riscv/qemu - qemu - Gitea: Git with a cup of tea

118747 Commits

68 Branches

448 Tags

581 MiB

Tree: 86f847a39a

Commit Graph

Author	SHA1	Message	Date
Emilio Cota	1ff4a81bd3	tcg: use QTree instead of GTree qemu-user can hang in a multi-threaded fork. One common reason is that when creating a TB, between fork and exec we manipulate a GTree whose memory allocator (GSlice) is not fork-safe. Although POSIX does not mandate it, the system's allocator (e.g. tcmalloc, libc malloc) is probably fork-safe. Fix some of these hangs by using QTree, which uses the system's allocator regardless of the Glib version that we used at configuration time. Tested with the test program in the original bug report, i.e.: ``` void garble() { int pid = fork(); if (pid == 0) { exit(0); } else { int wstatus; waitpid(pid, &wstatus, 0); } } void supragarble(unsigned depth) { if (depth == 0) return ; std::thread a(supragarble, depth-1); std::thread b(supragarble, depth-1); garble(); a.join(); b.join(); } int main() { supragarble(10); } ``` Resolves: https://gitlab.com/qemu-project/qemu/-/issues/285 Reported-by: Valentin David <me@valentindavid.com> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Emilio Cota <cota@braap.org> Message-Id: <20230205163758.416992-3-cota@braap.org> [rth: Add QEMU_DISABLE_CFI for all callback using functions.] Signed-off-by: Richard Henderson <richard.henderson@linaro.org>	3 years ago
Emilio Cota	e3feb2cc22	util: import GTree as QTree The only reason to add this implementation is to control the memory allocator used. Some users (e.g. TCG) cannot work reliably in multi-threaded environments (e.g. forking in user-mode) with GTree's allocator, GSlice. See https://gitlab.com/qemu-project/qemu/-/issues/285 for details. Importing GTree is a temporary workaround until GTree migrates away from GSlice. This implementation is identical to that in glib v2.75.0, except that we don't import recent additions to the API nor deprecated API calls, none of which are used in QEMU. I've imported tests from glib and added a benchmark just to make sure that performance is similar. Note: it cannot be identical because (1) we are not using GSlice, (2) we use different compilation flags (e.g. -fPIC) and (3) we're linking statically. $ cat /proc/cpuinfo\| grep 'model name' \| head -1 model name : AMD Ryzen 7 PRO 5850U with Radeon Graphics $ echo '0' \| sudo tee /sys/devices/system/cpu/cpufreq/boost $ tests/bench/qtree-bench Tree Op 32 1024 4096 131072 1048576 ------------------------------------------------------------------------------------------------ GTree Lookup 83.23 43.08 25.31 19.40 16.22 QTree Lookup 113.42 (1.36x) 53.83 (1.25x) 28.38 (1.12x) 17.64 (0.91x) 13.04 (0.80x) GTree Insert 44.23 29.37 25.83 19.49 17.03 QTree Insert 46.87 (1.06x) 25.62 (0.87x) 24.29 (0.94x) 16.83 (0.86x) 12.97 (0.76x) GTree Remove 53.27 35.15 31.43 24.64 16.70 QTree Remove 57.32 (1.08x) 41.76 (1.19x) 38.37 (1.22x) 29.30 (1.19x) 15.07 (0.90x) GTree RemoveAll 135.44 127.52 126.72 120.11 64.34 QTree RemoveAll 127.15 (0.94x) 110.37 (0.87x) 107.97 (0.85x) 97.13 (0.81x) 55.10 (0.86x) GTree Traverse 277.71 276.09 272.78 246.72 98.47 QTree Traverse 370.33 (1.33x) 411.97 (1.49x) 400.23 (1.47x) 262.82 (1.07x) 78.52 (0.80x) ------------------------------------------------------------------------------------------------ As a sanity check, the same benchmark when Glib's version is >= $glib_dropped_gslice_version (i.e. QTree == GTree): Tree Op 32 1024 4096 131072 1048576 ------------------------------------------------------------------------------------------------ GTree Lookup 82.72 43.09 24.18 19.73 16.09 QTree Lookup 81.82 (0.99x) 43.10 (1.00x) 24.20 (1.00x) 19.76 (1.00x) 16.26 (1.01x) GTree Insert 45.07 29.62 26.34 19.90 17.18 QTree Insert 45.72 (1.01x) 29.60 (1.00x) 26.38 (1.00x) 19.71 (0.99x) 17.20 (1.00x) GTree Remove 54.48 35.36 31.77 24.97 16.95 QTree Remove 54.46 (1.00x) 35.32 (1.00x) 31.77 (1.00x) 24.91 (1.00x) 17.15 (1.01x) GTree RemoveAll 140.68 127.36 125.43 121.45 68.20 QTree RemoveAll 140.65 (1.00x) 127.64 (1.00x) 125.01 (1.00x) 121.73 (1.00x) 67.06 (0.98x) GTree Traverse 278.68 276.05 266.75 251.65 104.93 QTree Traverse 278.31 (1.00x) 275.78 (1.00x) 266.42 (1.00x) 247.89 (0.99x) 104.58 (1.00x) ------------------------------------------------------------------------------------------------ Signed-off-by: Emilio Cota <cota@braap.org> Message-Id: <20230205163758.416992-2-cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>	3 years ago

Author

SHA1

Message

Date

Emilio Cota

1ff4a81bd3

tcg: use QTree instead of GTree

qemu-user can hang in a multi-threaded fork. One common
reason is that when creating a TB, between fork and exec
we manipulate a GTree whose memory allocator (GSlice) is
not fork-safe.

