Important ColdFire Lite for 52259 bug found!

[vier_kuifjes]

I just found an important bug in Coldfire Lite when compiled for the 52259 CPU. This bug will make the board hang during initialisation!

The bug is located in MCF52259_sysinit.c , near the end of the function void mcf52259_ePHY_init(void):

This is the original version:

 while(!(fec_mii_read(FEC_PHY0, 0x10, &reg0))) // read PHY status register
 {
  reg0=0;
 };
 DUPLEX_phy_r17_dpm = (int)((reg0&0x0004)>>2); // 1=full,0=half duplex...used in ifec.c

 do  //FSL read PHY status register
 {
  fec_mii_read(FEC_PHY0, 0x10, &reg0);
 }while (!(reg0&(PHY_R1_LS)));  //FSL exit while loop when Link Status up

 This is what should be there:

 while(!(fec_mii_read(FEC_PHY0, PHY_REG_SR, &reg0))) // read PHY status register
 {
  reg0=0;
 };
 DUPLEX_phy_r17_dpm = (int)((reg0&0x0004)>>2); // 1=full,0=half duplex...used in ifec.c

 do  //FSL read PHY status register
 {
  fec_mii_read(FEC_PHY0, PHY_REG_SR, &reg0);
 }while (!(reg0&(PHY_R1_LS)));  //FSL exit while loop when Link Status up

 

The actual value of PHY_REG_SR is 0x01. This means that, in the original version of MCF52259_sysinit.c, the program is reading the wrong register (0x10, which actually corresponds to PHY_REG_IR, the PHY interrupt register).

Message Edited by Marc VDH on 2009-02-25 02:52 PM

[vier_kuifjes]

I have checked out my "bug report" a little more closely. The original ePHY_init routine was written with only the52259EVB board in mind. This board uses a different PHY compared to the DEMO board. Both PHYs have a different register set, which means that the init routine for the EVB may not work for the DEMO board.

So I made a modification to make the init routine compatible with both EVB and DEMO boards, based on the board selection in processor.h

The first piece of code shows the original code from the init routine, the second piece is my new version.

 while(!(fec_mii_read(FEC_PHY0, 0x10, &reg0))) // read PHY status register
 {
  reg0=0;
 };
 DUPLEX_phy_r17_dpm = (int)((reg0&0x0004)>>2); // 1=full,0=half duplex...used in ifec.c

 do  //FSL read PHY status register
 {
  fec_mii_read(FEC_PHY0, 0x10, &reg0);
 }while (!(reg0&(PHY_R1_LS)));  //FSL exit while loop when Link Status up

 I still believe the original code contains a bug, by the way. There is a reference to a bit called PHY_R1_LS (link status) in PHY register 0x10.The link status bit is available in register 0x10, but not at the bit position PHY_R1_LS, which actually belongs to the PHY basic status register at adress 0x01.

 do  //FSL read PHY status register
 {
  fec_mii_read(FEC_PHY0, PHY_REG_SR, &reg0);
 }while (!(reg0&(PHY_R1_LS)));  //FSL exit while loop when Link Status up

#if M52259EVB
 while(!(fec_mii_read(FEC_PHY0, 0x10, &reg0))) // read DP83640 PHY status register
 {
  reg0=0;
 };
 DUPLEX_phy_r17_dpm = (int)((reg0&0x0004)>>2); // 1=full,0=half duplex...used in ifec.c
#else // M52259DEMO
 while ((reg0 & 0x1c) == 0)
 {
  while(!(fec_mii_read(FEC_PHY0, 0x1f, &reg0))) // read KSZ8041NL PHY control 2 register
  {
   reg0=0;
  };
 };
 DUPLEX_phy_r17_dpm = (int)((reg0&0x0010)>>4); // 1=full,0=half duplex...used in ifec.c
#endif

 In my new version, I switched the link status detect and the full/half duplex readout. I think it makes more sense to wait for the link to become active before figuring out if the link is full or half duplex.

Checking the duplex had to be done differently in the EVB and DEMO boards because of the different PHY chips used. See the corrsponding data sheets for th actual registers/bits used.

- Marc

[J2MEJediMaster]

I have checked with an enginner on this matter, and here is his response:

The M52259EVB has a National Semi DP83640 Ethernet PHY.

That PHY has a status register at 0x10 that contains many useful parameters all in one register.

Reference: http://www.national.com/ds/DP/DP83640.pdf 

14.1.10 PHY Status Register (PHYSTS)

This register provides a single location within the register set for quick access to commonly accessed information.

TABLE 23. PHY Status Register (PHYSTS), address 0x10

There are other registers that can be used, but we simply choose this one.

The 0x10 register we have in the code is a valid National Semi PHY register and has worked for us.

The 0x01 register he suggests is OK for Link Status but not duplex setting.

The 0x10 register has both the Link and Duplex status.

Having said that, we did notice a #if not set correctly that causes the board to fail to connect in the same function.

 MCF_FEC_MSCR = MCF_FEC_MSCR_MII_SPEED((uint32)(SYS_CLK_MHZ/5));
#if 0        <----- WAS SET TO 1.......
 do
 {
  while(!(fec_mii_read(FEC_PHY0, PHY_REG_CR, &reg0)))  //FSL read PHY control register
  {
   reg0=0;
  };
  
 } while(reg0&PHY_R0_RESET);  //Test RESET bit...1=in reset, 0=reset complete
#else
 for(myctr=0;myctr<32;myctr++)
 {
  fec_mii_read(myctr, PHY_REG_CR, &reg0);
  if(reg0 != 0xffff)
   break;
 }

A new version of ColdFire Lite was posted, and perhaps it will solve your problem.

The engineer is interested as why your change worked, and is willing to dicsuss the matter offline with you.

---Tom

[vier_kuifjes]

Maybe it might work with the EVB board then, but the smaller DEMO board uses a MICREL PHY. The MICREL PHY doesn't appear to have a register 0x10. Maybe that is the reason the original version didn't work for me?...

Message Edited by Marc VDH on 2009-02-26 08:29 PM

Message Edited by Marc VDH on 2009-02-26 08:30 PM

[vier_kuifjes]

I went checking for the new version. I found ColdFire_TCP-IP_Lite_Rev3.2 which is available since the beginning of February and which is the version I have been testing with.

I also found ColdFire_Lite_M52259EVB.

There appear to be quite a number of differences between the two. I'm not sure which is the one you're referencing to though, but the files in the archive which contains ColdFire_Lite_M52259EVB appear to be older then the files in the revision 3.2 archive.

Maybe I should check through all the files with differences and keep the best from both!

Contacting the engineer seems like a good idea to me too...

[J2MEJediMaster]

I'll have the engineer contact you at your Gmail address.

---Tom