Are external 10k pullups neccesary on MMC1 data/command lines when connected to eMMC device?

Hi,

Have you double-checked that the internal pullups are enabled? Also I don't understand your statement about U-boot - does eMMC work properly there or not?

Hi Biser,
Thanks for your answer
Yes, all the pullups are enabled.
eMMC does not work properly, eather in u-boot or in linux. Only the SD-Card attached to the MMC0 works properly.
Best Regards
Jean-Luc

This is strange. How long are the traces connecting to the eMMC? Do you have a serial resistor (~22Ohms) on the MMC_CLK line close to the processor?

Hi,

here are my schematics around this eMMC.

Yes we have this 22 Ohm. The traces are between 18-20 mm (Millimeters).

We see that some NC balls have been used by the layouter to contact the right balls, but this is allowed according eMMC datasheets.

Best Regards

Hello Jean-Luc,

What is the eMMC version?  i.e. what is the value for VDDQ/VCCQ on the eMMC datasheet?

Regards,

a2g

I can see nothing wrong in the schematics. You are using a valid I/O set on the AM335X. What I cannot see are the I/O power supplies. VDDSHV1 and VDDSHV2 must be 3.3V.

Can you try to place a 10kOhm pullup on the CMD line?

We have two possible chips, one from Hynix and one from MICRON,

Both are at least 4.5 compatible, both have VCC = 3.3V and dual  VCCq  (1.8 or 3.3V). And with both we have the same behavior.

Regards

Jean-Luc

We have all VDDSHVx = 3.3V.
Ok, I will try to add a pullup on CMD line.
But it works properly in mode 1 data bit (D0). We can see the manufacturer data, can correctly read/write in the whole domain. So for me the cmd and clk lines should be ok. Or not?
Thanks

I checked the Micron datasheet and it recommends a stronger pullup on CMD. I don't dare ask you to pullup all 8 data lines, as this will likely be impossible over the trace lengths that you have. Also I seem to remember that CMD is open-drain in some situations. Pullup on CLK is definitely not needed.

By the way, is this a single board or you see the same issue on multiple boards?

what about pinmux settings in your U-Boot files?
What version of U-boot are you using? from TI or from official from denx?
FYI, bus width is 4-bit in U-Boot by default.
Does U-boot recognize eMMC? ('mmc info')

Regards,
a2g

We have the same issue with our 15 prototypes, 5 having Micron and 10 having Hynix.

Yes, U-boot recognizes the eMMC, mmc info command gives the correct data. The only issue is the bus width limited to 1.
Our eMMC is also working under Linux if we force the data width to 1.

We are using U-boot based on the Beagleboard, but adapted to our board (pinmux, ...).
We were able to measure some data transfer on D0, but also on D1, D2, D3 lines during the eMMC initilization. So for me the pinmux is Ok. We also tried to enable the PD instead of PU on the datalines, it was worst, nothing anymore working, that's means PU are needed but may be our are not strong enough...

Tomorrow I will try to add the external PU on cmd line, and will be back to you. Today I am not in the office, so cannot do it...

Thanks again and Best Regards
Jean-Luc

Hi all,

I'm working together with Jean-Luc on this project. To be more specific we're using the TI flavour of U-Boot 2015.07. The config is indeed based on the config for TI AM335x boards with our own specific modifications.

We now that U-Boot only works with 4-bit width by default (such as with the Beagleboard config). But also 4 is not working. We're always falling back to 1-bit width because the initialization code fails to compare the EXT_CSD when configured with more that 1-bit width.

If it helps to identify the issue, here are the values we got when the driver compares the register values (2nd column is the correct value and 3rd one is the wrong value we get with 4-bit width set).

Register	1-bit width value	4-bit width value
EXT_CSD_PARTITIONING_SUPPORT	0x7	0x45
EXT_CSD_HC_WP_GRP_SIZE	0x10	0x54
EXT_CSD_REV	0x7	0x45
EXT_CSD_HC_ERASE_GRP_SIZE	0x1	0x45
EXT_CSD_SEC_CNT	0x0	0x44
EXT_CSD_SEC_CNT+1	0x80	0xC4
EXT_CSD_SEC_CNT+2	0x74	0x54
EXT_CSD_SEC_CNT+3	0x0	0x44

What is weird is that ALL our boards have the same behavior, even with two different MMC chip. On SKhynix one register is different  (EXT_CSD_REV=6) but the "mirrored" corresponding value is also different (i.e. 0x44 which is also the value minus 1...). That is, in U-Boot we ALWAYS have the same wrong values when trying more than one 1-bit width. But I can't come up with a good pattern out of those numbers.

PS: with Linux, it fails also while trying to initialize to 8- and 4- bit bus width with error -84 (EILSEQ) because the bit DEB_EN is set (i.e. DATA End Bit Error from AM335x TRM). Maybe this is also a hint for missing pull-ups or other wrong lines?? (but seems to be OK after layout verifications)

PS2: trying at lower speed doesn't seem to help. We went down to 600 kHz and still got the exact same behavior here.

Thanks a lot for any suggestion here

Hi,

Thanks for the details; did you tried with pull-up on cmd line? is it possible?

FYI, we're using AM3352 with a similar eMMC (MTFC4GACA) but on MMC2 port. I'm using U-Boot 2016-01 (no differences for this interface). we have pull-up on data lines, clk and CMD (CMD line is working in 2 modes, one is open-drain mode...). I have no linux OS, I checked with u-boot only (info/read/write access) and this working well.

If I'm not wrong, CSD register and other registers uses CLK/CMD lines only (I suppose this is not wrong because on our last prototype with a wrong eMMC with different Vccq, it was possible to get info but no read/write was possible)

I can try removing pull-up on different lines on my board (prototype) just for testing purpose if you're interested in, but not today as it's too dangerous to do it on a Friday!!! Please let me know!

Regards,
a2g.

Hi a2g and Biser,

Thanks for your reply.

I just tried with 10k PU on cmd line -> no changes
Then 10k PU on D1 and D2 -> no changes
But I cannot easely contact D3...

@a2g: Thanks for your proposal to remove your pull-up, it will be really great If you can and want do it to help us. ;-)

Best regards and nice weekend
Jean-Luc

Hello,
I checked this morning by removing pull-ups on my board on CMD, CLK and DAT0-3 lines and everything is still working correctly (info, read/write blocks).
So I can confirm that you would have had still an issue in your configuration as you found it! ^^
Regards.

Hi a2g,

Thanks a lot for your time and your confirmation. It was nice helping us solving this issue ;-)

Have a nice day and best Regards.