spi-imx: wait_for_completion should timeout even in non-DMA transfer cases

spi-imx: wait_for_completion should timeout even in non-DMA transfer cases

[Jean-Michel Hautbois]

Hi,

I am wondering why in spi-imx the spi_imx_pio_transfer() function is
calling wait_for_completion() and not wait_for_completion_timeout() as
in the spi_imx_dma_transfer() one.
I can't see a good reason for this, maybe should it be calculated
based on the spi clock and transfer->len, or at least be the same as
IMX_DMA_TIMEOUT which is 3 seconds ?
If you want it, I have a patch, just tell me if you are interested and
if you prefer a calculated timeout or even something else ?

Thanks,
JM

Re: spi-imx: wait_for_completion should timeout even in non-DMA transfer cases

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 626 bytes --]

On Fri, Sep 18, 2015 at 02:38:27PM +0200, Jean-Michel Hautbois wrote:
> Hi,
> 
> I am wondering why in spi-imx the spi_imx_pio_transfer() function is
> calling wait_for_completion() and not wait_for_completion_timeout() as
> in the spi_imx_dma_transfer() one.
> I can't see a good reason for this, maybe should it be calculated
> based on the spi clock and transfer->len, or at least be the same as
> IMX_DMA_TIMEOUT which is 3 seconds ?
> If you want it, I have a patch, just tell me if you are interested and
> if you prefer a calculated timeout or even something else ?

A calculated timeout is probably best.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]

Re: spi-imx: wait_for_completion should timeout even in non-DMA transfer cases

[Jean-Michel Hautbois]

2015-09-19 19:56 GMT+02:00 Mark Brown <broonie@kernel.org>:
> On Fri, Sep 18, 2015 at 02:38:27PM +0200, Jean-Michel Hautbois wrote:
>> Hi,
>>
>> I am wondering why in spi-imx the spi_imx_pio_transfer() function is
>> calling wait_for_completion() and not wait_for_completion_timeout() as
>> in the spi_imx_dma_transfer() one.
>> I can't see a good reason for this, maybe should it be calculated
>> based on the spi clock and transfer->len, or at least be the same as
>> IMX_DMA_TIMEOUT which is 3 seconds ?
>> If you want it, I have a patch, just tell me if you are interested and
>> if you prefer a calculated timeout or even something else ?
>
> A calculated timeout is probably best.

Well, it depends...
If you have a 20MHz clock, then you can transfer a byte in 4µs.
As most of the time pio_transfer is used for 8 to 32 bytes max, it
would be irrelevant to use such a small timeout.
I can obviously do it, but not sure if using a small (say, 100ms)
timeout would not be better.

JM

Re: spi-imx: wait_for_completion should timeout even in non-DMA transfer cases

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 651 bytes --]

On Mon, Sep 21, 2015 at 11:08:55AM +0200, Jean-Michel Hautbois wrote:
> 2015-09-19 19:56 GMT+02:00 Mark Brown <broonie@kernel.org>:

> > A calculated timeout is probably best.

> Well, it depends...
> If you have a 20MHz clock, then you can transfer a byte in 4µs.
> As most of the time pio_transfer is used for 8 to 32 bytes max, it
> would be irrelevant to use such a small timeout.
> I can obviously do it, but not sure if using a small (say, 100ms)
> timeout would not be better.

You want a minimum threashold, yes, but you also need to be able toc ope
with larger transfers too so need to be able to raise that where
appropriate.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]