Re: [PATCH V3 7/9] aacraid: Fix AIF triggered IOP_RESET

[Tomas Henzl <thenzl@redhat.com>]

On 20.1.2016 21:32, Raghava Aditya Renukunta wrote:
>
>> -----Original Message-----
>> From: Tomas Henzl [mailto:thenzl@redhat.com]
>> Sent: Tuesday, January 19, 2016 8:14 AM
>> To: Raghava Aditya Renukunta; James.Bottomley@HansenPartnership.com;
>> martin.petersen@oracle.com; linux-scsi@vger.kernel.org
>> Cc: Mahesh Rajashekhara; Murthy Bhat; Gana Sridaran; aacraid@pmc-
>> sierra.com; Scott Benesh; jthumshirn@suse.de; shane.seymour@hpe.com;
>> David Carroll
>> Subject: Re: [PATCH V3 7/9] aacraid: Fix AIF triggered IOP_RESET
>>
>> On 15.1.2016 08:16, Raghava Aditya Renukunta wrote:
>>> From: Raghava Aditya Renukunta <raghavaaditya.renukunta@pmcs.com>
>>>
>>> while driver removal is in progress or PCI shutdown is invoked, driver
>>> kills AIF aacraid thread, but IOCTL requests from the management tools
>>> re-start AIF thread leading to IOP_RESET.
>>>
>>> Fixed by setting adapter_shutdown flag when PCI shutdown is invoked.
>>>
>>> Changes in V2:
>>> Set adapter_shutdown flag before shutdown command is sent to \
>>> controller
>>>
>>> Changes in V3:
>>> Call aac_send_shut_shutdown first thing in __aac_shutdown
>>> Convert adapter_shutdown to atomic_t variable to prevent \
>>> SMP coherency issues(race conditions)
>> Hi,
>> I don't think that an atomic variable can change it, imagine that
>> thread just passed your test in aac_cfg_ioctl and another thread
>> enter a bit later the adapter_shutdown so both - an I/O and your shutdown
>> code
>> will run in parallel.
>> Also you need to fix your compat_ioctl path too.
>>
>> --tm
> Hello Tomas,
> Yes that would not solve this problem.
> I can think of 2 solutions
>
> 1.Place a delay after setting adapter_shutdown and before sending the actual
> shutdown command in aac_send_shutdown. This way any impending commands will be 
> processed before the adapter  actually receives the shutdown command. Since management 
> commands are sync only , shutdown command will be sent last. This solution uses an
> estimation of the delay
>
> 2.Since  commands are sync , place a check in aac_fib_send to block 
>  commands once adapter_shutdown is set(only shutdown command will be sent thru)

This option looks better but I guess you still can find a tiny race window.
What do you think about a mutual exclusive access using a mutex, do you think this could work?

diff --git a/drivers/scsi/aacraid/commctrl.c b/drivers/scsi/aacraid/commctrl.c
index 54195a117f..b9505c58db 100644
--- a/drivers/scsi/aacraid/commctrl.c
+++ b/drivers/scsi/aacraid/commctrl.c
@@ -858,6 +858,8 @@ int aac_do_ioctl(struct aac_dev * dev, int cmd, void __user *arg)
 	/*
 	 *	HBA gets first crack
 	 */
+	if (dev->adapter_shutdown)
+		return -EACCES;
 
 	status = aac_dev_ioctl(dev, cmd, arg);
 	if (status != -ENOTTY)
diff --git a/drivers/scsi/aacraid/comminit.c b/drivers/scsi/aacraid/comminit.c
index 0e954e37f0..02535c07a4 100644
--- a/drivers/scsi/aacraid/comminit.c
+++ b/drivers/scsi/aacraid/comminit.c
@@ -212,6 +212,10 @@ int aac_send_shutdown(struct aac_dev * dev)
 		return -ENOMEM;
 	aac_fib_init(fibctx);
 
