Re: vmd testers: serial console hangs fix
Hello,
I've applied diff to 7.3. it works I can paste to serial console,
hitting tab, etc.. Everything seems to work as expected.
thanks and
regards
sashan
On Mon, Oct 09, 2023 at 10:51:05AM -0400, Dave Voutila wrote:
> Looking for folks that use the serial console connection in vmd(8) and
> experience the host CPU spikes & delays associated with things like
> hitting the up key (to cycle shell command history) or copy/paste in X.
>
> vmd(8) approximates edge triggered interrupts with the emulated i8259
> interrupt controller. Some core devices (serial console, rtc, timer) try
> "toggling" an irq...but it's pointless in how our emulated i8259 is
> implemented. The deassert does nothing for edge triggered interrupts
> other than causing race conditions and extra syscalls.
>
> I've tested on two of my machines and this resolves the issue I've been
> seeing off and on. It should fix the report[1] on bugs@ from months ago.
>
> -dv
>
> [1] https://marc.info/?l=openbsd-bugs=169669132915760=2
>
> diff refs/heads/master refs/heads/vmd-edge
> commit - 7869b2fdaac7e118bfd1783874fe25ce3b8b0f09
> commit + 00d448cdbe2461f27419aaf79520a0cef720aefc
> blob - b98e7bdc69ac6a12eb84eaaf97ec43ecdbe83733
> blob + 248e3b161e88be505a88411156dc57a2772bd4fa
> --- usr.sbin/vmd/i8253.c
> +++ usr.sbin/vmd/i8253.c
> @@ -371,7 +371,6 @@ i8253_fire(int fd, short type, void *arg)
> struct i8253_channel *ctr = (struct i8253_channel *)arg;
>
> vcpu_assert_pic_irq(ctr->vm_id, 0, 0);
> - vcpu_deassert_pic_irq(ctr->vm_id, 0, 0);
>
> if (ctr->mode != TIMER_INTTC) {
> timerclear();
> blob - 43dce7b10d1467a5b7ac7f3308d01e32b4d0b9ee
> blob + 4fc147b19c99627e386ab26355b8f90a6ae5872b
> --- usr.sbin/vmd/mc146818.c
> +++ usr.sbin/vmd/mc146818.c
> @@ -150,7 +150,6 @@ rtc_fireper(int fd, short type, void *arg)
> rtc.regs[MC_REGC] |= MC_REGC_PF;
>
> vcpu_assert_pic_irq((ptrdiff_t)arg, 0, 8);
> - vcpu_deassert_pic_irq((ptrdiff_t)arg, 0, 8);
>
> evtimer_add(, _tv);
> }
> blob - bc23876bf0392312335da0d00e143583a87549af
> blob + 98ed7dbecf2120ccbf9a16198ee479cb15aebc5f
> --- usr.sbin/vmd/ns8250.c
> +++ usr.sbin/vmd/ns8250.c
> @@ -82,7 +82,6 @@ ratelimit(int fd, short type, void *arg)
> com1_dev.regs.iir &= ~IIR_NOPEND;
>
> vcpu_assert_pic_irq(com1_dev.vmid, 0, com1_dev.irq);
> - vcpu_deassert_pic_irq(com1_dev.vmid, 0, com1_dev.irq);
> mutex_unlock(_dev.mutex);
> }
>
> @@ -160,7 +159,6 @@ com_rcv_event(int fd, short kind, void *arg)
> if ((com1_dev.regs.iir & IIR_NOPEND) == 0) {
> /* XXX: vcpu_id */
> vcpu_assert_pic_irq((uintptr_t)arg, 0, com1_dev.irq);
> - vcpu_deassert_pic_irq((uintptr_t)arg, 0, com1_dev.irq);
> }
>
> mutex_unlock(_dev.mutex);
>
Re: vmd testers: serial console hangs fix
Klemens Nanni writes:
> On Sun, Oct 15, 2023 at 07:51:00AM -0400, Dave Voutila wrote:
>>
>> Dave Voutila writes:
>>
>> > Looking for folks that use the serial console connection in vmd(8) and
>> > experience the host CPU spikes & delays associated with things like
>> > hitting the up key (to cycle shell command history) or copy/paste in X.
