]> err.no Git - linux-2.6/commitdiff
[S390] zcrypt adjunct processor bus.
authorMartin Schwidefsky <schwidefsky@de.ibm.com>
Wed, 20 Sep 2006 13:58:25 +0000 (15:58 +0200)
committerMartin Schwidefsky <schwidefsky@de.ibm.com>
Wed, 20 Sep 2006 13:58:25 +0000 (15:58 +0200)
Add a bus for the adjunct processor interface. Up to 64 devices can
be connect to the ap bus interface, each device with 16 domains. That
makes 1024 message queues. The interface is asynchronous, the answer
to a message sent to a queue needs to be received at some later point
in time. Unfortunately the interface does not provide interrupts when
a message reply is pending. So the ap bus needs to implement some
fancy polling, each active queue is polled once per 1/HZ second or
continuously if an idle cpus exsists and the poll thread is activ
(see poll_thread parameter).

The ap bus uses the sysfs path /sys/bus/ap and has two bus attributes,
ap_domain and config_time. The ap_domain selects one of the 16 domains
to be used for this system. This limits the maximum number of ap devices
to 64. The config_time attribute contains the number of seconds between
two ap bus scans to find new devices.

The ap bus uses the modalias entries of the form "ap:tN" to autoload
the ap driver for hardware type N. Currently known types are:
3 - PCICC, 4 - PCICA, 5 - PCIXCC, 6 - CEX2A and 7 - CEX2C.

Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Ralph Wuerthner <rwuerthn@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
drivers/s390/crypto/ap_bus.c [new file with mode: 0644]
drivers/s390/crypto/ap_bus.h [new file with mode: 0644]
include/linux/mod_devicetable.h
scripts/mod/file2alias.c

diff --git a/drivers/s390/crypto/ap_bus.c b/drivers/s390/crypto/ap_bus.c
new file mode 100644 (file)
index 0000000..6ed0985
--- /dev/null
@@ -0,0 +1,1221 @@
+/*
+ * linux/drivers/s390/crypto/ap_bus.c
+ *
+ * Copyright (C) 2006 IBM Corporation
+ * Author(s): Cornelia Huck <cornelia.huck@de.ibm.com>
+ *           Martin Schwidefsky <schwidefsky@de.ibm.com>
+ *           Ralph Wuerthner <rwuerthn@de.ibm.com>
+ *
+ * Adjunct processor bus.
+ *
+ * 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/module.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/err.h>
+#include <linux/interrupt.h>
+#include <linux/workqueue.h>
+#include <linux/notifier.h>
+#include <linux/kthread.h>
+#include <linux/mutex.h>
+#include <asm/s390_rdev.h>
+
+#include "ap_bus.h"
+
+/* Some prototypes. */
+static void ap_scan_bus(void *);
+static void ap_poll_all(unsigned long);
+static void ap_poll_timeout(unsigned long);
+static int ap_poll_thread_start(void);
+static void ap_poll_thread_stop(void);
+
+/**
+ * Module description.
+ */
+MODULE_AUTHOR("IBM Corporation");
+MODULE_DESCRIPTION("Adjunct Processor Bus driver, "
+                  "Copyright 2006 IBM Corporation");
+MODULE_LICENSE("GPL");
+
+/**
+ * Module parameter
+ */
+int ap_domain_index = -1;      /* Adjunct Processor Domain Index */
+module_param_named(domain, ap_domain_index, int, 0000);
+MODULE_PARM_DESC(domain, "domain index for ap devices");
+EXPORT_SYMBOL(ap_domain_index);
+
+static int ap_thread_flag = 1;
+module_param_named(poll_thread, ap_thread_flag, int, 0000);
+MODULE_PARM_DESC(poll_thread, "Turn on/off poll thread, default is 1 (on).");
+
+static struct device *ap_root_device = NULL;
+
+/**
+ * Workqueue & timer for bus rescan.
+ */
+static struct workqueue_struct *ap_work_queue;
+static struct timer_list ap_config_timer;
+static int ap_config_time = AP_CONFIG_TIME;
+static DECLARE_WORK(ap_config_work, ap_scan_bus, NULL);
+
+/**
+ * Tasklet & timer for AP request polling.
+ */
+static struct timer_list ap_poll_timer = TIMER_INITIALIZER(ap_poll_timeout,0,0);
+static DECLARE_TASKLET(ap_tasklet, ap_poll_all, 0);
+static atomic_t ap_poll_requests = ATOMIC_INIT(0);
+static DECLARE_WAIT_QUEUE_HEAD(ap_poll_wait);
+static struct task_struct *ap_poll_kthread = NULL;
+static DEFINE_MUTEX(ap_poll_thread_mutex);
+
+/**
+ * Test if ap instructions are available.
+ *
+ * Returns 0 if the ap instructions are installed.
+ */
+static inline int ap_instructions_available(void)
+{
+       register unsigned long reg0 asm ("0") = AP_MKQID(0,0);
+       register unsigned long reg1 asm ("1") = -ENODEV;
+       register unsigned long reg2 asm ("2") = 0UL;
+
+       asm volatile(
+               "   .long 0xb2af0000\n"         /* PQAP(TAPQ) */
+               "0: la    %1,0\n"
+               "1:\n"
+               EX_TABLE(0b, 1b)
+               : "+d" (reg0), "+d" (reg1), "+d" (reg2) : : "cc" );
+       return reg1;
+}
+
+/**
+ * Test adjunct processor queue.
+ * @qid: the ap queue number
+ * @queue_depth: pointer to queue depth value
+ * @device_type: pointer to device type value
+ *
+ * Returns ap queue status structure.
+ */
+static inline struct ap_queue_status
+ap_test_queue(ap_qid_t qid, int *queue_depth, int *device_type)
+{
+       register unsigned long reg0 asm ("0") = qid;
+       register struct ap_queue_status reg1 asm ("1");
+       register unsigned long reg2 asm ("2") = 0UL;
+
+       asm volatile(".long 0xb2af0000"         /* PQAP(TAPQ) */
+                    : "+d" (reg0), "=d" (reg1), "+d" (reg2) : : "cc");
+       *device_type = (int) (reg2 >> 24);
+       *queue_depth = (int) (reg2 & 0xff);
+       return reg1;
+}
+
+/**
+ * Reset adjunct processor queue.
+ * @qid: the ap queue number
+ *
+ * Returns ap queue status structure.
+ */
+static inline struct ap_queue_status ap_reset_queue(ap_qid_t qid)
+{
+       register unsigned long reg0 asm ("0") = qid | 0x01000000UL;
+       register struct ap_queue_status reg1 asm ("1");
+       register unsigned long reg2 asm ("2") = 0UL;
+
+       asm volatile(
+               ".long 0xb2af0000"              /* PQAP(RAPQ) */
+               : "+d" (reg0), "=d" (reg1), "+d" (reg2) : : "cc");
+       return reg1;
+}
+
+/**
+ * Send message to adjunct processor queue.
