2009-03-09 22:57:09

by Altobelli, David

Subject: [PATCH] tweak spin logic for hpilo

Change hpilo open and close logic to spin for 10usec between checking device,
rather than every usec. Applies to 2.6.29.

Please CC me on any replies.

Signed-off-by: David Altobelli <[email protected]>
---

--- linux-2.6.28/drivers/misc/hpilo.h.orig 2009-03-09 13:24:25.000000000 -0500
+++ linux-2.6.28/drivers/misc/hpilo.h 2009-03-09 13:28:22.000000000 -0500
@@ -19,8 +19,12 @@
#define MAX_ILO_DEV 1
/* max number of files */
#define MAX_OPEN (MAX_CCB * MAX_ILO_DEV)
+/* total wait time in usec */
+#define MAX_WAIT_TIME 10000
+/* per spin wait time in usec */
+#define WAIT_TIME 10
/* spin counter for open/close delay */
-#define MAX_WAIT 10000
+#define MAX_WAIT (MAX_WAIT_TIME / WAIT_TIME)

/*
* Per device, used to track global memory allocations.
--- linux-2.6.28/drivers/misc/hpilo.c.orig 2009-03-09 13:24:31.000000000 -0500
+++ linux-2.6.28/drivers/misc/hpilo.c 2009-03-09 13:57:08.000000000 -0500
@@ -209,7 +209,7 @@ static void ilo_ccb_close(struct pci_dev
/* give iLO some time to process stop request */
for (retries = MAX_WAIT; retries > 0; retries--) {
doorbell_set(driver_ccb);
- udelay(1);
+ udelay(WAIT_TIME);
if (!(ioread32(&device_ccb->send_ctrl) & (1 << CTRL_BITPOS_A))
&&
!(ioread32(&device_ccb->recv_ctrl) & (1 << CTRL_BITPOS_A)))
@@ -312,7 +312,7 @@ static int ilo_ccb_open(struct ilo_hwinf
for (i = MAX_WAIT; i > 0; i--) {
if (ilo_pkt_dequeue(hw, driver_ccb, SENDQ, &pkt_id, NULL, NULL))
break;
- udelay(1);
+ udelay(WAIT_TIME);
}

if (i) {
@@ -759,7 +759,7 @@ static void __exit ilo_exit(void)
class_destroy(ilo_class);
}

-MODULE_VERSION("1.0");
+MODULE_VERSION("1.1");
MODULE_ALIAS(ILO_NAME);
MODULE_DESCRIPTION(ILO_NAME);
MODULE_AUTHOR("David Altobelli <[email protected]>");

2009-03-09 23:04:53

by Andrew Morton

Subject: Re: [PATCH] tweak spin logic for hpilo

On Mon, 9 Mar 2009 16:56:52 -0600
David Altobelli <[email protected]> wrote:

> Change hpilo open and close logic to spin for 10usec between checking device,
> rather than every usec.

err.. Why?

2009-03-09 23:14:32

by Jiri Slaby

Subject: Re: [PATCH] tweak spin logic for hpilo

On 9.3.2009 23:56, David Altobelli wrote:
> Change hpilo open and close logic to spin for 10usec between checking device,
> rather than every usec. Applies to 2.6.29.

Hi,

the description misses a `why' part. In what is this approach better?

2009-03-10 13:44:01

by Altobelli, David

Subject: RE: [PATCH] tweak spin logic for hpilo

> From: Andrew Morton
> David Altobelli wrote:
> > Change hpilo open and close logic to spin for 10usec between checking
> > device, rather than every usec.
>
> err.. Why?

Because the loop is coded to take up to 10ms, it seemed prudent to increase
the interval between polling the device, to reduce the load on the system and
allow more other work to happen.

2009-03-10 23:37:31

by Robert Hancock

Subject: Re: [PATCH] tweak spin logic for hpilo

Altobelli, David wrote:
>> From: Andrew Morton
>> David Altobelli wrote:
>>> Change hpilo open and close logic to spin for 10usec between checking
>>> device, rather than every usec.
>> err.. Why?
>
> Because the loop is coded to take up to 10ms, it seemed prudent to increase
> the interval between polling the device, to reduce the load on the system and
> allow more other work to happen.

You're using udelay, which is a spin loop. Nothing else can happen on
that CPU while it's delaying anyway.

2009-03-10 23:56:50

by Alan

Subject: Re: [PATCH] tweak spin logic for hpilo

> You're using udelay, which is a spin loop. Nothing else can happen on
> that CPU while it's delaying anyway.

However if your polling is doing I/O its clogging up the I/O busses (and
quite often the bridge)