Re: [PATCH v3 2/7] mmc: sdhci: add support for using external DMA devices

From: Faiz Abbas
Date: Mon Dec 23 2019 - 09:24:13 EST


Hi Adrian,

On 16/12/19 7:15 pm, Adrian Hunter wrote:
> On 16/12/19 10:27 am, Faiz Abbas wrote:
>> Hi Adrian,
>>
>> On 12/12/19 6:25 pm, Adrian Hunter wrote:
>>> On 10/12/19 11:51 am, Faiz Abbas wrote:
>>>> From: Chunyan Zhang <zhang.chunyan@xxxxxxxxxx>
>>>>
>>>> Some standard SD host controllers can support both external dma
>>>> controllers as well as ADMA/SDMA in which the SD host controller
>>>> acts as DMA master. TI's omap controller is the case as an example.
>>>>
>>>> Currently the generic SDHCI code supports ADMA/SDMA integrated in
>>>> the host controller but does not have any support for external DMA
>>>> controllers implemented using dmaengine, meaning that custom code is
>>>> needed for any systems that use an external DMA controller with SDHCI.
>>>>
>>>> Fixes by Faiz Abbas <faiz_abbas@xxxxxx>:
>>>> 1. Map scatterlists before dmaengine_prep_slave_sg()
>>>> 2. Use dma_async() functions inside of the send_command() path and call
>>>> terminate_sync() in non-atomic context in case of an error.
>>>>
>>>> @@ -2652,6 +2845,18 @@ static bool sdhci_request_done(struct sdhci_host *host)
>>>> if (host->flags & SDHCI_REQ_USE_DMA) {
>>>> struct mmc_data *data = mrq->data;
>>>>
>>>> + spin_unlock_irqrestore(&host->lock, flags);
>>>> +
>>>> + /* Terminate and synchronize dma in case of an error */
>>>> + if (data && (mrq->cmd->error || data->error) &&
>>>> + host->use_external_dma) {
>>>> + struct dma_chan *chan = sdhci_external_dma_channel(host,
>>>> + data);
>>>> + dmaengine_terminate_sync(chan);
>>>> + }
>>>> +
>>>> + spin_lock_irqsave(&host->lock, flags);
>>>> +
>>>
>>> Need to take the mrq out of mrqs_done[] to ensure it is not processed again,
>>> and put it back again to be consistent with the remaining code. Also put
>>> host->use_external_dma as the first condition i.e.
>>>
>>> if (host->use_external_dma && data &&
>>> (mrq->cmd->error || data->error)) {
>>> struct dma_chan *chan = sdhci_external_dma_channel(host, data);
>>>
>>> host->mrqs_done[i] = NULL;
>>> spin_unlock_irqrestore(&host->lock, flags);
>>> dmaengine_terminate_sync(chan);
>>> spin_lock_irqsave(&host->lock, flags);
>>> sdhci_set_mrq_done(host, mrq);
>>> }
>>>
>>> where sdhci_set_mrq_done() is factored out from __sdhci_finish_mrq() i.e.
>>>
>>> static void sdhci_set_mrq_done(struct sdhci_host *host, struct mmc_request *mrq)
>>> {
>>> int i;
>>>
>>> for (i = 0; i < SDHCI_MAX_MRQS; i++) {
>>> if (host->mrqs_done[i] == mrq) {
>>> WARN_ON(1);
>>> return;
>>> }
>>> }
>>>
>>> for (i = 0; i < SDHCI_MAX_MRQS; i++) {
>>> if (!host->mrqs_done[i]) {
>>> host->mrqs_done[i] = mrq;
>>> break;
>>> }
>>> }
>>>
>>> WARN_ON(i >= SDHCI_MAX_MRQS);
>>> }
>>>
>>> sdhci_set_mrq_done() can be made in the refactoring patch.
>> Haven't we already done the sdhci_set_mrq_done() part in
>> __sdhci_finish_mrq()?
>>
>> We are picking up an already "done" mrq, looking at whether it had any
>> error and then sychronizing with external dma. Or at least that is my
>> understanding.
>
> sdhci supports having 2 requests (1 data, 1 cmd) at a time, so there is an
> error case where 1 request will wait for the 2nd request before doing a
> reset. That logic is further down in sdhci_request_done() so you have to
> put the mrq back into host->mrqs_done[] to make it work.

Sorry for the late response. I had to spend some time figuring out how
the mrqs_done handling works. Will add the new function above.

>
>>
>>>
>>>> if (data && data->host_cookie == COOKIE_MAPPED) {
>>>> if (host->bounce_buffer) {
>>>> /*
>>>> @@ -3758,12 +3963,28 @@ int sdhci_setup_host(struct sdhci_host *host)
>>>> mmc_hostname(mmc), host->version);
>>>> }
>>>>
>>>> - if (host->quirks & SDHCI_QUIRK_FORCE_DMA)
>>>> + if (host->use_external_dma) {
>>>> + ret = sdhci_external_dma_init(host);
>>>> + if (ret == -EPROBE_DEFER)
>>>> + goto unreg;
>>>> +
>>>> + /*
>>>> + * Fall back to use the DMA/PIO integrated in standard SDHCI
>>>> + * instead of external DMA devices.
>>>> + */
>>>> + if (ret)
>>>> + sdhci_switch_external_dma(host, false);
>>>> + }
>>>> +
>>>> + if (host->quirks & SDHCI_QUIRK_FORCE_DMA) {
>>>> host->flags |= SDHCI_USE_SDMA;
>>>> - else if (!(host->caps & SDHCI_CAN_DO_SDMA))
>>>> + } else if (!(host->caps & SDHCI_CAN_DO_SDMA)) {
>>>> DBG("Controller doesn't have SDMA capability\n");
>>>> - else
>>>> + } else if (host->use_external_dma) {
>>>> + /* Using dma-names to detect external dma capability */
>>>
>>> What is this change for? Do you expect for SDHCI_USE_SDMA and
>>> SDHCI_USE_ADMA flags to be clear?
>>
>> Yes. Today the code enables SDMA by default (in the else part below
>> this). I want it to not enable SDMA in the external dma case.
>
> What about moving the "if (host->use_external_dma) {" clause and explicitly
> clearing SDHCI_USE_SDMA and SDHCI_USE_ADMA?
>

I am ok with this as well. Sending a new version.

Thanks,
Faiz