+ * @qid: the ap queue number
+ * @psmid: the program supplied message identifier
+ * @msg: the message text
+ * @length: the message length
+ *
+ * Returns ap queue status structure.
+ *
+ * Condition code 1 on NQAP can't happen because the L bit is 1.
+ *
+ * Condition code 2 on NQAP also means the send is incomplete,
+ * because a segment boundary was reached. The NQAP is repeated.
+ */
+static inline struct ap_queue_status
+__ap_send(ap_qid_t qid, unsigned long long psmid, void *msg, size_t length)
+{
+       typedef struct { char _[length]; } msgblock;
+       register unsigned long reg0 asm ("0") = qid | 0x40000000UL;
+       register struct ap_queue_status reg1 asm ("1");
+       register unsigned long reg2 asm ("2") = (unsigned long) msg;
+       register unsigned long reg3 asm ("3") = (unsigned long) length;
+       register unsigned long reg4 asm ("4") = (unsigned int) (psmid >> 32);
+       register unsigned long reg5 asm ("5") = (unsigned int) psmid;
+
+       asm volatile (
+               "0: .long 0xb2ad0042\n"         /* DQAP */
+               "   brc   2,0b"
+               : "+d" (reg0), "=d" (reg1), "+d" (reg2), "+d" (reg3)
+               : "d" (reg4), "d" (reg5), "m" (*(msgblock *) msg)
+               : "cc" );
+       return reg1;
+}
+
+int ap_send(ap_qid_t qid, unsigned long long psmid, void *msg, size_t length)
+{
+       struct ap_queue_status status;
+
+       status = __ap_send(qid, psmid, msg, length);
+       switch (status.response_code) {
+       case AP_RESPONSE_NORMAL:
+               return 0;
+       case AP_RESPONSE_Q_FULL:
+               return -EBUSY;
+       default:        /* Device is gone. */
+               return -ENODEV;
+       }
+}
+EXPORT_SYMBOL(ap_send);
+
+/*
+ * Receive message from adjunct processor queue.
+ * @qid: the ap queue number
+ * @psmid: pointer to program supplied message identifier
+ * @msg: the message text
+ * @length: the message length
+ *
+ * Returns ap queue status structure.
+ *
+ * Condition code 1 on DQAP means the receive has taken place
+ * but only partially. The response is incomplete, hence the
+ * DQAP is repeated.
+ *
+ * Condition code 2 on DQAP also means the receive is incomplete,
+ * this time because a segment boundary was reached. Again, the
+ * DQAP is repeated.
+ *
+ * Note that gpr2 is used by the DQAP instruction to keep track of
+ * any 'residual' length, in case the instruction gets interrupted.
+ * Hence it gets zeroed before the instruction.
+ */
+static inline struct ap_queue_status
+__ap_recv(ap_qid_t qid, unsigned long long *psmid, void *msg, size_t length)
+{
+       typedef struct { char _[length]; } msgblock;
+       register unsigned long reg0 asm("0") = qid | 0x80000000UL;
+       register struct ap_queue_status reg1 asm ("1");
+       register unsigned long reg2 asm("2") = 0UL;
+       register unsigned long reg4 asm("4") = (unsigned long) msg;
+       register unsigned long reg5 asm("5") = (unsigned long) length;
+       register unsigned long reg6 asm("6") = 0UL;
+       register unsigned long reg7 asm("7") = 0UL;
+
+
+       asm volatile(
+               "0: .long 0xb2ae0064\n"
+               "   brc   6,0b\n"
+               : "+d" (reg0), "=d" (reg1), "+d" (reg2),
+               "+d" (reg4), "+d" (reg5), "+d" (reg6), "+d" (reg7),
+               "=m" (*(msgblock *) msg) : : "cc" );
+       *psmid = (((unsigned long long) reg6) << 32) + reg7;
+       return reg1;
+}
+
+int ap_recv(ap_qid_t qid, unsigned long long *psmid, void *msg, size_t length)
+{
+       struct ap_queue_status status;
+
+       status = __ap_recv(qid, psmid, msg, length);
+       switch (status.response_code) {
+       case AP_RESPONSE_NORMAL:
+               return 0;
+       case AP_RESPONSE_NO_PENDING_REPLY:
+               if (status.queue_empty)
+                       return -ENOENT;
+               return -EBUSY;
+       default:
+               return -ENODEV;
+       }
+}
+EXPORT_SYMBOL(ap_recv);
+
+/**
+ * Check if an AP queue is available. The test is repeated for
+ * AP_MAX_RESET times.
+ * @qid: the ap queue number
+ * @queue_depth: pointer to queue depth value
+ * @device_type: pointer to device type value
+ */
+static int ap_query_queue(ap_qid_t qid, int *queue_depth, int *device_type)
+{
+       struct ap_queue_status status;
+       int t_depth, t_device_type, rc, i;
+
+       rc = -EBUSY;
+       for (i = 0; i < AP_MAX_RESET; i++) {
+               status = ap_test_queue(qid, &t_depth, &t_device_type);
+               switch (status.response_code) {
+               case AP_RESPONSE_NORMAL:
+                       *queue_depth = t_depth + 1;
+                       *device_type = t_device_type;
+                       rc = 0;
+                       break;
+               case AP_RESPONSE_Q_NOT_AVAIL:
+                       rc = -ENODEV;
+                       break;
+               case AP_RESPONSE_RESET_IN_PROGRESS:
+                       break;
+               case AP_RESPONSE_DECONFIGURED:
+                       rc = -ENODEV;
+                       break;
+               case AP_RESPONSE_CHECKSTOPPED:
+                       rc = -ENODEV;
+                       break;
+               case AP_RESPONSE_BUSY:
+                       break;
+               default:
+                       BUG();
+               }
+               if (rc != -EBUSY)
+                       break;
+               if (i < AP_MAX_RESET - 1)
+                       udelay(5);
+       }
+       return rc;
+}
+
+/**
+ * Reset an AP queue and wait for it to become available again.
