Merge branch 'for-linus' of git://git390.marist.edu/pub/scm/linux-2.6

* 'for-linus' of git://git390.marist.edu/pub/scm/linux-2.6: (65 commits)
  [S390] prevent unneccesary loops_per_jiffy recalculation
  [S390] cpuinfo: use get_online_cpus() instead of preempt_disable()
  [S390] smp: remove cpu hotplug messages
  [S390] mutex: enable spinning mutex on s390
  [S390] mutex: Introduce arch_mutex_cpu_relax()
  [S390] cio: fix ccwgroup unregistration race condition
  [S390] perf: add DWARF register lookup for s390
  [S390] cleanup ftrace backend functions
  [S390] ptrace cleanup
  [S390] smp/idle: call init_idle() before starting a new cpu
  [S390] smp: delay idle task creation
  [S390] dasd: Correct retry counter for terminated I/O.
  [S390] dasd: Add support for raw ECKD access.
  [S390] dasd: Prevent deadlock during suspend/resume.
  [S390] dasd: Improve handling of stolen DASD reservation
  [S390] dasd: do path verification for paths added at runtime
  [S390] dasd: add High Performance FICON multitrack support
  [S390] cio: reduce memory consumption of itcw structures
  [S390] nmi: enable machine checks early
  [S390] qeth: buffer count imbalance
  ...

11 files changed, 300 insertions, 210 deletions

diff --git a/drivers/s390/cio/ccwgroup.c b/drivers/s390/cio/ccwgroup.c
index 97b25d68e3e7..2864581d8ecb 100644
--- a/drivers/s390/cio/ccwgroup.c
+++ b/drivers/s390/cio/ccwgroup.c
@@ -67,6 +67,27 @@ __ccwgroup_remove_symlinks(struct ccwgroup_device *gdev)
 }
 
 /*
+ * Remove references from ccw devices to ccw group device and from
+ * ccw group device to ccw devices.
+ */
+static void __ccwgroup_remove_cdev_refs(struct ccwgroup_device *gdev)
+{
+        struct ccw_device *cdev;
+        int i;
+
+        for (i = 0; i < gdev->count; i++) {
+                cdev = gdev->cdev[i];
+                if (!cdev)
+                        continue;
+                spin_lock_irq(cdev->ccwlock);
+                dev_set_drvdata(&cdev->dev, NULL);
+                spin_unlock_irq(cdev->ccwlock);
+                gdev->cdev[i] = NULL;
+                put_device(&cdev->dev);
+        }
+}
+
+/*
  * Provide an 'ungroup' attribute so the user can remove group devices no
  * longer needed or accidentially created. Saves memory :)
  */
@@ -78,6 +99,7 @@ static void ccwgroup_ungroup_callback(struct device *dev)
         if (device_is_registered(&gdev->dev)) {
                 __ccwgroup_remove_symlinks(gdev);
                 device_unregister(dev);
+                __ccwgroup_remove_cdev_refs(gdev);
         }
         mutex_unlock(&gdev->reg_mutex);
 }
@@ -116,21 +138,7 @@ static DEVICE_ATTR(ungroup, 0200, NULL, ccwgroup_ungroup_store);
 static void
 ccwgroup_release (struct device *dev)
 {
-        struct ccwgroup_device *gdev;
-        int i;
-
-        gdev = to_ccwgroupdev(dev);
-
-        for (i = 0; i < gdev->count; i++) {
-                if (gdev->cdev[i]) {
-                        spin_lock_irq(gdev->cdev[i]->ccwlock);
-                        if (dev_get_drvdata(&gdev->cdev[i]->dev) == gdev)
-                                dev_set_drvdata(&gdev->cdev[i]->dev, NULL);
-                        spin_unlock_irq(gdev->cdev[i]->ccwlock);
-                        put_device(&gdev->cdev[i]->dev);
-                }
-        }
-        kfree(gdev);
+        kfree(to_ccwgroupdev(dev));
 }
 
 static int
@@ -639,6 +647,7 @@ void ccwgroup_driver_unregister(struct ccwgroup_driver *cdriver)
                 mutex_lock(&gdev->reg_mutex);
                 __ccwgroup_remove_symlinks(gdev);
                 device_unregister(dev);
+                __ccwgroup_remove_cdev_refs(gdev);
                 mutex_unlock(&gdev->reg_mutex);
                 put_device(dev);
         }
@@ -660,25 +669,6 @@ int ccwgroup_probe_ccwdev(struct ccw_device *cdev)
         return 0;
 }
 
-static struct ccwgroup_device *
-__ccwgroup_get_gdev_by_cdev(struct ccw_device *cdev)
-{
-        struct ccwgroup_device *gdev;
-
-        gdev = dev_get_drvdata(&cdev->dev);
-        if (gdev) {
-                if (get_device(&gdev->dev)) {
-                        mutex_lock(&gdev->reg_mutex);
-                        if (device_is_registered(&gdev->dev))
-                                return gdev;
-                        mutex_unlock(&gdev->reg_mutex);
-                        put_device(&gdev->dev);
-                }
-                return NULL;
-        }
-        return NULL;
-}
-
 /**
  * ccwgroup_remove_ccwdev() - remove function for slave devices
  * @cdev: ccw device to be removed
@@ -694,13 +684,25 @@ void ccwgroup_remove_ccwdev(struct ccw_device *cdev)
         /* Ignore offlining errors, device is gone anyway. */
         ccw_device_set_offline(cdev);
         /* If one of its devices is gone, the whole group is done for. */
-        gdev = __ccwgroup_get_gdev_by_cdev(cdev);
-        if (gdev) {
+        spin_lock_irq(cdev->ccwlock);
+        gdev = dev_get_drvdata(&cdev->dev);
+        if (!gdev) {
+                spin_unlock_irq(cdev->ccwlock);
+                return;
+        }
+        /* Get ccwgroup device reference for local processing. */
+        get_device(&gdev->dev);
+        spin_unlock_irq(cdev->ccwlock);
+        /* Unregister group device. */
+        mutex_lock(&gdev->reg_mutex);
+        if (device_is_registered(&gdev->dev)) {
                 __ccwgroup_remove_symlinks(gdev);
                 device_unregister(&gdev->dev);
-                mutex_unlock(&gdev->reg_mutex);
-                put_device(&gdev->dev);
+                __ccwgroup_remove_cdev_refs(gdev);
         }
+        mutex_unlock(&gdev->reg_mutex);
+        /* Release ccwgroup device reference for local processing. */
+        put_device(&gdev->dev);
 }
 
