On Thu, 12 Mar 2026 09:44:13 +0100
Corinna Vinschen wrote:
> On Mar 12 16:55, Takashi Yano wrote:
> > Hi Corinna,
> >
> > On Wed, 11 Mar 2026 16:11:08 +0100
> > Corinna Vinschen wrote:
> > > On Mar 10 17:50, Takashi Yano wrote:
> > > > The current code waits for `sendsig` by `for` loop in sigproc.cc,
> > > > however, the wait time might be insufficient for recent CPU.
> > > > The current code is as follows.
> > > >
> > > > for (int i = 0; !p->sendsig && i < 10000; i++)
> > > > yield ();
> > > >
> > > > Due to this problem, in tcsh, the following command occasionally
> > > > cannot be terminated by Ctrl-C. This is because, SIGCONT does not
> > > > wake-up `sleep` process correctly.
> > > >
> > > > $ cat | sleep 100 &
> > > > $ fg
> > > > $ (type Ctrl-C)
> > > >
> > > > With this patch, the wait time for `sendsig` is guaranteed to be
> > > > up to 100ms instead of looping for 10000 times.
> > > >
> > > > Fixes: d584454c8231 ("* sigproc.cc (sig_send): Wait for dwProcessId to
> > > > be non-zero as well as sendsig.")
> > > > Signed-off-by: Takashi Yano <[email protected]>
> > > > Reviewed-by:
> > > > ---
> > > > winsup/cygwin/sigproc.cc | 3 ++-
> > > > 1 file changed, 2 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/winsup/cygwin/sigproc.cc b/winsup/cygwin/sigproc.cc
> > > > index 30779cf8e..0fd7ed3ba 100644
> > > > --- a/winsup/cygwin/sigproc.cc
> > > > +++ b/winsup/cygwin/sigproc.cc
> > > > @@ -646,7 +646,8 @@ sig_send (_pinfo *p, siginfo_t& si, _cygtls *tls)
> > > > {
> > > > HANDLE dupsig;
> > > > DWORD dwProcessId;
> > > > - for (int i = 0; !p->sendsig && i < 10000; i++)
> > > > + DWORD t0 = GetTickCount ();
> > >
> > > Again a case where GetTickCount is sufficient? I'd suggest
> > > to use GetTickCount64 instead.
> > >
> > > Other than that, LGTM.
> >
> > Thanks for reviewing.
> >
> > The usage of GetTickCount() like this is safe, because
> > DWORD (unsigned integer) wraps on overflow. For example:
> >
> > If the first call of GetTickCount() returns 0xFFFFFFF0,
> > and the second one returns 0x00000010, the result of
> > subtranction is:
> >
> > 0x00000010 - 0xFFFFFFF0 = 0x00000020
> >
> > Therefore, if yield() returns within 49 days, the result
> > will be as expected.
> >
> > Am I overlooking something?
>
> No, it's fine as is.
>
> It's a personal thing. I'm always cringing a bit when I see
> GetTickCount(). IMHO using GetTickCount64() is "the right thing to do"
> because the counter is 64 bit anyway and 64 bit int arithmetic isn't
> slower than 32 bit int arithmetic on 64 bit machines. Or, is it?
No. I have compared GetTickCount() and GetTickCount64(). The result is
that they are almost the same performance as you expected.
--
Takashi Yano <[email protected]>
