Skip to contentSkip to conversation

Use semaphore instead of spinlock

UBUNTU:acpi/ec Voluspa 5 participants 9 messages Jun 14, 2006 – Jun 17, 2006 ACPI

lore
Share & tools
lore

Voluspa

author
#0 lore raw

CC drivers/acpi/glue.o
CC drivers/acpi/ec.o
drivers/acpi/ec.c: In function `acpi_ec_poll_read':
drivers/acpi/ec.c:341: error: union has no member named polling' drivers/acpi/ec.c:363: error: union has no member named polling' drivers/acpi/ec.c: In function `acpi_ec_poll_write':
drivers/acpi/ec.c:390: error: union has no member named polling' drivers/acpi/ec.c:415: error: union has no member named polling' drivers/acpi/ec.c:376: warning: unused variable flags' drivers/acpi/ec.c: In function acpi_ec_poll_query':
drivers/acpi/ec.c:599: error: union has no member named polling' drivers/acpi/ec.c:615: error: union has no member named polling' drivers/acpi/ec.c:578: warning: unused variable flags' drivers/acpi/ec.c: In function acpi_ec_gpe_poll_query':
drivers/acpi/ec.c:705: error: union has no member named polling' drivers/acpi/ec.c:708: error: union has no member named polling' drivers/acpi/ec.c:694: warning: unused variable flags' drivers/acpi/ec.c: In function acpi_ec_poll_add':
drivers/acpi/ec.c:1020: error: union has no member named polling' drivers/acpi/ec.c: In function acpi_fake_ecdt_poll_callback':
drivers/acpi/ec.c:1315: error: union has no member named polling' drivers/acpi/ec.c: In function acpi_ec_poll_get_real_ecdt':
drivers/acpi/ec.c:1431: error: union has no member named `polling' make[2]: *** [drivers/acpi/ec.o] Error 1
make[1]: *** [drivers/acpi] Error 2
make: *** [drivers] Error 2

Mvh
Mats Johannesson
--

#1 lore raw
Quoted 18 lines · Voluspa

CC drivers/acpi/glue.o
CC drivers/acpi/ec.o
drivers/acpi/ec.c: In function `acpi_ec_poll_read':
drivers/acpi/ec.c:341: error: union has no member named polling' drivers/acpi/ec.c:363: error: union has no member named polling' drivers/acpi/ec.c: In function `acpi_ec_poll_write':
drivers/acpi/ec.c:390: error: union has no member named polling' drivers/acpi/ec.c:415: error: union has no member named polling' drivers/acpi/ec.c:376: warning: unused variable flags' drivers/acpi/ec.c: In function acpi_ec_poll_query':
drivers/acpi/ec.c:599: error: union has no member named polling' drivers/acpi/ec.c:615: error: union has no member named polling' drivers/acpi/ec.c:578: warning: unused variable flags' drivers/acpi/ec.c: In function acpi_ec_gpe_poll_query':
drivers/acpi/ec.c:705: error: union has no member named polling' drivers/acpi/ec.c:708: error: union has no member named polling' drivers/acpi/ec.c:694: warning: unused variable flags' drivers/acpi/ec.c: In function acpi_ec_poll_add':

Quoted 8 lines

drivers/acpi/ec.c:1020: error: union has no member named polling' drivers/acpi/ec.c: In function acpi_fake_ecdt_poll_callback':
drivers/acpi/ec.c:1315: error: union has no member named polling' drivers/acpi/ec.c: In function acpi_ec_poll_get_real_ecdt':
drivers/acpi/ec.c:1431: error: union has no member named `polling' make[2]: *** [drivers/acpi/ec.o] Error 1
make[1]: *** [drivers/acpi] Error 2
make: *** [drivers] Error 2

Thanks, I'll work on that.

