Revision 15f9850d3c2d46f5851a424d2990a18b5bb5ebfd authored by David S. Miller on 19 May 2005, 05:49:26 UTC, committed by David S. Miller on 19 May 2005, 05:49:26 UTC
Even though we do software interrupt mitigation
via NAPI, it still helps to have some minimal
hw assisted mitigation.

This helps, particularly, on systems where register
I/O overhead is much greater than the CPU horsepower.

For example, it helps on NUMA systems.  In such cases
the PIO overhead to disable interrupts for NAPI accounts
for the majority of the packet processing cost.  The
CPU is fast enough such that only a single packet is
processed by each NAPI poll call.

Thanks to Michael Chan for reviewing this patch.

Signed-off-by: David S. Miller <davem@davemloft.net>
1 parent fac9b83
Raw File
pm.c
/*
 *  pm.c - Power management interface
 *
 *  Copyright (C) 2000 Andrew Henroid
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, write to the Free Software
 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */
#include <linux/init.h>
#include <linux/module.h>
#include <linux/spinlock.h>
#include <linux/mm.h>
#include <linux/slab.h>
#include <linux/pm.h>
#include <linux/interrupt.h>

int pm_active;

/*
 *	Locking notes:
 *		pm_devs_lock can be a semaphore providing pm ops are not called
 *	from an interrupt handler (already a bad idea so no change here). Each
 *	change must be protected so that an unlink of an entry doesn't clash
 *	with a pm send - which is permitted to sleep in the current architecture
 *
 *	Module unloads clashing with pm events now work out safely, the module 
 *	unload path will block until the event has been sent. It may well block
 *	until a resume but that will be fine.
 */
 
static DECLARE_MUTEX(pm_devs_lock);
static LIST_HEAD(pm_devs);

/**
 *	pm_register - register a device with power management
 *	@type: device type 
 *	@id: device ID
 *	@callback: callback function
 *
 *	Add a device to the list of devices that wish to be notified about
 *	power management events. A &pm_dev structure is returned on success,
 *	on failure the return is %NULL.
 *
 *      The callback function will be called in process context and
 *      it may sleep.
 */
 
struct pm_dev *pm_register(pm_dev_t type,
			   unsigned long id,
			   pm_callback callback)
{
	struct pm_dev *dev = kmalloc(sizeof(struct pm_dev), GFP_KERNEL);
	if (dev) {
		memset(dev, 0, sizeof(*dev));
		dev->type = type;
		dev->id = id;
		dev->callback = callback;

		down(&pm_devs_lock);
		list_add(&dev->entry, &pm_devs);
		up(&pm_devs_lock);
	}
	return dev;
}

/**
 *	pm_unregister -  unregister a device with power management
 *	@dev: device to unregister
 *
 *	Remove a device from the power management notification lists. The
 *	dev passed must be a handle previously returned by pm_register.
 */
 
void pm_unregister(struct pm_dev *dev)
{
	if (dev) {
		down(&pm_devs_lock);
		list_del(&dev->entry);
		up(&pm_devs_lock);

		kfree(dev);
	}
}

static void __pm_unregister(struct pm_dev *dev)
{
	if (dev) {
		list_del(&dev->entry);
		kfree(dev);
	}
}

/**
 *	pm_unregister_all - unregister all devices with matching callback
 *	@callback: callback function pointer
 *
 *	Unregister every device that would call the callback passed. This
 *	is primarily meant as a helper function for loadable modules. It
 *	enables a module to give up all its managed devices without keeping
 *	its own private list.
 */
 
void pm_unregister_all(pm_callback callback)
{
	struct list_head *entry;

	if (!callback)
		return;

	down(&pm_devs_lock);
	entry = pm_devs.next;
	while (entry != &pm_devs) {
		struct pm_dev *dev = list_entry(entry, struct pm_dev, entry);
		entry = entry->next;
		if (dev->callback == callback)
			__pm_unregister(dev);
	}
	up(&pm_devs_lock);
}