 MODULE_LICENSE("GPL");
diff --git a/drivers/s390/cio/chsc.c b/drivers/s390/cio/chsc.c
index 1aaddea673e0..0689fcf23a11 100644
--- a/drivers/s390/cio/chsc.c
+++ b/drivers/s390/cio/chsc.c
@@ -695,6 +695,25 @@ out:
         return ret;
 }
 
+int chsc_determine_fmt1_channel_path_desc(struct chp_id chpid,
+                                          struct channel_path_desc_fmt1 *desc)
+{
+        struct chsc_response_struct *chsc_resp;
+        struct chsc_scpd *scpd_area;
+        int ret;
+
+        spin_lock_irq(&chsc_page_lock);
+        scpd_area = chsc_page;
+        ret = chsc_determine_channel_path_desc(chpid, 0, 0, 1, 0, scpd_area);
+        if (ret)
+                goto out;
+        chsc_resp = (void *)&scpd_area->response;
+        memcpy(desc, &chsc_resp->data, sizeof(*desc));
+out:
+        spin_unlock_irq(&chsc_page_lock);
+        return ret;
+}
+
 static void
 chsc_initialize_cmg_chars(struct channel_path *chp, u8 cmcv,
                           struct cmg_chars *chars)
diff --git a/drivers/s390/cio/chsc.h b/drivers/s390/cio/chsc.h
index 6693f5e3176f..3f15b2aaeaea 100644
--- a/drivers/s390/cio/chsc.h
+++ b/drivers/s390/cio/chsc.h
@@ -35,6 +35,22 @@ struct channel_path_desc {
         u8 chpp;
 } __attribute__ ((packed));
 
+struct channel_path_desc_fmt1 {
+        u8 flags;
+        u8 lsn;
+        u8 desc;
+        u8 chpid;
+        u32:24;
+        u8 chpp;
+        u32 unused[3];
+        u16 mdc;
+        u16:13;
+        u8 r:1;
+        u8 s:1;
+        u8 f:1;
+        u32 zeros[2];
+} __attribute__ ((packed));
+
 struct channel_path;
 
 struct css_chsc_char {
@@ -92,6 +108,8 @@ int chsc_determine_channel_path_desc(struct chp_id chpid, int fmt, int rfmt,
                                      int c, int m, void *page);
 int chsc_determine_base_channel_path_desc(struct chp_id chpid,
                                           struct channel_path_desc *desc);
+int chsc_determine_fmt1_channel_path_desc(struct chp_id chpid,
+                                          struct channel_path_desc_fmt1 *desc);
 void chsc_chp_online(struct chp_id chpid);
 void chsc_chp_offline(struct chp_id chpid);
 int chsc_get_channel_measurement_chars(struct channel_path *chp);
diff --git a/drivers/s390/cio/css.c b/drivers/s390/cio/css.c
index 825951b6b83f..24d8e97355b9 100644
--- a/drivers/s390/cio/css.c
+++ b/drivers/s390/cio/css.c
@@ -618,6 +618,7 @@ EXPORT_SYMBOL_GPL(css_schedule_reprobe);
 static void css_process_crw(struct crw *crw0, struct crw *crw1, int overflow)
 {
         struct subchannel_id mchk_schid;
+        struct subchannel *sch;
 
         if (overflow) {
                 css_schedule_eval_all();
@@ -637,6 +638,13 @@ static void css_process_crw(struct crw *crw0, struct crw *crw1, int overflow)
         if (crw1)
                 mchk_schid.ssid = (crw1->rsid >> 4) & 3;
 
+        if (crw0->erc == CRW_ERC_PMOD) {
+                sch = get_subchannel_by_schid(mchk_schid);
+                if (sch) {
+                        css_update_ssd_info(sch);
+                        put_device(&sch->dev);
+                }
+        }
         /*
          * Since we are always presented with IPI in the CRW, we have to
          * use stsch() to find out if the subchannel in question has come
diff --git a/drivers/s390/cio/device_ops.c b/drivers/s390/cio/device_ops.c
index 6da84543dfe9..651976b54af8 100644
--- a/drivers/s390/cio/device_ops.c
+++ b/drivers/s390/cio/device_ops.c
@@ -687,6 +687,46 @@ int ccw_device_tm_start_timeout(struct ccw_device *cdev, struct tcw *tcw,
 EXPORT_SYMBOL(ccw_device_tm_start_timeout);
 
