Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
authorLinus Torvalds <torvalds@linux-foundation.org>
Thu, 23 Jul 2015 20:07:04 +0000 (13:07 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 23 Jul 2015 20:07:04 +0000 (13:07 -0700)
Pull virtio/vhost fixes from Michael Tsirkin:
 "Bugfixes and documentation fixes.

  Igor's patch that allows users to tweak memory table size is
  borderline, but it does fix known crashes, so I merged it"

* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
  vhost: add max_mem_regions module parameter
  vhost: extend memory regions allocation to vmalloc
  9p/trans_virtio: reset virtio device on remove
  virtio/s390: rename drivers/s390/kvm -> drivers/s390/virtio
  MAINTAINERS: separate section for s390 virtio drivers
  virtio: define virtio_pci_cfg_cap in header.
  virtio: Fix typecast of pointer in vring_init()
  virtio scsi: fix unused variable warning
  vhost: use binary search instead of linear in find_region()
  virtio_net: document VIRTIO_NET_CTRL_GUEST_OFFLOADS

1  2 
MAINTAINERS
drivers/s390/virtio/kvm_virtio.c
drivers/s390/virtio/virtio_ccw.c
drivers/scsi/virtio_scsi.c

diff --cc MAINTAINERS
index a2264167791acd0d309b1097937ad1ba8aa36cb8,fbef7d0a4dd5f9793baa9c1448413ace56a451b7..2eb627252239f6d71590636ab0068fe42d87f311
@@@ -10896,15 -10570,15 +10895,24 @@@ F:        drivers/block/virtio_blk.
  F:    include/linux/virtio_*.h
  F:    include/uapi/linux/virtio_*.h
  
+ VIRTIO DRIVERS FOR S390
+ M:    Christian Borntraeger <borntraeger@de.ibm.com>
+ M:    Cornelia Huck <cornelia.huck@de.ibm.com>
+ L:    linux-s390@vger.kernel.org
+ L:    virtualization@lists.linux-foundation.org
+ L:    kvm@vger.kernel.org
+ S:    Supported
+ F:    drivers/s390/virtio/
 +VIRTIO GPU DRIVER
 +M:    David Airlie <airlied@linux.ie>
 +M:    Gerd Hoffmann <kraxel@redhat.com>
 +L:    dri-devel@lists.freedesktop.org
 +L:    virtualization@lists.linux-foundation.org
 +S:    Maintained
 +F:    drivers/gpu/drm/virtio/
 +F:    include/uapi/linux/virtio_gpu.h
 +
  VIRTIO HOST (VHOST)
  M:    "Michael S. Tsirkin" <mst@redhat.com>
  L:    kvm@vger.kernel.org
index 0000000000000000000000000000000000000000,dd65c8b4c7fe983336a7ff1596c7655837d851b7..53fb975c404b19b3f1d50f1f226cd0155accd683
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,510 +1,510 @@@
 -      unsigned long total_memory_size = sclp_get_rzm() * sclp_get_rnmax();
+ /*
+  * virtio for kvm on s390
+  *
+  * Copyright IBM Corp. 2008
+  *
+  * This program is free software; you can redistribute it and/or modify
+  * it under the terms of the GNU General Public License (version 2 only)
+  * as published by the Free Software Foundation.
+  *
+  *    Author(s): Christian Borntraeger <borntraeger@de.ibm.com>
+  */
+ #include <linux/kernel_stat.h>
+ #include <linux/init.h>
+ #include <linux/bootmem.h>
+ #include <linux/err.h>
+ #include <linux/virtio.h>
+ #include <linux/virtio_config.h>
+ #include <linux/slab.h>
+ #include <linux/virtio_console.h>
+ #include <linux/interrupt.h>
+ #include <linux/virtio_ring.h>
+ #include <linux/export.h>
+ #include <linux/pfn.h>
+ #include <asm/io.h>
+ #include <asm/kvm_para.h>
+ #include <asm/kvm_virtio.h>
+ #include <asm/sclp.h>
+ #include <asm/setup.h>
+ #include <asm/irq.h>
+ #define VIRTIO_SUBCODE_64 0x0D00
+ /*
+  * The pointer to our (page) of device descriptions.
+  */
+ static void *kvm_devices;
+ static struct work_struct hotplug_work;
+ struct kvm_device {
+       struct virtio_device vdev;
+       struct kvm_device_desc *desc;
+ };
+ #define to_kvmdev(vd) container_of(vd, struct kvm_device, vdev)
+ /*
+  * memory layout:
+  * - kvm_device_descriptor
+  *        struct kvm_device_desc
+  * - configuration
+  *        struct kvm_vqconfig
+  * - feature bits
+  * - config space
+  */
+ static struct kvm_vqconfig *kvm_vq_config(const struct kvm_device_desc *desc)
+ {
+       return (struct kvm_vqconfig *)(desc + 1);
+ }
+ static u8 *kvm_vq_features(const struct kvm_device_desc *desc)
+ {
+       return (u8 *)(kvm_vq_config(desc) + desc->num_vq);
+ }
+ static u8 *kvm_vq_configspace(const struct kvm_device_desc *desc)
+ {
+       return kvm_vq_features(desc) + desc->feature_len * 2;
+ }
+ /*
+  * The total size of the config page used by this device (incl. desc)
+  */
+ static unsigned desc_size(const struct kvm_device_desc *desc)
+ {
+       return sizeof(*desc)
+               + desc->num_vq * sizeof(struct kvm_vqconfig)
+               + desc->feature_len * 2
+               + desc->config_len;
+ }
+ /* This gets the device's feature bits. */
+ static u64 kvm_get_features(struct virtio_device *vdev)
+ {
+       unsigned int i;
+       u32 features = 0;
+       struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
+       u8 *in_features = kvm_vq_features(desc);
+       for (i = 0; i < min(desc->feature_len * 8, 32); i++)
+               if (in_features[i / 8] & (1 << (i % 8)))
+                       features |= (1 << i);
+       return features;
+ }
+ static int kvm_finalize_features(struct virtio_device *vdev)
+ {
+       unsigned int i, bits;
+       struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
+       /* Second half of bitmap is features we accept. */
+       u8 *out_features = kvm_vq_features(desc) + desc->feature_len;
+       /* Give virtio_ring a chance to accept features. */
+       vring_transport_features(vdev);
+       /* Make sure we don't have any features > 32 bits! */
+       BUG_ON((u32)vdev->features != vdev->features);
+       memset(out_features, 0, desc->feature_len);
+       bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8;
+       for (i = 0; i < bits; i++) {
+               if (__virtio_test_bit(vdev, i))
+                       out_features[i / 8] |= (1 << (i % 8));
+       }
+       return 0;
+ }
+ /*
+  * Reading and writing elements in config space
+  */
+ static void kvm_get(struct virtio_device *vdev, unsigned int offset,
+                  void *buf, unsigned len)
+ {
+       struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
+       BUG_ON(offset + len > desc->config_len);
+       memcpy(buf, kvm_vq_configspace(desc) + offset, len);
+ }
+ static void kvm_set(struct virtio_device *vdev, unsigned int offset,
+                  const void *buf, unsigned len)
+ {
+       struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
+       BUG_ON(offset + len > desc->config_len);
+       memcpy(kvm_vq_configspace(desc) + offset, buf, len);
+ }
+ /*
+  * The operations to get and set the status word just access
+  * the status field of the device descriptor. set_status will also
+  * make a hypercall to the host, to tell about status changes
+  */
+ static u8 kvm_get_status(struct virtio_device *vdev)
+ {
+       return to_kvmdev(vdev)->desc->status;
+ }
+ static void kvm_set_status(struct virtio_device *vdev, u8 status)
+ {
+       BUG_ON(!status);
+       to_kvmdev(vdev)->desc->status = status;
+       kvm_hypercall1(KVM_S390_VIRTIO_SET_STATUS,
+                      (unsigned long) to_kvmdev(vdev)->desc);
+ }
+ /*
+  * To reset the device, we use the KVM_VIRTIO_RESET hypercall, using the
+  * descriptor address. The Host will zero the status and all the
+  * features.
+  */
+ static void kvm_reset(struct virtio_device *vdev)
+ {
+       kvm_hypercall1(KVM_S390_VIRTIO_RESET,
+                      (unsigned long) to_kvmdev(vdev)->desc);
+ }
+ /*
+  * When the virtio_ring code wants to notify the Host, it calls us here and we
+  * make a hypercall.  We hand the address  of the virtqueue so the Host
+  * knows which virtqueue we're talking about.