~Randy

#2 lore raw
In reply to Voluspa

updated version:

From: Ben Collins <[email protected]>

[UBUNTU:acpi/ec] Use semaphore instead of spinlock to get rid of missed interrupts on ACPI EC (embedded controller)

Reference: https://launchpad.net/bugs/39315
http://www.kernel.org/git/?p=linux/kernel/git/bcollins/ubuntu-dapper.git;a=commitdiff;h=c484728a760fcfcbad2319ed5364414bc86c3d38

---

---
drivers/acpi/ec.c | 48 ++++++++++++++++++++++++++++++------------------

1 files changed, 30 insertions(+), 18 deletions(-)
drivers/acpi/ec.c +0 -0
@@ -53,8 +53,8 @@ ACPI_MODULE_NAME("acpi_ec")

#define ACPI_EC_EVENT_IBE 0x02 /* Input buffer empty */
#define ACPI_EC_DELAY 50 /* Wait 50ms max. during EC ops */
#define ACPI_EC_UDELAY_GLK 1000 /* Wait 1ms max. to get global lock */
-#define ACPI_EC_UDELAY 100 /* Poll @ 100us increments */
-#define ACPI_EC_UDELAY_COUNT 1000 /* Wait 10ms max. during EC ops */
+#define ACPI_EC_MSLEEP 1 /* Poll @ 1ms increments */
+#define ACPI_EC_MSLEEP_COUNT 10 /* Wait 10ms max. during EC ops */
#define ACPI_EC_COMMAND_READ 0x80
#define ACPI_EC_COMMAND_WRITE 0x81
#define ACPI_EC_BURST_ENABLE 0x82

@@ -116,7 +116,7 @@ union acpi_ec {
		struct acpi_generic_address command_addr;
		struct acpi_generic_address data_addr;
		unsigned long global_lock;
-		spinlock_t lock;
+		struct semaphore sem;
	} poll;
};

@@ -172,7 +172,7 @@ static int acpi_ec_wait(union acpi_ec *e static int acpi_ec_poll_wait(union acpi_ec *ec, u8 event)

{
	u32 acpi_ec_status = 0;

- u32 i = ACPI_EC_UDELAY_COUNT;
+ u32 i = ACPI_EC_MSLEEP_COUNT;

	if (!ec)
		return -EINVAL;
@@ -185,7 +185,7 @@ static int acpi_ec_poll_wait(union acpi_
					       &ec->common.status_addr);
			if (acpi_ec_status & ACPI_EC_FLAG_OBF)
				return 0;
-			udelay(ACPI_EC_UDELAY);
+			msleep(ACPI_EC_MSLEEP);
		} while (--i > 0);
		break;
	case ACPI_EC_EVENT_IBE:
@@ -194,7 +194,7 @@ static int acpi_ec_poll_wait(union acpi_
					       &ec->common.status_addr);
			if (!(acpi_ec_status & ACPI_EC_FLAG_IBF))
				return 0;
-			udelay(ACPI_EC_UDELAY);
+			msleep(ACPI_EC_MSLEEP);
		} while (--i > 0);
		break;
	default:
@@ -323,7 +323,6 @@ static int acpi_ec_poll_read(union acpi_
{
	acpi_status status = AE_OK;
	int result = 0;
-	unsigned long flags = 0;
	u32 glk = 0;

	ACPI_FUNCTION_TRACE("acpi_ec_read");
@@ -339,7 +338,10 @@ static int acpi_ec_poll_read(union acpi_
			return_VALUE(-ENODEV);
	}

- spin_lock_irqsave(&ec->poll.lock, flags);
+ if (down_interruptible(&ec->poll.sem)) {
+ result = -ERESTARTSYS;
+ goto end_nosem;
+ }

	acpi_hw_low_level_write(8, ACPI_EC_COMMAND_READ,
				&ec->common.command_addr);
@@ -358,7 +360,8 @@ static int acpi_ec_poll_read(union acpi_
			  *data, address));

      end:

- spin_unlock_irqrestore(&ec->poll.lock, flags);
+ up(&ec->poll.sem);
+end_nosem:

	if (ec->common.global_lock)
		acpi_release_global_lock(glk);
@@ -384,7 +387,10 @@ static int acpi_ec_poll_write(union acpi
			return_VALUE(-ENODEV);
	}

- spin_lock_irqsave(&ec->poll.lock, flags);
+ if (down_interruptible(&ec->poll.sem)) {
+ result = -ERESTARTSYS;
+ goto end_nosem;
+ }

	acpi_hw_low_level_write(8, ACPI_EC_COMMAND_WRITE,
				&ec->common.command_addr);
@@ -406,7 +412,8 @@ static int acpi_ec_poll_write(union acpi
			  data, address));

      end:

- spin_unlock_irqrestore(&ec->poll.lock, flags);
+ up(&ec->poll.sem);
+end_nosem:

	if (ec->common.global_lock)
		acpi_release_global_lock(glk);

