mpt2sas: setpci reset on nytro warpdrive card along with sysfs access and
cli ioctl access resulted in kernel oops
1. pci_access_mutex lock added to provide synchronization between IOCTL,
sysfs, PCI resource handling path
2. gioc_lock spinlock to protect list operations over multiple
controllers
>From
c53a1cff4c07528b8b9ec7f6716e94950283e8f9 Mon Sep 17 00:00:00 2001
From: Nagarajkumar Narayanan <nagarajkumar.narayanan@seagate.com>
Date: Tue, 18 Aug 2015 11:58:13 +0530
Subject: [PATCH] mpt2sas setpci reset oops fix
In mpt2sas driver due to lack of synchronization between ioctl,
BRM status access through sysfs, pci resource removal kernel oops
happen as ioctl path and BRM status sysfs access path still tries
to access the removed resources
Two locks added to provide syncrhonization
1. pci_access_mutex: Mutex to synchronize ioctl,sysfs show path and
pci resource handling. PCI resource freeing will lead to free
vital hardware/memory resource, which might be in use by cli/sysfs
path functions resulting in Null pointer reference followed by kernel
crash. To avoid the above race condition we use mutex syncrhonization
which ensures the syncrhonization between cli/sysfs_show path
Note: pci_access_mutex is used only if nytro warpdrive cards
(ioc->is_warpdrive based on device id) are used
as we could not test this case with other SAS2 HBA cards
We can remove this check if this behaviour confirmed from other
cards.
2. spinlock on list operations over IOCs
Case: when multiple warpdrive cards(IOCs) are in use
Each IOC will added to the ioc list stucture on initialization.
Watchdog threads run at regular intervals to check IOC for any
fault conditions which will trigger the dead_ioc thread to
deallocate pci resource, resulting deleting the IOC netry from list,
this deletion need to protected by spinlock to enusre that
ioc removal is syncrhonized, if not synchronized it might lead to
list_del corruption as the ioc list is traversed in cli path
Signed-off-by: Nagarajkumar Narayanan <nagarajkumar.narayanan@seagate.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Acked-by: Sreekanth Reddy <sreekanth.reddy@avagotech.com>
Signed-off-by: James Bottomley <JBottomley@Odin.com>
if (ret)
return ret;
+ /* global ioc spinlock to protect controller list on list operations */
printk(KERN_INFO "setting fwfault_debug(%d)\n", mpt2sas_fwfault_debug);
+ spin_lock(&gioc_lock);
list_for_each_entry(ioc, &mpt2sas_ioc_list, list)
ioc->fwfault_debug = mpt2sas_fwfault_debug;
+ spin_unlock(&gioc_lock);
return 0;
}
dexitprintk(ioc, printk(MPT2SAS_INFO_FMT "%s\n", ioc->name,
__func__));
+ /* synchronizing freeing resource with pci_access_mutex lock */
+ mutex_lock(&ioc->pci_access_mutex);
if (ioc->chip_phys && ioc->chip) {
_base_mask_interrupts(ioc);
ioc->shost_recovery = 1;
pci_disable_pcie_error_reporting(pdev);
pci_disable_device(pdev);
}
+ mutex_unlock(&ioc->pci_access_mutex);
return;
}
* @delayed_tr_list: target reset link list
* @delayed_tr_volume_list: volume target reset link list
* @@temp_sensors_count: flag to carry the number of temperature sensors
+ * @pci_access_mutex: Mutex to synchronize ioctl,sysfs show path and
+ * pci resource handling. PCI resource freeing will lead to free
+ * vital hardware/memory resource, which might be in use by cli/sysfs
+ * path functions resulting in Null pointer reference followed by kernel
+ * crash. To avoid the above race condition we use mutex syncrhonization
+ * which ensures the syncrhonization between cli/sysfs_show path
*/
struct MPT2SAS_ADAPTER {
struct list_head list;
u8 mfg_pg10_hide_flag;
u8 hide_drives;
+ struct mutex pci_access_mutex;
};
typedef u8 (*MPT_CALLBACK)(struct MPT2SAS_ADAPTER *ioc, u16 smid, u8 msix_index,
/* base shared API */
extern struct list_head mpt2sas_ioc_list;
+/* spinlock on list operations over IOCs
+ * Case: when multiple warpdrive cards(IOCs) are in use
+ * Each IOC will added to the ioc list stucture on initialization.
