpf>)
endif()
-add_library(dwarves SHARED ${dwarves_LIB_SRCS})
+add_library(dwarves ${dwarves_LIB_SRCS})
set_target_properties(dwarves PROPERTIES VERSION 1.0.0 SOVERSION 1)
set_target_properties(dwarves PROPERTIES INTERFACE_LINK_LIBRARIES "")
target_link_libraries(dwarves ${DWARF_LIBRARIES} ${ZLIB_LIBRARIES} ${LIBBPF_LIBRARIES})
set(dwarves_emit_LIB_SRCS dwarves_emit.c)
-add_library(dwarves_emit SHARED ${dwarves_emit_LIB_SRCS})
+add_library(dwarves_emit ${dwarves_emit_LIB_SRCS})
set_target_properties(dwarves_emit PROPERTIES VERSION 1.0.0 SOVERSION 1)
target_link_libraries(dwarves_emit dwarves)
set(dwarves_reorganize_LIB_SRCS dwarves_reorganize.c)
-add_library(dwarves_reorganize SHARED ${dwarves_reorganize_LIB_SRCS})
+add_library(dwarves_reorganize ${dwarves_reorganize_LIB_SRCS})
set_target_properties(dwarves_reorganize PROPERTIES VERSION 1.0.0 SOVERSION 1)
target_link_libraries(dwarves_reorganize dwarves)
@@ -169,8 +174,8 @@ target_link_libraries(syscse dwarves)
install(TARGETS codiff ctracer dtagnames pahole pdwtags
pfunct pglobal prefcnt scncopy syscse RUNTIME DESTINATION
${CMAKE_INSTALL_PREFIX}/bin)
-install(TARGETS dwarves LIBRARY DESTINATION ${LIB_INSTALL_DIR})
-install(TARGETS dwarves dwarves_emit dwarves_reorganize LIBRARY DESTINATION ${LIB_INSTALL_DIR})
+install(TARGETS dwarves LIBRARY DESTINATION ${LIB_INSTALL_DIR} ARCHIVE DESTINATION ${LIB_INSTALL_DIR})
+install(TARGETS dwarves dwarves_emit dwarves_reorganize LIBRARY DESTINATION ${LIB_INSTALL_DIR} ARCHIVE DESTINATION ${LIB_INSTALL_DIR})
install(FILES dwarves.h dwarves_emit.h dwarves_reorganize.h
dutil.h gobuffer.h list.h rbtree.h pahole_strings.h
btf_encoder.h config.h ctf_encoder.h ctf.h
