Re: [PATCH 4/4] [SCSI] ufshcd: SCSI error handling

From: Amit Sahrawat
Date: Sat Feb 11 2012 - 12:39:31 EST


Hi Santosh,
Below is the original function in your patch:
+static int ufshcd_abort(struct scsi_cmnd *cmd)
+{
+ struct Scsi_Host *host;
+ struct ufs_hba *hba;
+ unsigned long flags;
+ unsigned int tag;
+ unsigned int pos;
+ int err;
+
+ host = cmd->device->host;
+ hba = (struct ufs_hba *) host->hostdata;
+ tag = cmd->request->tag;
+
+ spin_lock_irqsave(host->host_lock, flags);
+ pos = (1 << tag);
+
+ /* check if command is still pending */
+ if (!(hba->outstanding_reqs & pos)) {
+ err = -1;
+ spin_unlock_irqrestore(host->host_lock, flags);
at this place, you are taking care of unlocking - irqrestore - in case
of error. which is ok.
+ goto out;
+ }
+
+ err = ufshcd_issue_tm_cmd(hba, &hba->lrb[tag], UFS_ABORT_TASK);
+ if (!err) {
+ spin_lock_irqsave(host->host_lock, flags);
+ scsi_dma_unmap(cmd);
+
+ /* clear the respective UTRLCLR bit */
+ writel(~pos,
+ (UFSHCD_MMIO_BASE +
+ REG_UTP_TRANSFER_REQ_LIST_CLEAR));
+ hba->outstanding_reqs &= ~pos;
+ hba->lrb[tag].cmd = NULL;
+ spin_unlock_irqrestore(host->host_lock, flags);
this unlocking - irqrestore - will work for your irqsave inside this
'if' - there is no return inside this
+ }
+out:
+ return err;
At this point, there is no unlocking-restore for spin_lock_irqsave
done at the entry of this function
+}
So, the changes I suggested will take care of handling the error
condition as well as the conditions for 'ufshcd_issue_tm_cmd'.

Please let me know in case I am missing something.

Regards,
Amit Sahrawat

On 2/7/12, Santosh Y <santoshsy@xxxxxxxxx> wrote:
> On Mon, Feb 6, 2012 at 12:46 PM, Amit Sahrawat
> <amit.sahrawat83@xxxxxxxxx> wrote:
>> Hi,
>>
>> In function:
>> +static int ufshcd_abort(struct scsi_cmnd *cmd)
>> +{...
>> - int err;
>> + int err = -1;
>> ...
>> + spin_lock_irqsave(host->host_lock, flags);
>> + pos = (1 << tag);
>> +
>> + /* check if command is still pending */
>> + if (!(hba->outstanding_reqs & pos)) {
>> - err = -1;
>> - spin_unlock_irqrestore(host->host_lock, flags);
>> + goto out;
>> + }
>> ...
>> ...
>> +out:
>> + spin_unlock_irqrestore(host->host_lock, flags);
>> + return err;
>> +}
>> this will also take of matching out
>> spin_lock_irqsave()->spin_unlock_irqrestore() before exiting the
>> function.
>
> That will not work again.
> In case "if (!(hba->outstanding_reqs & pos))" is false,
> i.e if the command for which the abort is issued still exists,
>
> ufshcd_issue_tm_cmd() will be called and if no error returned
> by the function, the code after "out:" will be executed. Which will
> try to "spin_unlock_irqrestore",
> even though no lock has been acquired.
I think you are just trying to disable interrupts and disable
preemtion - with spin_lock_irqsave/spin_unlock_irqsave
if you are looking to acquire lock - then may be
spin_lock_irq/spin_unlock_irq - will work.
>
>>
>> In function:
>> +static int
>> +ufshcd_issue_tm_cmd(struct ufs_hba *hba,
>> + struct ufshcd_lrb *lrbp,
>> + u8 tm_function)
>> ...
>> + spin_lock_irqsave(host->host_lock, flags);
>> +
>> + /* If task management queue is full */
>> + if (!ufshcd_is_tmq_full(hba->outstanding_tasks)) {
>> + dev_err(&hba->pdev->dev, "Task management queue full\n");
>> I think this will map to a Buffered printf - which can potentially
>> result in BUG: Schedule while atomic()...