+ * Watchdog threads run at regular intervals to check IOC for any
+ * fault conditions which will trigger the dead_ioc thread to
+ * deallocate pci resource, resulting deleting the IOC netry from list,
+ * this deletion need to protected by spinlock to enusre that
+ * ioc removal is syncrhonized, if not synchronized it might lead to
+ * list_del corruption as the ioc list is traversed in cli path
+ */
+extern spinlock_t gioc_lock;
void mpt2sas_base_start_watchdog(struct MPT2SAS_ADAPTER *ioc);
void mpt2sas_base_stop_watchdog(struct MPT2SAS_ADAPTER *ioc);
struct MPT2SAS_ADAPTER *ioc, u64 sas_address);
void mpt2sas_port_enable_complete(struct MPT2SAS_ADAPTER *ioc);
-
void mpt2sas_scsih_reset_handler(struct MPT2SAS_ADAPTER *ioc, int reset_phase);
/* config shared API */
_ctl_verify_adapter(int ioc_number, struct MPT2SAS_ADAPTER **iocpp)
{
struct MPT2SAS_ADAPTER *ioc;
-
+ /* global ioc lock to protect controller on list operations */
+ spin_lock(&gioc_lock);
list_for_each_entry(ioc, &mpt2sas_ioc_list, list) {
if (ioc->id != ioc_number)
continue;
+ spin_unlock(&gioc_lock);
*iocpp = ioc;
return ioc_number;
}
+ spin_unlock(&gioc_lock);
*iocpp = NULL;
return -1;
}
poll_wait(filep, &ctl_poll_wait, wait);
+ /* global ioc lock to protect controller on list operations */
+ spin_lock(&gioc_lock);
list_for_each_entry(ioc, &mpt2sas_ioc_list, list) {
- if (ioc->aen_event_read_flag)
+ if (ioc->aen_event_read_flag) {
+ spin_unlock(&gioc_lock);
return POLLIN | POLLRDNORM;
+ }
}
+ spin_unlock(&gioc_lock);
return 0;
}
if (_ctl_verify_adapter(ioctl_header.ioc_number, &ioc) == -1 || !ioc)
return -ENODEV;
+ /* pci_access_mutex lock acquired by ioctl path */
+ mutex_lock(&ioc->pci_access_mutex);
if (ioc->shost_recovery || ioc->pci_error_recovery ||
- ioc->is_driver_loading)
- return -EAGAIN;
+ ioc->is_driver_loading || ioc->remove_host) {
+ ret = -EAGAIN;
+ goto out_unlock_pciaccess;
+ }
state = (file->f_flags & O_NONBLOCK) ? NON_BLOCKING : BLOCKING;
if (state == NON_BLOCKING) {
- if (!mutex_trylock(&ioc->ctl_cmds.mutex))
- return -EAGAIN;
+ if (!mutex_trylock(&ioc->ctl_cmds.mutex)) {
+ ret = -EAGAIN;
+ goto out_unlock_pciaccess;
+ }
} else if (mutex_lock_interruptible(&ioc->ctl_cmds.mutex)) {
- return -ERESTARTSYS;
+ ret = -ERESTARTSYS;
+ goto out_unlock_pciaccess;
}
switch (cmd) {
}
mutex_unlock(&ioc->ctl_cmds.mutex);
+out_unlock_pciaccess:
+ mutex_unlock(&ioc->pci_access_mutex);
return ret;
}
"warpdrive\n", ioc->name, __func__);
goto out;
}
+ /* pci_access_mutex lock acquired by sysfs show path */
+ mutex_lock(&ioc->pci_access_mutex);
+ if (ioc->pci_error_recovery || ioc->remove_host) {
+ mutex_unlock(&ioc->pci_access_mutex);
+ return 0;
+ }
/* allocate upto GPIOVal 36 entries */
sz = offsetof(Mpi2IOUnitPage3_t, GPIOVal) + (sizeof(u16) * 36);
out:
kfree(io_unit_pg3);
+ mutex_unlock(&ioc->pci_access_mutex);
return rc;
}
static DEVICE_ATTR(BRM_status, S_IRUGO, _ctl_BRM_status_show, NULL);
/* global parameters */
LIST_HEAD(mpt2sas_ioc_list);
-
+/* global ioc lock for list operations */
+DEFINE_SPINLOCK(gioc_lock);
/* local parameters */
static u8 scsi_io_cb_idx = -1;
static u8 tm_cb_idx = -1;
return ret;
printk(KERN_INFO "setting logging_level(0x%08x)\n", logging_level);
+ spin_lock(&gioc_lock);
list_for_each_entry(ioc, &mpt2sas_ioc_list, list)
ioc->logging_level = logging_level;
+ spin_unlock(&gioc_lock);
return 0;
}
module_param_call(logging_level, _scsih_set_debug_level, param_get_int,
sas_remove_host(shost);
scsi_remove_host(shost);
mpt2sas_base_detach(ioc);
+ spin_lock(&gioc_lock);
list_del(&ioc->list);
+ spin_unlock(&gioc_lock);
scsi_host_put(shost);
}
ioc = shost_priv(shost);
memset(ioc, 0, sizeof(struct MPT2SAS_ADAPTER));
INIT_LIST_HEAD(&ioc->list);
+ spin_lock(&gioc_lock);
list_add_tail(&ioc->list, &mpt2sas_ioc_list);
+ spin_unlock(&gioc_lock);
ioc->shost = shost;
ioc->id = mpt_ids++;
sprintf(ioc->name, "%s%d", MPT2SAS_DRIVER_NAME, ioc->id);
ioc->schedule_dead_ioc_flush_running_cmds = &_scsih_flush_running_cmds;
/* misc semaphores and spin locks */
mutex_init(&ioc->reset_in_progress_mutex);
+ /* initializing pci_access_mutex lock */
+ mutex_init(&ioc->pci_access_mutex);
spin_lock_init(&ioc->ioc_reset_in_progress_lock);
spin_lock_init(&ioc->scsi_lookup_lock);
spin_lock_init(&ioc->sas_device_lock);
out_attach_fail:
destroy_workqueue(ioc->firmware_event_thread);
out_thread_fail:
+ spin_lock(&gioc_lock);
list_del(&ioc->list);
+ spin_unlock(&gioc_lock);
scsi_host_put(shost);
return rv;
}
projects / karo-tx-linux.git / commitdiff