OPP selection issue on AM335x Starter Kit

Just a heads-up to those working with the AM335x Starter Kit and the WEC 7.0 BSP from Adeneo.

There seems to be an issue where selecting the highest OPP (OPP4, MPU @ 720 MHz), will actually make the MPU run at OPP2, at 500 MHz. I believe there is also a problem with configuring Vdd1 and Vdd2 on the CPU in XLDR, related to the OPP issue.

To confirm this, boot the binary WEC 7.0 images supplied by Adeneo on a Starter Kit. Then look at the output from IOCTL_HAL_GET_CPUSPEED (this returns the current CPU speed in MHz). This is what I am seeing, after selecting each OPP in the bootloader menu:

• OPP1: 275 MHz (OK, as expected)
• OPP2: 500 MHz (OK)
• OPP3: 600 MHz (OK)
• OPP4: 500 MHz (problem - expected 720 MHz)

The troubles start in XLDR, platform.c file, board_init function.

Since g_dwBoardId on a Starter Kit doesn't match the BeagleBone ID, we end up in the else branch, where the following lines can be found:

        if (TWLSetOPVoltage(kVdd1,opp_setting->VDD1Init))
        {
            mpu_pll_config(opp_setting);   
        }

The TWLSetOPVoltage call fails, which means that XLDR does not configure the OPP, but leaves it as is.

The CE OAL would normally set up the OPP again, but the SetOpp function (opp_map.c) assumes that XLDR succeeded in initially setting the OPP to whatever was selected at compile time, so if OPP4 was selected (by means of BSP_OPM_SELECT=4 in am33x_sk_bsp.bat), and a request comes in to change to OPP4, SetOpp ignores it, believing that the platform is already running at OPP4.

So the root of the problem is the failing TWLSetOPVoltage call in XLDR.

To fix it, change the following:

• In bsp_def.h, the I2C address of the power management device is incorrect.
  • Change the definition of TPS65217_I2C_SLAVE_ADDRESS from 0x24 to 0x2d
  • While you are at it, you might want to rename TPS65217_I2C_SLAVE_ADDRESS to TPS65910_I2C_SLAVE_ADDRESS, since the Starter Kit uses a TPS65910; same with TPS65217_I2C_BUS_ID
• In platform.c, the writes to the Vdd1/Vdd2 registers in the TPS65910 fail; specifically, the writes appear to succeed, but when the code reads back the values they don't match what was written. The Vdd1/Vdd2 registers are by default only accessible from the SmartReflex I2C interface, but this interface is not connected on the Starter Kit; only the "Control I2C" interface is. To make the registers accessible through the Control I2C interface, add a new branch to the if/else statement in board_init as shown:

    else if (g_dwBoardId == AM33X_BOARDID_SK_BOARD)
    {
        // Need to configure TPS65910 to access voltage setup registers
        // through the control I2C interface rather than the SmartReflex
        // I2C interface, since the latter is not connected on the AM335x
        // Starter Kit.
        // We do this by setting the SR_CTL_I2C_SEL bit in the
        // DEVCTRL_REG register in the TPS65910.
        hTwl = TWLOpen();
        if(hTwl)
        {
            if(TWLReadRegs(hTwl, PMIC_DEVCTRL_REG, &twlDevCtrlValue, sizeof(twlDevCtrlValue)))
            {
                twlDevCtrlValue |=  PMIC_DEVCTRL_REG_SR_CTL_I2C_SEL_CTL_I2C;
                TWLWriteRegs(hTwl, PMIC_DEVCTRL_REG, &twlDevCtrlValue, sizeof(twlDevCtrlValue));
            }

            TWLClose(hTwl);
            hTwl = NULL;
        }

        TWLSetOPVoltage(kVdd2,VDD2_INIT);

        if(TWLSetOPVoltage(kVdd1,opp_setting->VDD1Init))
        {
            mpu_pll_config(opp_setting);
        }
    }

This appears to fix the issues with setting up Vdd1/Vdd2 voltages and selecting OPP4 in XLDR.

Finally, to minimize confusion, you might want to also modify the debug message in opp_map.c, SetOpp so that the Opp index printed matches the Eboot entry (the 'opp' index is 0-based, Eboot entries are 1-based):
    OALMSG(1, (L"SetOpp to %d\r\n", opp + 1));