Re: [PATCH 3/6] i2c: qup: Add bam dma capabilities

[Sricharan R <sricharan@codeaurora.org>]

Hi Ivan,

On 03/25/2015 06:40 PM, Ivan T. Ivanov wrote:
>
> Hi Sricharan,
>
> On Fri, 2015-03-13 at 23:19 +0530, Sricharan R wrote:
>
>
>>   #define QUP_I2C_MASTER_GEN     0x408
>> +#define QUP_I2C_MASTER_CONFIG  0x408
>>
>
> Unused.
>
Ok, will remove it
>>   #define QUP_READ_LIMIT                 256
>> +#define MX_TX_RX_LEN                   SZ_64K
>> +#define MX_BLOCKS                      (MX_TX_RX_LEN / QUP_READ_LIMIT)
>> +
>> +#define TOUT_MAX                       300 /* Max timeout for 32k tx/tx */
>>
>
> seconds, muliseconds?
>
  Ok. Will add milliseconds.
>>   struct qup_i2c_config {
>>          int tag_ver;
>>          int max_freq;
>>   };
>>
>> +struct tag {
>
> Please use consistent naming convention.
>
ok.
>> +       u8 *start;
>> +       dma_addr_t addr;
>> +};
>> +
>>   struct qup_i2c_dev {
>>          struct device*dev;
>>          void __iomem*base;
>> @@ -157,9 +181,35 @@ struct qup_i2c_dev {
>>          /* QUP core errors */
>>          u32     qup_err;
>>
>> +       /* dma parameters */
>> +       bool    is_dma;
>> +       struct  dma_pool *dpool;
>> +       struct  tag start_tag;
>> +       struct  tag scratch_tag;
>> +       struct  tag footer_tag;
>> +       struct  dma_chan *dma_tx;
>> +       struct  dma_chan *dma_rx;
>> +       struct  scatterlist *sg_tx;
>> +       struct  scatterlist *sg_rx;
>> +       dma_addr_tsg_tx_phys;
>> +       dma_addr_tsg_rx_phys;
>
> Maybe these could be organized in structure per direction.
>
ok, will group it.
>> +
>>          struct completionxfer;
>>   };
>>
>> +struct i2c_bam_xfer {
>
> Unused.
>
  Right. Will remove the whole thing. Thanks.
>> +       struct qup_i2c_dev *qup;
>> +       u32 start_len;
>> +
>> +       u32 rx_nents;
>> +       u32 tx_nents;
>> +
>> +       struct dma_async_tx_descriptor *tx_desc;
>> +       dma_cookie_t tx_cookie;
>> +       struct dma_async_tx_descriptor *rx_desc;
>> +       dma_cookie_t rx_cookie;
>
> structure per direction.
>
>> +};
>> +
>>
  Infact should be removed.
>
>> +static void bam_i2c_callback(void *data)
>> +{
>
> Please use consistent naming, here and bellow.
>
ok, will change.
>> +       struct qup_i2c_dev *qup = data;
>> +
>> +       complete(&qup->xfer);
>> +}
>> +
>> +static int get_start_tag(u8 *tag, struct i2c_msg *msg, int first, int last,
>> +                       int blen)
>> +{
>> +       u8 addr = (msg->addr << 1) | ((msg->flags & I2C_M_RD) == I2C_M_RD);
>> +       u8 op;
>> +       int len = 0;
>> +
>> +       /* always issue stop for each read block */
>> +       if (last) {
>> +               if (msg->flags & I2C_M_RD)
>> +                       op = QUP_TAG_V2_DATARD_STOP;
>> +               else
>> +                       op = QUP_TAG_V2_DATAWR_STOP;
>> +       } else {
>> +               if (msg->flags & I2C_M_RD)
>> +                       op = QUP_TAG_V2_DATARD;
>> +               else
>> +                       op = QUP_TAG_V2_DATAWR;
>> +       }
>> +
>> +       if (msg->flags & I2C_M_TEN) {
>> +               len += 5;
>> +               tag[0] = QUP_TAG_V2_START;
>> +               tag[1] = addr;
>> +               tag[2] = op;
>> +               tag[3] = blen;
>> +
>> +               if (msg->flags & I2C_M_RD && last) {
>> +                       len += 2;
>> +                       tag[4] = QUP_BAM_INPUT_EOT;
>> +                       tag[5] = QUP_BAM_FLUSH_STOP;
>> +               }
>> +       } else {
>> +               if (first) {
>> +                       tag[len++] = QUP_TAG_V2_START;
>> +                       tag[len++] = addr;
>> +               }
>> +
>> +               tag[len++] = op;
>> +               tag[len++] = blen;
>> +
>> +               if (msg->flags & I2C_M_RD & last) {
>> +                       tag[len++] = QUP_BAM_INPUT_EOT;
>> +                       tag[len++] = QUP_BAM_FLUSH_STOP;
>> +               }
>> +       }
>> +
>> +       return len;
>> +}
>
> Maybe could be split to 2 functions?
>
hmm, ok will split couple of small ones.
>> -static const struct i2c_algorithm qup_i2c_algo = {
>> +static struct i2c_algorithm qup_i2c_algo = {
>
> Why?
>
  oops. Should not have been changed. Will fix.
>>          .master_xfer= qup_i2c_xfer,
>>          .functionality= qup_i2c_func,
>>   };
>> @@ -839,6 +1136,7 @@ static int qup_i2c_probe(struct platform_device *pdev)
>>          u32 clk_freq = 100000;
>>          const struct qup_i2c_config *config;
>>          const struct of_device_id *of_id;
>> +       int blocks;
>>
>>          qup = devm_kzalloc(&pdev->dev, sizeof(*qup), GFP_KERNEL);
>>          if (!qup)
>> @@ -875,6 +1173,53 @@ static int qup_i2c_probe(struct platform_device *pdev)
>>                  return qup->irq;
>>          }
>>
>> +       if (of_device_is_compatible(pdev->dev.of_node, "qcom,i2c-qup-v2.1.1") ||
>> +               of_device_is_compatible(pdev->dev.of_node,
>> +                       "qcom,i2c-qup-v2.2.1")) {
>
> Logic will be simpler if you check just for version 1 of the controller.
  yes. Will fix it.
>
>> +               qup->dma_rx = dma_request_slave_channel(&pdev->dev, "bam-rx");
>> +
>
> Please use dma_request_slave_channel_reason.
>
> As Andy noted, please use just "rx", "tx"
>
ok. will change it.
>> +               if (!qup->dma_rx)
>> +                       return -EPROBE_DEFER;
>
> Don't mask other errors, here and bellow. DMA support should be optional.
>
  Ok, will fix here.
>>                  dev_err(qup->dev, "Could not get core clock\n");
>> @@ -989,6 +1334,14 @@ static int qup_i2c_remove(struct platform_device *pdev)
>>   {
>>          struct qup_i2c_dev *qup = platform_get_drvdata(pdev);
>>
>> +       dma_pool_free(qup->dpool, qup->start_tag.start,
>> +                                                               qup->start_tag.addr);
>> +       dma_pool_free(qup->dpool, qup->scratch_tag.start,
>> +                                                               qup->scratch_tag.addr);
>> +       dma_pool_free(qup->dpool, qup->footer_tag.start,
>> +                                                               qup->footer_tag.addr);
>> +       dma_pool_destroy(qup->dpool);
>
> Only if is_dma, right?
>
Right. Will add the check.
> DMA mapping code omitted from review. I don't understand it well enough, sorry.
>
   Ok, thanks for reviewing the rest.

Regards,
  Sricharan

QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a 
member of Code Aurora Forum, hosted by The Linux Foundation