From: Julian Wiedmann <jwi@linux.ibm.com>
Date: Mon, 31 May 2021 18:40:06 +0300
Subject: s390/qdio: fix roll-back after timeout on ESTABLISH ccw
Git-commit: 2c197870e4701610ec3b1143808d4e31152caf30
Patch-mainline: v5.15-rc1
References: git-fixes
When qdio_establish() times out while waiting for the ESTABLISH ccw to
complete, it calls qdio_shutdown() to roll back all of its previous
actions. But at this point the qdio_irq's state is still
QDIO_IRQ_STATE_INACTIVE, so qdio_shutdown() will exit immediately
without doing any actual work.
Which means that eg. the qdio_irq's thinint-indicator stays registered,
and cdev->handler isn't restored to its old value. And since
commit 954d6235be41 ("s390/qdio: make thinint registration symmetric")
the qdio_irq also stays on the tiq_list, so on the next qdio_establish()
we might get a helpful BUG from the list-debugging code:
...
[ 4633.512591] list_add double add: new=00000000005a4110, prev=00000001b357db78, next=00000000005a4110.
[ 4633.512621] ------------[ cut here ]------------
[ 4633.512623] kernel BUG at lib/list_debug.c:29!
...
[ 4633.512796] [<00000001b2c6ee9a>] __list_add_valid+0x82/0xa0
[ 4633.512798] ([<00000001b2c6ee96>] __list_add_valid+0x7e/0xa0)
[ 4633.512800] [<00000001b2fcecce>] qdio_establish_thinint+0x116/0x190
[ 4633.512805] [<00000001b2fcbe58>] qdio_establish+0x128/0x498
...
Fix this by extracting a goto-chain from the existing error exits in
qdio_establish(), and check the return value of the wait_event_...()
to detect the timeout condition.
Fixes: 779e6e1c724d ("[S390] qdio: new qdio driver.")
Root-caused-by: Benjamin Block <bblock@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Reviewed-by: Benjamin Block <bblock@linux.ibm.com>
Cc: <stable@vger.kernel.org> # 2.6.27
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Acked-by: Petr Tesarik <ptesarik@suse.com>
---
drivers/s390/cio/qdio_main.c | 31 +++++++++++++++++++------------
1 file changed, 19 insertions(+), 12 deletions(-)
--- a/drivers/s390/cio/qdio_main.c
+++ b/drivers/s390/cio/qdio_main.c
@@ -1340,6 +1340,7 @@ int qdio_establish(struct qdio_initializ
struct ccw_device *cdev = init_data->cdev;
struct subchannel_id schid;
struct qdio_irq *irq_ptr;
+ long timeout;
int rc;
ccw_device_get_schid(cdev, &schid);
@@ -1353,11 +1354,8 @@ int qdio_establish(struct qdio_initializ
qdio_setup_irq(init_data);
rc = qdio_establish_thinint(irq_ptr);
- if (rc) {
- qdio_shutdown_irq(irq_ptr);
- mutex_unlock(&irq_ptr->setup_mutex);
- return rc;
- }
+ if (rc)
+ goto err_thinint;
/* establish q */
irq_ptr->ccw.cmd_code = irq_ptr->equeue.cmd;
@@ -1373,15 +1371,16 @@ int qdio_establish(struct qdio_initializ
if (rc) {
DBF_ERROR("%4x est IO ERR", irq_ptr->schid.sch_no);
DBF_ERROR("rc:%4x", rc);
- qdio_shutdown_thinint(irq_ptr);
- qdio_shutdown_irq(irq_ptr);
- mutex_unlock(&irq_ptr->setup_mutex);
- return rc;
+ goto err_ccw_start;
}
- wait_event_interruptible_timeout(cdev->private->wait_q,
- irq_ptr->state == QDIO_IRQ_STATE_ESTABLISHED ||
- irq_ptr->state == QDIO_IRQ_STATE_ERR, HZ);
+ timeout = wait_event_interruptible_timeout(cdev->private->wait_q,
+ irq_ptr->state == QDIO_IRQ_STATE_ESTABLISHED ||
+ irq_ptr->state == QDIO_IRQ_STATE_ERR, HZ);
+ if (timeout <= 0) {
+ rc = (timeout == -ERESTARTSYS) ? -EINTR : -ETIME;
+ goto err_ccw_timeout;
+ }
if (irq_ptr->state != QDIO_IRQ_STATE_ESTABLISHED) {
mutex_unlock(&irq_ptr->setup_mutex);
@@ -1400,6 +1399,14 @@ int qdio_establish(struct qdio_initializ
qdio_print_subchannel_info(irq_ptr, cdev);
qdio_setup_debug_entries(irq_ptr, cdev);
return 0;
+
+err_ccw_timeout:
+err_ccw_start:
+ qdio_shutdown_thinint(irq_ptr);
+err_thinint:
+ qdio_shutdown_irq(irq_ptr);
+ mutex_unlock(&irq_ptr->setup_mutex);
+ return rc;
}
EXPORT_SYMBOL_GPL(qdio_establish);