>> + spin_unlock_irqrestore(host->host_lock, flags);
>> + return err;
>> + }
>> So, it could be like this:
>
> ok, Thanks, I'll look into it.
>
>> + spin_lock_irqsave(host->host_lock, flags);
>> +
>> + /* If task management queue is full */
>> + if (!ufshcd_is_tmq_full(hba->outstanding_tasks)) {
>> + spin_unlock_irqrestore(host->host_lock, flags);
>> + dev_err(&hba->pdev->dev, "Task management queue full\n");
>> + return err;
>> + }
>>
>> Thanks & Regards,
>> Amit Sahrawat
>>
>>
>> On Thu, Feb 2, 2012 at 10:27 AM, Vinayak Holikatti
>> <vinholikatti@xxxxxxxxx> wrote:
>>> From: Santosh Yaraganavi <santoshsy@xxxxxxxxx>
>>>
>>> UFSHCD SCSI error handling includes following implementations,
>>> - Abort task
>>> - Device reset
>>> - Host reset
>>>
>>> Signed-off-by: Santosh Yaraganavi <santoshsy@xxxxxxxxx>
>>> Signed-off-by: Vinayak Holikatti <vinholikatti@xxxxxxxxx>
>>> Reviewed-by: Arnd Bergmann <arnd@xxxxxxxx>
>>> Reviewed-by: Saugata Das <saugata.das@xxxxxxxxxx>
>>> Reviewed-by: Vishak G <vishak.g@xxxxxxxxxxx>
>>> Reviewed-by: Girish K S <girish.shivananjappa@xxxxxxxxxx>
>>> ---
>>> drivers/scsi/ufs/ufshcd.c | 312
>>> +++++++++++++++++++++++++++++++++++++++++++++
>>> 1 files changed, 312 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
>>> index 1bfae84..7589dd1 100644
>>> --- a/drivers/scsi/ufs/ufshcd.c
>>> +++ b/drivers/scsi/ufs/ufshcd.c
>>> @@ -76,6 +76,8 @@
>>> #define NULL 0
>>> #endif /* NULL */
>>>
>>> +#define UFSHCD_MAX_TM_SLOTS 0xFF
>>> +
>>> #define BYTES_TO_DWORDS(p) (p >> 2)
>>> #define UFSHCD_MMIO_BASE (hba->mmio_base)
>>>
>>> @@ -83,6 +85,7 @@ enum {
>>> UFSHCD_MAX_CHANNEL = 1,
>>> UFSHCD_MAX_ID = 1,
>>> UFSHCD_MAX_LUNS = 8,
>>> + UFSHCD_MAX_TM_CMDS = 8,
>>> UFSHCD_CMD_PER_LUN = 16,
>>> UFSHCD_CAN_QUEUE = 32,
>>> BYTES_128 = 128,
>>> @@ -149,6 +152,7 @@ struct uic_command {
>>> * @host: Scsi_Host instance of the driver
>>> * @pdev: PCI device handle
>>> * @lrb: local reference block
>>> + * @outstanding_tasks: Bits representing outstanding task requests
>>> * @outstanding_reqs: Bits representing outstanding transfer requests
>>> * @capabilities: UFS Controller Capabilities
>>> * @nutrs: Transfer Request Queue depth supported by controller
>>> @@ -192,6 +196,7 @@ struct ufs_hba {
>>>
>>> struct ufshcd_lrb *lrb;
>>>
>>> + u32 outstanding_tasks;
>>> u32 outstanding_reqs;
>>>
>>> u32 capabilities;
>>> @@ -200,6 +205,8 @@ struct ufs_hba {
>>> u32 ufs_version;
>>>
>>> struct uic_command active_uic_cmd;
>>> + wait_queue_head_t ufshcd_tm_wait_queue;
>>> + u8 tm_condition[UFSHCD_MAX_TM_CMDS];
>>>
>>> u32 ufshcd_state;
>>> u32 int_enable_mask;
>>> @@ -278,6 +285,30 @@ static inline int ufshcd_get_tr_ocs(struct
>>> ufshcd_lrb *lrbp)
>>> }
>>>
>>> /**
>>> + * ufshcd_get_tmr_ocs - Get the UTMRD Overall Command Status
>>> + * @task_req_descp: pointer to utp_task_req_desc structure
>>> + *
>>> + * This function is used to get the OCS field from UTMRD
