On Mon, 13 Jun 2005, Christophe Varoqui wrote:
>
> I'm testing Mike Christie's START_STOP hwhandler and discovered a bunch of new, interesting, phenomenons :
>
> A little context first :
> o kernel 2.6.12-rc6 + qlogic discovery patch
> o qla2342 (dual 2GB)
> o EVA5000, Solaris-tagged connections
>
> Here is a map create by multipath, fresh from boot :
>
> eva1_lun2 (3600508b400014ba7000120000cf00000)
> [size=50 GB][features="1 queue_if_no_path"][hwhandler="1 hp_sw"]
> \_ round-robin 0 [active][best]
> \_ 0:0:0:2 sdb 8:16 [ready ][active]
> \_ 1:0:0:2 sdf 8:80 [ready ][active]
> \_ round-robin 0 [enabled]
> \_ 0:0:1:2 sdd 8:48 [faulty][active]
> \_ 1:0:1:2 sdh 8:112 [faulty][active]
>
> Start a background stream read with dd on that map.
>
> Do a port disable on the FC switch port connected to HBA 0
> Consistently at this moment I get the following in the logs :
>
> qla2300 0000:05:0d.0: LOOP DOWN detected.
> Debug: sleeping function called from invalid context at include/linux/rwsem.h:43
> in_atomic():1, irqs_disabled():1
> [<c0120a74>] __might_sleep+0xa4/0xc0
> [<c026a466>] device_for_each_child+0x26/0x80
> [<c02b3180>] target_block+0x0/0x30
> [<c02bbdae>] fc_remote_port_block+0x2e/0x60
> [<c02bdbf5>] qla2x00_mark_all_devices_lost+0x55/0x60
> [<c02c597e>] qla2x00_async_event+0x83e/0xd60
> [<c011dd2b>] find_busiest_group+0xbb/0x310
> [<c02cdce4>] sd_rw_intr+0x164/0x320
> [<c02c4e37>] qla2300_intr_handler+0x77/0x240
> [<c0144882>] handle_IRQ_event+0x32/0x70
Without wanting to making a number of large changes to the qla2xxx
internals to deal with these pre-qualifications, could you try the
following patch (lightly tested with latest linus git tree).
We'll need to update the fc_remote_port docs in order to account for
this semantic change in device_for_each_child().
--
av
Postpone fc_rport block/unblock to scheduled work.
diff --git a/drivers/scsi/qla2xxx/qla_def.h b/drivers/scsi/qla2xxx/qla_def.h
--- a/drivers/scsi/qla2xxx/qla_def.h
+++ b/drivers/scsi/qla2xxx/qla_def.h
@@ -33,6 +33,7 @@
#include <linux/mempool.h>
#include <linux/spinlock.h>
#include <linux/completion.h>
+#include <linux/workqueue.h>
#include <asm/semaphore.h>
#include <scsi/scsi.h>
@@ -1644,6 +1645,8 @@ typedef struct fc_port {
uint8_t cur_path; /* current path id */
struct fc_rport *rport;
+ struct work_struct block_work;
+ struct work_struct unblock_work;
} fc_port_t;
/*
diff --git a/drivers/scsi/qla2xxx/qla_gbl.h b/drivers/scsi/qla2xxx/qla_gbl.h
--- a/drivers/scsi/qla2xxx/qla_gbl.h
+++ b/drivers/scsi/qla2xxx/qla_gbl.h
@@ -82,6 +82,8 @@ extern void qla2x00_cmd_timeout(srb_t *)
extern void qla2x00_mark_device_lost(scsi_qla_host_t *, fc_port_t *, int);
extern void qla2x00_mark_all_devices_lost(scsi_qla_host_t *);
+extern void qla2x00_block_fcport(void *);
+extern void qla2x00_unblock_fcport(void *);
extern void qla2x00_blink_led(scsi_qla_host_t *);
diff --git a/drivers/scsi/qla2xxx/qla_init.c b/drivers/scsi/qla2xxx/qla_init.c
--- a/drivers/scsi/qla2xxx/qla_init.c
+++ b/drivers/scsi/qla2xxx/qla_init.c
@@ -1534,6 +1534,8 @@ qla2x00_alloc_fcport(scsi_qla_host_t *ha
fcport->iodesc_idx_sent = IODESC_INVALID_INDEX;
atomic_set(&fcport->state, FCS_UNCONFIGURED);
fcport->flags = FCF_RLC_SUPPORT;
+ INIT_WORK(&fcport->block_work, qla2x00_block_fcport, fcport);
+ INIT_WORK(&fcport->unblock_work, qla2x00_unblock_fcport, fcport);
return (fcport);
}
@@ -1899,7 +1901,7 @@ qla2x00_reg_remote_port(scsi_qla_host_t
struct fc_rport *rport;
if (fcport->rport) {
- fc_remote_port_unblock(fcport->rport);
+ schedule_work(&fcport->unblock_work);
return;
}
diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
--- a/drivers/scsi/qla2xxx/qla_os.c
+++ b/drivers/scsi/qla2xxx/qla_os.c
@@ -1407,6 +1407,8 @@ void qla2x00_remove_one(struct pci_dev *
qla2x00_free_sysfs_attr(ha);
+ flush_scheduled_work();
+
fc_remove_host(ha->host);
scsi_remove_host(ha->host);
@@ -1481,7 +1483,7 @@ void qla2x00_mark_device_lost(scsi_qla_h
int do_login)
{
if (atomic_read(&fcport->state) == FCS_ONLINE && fcport->rport)
- fc_remote_port_block(fcport->rport);
+ schedule_work(&fcport->block_work);
/*
* We may need to retry the login, so don't change the state of the
* port but do the retries.
@@ -1542,11 +1544,25 @@ qla2x00_mark_all_devices_lost(scsi_qla_h
if (atomic_read(&fcport->state) == FCS_DEVICE_DEAD)
continue;
if (atomic_read(&fcport->state) == FCS_ONLINE && fcport->rport)
- fc_remote_port_block(fcport->rport);
+ schedule_work(&fcport->block_work);
atomic_set(&fcport->state, FCS_DEVICE_LOST);
}
}
+void
+qla2x00_block_fcport(void *data)
+{
+ fc_port_t *fcport = (fc_port_t *)data;
+ fc_remote_port_block(fcport->rport);
+}
+
+void
+qla2x00_unblock_fcport(void *data)
+{
+ fc_port_t *fcport = (fc_port_t *)data;
+ fc_remote_port_unblock(fcport->rport);
+}
+
/*
* qla2x00_mem_alloc
* Allocates adapter memory.
-
: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