+  */
+ static bool kvm_notify(struct virtqueue *vq)
+ {
+       long rc;
+       struct kvm_vqconfig *config = vq->priv;
+       rc = kvm_hypercall1(KVM_S390_VIRTIO_NOTIFY, config->address);
+       if (rc < 0)
+               return false;
+       return true;
+ }
+ /*
+  * This routine finds the first virtqueue described in the configuration of
+  * this device and sets it up.
+  */
+ static struct virtqueue *kvm_find_vq(struct virtio_device *vdev,
+                                    unsigned index,
+                                    void (*callback)(struct virtqueue *vq),
+                                    const char *name)
+ {
+       struct kvm_device *kdev = to_kvmdev(vdev);
+       struct kvm_vqconfig *config;
+       struct virtqueue *vq;
+       int err;
+       if (index >= kdev->desc->num_vq)
+               return ERR_PTR(-ENOENT);
+       if (!name)
+               return NULL;
+       config = kvm_vq_config(kdev->desc)+index;
+       err = vmem_add_mapping(config->address,
+                              vring_size(config->num,
+                                         KVM_S390_VIRTIO_RING_ALIGN));
+       if (err)
+               goto out;
+       vq = vring_new_virtqueue(index, config->num, KVM_S390_VIRTIO_RING_ALIGN,
+                                vdev, true, (void *) config->address,
+                                kvm_notify, callback, name);
+       if (!vq) {
+               err = -ENOMEM;
+               goto unmap;
+       }
+       /*
+        * register a callback token
+        * The host will sent this via the external interrupt parameter
+        */
+       config->token = (u64) vq;
+       vq->priv = config;
+       return vq;
+ unmap:
+       vmem_remove_mapping(config->address,
+                           vring_size(config->num,
+                                      KVM_S390_VIRTIO_RING_ALIGN));
+ out:
+       return ERR_PTR(err);
+ }
+ static void kvm_del_vq(struct virtqueue *vq)
+ {
+       struct kvm_vqconfig *config = vq->priv;
+       vring_del_virtqueue(vq);
+       vmem_remove_mapping(config->address,
+                           vring_size(config->num,
+                                      KVM_S390_VIRTIO_RING_ALIGN));
+ }
+ static void kvm_del_vqs(struct virtio_device *vdev)
+ {
+       struct virtqueue *vq, *n;
+       list_for_each_entry_safe(vq, n, &vdev->vqs, list)
+               kvm_del_vq(vq);
+ }
+ static int kvm_find_vqs(struct virtio_device *vdev, unsigned nvqs,
+                       struct virtqueue *vqs[],
+                       vq_callback_t *callbacks[],
+                       const char *names[])
+ {
+       struct kvm_device *kdev = to_kvmdev(vdev);
+       int i;
+       /* We must have this many virtqueues. */
+       if (nvqs > kdev->desc->num_vq)
+               return -ENOENT;
+       for (i = 0; i < nvqs; ++i) {
+               vqs[i] = kvm_find_vq(vdev, i, callbacks[i], names[i]);
+               if (IS_ERR(vqs[i]))
+                       goto error;
+       }
+       return 0;
+ error:
+       kvm_del_vqs(vdev);
+       return PTR_ERR(vqs[i]);
+ }
+ static const char *kvm_bus_name(struct virtio_device *vdev)
+ {
+       return "";
+ }
+ /*
+  * The config ops structure as defined by virtio config
+  */
+ static const struct virtio_config_ops kvm_vq_configspace_ops = {
+       .get_features = kvm_get_features,
+       .finalize_features = kvm_finalize_features,
+       .get = kvm_get,
+       .set = kvm_set,
+       .get_status = kvm_get_status,
+       .set_status = kvm_set_status,
+       .reset = kvm_reset,
+       .find_vqs = kvm_find_vqs,
+       .del_vqs = kvm_del_vqs,
+       .bus_name = kvm_bus_name,
+ };
+ /*
+  * The root device for the kvm virtio devices.
+  * This makes them appear as /sys/devices/kvm_s390/0,1,2 not /sys/devices/0,1,2.
+  */
+ static struct device *kvm_root;
+ /*
+  * adds a new device and register it with virtio
+  * appropriate drivers are loaded by the device model
+  */
+ static void add_kvm_device(struct kvm_device_desc *d, unsigned int offset)
+ {
+       struct kvm_device *kdev;
+       kdev = kzalloc(sizeof(*kdev), GFP_KERNEL);
+       if (!kdev) {
+               printk(KERN_EMERG "Cannot allocate kvm dev %u type %u\n",
+                      offset, d->type);
+               return;
+       }
+       kdev->vdev.dev.parent = kvm_root;
+       kdev->vdev.id.device = d->type;
+       kdev->vdev.config = &kvm_vq_configspace_ops;
+       kdev->desc = d;
+       if (register_virtio_device(&kdev->vdev) != 0) {
+               printk(KERN_ERR "Failed to register kvm device %u type %u\n",
+                      offset, d->type);
+               kfree(kdev);
+       }
+ }
+ /*
+  * scan_devices() simply iterates through the device page.
+  * The type 0 is reserved to mean "end of devices".
+  */
+ static void scan_devices(void)
+ {
+       unsigned int i;
+       struct kvm_device_desc *d;
+       for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
+               d = kvm_devices + i;
+               if (d->type == 0)
+                       break;
+               add_kvm_device(d, i);
+       }
+ }
+ /*
+  * match for a kvm device with a specific desc pointer
+  */
+ static int match_desc(struct device *dev, void *data)
+ {
+       struct virtio_device *vdev = dev_to_virtio(dev);
+       struct kvm_device *kdev = to_kvmdev(vdev);
+       return kdev->desc == data;
+ }
+ /*
+  * hotplug_device tries to find changes in the device page.
+  */
+ static void hotplug_devices(struct work_struct *dummy)
+ {
+       unsigned int i;
+       struct kvm_device_desc *d;
+       struct device *dev;
+       for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
+               d = kvm_devices + i;
+               /* end of list */
+               if (d->type == 0)
+                       break;
+               /* device already exists */
+               dev = device_find_child(kvm_root, d, match_desc);
+               if (dev) {
+                       /* XXX check for hotplug remove */
+                       put_device(dev);
+                       continue;
+               }
+               /* new device */
+               printk(KERN_INFO "Adding new virtio device %p\n", d);
+               add_kvm_device(d, i);
+       }
+ }
+ /*
+  * we emulate the request_irq behaviour on top of s390 extints
+  */
+ static void kvm_extint_handler(struct ext_code ext_code,
+                              unsigned int param32, unsigned long param64)
+ {
+       struct virtqueue *vq;
+       u32 param;
+       if ((ext_code.subcode & 0xff00) != VIRTIO_SUBCODE_64)
+               return;
+       inc_irq_stat(IRQEXT_VRT);
+       /* The LSB might be overloaded, we have to mask it */
+       vq = (struct virtqueue *)(param64 & ~1UL);
+       /* We use ext_params to decide what this interrupt means */
+       param = param32 & VIRTIO_PARAM_MASK;
+       switch (param) {
+       case VIRTIO_PARAM_CONFIG_CHANGED:
+               virtio_config_changed(vq->vdev);
+               break;
+       case VIRTIO_PARAM_DEV_ADD:
+               schedule_work(&hotplug_work);
+               break;
+       case VIRTIO_PARAM_VRING_INTERRUPT:
+       default:
+               vring_interrupt(0, vq);
+               break;
+       }
+ }
+ /*
+  * For s390-virtio, we expect a page above main storage containing
+  * the virtio configuration. Try to actually load from this area
+  * in order to figure out if the host provides this page.
+  */
+ static int __init test_devices_support(unsigned long addr)
+ {
+       int ret = -EIO;
+       asm volatile(
+               "0:     lura    0,%1\n"
+               "1:     xgr     %0,%0\n"
+               "2:\n"
+               EX_TABLE(0b,2b)
+               EX_TABLE(1b,2b)
+               : "+d" (ret)
+               : "a" (addr)
+               : "0", "cc");
+       return ret;
+ }
+ /*
+  * Init function for virtio
+  * devices are in a single page above top of "normal" + standby mem
+  */
+ static int __init kvm_devices_init(void)
+ {
+       int rc;
 -      if (sclp_has_vt220() || sclp_has_linemode())
++      unsigned long total_memory_size = sclp.rzm * sclp.rnmax;
+       if (!MACHINE_IS_KVM)
+               return -ENODEV;
+       if (test_devices_support(total_memory_size) < 0)
+               return -ENODEV;
+       rc = vmem_add_mapping(total_memory_size, PAGE_SIZE);
+       if (rc)
+               return rc;
+       kvm_devices = (void *) total_memory_size;
+       kvm_root = root_device_register("kvm_s390");
+       if (IS_ERR(kvm_root)) {
+               rc = PTR_ERR(kvm_root);
+               printk(KERN_ERR "Could not register kvm_s390 root device");
+               vmem_remove_mapping(total_memory_size, PAGE_SIZE);
+               return rc;
+       }
+       INIT_WORK(&hotplug_work, hotplug_devices);
+       irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
+       register_external_irq(EXT_IRQ_CP_SERVICE, kvm_extint_handler);
+       scan_devices();
+       return 0;
+ }
+ /* code for early console output with virtio_console */
+ static __init int early_put_chars(u32 vtermno, const char *buf, int count)
+ {
+       char scratch[17];
+       unsigned int len = count;
+       if (len > sizeof(scratch) - 1)
+               len = sizeof(scratch) - 1;
+       scratch[len] = '\0';
+       memcpy(scratch, buf, len);
+       kvm_hypercall1(KVM_S390_VIRTIO_NOTIFY, __pa(scratch));
+       return len;
+ }
+ static int __init s390_virtio_console_init(void)
+ {
++      if (sclp.has_vt220 || sclp.has_linemode)
+               return -ENODEV;
+       return virtio_cons_early_init(early_put_chars);
+ }
+ console_initcall(s390_virtio_console_init);
+ /*
+  * We do this after core stuff, but before the drivers.
