[dwi2c PATCH v1] dwi2c add offsets to reads

[Simon Glass]

Hi CooperX,

On Fri, 6 Nov 2020 at 16:08, Duffin, CooperX <cooperx.duffin at intel.com> wrote:
>
> Hello Simon,
>
> I wasn’t using the test/dm/i2c I this was tested using hardware where I was using the dm_i2c_read() function. I just tried to use the test/dm/i2c where I am following the README but I keep getting  "sdl2-config: Command not found". I suspect it would also fail where it is trying to read but I will have to get test/dm/i2c working before I know for sure.
>
> Essentially my test was
>
> int uboot_app (int argc, char * const argv[])
> {
>         uint8_t buf[10];
>         uint8_t read_buf[5];
>         uint8_t dev_addr;
>         uint32_t bus_speed;
>         //i2c_bus device pointer
>         struct udevice *i2c_led;
>         struct udevice *bus;
>         app_startup(argv);
>         /* Print the ABI version */
>         printf ("Example expects ABI version %d\n", XF_VERSION);
>         printf ("Actual U-Boot ABI version %d\n", (int)get_version());
>
>         dev_addr = 0x60;
>         bus_speed = 100000; //100KHz
>         printf("starting test i2c\n");
>         buf[0] = 'b';
>         buf[1] = 'o';
>         buf[2] = 'o';
>         buf[3] = 't';
>         //Get i2c chip, init the code
>         printf("init starting\n");
>         if(i2c_get_chip_for_busnum(0 , 0x60, 1, &i2c_led)!=0){
>                 printf("ERROR: no device found at %x \n",dev_addr);
>                 return (0);
>         }
>         uclass_get_device_by_seq(UCLASS_I2C, 0, &bus);
>         printf("Setting bus speed to %d\n", bus_speed);
>         if(dm_i2c_set_bus_speed(bus,bus_speed)!=0){
>                 printf("ERROR: Cannot set buspeed\n");
>                 return (0);
>         }
>         printf("i2c_led name is %s\n",i2c_led->name);
>         if(i2c_led == NULL){
>                 printf("ERROR i2c_led 0 is null\n");
>                 return (0);
>         }
>         printf("Writing\n");
>         for(unsigned int a =0; a< 4; a++){
>                 if(dm_i2c_write(i2c_led,a,&buf[a],1)!=0){
>                         printf("ERROR writing\n");
>                         return (0);
>                 }
>         }
>
>         printf("Reading\n");
>         for(unsigned int a =0; a< 4; a++){
>                 if(dm_i2c_read(i2c_led,a,&read_buf[a],1)!=0){
>                         printf("ERROR writing\n");
>                         return (0);
>                 }
>         }
>         printf("read buffer is %c, %c, %c, %c \n",read_buf[0],read_buf[1], read_buf[2],read_buf[3]);
>         printf ("\n\n");
>         return (0);
> }
>
> ## Starting application at 0x0C100000 ...
> Example expects ABI version 10
> Actual U-Boot ABI version 10
> starting test i2c
> init starting
> Setting bus speed to 100000
> i2c_led name is generic_60
> Writing
> Reading
> read buffer is o, o, o, o
>
> with the fix I get:
>
> ## Starting application at 0x0C100000 ...
> Example expects ABI version 10
> Actual U-Boot ABI version 10
> starting test i2c
> init starting
> Setting bus speed to 100000
> i2c_led name is generic_60
> Writing
> Reading
> read buffer is b, o, o, t
>
> which is correct. Similarly if I stop auto boot I get a similar result:
>
> Hit any key to stop autoboot:  0
> device # i2c
> i2c - I2C sub-system
>
> Assuming I write "boot" to the device
> device # i2c dev 0
> Setting bus to 0
> device # i2c md 0x60 0 4
> 0000: 6f 6f 6f 6f    oooo
>
> With the fix/patch I get:
>
> device# i2c md 0x60 0 4
> 0000: 62 6f 6f 74    boot
>
> Which is correct. Also the reads were working in linux leading to my original suspicion that there might be something going on in the driver. Hopefully that answers your question let me know if you have anymore.

I had a bit of a look at this. If you look at dm_i2c_read() it
actually builds the address into the buffer it sends. So when it gets
to __dw_i2c_write() the alen parameter is always 0. That function is
actually a holdover from before driver model, so one day the alen and
addr parameters will go away.

I wonder if your device does not support multiple-byte reads or
writes? Can you try the i2c md with a single byte? There is a
DM_I2C_CHIP_WR_ADDRESS option to handle this - see the 'i2c flags'
command.

In the designware_i2c_xfer(), try printing out the bytes that it gets
in each message using i2c_dump_msgs().

Also check the i2c_get_chip() function which you are using. See if
chip->offset_len is set to 1 as it should be, for your device. If not,
perhaps something is missing.

In short I am not really sure what is going on, but I think it needs
more investigation.

BTW with this mailing list we like people to reply inline or at the
bottom, not at the top.

Regards,
Simon

>
> Best regards,
>
> -Cooper Duffin
> -----Original Message-----
> From: Simon Glass <sjg at chromium.org>
> Sent: Friday, November 6, 2020 10:51 AM
> To: Duffin, CooperX <cooperx.duffin at intel.com>
> Cc: U-Boot Mailing List <u-boot at lists.denx.de>; uboot-snps-arc at synopsys.com; Tom Rini <trini at konsulko.com>; Robert Beckett <bob.beckett at collabora.com>; Heiko Schocher <hs at denx.de>; Wolgang Denk <wd at denx.de>; Ian Ray <ian.ray at ge.com>
> Subject: Re: [dwi2c PATCH v1] dwi2c add offsets to reads
>
> Hi Cduffinx,
>
> On Thu, 5 Nov 2020 at 14:26, cduffinx <cooperx.duffin at intel.com> wrote:
> >
> > modify the designware_i2c_xfer function to use 1 byte per address, it
> > was set to 0 before which makes it think its reading a register type.
> > Added offset of where it is supposed to read from. Before it was
> > always reading from offset 0 despite specifying the offset in the
> > higher level function.
> >
> > Signed-off-by: Cooper Duffin <cooperx.duffin at intel.com>
> > Signed-off-by: cduffinx <cooperx.duffin at intel.com>
> > ---
> >
> >  drivers/i2c/designware_i2c.c | 5 +++--
> >  drivers/i2c/i2c-uclass.c     | 1 +
> >  include/i2c.h                | 1 +
> >  3 files changed, 5 insertions(+), 2 deletions(-)
> >
>
> Thanks for the patch!
>
> Is there a test that was failing before (test/dm/i2c.c) or should we add a new one to catch this?
>
> Regards,
> Simon