>>> + * Returns the OCS field in the UTMRD
>>> + */
>>> +static inline int
>>> +ufshcd_get_tmr_ocs(struct utp_task_req_desc *task_req_descp)
>>> +{
>>> + return task_req_descp->header.dword_2 & MASK_OCS;
>>> +}
>>> +
>>> +/**
>>> + * ufshcd_is_tmq_full - checks if the task management slots are full
>>> + * @outstanding_tasks: contains the task management doorbell value
>>> + *
>>> + * Returns 1 if a free slot available, 0 if task slots are full
>>> + */
>>> +static inline int ufshcd_is_tmq_full(u32 outstanding_tasks)
>>> +{
>>> + return (UFSHCD_MAX_TM_SLOTS == outstanding_tasks) ? 0 : 1;
>>> +}
>>> +
>>> +/**
>>> * ufshcd_get_lists_status - Check UCRDY, UTRLRDY and UTMRLRDY
>>> * @reg: Register value of host controller status
>>> *
>>> @@ -1098,6 +1129,45 @@ static void ufshcd_slave_destroy(struct
>>> scsi_device *sdev)
>>> }
>>>
>>> /**
>>> + * ufshcd_task_req_compl - handle task management request completion
>>> + * @hba: per adapter instance
>>> + * @index: index of the completed request
>>> + *
>>> + * Returns 0 on success, non-zero value on failure
>>> + */
>>> +static int ufshcd_task_req_compl(struct ufs_hba *hba, u32 index)
>>> +{
>>> + struct utp_upiu_task_rsp *task_rsp_upiup;
>>> + struct utp_task_req_desc *task_req_descp;
>>> + unsigned long flags;
>>> + int ocs_value;
>>> + int task_response = -1;
>>> +
>>> + spin_lock_irqsave(hba->host->host_lock, flags);
>>> +
>>> + /* clear completed tasks from outstanding_tasks */
>>> + hba->outstanding_tasks ^= index;
>>> +
>>> + task_req_descp =
>>> + (struct utp_task_req_desc
>>> *)hba->utmrdl_virt_addr_aligned;
>>> + ocs_value = ufshcd_get_tmr_ocs(&task_req_descp[index]);
>>> +
>>> + if (OCS_SUCCESS == ocs_value) {
>>> +
>>> + task_rsp_upiup = (struct utp_upiu_task_rsp *)
>>> + task_req_descp[index].task_rsp_upiu;
>>> + task_response =
>>> be32_to_cpu(task_rsp_upiup->header.dword_1);
>>> + task_response = ((task_response & MASK_TASK_RESPONSE) >>
>>> 8);
>>> +
>>> + /* clear task response */
>>> + memset(task_rsp_upiup, 0, sizeof(struct
>>> utp_upiu_task_rsp));
>>> + }
>>> + spin_unlock_irqrestore(hba->host->host_lock, flags);
>>> +
>>> + return task_response;
>>> +}
>>> +
>>> +/**
>>> * ufshcd_scsi_cmd_status - Update SCSI command result based on scsi
>>> status
>>> * @lrb: pointer to local reference block of completed command
>>> * @scsi_status: SCSI command status
>>> @@ -1314,6 +1384,31 @@ fatal_eh:
>>> }
>>>
>>> /**
>>> + * ufshcd_tmc_handler - handle task management function completion
>>> + * @hba: per adapter instance
>>> + */
>>> +static void ufshcd_tmc_handler(struct ufs_hba *hba)
>>> +{
>>> + u32 completed_reqs;
>>> + u32 tm_doorbell;
>>> + int i;
>>> +
>>> + tm_doorbell = readl(hba->mmio_base + REG_UTP_TASK_REQ_DOOR_BELL);
>>> + completed_reqs = tm_doorbell ^ hba->outstanding_tasks;
>>> + for (i = 0; i < hba->nutmrs; i++) {
>>> + if (completed_reqs & (1 << i)) {
