salinfo.c

linux-2.6.15.6

/*
 * salinfo.c
 *
 * Creates entries in /proc/sal for various system features.
 *
 * Copyright (c) 2003 Silicon Graphics, Inc.  All rights reserved.
 * Copyright (c) 2003 Hewlett-Packard Co
 *	Bjorn Helgaas <[e-mail]>
 *
 * 10/30/2001	[e-mail]		copied much of Stephane's palinfo
 *					code to create this file
 * Oct 23 2003	[e-mail]
 *   Replace IPI with set_cpus_allowed() to read a record from the required cpu.
 *   Redesign salinfo log processing to separate interrupt and user space
 *   contexts.
 *   Cache the record across multi-block reads from user space.
 *   Support > 64 cpus.
 *   Delete module_exit and MOD_INC/DEC_COUNT, salinfo cannot be a module.
 *
 * Jan 28 2004	[e-mail]
 *   Periodically check for outstanding MCA or INIT records.
 *
 * Dec  5 2004	[e-mail]
 *   Standardize which records are cleared automatically.
 *
 * Aug 18 2005	[e-mail]
 *   mca.c may not pass a buffer, a NULL buffer just indicates that a new
 *   record is available in SAL.
 *   Replace some NR_CPUS by cpus_online, for hotplug cpu.
 */

#include <linux/types.h>
#include <linux/proc_fs.h>
#include <linux/module.h>
#include <linux/smp.h>
#include <linux/smp_lock.h>
#include <linux/timer.h>
#include <linux/vmalloc.h>

#include <asm/semaphore.h>
#include <asm/sal.h>
#include <asm/uaccess.h>

MODULE_AUTHOR("Jesse Barnes <[e-mail]>");
MODULE_DESCRIPTION("/proc interface to IA-64 SAL features");
MODULE_LICENSE("GPL");

static int salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data);

typedef struct {
	const char		*name;		/* name of the proc entry */
	unsigned long           feature;        /* feature bit */
	struct proc_dir_entry	*entry;		/* registered entry (removal) */
} salinfo_entry_t;

/*
 * List {name,feature} pairs for every entry in /proc/sal/<feature>
 * that this module exports
 */
static salinfo_entry_t salinfo_entries[]={
	{ "bus_lock",           IA64_SAL_PLATFORM_FEATURE_BUS_LOCK, },
	{ "irq_redirection",	IA64_SAL_PLATFORM_FEATURE_IRQ_REDIR_HINT, },
	{ "ipi_redirection",	IA64_SAL_PLATFORM_FEATURE_IPI_REDIR_HINT, },
	{ "itc_drift",		IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT, },
};

#define NR_SALINFO_ENTRIES ARRAY_SIZE(salinfo_entries)

static char *salinfo_log_name[] = {
	"mca",
	"init",
	"cmc",
	"cpe",
};

static struct proc_dir_entry *salinfo_proc_entries[
	ARRAY_SIZE(salinfo_entries) +			/* /proc/sal/bus_lock */
	ARRAY_SIZE(salinfo_log_name) +			/* /proc/sal/{mca,...} */
	(2 * ARRAY_SIZE(salinfo_log_name)) +		/* /proc/sal/mca/{event,data} */
	1];						/* /proc/sal */

/* Some records we get ourselves, some are accessed as saved data in buffers
 * that are owned by mca.c.
 */
struct salinfo_data_saved {
	u8*			buffer;
	u64			size;
	u64			id;
	int			cpu;
};

/* State transitions.  Actions are :-
 *   Write "read <cpunum>" to the data file.
 *   Write "clear <cpunum>" to the data file.
 *   Write "oemdata <cpunum> <offset> to the data file.
 *   Read from the data file.
 *   Close the data file.
 *
 * Start state is NO_DATA.
 *
 * NO_DATA
 *    write "read <cpunum>" -> NO_DATA or LOG_RECORD.
 *    write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
 *    write "oemdata <cpunum> <offset> -> return -EINVAL.
 *    read data -> return EOF.
 *    close -> unchanged.  Free record areas.
 *
 * LOG_RECORD
 *    write "read <cpunum>" -> NO_DATA or LOG_RECORD.
 *    write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
 *    write "oemdata <cpunum> <offset> -> format the oem data, goto OEMDATA.
 *    read data -> return the INIT/MCA/CMC/CPE record.
 *    close -> unchanged.  Keep record areas.
 *
 * OEMDATA
 *    write "read <cpunum>" -> NO_DATA or LOG_RECORD.
 *    write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
 *    write "oemdata <cpunum> <offset> -> format the oem data, goto OEMDATA.
 *    read data -> return the formatted oemdata.
 *    close -> unchanged.  Keep record areas.
 *
 * Closing the data file does not change the state.  This allows shell scripts
 * to manipulate salinfo data, each shell redirection opens the file, does one
 * action then closes it again.  The record areas are only freed at close when
 * the state is NO_DATA.
 */
enum salinfo_state {
	STATE_NO_DATA,
	STATE_LOG_RECORD,
	STATE_OEMDATA,
};

struct salinfo_data {
	volatile cpumask_t	cpu_event;	/* which cpus have outstanding events */
	struct semaphore	sem;		/* count of cpus with outstanding events (bits set in cpu_event) */
	u8			*log_buffer;
	u64			log_size;
	u8			*oemdata;	/* decoded oem data */
	u64			oemdata_size;
	int			open;		/* single-open to prevent races */
	u8			type;
	u8			saved_num;	/* using a saved record? */
	enum salinfo_state	state :8;	/* processing state */
	u8			padding;
	int			cpu_check;	/* next CPU to check */
	struct salinfo_data_saved data_saved[5];/* save last 5 records from mca.c, must be < 255 */
};

static struct salinfo_data salinfo_data[ARRAY_SIZE(salinfo_log_name)];

static DEFINE_SPINLOCK(data_lock);
static DEFINE_SPINLOCK(data_saved_lock);