+  */
+ postcore_initcall(kvm_devices_init);
index 0000000000000000000000000000000000000000,6f1fa1773e76418ef3d55917fb3146875dd4f5b7..f8d8fdb26b72a593260fa67b98337414a684e9b5
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,1380 +1,1387 @@@
 -      memcpy(vcdev->config, config_area, sizeof(vcdev->config));
 -      memcpy(buf, &vcdev->config[offset], len);
+ /*
+  * ccw based virtio transport
+  *
+  * Copyright IBM Corp. 2012, 2014
+  *
+  * This program is free software; you can redistribute it and/or modify
+  * it under the terms of the GNU General Public License (version 2 only)
+  * as published by the Free Software Foundation.
+  *
+  *    Author(s): Cornelia Huck <cornelia.huck@de.ibm.com>
+  */
+ #include <linux/kernel_stat.h>
+ #include <linux/init.h>
+ #include <linux/bootmem.h>
+ #include <linux/err.h>
+ #include <linux/virtio.h>
+ #include <linux/virtio_config.h>
+ #include <linux/slab.h>
+ #include <linux/interrupt.h>
+ #include <linux/virtio_ring.h>
+ #include <linux/pfn.h>
+ #include <linux/async.h>
+ #include <linux/wait.h>
+ #include <linux/list.h>
+ #include <linux/bitops.h>
+ #include <linux/module.h>
+ #include <linux/io.h>
+ #include <linux/kvm_para.h>
+ #include <linux/notifier.h>
+ #include <asm/setup.h>
+ #include <asm/irq.h>
+ #include <asm/cio.h>
+ #include <asm/ccwdev.h>
+ #include <asm/virtio-ccw.h>
+ #include <asm/isc.h>
+ #include <asm/airq.h>
+ /*
+  * virtio related functions
+  */
+ struct vq_config_block {
+       __u16 index;
+       __u16 num;
+ } __packed;
+ #define VIRTIO_CCW_CONFIG_SIZE 0x100
+ /* same as PCI config space size, should be enough for all drivers */
+ struct virtio_ccw_device {
+       struct virtio_device vdev;
+       __u8 *status;
+       __u8 config[VIRTIO_CCW_CONFIG_SIZE];
+       struct ccw_device *cdev;
+       __u32 curr_io;
+       int err;
+       unsigned int revision; /* Transport revision */
+       wait_queue_head_t wait_q;
+       spinlock_t lock;
+       struct list_head virtqueues;
+       unsigned long indicators;
+       unsigned long indicators2;
+       struct vq_config_block *config_block;
+       bool is_thinint;
+       bool going_away;
+       bool device_lost;
++      unsigned int config_ready;
+       void *airq_info;
+ };
+ struct vq_info_block_legacy {
+       __u64 queue;
+       __u32 align;
+       __u16 index;
+       __u16 num;
+ } __packed;
+ struct vq_info_block {
+       __u64 desc;
+       __u32 res0;
+       __u16 index;
+       __u16 num;
+       __u64 avail;
+       __u64 used;
+ } __packed;
+ struct virtio_feature_desc {
+       __u32 features;
+       __u8 index;
+ } __packed;
+ struct virtio_thinint_area {
+       unsigned long summary_indicator;
+       unsigned long indicator;
+       u64 bit_nr;
+       u8 isc;
+ } __packed;
+ struct virtio_rev_info {
+       __u16 revision;
+       __u16 length;
+       __u8 data[];
+ };
+ /* the highest virtio-ccw revision we support */
+ #define VIRTIO_CCW_REV_MAX 1
+ struct virtio_ccw_vq_info {
+       struct virtqueue *vq;
+       int num;
+       void *queue;
+       union {
+               struct vq_info_block s;
+               struct vq_info_block_legacy l;
+       } *info_block;
+       int bit_nr;
+       struct list_head node;
+       long cookie;
+ };
+ #define VIRTIO_AIRQ_ISC IO_SCH_ISC /* inherit from subchannel */
+ #define VIRTIO_IV_BITS (L1_CACHE_BYTES * 8)
+ #define MAX_AIRQ_AREAS 20
+ static int virtio_ccw_use_airq = 1;
+ struct airq_info {
+       rwlock_t lock;
+       u8 summary_indicator;
+       struct airq_struct airq;
+       struct airq_iv *aiv;
+ };
+ static struct airq_info *airq_areas[MAX_AIRQ_AREAS];
+ #define CCW_CMD_SET_VQ 0x13
+ #define CCW_CMD_VDEV_RESET 0x33
+ #define CCW_CMD_SET_IND 0x43
+ #define CCW_CMD_SET_CONF_IND 0x53
+ #define CCW_CMD_READ_FEAT 0x12
+ #define CCW_CMD_WRITE_FEAT 0x11
+ #define CCW_CMD_READ_CONF 0x22
+ #define CCW_CMD_WRITE_CONF 0x21
+ #define CCW_CMD_WRITE_STATUS 0x31
+ #define CCW_CMD_READ_VQ_CONF 0x32
+ #define CCW_CMD_SET_IND_ADAPTER 0x73
+ #define CCW_CMD_SET_VIRTIO_REV 0x83
+ #define VIRTIO_CCW_DOING_SET_VQ 0x00010000
+ #define VIRTIO_CCW_DOING_RESET 0x00040000
+ #define VIRTIO_CCW_DOING_READ_FEAT 0x00080000
+ #define VIRTIO_CCW_DOING_WRITE_FEAT 0x00100000
+ #define VIRTIO_CCW_DOING_READ_CONFIG 0x00200000
+ #define VIRTIO_CCW_DOING_WRITE_CONFIG 0x00400000
+ #define VIRTIO_CCW_DOING_WRITE_STATUS 0x00800000
+ #define VIRTIO_CCW_DOING_SET_IND 0x01000000
+ #define VIRTIO_CCW_DOING_READ_VQ_CONF 0x02000000
+ #define VIRTIO_CCW_DOING_SET_CONF_IND 0x04000000
+ #define VIRTIO_CCW_DOING_SET_IND_ADAPTER 0x08000000
+ #define VIRTIO_CCW_DOING_SET_VIRTIO_REV 0x10000000
+ #define VIRTIO_CCW_INTPARM_MASK 0xffff0000
+ static struct virtio_ccw_device *to_vc_device(struct virtio_device *vdev)
+ {
+       return container_of(vdev, struct virtio_ccw_device, vdev);
+ }
+ static void drop_airq_indicator(struct virtqueue *vq, struct airq_info *info)
+ {
+       unsigned long i, flags;
+       write_lock_irqsave(&info->lock, flags);
+       for (i = 0; i < airq_iv_end(info->aiv); i++) {
+               if (vq == (void *)airq_iv_get_ptr(info->aiv, i)) {
+                       airq_iv_free_bit(info->aiv, i);
+                       airq_iv_set_ptr(info->aiv, i, 0);
+                       break;
+               }
+       }
+       write_unlock_irqrestore(&info->lock, flags);
+ }
+ static void virtio_airq_handler(struct airq_struct *airq)
+ {
+       struct airq_info *info = container_of(airq, struct airq_info, airq);
+       unsigned long ai;
+       inc_irq_stat(IRQIO_VAI);
+       read_lock(&info->lock);
+       /* Walk through indicators field, summary indicator active. */
+       for (ai = 0;;) {
+               ai = airq_iv_scan(info->aiv, ai, airq_iv_end(info->aiv));
+               if (ai == -1UL)
+                       break;
+               vring_interrupt(0, (void *)airq_iv_get_ptr(info->aiv, ai));
+       }
+       info->summary_indicator = 0;
+       smp_wmb();
+       /* Walk through indicators field, summary indicator not active. */
+       for (ai = 0;;) {
+               ai = airq_iv_scan(info->aiv, ai, airq_iv_end(info->aiv));
+               if (ai == -1UL)
+                       break;
+               vring_interrupt(0, (void *)airq_iv_get_ptr(info->aiv, ai));
+       }
+       read_unlock(&info->lock);
+ }
+ static struct airq_info *new_airq_info(void)
+ {
+       struct airq_info *info;
+       int rc;
+       info = kzalloc(sizeof(*info), GFP_KERNEL);
+       if (!info)
+               return NULL;
+       rwlock_init(&info->lock);
+       info->aiv = airq_iv_create(VIRTIO_IV_BITS, AIRQ_IV_ALLOC | AIRQ_IV_PTR);
+       if (!info->aiv) {
+               kfree(info);
+               return NULL;
+       }
+       info->airq.handler = virtio_airq_handler;
+       info->airq.lsi_ptr = &info->summary_indicator;
+       info->airq.lsi_mask = 0xff;
+       info->airq.isc = VIRTIO_AIRQ_ISC;
+       rc = register_adapter_interrupt(&info->airq);
+       if (rc) {
+               airq_iv_release(info->aiv);
+               kfree(info);
+               return NULL;
+       }
+       return info;
+ }
+ static void destroy_airq_info(struct airq_info *info)
+ {
+       if (!info)
+               return;
+       unregister_adapter_interrupt(&info->airq);
+       airq_iv_release(info->aiv);
+       kfree(info);
+ }
+ static unsigned long get_airq_indicator(struct virtqueue *vqs[], int nvqs,
+                                       u64 *first, void **airq_info)
+ {
+       int i, j;
+       struct airq_info *info;
+       unsigned long indicator_addr = 0;
+       unsigned long bit, flags;
+       for (i = 0; i < MAX_AIRQ_AREAS && !indicator_addr; i++) {
+               if (!airq_areas[i])
+                       airq_areas[i] = new_airq_info();
+               info = airq_areas[i];
+               if (!info)
+                       return 0;
+               write_lock_irqsave(&info->lock, flags);
+               bit = airq_iv_alloc(info->aiv, nvqs);
+               if (bit == -1UL) {
+                       /* Not enough vacancies. */
+                       write_unlock_irqrestore(&info->lock, flags);
+                       continue;
+               }
+               *first = bit;
+               *airq_info = info;
+               indicator_addr = (unsigned long)info->aiv->vector;
+               for (j = 0; j < nvqs; j++) {
+                       airq_iv_set_ptr(info->aiv, bit + j,
+                                       (unsigned long)vqs[j]);
+               }
+               write_unlock_irqrestore(&info->lock, flags);
+       }
+       return indicator_addr;
+ }
+ static void virtio_ccw_drop_indicators(struct virtio_ccw_device *vcdev)
+ {
+       struct virtio_ccw_vq_info *info;
+       list_for_each_entry(info, &vcdev->virtqueues, node)
+               drop_airq_indicator(info->vq, vcdev->airq_info);
+ }
+ static int doing_io(struct virtio_ccw_device *vcdev, __u32 flag)
+ {
+       unsigned long flags;
+       __u32 ret;
+       spin_lock_irqsave(get_ccwdev_lock(vcdev->cdev), flags);
+       if (vcdev->err)
+               ret = 0;
+       else
+               ret = vcdev->curr_io & flag;
+       spin_unlock_irqrestore(get_ccwdev_lock(vcdev->cdev), flags);
+       return ret;
+ }
+ static int ccw_io_helper(struct virtio_ccw_device *vcdev,
+                        struct ccw1 *ccw, __u32 intparm)
+ {
+       int ret;
+       unsigned long flags;
+       int flag = intparm & VIRTIO_CCW_INTPARM_MASK;
+       do {
+               spin_lock_irqsave(get_ccwdev_lock(vcdev->cdev), flags);
+               ret = ccw_device_start(vcdev->cdev, ccw, intparm, 0, 0);
+               if (!ret) {
+                       if (!vcdev->curr_io)
+                               vcdev->err = 0;
+                       vcdev->curr_io |= flag;
+               }
+               spin_unlock_irqrestore(get_ccwdev_lock(vcdev->cdev), flags);
+               cpu_relax();
+       } while (ret == -EBUSY);
+       wait_event(vcdev->wait_q, doing_io(vcdev, flag) == 0);
+       return ret ? ret : vcdev->err;
+ }
+ static void virtio_ccw_drop_indicator(struct virtio_ccw_device *vcdev,
+                                     struct ccw1 *ccw)
+ {
+       int ret;
+       unsigned long *indicatorp = NULL;
+       struct virtio_thinint_area *thinint_area = NULL;
+       struct airq_info *airq_info = vcdev->airq_info;
+       if (vcdev->is_thinint) {
+               thinint_area = kzalloc(sizeof(*thinint_area),
+                                      GFP_DMA | GFP_KERNEL);
+               if (!thinint_area)
+                       return;
+               thinint_area->summary_indicator =
+                       (unsigned long) &airq_info->summary_indicator;
+               thinint_area->isc = VIRTIO_AIRQ_ISC;
+               ccw->cmd_code = CCW_CMD_SET_IND_ADAPTER;
+               ccw->count = sizeof(*thinint_area);
+               ccw->cda = (__u32)(unsigned long) thinint_area;
+       } else {
+               indicatorp = kmalloc(sizeof(&vcdev->indicators),
+                                    GFP_DMA | GFP_KERNEL);
+               if (!indicatorp)
+                       return;
+               *indicatorp = 0;
+               ccw->cmd_code = CCW_CMD_SET_IND;
+               ccw->count = sizeof(vcdev->indicators);
+               ccw->cda = (__u32)(unsigned long) indicatorp;
+       }
+       /* Deregister indicators from host. */
+       vcdev->indicators = 0;
+       ccw->flags = 0;
+       ret = ccw_io_helper(vcdev, ccw,
+                           vcdev->is_thinint ?
+                           VIRTIO_CCW_DOING_SET_IND_ADAPTER :
+                           VIRTIO_CCW_DOING_SET_IND);
+       if (ret && (ret != -ENODEV))
+               dev_info(&vcdev->cdev->dev,
+                        "Failed to deregister indicators (%d)\n", ret);
+       else if (vcdev->is_thinint)
+               virtio_ccw_drop_indicators(vcdev);
+       kfree(indicatorp);
+       kfree(thinint_area);
+ }
+ static inline long do_kvm_notify(struct subchannel_id schid,
+                                unsigned long queue_index,
+                                long cookie)
+ {
+       register unsigned long __nr asm("1") = KVM_S390_VIRTIO_CCW_NOTIFY;
+       register struct subchannel_id __schid asm("2") = schid;
+       register unsigned long __index asm("3") = queue_index;
+       register long __rc asm("2");
+       register long __cookie asm("4") = cookie;
+       asm volatile ("diag 2,4,0x500\n"
+                     : "=d" (__rc) : "d" (__nr), "d" (__schid), "d" (__index),
+                     "d"(__cookie)
+                     : "memory", "cc");
+       return __rc;
+ }
+ static bool virtio_ccw_kvm_notify(struct virtqueue *vq)
+ {
+       struct virtio_ccw_vq_info *info = vq->priv;
+       struct virtio_ccw_device *vcdev;
+       struct subchannel_id schid;
+       vcdev = to_vc_device(info->vq->vdev);
+       ccw_device_get_schid(vcdev->cdev, &schid);
+       info->cookie = do_kvm_notify(schid, vq->index, info->cookie);
+       if (info->cookie < 0)
+               return false;
+       return true;
+ }
+ static int virtio_ccw_read_vq_conf(struct virtio_ccw_device *vcdev,
+                                  struct ccw1 *ccw, int index)
+ {
+       vcdev->config_block->index = index;
+       ccw->cmd_code = CCW_CMD_READ_VQ_CONF;
+       ccw->flags = 0;
+       ccw->count = sizeof(struct vq_config_block);
+       ccw->cda = (__u32)(unsigned long)(vcdev->config_block);
+       ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_READ_VQ_CONF);
+       return vcdev->config_block->num;
+ }
+ static void virtio_ccw_del_vq(struct virtqueue *vq, struct ccw1 *ccw)
+ {
+       struct virtio_ccw_device *vcdev = to_vc_device(vq->vdev);
+       struct virtio_ccw_vq_info *info = vq->priv;
+       unsigned long flags;
+       unsigned long size;
+       int ret;
+       unsigned int index = vq->index;
+       /* Remove from our list. */
+       spin_lock_irqsave(&vcdev->lock, flags);
+       list_del(&info->node);
+       spin_unlock_irqrestore(&vcdev->lock, flags);
+       /* Release from host. */
+       if (vcdev->revision == 0) {
+               info->info_block->l.queue = 0;
+               info->info_block->l.align = 0;
+               info->info_block->l.index = index;
+               info->info_block->l.num = 0;
+               ccw->count = sizeof(info->info_block->l);
+       } else {
+               info->info_block->s.desc = 0;
+               info->info_block->s.index = index;
+               info->info_block->s.num = 0;
+               info->info_block->s.avail = 0;
+               info->info_block->s.used = 0;
+               ccw->count = sizeof(info->info_block->s);
+       }
+       ccw->cmd_code = CCW_CMD_SET_VQ;
+       ccw->flags = 0;
+       ccw->cda = (__u32)(unsigned long)(info->info_block);
+       ret = ccw_io_helper(vcdev, ccw,
+                           VIRTIO_CCW_DOING_SET_VQ | index);
+       /*
+        * -ENODEV isn't considered an error: The device is gone anyway.