+	mutex_lock(&aac_mutex);
+	dev->adapter_shutdown = 1;
+	mutex_unlock(&aac_mutex);
+
 	cmd = (struct aac_close *) fib_data(fibctx);
 
 	cmd->command = cpu_to_le32(VM_CloseAll);
@@ -229,7 +233,6 @@ int aac_send_shutdown(struct aac_dev * dev)
 	/* FIB should be freed only after getting the response from the F/W */
 	if (status != -ERESTARTSYS)
 		aac_fib_free(fibctx);
-	dev->adapter_shutdown = 1;
 	if ((dev->pdev->device == PMC_DEVICE_S7 ||
 	     dev->pdev->device == PMC_DEVICE_S8 ||
 	     dev->pdev->device == PMC_DEVICE_S9) &&
diff --git a/drivers/scsi/aacraid/linit.c b/drivers/scsi/aacraid/linit.c
index 6944560e22..a87880ab32 100644
--- a/drivers/scsi/aacraid/linit.c
+++ b/drivers/scsi/aacraid/linit.c
@@ -706,8 +706,9 @@ static long aac_cfg_ioctl(struct file *file,
 	int ret;
 	struct aac_dev *aac;
 	aac = (struct aac_dev *)file->private_data;
-	if (!capable(CAP_SYS_RAWIO) || aac->adapter_shutdown)
+	if (!capable(CAP_SYS_RAWIO))
 		return -EPERM;
+
 	mutex_lock(&aac_mutex);
 	ret = aac_do_ioctl(file->private_data, cmd, (void __user *)arg);
 	mutex_unlock(&aac_mutex);