@@ -589,7 +596,10 @@ static int acpi_ec_poll_query(union acpi
* Note that successful completion of the query causes the ACPI_EC_SCI
* bit to be cleared (and thus clearing the interrupt source). */
- spin_lock_irqsave(&ec->poll.lock, flags);
+ if (down_interruptible(&ec->poll.sem)) {
+ result = -ERESTARTSYS;
+ goto end_nosem;
+ }

	acpi_hw_low_level_write(8, ACPI_EC_COMMAND_QUERY,
				&ec->common.command_addr);
@@ -602,7 +612,8 @@ static int acpi_ec_poll_query(union acpi
		result = -ENODATA;

      end:

- spin_unlock_irqrestore(&ec->poll.lock, flags);
+ up(&ec->poll.sem);
+end_nosem:

	if (ec->common.global_lock)
		acpi_release_global_lock(glk);
@@ -691,9 +702,10 @@ static void acpi_ec_gpe_poll_query(void 
	if (!ec_cxt)
		goto end;

- spin_lock_irqsave(&ec->poll.lock, flags);
+ if (down_interruptible (&ec->poll.sem))

+		return_VOID;
	acpi_hw_low_level_read(8, &value, &ec->common.command_addr);

- spin_unlock_irqrestore(&ec->poll.lock, flags);
+ up(&ec->poll.sem);

	/* TBD: Implement asynch events!
	 * NOTE: All we care about are EC-SCI's.  Other EC events are

@@ -1005,7 +1017,7 @@ static int acpi_ec_poll_add(struct acpi_

	ec->common.handle = device->handle;
	ec->common.uid = -1;
-	spin_lock_init(&ec->poll.lock);
+	init_MUTEX(&ec->poll.sem);
	strcpy(acpi_device_name(device), ACPI_EC_DEVICE_NAME);
	strcpy(acpi_device_class(device), ACPI_EC_CLASS);
	acpi_driver_data(device) = ec;
@@ -1300,7 +1312,7 @@ acpi_fake_ecdt_poll_callback(acpi_handle
				  &ec_ecdt->common.gpe_bit);
	if (ACPI_FAILURE(status))
		return status;
-	spin_lock_init(&ec_ecdt->poll.lock);
+	init_MUTEX(&ec_ecdt->poll.sem);
	ec_ecdt->common.global_lock = TRUE;
	ec_ecdt->common.handle = handle;

@@ -1416,7 +1428,7 @@ static int __init acpi_ec_poll_get_real_

	ec_ecdt->common.status_addr = ecdt_ptr->ec_control;
	ec_ecdt->common.data_addr = ecdt_ptr->ec_data;
	ec_ecdt->common.gpe_bit = ecdt_ptr->gpe_bit;
-	spin_lock_init(&ec_ecdt->poll.lock);
+	init_MUTEX(&ec_ecdt->poll.sem);
	/* use the GL just to be safe */
	ec_ecdt->common.global_lock = TRUE;
	ec_ecdt->common.uid = ecdt_ptr->uid;
#3 lore raw

Review

Quoted 7 lines

[UBUNTU:acpi/ec] Use semaphore instead of spinlock to get rid of missed interrupts on ACPI EC (embedded controller)

(quoted diff) +2 -1
-		spinlock_t lock;
+		struct semaphore sem;
 
+	init_MUTEX(&ec->poll.sem);

I think nowadays this should be a struct mutex...

author
#4 lore raw
In reply to Randy Dunlap
Quoted 1 line · On Wed, 14 Jun 2006 18:14:54 -0700 Randy Dunlap wrote:

updated version:

As a user, it's great if this fixes people's keyboards and mice. But it's not a panacea. Gkrellm reads CPU temperatures from /proc/acpi/thermal_zone/*/temperature and that disturbs a time-critical application like mplayer, both when reading normal video and hacked mms:
sound streams (ogg sound is OK, though):

rtc: lost some interrupts at 1024Hz.
rtc: lost some interrupts at 1024Hz.

The loss frequency is much lower with this patch, by something like an estimated 5-10 times.

