[PATCH 09/23] mmc: sdhci: fix incorrect get data interrupt during no data transfer
Ritesh Harjani
ritesh.list at gmail.com
Thu May 26 07:59:05 PDT 2016
Hi Adrian,
On Tue, May 17, 2016 at 11:28 AM, Adrian Hunter <adrian.hunter at intel.com> wrote:
> On 17/05/16 07:31, Ritesh Harjani wrote:
>> On Tue, May 10, 2016 at 12:21 PM, Adrian Hunter <adrian.hunter at intel.com> wrote:
>>> On 15/04/16 20:29, Dong Aisheng wrote:
>>>> Current code will report the wrong data interrupt got when no
>>>> data operation in progress assumed by getting !host->data in
>>>> sdhci_data_irq().
>>>>
>>>> For a data command handling process, the driver will call
>>>> sdhci_finish_data() and clear host->data in case any data error,
>>>> then card finish_tasklet will do the rest controller reset work.
>>>>
>>>> Before the tasklet got run, however, controllers may report the
>>>> TC(Transfer Complete) interrupt (SDHCI_INT_DATA_END) a bit later
>>>> than data CRC error and data end bit error interrupts for single
>>>> block transfer or the last block of multiblock transfer.
>>
>> What if the controller generates Transfer complete interrupt bit more slower
>> and the tasklet do get a chance to run. In that case even host->mrq
>> will become NULL.
>> Actually if that happens it's even more worse since the driver may
>> start issuing other commands to controller.
>>
>> Are we covering that case here? Do you think that this case is possible?
>
> After data error, the host driver resets the controller so there will be no
> more interrupts relating to that mrq.
Resetting will happen in sdhci_tasklet_finish.
My case was -
1. Controller raised a Data CRC error.
2. sdhci_irq made host->data = NULL in sdhci_finish_data and scheduled
the finish_tasklet.
3. while the tasklet is running and before resetting CMD and Data
machine, TC interrupt is raised by controller.
4. sdhci_irq will call for sdhci_data_irq.
But I get it that the second interrupt will end up printing the same
error since host->data will be found as NULL in sdhci_data_irq and
will return.
So below patch should do the job :)
>
> This patch is avoiding unnecessary warnings until that happens.
Yes.
>
>>
>>
>>>>
>>>> Controller usually detects and generates data CRC/end bit error
>>>> interrupts once one block on the bus is transferred completely.
>>>> For single block transfer, since there's only one bock to transfer,
>>>> the controller will report transfer complete interrupt as well,
>>>> but until the data in controller FIFO has been successfully
>>>> transferred to memory. The time gap of TC and CRC interrupt depends on
>>>> the system busy state at that point and memory bus access speed.
>>>>
>>>> So it is possible when TC interrupt generated, host->data is already
>>>> equal to NULL due to cleared by former CRC/Data End Bit error which
>>>> is reasonable.
>>>>
>>>> Thus we DO NOT report the weird data interrupt event for this case.
>>>>
>>>> Else we may easily see warning below during SD3.0 card manually tuning
>>>> process (calling mmc_send_tuning() which is a single block transfer)
>>>> mmc0: Got data interrupt 0x00000002 even though no data operation was in progress
>>>>
>>>> The detailed command log is as follows:
>>>> [ 1657.920983] mmc0: starting CMD19 arg 00000000 flags 00000035
>>>> [ 1657.921009] mmc0: blksz 64 blocks 1 flags 00000200 tsac 150 ms nsac 0
>>>> [ 1657.921085] sdhci [sdhci_irq()]: *** mmc0 got interrupt: 0x00200001
>>>> [ 1657.921112] sdhci [sdhci_irq()]: *** mmc0 got interrupt: 0x00000002
>>>> [ 1657.921131] mmc0: Got data interrupt 0x00000002 even though no data operation was in progress.
>>>> [ 1657.929761] sdhci: =========== REGISTER DUMP (mmc0)===========
>>>> [ 1657.929780] sdhci: Sys addr: 0x3d5d6380 | Version: 0x00000002
>>>> [ 1657.929796] sdhci: Blk size: 0x00000040 | Blk cnt: 0x00000001
>>>> [ 1657.929814] sdhci: Argument: 0x00000000 | Trn mode: 0x00000013
>>>> [ 1657.929831] sdhci: Present: 0x01fd8008 | Host ctl: 0x00000023
>>>> [ 1657.929847] sdhci: Power: 0x00000002 | Blk gap: 0x00000080
>>>> [ 1657.929863] sdhci: Wake-up: 0x00000008 | Clock: 0x0000000f
>>>> [ 1657.929879] sdhci: Timeout: 0x0000000f | Int stat: 0x00000000
>>>> [ 1657.929896] sdhci: Int enab: 0x107f008b | Sig enab: 0x107f008b
>>>> [ 1657.929914] sdhci: AC12 err: 0x00000000 | Slot int: 0x00000003
>>>> [ 1657.929932] sdhci: Caps: 0x07eb0000 | Caps_1: 0x00002007
>>>> [ 1657.929949] sdhci: Cmd: 0x0000133a | Max curr: 0x00ffffff
>>>> [ 1657.929965] sdhci: Host ctl2: 0x000000c8
>>>> [ 1657.929981] sdhci: ADMA Err: 0x00000000 | ADMA Ptr: 0x8f042208
>>>> [ 1657.929995] sdhci: ===========================================
>>>> [ 1657.930156] mmc0: req done (CMD19): 0: 00000900 00000000 00000000 00000000
>>>> [ 1657.930179] mmc0: 0 bytes transferred: -84
>>>>
>>>> It shows we first have a data CRC error interrupt then a data transfer
>>>> complete interrupt.
>>>> Then we got the !host->data case in sdhci_data_irq().
>>>>
>>>> CC: stable <stable at vger.kernel.org>
>>>> Signed-off-by: Dong Aisheng <aisheng.dong at nxp.com>
>>>> ---
>>>> drivers/mmc/host/sdhci.c | 11 +++++++++++
>>>> 1 file changed, 11 insertions(+)
>>>>
>>>> diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c
>>>> index 40e3551..2eb0e34 100644
>>>> --- a/drivers/mmc/host/sdhci.c
>>>> +++ b/drivers/mmc/host/sdhci.c
>>>> @@ -2325,6 +2325,17 @@ static void sdhci_data_irq(struct sdhci_host *host, u32 intmask)
>>>> }
>>>> }
>>>>
>>>> + /*
>>>> + * The "data complete" interrupt is possible to happen a bit
>>>> + * later than CRC error and data end bit error interrupts
>>>> + * separately for single block transfer or the last block of
>>>> + * multiblock transfer. For this case, we DO NOT report the
>>>> + * weird data interrupt event.
>>>> + */
>>>> + if ((intmask & SDHCI_INT_DATA_END) &&
>>>> + (host->mrq && host->mrq->data && host->mrq->data->error))
>>>> + return;
>>>
>>> This could be generalized a bit more i.e. what about:
>>>
>>> /*
>>> * After an error and before the the data circuit is reset in
>>> * sdhci_tasklet_finish, we could get more interrupts, but we
>>> * already have an error, so ignore them.
>>> */
>>> if (host->mrq && host->mrq->data && host->mrq->data->error)
>>> return;
>>>
>>>> +
>>>> pr_err("%s: Got data interrupt 0x%08x even though no data operation was in progress.\n",
>>>> mmc_hostname(host->mmc), (unsigned)intmask);
>>>> sdhci_dumpregs(host);
>>>>
>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
>>> the body of a message to majordomo at vger.kernel.org
>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>
>
More information about the linux-arm-kernel
mailing list