+ * @qid: the ap queue number
+ */
+static int ap_init_queue(ap_qid_t qid)
+{
+       struct ap_queue_status status;
+       int rc, dummy, i;
+
+       rc = -ENODEV;
+       status = ap_reset_queue(qid);
+       for (i = 0; i < AP_MAX_RESET; i++) {
+               switch (status.response_code) {
+               case AP_RESPONSE_NORMAL:
+                       if (status.queue_empty)
+                               rc = 0;
+                       break;
+               case AP_RESPONSE_Q_NOT_AVAIL:
+               case AP_RESPONSE_DECONFIGURED:
+               case AP_RESPONSE_CHECKSTOPPED:
+                       i = AP_MAX_RESET;       /* return with -ENODEV */
+                       break;
+               case AP_RESPONSE_RESET_IN_PROGRESS:
+               case AP_RESPONSE_BUSY:
+               default:
+                       break;
+               }
+               if (rc != -ENODEV)
+                       break;
+               if (i < AP_MAX_RESET - 1) {
+                       udelay(5);
+                       status = ap_test_queue(qid, &dummy, &dummy);
+               }
+       }
+       return rc;
+}
+
+/**
+ * AP device related attributes.
+ */
+static ssize_t ap_hwtype_show(struct device *dev,
+                             struct device_attribute *attr, char *buf)
+{
+       struct ap_device *ap_dev = to_ap_dev(dev);
+       return snprintf(buf, PAGE_SIZE, "%d\n", ap_dev->device_type);
+}
+static DEVICE_ATTR(hwtype, 0444, ap_hwtype_show, NULL);
+
+static ssize_t ap_depth_show(struct device *dev, struct device_attribute *attr,
+                            char *buf)
+{
+       struct ap_device *ap_dev = to_ap_dev(dev);
+       return snprintf(buf, PAGE_SIZE, "%d\n", ap_dev->queue_depth);
+}
+static DEVICE_ATTR(depth, 0444, ap_depth_show, NULL);
+
+static ssize_t ap_request_count_show(struct device *dev,
+                                    struct device_attribute *attr,
+                                    char *buf)
+{
+       struct ap_device *ap_dev = to_ap_dev(dev);
+       int rc;
+
+       spin_lock_bh(&ap_dev->lock);
+       rc = snprintf(buf, PAGE_SIZE, "%d\n", ap_dev->total_request_count);
+       spin_unlock_bh(&ap_dev->lock);
+       return rc;
+}
+
+static DEVICE_ATTR(request_count, 0444, ap_request_count_show, NULL);
+
+static ssize_t ap_modalias_show(struct device *dev,
+                               struct device_attribute *attr, char *buf)
+{
+       return sprintf(buf, "ap:t%02X", to_ap_dev(dev)->device_type);
+}
+
+static DEVICE_ATTR(modalias, 0444, ap_modalias_show, NULL);
+
+static struct attribute *ap_dev_attrs[] = {
+       &dev_attr_hwtype.attr,
+       &dev_attr_depth.attr,
+       &dev_attr_request_count.attr,
+       &dev_attr_modalias.attr,
+       NULL
+};
+static struct attribute_group ap_dev_attr_group = {
+       .attrs = ap_dev_attrs
+};
+
+/**
+ * AP bus driver registration/unregistration.
+ */
+static int ap_bus_match(struct device *dev, struct device_driver *drv)
+{
+       struct ap_device *ap_dev = to_ap_dev(dev);
+       struct ap_driver *ap_drv = to_ap_drv(drv);
+       struct ap_device_id *id;
+
+       /**
+        * Compare device type of the device with the list of
+        * supported types of the device_driver.
+        */
+       for (id = ap_drv->ids; id->match_flags; id++) {
+               if ((id->match_flags & AP_DEVICE_ID_MATCH_DEVICE_TYPE) &&
+                   (id->dev_type != ap_dev->device_type))
+                       continue;
+               return 1;
+       }
+       return 0;
+}
+
+/**
+ * uevent function for AP devices. It sets up a single environment
+ * variable DEV_TYPE which contains the hardware device type.
+ */
+static int ap_uevent (struct device *dev, char **envp, int num_envp,
+                      char *buffer, int buffer_size)
+{
+       struct ap_device *ap_dev = to_ap_dev(dev);
+       int length;
+
+       if (!ap_dev)
+               return -ENODEV;
+
+       /* Set up DEV_TYPE environment variable. */
+       envp[0] = buffer;
+       length = scnprintf(buffer, buffer_size, "DEV_TYPE=%04X",
+                          ap_dev->device_type);
+       if (buffer_size - length <= 0)
+               return -ENOMEM;
+       envp[1] = 0;
+       return 0;
+}
+
+static struct bus_type ap_bus_type = {
+       .name = "ap",
+       .match = &ap_bus_match,
+       .uevent = &ap_uevent,
+};
+
+static int ap_device_probe(struct device *dev)
+{
+       struct ap_device *ap_dev = to_ap_dev(dev);
+       struct ap_driver *ap_drv = to_ap_drv(dev->driver);
+       int rc;
+
+       ap_dev->drv = ap_drv;
+       rc = ap_drv->probe ? ap_drv->probe(ap_dev) : -ENODEV;
+       if (rc)
+               ap_dev->unregistered = 1;
+       return rc;
+}
+
+/**
+ * Flush all requests from the request/pending queue of an AP device.
+ * @ap_dev: pointer to the AP device.
+ */
+static inline void __ap_flush_queue(struct ap_device *ap_dev)
+{
+       struct ap_message *ap_msg, *next;
+
+       list_for_each_entry_safe(ap_msg, next, &ap_dev->pendingq, list) {
+               list_del_init(&ap_msg->list);
+               ap_dev->pendingq_count--;
+               ap_dev->drv->receive(ap_dev, ap_msg, ERR_PTR(-ENODEV));
+       }
+       list_for_each_entry_safe(ap_msg, next, &ap_dev->requestq, list) {
+               list_del_init(&ap_msg->list);
+               ap_dev->requestq_count--;
+               ap_dev->drv->receive(ap_dev, ap_msg, ERR_PTR(-ENODEV));
+       }
+}
+
+void ap_flush_queue(struct ap_device *ap_dev)
+{
+       spin_lock_bh(&ap_dev->lock);
+       __ap_flush_queue(ap_dev);
+       spin_unlock_bh(&ap_dev->lock);
+}
+EXPORT_SYMBOL(ap_flush_queue);
+
+static int ap_device_remove(struct device *dev)
+{
+       struct ap_device *ap_dev = to_ap_dev(dev);
+       struct ap_driver *ap_drv = ap_dev->drv;
+
+       spin_lock_bh(&ap_dev->lock);
+       __ap_flush_queue(ap_dev);
+       /**
+        * set ->unregistered to 1 while holding the lock. This prevents
+        * new messages to be put on the queue from now on.
