On 31/05/24 13:31, Victor Shih wrote:
> On Fri, May 24, 2024 at 2:54 PM Adrian Hunter <adrian.hunter@xxxxxxxxx> wrote:
>>
>> On 22/05/24 14:08, Victor Shih wrote:
>>> From: Victor Shih <victor.shih@xxxxxxxxxxxxxxxxxxx>
>>>
>>> Add UHS-II Auto Command Error Recovery functionality
>>> into the MMC request processing flow.
>>
>> Not sure what "auto" means here, but the commit message
>> should outline what the spec. requires for error recovery.
>>
>
> Hi, Adrian
>
> I will add instructions in the v17 version.
>
> Thanks, Victor Shih
>
>>>
>>> Signed-off-by: Ben Chuang <ben.chuang@xxxxxxxxxxxxxxxxxxx>
>>> Signed-off-by: Victor Shih <victor.shih@xxxxxxxxxxxxxxxxxxx>
>>> ---
>>>
>>> Updates in V16:
>>> - Separate the Error Recovery mechanism from patch#7 to patch#8.
>>>
>>> ---
>>>
>>> drivers/mmc/core/core.c | 4 ++
>>> drivers/mmc/core/core.h | 1 +
>>> drivers/mmc/core/sd_uhs2.c | 80 ++++++++++++++++++++++++++++++++++++++
>>> include/linux/mmc/host.h | 6 +++
>>> 4 files changed, 91 insertions(+)
>>>
>>> diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
>>> index 68496c51a521..18642afc405f 100644
>>> --- a/drivers/mmc/core/core.c
>>> +++ b/drivers/mmc/core/core.c
>>> @@ -403,6 +403,10 @@ void mmc_wait_for_req_done(struct mmc_host *host, struct mmc_request *mrq)
>>> while (1) {
>>> wait_for_completion(&mrq->completion);
>>>
>>> + if (host->ops->get_cd(host))
>>> + if (mrq->cmd->error || (mrq->data && mrq->data->error))
>>> + mmc_sd_uhs2_error_recovery(host, mrq);
>>
>> There are several issues with this:
>>
>> 1. It is not OK to start a request from within the request path
>> because it is recursive:
>>
>> mmc_wait_for_req_done() <--
>> mmc_sd_uhs2_error_recovery()
>> sd_uhs2_abort_trans()
>> mmc_wait_for_cmd()
>> mmc_wait_for_req()
>> mmc_wait_for_req_done() <--
>>
>> 2. The mmc block driver does not use this path
>>
>> 3. No need to always call ->get_cd() if there is no error
>>
>> It is worth considering whether the host controller could
>> send the abort command as part of the original request, as
>> is done with the stop command.
>>
>
> Hi, Adrian
>
> 1. It looks like just issuing a command in
> mmc_wait_for_req_done() will cause a recursion.
> I will drop sd_uhs2_abort_trans() and
> sd_uhs2_abort_status_read() in the v17 version.
> 2. I have no idea about this part, could you please give me some advice?
The mmc block driver sets the ->done() callback and so
mmc_wait_for_req_done() is never called for data transfers.
That won't matter if the host controller handles doing
the abort command, as was suggested elsewhere.
> 3. I will try to modify this part in the v17 version.
>
> Thanks, Victor Shih
>
>>> +
>>> cmd = mrq->cmd;
>>>
>>> if (!cmd->error || !cmd->retries ||
>>> diff --git a/drivers/mmc/core/core.h b/drivers/mmc/core/core.h
>>> index 920323faa834..259d47c8bb19 100644
>>> --- a/drivers/mmc/core/core.h
>>> +++ b/drivers/mmc/core/core.h
>>> @@ -82,6 +82,7 @@ int mmc_attach_mmc(struct mmc_host *host);
>>> int mmc_attach_sd(struct mmc_host *host);
>>> int mmc_attach_sdio(struct mmc_host *host);
>>> int mmc_attach_sd_uhs2(struct mmc_host *host);
>>> +void mmc_sd_uhs2_error_recovery(struct mmc_host *mmc, struct mmc_request *mrq);
>>>
>>> /* Module parameters */
>>> extern bool use_spi_crc;
>>> diff --git a/drivers/mmc/core/sd_uhs2.c b/drivers/mmc/core/sd_uhs2.c
>>> index 85939a2582dc..d5acb4e6ccac 100644
>>> --- a/drivers/mmc/core/sd_uhs2.c
>>> +++ b/drivers/mmc/core/sd_uhs2.c
>>> @@ -1324,3 +1324,83 @@ int mmc_attach_sd_uhs2(struct mmc_host *host)
>>>
>>> return err;
>>> }
>>> +
>>> +static void sd_uhs2_abort_trans(struct mmc_host *mmc)
>>> +{