 /**
+ * ccw_device_get_mdc - accumulate max data count
+ * @cdev: ccw device for which the max data count is accumulated
+ * @mask: mask of paths to use
+ *
+ * Return the number of 64K-bytes blocks all paths at least support
+ * for a transport command. Return values <= 0 indicate failures.
+ */
+int ccw_device_get_mdc(struct ccw_device *cdev, u8 mask)
+{
+        struct subchannel *sch = to_subchannel(cdev->dev.parent);
+        struct channel_path_desc_fmt1 desc;
+        struct chp_id chpid;
+        int mdc = 0, ret, i;
+
+        /* Adjust requested path mask to excluded varied off paths. */
+        if (mask)
+                mask &= sch->lpm;
+        else
+                mask = sch->lpm;
+
+        chp_id_init(&chpid);
+        for (i = 0; i < 8; i++) {
+                if (!(mask & (0x80 >> i)))
+                        continue;
+                chpid.id = sch->schib.pmcw.chpid[i];
+                ret = chsc_determine_fmt1_channel_path_desc(chpid, &desc);
+                if (ret)
+                        return ret;
+                if (!desc.f)
+                        return 0;
+                if (!desc.r)
+                        mdc = 1;
+                mdc = mdc ? min(mdc, (int)desc.mdc) : desc.mdc;
+        }
+
+        return mdc;
+}
+EXPORT_SYMBOL(ccw_device_get_mdc);
+
+/**
  * ccw_device_tm_intrg - perform interrogate function
  * @cdev: ccw device on which to perform the interrogate function
  *
diff --git a/drivers/s390/cio/itcw.c b/drivers/s390/cio/itcw.c
index a0ae29564774..358ee16d10a2 100644
--- a/drivers/s390/cio/itcw.c
+++ b/drivers/s390/cio/itcw.c
@@ -93,6 +93,7 @@ EXPORT_SYMBOL(itcw_get_tcw);
 size_t itcw_calc_size(int intrg, int max_tidaws, int intrg_max_tidaws)
 {
         size_t len;
+        int cross_count;
 
         /* Main data. */
         len = sizeof(struct itcw);
@@ -105,12 +106,27 @@ size_t itcw_calc_size(int intrg, int max_tidaws, int intrg_max_tidaws)
                        /* TSB */ sizeof(struct tsb) +
                        /* TIDAL */ intrg_max_tidaws * sizeof(struct tidaw);
         }
+
         /* Maximum required alignment padding. */
         len += /* Initial TCW */ 63 + /* Interrogate TCCB */ 7;
-        /* Maximum padding for structures that may not cross 4k boundary. */
-        if ((max_tidaws > 0) || (intrg_max_tidaws > 0))
-                len += max(max_tidaws, intrg_max_tidaws) *
-                       sizeof(struct tidaw) - 1;
+
+        /* TIDAW lists may not cross a 4k boundary. To cross a
+         * boundary we need to add a TTIC TIDAW. We need to reserve
+         * one additional TIDAW for a TTIC that we may need to add due
+         * to the placement of the data chunk in memory, and a further
+         * TIDAW for each page boundary that the TIDAW list may cross
+         * due to it's own size.
+         */
+        if (max_tidaws) {
+                cross_count = 1 + ((max_tidaws * sizeof(struct tidaw) - 1)
+                                   >> PAGE_SHIFT);
+                len += cross_count * sizeof(struct tidaw);
+        }
+        if (intrg_max_tidaws) {
+                cross_count = 1 + ((intrg_max_tidaws * sizeof(struct tidaw) - 1)
+                                   >> PAGE_SHIFT);
+                len += cross_count * sizeof(struct tidaw);
+        }
         return len;
 }
 EXPORT_SYMBOL(itcw_calc_size);
@@ -165,6 +181,7 @@ struct itcw *itcw_init(void *buffer, size_t size, int op, int intrg,
         void *chunk;
         addr_t start;
         addr_t end;
+        int cross_count;
 
         /* Check for 2G limit. */
         start = (addr_t) buffer;
@@ -177,8 +194,17 @@ struct itcw *itcw_init(void *buffer, size_t size, int op, int intrg,
         if (IS_ERR(chunk))
                 return chunk;
         itcw = chunk;
-        itcw->max_tidaws = max_tidaws;
-        itcw->intrg_max_tidaws = intrg_max_tidaws;
+        /* allow for TTIC tidaws that may be needed to cross a page boundary */
+        cross_count = 0;
+        if (max_tidaws)
+                cross_count = 1 + ((max_tidaws * sizeof(struct tidaw) - 1)
+                                   >> PAGE_SHIFT);
+        itcw->max_tidaws = max_tidaws + cross_count;
+        cross_count = 0;
+        if (intrg_max_tidaws)
+                cross_count = 1 + ((intrg_max_tidaws * sizeof(struct tidaw) - 1)
+                                   >> PAGE_SHIFT);
+        itcw->intrg_max_tidaws = intrg_max_tidaws + cross_count;
         /* Main TCW. */
         chunk = fit_chunk(&start, end, sizeof(struct tcw), 64, 0);
         if (IS_ERR(chunk))
@@ -198,7 +224,7 @@ struct itcw *itcw_init(void *buffer, size_t size, int op, int intrg,
         /* Data TIDAL. */
         if (max_tidaws > 0) {
                 chunk = fit_chunk(&start, end, sizeof(struct tidaw) *
-                                  max_tidaws, 16, 1);
+                                  itcw->max_tidaws, 16, 0);
                 if (IS_ERR(chunk))
                         return chunk;
                 tcw_set_data(itcw->tcw, chunk, 1);
@@ -206,7 +232,7 @@ struct itcw *itcw_init(void *buffer, size_t size, int op, int intrg,
         /* Interrogate data TIDAL. */
         if (intrg && (intrg_max_tidaws > 0)) {
                 chunk = fit_chunk(&start, end, sizeof(struct tidaw) *
-                                  intrg_max_tidaws, 16, 1);
+                                  itcw->intrg_max_tidaws, 16, 0);
                 if (IS_ERR(chunk))
                         return chunk;
                 tcw_set_data(itcw->intrg_tcw, chunk, 1);
@@ -283,13 +309,29 @@ EXPORT_SYMBOL(itcw_add_dcw);
  * the new tidaw on success or -%ENOSPC if the new tidaw would exceed the
  * available space.
  *
- * Note: the tidaw-list is assumed to be contiguous with no ttics. The
- * last-tidaw flag for the last tidaw in the list will be set by itcw_finalize.
+ * Note: TTIC tidaws are automatically added when needed, so explicitly calling
+ * this interface with the TTIC flag is not supported. The last-tidaw flag
+ * for the last tidaw in the list will be set by itcw_finalize.
  */
 struct tidaw *itcw_add_tidaw(struct itcw *itcw, u8 flags, void *addr, u32 count)
 {
+        struct tidaw *following;
+
         if (itcw->num_tidaws >= itcw->max_tidaws)
                 return ERR_PTR(-ENOSPC);
+        /*
+         * Is the tidaw, which follows the one we are about to fill, on the next
+         * page? Then we have to insert a TTIC tidaw first, that points to the
+         * tidaw on the new page.
+         */
+        following = ((struct tidaw *) tcw_get_data(itcw->tcw))
+                + itcw->num_tidaws + 1;
+        if (itcw->num_tidaws && !((unsigned long) following & ~PAGE_MASK)) {
+                tcw_add_tidaw(itcw->tcw, itcw->num_tidaws++,
+                              TIDAW_FLAGS_TTIC, following, 0);
+                if (itcw->num_tidaws >= itcw->max_tidaws)
+                        return ERR_PTR(-ENOSPC);
+        }
         return tcw_add_tidaw(itcw->tcw, itcw->num_tidaws++, flags, addr, count);
 }
 EXPORT_SYMBOL(itcw_add_tidaw);
diff --git a/drivers/s390/cio/qdio.h b/drivers/s390/cio/qdio.h
index 0f4ef8769a3d..7bc643f3f5ab 100644
--- a/drivers/s390/cio/qdio.h
+++ b/drivers/s390/cio/qdio.h
@@ -91,6 +91,12 @@ enum qdio_irq_states {
 #define AC1_SC_QEBSM_AVAILABLE          0x02    /* available for subchannel */
 #define AC1_SC_QEBSM_ENABLED            0x01    /* enabled for subchannel */
 