+        */
+       ap_dev->unregistered = 1;
+       spin_unlock_bh(&ap_dev->lock);
+       if (ap_drv->remove)
+               ap_drv->remove(ap_dev);
+       return 0;
+}
+
+int ap_driver_register(struct ap_driver *ap_drv, struct module *owner,
+                      char *name)
+{
+       struct device_driver *drv = &ap_drv->driver;
+
+       drv->bus = &ap_bus_type;
+       drv->probe = ap_device_probe;
+       drv->remove = ap_device_remove;
+       drv->owner = owner;
+       drv->name = name;
+       return driver_register(drv);
+}
+EXPORT_SYMBOL(ap_driver_register);
+
+void ap_driver_unregister(struct ap_driver *ap_drv)
+{
+       driver_unregister(&ap_drv->driver);
+}
+EXPORT_SYMBOL(ap_driver_unregister);
+
+/**
+ * AP bus attributes.
+ */
+static ssize_t ap_domain_show(struct bus_type *bus, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%d\n", ap_domain_index);
+}
+
+static BUS_ATTR(ap_domain, 0444, ap_domain_show, NULL);
+
+static ssize_t ap_config_time_show(struct bus_type *bus, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%d\n", ap_config_time);
+}
+
+static ssize_t ap_config_time_store(struct bus_type *bus,
+                                   const char *buf, size_t count)
+{
+       int time;
+
+       if (sscanf(buf, "%d\n", &time) != 1 || time < 5 || time > 120)
+               return -EINVAL;
+       ap_config_time = time;
+       if (!timer_pending(&ap_config_timer) ||
+           !mod_timer(&ap_config_timer, jiffies + ap_config_time * HZ)) {
+               ap_config_timer.expires = jiffies + ap_config_time * HZ;
+               add_timer(&ap_config_timer);
+       }
+       return count;
+}
+
+static BUS_ATTR(config_time, 0644, ap_config_time_show, ap_config_time_store);
+
+static ssize_t ap_poll_thread_show(struct bus_type *bus, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%d\n", ap_poll_kthread ? 1 : 0);
+}
+
+static ssize_t ap_poll_thread_store(struct bus_type *bus,
+                                   const char *buf, size_t count)
+{
+       int flag, rc;
+
+       if (sscanf(buf, "%d\n", &flag) != 1)
+               return -EINVAL;
+       if (flag) {
+               rc = ap_poll_thread_start();
+               if (rc)
+                       return rc;
+       }
+       else
+               ap_poll_thread_stop();
+       return count;
+}
+
+static BUS_ATTR(poll_thread, 0644, ap_poll_thread_show, ap_poll_thread_store);
+
+static struct bus_attribute *const ap_bus_attrs[] = {
+       &bus_attr_ap_domain,
+       &bus_attr_config_time,
+       &bus_attr_poll_thread,
+       NULL
+};
+
+/**
+ * Pick one of the 16 ap domains.
+ */
+static inline int ap_select_domain(void)
+{
+       int queue_depth, device_type, count, max_count, best_domain;
+       int rc, i, j;
+
+       /**
+        * We want to use a single domain. Either the one specified with
+        * the "domain=" parameter or the domain with the maximum number
+        * of devices.
+        */
+       if (ap_domain_index >= 0 && ap_domain_index < AP_DOMAINS)
+               /* Domain has already been selected. */
+               return 0;
+       best_domain = -1;
+       max_count = 0;
+       for (i = 0; i < AP_DOMAINS; i++) {
+               count = 0;
+               for (j = 0; j < AP_DEVICES; j++) {
+                       ap_qid_t qid = AP_MKQID(j, i);
+                       rc = ap_query_queue(qid, &queue_depth, &device_type);
+                       if (rc)
+                               continue;
+                       count++;
+               }
+               if (count > max_count) {
+                       max_count = count;
+                       best_domain = i;
+               }
+       }
+       if (best_domain >= 0){
+               ap_domain_index = best_domain;
+               return 0;
+       }
+       return -ENODEV;
+}
+
+/**
+ * Find the device type if query queue returned a device type of 0.
+ * @ap_dev: pointer to the AP device.
+ */
+static int ap_probe_device_type(struct ap_device *ap_dev)
+{
+       static unsigned char msg[] = {
+               0x00,0x06,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x58,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x01,0x00,0x43,0x43,0x41,0x2d,0x41,0x50,
+               0x50,0x4c,0x20,0x20,0x20,0x01,0x01,0x01,
+               0x00,0x00,0x00,0x00,0x50,0x4b,0x00,0x00,
+               0x00,0x00,0x01,0x1c,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x05,0xb8,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x70,0x00,0x41,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x54,0x32,0x01,0x00,0xa0,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0xb8,0x05,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x0a,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
+               0x00,0x00,0x00,0x00,0x00,0x00,0x08,0x00,
+               0x49,0x43,0x53,0x46,0x20,0x20,0x20,0x20,
+               0x50,0x4b,0x0a,0x00,0x50,0x4b,0x43,0x53,
+               0x2d,0x31,0x2e,0x32,0x37,0x00,0x11,0x22,
+               0x33,0x44,0x55,0x66,0x77,0x88,0x99,0x00,
+               0x11,0x22,0x33,0x44,0x55,0x66,0x77,0x88,
+               0x99,0x00,0x11,0x22,0x33,0x44,0x55,0x66,
+               0x77,0x88,0x99,0x00,0x11,0x22,0x33,0x44,
+               0x55,0x66,0x77,0x88,0x99,0x00,0x11,0x22,
+               0x33,0x44,0x55,0x66,0x77,0x88,0x99,0x00,
+               0x11,0x22,0x33,0x5d,0x00,0x5b,0x00,0x77,
+               0x88,0x1e,0x00,0x00,0x57,0x00,0x00,0x00,
+               0x00,0x04,0x00,0x00,0x4f,0x00,0x00,0x00,
+               0x03,0x02,0x00,0x00,0x40,0x01,0x00,0x01,
+               0xce,0x02,0x68,0x2d,0x5f,0xa9,0xde,0x0c,
+               0xf6,0xd2,0x7b,0x58,0x4b,0xf9,0x28,0x68,
+               0x3d,0xb4,0xf4,0xef,0x78,0xd5,0xbe,0x66,
+               0x63,0x42,0xef,0xf8,0xfd,0xa4,0xf8,0xb0,
+               0x8e,0x29,0xc2,0xc9,0x2e,0xd8,0x45,0xb8,
+               0x53,0x8c,0x6f,0x4e,0x72,0x8f,0x6c,0x04,
+               0x9c,0x88,0xfc,0x1e,0xc5,0x83,0x55,0x57,
+               0xf7,0xdd,0xfd,0x4f,0x11,0x36,0x95,0x5d,
+       };
+       struct ap_queue_status status;
+       unsigned long long psmid;
+       char *reply;
+       int rc, i;
+
+       reply = (void *) get_zeroed_page(GFP_KERNEL);
+       if (!reply) {
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       status = __ap_send(ap_dev->qid, 0x0102030405060708ULL,
+                          msg, sizeof(msg));
+       if (status.response_code != AP_RESPONSE_NORMAL) {
+               rc = -ENODEV;
+               goto out_free;
+       }
+
+       /* Wait for the test message to complete. */
+       for (i = 0; i < 6; i++) {
+               mdelay(300);
+               status = __ap_recv(ap_dev->qid, &psmid, reply, 4096);
+               if (status.response_code == AP_RESPONSE_NORMAL &&
+                   psmid == 0x0102030405060708ULL)
+                       break;
+       }
+       if (i < 6) {
+               /* Got an answer. */
+               if (reply[0] == 0x00 && reply[1] == 0x86)
+                       ap_dev->device_type = AP_DEVICE_TYPE_PCICC;
+               else
+                       ap_dev->device_type = AP_DEVICE_TYPE_PCICA;
+               rc = 0;
+       } else
+               rc = -ENODEV;
+
+out_free:
+       free_page((unsigned long) reply);
+out:
+       return rc;
+}
+
+/**
+ * Scan the ap bus for new devices.