+        * This may happen on device detach.
+        */
+       if (ret && (ret != -ENODEV))
+               dev_warn(&vq->vdev->dev, "Error %d while deleting queue %d",
+                        ret, index);
+       vring_del_virtqueue(vq);
+       size = PAGE_ALIGN(vring_size(info->num, KVM_VIRTIO_CCW_RING_ALIGN));
+       free_pages_exact(info->queue, size);
+       kfree(info->info_block);
+       kfree(info);
+ }
+ static void virtio_ccw_del_vqs(struct virtio_device *vdev)
+ {
+       struct virtqueue *vq, *n;
+       struct ccw1 *ccw;
+       struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+       ccw = kzalloc(sizeof(*ccw), GFP_DMA | GFP_KERNEL);
+       if (!ccw)
+               return;
+       virtio_ccw_drop_indicator(vcdev, ccw);
+       list_for_each_entry_safe(vq, n, &vdev->vqs, list)
+               virtio_ccw_del_vq(vq, ccw);
+       kfree(ccw);
+ }
+ static struct virtqueue *virtio_ccw_setup_vq(struct virtio_device *vdev,
+                                            int i, vq_callback_t *callback,
+                                            const char *name,
+                                            struct ccw1 *ccw)
+ {
+       struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+       int err;
+       struct virtqueue *vq = NULL;
+       struct virtio_ccw_vq_info *info;
+       unsigned long size = 0; /* silence the compiler */
+       unsigned long flags;
+       /* Allocate queue. */
+       info = kzalloc(sizeof(struct virtio_ccw_vq_info), GFP_KERNEL);
+       if (!info) {
+               dev_warn(&vcdev->cdev->dev, "no info\n");
+               err = -ENOMEM;
+               goto out_err;
+       }
+       info->info_block = kzalloc(sizeof(*info->info_block),
+                                  GFP_DMA | GFP_KERNEL);
+       if (!info->info_block) {
+               dev_warn(&vcdev->cdev->dev, "no info block\n");
+               err = -ENOMEM;
+               goto out_err;
+       }
+       info->num = virtio_ccw_read_vq_conf(vcdev, ccw, i);
+       size = PAGE_ALIGN(vring_size(info->num, KVM_VIRTIO_CCW_RING_ALIGN));
+       info->queue = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
+       if (info->queue == NULL) {
+               dev_warn(&vcdev->cdev->dev, "no queue\n");
+               err = -ENOMEM;
+               goto out_err;
+       }
+       vq = vring_new_virtqueue(i, info->num, KVM_VIRTIO_CCW_RING_ALIGN, vdev,
+                                true, info->queue, virtio_ccw_kvm_notify,
+                                callback, name);
+       if (!vq) {
+               /* For now, we fail if we can't get the requested size. */
+               dev_warn(&vcdev->cdev->dev, "no vq\n");
+               err = -ENOMEM;
+               goto out_err;
+       }
+       /* Register it with the host. */
+       if (vcdev->revision == 0) {
+               info->info_block->l.queue = (__u64)info->queue;
+               info->info_block->l.align = KVM_VIRTIO_CCW_RING_ALIGN;
+               info->info_block->l.index = i;
+               info->info_block->l.num = info->num;
+               ccw->count = sizeof(info->info_block->l);
+       } else {
+               info->info_block->s.desc = (__u64)info->queue;
+               info->info_block->s.index = i;
+               info->info_block->s.num = info->num;
+               info->info_block->s.avail = (__u64)virtqueue_get_avail(vq);
+               info->info_block->s.used = (__u64)virtqueue_get_used(vq);
+               ccw->count = sizeof(info->info_block->s);
+       }
+       ccw->cmd_code = CCW_CMD_SET_VQ;
+       ccw->flags = 0;
+       ccw->cda = (__u32)(unsigned long)(info->info_block);
+       err = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_SET_VQ | i);
+       if (err) {
+               dev_warn(&vcdev->cdev->dev, "SET_VQ failed\n");
+               goto out_err;
+       }
+       info->vq = vq;
+       vq->priv = info;
+       /* Save it to our list. */
+       spin_lock_irqsave(&vcdev->lock, flags);
+       list_add(&info->node, &vcdev->virtqueues);
+       spin_unlock_irqrestore(&vcdev->lock, flags);
+       return vq;
+ out_err:
+       if (vq)
+               vring_del_virtqueue(vq);
+       if (info) {
+               if (info->queue)
+                       free_pages_exact(info->queue, size);
+               kfree(info->info_block);
+       }
+       kfree(info);
+       return ERR_PTR(err);
+ }
+ static int virtio_ccw_register_adapter_ind(struct virtio_ccw_device *vcdev,
+                                          struct virtqueue *vqs[], int nvqs,
+                                          struct ccw1 *ccw)
+ {
+       int ret;
+       struct virtio_thinint_area *thinint_area = NULL;
+       struct airq_info *info;
+       thinint_area = kzalloc(sizeof(*thinint_area), GFP_DMA | GFP_KERNEL);
+       if (!thinint_area) {
+               ret = -ENOMEM;
+               goto out;
+       }
+       /* Try to get an indicator. */
+       thinint_area->indicator = get_airq_indicator(vqs, nvqs,
+                                                    &thinint_area->bit_nr,
+                                                    &vcdev->airq_info);
+       if (!thinint_area->indicator) {
+               ret = -ENOSPC;
+               goto out;
+       }
+       info = vcdev->airq_info;
+       thinint_area->summary_indicator =
+               (unsigned long) &info->summary_indicator;
+       thinint_area->isc = VIRTIO_AIRQ_ISC;
+       ccw->cmd_code = CCW_CMD_SET_IND_ADAPTER;
+       ccw->flags = CCW_FLAG_SLI;
+       ccw->count = sizeof(*thinint_area);
+       ccw->cda = (__u32)(unsigned long)thinint_area;
+       ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_SET_IND_ADAPTER);
+       if (ret) {
+               if (ret == -EOPNOTSUPP) {
+                       /*
+                        * The host does not support adapter interrupts
+                        * for virtio-ccw, stop trying.
+                        */
+                       virtio_ccw_use_airq = 0;
+                       pr_info("Adapter interrupts unsupported on host\n");
+               } else
+                       dev_warn(&vcdev->cdev->dev,
+                                "enabling adapter interrupts = %d\n", ret);
+               virtio_ccw_drop_indicators(vcdev);
+       }
+ out:
+       kfree(thinint_area);
+       return ret;
+ }
+ static int virtio_ccw_find_vqs(struct virtio_device *vdev, unsigned nvqs,
+                              struct virtqueue *vqs[],
+                              vq_callback_t *callbacks[],
+                              const char *names[])
+ {
+       struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+       unsigned long *indicatorp = NULL;
+       int ret, i;
+       struct ccw1 *ccw;
+       ccw = kzalloc(sizeof(*ccw), GFP_DMA | GFP_KERNEL);
+       if (!ccw)
+               return -ENOMEM;
+       for (i = 0; i < nvqs; ++i) {
+               vqs[i] = virtio_ccw_setup_vq(vdev, i, callbacks[i], names[i],
+                                            ccw);
+               if (IS_ERR(vqs[i])) {
+                       ret = PTR_ERR(vqs[i]);
+                       vqs[i] = NULL;
+                       goto out;
+               }
+       }
+       ret = -ENOMEM;
+       /* We need a data area under 2G to communicate. */
+       indicatorp = kmalloc(sizeof(&vcdev->indicators), GFP_DMA | GFP_KERNEL);
+       if (!indicatorp)
+               goto out;
+       *indicatorp = (unsigned long) &vcdev->indicators;
+       if (vcdev->is_thinint) {
+               ret = virtio_ccw_register_adapter_ind(vcdev, vqs, nvqs, ccw);
+               if (ret)
+                       /* no error, just fall back to legacy interrupts */
+                       vcdev->is_thinint = 0;
+       }
+       if (!vcdev->is_thinint) {
+               /* Register queue indicators with host. */
+               vcdev->indicators = 0;
+               ccw->cmd_code = CCW_CMD_SET_IND;
+               ccw->flags = 0;
+               ccw->count = sizeof(vcdev->indicators);
+               ccw->cda = (__u32)(unsigned long) indicatorp;
+               ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_SET_IND);
+               if (ret)
+                       goto out;
+       }
+       /* Register indicators2 with host for config changes */
+       *indicatorp = (unsigned long) &vcdev->indicators2;
+       vcdev->indicators2 = 0;
+       ccw->cmd_code = CCW_CMD_SET_CONF_IND;
+       ccw->flags = 0;
+       ccw->count = sizeof(vcdev->indicators2);
+       ccw->cda = (__u32)(unsigned long) indicatorp;
+       ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_SET_CONF_IND);
+       if (ret)
+               goto out;
+       kfree(indicatorp);
+       kfree(ccw);
+       return 0;
+ out:
+       kfree(indicatorp);
+       kfree(ccw);
+       virtio_ccw_del_vqs(vdev);
+       return ret;
+ }
+ static void virtio_ccw_reset(struct virtio_device *vdev)
+ {
+       struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+       struct ccw1 *ccw;
+       ccw = kzalloc(sizeof(*ccw), GFP_DMA | GFP_KERNEL);
+       if (!ccw)
+               return;
+       /* Zero status bits. */
+       *vcdev->status = 0;
+       /* Send a reset ccw on device. */
+       ccw->cmd_code = CCW_CMD_VDEV_RESET;
+       ccw->flags = 0;
+       ccw->count = 0;
+       ccw->cda = 0;
+       ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_RESET);
+       kfree(ccw);
+ }
+ static u64 virtio_ccw_get_features(struct virtio_device *vdev)
+ {
+       struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+       struct virtio_feature_desc *features;
+       int ret;
+       u64 rc;
+       struct ccw1 *ccw;
+       ccw = kzalloc(sizeof(*ccw), GFP_DMA | GFP_KERNEL);
+       if (!ccw)
+               return 0;
+       features = kzalloc(sizeof(*features), GFP_DMA | GFP_KERNEL);
+       if (!features) {
+               rc = 0;
+               goto out_free;
+       }
+       /* Read the feature bits from the host. */
+       features->index = 0;
+       ccw->cmd_code = CCW_CMD_READ_FEAT;
+       ccw->flags = 0;
+       ccw->count = sizeof(*features);
+       ccw->cda = (__u32)(unsigned long)features;
+       ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_READ_FEAT);
+       if (ret) {
+               rc = 0;
+               goto out_free;
+       }
+       rc = le32_to_cpu(features->features);
+       if (vcdev->revision == 0)
+               goto out_free;
+       /* Read second half of the feature bits from the host. */
+       features->index = 1;
+       ccw->cmd_code = CCW_CMD_READ_FEAT;
+       ccw->flags = 0;
+       ccw->count = sizeof(*features);
+       ccw->cda = (__u32)(unsigned long)features;
+       ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_READ_FEAT);
+       if (ret == 0)
+               rc |= (u64)le32_to_cpu(features->features) << 32;
+ out_free:
+       kfree(features);
+       kfree(ccw);
+       return rc;
+ }
+ static int virtio_ccw_finalize_features(struct virtio_device *vdev)
+ {
+       struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+       struct virtio_feature_desc *features;
+       struct ccw1 *ccw;
+       int ret;
+       if (vcdev->revision >= 1 &&
+           !__virtio_test_bit(vdev, VIRTIO_F_VERSION_1)) {
+               dev_err(&vdev->dev, "virtio: device uses revision 1 "
+                       "but does not have VIRTIO_F_VERSION_1\n");
+               return -EINVAL;
+       }
+       ccw = kzalloc(sizeof(*ccw), GFP_DMA | GFP_KERNEL);
+       if (!ccw)
+               return -ENOMEM;
+       features = kzalloc(sizeof(*features), GFP_DMA | GFP_KERNEL);
+       if (!features) {
+               ret = -ENOMEM;
+               goto out_free;
+       }
+       /* Give virtio_ring a chance to accept features. */
+       vring_transport_features(vdev);
+       features->index = 0;
+       features->features = cpu_to_le32((u32)vdev->features);
+       /* Write the first half of the feature bits to the host. */
+       ccw->cmd_code = CCW_CMD_WRITE_FEAT;
+       ccw->flags = 0;
+       ccw->count = sizeof(*features);
+       ccw->cda = (__u32)(unsigned long)features;
+       ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_FEAT);
+       if (ret)
+               goto out_free;
+       if (vcdev->revision == 0)
+               goto out_free;
+       features->index = 1;
+       features->features = cpu_to_le32(vdev->features >> 32);
+       /* Write the second half of the feature bits to the host. */
+       ccw->cmd_code = CCW_CMD_WRITE_FEAT;
+       ccw->flags = 0;
+       ccw->count = sizeof(*features);
+       ccw->cda = (__u32)(unsigned long)features;
+       ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_FEAT);
+ out_free:
+       kfree(features);
+       kfree(ccw);
+       return ret;
+ }
+ static void virtio_ccw_get_config(struct virtio_device *vdev,
+                                 unsigned int offset, void *buf, unsigned len)
+ {
+       struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+       int ret;
+       struct ccw1 *ccw;
+       void *config_area;
+       ccw = kzalloc(sizeof(*ccw), GFP_DMA | GFP_KERNEL);
+       if (!ccw)
+               return;
+       config_area = kzalloc(VIRTIO_CCW_CONFIG_SIZE, GFP_DMA | GFP_KERNEL);
+       if (!config_area)
+               goto out_free;
+       /* Read the config area from the host. */
+       ccw->cmd_code = CCW_CMD_READ_CONF;
+       ccw->flags = 0;
+       ccw->count = offset + len;
+       ccw->cda = (__u32)(unsigned long)config_area;
+       ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_READ_CONFIG);
+       if (ret)
+               goto out_free;
++      memcpy(vcdev->config, config_area, offset + len);
++      if (buf)
++              memcpy(buf, &vcdev->config[offset], len);
++      if (vcdev->config_ready < offset + len)
++              vcdev->config_ready = offset + len;
+ out_free:
+       kfree(config_area);
+       kfree(ccw);
+ }
+ static void virtio_ccw_set_config(struct virtio_device *vdev,
+                                 unsigned int offset, const void *buf,
+                                 unsigned len)
+ {
+       struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+       struct ccw1 *ccw;
+       void *config_area;
+       ccw = kzalloc(sizeof(*ccw), GFP_DMA | GFP_KERNEL);
+       if (!ccw)
+               return;
+       config_area = kzalloc(VIRTIO_CCW_CONFIG_SIZE, GFP_DMA | GFP_KERNEL);
+       if (!config_area)
+               goto out_free;
++      /* Make sure we don't overwrite fields. */
++      if (vcdev->config_ready < offset)
++              virtio_ccw_get_config(vdev, 0, NULL, offset);
+       memcpy(&vcdev->config[offset], buf, len);
+       /* Write the config area to the host. */
+       memcpy(config_area, vcdev->config, sizeof(vcdev->config));
+       ccw->cmd_code = CCW_CMD_WRITE_CONF;
+       ccw->flags = 0;
+       ccw->count = offset + len;
+       ccw->cda = (__u32)(unsigned long)config_area;
+       ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_CONFIG);
+ out_free:
+       kfree(config_area);
+       kfree(ccw);
+ }
+ static u8 virtio_ccw_get_status(struct virtio_device *vdev)
+ {
+       struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+       return *vcdev->status;
