The commit e10f822a2b39 has a problem that CPU load gets high if pending signal is not processed successfully for a long time. With this patch, wait_sig() calls Sleep(1), rather than yield(), if the pending signal has not been processed successfully for a predetermined time to prevent CPU from high load.
Addresses: https://cygwin.com/pipermail/cygwin/2024-December/256884.html Fixes: e10f822a2b39 ("Cygwin: signal: Handle queued signal without explicit __SIGFLUSH") Reported-by: 凯夏 <walke...@gmail.com> Reviewed-by: Signed-off-by: Takashi Yano <takashi.y...@nifty.ne.jp> --- winsup/cygwin/sigproc.cc | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) diff --git a/winsup/cygwin/sigproc.cc b/winsup/cygwin/sigproc.cc index 59b4208a6..e01a67ebe 100644 --- a/winsup/cygwin/sigproc.cc +++ b/winsup/cygwin/sigproc.cc @@ -1345,6 +1345,12 @@ wait_sig (VOID *) hntdll = GetModuleHandle ("ntdll.dll"); + /* GetTickCount() here is enough because GetTickCount() - t0 does + not overflow until 49 days psss. Even if GetTickCount() overflows, + GetTickCount() - t0 returns correct value, since underflow in + unsigned wraps correctly. Pending a signal for more thtn 49 + days would be noncense. */ + DWORD t0 = GetTickCount (); for (;;) { DWORD nb; @@ -1354,7 +1360,10 @@ wait_sig (VOID *) else if (sigq.start.next && PeekNamedPipe (my_readsig, NULL, 0, NULL, &nb, NULL) && !nb) { - yield (); + if (GetTickCount () - t0 > 10) + Sleep (1); + else + yield (); pack.si.si_signo = __SIGFLUSH; } else if (!ReadFile (my_readsig, &pack, sizeof (pack), &nb, NULL)) @@ -1364,6 +1373,8 @@ wait_sig (VOID *) system_printf ("garbled signal pipe data nb %u, sig %d", nb, pack.si.si_signo); continue; } + if (pack.si.si_signo != __SIGFLUSH) + t0 = GetTickCount (); sigq.retry = false; /* Don't process signals when we start exiting */ -- 2.45.1
