Re: [PATCH 1/2] i2c: qcom-geni: Provide an option to select FIFO processing

[Lee Jones]

On Wed, 04 Sep 2019, Wolfram Sang wrote:

> On Wed, Sep 04, 2019 at 01:35:48PM -0700, Bjorn Andersson wrote:
> > On Wed 04 Sep 04:36 PDT 2019, Lee Jones wrote:
> > 
> > The subject implies that we select FIFO mode instead of DMA, but that's
> > not really true, because with DMA enabled we still fall back to FIFO for
> > messages below 32 bytes. 

Do you mean, we fall back to DMA?

> > So what this does it to disable DMA, which neither the subject or the DT
> > property describes.
> > 
> > Also missing is a description of why this is needed.
> 
> Yes.
> 
> I am willing to help to get this resolved soonish. However, I have
> issues with the approach.
> 
> It looks like a workaround to me. It would be interesting to hear which
> I2C client breaks with DMA and if it's driver can't be fixed somehow
> instead. But even if we agree on a workaround short term, adding a
> binding for this workaround seems like a no-go to me. We have to live
> with this binding forever. Sidenote: I could think of a generic
> 'disable-dma' which could be reused everywhere but we probably won't get
> that upstream that late in the cycle.
> 
> Is there no other way to disable DMA which is local to this driver so we
> can easily revert the workaround later?

This is the most local low-impact solution (nomenclature aside).

The beautiful thing about this approach is that, *if* the Geni SE DMA
ever starts working, we can remove the C code and any old properties
left in older DTs just become NOOP.  Older kernels with newer DTs
(less of a priority) *still* won't work, but they don't work now
anyway.

NB: QCom have also made it pretty clear that DTBs *must* match their
kernel version.  I know this is controversial amongst DT purists, but
it's still how QCom operate.

The offending line can be found at [0].  There is no obvious bug to
fix and this code obviously works well on some of the hardware
platforms using it.  But on our platform (Lenovo Yoga C630 - QCom
SMD850) that final command, which initiates the DMA transaction, ends
up rebooting the machine.

With regards to the nomenclature, my original suggestion was
'qcom,geni-se-no-dma'.  Would that better suit your request?

[0] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/soc/qcom/qcom-geni-se.c#n644

> > > Signed-off-by: Lee Jones <lee.jones@xxxxxxxxxx>
> > > ---
> > >  drivers/i2c/busses/i2c-qcom-geni.c | 14 ++++++++++----
> > >  1 file changed, 10 insertions(+), 4 deletions(-)
> > > 
> > > diff --git a/drivers/i2c/busses/i2c-qcom-geni.c b/drivers/i2c/busses/i2c-qcom-geni.c
> > > index a89bfce5388e..dfdbce067827 100644
> > > --- a/drivers/i2c/busses/i2c-qcom-geni.c
> > > +++ b/drivers/i2c/busses/i2c-qcom-geni.c
> > > @@ -353,13 +353,16 @@ static void geni_i2c_tx_fsm_rst(struct geni_i2c_dev *gi2c)
> > >  static int geni_i2c_rx_one_msg(struct geni_i2c_dev *gi2c, struct i2c_msg *msg,
> > >  				u32 m_param)
> > >  {
> > > +	struct device_node *np = gi2c->se.dev->of_node;
> > >  	dma_addr_t rx_dma;
> > >  	unsigned long time_left;
> > > -	void *dma_buf;
> > > +	void *dma_buf = NULL;
> > >  	struct geni_se *se = &gi2c->se;
> > >  	size_t len = msg->len;
> > >  
> > > -	dma_buf = i2c_get_dma_safe_msg_buf(msg, 32);
> > > +	if (!of_property_read_bool(np, "qcom,geni-se-fifo"))
> > > +		dma_buf = i2c_get_dma_safe_msg_buf(msg, 32);
> > > +
> > >  	if (dma_buf)
> > >  		geni_se_select_mode(se, GENI_SE_DMA);
> > >  	else
> > > @@ -392,13 +395,16 @@ static int geni_i2c_rx_one_msg(struct geni_i2c_dev *gi2c, struct i2c_msg *msg,
> > >  static int geni_i2c_tx_one_msg(struct geni_i2c_dev *gi2c, struct i2c_msg *msg,
> > >  				u32 m_param)
> > >  {
> > > +	struct device_node *np = gi2c->se.dev->of_node;
> > >  	dma_addr_t tx_dma;
> > >  	unsigned long time_left;
> > > -	void *dma_buf;
> > > +	void *dma_buf = NULL;
> > >  	struct geni_se *se = &gi2c->se;
> > >  	size_t len = msg->len;
> > >  
> > > -	dma_buf = i2c_get_dma_safe_msg_buf(msg, 32);
> > > +	if (!of_property_read_bool(np, "qcom,geni-se-fifo"))
> > > +		dma_buf = i2c_get_dma_safe_msg_buf(msg, 32);
> > > +
> > >  	if (dma_buf)
> > >  		geni_se_select_mode(se, GENI_SE_DMA);
> > >  	else



-- 
Lee Jones [æçæ]
Linaro Services Technical Lead
Linaro.org â Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog