summaryrefslogtreecommitdiffstats
path: root/drivers/s390/cio/cmf.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/s390/cio/cmf.c')
-rw-r--r--drivers/s390/cio/cmf.c1042
1 files changed, 1042 insertions, 0 deletions
diff --git a/drivers/s390/cio/cmf.c b/drivers/s390/cio/cmf.c
new file mode 100644
index 000000000000..49def26ba383
--- /dev/null
+++ b/drivers/s390/cio/cmf.c
@@ -0,0 +1,1042 @@
+/*
+ * linux/drivers/s390/cio/cmf.c ($Revision: 1.16 $)
+ *
+ * Linux on zSeries Channel Measurement Facility support
+ *
+ * Copyright 2000,2003 IBM Corporation
+ *
+ * Author: Arnd Bergmann <arndb@de.ibm.com>
+ *
+ * original idea from Natarajan Krishnaswami <nkrishna@us.ibm.com>
+ *
+ * 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, 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., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/bootmem.h>
+#include <linux/device.h>
+#include <linux/init.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+
+#include <asm/ccwdev.h>
+#include <asm/cio.h>
+#include <asm/cmb.h>
+
+#include "cio.h"
+#include "css.h"
+#include "device.h"
+#include "ioasm.h"
+#include "chsc.h"
+
+/* parameter to enable cmf during boot, possible uses are:
+ * "s390cmf" -- enable cmf and allocate 2 MB of ram so measuring can be
+ * used on any subchannel
+ * "s390cmf=<num>" -- enable cmf and allocate enough memory to measure
+ * <num> subchannel, where <num> is an integer
+ * between 1 and 65535, default is 1024
+ */
+#define ARGSTRING "s390cmf"
+
+/* indices for READCMB */
+enum cmb_index {
+ /* basic and exended format: */
+ cmb_ssch_rsch_count,
+ cmb_sample_count,
+ cmb_device_connect_time,
+ cmb_function_pending_time,
+ cmb_device_disconnect_time,
+ cmb_control_unit_queuing_time,
+ cmb_device_active_only_time,
+ /* extended format only: */
+ cmb_device_busy_time,
+ cmb_initial_command_response_time,
+};
+
+/**
+ * enum cmb_format - types of supported measurement block formats
+ *
+ * @CMF_BASIC: traditional channel measurement blocks supported
+ * by all machines that we run on
+ * @CMF_EXTENDED: improved format that was introduced with the z990
+ * machine
+ * @CMF_AUTODETECT: default: use extended format when running on a z990
+ * or later machine, otherwise fall back to basic format
+ **/
+enum cmb_format {
+ CMF_BASIC,
+ CMF_EXTENDED,
+ CMF_AUTODETECT = -1,
+};
+/**
+ * format - actual format for all measurement blocks
+ *
+ * The format module parameter can be set to a value of 0 (zero)
+ * or 1, indicating basic or extended format as described for
+ * enum cmb_format.
+ */
+static int format = CMF_AUTODETECT;
+module_param(format, bool, 0444);
+
+/**
+ * struct cmb_operations - functions to use depending on cmb_format
+ *
+ * all these functions operate on a struct cmf_device. There is only
+ * one instance of struct cmb_operations because all cmf_device
+ * objects are guaranteed to be of the same type.
+ *
+ * @alloc: allocate memory for a channel measurement block,
+ * either with the help of a special pool or with kmalloc
+ * @free: free memory allocated with @alloc
+ * @set: enable or disable measurement
+ * @readall: read a measurement block in a common format
+ * @reset: clear the data in the associated measurement block and
+ * reset its time stamp
+ */
+struct cmb_operations {
+ int (*alloc) (struct ccw_device*);
+ void(*free) (struct ccw_device*);
+ int (*set) (struct ccw_device*, u32);
+ u64 (*read) (struct ccw_device*, int);
+ int (*readall)(struct ccw_device*, struct cmbdata *);
+ void (*reset) (struct ccw_device*);
+
+ struct attribute_group *attr_group;
+};
+static struct cmb_operations *cmbops;
+
+/* our user interface is designed in terms of nanoseconds,
+ * while the hardware measures total times in its own
+ * unit.*/
+static inline u64 time_to_nsec(u32 value)
+{
+ return ((u64)value) * 128000ull;
+}
+
+/*
+ * Users are usually interested in average times,
+ * not accumulated time.
+ * This also helps us with atomicity problems
+ * when reading sinlge values.
+ */
+static inline u64 time_to_avg_nsec(u32 value, u32 count)
+{
+ u64 ret;
+
+ /* no samples yet, avoid division by 0 */
+ if (count == 0)
+ return 0;
+
+ /* value comes in units of 128 µsec */
+ ret = time_to_nsec(value);
+ do_div(ret, count);
+
+ return ret;
+}
+
+/* activate or deactivate the channel monitor. When area is NULL,
+ * the monitor is deactivated. The channel monitor needs to
+ * be active in order to measure subchannels, which also need
+ * to be enabled. */
+static inline void
+cmf_activate(void *area, unsigned int onoff)
+{
+ register void * __gpr2 asm("2");
+ register long __gpr1 asm("1");
+
+ __gpr2 = area;
+ __gpr1 = onoff ? 2 : 0;
+ /* activate channel measurement */
+ asm("schm" : : "d" (__gpr2), "d" (__gpr1) );
+}
+
+static int
+set_schib(struct ccw_device *cdev, u32 mme, int mbfc, unsigned long address)
+{
+ int ret;
+ int retry;
+ struct subchannel *sch;
+ struct schib *schib;
+
+ sch = to_subchannel(cdev->dev.parent);
+ schib = &sch->schib;
+ /* msch can silently fail, so do it again if necessary */
+ for (retry = 0; retry < 3; retry++) {
+ /* prepare schib */
+ stsch(sch->irq, schib);
+ schib->pmcw.mme = mme;
+ schib->pmcw.mbfc = mbfc;
+ /* address can be either a block address or a block index */
+ if (mbfc)
+ schib->mba = address;
+ else
+ schib->pmcw.mbi = address;
+
+ /* try to submit it */
+ switch(ret = msch_err(sch->irq, schib)) {
+ case 0:
+ break;
+ case 1:
+ case 2: /* in I/O or status pending */
+ ret = -EBUSY;
+ break;
+ case 3: /* subchannel is no longer valid */
+ ret = -ENODEV;
+ break;
+ default: /* msch caught an exception */
+ ret = -EINVAL;
+ break;
+ }
+ stsch(sch->irq, schib); /* restore the schib */
+
+ if (ret)
+ break;
+
+ /* check if it worked */
+ if (schib->pmcw.mme == mme &&
+ schib->pmcw.mbfc == mbfc &&
+ (mbfc ? (schib->mba == address)
+ : (schib->pmcw.mbi == address)))
+ return 0;
+
+ ret = -EINVAL;
+ }
+
+ return ret;
+}
+
+struct set_schib_struct {
+ u32 mme;
+ int mbfc;
+ unsigned long address;
+ wait_queue_head_t wait;
+ int ret;
+};
+
+static int set_schib_wait(struct ccw_device *cdev, u32 mme,
+ int mbfc, unsigned long address)
+{
+ struct set_schib_struct s = {
+ .mme = mme,
+ .mbfc = mbfc,
+ .address = address,
+ .wait = __WAIT_QUEUE_HEAD_INITIALIZER(s.wait),
+ };
+
+ spin_lock_irq(cdev->ccwlock);
+ s.ret = set_schib(cdev, mme, mbfc, address);
+ if (s.ret != -EBUSY) {
+ goto out_nowait;
+ }
+
+ if (cdev->private->state != DEV_STATE_ONLINE) {
+ s.ret = -EBUSY;
+ /* if the device is not online, don't even try again */
+ goto out_nowait;
+ }
+ cdev->private->state = DEV_STATE_CMFCHANGE;
+ cdev->private->cmb_wait = &s;
+ s.ret = 1;
+
+ spin_unlock_irq(cdev->ccwlock);
+ if (wait_event_interruptible(s.wait, s.ret != 1)) {
+ spin_lock_irq(cdev->ccwlock);
+ if (s.ret == 1) {
+ s.ret = -ERESTARTSYS;
+ cdev->private->cmb_wait = 0;
+ if (cdev->private->state == DEV_STATE_CMFCHANGE)
+ cdev->private->state = DEV_STATE_ONLINE;
+ }
+ spin_unlock_irq(cdev->ccwlock);
+ }
+ return s.ret;
+
+out_nowait:
+ spin_unlock_irq(cdev->ccwlock);
+ return s.ret;
+}
+
+void retry_set_schib(struct ccw_device *cdev)
+{
+ struct set_schib_struct *s;
+
+ s = cdev->private->cmb_wait;
+ cdev->private->cmb_wait = 0;
+ if (!s) {
+ WARN_ON(1);
+ return;
+ }
+ s->ret = set_schib(cdev, s->mme, s->mbfc, s->address);
+ wake_up(&s->wait);
+}
+
+/**
+ * struct cmb_area - container for global cmb data
+ *
+ * @mem: pointer to CMBs (only in basic measurement mode)
+ * @list: contains a linked list of all subchannels
+ * @lock: protect concurrent access to @mem and @list
+ */
+struct cmb_area {
+ struct cmb *mem;
+ struct list_head list;
+ int num_channels;
+ spinlock_t lock;
+};
+
+static struct cmb_area cmb_area = {
+ .lock = SPIN_LOCK_UNLOCKED,
+ .list = LIST_HEAD_INIT(cmb_area.list),
+ .num_channels = 1024,
+};
+
+
+/* ****** old style CMB handling ********/
+
+/** int maxchannels
+ *
+ * Basic channel measurement blocks are allocated in one contiguous
+ * block of memory, which can not be moved as long as any channel
+ * is active. Therefore, a maximum number of subchannels needs to
+ * be defined somewhere. This is a module parameter, defaulting to
+ * a resonable value of 1024, or 32 kb of memory.
+ * Current kernels don't allow kmalloc with more than 128kb, so the
+ * maximum is 4096
+ */
+
+module_param_named(maxchannels, cmb_area.num_channels, uint, 0444);
+
+/**
+ * struct cmb - basic channel measurement block
+ *
+ * cmb as used by the hardware the fields are described in z/Architecture
+ * Principles of Operation, chapter 17.
+ * The area to be a contiguous array and may not be reallocated or freed.
+ * Only one cmb area can be present in the system.
+ */
+struct cmb {
+ u16 ssch_rsch_count;
+ u16 sample_count;
+ u32 device_connect_time;
+ u32 function_pending_time;
+ u32 device_disconnect_time;
+ u32 control_unit_queuing_time;
+ u32 device_active_only_time;
+ u32 reserved[2];
+};
+
+/* insert a single device into the cmb_area list
+ * called with cmb_area.lock held from alloc_cmb
+ */
+static inline int
+alloc_cmb_single (struct ccw_device *cdev)
+{
+ struct cmb *cmb;
+ struct ccw_device_private *node;
+ int ret;
+
+ spin_lock_irq(cdev->ccwlock);
+ if (!list_empty(&cdev->private->cmb_list)) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ /* find first unused cmb in cmb_area.mem.
+ * this is a little tricky: cmb_area.list
+ * remains sorted by ->cmb pointers */
+ cmb = cmb_area.mem;
+ list_for_each_entry(node, &cmb_area.list, cmb_list) {
+ if ((struct cmb*)node->cmb > cmb)
+ break;
+ cmb++;
+ }
+ if (cmb - cmb_area.mem >= cmb_area.num_channels) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ /* insert new cmb */
+ list_add_tail(&cdev->private->cmb_list, &node->cmb_list);
+ cdev->private->cmb = cmb;
+ ret = 0;
+out:
+ spin_unlock_irq(cdev->ccwlock);
+ return ret;
+}
+
+static int
+alloc_cmb (struct ccw_device *cdev)
+{
+ int ret;
+ struct cmb *mem;
+ ssize_t size;
+
+ spin_lock(&cmb_area.lock);
+
+ if (!cmb_area.mem) {
+ /* there is no user yet, so we need a new area */
+ size = sizeof(struct cmb) * cmb_area.num_channels;
+ WARN_ON(!list_empty(&cmb_area.list));
+
+ spin_unlock(&cmb_area.lock);
+ mem = (void*)__get_free_pages(GFP_KERNEL | GFP_DMA,
+ get_order(size));
+ spin_lock(&cmb_area.lock);
+
+ if (cmb_area.mem) {
+ /* ok, another thread was faster */
+ free_pages((unsigned long)mem, get_order(size));
+ } else if (!mem) {
+ /* no luck */
+ ret = -ENOMEM;
+ goto out;
+ } else {
+ /* everything ok */
+ memset(mem, 0, size);
+ cmb_area.mem = mem;
+ cmf_activate(cmb_area.mem, 1);
+ }
+ }
+
+ /* do the actual allocation */
+ ret = alloc_cmb_single(cdev);
+out:
+ spin_unlock(&cmb_area.lock);
+
+ return ret;
+}
+
+static void
+free_cmb(struct ccw_device *cdev)
+{
+ struct ccw_device_private *priv;
+
+ priv = cdev->private;
+
+ spin_lock(&cmb_area.lock);
+ spin_lock_irq(cdev->ccwlock);
+
+ if (list_empty(&priv->cmb_list)) {
+ /* already freed */
+ goto out;
+ }
+
+ priv->cmb = NULL;
+ list_del_init(&priv->cmb_list);
+
+ if (list_empty(&cmb_area.list)) {
+ ssize_t size;
+ size = sizeof(struct cmb) * cmb_area.num_channels;
+ cmf_activate(NULL, 0);
+ free_pages((unsigned long)cmb_area.mem, get_order(size));
+ cmb_area.mem = NULL;
+ }
+out:
+ spin_unlock_irq(cdev->ccwlock);
+ spin_unlock(&cmb_area.lock);
+}
+
+static int
+set_cmb(struct ccw_device *cdev, u32 mme)
+{
+ u16 offset;
+
+ if (!cdev->private->cmb)
+ return -EINVAL;
+
+ offset = mme ? (struct cmb *)cdev->private->cmb - cmb_area.mem : 0;
+
+ return set_schib_wait(cdev, mme, 0, offset);
+}
+
+static u64
+read_cmb (struct ccw_device *cdev, int index)
+{
+ /* yes, we have to put it on the stack
+ * because the cmb must only be accessed
+ * atomically, e.g. with mvc */
+ struct cmb cmb;
+ unsigned long flags;
+ u32 val;
+
+ spin_lock_irqsave(cdev->ccwlock, flags);
+ if (!cdev->private->cmb) {
+ spin_unlock_irqrestore(cdev->ccwlock, flags);
+ return 0;
+ }
+
+ cmb = *(struct cmb*)cdev->private->cmb;
+ spin_unlock_irqrestore(cdev->ccwlock, flags);
+
+ switch (index) {
+ case cmb_ssch_rsch_count:
+ return cmb.ssch_rsch_count;
+ case cmb_sample_count:
+ return cmb.sample_count;
+ case cmb_device_connect_time:
+ val = cmb.device_connect_time;
+ break;
+ case cmb_function_pending_time:
+ val = cmb.function_pending_time;
+ break;
+ case cmb_device_disconnect_time:
+ val = cmb.device_disconnect_time;
+ break;
+ case cmb_control_unit_queuing_time:
+ val = cmb.control_unit_queuing_time;
+ break;
+ case cmb_device_active_only_time:
+ val = cmb.device_active_only_time;
+ break;
+ default:
+ return 0;
+ }
+ return time_to_avg_nsec(val, cmb.sample_count);
+}
+
+static int
+readall_cmb (struct ccw_device *cdev, struct cmbdata *data)
+{
+ /* yes, we have to put it on the stack
+ * because the cmb must only be accessed
+ * atomically, e.g. with mvc */
+ struct cmb cmb;
+ unsigned long flags;
+ u64 time;
+
+ spin_lock_irqsave(cdev->ccwlock, flags);
+ if (!cdev->private->cmb) {
+ spin_unlock_irqrestore(cdev->ccwlock, flags);
+ return -ENODEV;
+ }
+
+ cmb = *(struct cmb*)cdev->private->cmb;
+ time = get_clock() - cdev->private->cmb_start_time;
+ spin_unlock_irqrestore(cdev->ccwlock, flags);
+
+ memset(data, 0, sizeof(struct cmbdata));
+
+ /* we only know values before device_busy_time */
+ data->size = offsetof(struct cmbdata, device_busy_time);
+
+ /* convert to nanoseconds */
+ data->elapsed_time = (time * 1000) >> 12;
+
+ /* copy data to new structure */
+ data->ssch_rsch_count = cmb.ssch_rsch_count;
+ data->sample_count = cmb.sample_count;
+
+ /* time fields are converted to nanoseconds while copying */
+ data->device_connect_time = time_to_nsec(cmb.device_connect_time);
+ data->function_pending_time = time_to_nsec(cmb.function_pending_time);
+ data->device_disconnect_time = time_to_nsec(cmb.device_disconnect_time);
+ data->control_unit_queuing_time
+ = time_to_nsec(cmb.control_unit_queuing_time);
+ data->device_active_only_time
+ = time_to_nsec(cmb.device_active_only_time);
+
+ return 0;
+}
+
+static void
+reset_cmb(struct ccw_device *cdev)
+{
+ struct cmb *cmb;
+ spin_lock_irq(cdev->ccwlock);
+ cmb = cdev->private->cmb;
+ if (cmb)
+ memset (cmb, 0, sizeof (*cmb));
+ cdev->private->cmb_start_time = get_clock();
+ spin_unlock_irq(cdev->ccwlock);
+}
+
+static struct attribute_group cmf_attr_group;
+
+static struct cmb_operations cmbops_basic = {
+ .alloc = alloc_cmb,
+ .free = free_cmb,
+ .set = set_cmb,
+ .read = read_cmb,
+ .readall = readall_cmb,
+ .reset = reset_cmb,
+ .attr_group = &cmf_attr_group,
+};
+
+/* ******** extended cmb handling ********/
+
+/**
+ * struct cmbe - extended channel measurement block
+ *
+ * cmb as used by the hardware, may be in any 64 bit physical location,
+ * the fields are described in z/Architecture Principles of Operation,
+ * third edition, chapter 17.
+ */
+struct cmbe {
+ u32 ssch_rsch_count;
+ u32 sample_count;
+ u32 device_connect_time;
+ u32 function_pending_time;
+ u32 device_disconnect_time;
+ u32 control_unit_queuing_time;
+ u32 device_active_only_time;
+ u32 device_busy_time;
+ u32 initial_command_response_time;
+ u32 reserved[7];
+};
+
+/* kmalloc only guarantees 8 byte alignment, but we need cmbe
+ * pointers to be naturally aligned. Make sure to allocate
+ * enough space for two cmbes */
+static inline struct cmbe* cmbe_align(struct cmbe *c)
+{
+ unsigned long addr;
+ addr = ((unsigned long)c + sizeof (struct cmbe) - sizeof(long)) &
+ ~(sizeof (struct cmbe) - sizeof(long));
+ return (struct cmbe*)addr;
+}
+
+static int
+alloc_cmbe (struct ccw_device *cdev)
+{
+ struct cmbe *cmbe;
+ cmbe = kmalloc (sizeof (*cmbe) * 2, GFP_KERNEL);
+ if (!cmbe)
+ return -ENOMEM;
+
+ spin_lock_irq(cdev->ccwlock);
+ if (cdev->private->cmb) {
+ kfree(cmbe);
+ spin_unlock_irq(cdev->ccwlock);
+ return -EBUSY;
+ }
+
+ cdev->private->cmb = cmbe;
+ spin_unlock_irq(cdev->ccwlock);
+
+ /* activate global measurement if this is the first channel */
+ spin_lock(&cmb_area.lock);
+ if (list_empty(&cmb_area.list))
+ cmf_activate(NULL, 1);
+ list_add_tail(&cdev->private->cmb_list, &cmb_area.list);
+ spin_unlock(&cmb_area.lock);
+
+ return 0;
+}
+
+static void
+free_cmbe (struct ccw_device *cdev)
+{
+ spin_lock_irq(cdev->ccwlock);
+ if (cdev->private->cmb)
+ kfree(cdev->private->cmb);
+ cdev->private->cmb = NULL;
+ spin_unlock_irq(cdev->ccwlock);
+
+ /* deactivate global measurement if this is the last channel */
+ spin_lock(&cmb_area.lock);
+ list_del_init(&cdev->private->cmb_list);
+ if (list_empty(&cmb_area.list))
+ cmf_activate(NULL, 0);
+ spin_unlock(&cmb_area.lock);
+}
+
+static int
+set_cmbe(struct ccw_device *cdev, u32 mme)
+{
+ unsigned long mba;
+
+ if (!cdev->private->cmb)
+ return -EINVAL;
+ mba = mme ? (unsigned long) cmbe_align(cdev->private->cmb) : 0;
+
+ return set_schib_wait(cdev, mme, 1, mba);
+}
+
+
+u64
+read_cmbe (struct ccw_device *cdev, int index)
+{
+ /* yes, we have to put it on the stack
+ * because the cmb must only be accessed
+ * atomically, e.g. with mvc */
+ struct cmbe cmb;
+ unsigned long flags;
+ u32 val;
+
+ spin_lock_irqsave(cdev->ccwlock, flags);
+ if (!cdev->private->cmb) {
+ spin_unlock_irqrestore(cdev->ccwlock, flags);
+ return 0;
+ }
+
+ cmb = *cmbe_align(cdev->private->cmb);
+ spin_unlock_irqrestore(cdev->ccwlock, flags);
+
+ switch (index) {
+ case cmb_ssch_rsch_count:
+ return cmb.ssch_rsch_count;
+ case cmb_sample_count:
+ return cmb.sample_count;
+ case cmb_device_connect_time:
+ val = cmb.device_connect_time;
+ break;
+ case cmb_function_pending_time:
+ val = cmb.function_pending_time;
+ break;
+ case cmb_device_disconnect_time:
+ val = cmb.device_disconnect_time;
+ break;
+ case cmb_control_unit_queuing_time:
+ val = cmb.control_unit_queuing_time;
+ break;
+ case cmb_device_active_only_time:
+ val = cmb.device_active_only_time;
+ break;
+ case cmb_device_busy_time:
+ val = cmb.device_busy_time;
+ break;
+ case cmb_initial_command_response_time:
+ val = cmb.initial_command_response_time;
+ break;
+ default:
+ return 0;
+ }
+ return time_to_avg_nsec(val, cmb.sample_count);
+}
+
+static int
+readall_cmbe (struct ccw_device *cdev, struct cmbdata *data)
+{
+ /* yes, we have to put it on the stack
+ * because the cmb must only be accessed
+ * atomically, e.g. with mvc */
+ struct cmbe cmb;
+ unsigned long flags;
+ u64 time;
+
+ spin_lock_irqsave(cdev->ccwlock, flags);
+ if (!cdev->private->cmb) {
+ spin_unlock_irqrestore(cdev->ccwlock, flags);
+ return -ENODEV;
+ }
+
+ cmb = *cmbe_align(cdev->private->cmb);
+ time = get_clock() - cdev->private->cmb_start_time;
+ spin_unlock_irqrestore(cdev->ccwlock, flags);
+
+ memset (data, 0, sizeof(struct cmbdata));
+
+ /* we only know values before device_busy_time */
+ data->size = offsetof(struct cmbdata, device_busy_time);
+
+ /* conver to nanoseconds */
+ data->elapsed_time = (time * 1000) >> 12;
+
+ /* copy data to new structure */
+ data->ssch_rsch_count = cmb.ssch_rsch_count;
+ data->sample_count = cmb.sample_count;
+
+ /* time fields are converted to nanoseconds while copying */
+ data->device_connect_time = time_to_nsec(cmb.device_connect_time);
+ data->function_pending_time = time_to_nsec(cmb.function_pending_time);
+ data->device_disconnect_time = time_to_nsec(cmb.device_disconnect_time);
+ data->control_unit_queuing_time
+ = time_to_nsec(cmb.control_unit_queuing_time);
+ data->device_active_only_time
+ = time_to_nsec(cmb.device_active_only_time);
+ data->device_busy_time = time_to_nsec(cmb.device_busy_time);
+ data->initial_command_response_time
+ = time_to_nsec(cmb.initial_command_response_time);
+
+ return 0;
+}
+
+static void
+reset_cmbe(struct ccw_device *cdev)
+{
+ struct cmbe *cmb;
+ spin_lock_irq(cdev->ccwlock);
+ cmb = cmbe_align(cdev->private->cmb);
+ if (cmb)
+ memset (cmb, 0, sizeof (*cmb));
+ cdev->private->cmb_start_time = get_clock();
+ spin_unlock_irq(cdev->ccwlock);
+}
+
+static struct attribute_group cmf_attr_group_ext;
+
+static struct cmb_operations cmbops_extended = {
+ .alloc = alloc_cmbe,
+ .free = free_cmbe,
+ .set = set_cmbe,
+ .read = read_cmbe,
+ .readall = readall_cmbe,
+ .reset = reset_cmbe,
+ .attr_group = &cmf_attr_group_ext,
+};
+
+
+static ssize_t
+cmb_show_attr(struct device *dev, char *buf, enum cmb_index idx)
+{
+ return sprintf(buf, "%lld\n",
+ (unsigned long long) cmf_read(to_ccwdev(dev), idx));
+}
+
+static ssize_t
+cmb_show_avg_sample_interval(struct device *dev, char *buf)
+{
+ struct ccw_device *cdev;
+ long interval;
+ unsigned long count;
+
+ cdev = to_ccwdev(dev);
+ interval = get_clock() - cdev->private->cmb_start_time;
+ count = cmf_read(cdev, cmb_sample_count);
+ if (count)
+ interval /= count;
+ else
+ interval = -1;
+ return sprintf(buf, "%ld\n", interval);
+}
+
+static ssize_t
+cmb_show_avg_utilization(struct device *dev, char *buf)
+{
+ struct cmbdata data;
+ u64 utilization;
+ unsigned long t, u;
+ int ret;
+
+ ret = cmf_readall(to_ccwdev(dev), &data);
+ if (ret)
+ return ret;
+
+ utilization = data.device_connect_time +
+ data.function_pending_time +
+ data.device_disconnect_time;
+
+ /* shift to avoid long long division */
+ while (-1ul < (data.elapsed_time | utilization)) {
+ utilization >>= 8;
+ data.elapsed_time >>= 8;
+ }
+
+ /* calculate value in 0.1 percent units */
+ t = (unsigned long) data.elapsed_time / 1000;
+ u = (unsigned long) utilization / t;
+
+ return sprintf(buf, "%02ld.%01ld%%\n", u/ 10, u - (u/ 10) * 10);
+}
+
+#define cmf_attr(name) \
+static ssize_t show_ ## name (struct device * dev, char * buf) \
+{ return cmb_show_attr((dev), buf, cmb_ ## name); } \
+static DEVICE_ATTR(name, 0444, show_ ## name, NULL);
+
+#define cmf_attr_avg(name) \
+static ssize_t show_avg_ ## name (struct device * dev, char * buf) \
+{ return cmb_show_attr((dev), buf, cmb_ ## name); } \
+static DEVICE_ATTR(avg_ ## name, 0444, show_avg_ ## name, NULL);
+
+cmf_attr(ssch_rsch_count);
+cmf_attr(sample_count);
+cmf_attr_avg(device_connect_time);
+cmf_attr_avg(function_pending_time);
+cmf_attr_avg(device_disconnect_time);
+cmf_attr_avg(control_unit_queuing_time);
+cmf_attr_avg(device_active_only_time);
+cmf_attr_avg(device_busy_time);
+cmf_attr_avg(initial_command_response_time);
+
+static DEVICE_ATTR(avg_sample_interval, 0444, cmb_show_avg_sample_interval, NULL);
+static DEVICE_ATTR(avg_utilization, 0444, cmb_show_avg_utilization, NULL);
+
+static struct attribute *cmf_attributes[] = {
+ &dev_attr_avg_sample_interval.attr,
+ &dev_attr_avg_utilization.attr,
+ &dev_attr_ssch_rsch_count.attr,
+ &dev_attr_sample_count.attr,
+ &dev_attr_avg_device_connect_time.attr,
+ &dev_attr_avg_function_pending_time.attr,
+ &dev_attr_avg_device_disconnect_time.attr,
+ &dev_attr_avg_control_unit_queuing_time.attr,
+ &dev_attr_avg_device_active_only_time.attr,
+ 0,
+};
+
+static struct attribute_group cmf_attr_group = {
+ .name = "cmf",
+ .attrs = cmf_attributes,
+};
+
+static struct attribute *cmf_attributes_ext[] = {
+ &dev_attr_avg_sample_interval.attr,
+ &dev_attr_avg_utilization.attr,
+ &dev_attr_ssch_rsch_count.attr,
+ &dev_attr_sample_count.attr,
+ &dev_attr_avg_device_connect_time.attr,
+ &dev_attr_avg_function_pending_time.attr,
+ &dev_attr_avg_device_disconnect_time.attr,
+ &dev_attr_avg_control_unit_queuing_time.attr,
+ &dev_attr_avg_device_active_only_time.attr,
+ &dev_attr_avg_device_busy_time.attr,
+ &dev_attr_avg_initial_command_response_time.attr,
+ 0,
+};
+
+static struct attribute_group cmf_attr_group_ext = {
+ .name = "cmf",
+ .attrs = cmf_attributes_ext,
+};
+
+static ssize_t cmb_enable_show(struct device *dev, char *buf)
+{
+ return sprintf(buf, "%d\n", to_ccwdev(dev)->private->cmb ? 1 : 0);
+}
+
+static ssize_t cmb_enable_store(struct device *dev, const char *buf, size_t c)
+{
+ struct ccw_device *cdev;
+ int ret;
+
+ cdev = to_ccwdev(dev);
+
+ switch (buf[0]) {
+ case '0':
+ ret = disable_cmf(cdev);
+ if (ret)
+ printk(KERN_INFO "disable_cmf failed (%d)\n", ret);
+ break;
+ case '1':
+ ret = enable_cmf(cdev);
+ if (ret && ret != -EBUSY)
+ printk(KERN_INFO "enable_cmf failed (%d)\n", ret);
+ break;
+ }
+
+ return c;
+}
+
+DEVICE_ATTR(cmb_enable, 0644, cmb_enable_show, cmb_enable_store);
+
+/* enable_cmf/disable_cmf: module interface for cmf (de)activation */
+int
+enable_cmf(struct ccw_device *cdev)
+{
+ int ret;
+
+ ret = cmbops->alloc(cdev);
+ cmbops->reset(cdev);
+ if (ret)
+ return ret;
+ ret = cmbops->set(cdev, 2);
+ if (ret) {
+ cmbops->free(cdev);
+ return ret;
+ }
+ ret = sysfs_create_group(&cdev->dev.kobj, cmbops->attr_group);
+ if (!ret)
+ return 0;
+ cmbops->set(cdev, 0); //FIXME: this can fail
+ cmbops->free(cdev);
+ return ret;
+}
+
+int
+disable_cmf(struct ccw_device *cdev)
+{
+ int ret;
+
+ ret = cmbops->set(cdev, 0);
+ if (ret)
+ return ret;
+ cmbops->free(cdev);
+ sysfs_remove_group(&cdev->dev.kobj, cmbops->attr_group);
+ return ret;
+}
+
+u64
+cmf_read(struct ccw_device *cdev, int index)
+{
+ return cmbops->read(cdev, index);
+}
+
+int
+cmf_readall(struct ccw_device *cdev, struct cmbdata *data)
+{
+ return cmbops->readall(cdev, data);
+}
+
+static int __init
+init_cmf(void)
+{
+ char *format_string;
+ char *detect_string = "parameter";
+
+ /* We cannot really autoprobe this. If the user did not give a parameter,
+ see if we are running on z990 or up, otherwise fall back to basic mode. */
+
+ if (format == CMF_AUTODETECT) {
+ if (!css_characteristics_avail ||
+ !css_general_characteristics.ext_mb) {
+ format = CMF_BASIC;
+ } else {
+ format = CMF_EXTENDED;
+ }
+ detect_string = "autodetected";
+ } else {
+ detect_string = "parameter";
+ }
+
+ switch (format) {
+ case CMF_BASIC:
+ format_string = "basic";
+ cmbops = &cmbops_basic;
+ if (cmb_area.num_channels > 4096 || cmb_area.num_channels < 1) {
+ printk(KERN_ERR "Basic channel measurement facility"
+ " can only use 1 to 4096 devices\n"
+ KERN_ERR "when the cmf driver is built"
+ " as a loadable module\n");
+ return 1;
+ }
+ break;
+ case CMF_EXTENDED:
+ format_string = "extended";
+ cmbops = &cmbops_extended;
+ break;
+ default:
+ printk(KERN_ERR "Invalid format %d for channel "
+ "measurement facility\n", format);
+ return 1;
+ }
+
+ printk(KERN_INFO "Channel measurement facility using %s format (%s)\n",
+ format_string, detect_string);
+ return 0;
+}
+
+module_init(init_cmf);
+
+
+MODULE_AUTHOR("Arnd Bergmann <arndb@de.ibm.com>");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("channel measurement facility base driver\n"
+ "Copyright 2003 IBM Corporation\n");
+
+EXPORT_SYMBOL_GPL(enable_cmf);
+EXPORT_SYMBOL_GPL(disable_cmf);
+EXPORT_SYMBOL_GPL(cmf_read);
+EXPORT_SYMBOL_GPL(cmf_readall);