Skip to contentSkip to conversation

powerpc/64s: Fix doorbell wakeup msgclr optimisation

PATCH Applied Nicholas Piggin 2 participants 2 messages Apr 2, 2020 – Apr 6, 2020 PowerPC

lore
Share & tools
lore

Nicholas Piggin

author reviewer
#0 lore raw

PATCH 1/1 powerpc/64s: Fix doorbell wakeup msgclr optimisation

Commit 3282a3da25bd ("powerpc/64: Implement soft interrupt replay in C") broke the doorbell wakeup optimisation introduced by commit a9af97aa0a12 ("powerpc/64s: msgclr when handling doorbell exceptions from system reset").

This patch restores it, in C code. It's moved explicitly to the system reset wakeup path, rather than the doorbell replay path, because it is always the right thing to do in the wakeup case, but it could be rare to have a pending message in other cases in which case it's wasted work -- we would have to be done to see if that was a worthwhile optimisation, and I suspect it would not be.

The results are similar to those in the original commit, test on POWER8 of context_switch selftests benchmark with polling idle disabled (e.g., always nap, giving cross-CPU IPIs) gives the following results:

                                    broken           patched
    Different threads, same core:   317k/s           375k/s    +18.7%
    Different cores:                280k/s           282k/s     +1.0%
Fixes 3282a3da25bd ("powerpc/64: Implement soft interrupt replay in C")
---
 arch/powerpc/kernel/exceptions-64s.S | 19 -------------------
 arch/powerpc/kernel/irq.c            | 13 +++++++++++++
 2 files changed, 13 insertions(+), 19 deletions(-)
arch/powerpc/kernel/exceptions-64s.S +0 -19
index 18bbce143084..728ccb0f560c 100644
@@ -3121,22 +3121,3 @@ handle_dabr_fault:
 	li	r5,SIGSEGV
 	bl	bad_page_fault
 	b	interrupt_return
-
-/*
- * When doorbell is triggered from system reset wakeup, the message is
- * not cleared, so it would fire again when EE is enabled.
- *
- * When coming from local_irq_enable, there may be the same problem if
- * we were hard disabled.
- *
- * Execute msgclr to clear pending exceptions before handling it.
- */
-h_doorbell_common_msgclr:
-	LOAD_REG_IMMEDIATE(r3, PPC_DBELL_MSGTYPE << (63-36))
-	PPC_MSGCLR(3)
-	b 	h_doorbell_common_virt
-
-doorbell_super_common_msgclr:
-	LOAD_REG_IMMEDIATE(r3, PPC_DBELL_MSGTYPE << (63-36))
-	PPC_MSGCLRP(3)
-	b 	doorbell_super_common_virt
arch/powerpc/kernel/irq.c +13 -0
index 6ea27dbcb872..ed6230ba0c43 100644
@@ -527,6 +527,19 @@ void irq_set_pending_from_srr1(unsigned long srr1)
 		return;
 	}
 
+	if (reason == PACA_IRQ_DBELL) {
+		/*
+		 * When doorbell triggers a system reset wakeup, the message
+		 * is not cleared, so if the doorbell interrupt is replayed
+		 * and the IPI handled, the doorbell interrupt would still
+		 * fire when EE is enabled.
+		 *
+		 * To avoid taking the superfluous doorbell interrupt,
+		 * execute a msgclr here before the interrupt is replayed.
+		 */
+		ppc_msgclr(PPC_DBELL_MSGTYPE);
+	}
+
 	/*
 	 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
 	 * so this can be called unconditionally with the SRR1 wake
Signature · 1 line
2.23.0
#1 lore raw
Quoted 16 lines · Nicholas Piggin

Commit 3282a3da25bd ("powerpc/64: Implement soft interrupt replay in C") broke the doorbell wakeup optimisation introduced by commit a9af97aa0a12 ("powerpc/64s: msgclr when handling doorbell exceptions from system reset").

This patch restores it, in C code. It's moved explicitly to the system reset wakeup path, rather than the doorbell replay path, because it is always the right thing to do in the wakeup case, but it could be rare to have a pending message in other cases in which case it's wasted work -- we would have to be done to see if that was a worthwhile optimisation, and I suspect it would not be.

The results are similar to those in the original commit, test on POWER8 of context_switch selftests benchmark with polling idle disabled (e.g., always nap, giving cross-CPU IPIs) gives the following results:

Quoted 6 lines
                                    broken           patched
    Different threads, same core:   317k/s           375k/s    +18.7%
    Different cores:                280k/s           282k/s     +1.0%
Fixes 3282a3da25bd ("powerpc/64: Implement soft interrupt replay in C")

Applied to powerpc next, thanks.

https://git.kernel.org/powerpc/c/0c89649a70bed679fd408c1eb82fa99dbe1354a0

cheers