Although POSIX does not mandate it, the system's allocator
(e.g. tcmalloc, libc malloc) is probably fork-safe.

Fix some of these hangs by using QTree, which uses the system's
allocator regardless of the Glib version that we used at
configuration time.

Tested with the test program in the original bug report, i.e.:
```

void garble() {
  int pid = fork();
  if (pid == 0) {
    exit(0);
  } else {
    int wstatus;
    waitpid(pid, &wstatus, 0);
  }
}

void supragarble(unsigned depth) {
  if (depth == 0)
    return ;

  std::thread a(supragarble, depth-1);
  std::thread b(supragarble, depth-1);
  garble();
  a.join();
  b.join();
}

int main() {
  supragarble(10);
}
```

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/285
Reported-by: Valentin David <me@valentindavid.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Emilio Cota <cota@braap.org>
Message-Id: <20230205163758.416992-3-cota@braap.org>
[rth: Add QEMU_DISABLE_CFI for all callback using functions.]
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

3 years ago

Emilio Cota

e3feb2cc22

util: import GTree as QTree

The only reason to add this implementation is to control the memory allocator
used. Some users (e.g. TCG) cannot work reliably in multi-threaded
environments (e.g. forking in user-mode) with GTree's allocator, GSlice.
See https://gitlab.com/qemu-project/qemu/-/issues/285 for details.

Importing GTree is a temporary workaround until GTree migrates away
from GSlice.

This implementation is identical to that in glib v2.75.0, except that
we don't import recent additions to the API nor deprecated API calls,
none of which are used in QEMU.

I've imported tests from glib and added a benchmark just to
make sure that performance is similar. Note: it cannot be identical
because (1) we are not using GSlice, (2) we use different compilation flags
(e.g. -fPIC) and (3) we're linking statically.

$ cat /proc/cpuinfo| grep 'model name' | head -1
model name      : AMD Ryzen 7 PRO 5850U with Radeon Graphics
$ echo '0' | sudo tee /sys/devices/system/cpu/cpufreq/boost
$ tests/bench/qtree-bench

 Tree         Op      32            1024            4096          131072         1048576
------------------------------------------------------------------------------------------------
GTree     Lookup   83.23           43.08           25.31           19.40           16.22
QTree     Lookup  113.42 (1.36x)   53.83 (1.25x)   28.38 (1.12x)   17.64 (0.91x)   13.04 (0.80x)
GTree     Insert   44.23           29.37           25.83           19.49           17.03
QTree     Insert   46.87 (1.06x)   25.62 (0.87x)   24.29 (0.94x)   16.83 (0.86x)   12.97 (0.76x)
GTree     Remove   53.27           35.15           31.43           24.64           16.70
QTree     Remove   57.32 (1.08x)   41.76 (1.19x)   38.37 (1.22x)   29.30 (1.19x)   15.07 (0.90x)
GTree  RemoveAll  135.44          127.52          126.72          120.11           64.34
QTree  RemoveAll  127.15 (0.94x)  110.37 (0.87x)  107.97 (0.85x)   97.13 (0.81x)   55.10 (0.86x)
GTree   Traverse  277.71          276.09          272.78          246.72           98.47
QTree   Traverse  370.33 (1.33x)  411.97 (1.49x)  400.23 (1.47x)  262.82 (1.07x)   78.52 (0.80x)
------------------------------------------------------------------------------------------------

As a sanity check, the same benchmark when Glib's version
is >= $glib_dropped_gslice_version (i.e. QTree == GTree):

 Tree         Op      32            1024            4096          131072         1048576
------------------------------------------------------------------------------------------------
GTree     Lookup   82.72           43.09           24.18           19.73           16.09
QTree     Lookup   81.82 (0.99x)   43.10 (1.00x)   24.20 (1.00x)   19.76 (1.00x)   16.26 (1.01x)
GTree     Insert   45.07           29.62           26.34           19.90           17.18
QTree     Insert   45.72 (1.01x)   29.60 (1.00x)   26.38 (1.00x)   19.71 (0.99x)   17.20 (1.00x)
GTree     Remove   54.48           35.36           31.77           24.97           16.95
QTree     Remove   54.46 (1.00x)   35.32 (1.00x)   31.77 (1.00x)   24.91 (1.00x)   17.15 (1.01x)
GTree  RemoveAll  140.68          127.36          125.43          121.45           68.20
QTree  RemoveAll  140.65 (1.00x)  127.64 (1.00x)  125.01 (1.00x)  121.73 (1.00x)   67.06 (0.98x)
GTree   Traverse  278.68          276.05          266.75          251.65          104.93
QTree   Traverse  278.31 (1.00x)  275.78 (1.00x)  266.42 (1.00x)  247.89 (0.99x)  104.58 (1.00x)
------------------------------------------------------------------------------------------------

Signed-off-by: Emilio Cota <cota@braap.org>
Message-Id: <20230205163758.416992-2-cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

3 years ago

2 Commits (86f847a39aef93bcfbea65a702ba76762ae54d61)