It took me a while, but now I understand why SB600 SPI accesses hang on some boards.
The problem is not related to the presence or absence of a SPI chip. It is also not related to usage of GPIO lines for SPI or GPIO.
On some boards (with LPC flash) the SPI_BaseAddr is 0. The code did not check for that condition and assumed the BAR was really at address zero. Of course, every machine has RAM there and all writes to the SB600 SPI controller ended up scribbling over the zero page in RAM. That also means the SB600 SPI driver never got the expected response from the SPI controller because the controller never saw the writes.
Fortunately, the bug is fixed in r491 in the new flashrom repository.
If anyone still wondered why we were seeing these hangs, the above explanation should hopefully be sufficient.
Regards, Carl-Daniel
Carl-Daniel Hailfinger wrote:
On some boards (with LPC flash) the SPI_BaseAddr is 0. The code did not check for that condition and assumed the BAR was really at address zero.
The problem is of course, still, that it's impossible to really detect what the chipset is using. Is this the same also with SB700?
I guess you're using the SPI BAR value as part of the heuristic.
//Peter
On 14.05.2009 03:31, Peter Stuge wrote:
Carl-Daniel Hailfinger wrote:
On some boards (with LPC flash) the SPI_BaseAddr is 0. The code did not check for that condition and assumed the BAR was really at address zero.
The problem is of course, still, that it's impossible to really detect what the chipset is using. Is this the same also with SB700?
The docs are not yet public, so I don't know.
I guess you're using the SPI BAR value as part of the heuristic.
Since accessing the SPI master is completely impossible if the SPI BAR is located at 0x0 (unless flashrom includes code which can tell the OS that a new BAR appeared and force the OS to reallocate resources), it is a perfectly reliable indicator of SPI probe usability. Note that even if a SPI flash chip were attached, probing for it would still hang with the BAR at 0x0. OTOH, even if no SPI chip is attached, probing will not hang as long as the BAR is not at 0x0.
To summarize, the code does not check for presence of a SPI chip but just if SPI accesses will hang. And since the hang is the real problem at hand, the BAR checking works out just fine.
Regards, Carl-Daniel
Carl-Daniel Hailfinger wrote:
The docs are not yet public, so I don't know.
The question was for someone else who knows.
I guess you're using the SPI BAR value as part of the heuristic.
Since accessing the SPI master is completely impossible if the SPI BAR is located at 0x0
Sure, but a valid BAR does not a flash chip present make.
OTOH, even if no SPI chip is attached, probing will not hang as long as the BAR is not at 0x0.
Do you know what will actually happen in that case?
//Peter
-
Carl-Daniel Hailfinger -
Peter Stuge