+ }
+ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
+ {
+       struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+       u8 old_status = *vcdev->status;
+       struct ccw1 *ccw;
+       int ret;
+       ccw = kzalloc(sizeof(*ccw), GFP_DMA | GFP_KERNEL);
+       if (!ccw)
+               return;
+       /* Write the status to the host. */
+       *vcdev->status = status;
+       ccw->cmd_code = CCW_CMD_WRITE_STATUS;
+       ccw->flags = 0;
+       ccw->count = sizeof(status);
+       ccw->cda = (__u32)(unsigned long)vcdev->status;
+       ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
+       /* Write failed? We assume status is unchanged. */
+       if (ret)
+               *vcdev->status = old_status;
+       kfree(ccw);
+ }
+ static struct virtio_config_ops virtio_ccw_config_ops = {
+       .get_features = virtio_ccw_get_features,
+       .finalize_features = virtio_ccw_finalize_features,
+       .get = virtio_ccw_get_config,
+       .set = virtio_ccw_set_config,
+       .get_status = virtio_ccw_get_status,
+       .set_status = virtio_ccw_set_status,
+       .reset = virtio_ccw_reset,
+       .find_vqs = virtio_ccw_find_vqs,
+       .del_vqs = virtio_ccw_del_vqs,
+ };
+ /*
+  * ccw bus driver related functions
+  */
+ static void virtio_ccw_release_dev(struct device *_d)
+ {
+       struct virtio_device *dev = container_of(_d, struct virtio_device,
+                                                dev);
+       struct virtio_ccw_device *vcdev = to_vc_device(dev);
+       kfree(vcdev->status);
+       kfree(vcdev->config_block);
+       kfree(vcdev);
+ }
+ static int irb_is_error(struct irb *irb)
+ {
+       if (scsw_cstat(&irb->scsw) != 0)
+               return 1;
+       if (scsw_dstat(&irb->scsw) & ~(DEV_STAT_CHN_END | DEV_STAT_DEV_END))
+               return 1;
+       if (scsw_cc(&irb->scsw) != 0)
+               return 1;
+       return 0;
+ }
+ static struct virtqueue *virtio_ccw_vq_by_ind(struct virtio_ccw_device *vcdev,
+                                             int index)
+ {
+       struct virtio_ccw_vq_info *info;
+       unsigned long flags;
+       struct virtqueue *vq;
+       vq = NULL;
+       spin_lock_irqsave(&vcdev->lock, flags);
+       list_for_each_entry(info, &vcdev->virtqueues, node) {
+               if (info->vq->index == index) {
+                       vq = info->vq;
+                       break;
+               }
+       }
+       spin_unlock_irqrestore(&vcdev->lock, flags);
+       return vq;
+ }
+ static void virtio_ccw_int_handler(struct ccw_device *cdev,
+                                  unsigned long intparm,
+                                  struct irb *irb)
+ {
+       __u32 activity = intparm & VIRTIO_CCW_INTPARM_MASK;
+       struct virtio_ccw_device *vcdev = dev_get_drvdata(&cdev->dev);
+       int i;
+       struct virtqueue *vq;
+       if (!vcdev)
+               return;
+       /* Check if it's a notification from the host. */
+       if ((intparm == 0) &&
+           (scsw_stctl(&irb->scsw) ==
+            (SCSW_STCTL_ALERT_STATUS | SCSW_STCTL_STATUS_PEND))) {
+               /* OK */
+       }
+       if (irb_is_error(irb)) {
+               /* Command reject? */
+               if ((scsw_dstat(&irb->scsw) & DEV_STAT_UNIT_CHECK) &&
+                   (irb->ecw[0] & SNS0_CMD_REJECT))
+                       vcdev->err = -EOPNOTSUPP;
+               else
+                       /* Map everything else to -EIO. */
+                       vcdev->err = -EIO;
+       }
+       if (vcdev->curr_io & activity) {
+               switch (activity) {
+               case VIRTIO_CCW_DOING_READ_FEAT:
+               case VIRTIO_CCW_DOING_WRITE_FEAT:
+               case VIRTIO_CCW_DOING_READ_CONFIG:
+               case VIRTIO_CCW_DOING_WRITE_CONFIG:
+               case VIRTIO_CCW_DOING_WRITE_STATUS:
+               case VIRTIO_CCW_DOING_SET_VQ:
+               case VIRTIO_CCW_DOING_SET_IND:
+               case VIRTIO_CCW_DOING_SET_CONF_IND:
+               case VIRTIO_CCW_DOING_RESET:
+               case VIRTIO_CCW_DOING_READ_VQ_CONF:
+               case VIRTIO_CCW_DOING_SET_IND_ADAPTER:
+               case VIRTIO_CCW_DOING_SET_VIRTIO_REV:
+                       vcdev->curr_io &= ~activity;
+                       wake_up(&vcdev->wait_q);
+                       break;
+               default:
+                       /* don't know what to do... */
+                       dev_warn(&cdev->dev, "Suspicious activity '%08x'\n",
+                                activity);
+                       WARN_ON(1);
+                       break;
+               }
+       }
+       for_each_set_bit(i, &vcdev->indicators,
+                        sizeof(vcdev->indicators) * BITS_PER_BYTE) {
+               /* The bit clear must happen before the vring kick. */
+               clear_bit(i, &vcdev->indicators);
+               barrier();
+               vq = virtio_ccw_vq_by_ind(vcdev, i);
+               vring_interrupt(0, vq);
+       }
+       if (test_bit(0, &vcdev->indicators2)) {
+               virtio_config_changed(&vcdev->vdev);
+               clear_bit(0, &vcdev->indicators2);
+       }
+ }
+ /*
+  * We usually want to autoonline all devices, but give the admin
+  * a way to exempt devices from this.
+  */
+ #define __DEV_WORDS ((__MAX_SUBCHANNEL + (8*sizeof(long) - 1)) / \
+                    (8*sizeof(long)))
+ static unsigned long devs_no_auto[__MAX_SSID + 1][__DEV_WORDS];
+ static char *no_auto = "";
+ module_param(no_auto, charp, 0444);
+ MODULE_PARM_DESC(no_auto, "list of ccw bus id ranges not to be auto-onlined");
+ static int virtio_ccw_check_autoonline(struct ccw_device *cdev)
+ {
+       struct ccw_dev_id id;
+       ccw_device_get_id(cdev, &id);
+       if (test_bit(id.devno, devs_no_auto[id.ssid]))
+               return 0;
+       return 1;
+ }
+ static void virtio_ccw_auto_online(void *data, async_cookie_t cookie)
+ {
+       struct ccw_device *cdev = data;
+       int ret;
+       ret = ccw_device_set_online(cdev);
+       if (ret)
+               dev_warn(&cdev->dev, "Failed to set online: %d\n", ret);
+ }
+ static int virtio_ccw_probe(struct ccw_device *cdev)
+ {
+       cdev->handler = virtio_ccw_int_handler;
+       if (virtio_ccw_check_autoonline(cdev))
+               async_schedule(virtio_ccw_auto_online, cdev);
+       return 0;
+ }
+ static struct virtio_ccw_device *virtio_grab_drvdata(struct ccw_device *cdev)
+ {
+       unsigned long flags;
+       struct virtio_ccw_device *vcdev;
+       spin_lock_irqsave(get_ccwdev_lock(cdev), flags);
+       vcdev = dev_get_drvdata(&cdev->dev);
+       if (!vcdev || vcdev->going_away) {
+               spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
+               return NULL;
+       }
+       vcdev->going_away = true;
+       spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
+       return vcdev;
+ }
+ static void virtio_ccw_remove(struct ccw_device *cdev)
+ {
+       unsigned long flags;
+       struct virtio_ccw_device *vcdev = virtio_grab_drvdata(cdev);
+       if (vcdev && cdev->online) {
+               if (vcdev->device_lost)
+                       virtio_break_device(&vcdev->vdev);
+               unregister_virtio_device(&vcdev->vdev);
+               spin_lock_irqsave(get_ccwdev_lock(cdev), flags);
+               dev_set_drvdata(&cdev->dev, NULL);
+               spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
+       }
+       cdev->handler = NULL;
+ }
+ static int virtio_ccw_offline(struct ccw_device *cdev)
+ {
+       unsigned long flags;
+       struct virtio_ccw_device *vcdev = virtio_grab_drvdata(cdev);
+       if (!vcdev)
+               return 0;
+       if (vcdev->device_lost)
+               virtio_break_device(&vcdev->vdev);
+       unregister_virtio_device(&vcdev->vdev);
+       spin_lock_irqsave(get_ccwdev_lock(cdev), flags);
+       dev_set_drvdata(&cdev->dev, NULL);
+       spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
+       return 0;
+ }
+ static int virtio_ccw_set_transport_rev(struct virtio_ccw_device *vcdev)
+ {
+       struct virtio_rev_info *rev;
+       struct ccw1 *ccw;
+       int ret;
+       ccw = kzalloc(sizeof(*ccw), GFP_DMA | GFP_KERNEL);
+       if (!ccw)
+               return -ENOMEM;
+       rev = kzalloc(sizeof(*rev), GFP_DMA | GFP_KERNEL);
+       if (!rev) {
+               kfree(ccw);
+               return -ENOMEM;
+       }
+       /* Set transport revision */
+       ccw->cmd_code = CCW_CMD_SET_VIRTIO_REV;
+       ccw->flags = 0;
+       ccw->count = sizeof(*rev);
+       ccw->cda = (__u32)(unsigned long)rev;
+       vcdev->revision = VIRTIO_CCW_REV_MAX;
+       do {
+               rev->revision = vcdev->revision;
+               /* none of our supported revisions carry payload */
+               rev->length = 0;
+               ret = ccw_io_helper(vcdev, ccw,
+                                   VIRTIO_CCW_DOING_SET_VIRTIO_REV);
+               if (ret == -EOPNOTSUPP) {
+                       if (vcdev->revision == 0)
+                               /*
+                                * The host device does not support setting
+                                * the revision: let's operate it in legacy
+                                * mode.