>>> +
>>> + /*
>>> + * Change the default value 0xFF to 0 to indicate
>>> + * completion of respective task management
>>> command
>>> + */
>>> + hba->tm_condition[i] = 0;
>>> +
>>> wake_up_interruptible(&hba->ufshcd_tm_wait_queue);
>>> + }
>>> + }
>>> +}
>>> +
>>> +/**
>>> * ufshcd_sl_intr - Interrupt service routine
>>> * @hba: per adapter instance
>>> * @intr_status: contains interrupts generated by the controller
>>> @@ -1327,6 +1422,9 @@ static void ufshcd_sl_intr(struct ufs_hba *hba, u32
>>> intr_status)
>>> if (intr_status & UIC_COMMAND_COMPL)
>>> schedule_work(&hba->uic_workq);
>>>
>>> + if (intr_status & UTP_TASK_REQ_COMPL)
>>> + ufshcd_tmc_handler(hba);
>>> +
>>> if (intr_status & UTP_TRANSFER_REQ_COMPL)
>>> ufshcd_transfer_req_compl(hba);
>>> }
>>> @@ -1362,6 +1460,209 @@ static irqreturn_t ufshcd_intr(int irq, void
>>> *__hba)
>>> return retval;
>>> }
>>>
>>> +/**
>>> + * ufshcd_issue_tm_cmd - issues task management commands to controller
>>> + * @hba: per adapter instance
>>> + * @lrbp: pointer to local reference block
>>> + *
>>> + * Returns 0 on success, non-zero value on failure
>>> + */
>>> +static int
>>> +ufshcd_issue_tm_cmd(struct ufs_hba *hba,
>>> + struct ufshcd_lrb *lrbp,
>>> + u8 tm_function)
>>> +{
>>> + struct utp_task_req_desc *task_req_descp;
>>> + struct utp_upiu_task_req *task_req_upiup;
>>> + struct Scsi_Host *host;
>>> + unsigned long flags;
>>> + int i;
>>> + int free_slot = 0;
>>> + int err = -1;
>>> +
>>> + host = hba->host;
>>> +
>>> + spin_lock_irqsave(host->host_lock, flags);
>>> +
>>> + /* If task management queue is full */
>>> + if (!ufshcd_is_tmq_full(hba->outstanding_tasks)) {
>>> + dev_err(&hba->pdev->dev, "Task management queue full\n");
>>> + spin_unlock_irqrestore(host->host_lock, flags);
>>> + return err;
>>> + }
>>> +
>>> + for (i = 0; i < hba->nutmrs; i++) {
>>> + if (!(hba->outstanding_tasks & (1 << i))) {
>>> + free_slot = i;
>>> + break;
>>> + }
>>> + }
>>> +
>>> + /* Configure task request descriptor */
>>> + task_req_descp =
>>> + (struct utp_task_req_desc *)
>>> hba->utmrdl_virt_addr_aligned;
>>> + task_req_descp += free_slot;
>>> +
>>> + memset(task_req_descp, 0, sizeof(struct utp_task_req_desc));
>>> + task_req_descp->header.dword_0 =
>>> cpu_to_le32(UTP_REQ_DESC_INT_CMD);
>>> + task_req_descp->header.dword_2 =
>>> + cpu_to_le32(OCS_INVALID_COMMAND_STATUS);
>>> +
>>> + /* Configure task request UPIU */
>>> + task_req_upiup =
>>> + (struct utp_upiu_task_req *)
>>> task_req_descp->task_req_upiu;
>>> + task_req_upiup->header.dword_0 =
>>> + cpu_to_be32(UPIU_HEADER_DWORD(UPIU_TRANSACTION_TASK_REQ,
>>> 0,
>>> + lrbp->lun,
>>> lrbp->task_tag));
>>> + task_req_upiup->header.dword_1 =
>>> + cpu_to_be32(UPIU_HEADER_DWORD(0, tm_function, 0, 0));
>>> +
>>> + task_req_upiup->input_param1 = lrbp->lun;
>>> + task_req_upiup->input_param1 =
>>> + cpu_to_be32(task_req_upiup->input_param1);