>>> + struct mmc_request mrq = {};
>>> + struct mmc_command cmd = {0};
>>> + struct uhs2_command uhs2_cmd = {};
>>> + int err;
>>> +
>>> + mrq.cmd = &cmd;
>>> + mmc->ongoing_mrq = &mrq;
>>> +
>>> + uhs2_cmd.header = UHS2_NATIVE_PACKET | UHS2_PACKET_TYPE_CCMD |
>>> + mmc->card->uhs2_config.node_id;
>>> + uhs2_cmd.arg = ((UHS2_DEV_CMD_TRANS_ABORT & 0xFF) << 8) |
>>> + UHS2_NATIVE_CMD_WRITE |
>>> + (UHS2_DEV_CMD_TRANS_ABORT >> 8);
>>> +
>>> + sd_uhs2_cmd_assemble(&cmd, &uhs2_cmd, 0, 0);
>>> + err = mmc_wait_for_cmd(mmc, &cmd, 0);
>>> +
>>> + if (err)
>>> + pr_err("%s: %s: UHS2 CMD send fail, err= 0x%x!\n",
>>> + mmc_hostname(mmc), __func__, err);
>>> +}
>>> +
>>> +static void sd_uhs2_abort_status_read(struct mmc_host *mmc)
>>> +{
>>> + struct mmc_request mrq = {};
>>> + struct mmc_command cmd = {0};
>>> + struct uhs2_command uhs2_cmd = {};
>>> + int err;
>>> +
>>> + mrq.cmd = &cmd;
>>> + mmc->ongoing_mrq = &mrq;
>>> +
>>> + uhs2_cmd.header = UHS2_NATIVE_PACKET |
>>> + UHS2_PACKET_TYPE_CCMD |
>>> + mmc->card->uhs2_config.node_id;
>>> + uhs2_cmd.arg = ((UHS2_DEV_STATUS_REG & 0xFF) << 8) |
>>> + UHS2_NATIVE_CMD_READ |
>>> + UHS2_NATIVE_CMD_PLEN_4B |
>>> + (UHS2_DEV_STATUS_REG >> 8);
>>> +
>>> + sd_uhs2_cmd_assemble(&cmd, &uhs2_cmd, 0, 0);
>>> + err = mmc_wait_for_cmd(mmc, &cmd, 0);
>>> +
>>> + if (err)
>>> + pr_err("%s: %s: UHS2 CMD send fail, err= 0x%x!\n",
>>> + mmc_hostname(mmc), __func__, err);
>>> +}
>>> +
>>> +void mmc_sd_uhs2_error_recovery(struct mmc_host *mmc, struct mmc_request *mrq)
>>> +{
>>> + mmc->ops->uhs2_reset_cmd_data(mmc);
>>
>> The host controller should already have done any resets needed.
>> sdhci already has support for doing that - see host->pending_reset
>>
>
> Hi, Adrian
>
> I'm not sure what this means. Could you please give me more information?
sdhci_uhs2_request_done() checks sdhci_needs_reset() and does
sdhci_uhs2_reset().
sdhci_needs_reset() does not cater for data errors because
the reset for data errors is done directly in what becomes
__sdhci_finish_data_common().
You may need to:
1. add a parameter to __sdhci_finish_data_common() to
skip doing the sdhci reset and instead set
host->pending_reset
2. amend sdhci_uhs2_request_done() to check for data error
also to decide if a reset is needed
>
> Thanks, Victor Shih
>
>>> +
>>> + if (mrq->data) {
>>> + if (mrq->data->error && mmc_card_uhs2(mmc)) {
>>> + if (mrq->cmd) {
>>> + switch (mrq->cmd->error) {
>>> + case ETIMEDOUT:
>>> + case EILSEQ:
>>> + case EIO:
>>> + sd_uhs2_abort_trans(mmc);
>>> + sd_uhs2_abort_status_read(mmc);
>>
>> What is the purpose of sd_uhs2_abort_status_read() here?
>> It is not obvious it does anything.
>>
>
> Hi, Adrian
>
> sd_uhs2_abort_status_read() seems to only have read status,
> I will drop this in the v17 version.
>
> Thanks, Victor Shih
>
>>> + break;
>>> + default:
>>> + break;
>>> + }
>>> + }
>>> + }
>>> + } else {
>>> + if (mrq->cmd) {
>>> + switch (mrq->cmd->error) {
>>> + case ETIMEDOUT:
>>> + sd_uhs2_abort_trans(mmc);
>>> + break;
>>> + }
>>> + }
>>> + }
>>> +}
>>> diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
>>> index fc9520b3bfa4..c914a58f7e1e 100644
>>> --- a/include/linux/mmc/host.h
>>> +++ b/include/linux/mmc/host.h
>>> @@ -271,6 +271,12 @@ struct mmc_host_ops {
>>> * negative errno in case of a failure or zero for success.
>>> */
>>> int (*uhs2_control)(struct mmc_host *host, enum sd_uhs2_operation op);
>>> +
>>> + /*
>>> + * The uhs2_reset_cmd_data callback is used to excute reset
>>> + * when a auto command error occurs.
>>> + */
>>> + void (*uhs2_reset_cmd_data)(struct mmc_host *host);
>>> };
>>>
>>> struct mmc_cqe_ops {
>>