+ */
+static int __ap_scan_bus(struct device *dev, void *data)
+{
+       return to_ap_dev(dev)->qid == (ap_qid_t)(unsigned long) data;
+}
+
+static void ap_device_release(struct device *dev)
+{
+       struct ap_device *ap_dev = to_ap_dev(dev);
+
+       kfree(ap_dev);
+}
+
+static void ap_scan_bus(void *data)
+{
+       struct ap_device *ap_dev;
+       struct device *dev;
+       ap_qid_t qid;
+       int queue_depth, device_type;
+       int rc, i;
+
+       if (ap_select_domain() != 0)
+               return;
+       for (i = 0; i < AP_DEVICES; i++) {
+               qid = AP_MKQID(i, ap_domain_index);
+               dev = bus_find_device(&ap_bus_type, NULL,
+                                     (void *)(unsigned long)qid,
+                                     __ap_scan_bus);
+               if (dev) {
+                       put_device(dev);
+                       continue;
+               }
+               rc = ap_query_queue(qid, &queue_depth, &device_type);
+               if (rc)
+                       continue;
+               rc = ap_init_queue(qid);
+               if (rc)
+                       continue;
+               ap_dev = kzalloc(sizeof(*ap_dev), GFP_KERNEL);
+               if (!ap_dev)
+                       break;
+               ap_dev->qid = qid;
+               ap_dev->queue_depth = queue_depth;
+               spin_lock_init(&ap_dev->lock);
+               INIT_LIST_HEAD(&ap_dev->pendingq);
+               INIT_LIST_HEAD(&ap_dev->requestq);
+               if (device_type == 0)
+                       ap_probe_device_type(ap_dev);
+               else
+                       ap_dev->device_type = device_type;
+
+               ap_dev->device.bus = &ap_bus_type;
+               ap_dev->device.parent = ap_root_device;
+               snprintf(ap_dev->device.bus_id, BUS_ID_SIZE, "card%02x",
+                        AP_QID_DEVICE(ap_dev->qid));
+               ap_dev->device.release = ap_device_release;
+               rc = device_register(&ap_dev->device);
+               if (rc) {
+                       kfree(ap_dev);
+                       continue;
+               }
+               /* Add device attributes. */
+               rc = sysfs_create_group(&ap_dev->device.kobj,
+                                       &ap_dev_attr_group);
+               if (rc)
+                       device_unregister(&ap_dev->device);
+       }
+}
+
+static void
+ap_config_timeout(unsigned long ptr)
+{
+       queue_work(ap_work_queue, &ap_config_work);
+       ap_config_timer.expires = jiffies + ap_config_time * HZ;
+       add_timer(&ap_config_timer);
+}
+
+/**
+ * Set up the timer to run the poll tasklet
+ */
+static inline void ap_schedule_poll_timer(void)
+{
+       if (timer_pending(&ap_poll_timer))
+               return;
+       mod_timer(&ap_poll_timer, jiffies + AP_POLL_TIME);
+}
+
+/**
+ * Receive pending reply messages from an AP device.
+ * @ap_dev: pointer to the AP device
+ * @flags: pointer to control flags, bit 2^0 is set if another poll is
+ *        required, bit 2^1 is set if the poll timer needs to get armed
+ * Returns 0 if the device is still present, -ENODEV if not.
+ */
+static inline int ap_poll_read(struct ap_device *ap_dev, unsigned long *flags)
+{
+       struct ap_queue_status status;
+       struct ap_message *ap_msg;
+
+       if (ap_dev->queue_count <= 0)
+               return 0;
+       status = __ap_recv(ap_dev->qid, &ap_dev->reply->psmid,
+                          ap_dev->reply->message, ap_dev->reply->length);
+       switch (status.response_code) {
+       case AP_RESPONSE_NORMAL:
+               atomic_dec(&ap_poll_requests);
+               ap_dev->queue_count--;
+               list_for_each_entry(ap_msg, &ap_dev->pendingq, list) {
+                       if (ap_msg->psmid != ap_dev->reply->psmid)
+                               continue;
+                       list_del_init(&ap_msg->list);
+                       ap_dev->pendingq_count--;
+                       ap_dev->drv->receive(ap_dev, ap_msg, ap_dev->reply);
+                       break;
+               }
+               if (ap_dev->queue_count > 0)
+                       *flags |= 1;
+               break;
+       case AP_RESPONSE_NO_PENDING_REPLY:
+               if (status.queue_empty) {
+                       /* The card shouldn't forget requests but who knows. */
+                       ap_dev->queue_count = 0;
+                       list_splice_init(&ap_dev->pendingq, &ap_dev->requestq);
+                       ap_dev->requestq_count += ap_dev->pendingq_count;
+                       ap_dev->pendingq_count = 0;
+               } else
+                       *flags |= 2;
+               break;
+       default:
+               return -ENODEV;
+       }
+       return 0;
+}
+
+/**
+ * Send messages from the request queue to an AP device.
+ * @ap_dev: pointer to the AP device
+ * @flags: pointer to control flags, bit 2^0 is set if another poll is
+ *        required, bit 2^1 is set if the poll timer needs to get armed
+ * Returns 0 if the device is still present, -ENODEV if not.
