Re: [V3] i2c: i2c-qcom-geni: Correct I2C TRE sequence

From: Viken Dadhaniya
Date: Mon Feb 12 2024 - 07:59:00 EST

On 2/9/2024 8:11 PM, Konrad Dybcio wrote:
On 8.02.2024 12:59, Andi Shyti wrote:
Hi Dmitry,

On Thu, Feb 08, 2024 at 01:04:14PM +0200, Dmitry Baryshkov wrote:
On Thu, 8 Feb 2024 at 12:02, Andi Shyti <andi.shyti@xxxxxxxxxx> wrote:

Hi Viken, Dmitry,

On Fri, Feb 02, 2024 at 04:13:06PM +0530, Viken Dadhaniya wrote:

On 2/1/2024 5:24 PM, Dmitry Baryshkov wrote:
On Thu, 1 Feb 2024 at 12:13, Viken Dadhaniya <quic_vdadhani@xxxxxxxxxxx> wrote:

For i2c read operation in GSI mode, we are getting timeout
due to malformed TRE basically incorrect TRE sequence
in gpi(drivers/dma/qcom/gpi.c) driver.

TRE stands for Transfer Ring Element - which is basically an element with
size of 4 words. It contains all information like slave address,
clk divider, dma address value data size etc).

Mainly we have 3 TREs(Config, GO and DMA tre).
- CONFIG TRE : consists of internal register configuration which is
required before start of the transfer.
- DMA TRE : contains DDR/Memory address, called as DMA descriptor.
- GO TRE : contains Transfer directions, slave ID, Delay flags, Length
of the transfer.

Driver calls GPI driver API to config each TRE depending on the protocol.
If we see GPI driver, for RX operation we are configuring DMA tre and
for TX operation we are configuring GO tre.

For read operation tre sequence will be as below which is not aligned
to hardware programming guide.

- CONFIG tre
- DMA tre
- GO tre

As per Qualcomm's internal Hardware Programming Guide, we should configure
TREs in below sequence for any RX only transfer.

- CONFIG tre
- GO tre
- DMA tre

In summary, for RX only transfers, we are reordering DMA and GO TREs.
Tested covering i2c read/write transfer on QCM6490 RB3 board.

This hasn't improved. You must describe what is the connection between
TRE types and the geni_i2c_gpi calls.
It is not obvious until somebody looks into the GPI DMA driver.

Another point, for some reason you are still using just the patch
version in email subject. Please fix your setup so that the email
subject also includes the `[PATCH` part in the subject, which is there
by default.
Hint: git format-patch -1 -v4 will do that for you without a need to
correct anything afterwards.

At high level, let me explain the I2C to GPI driver flow in general.

I2C driver calls GPI driver exposed functions which will prepare all the
TREs as per programming guide and
queues to the GPI DMA engine for execution. Upon completion of the Transfer,
GPI DMA engine will generate an
interrupt which will be handled inside the GPIO driver. Then GPI driver will
call DMA framework registered callback by i2c.
Upon receiving this callback, i2c driver marks the transfer completion.

Any news about this? Dmitry do you still have concerns? We can
add this last description in the commit log, as well, if needed.

I was looking for pretty simple addition to the commit message, that
links existing commit message to the actual source code change: that
geni_i2c_gpi(I2C_WRITE) results in the GO TRE and
geni_i2c_gpi(I2C_READ) generates DMA TRE. But I haven't seen anything
sensible up to now. So far we have a nice description of required
programming sequence in terms of CONFIG, GO, DMA TREs and then source
code change that seems completely unrelated to the commit message,
unless one actually goes deep into the corresponding GPI DMA driver.

Agree. I can't take this patch until the commit message has a
proper description and until Dmitry doesn't have any concerns

And please, please, include the word PATCH in the square brackets in
the subject, it's landing in the wrong email folders for a number of

Included "PATCH" string in subject in V4.