>
> I am more inclined to go with option 2.
>
> Regards,
> Raghava Aditya
>
>>> Signed-off-by: Raghava Aditya Renukunta
>> <raghavaaditya.renukunta@pmcs.com>
>>> ---
>>>  drivers/scsi/aacraid/aacraid.h  |  2 +-
>>>  drivers/scsi/aacraid/comminit.c |  6 +++---
>>>  drivers/scsi/aacraid/linit.c    | 15 ++++++++-------
>>>  3 files changed, 12 insertions(+), 11 deletions(-)
>>>
>>> diff --git a/drivers/scsi/aacraid/aacraid.h b/drivers/scsi/aacraid/aacraid.h
>>> index 2916288..3473668 100644
>>> --- a/drivers/scsi/aacraid/aacraid.h
>>> +++ b/drivers/scsi/aacraid/aacraid.h
>>> @@ -1234,7 +1234,7 @@ struct aac_dev
>>>  	int			msi_enabled;	/* MSI/MSI-X enabled */
>>>  	struct msix_entry	msixentry[AAC_MAX_MSIX];
>>>  	struct aac_msix_ctx	aac_msix[AAC_MAX_MSIX]; /* context */
>>> -	u8			adapter_shutdown;
>>> +	atomic_t		adapter_shutdown;
>>>  	u32			handle_pci_error;
>>>  };
>>>
>>> diff --git a/drivers/scsi/aacraid/comminit.c
>> b/drivers/scsi/aacraid/comminit.c
>>> index 0e954e3..d361673 100644
>>> --- a/drivers/scsi/aacraid/comminit.c
>>> +++ b/drivers/scsi/aacraid/comminit.c
>>> @@ -212,8 +212,9 @@ int aac_send_shutdown(struct aac_dev * dev)
>>>  		return -ENOMEM;
>>>  	aac_fib_init(fibctx);
>>>
>>> -	cmd = (struct aac_close *) fib_data(fibctx);
>>> +	atomic_set(&dev->adapter_shutdown, 1);
>>>
>>> +	cmd = (struct aac_close *) fib_data(fibctx);
>>>  	cmd->command = cpu_to_le32(VM_CloseAll);
>>>  	cmd->cid = cpu_to_le32(0xfffffffe);
>>>
>>> @@ -229,7 +230,6 @@ int aac_send_shutdown(struct aac_dev * dev)
>>>  	/* FIB should be freed only after getting the response from the F/W
>> */
>>>  	if (status != -ERESTARTSYS)
>>>  		aac_fib_free(fibctx);
>>> -	dev->adapter_shutdown = 1;
>>>  	if ((dev->pdev->device == PMC_DEVICE_S7 ||
>>>  	     dev->pdev->device == PMC_DEVICE_S8 ||
>>>  	     dev->pdev->device == PMC_DEVICE_S9) &&
>>> @@ -468,7 +468,7 @@ struct aac_dev *aac_init_adapter(struct aac_dev
>> *dev)
>>>  	}
>>>  	dev->max_msix = 0;
>>>  	dev->msi_enabled = 0;
>>> -	dev->adapter_shutdown = 0;
>>> +	atomic_set(&dev->adapter_shutdown, 0);
>>>  	if ((!aac_adapter_sync_cmd(dev,
>> GET_COMM_PREFERRED_SETTINGS,
>>>  	  0, 0, 0, 0, 0, 0,
>>>  	  status+0, status+1, status+2, status+3, status+4))
>>> diff --git a/drivers/scsi/aacraid/linit.c b/drivers/scsi/aacraid/linit.c
>>> index 6944560..27b3fcd 100644
>>> --- a/drivers/scsi/aacraid/linit.c
>>> +++ b/drivers/scsi/aacraid/linit.c
>>> @@ -706,7 +706,7 @@ static long aac_cfg_ioctl(struct file *file,
>>>  	int ret;
>>>  	struct aac_dev *aac;
>>>  	aac = (struct aac_dev *)file->private_data;
>>> -	if (!capable(CAP_SYS_RAWIO) || aac->adapter_shutdown)
>>> +	if (!capable(CAP_SYS_RAWIO) || atomic_read(&aac-
>>> adapter_shutdown))
>>>  		return -EPERM;
>>>  	mutex_lock(&aac_mutex);
>>>  	ret = aac_do_ioctl(file->private_data, cmd, (void __user *)arg);
>>> @@ -1078,6 +1078,8 @@ static void __aac_shutdown(struct aac_dev * aac)
>>>  	int i;
>>>  	int cpu;
>>>
>>> +	aac_send_shutdown(aac);
>>> +
>>>  	if (aac->aif_thread) {
>>>  		int i;
>>>  		/* Clear out events first */
>>> @@ -1089,7 +1091,7 @@ static void __aac_shutdown(struct aac_dev * aac)
>>>  		}
>>>  		kthread_stop(aac->thread);
>>>  	}
>>> -	aac_send_shutdown(aac);
>>> +
>>>  	aac_adapter_disable_int(aac);
>>>  	cpu = cpumask_first(cpu_online_mask);
>>>  	if (aac->pdev->device == PMC_DEVICE_S6 ||
>>> @@ -1474,7 +1476,7 @@ static int aac_resume(struct pci_dev *pdev)
>>>  	* reset this flag to unblock ioctl() as it was set at
>>>  	* aac_send_shutdown() to block ioctls from upperlayer
>>>  	*/
>>> -	aac->adapter_shutdown = 0;
>>> +	atomic_set(&aac->adapter_shutdown, 0);
>>>  	scsi_unblock_requests(shost);
>>>
>>>  	return 0;
>>> @@ -1553,9 +1555,8 @@ static pci_ers_result_t
>> aac_pci_error_detected(struct pci_dev *pdev,
>>>  	case pci_channel_io_normal:
>>>  		return PCI_ERS_RESULT_CAN_RECOVER;
>>>  	case pci_channel_io_frozen:
>>> -
>>>  		aac->handle_pci_error = 1;
>>> -		aac->adapter_shutdown = 1;
>>> +		atomic_set(&aac->adapter_shutdown, 1);
>>>
>>>  		scsi_block_requests(aac->scsi_host_ptr);
>>>  		aac_flush_ios(aac);
>>> @@ -1567,7 +1568,7 @@ static pci_ers_result_t
>> aac_pci_error_detected(struct pci_dev *pdev,
>>>  		return PCI_ERS_RESULT_NEED_RESET;
>>>  	case pci_channel_io_perm_failure:
>>>  		aac->handle_pci_error = 1;
>>> -		aac->adapter_shutdown = 1;
>>> +		atomic_set(&aac->adapter_shutdown, 1);
>>>
>>>  		aac_flush_ios(aac);
>>>  		return PCI_ERS_RESULT_DISCONNECT;
>>> @@ -1636,7 +1637,7 @@ static void aac_pci_resume(struct pci_dev *pdev)
>>>  	 * reset this flag to unblock ioctl() as it was set
>>>  	 * at aac_send_shutdown() to block ioctls from upperlayer
>>>  	 */
>>> -	aac->adapter_shutdown = 0;
>>> +	atomic_set(&aac->adapter_shutdown, 0);
>>>  	aac->handle_pci_error = 0;
>>>
>>>  	shost_for_each_device(sdev, shost)
> --
> To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html