+/* SIGA flags */
+#define QDIO_SIGA_WRITE         0x00
+#define QDIO_SIGA_READ          0x01
+#define QDIO_SIGA_SYNC          0x02
+#define QDIO_SIGA_QEBSM_FLAG    0x80
+
 #ifdef CONFIG_64BIT
 static inline int do_sqbs(u64 token, unsigned char state, int queue,
                           int *start, int *count)
@@ -142,10 +148,9 @@ struct siga_flag {
         u8 input:1;
         u8 output:1;
         u8 sync:1;
-        u8 no_sync_ti:1;
-        u8 no_sync_out_ti:1;
-        u8 no_sync_out_pci:1;
-        u8:2;
+        u8 sync_after_ai:1;
+        u8 sync_out_after_pci:1;
+        u8:3;
 } __attribute__ ((packed));
 
 struct chsc_ssqd_area {
@@ -202,6 +207,7 @@ struct qdio_dev_perf_stat {
         unsigned int inbound_queue_full;
         unsigned int outbound_call;
         unsigned int outbound_handler;
+        unsigned int outbound_queue_full;
         unsigned int fast_requeue;
         unsigned int target_full;
         unsigned int eqbs;
@@ -245,10 +251,10 @@ struct qdio_input_q {
 struct qdio_output_q {
         /* PCIs are enabled for the queue */
         int pci_out_enabled;
-        /* IQDIO: output multiple buffers (enhanced SIGA) */
-        int use_enh_siga;
         /* timer to check for more outbound work */
         struct timer_list timer;
+        /* used SBALs before tasklet schedule */
+        int scan_threshold;
 };
 
 /*
@@ -383,12 +389,13 @@ static inline int multicast_outbound(struct qdio_q *q)
         (q->irq_ptr->qib.ac & QIB_AC_OUTBOUND_PCI_SUPPORTED)
 #define is_qebsm(q)                     (q->irq_ptr->sch_token != 0)
 
-#define need_siga_sync_thinint(q)       (!q->irq_ptr->siga_flag.no_sync_ti)
-#define need_siga_sync_out_thinint(q)   (!q->irq_ptr->siga_flag.no_sync_out_ti)
 #define need_siga_in(q)                 (q->irq_ptr->siga_flag.input)
 #define need_siga_out(q)                (q->irq_ptr->siga_flag.output)
-#define need_siga_sync(q)               (q->irq_ptr->siga_flag.sync)
-#define siga_syncs_out_pci(q)           (q->irq_ptr->siga_flag.no_sync_out_pci)
+#define need_siga_sync(q)               (unlikely(q->irq_ptr->siga_flag.sync))
+#define need_siga_sync_after_ai(q)      \
+        (unlikely(q->irq_ptr->siga_flag.sync_after_ai))
+#define need_siga_sync_out_after_pci(q) \
+        (unlikely(q->irq_ptr->siga_flag.sync_out_after_pci))
 
 #define for_each_input_queue(irq_ptr, q, i)     \
         for (i = 0, q = irq_ptr->input_qs[0];   \
@@ -423,9 +430,9 @@ struct indicator_t {
 
 extern struct indicator_t *q_indicators;
 
-static inline int shared_ind(struct qdio_irq *irq_ptr)
+static inline int shared_ind(u32 *dsci)
 {
-        return irq_ptr->dsci == &q_indicators[TIQDIO_SHARED_IND].ind;
+        return dsci == &q_indicators[TIQDIO_SHARED_IND].ind;
 }
 
 /* prototypes for thin interrupt */
diff --git a/drivers/s390/cio/qdio_debug.c b/drivers/s390/cio/qdio_debug.c
index 28868e7471a5..f8b03a636e49 100644
--- a/drivers/s390/cio/qdio_debug.c
+++ b/drivers/s390/cio/qdio_debug.c
@@ -151,6 +151,7 @@ static char *qperf_names[] = {
         "Inbound queue full",
         "Outbound calls",
         "Outbound handler",
+        "Outbound queue full",
         "Outbound fast_requeue",
         "Outbound target_full",
         "QEBSM eqbs",
diff --git a/drivers/s390/cio/qdio_main.c b/drivers/s390/cio/qdio_main.c
index 5fcfa7f9e9ef..e9fff2b9bce2 100644
--- a/drivers/s390/cio/qdio_main.c
+++ b/drivers/s390/cio/qdio_main.c
@@ -14,6 +14,7 @@
 #include <linux/timer.h>
 #include <linux/delay.h>
 #include <linux/gfp.h>
+#include <linux/kernel_stat.h>
 #include <asm/atomic.h>
 #include <asm/debug.h>
 #include <asm/qdio.h>
@@ -29,11 +30,12 @@ MODULE_AUTHOR("Utz Bacher <utz.bacher@de.ibm.com>,"\
 MODULE_DESCRIPTION("QDIO base support");
 MODULE_LICENSE("GPL");
 
-static inline int do_siga_sync(struct subchannel_id schid,
-                               unsigned int out_mask, unsigned int in_mask)
+static inline int do_siga_sync(unsigned long schid,
+                               unsigned int out_mask, unsigned int in_mask,
+                               unsigned int fc)
 {
-        register unsigned long __fc asm ("0") = 2;
-        register struct subchannel_id __schid asm ("1") = schid;
+        register unsigned long __fc asm ("0") = fc;
+        register unsigned long __schid asm ("1") = schid;
         register unsigned long out asm ("2") = out_mask;
         register unsigned long in asm ("3") = in_mask;
         int cc;
@@ -47,10 +49,11 @@ static inline int do_siga_sync(struct subchannel_id schid,
         return cc;
 }
 
-static inline int do_siga_input(struct subchannel_id schid, unsigned int mask)
+static inline int do_siga_input(unsigned long schid, unsigned int mask,
+                                unsigned int fc)
 {
-        register unsigned long __fc asm ("0") = 1;
-        register struct subchannel_id __schid asm ("1") = schid;
+        register unsigned long __fc asm ("0") = fc;
+        register unsigned long __schid asm ("1") = schid;
         register unsigned long __mask asm ("2") = mask;
         int cc;
 
@@ -279,16 +282,20 @@ void qdio_init_buf_states(struct qdio_irq *irq_ptr)
 static inline int qdio_siga_sync(struct qdio_q *q, unsigned int output,
                           unsigned int input)
 {
+        unsigned long schid = *((u32 *) &q->irq_ptr->schid);
+        unsigned int fc = QDIO_SIGA_SYNC;
         int cc;
 
-        if (!need_siga_sync(q))
-                return 0;
-
         DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-s:%1d", q->nr);
         qperf_inc(q, siga_sync);
 
-        cc = do_siga_sync(q->irq_ptr->schid, output, input);
-        if (cc)
+        if (is_qebsm(q)) {
+                schid = q->irq_ptr->sch_token;
+                fc |= QDIO_SIGA_QEBSM_FLAG;
+        }
+
+        cc = do_siga_sync(schid, output, input, fc);
+        if (unlikely(cc))
                 DBF_ERROR("%4x SIGA-S:%2d", SCH_NO(q), cc);
         return cc;
 }
@@ -301,38 +308,22 @@ static inline int qdio_siga_sync_q(struct qdio_q *q)
                 return qdio_siga_sync(q, q->mask, 0);
 }
 
-static inline int qdio_siga_sync_out(struct qdio_q *q)
-{
-        return qdio_siga_sync(q, ~0U, 0);
-}
-
-static inline int qdio_siga_sync_all(struct qdio_q *q)
-{
-        return qdio_siga_sync(q, ~0U, ~0U);
-}
-
 static int qdio_siga_output(struct qdio_q *q, unsigned int *busy_bit)
 {
-        unsigned long schid;
-        unsigned int fc = 0;
+        unsigned long schid = *((u32 *) &q->irq_ptr->schid);
+        unsigned int fc = QDIO_SIGA_WRITE;
         u64 start_time = 0;
         int cc;
 
-        if (q->u.out.use_enh_siga)
-                fc = 3;
-
         if (is_qebsm(q)) {
                 schid = q->irq_ptr->sch_token;
-                fc |= 0x80;
+                fc |= QDIO_SIGA_QEBSM_FLAG;
         }
-        else
-                schid = *((u32 *)&q->irq_ptr->schid);
-
 again:
         cc = do_siga_output(schid, q->mask, busy_bit, fc);
 
         /* hipersocket busy condition */
-        if (*busy_bit) {
+        if (unlikely(*busy_bit)) {
                 WARN_ON(queue_type(q) != QDIO_IQDIO_QFMT || cc != 2);
 
                 if (!start_time) {
@@ -347,32 +338,41 @@ again:
 
 static inline int qdio_siga_input(struct qdio_q *q)
 {
+        unsigned long schid = *((u32 *) &q->irq_ptr->schid);
+        unsigned int fc = QDIO_SIGA_READ;
         int cc;
 
         DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-r:%1d", q->nr);
         qperf_inc(q, siga_read);
 
-        cc = do_siga_input(q->irq_ptr->schid, q->mask);
-        if (cc)
+        if (is_qebsm(q)) {
+                schid = q->irq_ptr->sch_token;
+                fc |= QDIO_SIGA_QEBSM_FLAG;
+        }
+
+        cc = do_siga_input(schid, q->mask, fc);
+        if (unlikely(cc))
                 DBF_ERROR("%4x SIGA-R:%2d", SCH_NO(q), cc);
         return cc;
 }
 
-static inline void qdio_sync_after_thinint(struct qdio_q *q)
+#define qdio_siga_sync_out(q) qdio_siga_sync(q, ~0U, 0)
+#define qdio_siga_sync_all(q) qdio_siga_sync(q, ~0U, ~0U)
+
+static inline void qdio_sync_queues(struct qdio_q *q)
 {
-        if (pci_out_supported(q)) {
-                if (need_siga_sync_thinint(q))
-                        qdio_siga_sync_all(q);
-                else if (need_siga_sync_out_thinint(q))
-                        qdio_siga_sync_out(q);
-        } else
+        /* PCI capable outbound queues will also be scanned so sync them too */
+        if (pci_out_supported(q))
+                qdio_siga_sync_all(q);
+        else
                 qdio_siga_sync_q(q);
 }
 
 int debug_get_buf_state(struct qdio_q *q, unsigned int bufnr,
                         unsigned char *state)
 {
-        qdio_siga_sync_q(q);
+        if (need_siga_sync(q))
+                qdio_siga_sync_q(q);
         return get_buf_states(q, bufnr, state, 1, 0);
 }
 
@@ -549,7 +549,8 @@ static inline int qdio_inbound_q_done(struct qdio_q *q)
         if (!atomic_read(&q->nr_buf_used))
                 return 1;
 
-        qdio_siga_sync_q(q);
+        if (need_siga_sync(q))
+                qdio_siga_sync_q(q);
         get_buf_state(q, q->first_to_check, &state, 0);
 
         if (state == SLSB_P_INPUT_PRIMED || state == SLSB_P_INPUT_ERROR)
@@ -644,9 +645,12 @@ static int get_outbound_buffer_frontier(struct qdio_q *q)
         int count, stop;
         unsigned char state;
 
-        if (((queue_type(q) != QDIO_IQDIO_QFMT) && !pci_out_supported(q)) ||
-            (queue_type(q) == QDIO_IQDIO_QFMT && multicast_outbound(q)))
-                qdio_siga_sync_q(q);
+        if (need_siga_sync(q))
+                if (((queue_type(q) != QDIO_IQDIO_QFMT) &&
+                    !pci_out_supported(q)) ||
+                    (queue_type(q) == QDIO_IQDIO_QFMT &&
+                    multicast_outbound(q)))
+                        qdio_siga_sync_q(q);
 
         /*
          * Don't check 128 buffers, as otherwise qdio_inbound_q_moved
@@ -818,7 +822,8 @@ static inline void qdio_check_outbound_after_thinint(struct qdio_q *q)
 static void __tiqdio_inbound_processing(struct qdio_q *q)
 {
         qperf_inc(q, tasklet_inbound);
-        qdio_sync_after_thinint(q);
+        if (need_siga_sync(q) && need_siga_sync_after_ai(q))
+                qdio_sync_queues(q);
 
         /*
          * The interrupt could be caused by a PCI request. Check the
@@ -898,16 +903,14 @@ static void qdio_int_handler_pci(struct qdio_irq *irq_ptr)
                         tasklet_schedule(&q->tasklet);
         }
 
-        if (!(irq_ptr->qib.ac & QIB_AC_OUTBOUND_PCI_SUPPORTED))
+        if (!pci_out_supported(q))
                 return;
 
         for_each_output_queue(irq_ptr, q, i) {
                 if (qdio_outbound_q_done(q))
                         continue;
-
-                if (!siga_syncs_out_pci(q))
+                if (need_siga_sync(q) && need_siga_sync_out_after_pci(q))
                         qdio_siga_sync_q(q);
-
                 tasklet_schedule(&q->tasklet);
         }
 }
@@ -970,6 +973,7 @@ void qdio_int_handler(struct ccw_device *cdev, unsigned long intparm,
                 return;
         }
 
+        kstat_cpu(smp_processor_id()).irqs[IOINT_QDI]++;
         if (irq_ptr->perf_stat_enabled)
                 irq_ptr->perf_stat.qdio_int++;
 
@@ -1273,7 +1277,6 @@ int qdio_establish(struct qdio_initialize *init_data)
         }
 
         qdio_setup_ssqd_info(irq_ptr);
-        DBF_EVENT("qDmmwc:%2x", irq_ptr->ssqd_desc.mmwc);
         DBF_EVENT("qib ac:%4x", irq_ptr->qib.ac);
 
         /* qebsm is now setup if available, initialize buffer states */
@@ -1445,52 +1448,38 @@ static int handle_outbound(struct qdio_q *q, unsigned int callflags,
         used = atomic_add_return(count, &q->nr_buf_used);
         BUG_ON(used > QDIO_MAX_BUFFERS_PER_Q);
 
+        if (used == QDIO_MAX_BUFFERS_PER_Q)
+                qperf_inc(q, outbound_queue_full);
+
         if (callflags & QDIO_FLAG_PCI_OUT) {
                 q->u.out.pci_out_enabled = 1;
                 qperf_inc(q, pci_request_int);
-        }
-        else
+        } else
                 q->u.out.pci_out_enabled = 0;
 
         if (queue_type(q) == QDIO_IQDIO_QFMT) {
-                if (multicast_outbound(q))
+                /* One SIGA-W per buffer required for unicast HiperSockets. */
+                WARN_ON_ONCE(count > 1 && !multicast_outbound(q));
+
+                rc = qdio_kick_outbound_q(q);
+        } else if (need_siga_sync(q)) {
+                rc = qdio_siga_sync_q(q);
+        } else {
+                /* try to fast requeue buffers */
+                get_buf_state(q, prev_buf(bufnr), &state, 0);
+                if (state != SLSB_CU_OUTPUT_PRIMED)
                         rc = qdio_kick_outbound_q(q);
                 else
-                        if ((q->irq_ptr->ssqd_desc.mmwc > 1) &&
-                            (count > 1) &&
-                            (count <= q->irq_ptr->ssqd_desc.mmwc)) {
-                                /* exploit enhanced SIGA */
-                                q->u.out.use_enh_siga = 1;
-                                rc = qdio_kick_outbound_q(q);
-                        } else {
-                                /*
-                                * One siga-w per buffer required for unicast
-                                * HiperSockets.
-                                */
-                                q->u.out.use_enh_siga = 0;
-                                while (count--) {
-                                        rc = qdio_kick_outbound_q(q);
-                                        if (rc)
-                                                goto out;
-                                }
-                        }
-                goto out;
-        }
-
-        if (need_siga_sync(q)) {
-                qdio_siga_sync_q(q);
-                goto out;
+                        qperf_inc(q, fast_requeue);
         }
 