>>> + task_req_upiup->input_param2 = lrbp->task_tag;
>>> + task_req_upiup->input_param2 =
>>> + cpu_to_be32(task_req_upiup->input_param2);
>>> +
>>> + /* send command to the controller */
>>> + hba->outstanding_tasks |= (1 << free_slot);
>>> + writel((1 << free_slot),
>>> + (UFSHCD_MMIO_BASE + REG_UTP_TASK_REQ_DOOR_BELL));
>>> +
>>> + spin_unlock_irqrestore(host->host_lock, flags);
>>> +
>>> + /* wait until the task management command is completed */
>>> + err = wait_event_interruptible_timeout(hba->ufshcd_tm_wait_queue,
>>> + (hba->tm_condition[free_slot] !=
>>> 0),
>>> + 60 * HZ);
>>> + hba->tm_condition[free_slot] = 0xFF;
>>> + if (!err) {
>>> + dev_err(&hba->pdev->dev,
>>> + "Task management command timed-out\n");
>>> + return err;
>>> + }
>>> + return ufshcd_task_req_compl(hba, free_slot);
>>> +}
>>> +
>>> +/**
>>> + * ufshcd_device_reset - reset device and abort all the pending commands
>>> + * @cmd: SCSI command pointer
>>> + *
>>> + * Returns 0 on success, non-zero value on failure
>>> + */
>>> +static int ufshcd_device_reset(struct scsi_cmnd *cmd)
>>> +{
>>> + struct Scsi_Host *host;
>>> + struct ufs_hba *hba;
>>> + unsigned long flags;
>>> + int reset_lun;
>>> + unsigned int tag;
>>> + u32 i;
>>> + u32 pos; /* pos: represents a bit in a register */
>>> + int err = -1;
>>> +
>>> + host = cmd->device->host;
>>> + hba = (struct ufs_hba *)host->hostdata;
>>> + tag = cmd->request->tag;
>>> +
>>> + if ((hba->lrb[tag].cmd == cmd))
>>> + reset_lun = hba->lrb[tag].lun;
>>> + else
>>> + goto out;
>>> +
>>> + err = ufshcd_issue_tm_cmd(hba, &hba->lrb[tag],
>>> UFS_LOGICAL_RESET);
>>> + if (!err) {
>>> + spin_lock_irqsave(host->host_lock, flags);
>>> + for (i = 0; i < hba->nutrs; i++) {
>>> + pos = (1 << i);
>>> + if ((pos & hba->outstanding_reqs) &&
>>> + (reset_lun == hba->lrb[i].lun)) {
>>> +
>>> + /* clear the respective UTRLCLR bit */
>>> + writel(~pos, (UFSHCD_MMIO_BASE +
>>> + REG_UTP_TRANSFER_REQ_LIST_CLEAR));
>>> +
>>> + hba->outstanding_reqs ^= pos;
>>> +
>>> + if (hba->lrb[i].cmd) {
>>> + scsi_dma_unmap(hba->lrb[i].cmd);
>>> + hba->lrb[i].cmd->result =
>>> + DID_ABORT << 16;
>>> + hba->lrb[i].cmd->scsi_done(cmd);
>>> + hba->lrb[i].cmd = NULL;
>>> + }
>>> + }
>>> + } /* end of for */
>>> + spin_unlock_irqrestore(host->host_lock, flags);
>>> + } /*end of if */
>>> +out:
>>> + return err;
>>> +}
>>> +
>>> +/**
>>> + * ufshcd_host_reset - Main reset function registered with scsi layer
>>> + * @cmd: SCSI command pointer
>>> + *
>>> + * Returns 0 on success, non-zero value on failure
>>> + */
>>> +static int ufshcd_host_reset(struct scsi_cmnd *cmd)
>>> +{
>>> + struct ufs_hba *hba = NULL;
>>> +
>>> + hba = (struct ufs_hba *) &cmd->device->host->hostdata[0];
>>> +
>>> + if (UFSHCD_STATE_RESET == hba->ufshcd_state)
>>> + return 0;
>>> +
>>> + return ufshcd_do_reset(hba) ? -1 : 0;
>>> +}
>>> +
>>> +/**
>>> + * ufshcd_abort - abort a specific command
