PXA270 board startup: printf does not work

Starting up a new board with a PXA270, the console output from printf calls in board.c contains garbage, whereas puts works fine. It appears the vsprintf call in printf works correctly, but printf's local char buffer has a few alternate correct characters at its very beginning followed by junk.

Everything else seems to work correctly: this SDRAM area can be written
to with a debugger. But test functions (similar to printf) that copy a
string parameter to a local buffer and then call puts also produce garbage.

Buffers initialized (i.e., char t [12] = "Testing...";) in these functions will print (with puts) ONLY if their size is <= 32. Their contents become garbage when other data are copied to them, regardless of size.

I haven't yet tried to start a ramdisk or image; without correct output from printf or the debug () macro it would be futile.

Thanks for any ideas.

JP

Your talk about having a corrupted local char buffer confused me. If it were not for that detail, I would have been positive that you have a problem with handling your UART's Tx busy flag.

Your symptoms are typical of not waiting for the UART to complete transmitting a character before stuffing the next one in once the UART FIFO is full. The result is that, if the string has fewer than /n/ characters (/n/ being the depth of the UART's Tx FIFO, possibly a few more), it works OK. If the string is longer than /n/, it gets corrupted by your s/w overwriting characters in the UART. Depending on the UART implementation, this could include the one currently being shifted out, resulting in garbage characters.

What in your processor is 32 bytes long? UART FIFO? Cache line? Hmmmm, could you be having cache consistency problems? Cache problems would be consistent with garbage in memory.

BIG WARNING NOTE: Writing memory with a debugger is typically very benign (s.l.o.w.) compared to writing with the processor. Writing single memory locations with the processor is typically benign compared to a cache line burst read/write. SDRAM (including DDR/DDR2) initialization problems typically do *not* show up until cache is enabled because it is the burst read/write that violates the "S" in SDRAM (you end up out of sync "synchronous").
FAQ: <http://www.denx.de/wiki/DULG/SDRAM>

Quoted 4 lines · Jerry Van Baren

Your talk about having a corrupted local char buffer confused me. If it were not for that detail, I would have been positive that you have a problem with handling your UART's Tx busy flag.

Quoted 8 lines

Your symptoms are typical of not waiting for the UART to complete transmitting a character before stuffing the next one in once the UART FIFO is full. The result is that, if the string has fewer than /n/ characters (/n/ being the depth of the UART's Tx FIFO, possibly a few more), it works OK. If the string is longer than /n/, it gets corrupted by your s/w overwriting characters in the UART. Depending on the UART implementation, this could include the one currently being shifted out, resulting in garbage characters.

A very long (already initialized) string prints correctly with puts, so
I suspect the FIFO implementation is OK. We're using the FFUART.

Quoted 4 lines

What in your processor is 32 bytes long? UART FIFO? Cache line? Hmmmm, could you be having cache consistency problems? Cache problems would be consistent with garbage in memory.

Quoted 8 lines

BIG WARNING NOTE: Writing memory with a debugger is typically very benign (s.l.o.w.) compared to writing with the processor. Writing single memory locations with the processor is typically benign compared to a cache line burst read/write. SDRAM (including DDR/DDR2) initialization problems typically do *not* show up until cache is enabled because it is the burst read/write that violates the "S" in SDRAM (you end up out of sync "synchronous").
FAQ: <http://www.denx.de/wiki/DULG/SDRAM>

Thanks for those ideas. We haven't found many examples to compare our configuration to, but we did have trouble initially getting the SDRAM startup sequence working. Your description and the wiki entry describe what could be happening.

JP

Quoted 1 line

JP

This turned out to be a hardware problem: the DQM 0/1 lines that select 8 bits of a 32 bit location were reversed.