Skip to contentSkip to conversation

clocksource/drivers/fttmr010: Fix set_next_event handler

PATCH In review Tao Ren 2 participants 3 messages Oct 15, 2018 – Oct 21, 2018 Clocksource, clockevent drivers

lore
Share & tools
lore

Tao Ren

author
#0 lore raw

PATCH 1/1 clocksource/drivers/fttmr010: Fix set_next_event handler

Currently, the aspeed MATCH1 register is updated to <current_count - cycles> in set_next_event handler, with the assumption that COUNT register value is preserved when the timer is disabled and it continues decrementing after the timer is enabled. But the assumption is wrong:
RELOAD register is loaded into COUNT register when the aspeed timer is enabled, which means the next event may be delayed because timer interrupt won't be generated until <0xFFFFFFFF - current_count + cycles>.

The problem can be fixed by updating RELOAD register to <cycles>, and COUNT register will be re-loaded when the timer is enabled and interrupt is generated when COUNT register overflows.

The test result on Facebook Backpack-CMM BMC hardware (AST2500) shows the issue is fixed: without the patch, usleep(100) suspends the process for several milliseconds (and sometimes even over 40 milliseconds); after applying the fix, usleep(100) takes averagely 240 microseconds to return under the same workload level.

---
 drivers/clocksource/timer-fttmr010.c | 18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)
drivers/clocksource/timer-fttmr010.c +11 -7
index c020038ebfab..cf93f6419b51 100644
@@ -130,13 +130,17 @@ static int fttmr010_timer_set_next_event(unsigned long cycles,
 	cr &= ~fttmr010->t1_enable_val;
 	writel(cr, fttmr010->base + TIMER_CR);
 
-	/* Setup the match register forward/backward in time */
-	cr = readl(fttmr010->base + TIMER1_COUNT);
-	if (fttmr010->count_down)
-		cr -= cycles;
-	else
-		cr += cycles;
-	writel(cr, fttmr010->base + TIMER1_MATCH1);
+	if (fttmr010->count_down) {
+		/*
+		 * ASPEED Timer Controller will load TIMER1_LOAD register
+		 * into TIMER1_COUNT register when the timer is re-enabled.
+		 */
+		writel(cycles, fttmr010->base + TIMER1_LOAD);
+	} else {
+		/* Setup the match register forward in time */
+		cr = readl(fttmr010->base + TIMER1_COUNT);
+		writel(cr + cycles, fttmr010->base + TIMER1_MATCH1);
+	}
 
 	/* Start */
 	cr = readl(fttmr010->base + TIMER_CR);
Signature · 1 line
2.17.1
author
#1 lore raw
Quoted 19 lines · Tao Ren

Currently, the aspeed MATCH1 register is updated to <current_count - cycles> in set_next_event handler, with the assumption that COUNT register value is preserved when the timer is disabled and it continues decrementing after the timer is enabled. But the assumption is wrong:
RELOAD register is loaded into COUNT register when the aspeed timer is enabled, which means the next event may be delayed because timer interrupt won't be generated until <0xFFFFFFFF - current_count + cycles>.

The problem can be fixed by updating RELOAD register to <cycles>, and COUNT register will be re-loaded when the timer is enabled and interrupt is generated when COUNT register overflows.

The test result on Facebook Backpack-CMM BMC hardware (AST2500) shows the issue is fixed: without the patch, usleep(100) suspends the process for several milliseconds (and sometimes even over 40 milliseconds); after applying the fix, usleep(100) takes averagely 240 microseconds to return under the same workload level.

Quoted 4 lines

Hi Joel / Andrew,

The patch is included in v4.19-rc6 (commit 4451d3f59f2a), so you could either apply the patch to dev-4.18 now, or it can also be included when kernel is upgraded to 4.19.

Thanks,

Tao Ren

#2 lore raw
Quoted 22 lines · Tao Ren
Quoted 19 lines · Tao Ren

Currently, the aspeed MATCH1 register is updated to <current_count - cycles> in set_next_event handler, with the assumption that COUNT register value is preserved when the timer is disabled and it continues decrementing after the timer is enabled. But the assumption is wrong:
RELOAD register is loaded into COUNT register when the aspeed timer is enabled, which means the next event may be delayed because timer interrupt won't be generated until <0xFFFFFFFF - current_count + cycles>.

The problem can be fixed by updating RELOAD register to <cycles>, and COUNT register will be re-loaded when the timer is enabled and interrupt is generated when COUNT register overflows.

The test result on Facebook Backpack-CMM BMC hardware (AST2500) shows the issue is fixed: without the patch, usleep(100) suspends the process for several milliseconds (and sometimes even over 40 milliseconds); after applying the fix, usleep(100) takes averagely 240 microseconds to return under the same workload level.

Quoted 8 lines
Quoted 4 lines

Hi Joel / Andrew,

The patch is included in v4.19-rc6 (commit 4451d3f59f2a), so you could either apply the patch to dev-4.18 now, or it can also be included when kernel is upgraded to 4.19.

This patch was included in stable, so when openbmc moved to 4.18.16 we got it for "free".

Thanks for submitting the fix upstream. This saves a everyone from duplicated effort.

Cheers,

Joel