>> >
>> > vmd(8) approximates edge triggered interrupts with the emulated i8259
>> > interrupt controller. Some core devices (serial console, rtc, timer) try
>> > "toggling" an irq...but it's pointless in how our emulated i8259 is
>> > implemented. The deassert does nothing for edge triggered interrupts
>> > other than causing race conditions and extra syscalls.
>> >
>> > I've tested on two of my machines and this resolves the issue I've been
>> > seeing off and on. It should fix the report[1] on bugs@ from months ago.
>>
>> Still looking for testers as the origin bug reporter has not responded.
>
> I know and saw that typing certain things or pasting into the serial
> console had really annoying issues, but now during tests I failed to
> reproduce this, with or without your diff.
Some of my recent fixes probably reduced the probability of hitting the
race condition.
>
> Interacting wth OpenBSD and Alpine Linux guests via 'vmctl console' works
> fine for me with this diff.
>
That's good news. Thanks.
>>
>> >
>> > -dv
>> >
>> > [1] https://marc.info/?l=openbsd-bugs=169669132915760=2
>> >
>> > diff refs/heads/master refs/heads/vmd-edge
>> > commit - 7869b2fdaac7e118bfd1783874fe25ce3b8b0f09
>> > commit + 00d448cdbe2461f27419aaf79520a0cef720aefc
>> > blob - b98e7bdc69ac6a12eb84eaaf97ec43ecdbe83733
>> > blob + 248e3b161e88be505a88411156dc57a2772bd4fa
>> > --- usr.sbin/vmd/i8253.c
>> > +++ usr.sbin/vmd/i8253.c
>> > @@ -371,7 +371,6 @@ i8253_fire(int fd, short type, void *arg)
>> >struct i8253_channel *ctr = (struct i8253_channel *)arg;
>> >
>> >vcpu_assert_pic_irq(ctr->vm_id, 0, 0);
>> > - vcpu_deassert_pic_irq(ctr->vm_id, 0, 0);
>> >
>> >if (ctr->mode != TIMER_INTTC) {
>> >timerclear();
>> > blob - 43dce7b10d1467a5b7ac7f3308d01e32b4d0b9ee
>> > blob + 4fc147b19c99627e386ab26355b8f90a6ae5872b
>> > --- usr.sbin/vmd/mc146818.c
>> > +++ usr.sbin/vmd/mc146818.c
>> > @@ -150,7 +150,6 @@ rtc_fireper(int fd, short type, void *arg)
>> >rtc.regs[MC_REGC] |= MC_REGC_PF;
>> >
>> >vcpu_assert_pic_irq((ptrdiff_t)arg, 0, 8);
>> > - vcpu_deassert_pic_irq((ptrdiff_t)arg, 0, 8);
>> >
>> >evtimer_add(, _tv);
>> > }
>> > blob - bc23876bf0392312335da0d00e143583a87549af
>> > blob + 98ed7dbecf2120ccbf9a16198ee479cb15aebc5f
>> > --- usr.sbin/vmd/ns8250.c
>> > +++ usr.sbin/vmd/ns8250.c
>> > @@ -82,7 +82,6 @@ ratelimit(int fd, short type, void *arg)
>> >com1_dev.regs.iir &= ~IIR_NOPEND;
>> >
>> >vcpu_assert_pic_irq(com1_dev.vmid, 0, com1_dev.irq);
>> > - vcpu_deassert_pic_irq(com1_dev.vmid, 0, com1_dev.irq);
>> >mutex_unlock(_dev.mutex);
>> > }
>> >
>> > @@ -160,7 +159,6 @@ com_rcv_event(int fd, short kind, void *arg)
>> >if ((com1_dev.regs.iir & IIR_NOPEND) == 0) {
>> >/* XXX: vcpu_id */
>> >vcpu_assert_pic_irq((uintptr_t)arg, 0, com1_dev.irq);
>> > - vcpu_deassert_pic_irq((uintptr_t)arg, 0, com1_dev.irq);
>> >}
>> >
>> >mutex_unlock(_dev.mutex);
>>
Re: vmd testers: serial console hangs fix
On Sun, Oct 15, 2023 at 07:51:00AM -0400, Dave Voutila wrote:
>
> Dave Voutila writes:
>
> > Looking for folks that use the serial console connection in vmd(8) and
> > experience the host CPU spikes & delays associated with things like
> > hitting the up key (to cycle shell command history) or copy/paste in X.