+ */
+static inline int ap_poll_write(struct ap_device *ap_dev, unsigned long *flags)
+{
+       struct ap_queue_status status;
+       struct ap_message *ap_msg;
+
+       if (ap_dev->requestq_count <= 0 ||
+           ap_dev->queue_count >= ap_dev->queue_depth)
+               return 0;
+       /* Start the next request on the queue. */
+       ap_msg = list_entry(ap_dev->requestq.next, struct ap_message, list);
+       status = __ap_send(ap_dev->qid, ap_msg->psmid,
+                          ap_msg->message, ap_msg->length);
+       switch (status.response_code) {
+       case AP_RESPONSE_NORMAL:
+               atomic_inc(&ap_poll_requests);
+               ap_dev->queue_count++;
+               list_move_tail(&ap_msg->list, &ap_dev->pendingq);
+               ap_dev->requestq_count--;
+               ap_dev->pendingq_count++;
+               if (ap_dev->queue_count < ap_dev->queue_depth &&
+                   ap_dev->requestq_count > 0)
+                       *flags |= 1;
+               *flags |= 2;
+               break;
+       case AP_RESPONSE_Q_FULL:
+               *flags |= 2;
+               break;
+       case AP_RESPONSE_MESSAGE_TOO_BIG:
+               return -EINVAL;
+       default:
+               return -ENODEV;
+       }
+       return 0;
+}
+
+/**
+ * Poll AP device for pending replies and send new messages. If either
+ * ap_poll_read or ap_poll_write returns -ENODEV unregister the device.
+ * @ap_dev: pointer to the bus device
+ * @flags: pointer to control flags, bit 2^0 is set if another poll is
+ *        required, bit 2^1 is set if the poll timer needs to get armed
+ * Returns 0.
+ */
+static inline int ap_poll_queue(struct ap_device *ap_dev, unsigned long *flags)
+{
+       int rc;
+
+       rc = ap_poll_read(ap_dev, flags);
+       if (rc)
+               return rc;
+       return ap_poll_write(ap_dev, flags);
+}
+
+/**
+ * Queue a message to a device.
+ * @ap_dev: pointer to the AP device
+ * @ap_msg: the message to be queued
+ */
+static int __ap_queue_message(struct ap_device *ap_dev, struct ap_message *ap_msg)
+{
+       struct ap_queue_status status;
+
+       if (list_empty(&ap_dev->requestq) &&
+           ap_dev->queue_count < ap_dev->queue_depth) {
+               status = __ap_send(ap_dev->qid, ap_msg->psmid,
+                                  ap_msg->message, ap_msg->length);
+               switch (status.response_code) {
+               case AP_RESPONSE_NORMAL:
+                       list_add_tail(&ap_msg->list, &ap_dev->pendingq);
+                       atomic_inc(&ap_poll_requests);
+                       ap_dev->pendingq_count++;
+                       ap_dev->queue_count++;
+                       ap_dev->total_request_count++;
+                       break;
+               case AP_RESPONSE_Q_FULL:
+                       list_add_tail(&ap_msg->list, &ap_dev->requestq);
+                       ap_dev->requestq_count++;
+                       ap_dev->total_request_count++;
+                       return -EBUSY;
+               case AP_RESPONSE_MESSAGE_TOO_BIG:
+                       ap_dev->drv->receive(ap_dev, ap_msg, ERR_PTR(-EINVAL));
+                       return -EINVAL;
+               default:        /* Device is gone. */
+                       ap_dev->drv->receive(ap_dev, ap_msg, ERR_PTR(-ENODEV));
+                       return -ENODEV;
+               }
+       } else {
+               list_add_tail(&ap_msg->list, &ap_dev->requestq);
+               ap_dev->requestq_count++;
+               ap_dev->total_request_count++;
+               return -EBUSY;
+       }
+       ap_schedule_poll_timer();
+       return 0;
+}
+
+void ap_queue_message(struct ap_device *ap_dev, struct ap_message *ap_msg)
+{
+       unsigned long flags;
+       int rc;
+
+       spin_lock_bh(&ap_dev->lock);
+       if (!ap_dev->unregistered) {
+               /* Make room on the queue by polling for finished requests. */
+               rc = ap_poll_queue(ap_dev, &flags);
+               if (!rc)
+                       rc = __ap_queue_message(ap_dev, ap_msg);
+               if (!rc)
+                       wake_up(&ap_poll_wait);
+       } else {
+               ap_dev->drv->receive(ap_dev, ap_msg, ERR_PTR(-ENODEV));
+               rc = 0;
+       }
+       spin_unlock_bh(&ap_dev->lock);
+       if (rc == -ENODEV)
+               device_unregister(&ap_dev->device);
+}
+EXPORT_SYMBOL(ap_queue_message);
+
+/**
+ * Cancel a crypto request. This is done by removing the request
+ * from the devive pendingq or requestq queue. Note that the
+ * request stays on the AP queue. When it finishes the message
+ * reply will be discarded because the psmid can't be found.
+ * @ap_dev: AP device that has the message queued
+ * @ap_msg: the message that is to be removed
+ */
+void ap_cancel_message(struct ap_device *ap_dev, struct ap_message *ap_msg)
+{
+       struct ap_message *tmp;
+
+       spin_lock_bh(&ap_dev->lock);
+       if (!list_empty(&ap_msg->list)) {
+               list_for_each_entry(tmp, &ap_dev->pendingq, list)
+                       if (tmp->psmid == ap_msg->psmid) {
+                               ap_dev->pendingq_count--;
+                               goto found;
+                       }
+               ap_dev->requestq_count--;
+       found:
+               list_del_init(&ap_msg->list);
+       }
+       spin_unlock_bh(&ap_dev->lock);
+}
+EXPORT_SYMBOL(ap_cancel_message);
+
+/**
+ * AP receive polling for finished AP requests
+ */
+static void ap_poll_timeout(unsigned long unused)
+{
+       tasklet_schedule(&ap_tasklet);
+}
+
+/**
+ * Poll all AP devices on the bus in a round robin fashion. Continue
+ * polling until bit 2^0 of the control flags is not set. If bit 2^1
+ * of the control flags has been set arm the poll timer.
+ */
+static int __ap_poll_all(struct device *dev, void *data)
+{
+       struct ap_device *ap_dev = to_ap_dev(dev);
+       int rc;
+
+       spin_lock(&ap_dev->lock);
+       if (!ap_dev->unregistered) {
+               rc = ap_poll_queue(to_ap_dev(dev), (unsigned long *) data);
+       } else
+               rc = 0;
+       spin_unlock(&ap_dev->lock);
+       if (rc)
+               device_unregister(&ap_dev->device);
+       return 0;
+}
+
+static void ap_poll_all(unsigned long dummy)
+{
+       unsigned long flags;
+
+       do {
+               flags = 0;
+               bus_for_each_dev(&ap_bus_type, NULL, &flags, __ap_poll_all);
+       } while (flags & 1);
+       if (flags & 2)
+               ap_schedule_poll_timer();
+}
+
+/**
+ * AP bus poll thread. The purpose of this thread is to poll for
+ * finished requests in a loop if there is a "free" cpu - that is
+ * a cpu that doesn't have anything better to do. The polling stops
+ * as soon as there is another task or if all messages have been
+ * delivered.