+                                */
+                               ret = 0;
+                       else
+                               vcdev->revision--;
+               }
+       } while (ret == -EOPNOTSUPP);
+       kfree(ccw);
+       kfree(rev);
+       return ret;
+ }
+ static int virtio_ccw_online(struct ccw_device *cdev)
+ {
+       int ret;
+       struct virtio_ccw_device *vcdev;
+       unsigned long flags;
+       vcdev = kzalloc(sizeof(*vcdev), GFP_KERNEL);
+       if (!vcdev) {
+               dev_warn(&cdev->dev, "Could not get memory for virtio\n");
+               ret = -ENOMEM;
+               goto out_free;
+       }
+       vcdev->config_block = kzalloc(sizeof(*vcdev->config_block),
+                                  GFP_DMA | GFP_KERNEL);
+       if (!vcdev->config_block) {
+               ret = -ENOMEM;
+               goto out_free;
+       }
+       vcdev->status = kzalloc(sizeof(*vcdev->status), GFP_DMA | GFP_KERNEL);
+       if (!vcdev->status) {
+               ret = -ENOMEM;
+               goto out_free;
+       }
+       vcdev->is_thinint = virtio_ccw_use_airq; /* at least try */
+       vcdev->vdev.dev.parent = &cdev->dev;
+       vcdev->vdev.dev.release = virtio_ccw_release_dev;
+       vcdev->vdev.config = &virtio_ccw_config_ops;
+       vcdev->cdev = cdev;
+       init_waitqueue_head(&vcdev->wait_q);
+       INIT_LIST_HEAD(&vcdev->virtqueues);
+       spin_lock_init(&vcdev->lock);
+       spin_lock_irqsave(get_ccwdev_lock(cdev), flags);
+       dev_set_drvdata(&cdev->dev, vcdev);
+       spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
+       vcdev->vdev.id.vendor = cdev->id.cu_type;
+       vcdev->vdev.id.device = cdev->id.cu_model;
+       ret = virtio_ccw_set_transport_rev(vcdev);
+       if (ret)
+               goto out_free;
+       ret = register_virtio_device(&vcdev->vdev);
+       if (ret) {
+               dev_warn(&cdev->dev, "Failed to register virtio device: %d\n",
+                        ret);
+               goto out_put;
+       }
+       return 0;
+ out_put:
+       spin_lock_irqsave(get_ccwdev_lock(cdev), flags);
+       dev_set_drvdata(&cdev->dev, NULL);
+       spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
+       put_device(&vcdev->vdev.dev);
+       return ret;
+ out_free:
+       if (vcdev) {
+               kfree(vcdev->status);
+               kfree(vcdev->config_block);
+       }
+       kfree(vcdev);
+       return ret;
+ }
+ static int virtio_ccw_cio_notify(struct ccw_device *cdev, int event)
+ {
+       int rc;
+       struct virtio_ccw_device *vcdev = dev_get_drvdata(&cdev->dev);
+       /*
+        * Make sure vcdev is set
+        * i.e. set_offline/remove callback not already running
+        */
+       if (!vcdev)
+               return NOTIFY_DONE;
+       switch (event) {
+       case CIO_GONE:
+               vcdev->device_lost = true;
+               rc = NOTIFY_DONE;
+               break;
+       default:
+               rc = NOTIFY_DONE;
+               break;
+       }
+       return rc;
+ }
+ static struct ccw_device_id virtio_ids[] = {
+       { CCW_DEVICE(0x3832, 0) },
+       {},
+ };
+ MODULE_DEVICE_TABLE(ccw, virtio_ids);
+ static struct ccw_driver virtio_ccw_driver = {
+       .driver = {
+               .owner = THIS_MODULE,
+               .name = "virtio_ccw",
+       },
+       .ids = virtio_ids,
+       .probe = virtio_ccw_probe,
+       .remove = virtio_ccw_remove,
+       .set_offline = virtio_ccw_offline,
+       .set_online = virtio_ccw_online,
+       .notify = virtio_ccw_cio_notify,
+       .int_class = IRQIO_VIR,
+ };
+ static int __init pure_hex(char **cp, unsigned int *val, int min_digit,
+                          int max_digit, int max_val)
+ {
+       int diff;
+       diff = 0;
+       *val = 0;
+       while (diff <= max_digit) {
+               int value = hex_to_bin(**cp);
+               if (value < 0)
+                       break;
+               *val = *val * 16 + value;
+               (*cp)++;
+               diff++;
+       }
+       if ((diff < min_digit) || (diff > max_digit) || (*val > max_val))
+               return 1;
+       return 0;
+ }
+ static int __init parse_busid(char *str, unsigned int *cssid,
+                             unsigned int *ssid, unsigned int *devno)
+ {
+       char *str_work;
+       int rc, ret;
+       rc = 1;
+       if (*str == '\0')
+               goto out;
+       str_work = str;
+       ret = pure_hex(&str_work, cssid, 1, 2, __MAX_CSSID);
+       if (ret || (str_work[0] != '.'))
+               goto out;
+       str_work++;
+       ret = pure_hex(&str_work, ssid, 1, 1, __MAX_SSID);
+       if (ret || (str_work[0] != '.'))
+               goto out;
+       str_work++;
+       ret = pure_hex(&str_work, devno, 4, 4, __MAX_SUBCHANNEL);
+       if (ret || (str_work[0] != '\0'))
+               goto out;
+       rc = 0;
+ out:
+       return rc;
+ }
+ static void __init no_auto_parse(void)
+ {
+       unsigned int from_cssid, to_cssid, from_ssid, to_ssid, from, to;
+       char *parm, *str;
+       int rc;
+       str = no_auto;
+       while ((parm = strsep(&str, ","))) {
+               rc = parse_busid(strsep(&parm, "-"), &from_cssid,
+                                &from_ssid, &from);
+               if (rc)
+                       continue;
+               if (parm != NULL) {
+                       rc = parse_busid(parm, &to_cssid,
+                                        &to_ssid, &to);
+                       if ((from_ssid > to_ssid) ||
+                           ((from_ssid == to_ssid) && (from > to)))
+                               rc = -EINVAL;
+               } else {
+                       to_cssid = from_cssid;
+                       to_ssid = from_ssid;
+                       to = from;
+               }
+               if (rc)
+                       continue;
+               while ((from_ssid < to_ssid) ||
+                      ((from_ssid == to_ssid) && (from <= to))) {
+                       set_bit(from, devs_no_auto[from_ssid]);
+                       from++;
+                       if (from > __MAX_SUBCHANNEL) {
+                               from_ssid++;
+                               from = 0;
+                       }
+               }
+       }
+ }
+ static int __init virtio_ccw_init(void)
+ {
+       /* parse no_auto string before we do anything further */
+       no_auto_parse();
+       return ccw_driver_register(&virtio_ccw_driver);
+ }
+ module_init(virtio_ccw_init);
+ static void __exit virtio_ccw_exit(void)
+ {
+       int i;
+       ccw_driver_unregister(&virtio_ccw_driver);
+       for (i = 0; i < MAX_AIRQ_AREAS; i++)
+               destroy_airq_info(airq_areas[i]);
+ }
+ module_exit(virtio_ccw_exit);
index 285f77544c36391a89b243001c3fc93d0703f46f,55441c7f3e832d4d95945511c331f6adbbfd644d..7dbbb29d24c6cf5290cd06a7a97df570a8860f18
@@@ -1007,8 -1002,9 +1007,10 @@@ static int virtscsi_probe(struct virtio
        shost->max_cmd_len = VIRTIO_SCSI_CDB_SIZE;
        shost->nr_hw_queues = num_queues;
  
 +#ifdef CONFIG_BLK_DEV_INTEGRITY
        if (virtio_has_feature(vdev, VIRTIO_SCSI_F_T10_PI)) {
+               int host_prot;
                host_prot = SHOST_DIF_TYPE1_PROTECTION | SHOST_DIF_TYPE2_PROTECTION |
                            SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE1_PROTECTION |
                            SHOST_DIX_TYPE2_PROTECTION | SHOST_DIX_TYPE3_PROTECTION;