-        /* try to fast requeue buffers */
-        get_buf_state(q, prev_buf(bufnr), &state, 0);
-        if (state != SLSB_CU_OUTPUT_PRIMED)
-                rc = qdio_kick_outbound_q(q);
+        /* in case of SIGA errors we must process the error immediately */
+        if (used >= q->u.out.scan_threshold || rc)
+                tasklet_schedule(&q->tasklet);
         else
-                qperf_inc(q, fast_requeue);
-
-out:
-        tasklet_schedule(&q->tasklet);
+                /* free the SBALs in case of no further traffic */
+                if (!timer_pending(&q->u.out.timer))
+                        mod_timer(&q->u.out.timer, jiffies + HZ);
         return rc;
 }
 
@@ -1550,7 +1539,7 @@ int qdio_start_irq(struct ccw_device *cdev, int nr)
 
         WARN_ON(queue_irqs_enabled(q));
 
-        if (!shared_ind(q->irq_ptr))
+        if (!shared_ind(q->irq_ptr->dsci))
                 xchg(q->irq_ptr->dsci, 0);
 
         qdio_stop_polling(q);
@@ -1560,7 +1549,7 @@ int qdio_start_irq(struct ccw_device *cdev, int nr)
          * We need to check again to not lose initiative after
          * resetting the ACK state.
          */
-        if (!shared_ind(q->irq_ptr) && *q->irq_ptr->dsci)
+        if (!shared_ind(q->irq_ptr->dsci) && *q->irq_ptr->dsci)
                 goto rescan;
         if (!qdio_inbound_q_done(q))
                 goto rescan;
@@ -1600,12 +1589,14 @@ int qdio_get_next_buffers(struct ccw_device *cdev, int nr, int *bufnr,
         q = irq_ptr->input_qs[nr];
         WARN_ON(queue_irqs_enabled(q));
 
-        qdio_sync_after_thinint(q);
-
         /*
-         * The interrupt could be caused by a PCI request. Check the
-         * PCI capable outbound queues.
+         * Cannot rely on automatic sync after interrupt since queues may
+         * also be examined without interrupt.
          */
+        if (need_siga_sync(q))
+                qdio_sync_queues(q);
+
+        /* check the PCI capable outbound queues. */
         qdio_check_outbound_after_thinint(q);
 
         if (!qdio_inbound_q_moved(q))
diff --git a/drivers/s390/cio/qdio_setup.c b/drivers/s390/cio/qdio_setup.c
index a13cf7ec64b2..89107d0938c4 100644
--- a/drivers/s390/cio/qdio_setup.c
+++ b/drivers/s390/cio/qdio_setup.c
@@ -178,6 +178,7 @@ static void setup_queues(struct qdio_irq *irq_ptr,
                 setup_queues_misc(q, irq_ptr, qdio_init->output_handler, i);
 
                 q->is_input_q = 0;
+                q->u.out.scan_threshold = qdio_init->scan_threshold;
                 setup_storage_lists(q, irq_ptr, output_sbal_array, i);
                 output_sbal_array += QDIO_MAX_BUFFERS_PER_Q;
 
@@ -196,14 +197,10 @@ static void process_ac_flags(struct qdio_irq *irq_ptr, unsigned char qdioac)
                 irq_ptr->siga_flag.output = 1;
         if (qdioac & AC1_SIGA_SYNC_NEEDED)
                 irq_ptr->siga_flag.sync = 1;
-        if (qdioac & AC1_AUTOMATIC_SYNC_ON_THININT)
-                irq_ptr->siga_flag.no_sync_ti = 1;
-        if (qdioac & AC1_AUTOMATIC_SYNC_ON_OUT_PCI)
-                irq_ptr->siga_flag.no_sync_out_pci = 1;
-
-        if (irq_ptr->siga_flag.no_sync_out_pci &&
-            irq_ptr->siga_flag.no_sync_ti)
-                irq_ptr->siga_flag.no_sync_out_ti = 1;
+        if (!(qdioac & AC1_AUTOMATIC_SYNC_ON_THININT))
+                irq_ptr->siga_flag.sync_after_ai = 1;
+        if (!(qdioac & AC1_AUTOMATIC_SYNC_ON_OUT_PCI))
+                irq_ptr->siga_flag.sync_out_after_pci = 1;
 }
 
 static void check_and_setup_qebsm(struct qdio_irq *irq_ptr,
@@ -451,7 +448,7 @@ void qdio_print_subchannel_info(struct qdio_irq *irq_ptr,
         char s[80];
 
         snprintf(s, 80, "qdio: %s %s on SC %x using "
-                 "AI:%d QEBSM:%d PCI:%d TDD:%d SIGA:%s%s%s%s%s%s\n",
+                 "AI:%d QEBSM:%d PCI:%d TDD:%d SIGA:%s%s%s%s%s\n",
                  dev_name(&cdev->dev),
                  (irq_ptr->qib.qfmt == QDIO_QETH_QFMT) ? "OSA" :
                         ((irq_ptr->qib.qfmt == QDIO_ZFCP_QFMT) ? "ZFCP" : "HS"),
@@ -463,9 +460,8 @@ void qdio_print_subchannel_info(struct qdio_irq *irq_ptr,
                  (irq_ptr->siga_flag.input) ? "R" : " ",
                  (irq_ptr->siga_flag.output) ? "W" : " ",
                  (irq_ptr->siga_flag.sync) ? "S" : " ",
-                 (!irq_ptr->siga_flag.no_sync_ti) ? "A" : " ",
-                 (!irq_ptr->siga_flag.no_sync_out_ti) ? "O" : " ",
-                 (!irq_ptr->siga_flag.no_sync_out_pci) ? "P" : " ");
+                 (irq_ptr->siga_flag.sync_after_ai) ? "A" : " ",
+                 (irq_ptr->siga_flag.sync_out_after_pci) ? "P" : " ");
         printk(KERN_INFO "%s", s);
 }
 