+ */
+static int ap_poll_thread(void *data)
+{
+       DECLARE_WAITQUEUE(wait, current);
+       unsigned long flags;
+       int requests;
+
+       set_user_nice(current, -20);
+       while (1) {
+               if (need_resched()) {
+                       schedule();
+                       continue;
+               }
+               add_wait_queue(&ap_poll_wait, &wait);
+               set_current_state(TASK_INTERRUPTIBLE);
+               if (kthread_should_stop())
+                       break;
+               requests = atomic_read(&ap_poll_requests);
+               if (requests <= 0)
+                       schedule();
+               set_current_state(TASK_RUNNING);
+               remove_wait_queue(&ap_poll_wait, &wait);
+
+               local_bh_disable();
+               flags = 0;
+               bus_for_each_dev(&ap_bus_type, NULL, &flags, __ap_poll_all);
+               local_bh_enable();
+       }
+       set_current_state(TASK_RUNNING);
+       remove_wait_queue(&ap_poll_wait, &wait);
+       return 0;
+}
+
+static int ap_poll_thread_start(void)
+{
+       int rc;
+
+       mutex_lock(&ap_poll_thread_mutex);
+       if (!ap_poll_kthread) {
+               ap_poll_kthread = kthread_run(ap_poll_thread, NULL, "appoll");
+               rc = IS_ERR(ap_poll_kthread) ? PTR_ERR(ap_poll_kthread) : 0;
+               if (rc)
+                       ap_poll_kthread = NULL;
+       }
+       else
+               rc = 0;
+       mutex_unlock(&ap_poll_thread_mutex);
+       return rc;
+}
+
+static void ap_poll_thread_stop(void)
+{
+       mutex_lock(&ap_poll_thread_mutex);
+       if (ap_poll_kthread) {
+               kthread_stop(ap_poll_kthread);
+               ap_poll_kthread = NULL;
+       }
+       mutex_unlock(&ap_poll_thread_mutex);
+}
+
+/**
+ * The module initialization code.
+ */
+int __init ap_module_init(void)
+{
+       int rc, i;
+
+       if (ap_domain_index < -1 || ap_domain_index >= AP_DOMAINS) {
+               printk(KERN_WARNING "Invalid param: domain = %d. "
+                      " Not loading.\n", ap_domain_index);
+               return -EINVAL;
+       }
+       if (ap_instructions_available() != 0) {
+               printk(KERN_WARNING "AP instructions not installed.\n");
+               return -ENODEV;
+       }
+
+       /* Create /sys/bus/ap. */
+       rc = bus_register(&ap_bus_type);
+       if (rc)
+               goto out;
+       for (i = 0; ap_bus_attrs[i]; i++) {
+               rc = bus_create_file(&ap_bus_type, ap_bus_attrs[i]);
+               if (rc)
+                       goto out_bus;
+       }
+
+       /* Create /sys/devices/ap. */
+       ap_root_device = s390_root_dev_register("ap");
+       rc = IS_ERR(ap_root_device) ? PTR_ERR(ap_root_device) : 0;
+       if (rc)
+               goto out_bus;
+
+       ap_work_queue = create_singlethread_workqueue("kapwork");
+       if (!ap_work_queue) {
+               rc = -ENOMEM;
+               goto out_root;
+       }
+
+       if (ap_select_domain() == 0)
+               ap_scan_bus(NULL);
+
+       /* Setup the ap bus rescan timer. */
+       init_timer(&ap_config_timer);
+       ap_config_timer.function = ap_config_timeout;
+       ap_config_timer.data = 0;
+       ap_config_timer.expires = jiffies + ap_config_time * HZ;
+       add_timer(&ap_config_timer);
+
+       /* Start the low priority AP bus poll thread. */
+       if (ap_thread_flag) {
+               rc = ap_poll_thread_start();
+               if (rc)
+                       goto out_work;
+       }
+
+       return 0;
+
+out_work:
+       del_timer_sync(&ap_config_timer);
+       del_timer_sync(&ap_poll_timer);
+       destroy_workqueue(ap_work_queue);
+out_root:
+       s390_root_dev_unregister(ap_root_device);
+out_bus:
+       while (i--)
+               bus_remove_file(&ap_bus_type, ap_bus_attrs[i]);
+       bus_unregister(&ap_bus_type);
+out:
+       return rc;
+}
+
+static int __ap_match_all(struct device *dev, void *data)
+{
+       return 1;
+}
+
+/**
+ * The module termination code
+ */
+void ap_module_exit(void)
+{
+       int i;
+       struct device *dev;
+
+       ap_poll_thread_stop();
+       del_timer_sync(&ap_config_timer);
+       del_timer_sync(&ap_poll_timer);
+       destroy_workqueue(ap_work_queue);
+       s390_root_dev_unregister(ap_root_device);
+       while ((dev = bus_find_device(&ap_bus_type, NULL, NULL,
+                   __ap_match_all)))
+       {
+               device_unregister(dev);
+               put_device(dev);
+       }
+       for (i = 0; ap_bus_attrs[i]; i++)
+               bus_remove_file(&ap_bus_type, ap_bus_attrs[i]);
+       bus_unregister(&ap_bus_type);
+}
+
+#ifndef CONFIG_ZCRYPT_MONOLITHIC
+module_init(ap_module_init);
+module_exit(ap_module_exit);
+#endif
diff --git a/drivers/s390/crypto/ap_bus.h b/drivers/s390/crypto/ap_bus.h
new file mode 100644 (file)
index 0000000..83b69c0
--- /dev/null
@@ -0,0 +1,158 @@
+/*
+ * linux/drivers/s390/crypto/ap_bus.h
+ *
+ * Copyright (C) 2006 IBM Corporation
+ * Author(s): Cornelia Huck <cornelia.huck@de.ibm.com>
+ *           Martin Schwidefsky <schwidefsky@de.ibm.com>
+ *           Ralph Wuerthner <rwuerthn@de.ibm.com>
+ *
+ * Adjunct processor bus header file.
+ *
+ * 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.
+ */
+
+#ifndef _AP_BUS_H_
+#define _AP_BUS_H_
+
+#include <linux/device.h>
+#include <linux/mod_devicetable.h>
+#include <linux/types.h>
+
+#define AP_DEVICES 64          /* Number of AP devices. */
+#define AP_DOMAINS 16          /* Number of AP domains. */
+#define AP_MAX_RESET 90                /* Maximum number of resets. */
+#define AP_CONFIG_TIME 30      /* Time in seconds between AP bus rescans. */
+#define AP_POLL_TIME 1         /* Time in ticks between receive polls. */
+
+extern int ap_domain_index;
+
+/**
+ * The ap_qid_t identifier of an ap queue. It contains a
+ * 6 bit device index and a 4 bit queue index (domain).
