The current documentation for fc_remote_port_block() and fc_remote_port_unblock() state that the functions can be called from either interrupt and process context. This is not the case, as del_timer_sync(), cancel_delayed_work(), and flush_workqueue() can all sleep: Debug: sleeping function called from invalid context at kernel/workqueue.c:264 in_atomic():1, irqs_disabled():1 [<c011b434>] __might_sleep+0xa4/0xc0 [<c0127b32>] try_to_del_timer_sync+0x52/0x60 [<c012faac>] flush_workqueue+0x1c/0xb0 [<c0127b68>] del_timer_sync+0x28/0x40 [<c02c00d0>] fc_remote_port_unblock+0x70/0xa0 [<c02c00d0>] fc_remote_port_unblock+0x70/0xa0 [<f8aefeca>] qla2x00_send_login_iocb_cb+0x59a/0x7c0 [qla2xxx] [<c03a3def>] _spin_lock_irqsave+0xf/0x20 [<c0127b32>] try_to_del_timer_sync+0x52/0x60 [<f8af079c>] qla2x00_process_iodesc+0x9c/0x150 [qla2xxx] [<f8ae76e3>] qla2x00_process_response_queue+0xd3/0x180 [qla2xxx] [<f8ae6c30>] qla2300_intr_handler+0x160/0x230 [qla2xxx] [<c0127ef0>] update_wall_time+0x10/0x50 [<c013ef79>] handle_IRQ_event+0x39/0x70 [<c013f047>] __do_IRQ+0x97/0x110 [<c01053e6>] do_IRQ+0x46/0x70 ======================= [<c0103b4e>] common_interrupt+0x1a/0x20 [<c023c82d>] acpi_processor_idle+0x107/0x29d [<c0100e89>] cpu_idle+0x69/0x80 [<c047da3a>] start_kernel+0x19a/0x1e0 [<c047d3b0>] unknown_bootoption+0x0/0x1e0 earlier in June we were indirectly hit by something similar during the klist transition: http://marc.theaimsgroup.com/?l=linux-scsi&m=111868700022257&w=2 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 The attached patch, defers the block/unblock() operations to the keventd thread. Signed-off-by: Andrew Vasquez <andrew.vasquez@xxxxxxxxxx> --- diff --git a/drivers/scsi/scsi_transport_fc.c b/drivers/scsi/scsi_transport_fc.c --- a/drivers/scsi/scsi_transport_fc.c +++ b/drivers/scsi/scsi_transport_fc.c @@ -215,6 +215,8 @@ fc_bitfield_name_search(remote_port_role static void fc_timeout_blocked_rport(void *data); static void fc_scsi_scan_rport(void *data); static void fc_rport_terminate(struct fc_rport *rport); +static void fc_block_rport(void *data); +static void fc_unblock_rport(void *data); /* * Attribute counts pre object type... @@ -1233,6 +1235,8 @@ fc_rport_create(struct Scsi_Host *shost, INIT_WORK(&rport->dev_loss_work, fc_timeout_blocked_rport, rport); INIT_WORK(&rport->scan_work, fc_scsi_scan_rport, rport); + INIT_WORK(&rport->block_work, fc_block_rport, rport); + INIT_WORK(&rport->unblock_work, fc_unblock_rport, rport); spin_lock_irqsave(shost->host_lock, flags); @@ -1585,6 +1589,19 @@ fc_timeout_blocked_rport(void *data) scsi_remove_target(&rport->dev); } +static void +fc_block_rport(void *data) +{ + struct fc_rport *rport = (struct fc_rport *)data; + + scsi_target_block(&rport->dev); + + /* cap the length the devices can be blocked */ + schedule_delayed_work(&rport->dev_loss_work, rport->dev_loss_tmo * HZ); + + rport->port_state = FC_PORTSTATE_BLOCKED; +} + /** * fc_remote_port_block - temporarily block any scsi traffic to a remote port. * @rport: remote port to be blocked. @@ -1611,35 +1628,19 @@ int fc_remote_port_block(struct fc_rport *rport) { int timeout = rport->dev_loss_tmo; - struct work_struct *work = &rport->dev_loss_work; if (timeout < 0 || timeout > SCSI_DEVICE_BLOCK_MAX_TIMEOUT) return -EINVAL; - scsi_target_block(&rport->dev); - - /* cap the length the devices can be blocked */ - schedule_delayed_work(work, timeout * HZ); - - rport->port_state = FC_PORTSTATE_BLOCKED; + schedule_work(&rport->block_work); return 0; } EXPORT_SYMBOL(fc_remote_port_block); -/** - * fc_remote_port_unblock - restart any blocked scsi traffic to a remote port. - * @rport: remote port to be unblocked. - * - * scsi lld's with a FC transport call this routine to restart IO to all - * devices associated with the caller's scsi target following a fc_target_block - * request. Called from interrupt or normal process context. - * - * Notes: - * This routine assumes no locks are held on entry. - **/ - void -fc_remote_port_unblock(struct fc_rport *rport) +static void +fc_unblock_rport(void *data) { + struct fc_rport *rport = (struct fc_rport *)data; struct work_struct *work = &rport->dev_loss_work; struct Scsi_Host *shost = rport_to_shost(rport); @@ -1662,6 +1663,24 @@ fc_remote_port_unblock(struct fc_rport * rport->port_state = FC_PORTSTATE_ONLINE; } + +/** + * fc_remote_port_unblock - restart any blocked scsi traffic to a remote port. + * @rport: remote port to be unblocked. + * + * scsi lld's with a FC transport call this routine to restart IO to all + * devices associated with the caller's scsi target following a fc_target_block + * request. Called from interrupt or normal process context. + * + * Notes: + * This routine assumes no locks are held on entry. + **/ +void +fc_remote_port_unblock(struct fc_rport *rport) +{ + schedule_work(&rport->unblock_work); + +} EXPORT_SYMBOL(fc_remote_port_unblock); /** diff --git a/include/scsi/scsi_transport_fc.h b/include/scsi/scsi_transport_fc.h --- a/include/scsi/scsi_transport_fc.h +++ b/include/scsi/scsi_transport_fc.h @@ -203,6 +203,8 @@ struct fc_rport { /* aka fc_starget_attr struct device dev; struct work_struct dev_loss_work; struct work_struct scan_work; + struct work_struct block_work; + struct work_struct unblock_work; } __attribute__((aligned(sizeof(unsigned long)))); #define dev_to_rport(d) \ - : 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
