Re: [RFC PATCH v5 3/3] SPI: Add virtio SPI driver.

Hi Mukesh,

Thank you for reviewing the patch and providing your feedback. Really
appreciate your detailed suggestions.

On 6/30/2025 2:54 PM, Mukesh Kumar Savaliya wrote:
> Hi, Haixu, Thanks !
> 
> On 6/20/2025 9:42 AM, Haixu Cui wrote:
>> This is the virtio SPI Linux kernel driver.
>>
>> Signed-off-by: Haixu Cui <[email protected]>
>> ---
>>   MAINTAINERS              |   6 +
>>   drivers/spi/Kconfig      |  11 +
>>   drivers/spi/Makefile     |   1 +
>>   drivers/spi/spi-virtio.c | 444 +++++++++++++++++++++++++++++++++++++++
>>   4 files changed, 462 insertions(+)
>>   create mode 100644 drivers/spi/spi-virtio.c
>>

>> + * So the corresponding relationship:
>> + * A   <===> cs_setup_ns (after CS asserted)
> And "before clock start" ? to be added in bracket as comment.

Here I refer to the cs_setup definition in include/linux/spi/spi.h,
where the cs_setup is described only in terms of delay after CS is
asserted, without referencing the clock signal.

> 
>> +
>> +static int virtio_spi_transfer_one(struct spi_controller *ctrl,
>> +                   struct spi_device *spi,
>> +                   struct spi_transfer *xfer)
>> +{
> 
> [...]
> 
>> +    wait_for_completion(&priv->spi_req.completion);
>> +
> I see init_completion(spi_req.completion) is called during probe() but 
> successive transfer doent have reinit_completion(spi_req.completion). 
> wondering how is this working for back to back transfers.

In current implementation, each SPI transfer uses the same spi_req
instance, which is a member of struct virtio_spi_priv.

I'm considering removing spi_req from the virtio_spi_priv structure,
instead dynamically allocation a new spi_req for each transfer. This
way, each transfer would have its own completion object, so we could
simply call init_completion() without worrying about reinitializing a
shared one. I believe this would make the design cleaner and avoid
potential issues.

Is this approach okay with you? If so I will update the patch
accordingly in the next revision.

>> +    /* Read result from message and translate return code */
>> +    switch (priv->spi_req.result.result) {
>> +    case VIRTIO_SPI_TRANS_OK:
>> +        /* ret is 0 */
>> +        break;
>> +    case VIRTIO_SPI_PARAM_ERR:
>> +        ret = -EINVAL;
>> +        break;
>> +    case VIRTIO_SPI_TRANS_ERR:
>> +        ret = -EIO;
>> +        break;
>> +    default: /* Protocol violation */
> Comment in new line ? following same method across.

This comment seems not particularly helpful. I’ll remove it to keep
the code clean.

>> +static void virtio_spi_remove(struct virtio_device *vdev)
>> +{
>> +    struct spi_controller *ctrl = dev_get_drvdata(&vdev->dev);
>> +
>> +    /* Order: 1.) unregister controller, 2.) remove virtqueue */
> Is this a specific flow for virtio OR generic ? if its generic, we can 
> remove the comments.

This is generic actually, I will remove it to keep code clean.
>> +    spi_unregister_controller(ctrl);
>> +    virtio_spi_del_vq(vdev);
>> +}
>> +

Thanks
Haixu Cui