PCIe on DM816x

Prateek,

EVM has a fixed 100MHz clock, no SSC.

Which x1 card is being used here? Can you send the dump of DEBUG0 register @0x51001728 (can use devmem2 utility).

Also, please ensure that SW5 switch settings are as described in the RC driver user guide.

   Hemant

Hi Hemant,

The customer was able to get Intel network card (details below) working with PSP version 4.00.01.13 however they need to have SW5 in ON state (switch is closer to r195 and away from PCIe slot). This is opposite of what is specified on link below - http://processors.wiki.ti.com/index.php/DM81xx_AM38xx_PCI_Express_Root_Complex_Driver_User_Guide. They are still not able to get video card working with this PSP version though.

With older PSP version v4.00.0010/12 they are not able to get either of cards working. Result of devmem2 utility below from those releases. I have requested customer to provide this information for new PSP version .

1)  Maxtron PCIe video card MGI G55-MDDE32F with old PSP verison 4.00.00.10/12.  This card is really old, but still working, and the read for 0x51001728 is below:

 root@dm816x-evm:~# devmem2 0x51001728

/dev/mem opened.

Memory mapped at address 0x4020d000.

Read at address  0x51001728 (0x4020d728): 0x00004A03

root@dm816x-evm:~# devmem2 0x51001728

/dev/mem opened.

Memory mapped at address 0x400cf000.

Read at address  0x51001728 (0x400cf728): 0x0000B503

root@dm816x-evm:~# devmem2 0x51001728

/dev/mem opened.

Memory mapped at address 0x4021e000.

Read at address  0x51001728 (0x4021e728): 0x00004A03

root@dm816x-evm:~# devmem2 0x51001728

/dev/mem opened.

Memory mapped at address 0x4032b000.

Read at address  0x51001728 (0x4032b728): 0x0000BC43

2)  Intel network card with version PSP 04.00.00.10/12.

http://www.newegg.com/Product/Product.aspx?Item=N82E16833106014

 root@dm816x-evm:~# devmem2 0x51001728

/dev/mem opened.

Memory mapped at address 0x4013c000.

Read at address  0x51001728 (0x4013c728): 0x0300C211

root@dm816x-evm:~# devmem2 0x51001728

/dev/mem opened.

Memory mapped at address 0x40297000.

Read at address  0x51001728 (0x40297728): 0x03004311

root@dm816x-evm:~# devmem2 0x51001728

/dev/mem opened.

Memory mapped at address 0x400a6000.

Read at address  0x51001728 (0x400a6728): 0x03000111

Hi Hemant,

CORRECTION to post above -

1) Both the devmem2 utility dumps above are using PSP v4.00.01.13.

2) Intel network card works 5% of time when SW5 is OFF as per PCIe driver user guide.

3) Intel network card works all the time when SW5 is ON without any modification to the board. However, the customer is looking at using DM816x as RC and not EP and would like SW5 set to INPUT mode (OFF).

Prateek

If the EVM has PCIe reset logic fixed (check EVM version and changes), that is, PERST# is maintained high after power on, then the SW5-1 should be set as output else if it is always driver low, then only option is to set the switch as input else the EP card will remain in reset if it samples PERST# pin.

   Hemant

Hi Hemant,

Which version of DM8168 EVM from link below fixes PCIe reset logic? http://support.spectrumdigital.com/boards/evm816x/revg/files/evm816x_faq.htm

Are you saying that if DM8168 has to work as RC, then PERST# should be maintained high after power on the board. If this fix is not there, then we have to boot as EP and then at run time change from EP to RC. ??

Thanks,

Prateek

Prateek,

I am not aware the EVM version which has fixed the PERST# (if at all).

Regarding toggle of PERST# (A11 line), as mentioned in the user guide, the correct thing to do on h/w is to toggle the line to high once the EVM (RC) is powered - generally the EP cards use this as indication that the power from slot is good and carry an internal reset on the EP board.

It is possible that, if the A11 remains low, the EP may not start and get detected - of course, note that the behavior depends on how the EP board h/w uses A11 input. 

I suggest to monitor the A11 line in the slot when SW5-1 is ON as well as OFF, please let me know HIGH/LOW status in both the cases.

   Hemant

Hi Hemant,

I am also working with Tri in this issue that we currently have. We have 2 EVM boards rev. D and rev. F, they both have PERST# high after power-on no matter which position SW-5 is at (ON/OFF). Also, resistor R218 is already 1K-ohm which is lower than 2K-ohm so I believe the PERST# getting always stuck at low issue is already fixed on our EVM boards.

On the EVM board, we have the following:

• SW-5 is ON (short – toward R195) EVM outputs PERST# to EP slot: DM8168 detects Intel Gigabit Controller card all the time. (CPU is RC mode)
• SW-5 is OFF (open – toward PCIe card) EP inputs PERST# to EVM: DM8168 detects Intel Gigabit Controller card “some time” (about 5%) (CPU is in EP mode)

On our engineering prototype board has an IDT PCIe switch with one x1 upstream port to the CPU and three x1 downstream ports, thus DM8168 needs to be in RC mode.

Kevin,

they both have PERST# high after power-on no matter which position SW-5 is at (ON/OFF). Also, resistor R218 is already 1K-ohm which is lower than 2K-ohm

But you are still having different observations depending upon the switch setting, perhaps when SW5-1 is ON (OUT), there is a momentary low to high transition? Can you upload signal trace for both switch settings?

   Hemant

Hi Hemant,

We figured it out. Most PCIe cards have their own independent "PERST#", but some have options where we can chose either "independent" or "from RC's reset." In our case, we have another PCIe switch EVM (EP) board that does not work with TI DM8168's EVM (RC) when EP's PERST# is set to "from RC's reset." It works just fine when we set it to "independent" due to a reset timing.

Thanks for the help.