> >
> > vmd(8) approximates edge triggered interrupts with the emulated i8259
> > interrupt controller. Some core devices (serial console, rtc, timer) try
> > "toggling" an irq...but it's pointless in how our emulated i8259 is
> > implemented. The deassert does nothing for edge triggered interrupts
> > other than causing race conditions and extra syscalls.
> >
> > I've tested on two of my machines and this resolves the issue I've been
> > seeing off and on. It should fix the report[1] on bugs@ from months ago.
>
> Still looking for testers as the origin bug reporter has not responded.
I know and saw that typing certain things or pasting into the serial
console had really annoying issues, but now during tests I failed to
reproduce this, with or without your diff.
Interacting wth OpenBSD and Alpine Linux guests via 'vmctl console' works
fine for me with this diff.
>
> >
> > -dv
> >
> > [1] https://marc.info/?l=openbsd-bugs=169669132915760=2
> >
> > diff refs/heads/master refs/heads/vmd-edge
> > commit - 7869b2fdaac7e118bfd1783874fe25ce3b8b0f09
> > commit + 00d448cdbe2461f27419aaf79520a0cef720aefc
> > blob - b98e7bdc69ac6a12eb84eaaf97ec43ecdbe83733
> > blob + 248e3b161e88be505a88411156dc57a2772bd4fa
> > --- usr.sbin/vmd/i8253.c
> > +++ usr.sbin/vmd/i8253.c
> > @@ -371,7 +371,6 @@ i8253_fire(int fd, short type, void *arg)
> > struct i8253_channel *ctr = (struct i8253_channel *)arg;
> >
> > vcpu_assert_pic_irq(ctr->vm_id, 0, 0);
> > - vcpu_deassert_pic_irq(ctr->vm_id, 0, 0);
> >
> > if (ctr->mode != TIMER_INTTC) {
> > timerclear();
> > blob - 43dce7b10d1467a5b7ac7f3308d01e32b4d0b9ee
> > blob + 4fc147b19c99627e386ab26355b8f90a6ae5872b
> > --- usr.sbin/vmd/mc146818.c
> > +++ usr.sbin/vmd/mc146818.c
> > @@ -150,7 +150,6 @@ rtc_fireper(int fd, short type, void *arg)
> > rtc.regs[MC_REGC] |= MC_REGC_PF;
> >
> > vcpu_assert_pic_irq((ptrdiff_t)arg, 0, 8);
> > - vcpu_deassert_pic_irq((ptrdiff_t)arg, 0, 8);
> >
> > evtimer_add(, _tv);
> > }
> > blob - bc23876bf0392312335da0d00e143583a87549af
> > blob + 98ed7dbecf2120ccbf9a16198ee479cb15aebc5f
> > --- usr.sbin/vmd/ns8250.c
> > +++ usr.sbin/vmd/ns8250.c
> > @@ -82,7 +82,6 @@ ratelimit(int fd, short type, void *arg)
> > com1_dev.regs.iir &= ~IIR_NOPEND;
> >
> > vcpu_assert_pic_irq(com1_dev.vmid, 0, com1_dev.irq);
> > - vcpu_deassert_pic_irq(com1_dev.vmid, 0, com1_dev.irq);
> > mutex_unlock(_dev.mutex);
> > }
> >
> > @@ -160,7 +159,6 @@ com_rcv_event(int fd, short kind, void *arg)
> > if ((com1_dev.regs.iir & IIR_NOPEND) == 0) {
> > /* XXX: vcpu_id */
> > vcpu_assert_pic_irq((uintptr_t)arg, 0, com1_dev.irq);
> > - vcpu_deassert_pic_irq((uintptr_t)arg, 0, com1_dev.irq);
> > }
> >
> > mutex_unlock(_dev.mutex);
>
Re: vmd testers: serial console hangs fix
Dave Voutila writes:
> Looking for folks that use the serial console connection in vmd(8) and
> experience the host CPU spikes & delays associated with things like
> hitting the up key (to cycle shell command history) or copy/paste in X.
