PowerPC: MPIC ack interrupts at mpic_teardown_this_cpu()

We really need to ack interrupts at mpic_teardown, since not all platforms reset mpic at kernel start-up. For example, kexec'ed kernel hangs on P.A. Semi if mpic_eoi() isn't called.

Signed-off-by Valentine Barshak <[email protected]>

---

 arch/powerpc/sysdev/mpic.c |    6 +-----
 1 files changed, 1 insertion(+), 5 deletions(-)

arch/powerpc/sysdev/mpic.c +1 -5

@@ -1410,11 +1410,6 @@ void mpic_cpu_set_priority(int prio)
 	mpic_cpu_write(MPIC_INFO(CPU_CURRENT_TASK_PRI), prio);
 }
 
-/*
- * XXX: someone who knows mpic should check this.
- * do we need to eoi the ipi including for kexec cpu here (see xics comments)?
- * or can we reset the mpic in the new kernel?
- */
 void mpic_teardown_this_cpu(int secondary)
 {
 	struct mpic *mpic = mpic_primary;

@@ -1434,6 +1429,7 @@ void mpic_teardown_this_cpu(int secondar
 
 	/* Set current processor priority to max */
 	mpic_cpu_write(MPIC_INFO(CPU_CURRENT_TASK_PRI), 0xf);
+	mpic_eoi(mpic);
 
 	spin_unlock_irqrestore(&mpic_lock, flags);
 }

_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev

Quoted 6 lines · Valentine Barshak

We really need to ack interrupts at mpic_teardown, since not all platforms reset mpic at kernel start-up. For example, kexec'ed kernel hangs on P.A. Semi if mpic_eoi() isn't called.

Signed-off-by Valentine Barshak <[email protected]>

---

Would be interesting to find out why it hangs tho... it shouldn't .

Quoted 8 lines

Quoted 6 lines

We really need to ack interrupts at mpic_teardown, since not all platforms reset mpic at kernel start-up. For example, kexec'ed kernel hangs on P.A. Semi if mpic_eoi() isn't called.

Signed-off-by Valentine Barshak <[email protected]>

---

Would be interesting to find out why it hangs tho... it shouldn't .

I haven't reproduced the problem and looked at hardware state yet, but I would expect it to be because the openpic won't send another interrupt until the previous is EOI'd, and the IPI is never EOI'd as far as I can tell.

The XICS code does it explicitly already. I'm surprised it doesn't break on the 970-based platforms actually...

Quoted 15 lines

Quoted 8 lines

Quoted 6 lines

We really need to ack interrupts at mpic_teardown, since not all platforms reset mpic at kernel start-up. For example, kexec'ed kernel hangs on P.A. Semi if mpic_eoi() isn't called.

Signed-off-by Valentine Barshak <[email protected]>

---

Would be interesting to find out why it hangs tho... it shouldn't .

I haven't reproduced the problem and looked at hardware state yet, but I
would expect it to be because the openpic won't send another interrupt until the previous is EOI'd, and the IPI is never EOI'd as far as I can tell.

Quoted 3 lines

The XICS code does it explicitly already. I'm surprised it doesn't break
on the 970-based platforms actually...

platforms/maple/setup.c and platforms/powermac/pic.c use MPIC_WANTS_RESET.
Is there a reason why we don't do that on every MPIC? If there is such a reason, the default should be to reset, only pseries and chrp and cell and now pasemi do not use it. It's the only sane way to get an MPIC into
a sane known state starting from <whatever> state.

Quoted 7 lines

platforms/maple/setup.c and platforms/powermac/pic.c use MPIC_WANTS_RESET.
Is there a reason why we don't do that on every MPIC? If there is such a reason, the default should be to reset, only pseries and chrp and cell and now pasemi do not use it. It's the only sane way to get an MPIC into
a sane known state starting from <whatever> state.

I used to have the reset there on pasemi but due to some other issues I ended up taking it out.

It's not like doing a final EOI is a big deal, I really don't see why there's hassle about this patch.

Quoted 13 lines

Quoted 9 lines

platforms/maple/setup.c and platforms/powermac/pic.c use MPIC_WANTS_RESET.
Is there a reason why we don't do that on every MPIC? If there is such
a reason, the default should be to reset, only pseries and chrp and cell
and now pasemi do not use it. It's the only sane way to get an MPIC into
a sane known state starting from <whatever> state.

I used to have the reset there on pasemi but due to some other issues I ended up taking it out.

Quoted 2 lines

It's not like doing a final EOI is a big deal, I really don't see why there's hassle about this patch.

The patch is fine, don't get me wrong. I'm just saying MPIC_WANTS_RESET should be the default since pretty much everything wants it. The few platforms that don't can document why; with the current code, we cannot see if you just forgot it or not.

We really need to ack interrupts at mpic_teardown, since not all platforms reset mpic at kernel start-up. For example, kexec'ed kernel hangs on P.A. Semi if mpic_eoi() isn't called.

Signed-off-by Valentine Barshak <[email protected]>