December 2000 - ixp1200 - lists.cs.princeton.edu

RE: IXP Bus Endian Mode
by Hooper, Donald 08 Dec '00

08 Dec '00

All the CD Beta releases were tested with IX Bus in big and little endian mode. There are customers using IX Bus in production systems little endian mode. You have to program the MAC device to be little endian to get bytes swapped. Setting the RDYBUS_TEMPLATE_CTL 5 only causes byte enables pins to be interpreted correctly for little endian. Don -----Original Message----- From: Naik, Uday Sent: Thursday, December 07, 2000 10:36 PM To: Joe, Robert; IXP1200 Users; 'ixp1200(a)CS.Princeton.EDU' Subject: RE: IXP Bus Endian Mode I believe setting the IXBUS in Little Endian mode does not work. But you can set the Endian registers in the PCI bus controller for the IXP1200, to swap the data corectly as it goes over the bus. Uday -----Original Message----- From: Joe, Robert [mailto:robert.joe@intel.com] Sent: Thursday, December 07, 2000 7:32 PM To: IXP1200 Users; 'ixp1200(a)CS.Princeton.EDU' Subject: IXP Bus Endian Mode I am setting bit 5 in RDYBUS_TEMPLATE_CTL, but the IXP bus still appears to be running in Big Endian mode. Here are the details: B0 board in eval chassis Proprietary OS with the ARM running in Little Endian mode. 1.0 pwd_2f with only changes to the SDRAM packet base address In hw_init.uc just after the ready bus has been programmed immed[$xfer2, 0xF20] csr[write, $xfer2, RDYBUS_TEMPLATE_CTL], CTX_SWAP I have ARM code running which dequeues packets off SRAM/SDRAM as they were placed by the microengine receive threads. Ultimately I want to DMA the packet data across the PCI bus. However, because the IX bus is running in Big Endian mode, the packets appear in SDRAM byte swapped on longword boundaries (see HRM, pg 6-55). I know if I do a longword copy of the packet, the copied packet will appear in the correct order. But I want to avoid this overhead. In fact, this is probably the reason the PETH driver works? My thinking is, if I can get the IX bus running in Little Endian mode, the packet data will be in the correct order for my DMA over the PCI bus. Agree? Has anyone run pwd_2f with LITTLE_ENDIAN defined and Little Endian (bit 5) set in the RDYBUS_TEMPLATE_CTL? If you have: Where did you set the RDYBUS_TEMPLATE_CTL? What OS are you running and do you have the ARM running in Little or Big Endian (although this should not matter??) Are you seeing the packet data in Little Endian format? Robert Joe (858) 391 1849

1 0

RE: IXP Bus Endian Mode
by Naik, Uday 07 Dec '00

07 Dec '00

I believe setting the IXBUS in Little Endian mode does not work. But you can set the Endian registers in the PCI bus controller for the IXP1200, to swap the data corectly as it goes over the bus. Uday -----Original Message----- From: Joe, Robert [mailto:robert.joe@intel.com] Sent: Thursday, December 07, 2000 7:32 PM To: IXP1200 Users; 'ixp1200(a)CS.Princeton.EDU' Subject: IXP Bus Endian Mode I am setting bit 5 in RDYBUS_TEMPLATE_CTL, but the IXP bus still appears to be running in Big Endian mode. Here are the details: B0 board in eval chassis Proprietary OS with the ARM running in Little Endian mode. 1.0 pwd_2f with only changes to the SDRAM packet base address In hw_init.uc just after the ready bus has been programmed immed[$xfer2, 0xF20] csr[write, $xfer2, RDYBUS_TEMPLATE_CTL], CTX_SWAP I have ARM code running which dequeues packets off SRAM/SDRAM as they were placed by the microengine receive threads. Ultimately I want to DMA the packet data across the PCI bus. However, because the IX bus is running in Big Endian mode, the packets appear in SDRAM byte swapped on longword boundaries (see HRM, pg 6-55). I know if I do a longword copy of the packet, the copied packet will appear in the correct order. But I want to avoid this overhead. In fact, this is probably the reason the PETH driver works? My thinking is, if I can get the IX bus running in Little Endian mode, the packet data will be in the correct order for my DMA over the PCI bus. Agree? Has anyone run pwd_2f with LITTLE_ENDIAN defined and Little Endian (bit 5) set in the RDYBUS_TEMPLATE_CTL? If you have: Where did you set the RDYBUS_TEMPLATE_CTL? What OS are you running and do you have the ARM running in Little or Big Endian (although this should not matter??) Are you seeing the packet data in Little Endian format? Robert Joe (858) 391 1849

1 0

IXP Bus Endian Mode
by Joe, Robert 07 Dec '00

07 Dec '00

I am setting bit 5 in RDYBUS_TEMPLATE_CTL, but the IXP bus still appears to be running in Big Endian mode. Here are the details: B0 board in eval chassis Proprietary OS with the ARM running in Little Endian mode. 1.0 pwd_2f with only changes to the SDRAM packet base address In hw_init.uc just after the ready bus has been programmed immed[$xfer2, 0xF20] csr[write, $xfer2, RDYBUS_TEMPLATE_CTL], CTX_SWAP I have ARM code running which dequeues packets off SRAM/SDRAM as they were placed by the microengine receive threads. Ultimately I want to DMA the packet data across the PCI bus. However, because the IX bus is running in Big Endian mode, the packets appear in SDRAM byte swapped on longword boundaries (see HRM, pg 6-55). I know if I do a longword copy of the packet, the copied packet will appear in the correct order. But I want to avoid this overhead. In fact, this is probably the reason the PETH driver works? My thinking is, if I can get the IX bus running in Little Endian mode, the packet data will be in the correct order for my DMA over the PCI bus. Agree? Has anyone run pwd_2f with LITTLE_ENDIAN defined and Little Endian (bit 5) set in the RDYBUS_TEMPLATE_CTL? If you have: Where did you set the RDYBUS_TEMPLATE_CTL? What OS are you running and do you have the ARM running in Little or Big Endian (although this should not matter??) Are you seeing the packet data in Little Endian format? Robert Joe (858) 391 1849

1 0