Also, here the notebook's touchpad gets clobbered by temperature reads when dragging a small xterm window around:

psmouse.c: TouchPad at isa0060/serio4/input0 lost sync at byte 1 psmouse.c: TouchPad at isa0060/serio4/input0 lost sync at byte 1 psmouse.c: TouchPad at isa0060/serio4/input0 lost sync at byte 1 psmouse.c: TouchPad at isa0060/serio4/input0 lost sync at byte 1 psmouse.c: TouchPad at isa0060/serio4/input0 lost sync at byte 1 psmouse.c: issuing reconnect request
psmouse.c: TouchPad at isa0060/serio4/input0 lost sync at byte 4 psmouse.c: TouchPad at isa0060/serio4/input0 lost sync at byte 1 psmouse.c: TouchPad at isa0060/serio4/input0 - driver resynched. psmouse.c: TouchPad at isa0060/serio4/input0 lost sync at byte 4 psmouse.c: TouchPad at isa0060/serio4/input0 lost sync at byte 1 psmouse.c: TouchPad at isa0060/serio4/input0 - driver resynched.

So, here temperature reading stays off even if this patch hits mainline. Battery reading is fine, strangely enough. Perhaps because I've set a period of 30 seconds between updates.

Or it might be that a lot of work has gone in regarding acpi BAT...

Mvh
Mats Johannesson
--

#5 lore raw
Quoted 9 lines · Voluspa
Quoted 1 line · On Wed, 14 Jun 2006 18:14:54 -0700 Randy Dunlap wrote:

updated version:

As a user, it's great if this fixes people's keyboards and mice. But it's not a panacea. Gkrellm reads CPU temperatures from /proc/acpi/thermal_zone/*/temperature and that disturbs a time-critical application like mplayer, both when reading normal video and hacked mms:
sound streams (ogg sound is OK, though):

It would be helpful to analyze this with Ingo's latency tracing patch.

Lee

author
#6 lore raw
Quoted 4 lines · On Thu, 15 Jun 2006 12:27:03 -0400 Lee Revell wrote:
Quoted 3 lines · Voluspa
Quoted 1 line · On Wed, 14 Jun 2006 18:14:54 -0700 Randy Dunlap wrote:

updated version:

Quoted 8 lines
Quoted 6 lines

As a user, it's great if this fixes people's keyboards and mice. But it's not a panacea. Gkrellm reads CPU temperatures from /proc/acpi/thermal_zone/*/temperature and that disturbs a time-critical application like mplayer, both when reading normal video and hacked mms:
sound streams (ogg sound is OK, though):

It would be helpful to analyze this with Ingo's latency tracing patch.

Do you mean proper in relation to ubuntu-patched, or just proper? And, hmmm, I do read a lot of archived lkml threads, but latency tracing patch... Is it buried in the -rt set?

Mvh
Mats Johannesson
--

#7 lore raw
Quoted 12 lines · Voluspa
Quoted 11 lines · On Thu, 15 Jun 2006 12:27:03 -0400 Lee Revell wrote:
Quoted 9 lines · Voluspa
Quoted 1 line · On Wed, 14 Jun 2006 18:14:54 -0700 Randy Dunlap wrote:

updated version:

As a user, it's great if this fixes people's keyboards and mice. But it's not a panacea. Gkrellm reads CPU temperatures from /proc/acpi/thermal_zone/*/temperature and that disturbs a time-critical application like mplayer, both when reading normal video and hacked mms:
sound streams (ogg sound is OK, though):

Quoted 3 lines
Quoted 1 line

It would be helpful to analyze this with Ingo's latency tracing patch.

Do you mean proper in relation to ubuntu-patched, or just proper?

Not sure, I'm not familiar with the exact reason that reading CPU temperatures would cause long latencies like this. The latency tracer can tell you the exact code path responsible.

Quoted 3 lines

And, hmmm,
I do read a lot of archived lkml threads, but latency tracing patch... Is it buried in the -rt set?

It's part of the -rt set and also available broken out for 2.6.16:

http://people.redhat.com/mingo/latency-tracing-patches/latency-tracing-v2.6.16.patch

Lee

#8 lore raw
In reply to Randy Dunlap
Quoted 6 lines · Randy Dunlap

updated version:

From: Ben Collins <[email protected]>

[UBUNTU:acpi/ec] Use semaphore instead of spinlock to get rid of missed interrupts on ACPI EC (embedded controller)

that is odd.

first of all this should use a mutex not a semaphore.

Second, if this isn't used in irq context, then just dropping the "_irqsave" from the spinlocks should be enough to actually get rid of the missed interrupts.... and if it is used there then I highly question your use of semaphores etc...