diff --git a/drivers/s390/cio/qdio_thinint.c b/drivers/s390/cio/qdio_thinint.c
index 5d9c66627b6e..5c4e741d8221 100644
--- a/drivers/s390/cio/qdio_thinint.c
+++ b/drivers/s390/cio/qdio_thinint.c
@@ -8,6 +8,7 @@
  */
 #include <linux/io.h>
 #include <linux/slab.h>
+#include <linux/kernel_stat.h>
 #include <asm/atomic.h>
 #include <asm/debug.h>
 #include <asm/qdio.h>
@@ -35,22 +36,8 @@ static u8 *tiqdio_alsi;
 
 struct indicator_t *q_indicators;
 
-static int css_qdio_omit_svs;
-
 static u64 last_ai_time;
 
-static inline unsigned long do_clear_global_summary(void)
-{
-        register unsigned long __fn asm("1") = 3;
-        register unsigned long __tmp asm("2");
-        register unsigned long __time asm("3");
-
-        asm volatile(
-                "       .insn   rre,0xb2650000,2,0"
-                : "+d" (__fn), "=d" (__tmp), "=d" (__time));
-        return __time;
-}
-
 /* returns addr for the device state change indicator */
 static u32 *get_indicator(void)
 {
@@ -83,10 +70,6 @@ void tiqdio_add_input_queues(struct qdio_irq *irq_ptr)
         struct qdio_q *q;
         int i;
 
-        /* No TDD facility? If we must use SIGA-s we can also omit SVS. */
-        if (!css_qdio_omit_svs && irq_ptr->siga_flag.sync)
-                css_qdio_omit_svs = 1;
-
         mutex_lock(&tiq_list_lock);
         for_each_input_queue(irq_ptr, q, i)
                 list_add_rcu(&q->entry, &tiq_list);
@@ -112,9 +95,9 @@ void tiqdio_remove_input_queues(struct qdio_irq *irq_ptr)
         }
 }
 
-static inline int shared_ind_used(void)
+static inline u32 shared_ind_set(void)
 {
-        return atomic_read(&q_indicators[TIQDIO_SHARED_IND].count);
+        return q_indicators[TIQDIO_SHARED_IND].ind;
 }
 
 /**
@@ -124,20 +107,11 @@ static inline int shared_ind_used(void)
  */
 static void tiqdio_thinint_handler(void *alsi, void *data)
 {
+        u32 si_used = shared_ind_set();
         struct qdio_q *q;
 
         last_ai_time = S390_lowcore.int_clock;
-
-        /*
-         * SVS only when needed: issue SVS to benefit from iqdio interrupt
-         * avoidance (SVS clears adapter interrupt suppression overwrite).
-         */
-        if (!css_qdio_omit_svs)
-                do_clear_global_summary();
-
-        /* reset local summary indicator */
-        if (shared_ind_used())
-                xchg(tiqdio_alsi, 0);
+        kstat_cpu(smp_processor_id()).irqs[IOINT_QAI]++;
 
         /* protect tiq_list entries, only changed in activate or shutdown */
         rcu_read_lock();
@@ -146,7 +120,10 @@ static void tiqdio_thinint_handler(void *alsi, void *data)
         list_for_each_entry_rcu(q, &tiq_list, entry) {
 
                 /* only process queues from changed sets */
-                if (!*q->irq_ptr->dsci)
+                if (unlikely(shared_ind(q->irq_ptr->dsci))) {
+                        if (!si_used)
+                                continue;
+                } else if (!*q->irq_ptr->dsci)
                         continue;
 
                 if (q->u.in.queue_start_poll) {
@@ -162,7 +139,7 @@ static void tiqdio_thinint_handler(void *alsi, void *data)
                                                  q->irq_ptr->int_parm);
                 } else {
                         /* only clear it if the indicator is non-shared */
-                        if (!shared_ind(q->irq_ptr))
+                        if (!shared_ind(q->irq_ptr->dsci))
                                 xchg(q->irq_ptr->dsci, 0);
                         /*
                          * Call inbound processing but not directly
@@ -178,13 +155,8 @@ static void tiqdio_thinint_handler(void *alsi, void *data)
          * If the shared indicator was used clear it now after all queues
          * were processed.
          */
-        if (shared_ind_used()) {
+        if (si_used && shared_ind_set())
                 xchg(&q_indicators[TIQDIO_SHARED_IND].ind, 0);
-
-                /* prevent racing */
-                if (*tiqdio_alsi)
-                        xchg(&q_indicators[TIQDIO_SHARED_IND].ind, 1 << 7);
-        }
 }
 
 static int set_subchannel_ind(struct qdio_irq *irq_ptr, int reset)
@@ -269,12 +241,6 @@ int qdio_establish_thinint(struct qdio_irq *irq_ptr)
 {
         if (!is_thinint_irq(irq_ptr))
                 return 0;
-
-        /* Check for aif time delay disablement. If installed,
-         * omit SVS even under LPAR
-         */
-        if (css_general_characteristics.aif_tdd)
-                css_qdio_omit_svs = 1;
         return set_subchannel_ind(irq_ptr, 0);
 }