/**
 *	pm_send - send request to a single device
 *	@dev: device to send to
 *	@rqst: power management request
 *	@data: data for the callback
 *
 *	Issue a power management request to a given device. The 
 *	%PM_SUSPEND and %PM_RESUME events are handled specially. The
 *	data field must hold the intended next state. No call is made
 *	if the state matches.
 *
 *	BUGS: what stops two power management requests occurring in parallel
 *	and conflicting.
 *
 *	WARNING: Calling pm_send directly is not generally recommended, in
 *	particular there is no locking against the pm_dev going away. The
 *	caller must maintain all needed locking or have 'inside knowledge'
 *	on the safety. Also remember that this function is not locked against
 *	pm_unregister. This means that you must handle SMP races on callback
 *	execution and unload yourself.
 */
 
static int pm_send(struct pm_dev *dev, pm_request_t rqst, void *data)
{
	int status = 0;
	unsigned long prev_state, next_state;

	if (in_interrupt())
		BUG();

	switch (rqst) {
	case PM_SUSPEND:
	case PM_RESUME:
		prev_state = dev->state;
		next_state = (unsigned long) data;
		if (prev_state != next_state) {
			if (dev->callback)
				status = (*dev->callback)(dev, rqst, data);
			if (!status) {
				dev->state = next_state;
				dev->prev_state = prev_state;
			}
		}
		else {
			dev->prev_state = prev_state;
		}
		break;
	default:
		if (dev->callback)
			status = (*dev->callback)(dev, rqst, data);
		break;
	}
	return status;
}

/*
 * Undo incomplete request
 */
static void pm_undo_all(struct pm_dev *last)
{
	struct list_head *entry = last->entry.prev;
	while (entry != &pm_devs) {
		struct pm_dev *dev = list_entry(entry, struct pm_dev, entry);
		if (dev->state != dev->prev_state) {
			/* previous state was zero (running) resume or
			 * previous state was non-zero (suspended) suspend
			 */
			pm_request_t undo = (dev->prev_state
					     ? PM_SUSPEND:PM_RESUME);
			pm_send(dev, undo, (void*) dev->prev_state);
		}
		entry = entry->prev;
	}
}

/**
 *	pm_send_all - send request to all managed devices
 *	@rqst: power management request
 *	@data: data for the callback
 *
 *	Issue a power management request to a all devices. The 
 *	%PM_SUSPEND events are handled specially. Any device is 
 *	permitted to fail a suspend by returning a non zero (error)
 *	value from its callback function. If any device vetoes a 
 *	suspend request then all other devices that have suspended 
 *	during the processing of this request are restored to their
 *	previous state.
 *
 *	WARNING:  This function takes the pm_devs_lock. The lock is not dropped until
 *	the callbacks have completed. This prevents races against pm locking
 *	functions, races against module unload pm_unregister code. It does
 *	mean however that you must not issue pm_ functions within the callback
 *	or you will deadlock and users will hate you.
 *
 *	Zero is returned on success. If a suspend fails then the status
 *	from the device that vetoes the suspend is returned.
 *
 *	BUGS: what stops two power management requests occurring in parallel
 *	and conflicting.
 */
 
int pm_send_all(pm_request_t rqst, void *data)
{
	struct list_head *entry;
	
	down(&pm_devs_lock);
	entry = pm_devs.next;
	while (entry != &pm_devs) {
		struct pm_dev *dev = list_entry(entry, struct pm_dev, entry);
		if (dev->callback) {
			int status = pm_send(dev, rqst, data);
			if (status) {
				/* return devices to previous state on
				 * failed suspend request
				 */
				if (rqst == PM_SUSPEND)
					pm_undo_all(dev);
				up(&pm_devs_lock);
				return status;
			}
		}
		entry = entry->next;
	}
	up(&pm_devs_lock);
	return 0;
}

EXPORT_SYMBOL(pm_register);
EXPORT_SYMBOL(pm_unregister);
EXPORT_SYMBOL(pm_unregister_all);
EXPORT_SYMBOL(pm_send_all);
EXPORT_SYMBOL(pm_active);


back to top