>>> + * @cmd: SCSI command pointer
>>> + *
>>> + * Returns 0 on success, non-zero value on failure
>>> + */
>>> +static int ufshcd_abort(struct scsi_cmnd *cmd)
>>> +{
>>> + struct Scsi_Host *host;
>>> + struct ufs_hba *hba;
>>> + unsigned long flags;
>>> + unsigned int tag;
>>> + unsigned int pos;
>>> + int err;
>>> +
>>> + host = cmd->device->host;
>>> + hba = (struct ufs_hba *) host->hostdata;
>>> + tag = cmd->request->tag;
>>> +
>>> + spin_lock_irqsave(host->host_lock, flags);
>>> + pos = (1 << tag);
>>> +
>>> + /* check if command is still pending */
>>> + if (!(hba->outstanding_reqs & pos)) {
>>> + err = -1;
>>> + spin_unlock_irqrestore(host->host_lock, flags);
>>> + goto out;
>>> + }
>>> +
>>> + err = ufshcd_issue_tm_cmd(hba, &hba->lrb[tag], UFS_ABORT_TASK);
>>> + if (!err) {
>>> + spin_lock_irqsave(host->host_lock, flags);
>>> + scsi_dma_unmap(cmd);
>>> +
>>> + /* clear the respective UTRLCLR bit */
>>> + writel(~pos,
>>> + (UFSHCD_MMIO_BASE +
>>> + REG_UTP_TRANSFER_REQ_LIST_CLEAR));
>>> + hba->outstanding_reqs &= ~pos;
>>> + hba->lrb[tag].cmd = NULL;
>>> + spin_unlock_irqrestore(host->host_lock, flags);
>>> + }
>>> +out:
>>> + return err;
>>> +}
>>> +
>>> static struct scsi_host_template ufshcd_driver_template = {
>>> .module = THIS_MODULE,
>>> .name = UFSHCD,
>>> @@ -1369,6 +1670,9 @@ static struct scsi_host_template
>>> ufshcd_driver_template = {
>>> .queuecommand = ufshcd_queuecommand,
>>> .slave_alloc = ufshcd_slave_alloc,
>>> .slave_destroy = ufshcd_slave_destroy,
>>> + .eh_abort_handler = ufshcd_abort,
>>> + .eh_device_reset_handler = ufshcd_device_reset,
>>> + .eh_host_reset_handler = ufshcd_host_reset,
>>> .this_id = -1,
>>> .sg_tablesize = SG_ALL,
>>> .cmd_per_lun = UFSHCD_CMD_PER_LUN,
>>> @@ -1483,6 +1787,7 @@ ufshcd_probe(struct pci_dev *pdev, const struct
>>> pci_device_id *id)
>>> struct Scsi_Host *host;
>>> struct ufs_hba *hba;
>>> int ufs_hba_len;
>>> + int index;
>>> int err;
>>>
>>> ufs_hba_len = sizeof(struct ufs_hba);
>>> @@ -1547,6 +1852,13 @@ ufshcd_probe(struct pci_dev *pdev, const struct
>>> pci_device_id *id)
>>> host->unique_id = host->host_no;
>>> host->max_cmd_len = MAX_CDB_SIZE;
>>>
>>> + /* Initailze wait queue for task management */
>>> + init_waitqueue_head(&hba->ufshcd_tm_wait_queue);
>>> +
>>> + /* Initially assign invalid value to tm_condition */
>>> + for (index = 0; index < hba->nutmrs; index++)
>>> + hba->tm_condition[index] = 0xFF;
>>> +
>>> /* Initialize work queues */
>>> INIT_WORK(&hba->uic_workq, ufshcd_uic_cc_handler);
>>> INIT_WORK(&hba->feh_workq, ufshcd_fatal_err_handler);
>>> --
>>> 1.7.5.4
>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-kernel"
>>> in
>>> the body of a message to majordomo@xxxxxxxxxxxxxxx
>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>> Please read the FAQ at http://www.tux.org/lkml/
>
>
>
> --
> ~Santosh
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/