/** salinfo_platform_oemdata - optional callback to decode oemdata from an error
 * record.
 * @sect_header: pointer to the start of the section to decode.
 * @oemdata: returns vmalloc area containing the decded output.
 * @oemdata_size: returns length of decoded output (strlen).
 *
 * Description: If user space asks for oem data to be decoded by the kernel
 * and/or prom and the platform has set salinfo_platform_oemdata to the address
 * of a platform specific routine then call that routine.  salinfo_platform_oemdata
 * vmalloc's and formats its output area, returning the address of the text
 * and its strlen.  Returns 0 for success, -ve for error.  The callback is
 * invoked on the cpu that generated the error record.
 */
int (*salinfo_platform_oemdata)(const u8 *sect_header, u8 **oemdata, u64 *oemdata_size);

struct salinfo_platform_oemdata_parms {
	const u8 *efi_guid;
	u8 **oemdata;
	u64 *oemdata_size;
	int ret;
};

static void
salinfo_platform_oemdata_cpu(void *context)
{
	struct salinfo_platform_oemdata_parms *parms = context;
	parms->ret = salinfo_platform_oemdata(parms->efi_guid, parms->oemdata, parms->oemdata_size);
}

static void
shift1_data_saved (struct salinfo_data *data, int shift)
{
	memcpy(data->data_saved+shift, data->data_saved+shift+1,
	       (ARRAY_SIZE(data->data_saved) - (shift+1)) * sizeof(data->data_saved[0]));
	memset(data->data_saved + ARRAY_SIZE(data->data_saved) - 1, 0,
	       sizeof(data->data_saved[0]));
}

/* This routine is invoked in interrupt context.  Note: mca.c enables
 * interrupts before calling this code for CMC/CPE.  MCA and INIT events are
 * not irq safe, do not call any routines that use spinlocks, they may deadlock.
 * MCA and INIT records are recorded, a timer event will look for any
 * outstanding events and wake up the user space code.
 *
 * The buffer passed from mca.c points to the output from ia64_log_get. This is
 * a persistent buffer but its contents can change between the interrupt and
 * when user space processes the record.  Save the record id to identify
 * changes.  If the buffer is NULL then just update the bitmap.
 */
void
salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
{
	struct salinfo_data *data = salinfo_data + type;
	struct salinfo_data_saved *data_saved;
	unsigned long flags = 0;
	int i;
	int saved_size = ARRAY_SIZE(data->data_saved);

	BUG_ON(type >= ARRAY_SIZE(salinfo_log_name));

	if (buffer) {
		if (irqsafe)
			spin_lock_irqsave(&data_saved_lock, flags);
		for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
			if (!data_saved->buffer)
				break;
		}
		if (i == saved_size) {
			if (!data->saved_num) {
				shift1_data_saved(data, 0);
				data_saved = data->data_saved + saved_size - 1;
			} else
				data_saved = NULL;
		}
		if (data_saved) {
			data_saved->cpu = smp_processor_id();
			data_saved->id = ((sal_log_record_header_t *)buffer)->id;
			data_saved->size = size;
			data_saved->buffer = buffer;
		}
		if (irqsafe)
			spin_unlock_irqrestore(&data_saved_lock, flags);
	}

	if (!test_and_set_bit(smp_processor_id(), &data->cpu_event)) {
		if (irqsafe)
			up(&data->sem);
	}
}

/* Check for outstanding MCA/INIT records every minute (arbitrary) */
#define SALINFO_TIMER_DELAY (60*HZ)
static struct timer_list salinfo_timer;

static void
salinfo_timeout_check(struct salinfo_data *data)
{
	int i;
	if (!data->open)
		return;
	for_each_online_cpu(i) {
		if (test_bit(i, &data->cpu_event)) {
			/* double up() is not a problem, user space will see no
			 * records for the additional "events".
			 */
			up(&data->sem);
		}
	}
}

static void 
salinfo_timeout (unsigned long arg)
{
	salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_MCA);
	salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_INIT);
	salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
	add_timer(&salinfo_timer);
}

static int
salinfo_event_open(struct inode *inode, struct file *file)
{
	if (!capable(CAP_SYS_ADMIN))
		return -EPERM;
	return 0;
}

static ssize_t
salinfo_event_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
{
	struct inode *inode = file->f_dentry->d_inode;
	struct proc_dir_entry *entry = PDE(inode);
	struct salinfo_data *data = entry->data;
	char cmd[32];
	size_t size;
	int i, n, cpu = -1;

retry:
	if (down_trylock(&data->sem)) {
		if (file->f_flags & O_NONBLOCK)
			return -EAGAIN;
		if (down_interruptible(&data->sem))
			return -EINTR;
	}

	n = data->cpu_check;
	for (i = 0; i < NR_CPUS; i++) {
		if (test_bit(n, &data->cpu_event) && cpu_online(n)) {
			cpu = n;
			break;
		}
		if (++n == NR_CPUS)
			n = 0;
	}

	if (cpu == -1)
		goto retry;

	/* events are sticky until the user says "clear" */
	up(&data->sem);

	/* for next read, start checking at next CPU */
	data->cpu_check = cpu;
	if (++data->cpu_check == NR_CPUS)
		data->cpu_check = 0;