+ */
+typedef unsigned int ap_qid_t;
+
+#define AP_MKQID(_device,_queue) (((_device) & 63) << 8 | ((_queue) & 15))
+#define AP_QID_DEVICE(_qid) (((_qid) >> 8) & 63)
+#define AP_QID_QUEUE(_qid) ((_qid) & 15)
+
+/**
+ * The ap queue status word is returned by all three AP functions
+ * (PQAP, NQAP and DQAP).  There's a set of flags in the first
+ * byte, followed by a 1 byte response code.
+ */
+struct ap_queue_status {
+       unsigned int queue_empty        : 1;
+       unsigned int replies_waiting    : 1;
+       unsigned int queue_full         : 1;
+       unsigned int pad1               : 5;
+       unsigned int response_code      : 8;
+       unsigned int pad2               : 16;
+};
+
+#define AP_RESPONSE_NORMAL             0x00
+#define AP_RESPONSE_Q_NOT_AVAIL                0x01
+#define AP_RESPONSE_RESET_IN_PROGRESS  0x02
+#define AP_RESPONSE_DECONFIGURED       0x03
+#define AP_RESPONSE_CHECKSTOPPED       0x04
+#define AP_RESPONSE_BUSY               0x05
+#define AP_RESPONSE_Q_FULL             0x10
+#define AP_RESPONSE_NO_PENDING_REPLY   0x10
+#define AP_RESPONSE_INDEX_TOO_BIG      0x11
+#define AP_RESPONSE_NO_FIRST_PART      0x13
+#define AP_RESPONSE_MESSAGE_TOO_BIG    0x15
+
+/**
+ * Known device types
+ */
+#define AP_DEVICE_TYPE_PCICC   3
+#define AP_DEVICE_TYPE_PCICA   4
+#define AP_DEVICE_TYPE_PCIXCC  5
+#define AP_DEVICE_TYPE_CEX2A   6
+#define AP_DEVICE_TYPE_CEX2C   7
+
+struct ap_device;
+struct ap_message;
+
+struct ap_driver {
+       struct device_driver driver;
+       struct ap_device_id *ids;
+
+       int (*probe)(struct ap_device *);
+       void (*remove)(struct ap_device *);
+       /* receive is called from tasklet context */
+       void (*receive)(struct ap_device *, struct ap_message *,
+                       struct ap_message *);
+};
+
+#define to_ap_drv(x) container_of((x), struct ap_driver, driver)
+
+int ap_driver_register(struct ap_driver *, struct module *, char *);
+void ap_driver_unregister(struct ap_driver *);
+
+struct ap_device {
+       struct device device;
+       struct ap_driver *drv;          /* Pointer to AP device driver. */
+       spinlock_t lock;                /* Per device lock. */
+
+       ap_qid_t qid;                   /* AP queue id. */
+       int queue_depth;                /* AP queue depth.*/
+       int device_type;                /* AP device type. */
+       int unregistered;               /* marks AP device as unregistered */
+
+       int queue_count;                /* # messages currently on AP queue. */
+
+       struct list_head pendingq;      /* List of message sent to AP queue. */
+       int pendingq_count;             /* # requests on pendingq list. */
+       struct list_head requestq;      /* List of message yet to be sent. */
+       int requestq_count;             /* # requests on requestq list. */
+       int total_request_count;        /* # requests ever for this AP device. */
+
+       struct ap_message *reply;       /* Per device reply message. */
+
+       void *private;                  /* ap driver private pointer. */
+};
+
+#define to_ap_dev(x) container_of((x), struct ap_device, device)
+
+struct ap_message {
+       struct list_head list;          /* Request queueing. */
+       unsigned long long psmid;       /* Message id. */
+       void *message;                  /* Pointer to message buffer. */
+       size_t length;                  /* Message length. */
+
+       void *private;                  /* ap driver private pointer. */
+};
+
+#define AP_DEVICE(dt)                                  \
+       .dev_type=(dt),                                 \
+       .match_flags=AP_DEVICE_ID_MATCH_DEVICE_TYPE,
+
+/**
+ * Note: don't use ap_send/ap_recv after using ap_queue_message
+ * for the first time. Otherwise the ap message queue will get
+ * confused.
+ */
+int ap_send(ap_qid_t, unsigned long long, void *, size_t);
+int ap_recv(ap_qid_t, unsigned long long *, void *, size_t);
+
+void ap_queue_message(struct ap_device *ap_dev, struct ap_message *ap_msg);
+void ap_cancel_message(struct ap_device *ap_dev, struct ap_message *ap_msg);
+void ap_flush_queue(struct ap_device *ap_dev);
+
+int ap_module_init(void);
+void ap_module_exit(void);
+
+#endif /* _AP_BUS_H_ */
index f6977708585cab80c7840db5920ae1c512578e3a..f7ca0b09075d960d54315b20c2f2cb1e29e3ec94 100644 (file)
@@ -148,6 +148,17 @@ struct ccw_device_id {
 #define CCW_DEVICE_ID_MATCH_DEVICE_TYPE                0x04
 #define CCW_DEVICE_ID_MATCH_DEVICE_MODEL       0x08
 
+/* s390 AP bus devices */
+struct ap_device_id {
+       __u16 match_flags;      /* which fields to match against */
+       __u8 dev_type;          /* device type */
+       __u8 pad1;
+       __u32 pad2;
+       kernel_ulong_t driver_info;
+};
+
+#define AP_DEVICE_ID_MATCH_DEVICE_TYPE         0x01
+
 
 #define PNP_ID_LEN     8
 #define PNP_MAX_DEVICES        8
index e2de650d3dbff82ea73de953aec6edffc63f28e5..de76da80443f757a8ed90ce4f27dd53e2ff2a8db 100644 (file)
@@ -265,6 +265,14 @@ static int do_ccw_entry(const char *filename,
        return 1;
 }
 
+/* looks like: "ap:tN" */
+static int do_ap_entry(const char *filename,
+                      struct ap_device_id *id, char *alias)
+{
+       sprintf(alias, "ap:t%02X", id->dev_type);
+       return 1;
+}
+
 /* Looks like: "serio:tyNprNidNexN" */
 static int do_serio_entry(const char *filename,
                          struct serio_device_id *id, char *alias)
@@ -503,6 +511,10 @@ void handle_moddevtable(struct module *mod, struct elf_info *info,
                do_table(symval, sym->st_size,
                         sizeof(struct ccw_device_id), "ccw",
                         do_ccw_entry, mod);
+       else if (sym_is(symname, "__mod_ap_device_table"))
+               do_table(symval, sym->st_size,
+                        sizeof(struct ap_device_id), "ap",
+                        do_ap_entry, mod);
        else if (sym_is(symname, "__mod_serio_device_table"))
                do_table(symval, sym->st_size,
                         sizeof(struct serio_device_id), "serio",