>
> vmd(8) approximates edge triggered interrupts with the emulated i8259
> interrupt controller. Some core devices (serial console, rtc, timer) try
> "toggling" an irq...but it's pointless in how our emulated i8259 is
> implemented. The deassert does nothing for edge triggered interrupts
> other than causing race conditions and extra syscalls.
>
> I've tested on two of my machines and this resolves the issue I've been
> seeing off and on. It should fix the report[1] on bugs@ from months ago.
Still looking for testers as the origin bug reporter has not responded.
>
> -dv
>
> [1] https://marc.info/?l=openbsd-bugs=169669132915760=2
>
> diff refs/heads/master refs/heads/vmd-edge
> commit - 7869b2fdaac7e118bfd1783874fe25ce3b8b0f09
> commit + 00d448cdbe2461f27419aaf79520a0cef720aefc
> blob - b98e7bdc69ac6a12eb84eaaf97ec43ecdbe83733
> blob + 248e3b161e88be505a88411156dc57a2772bd4fa
> --- usr.sbin/vmd/i8253.c
> +++ usr.sbin/vmd/i8253.c
> @@ -371,7 +371,6 @@ i8253_fire(int fd, short type, void *arg)
> struct i8253_channel *ctr = (struct i8253_channel *)arg;
>
> vcpu_assert_pic_irq(ctr->vm_id, 0, 0);
> - vcpu_deassert_pic_irq(ctr->vm_id, 0, 0);
>
> if (ctr->mode != TIMER_INTTC) {
> timerclear();
> blob - 43dce7b10d1467a5b7ac7f3308d01e32b4d0b9ee
> blob + 4fc147b19c99627e386ab26355b8f90a6ae5872b
> --- usr.sbin/vmd/mc146818.c
> +++ usr.sbin/vmd/mc146818.c
> @@ -150,7 +150,6 @@ rtc_fireper(int fd, short type, void *arg)
> rtc.regs[MC_REGC] |= MC_REGC_PF;
>
> vcpu_assert_pic_irq((ptrdiff_t)arg, 0, 8);
> - vcpu_deassert_pic_irq((ptrdiff_t)arg, 0, 8);
>
> evtimer_add(, _tv);
> }
> blob - bc23876bf0392312335da0d00e143583a87549af
> blob + 98ed7dbecf2120ccbf9a16198ee479cb15aebc5f
> --- usr.sbin/vmd/ns8250.c
> +++ usr.sbin/vmd/ns8250.c
> @@ -82,7 +82,6 @@ ratelimit(int fd, short type, void *arg)
> com1_dev.regs.iir &= ~IIR_NOPEND;
>
> vcpu_assert_pic_irq(com1_dev.vmid, 0, com1_dev.irq);
> - vcpu_deassert_pic_irq(com1_dev.vmid, 0, com1_dev.irq);
> mutex_unlock(_dev.mutex);
> }
>
> @@ -160,7 +159,6 @@ com_rcv_event(int fd, short kind, void *arg)
> if ((com1_dev.regs.iir & IIR_NOPEND) == 0) {
> /* XXX: vcpu_id */
> vcpu_assert_pic_irq((uintptr_t)arg, 0, com1_dev.irq);
> - vcpu_deassert_pic_irq((uintptr_t)arg, 0, com1_dev.irq);
> }
>
> mutex_unlock(_dev.mutex);
