src/patches/suse-2.6.27.31/patches.arch/s390-12-01-cio-reprobe-deadlock-fix.patch

   1 From: Gerald Schaefer <geraldsc@de.ibm.com>
   2 Subject: cio: incomplete device initialization on LPAR
   3 References: bnc#487755,LTC#52366
   4
   5 Symptom:     Random CCW devices are not usable after IPL. Sysfs shows
   6              subchannel directories without links to CCW devices. This occurs
   7              on LPARs with DASDs.
   8 Problem:     The DASD device driver triggers the subchannel reprobe function
   9              during PAV initialization. Subchannel reprobing blocks the kslowcrw
  10              workqueue until all outstanding device recognition requests have
  11              finished processing. Device recognition uses kslowcrw to finish
  12              processing which leads to a workqueue deadlock.
  13 Solution:    Move the waiting portion of subchannel reprobing to the cio
  14              workqueue.
  15
  16 Acked-by: John Jolly <jjolly@suse.de>
  17
  18 ---
  19  drivers/s390/cio/css.c |   19 ++++++++++++++++---
  20  1 file changed, 16 insertions(+), 3 deletions(-)
  21
  22 Index: linux-sles11/drivers/s390/cio/css.c
  23 ===================================================================
  24 --- linux-sles11.orig/drivers/s390/cio/css.c
  25 +++ linux-sles11/drivers/s390/cio/css.c
  26 @@ -509,6 +509,17 @@ static int reprobe_subchannel(struct sub
  27         return ret;
  28  }
  29
  30 +static void reprobe_after_idle(struct work_struct *unused)
  31 +{
  32 +       /* Make sure initial subchannel scan is done. */
  33 +       wait_event(ccw_device_init_wq,
  34 +                  atomic_read(&ccw_device_init_count) == 0);
  35 +       if (need_reprobe)
  36 +               css_schedule_reprobe();
  37 +}
  38 +
  39 +static DECLARE_WORK(reprobe_idle_work, reprobe_after_idle);
  40 +
  41  /* Work function used to reprobe all unregistered subchannels. */
  42  static void reprobe_all(struct work_struct *unused)
  43  {
  44 @@ -516,10 +527,12 @@ static void reprobe_all(struct work_stru
  45
  46         CIO_MSG_EVENT(4, "reprobe start\n");
  47
  48 -       need_reprobe = 0;
  49         /* Make sure initial subchannel scan is done. */
  50 -       wait_event(ccw_device_init_wq,
  51 -                  atomic_read(&ccw_device_init_count) == 0);
  52 +       if (atomic_read(&ccw_device_init_count) != 0) {
  53 +               queue_work(ccw_device_work, &reprobe_idle_work);
  54 +               return;
  55 +       }
  56 +       need_reprobe = 0;
  57         ret = for_each_subchannel_staged(NULL, reprobe_subchannel, NULL);
  58
  59         CIO_MSG_EVENT(4, "reprobe done (rc=%d, need_reprobe=%d)\n", ret,