2008-09-10 16:23:16 DMA has priority over core for ext. accesses on 526/527 broken?
Greg Semeraro (UNITED STATES)
Message: 61911
I am seeing PPI overrun errors with a BF526 and BF527 when the "DMA has priority over core for ext. accesses" bit is set in the "EBIU_AMGCTL Global Control" register. These same errors occur when using a BF537 but only if that bit turned off. The implication is that the 526/527 has the same bug as the BF548 (the 548 doesn't give priority to the DMA even when configured to do so). Has anyone else seen this?
In summary:
BF527 - 525MHz core w/ 131MHz peripheral w/ DMA priority set - Occasional PPI Errors
BF526 - 400MHz core w/ 80MHz peripheral w/ DMA priority set - Continuous PPI Errors
BF526 - 400MHz core w/ 100MHz peripheral w/ DMA priority set - Continuous PPI Errors (I know that the peripheral spec on the 526 is 80MHz Max).
BF526 - 400MHz core w/ 133MHz peripheral w/ DMA priority set - Occasional PPI Errors (I know that the peripheral spec on the 526 is 80MHz Max).
BF537 - 500MHz core w/ 100MHz peripheral w/ DMA priority set - No PPI Errors, ever
BF537 - 500MHz core w/ 100MHz peripheral w/ Core priority set - Continuous PPI Errors.
The PPI clock is 40MHz in all of these cases and the peripheral device is identical.
Greg
QuoteReplyEditDelete
2008-09-11 03:15:33 Re: DMA has priority over core for ext. accesses on 526/527 broken?
Mike Frysinger (UNITED STATES)
Message: 61930
i dont believe we've done extensive tests on BF527 along the lines of what you're describing. we've focused on the BF537 and BF548 platforms for proving out software stacks.
what sort of things are you doing ? i'm wondering if there's some simple tests we can perform on the BF527-EZKIT with the LCD.
also, what hardware platforms are you working with ? i'm guessing it isnt a BF527-EZKIT or BF526-EZBOARD ...
QuoteReplyEditDelete
2008-09-11 08:57:43 Re: DMA has priority over core for ext. accesses on 526/527 broken?
Greg Semeraro (UNITED STATES)
Message: 61988
We have custom boards that are based on the 537-STAMP and 527-EZKIT designs. We have 527 and 526 CPUs in the boards that are based on the 527-EZKIT design.
All we are trying to do is transfer an image into SDRAM over the PPI bus. The PPI clock is 40MHz, we get PPI FIFO overrun errors (0x1000 in the PPI Status register) as described above. The "test" would be whatever regression testing that you have for verifying that the Core yeilds priority to the DMA controller when configured to do so. Since this is a known issue with the 548 I have to think that there is already a test that you could use on the 527/526 CPU.
Greg
QuoteReplyEditDelete
2008-09-11 13:17:44 Re: DMA has priority over core for ext. accesses on 526/527 broken?
Robin Getz (UNITED STATES)
Message: 61998
Greg:
It appears as if you have contacted your local FAE about this - and they are trying to determine this for you - so we are considering this closed here.
-Robin
QuoteReplyEditDelete
2008-09-14 18:15:22 Re: DMA has priority over core for ext. accesses on 526/527 broken?
Ian Jeffray (UNITED KINGDOM)
Message: 62137
Hi Greg,
I'm plagued by PPI overrun errors on both the BF537-STAMP board and on our custom designed BF527 board. Setting or clearing the DMA priority never made any difference for me at all... I always get some level of PPI DMA error, even at modest PPI data rates (Eg 25Mhz PPI clock with 125Mhz SCLK, 525Mhz core).
As far as I understand it, there is no real way to avoid these errors, short of completely turning off interrupts and other peripherals whilst using PPI (which isn't really an option for us).
There's some simple hints on the Wiki you could check which may help reduce the impact: http://docs.blackfin.uclinux.org/doku.php?id=ppi
I'd love to know more about this problem - especially if there are further suggestions for ways of reducing PPI DMA errors further.
QuoteReplyEditDelete
2008-09-15 00:10:01 Re: DMA has priority over core for ext. accesses on 526/527 broken?
Robin Getz (UNITED STATES)
Message: 62143
Ian:
I would be interested in the configuration of your system (25MHz PPI / 125MHz SCLK) - we run similar things, and don't see PPI issues. What are your PPI settings?
-Robin
QuoteReplyEditDelete
2008-09-15 07:50:04 Re: DMA has priority over core for ext. accesses on 526/527 broken?
Michael Hennerich (GERMANY)
Message: 62168
Make sure you run ideal SDRAM configuration parameters.
Too conservative settings may explain the difference you’re seeing between BF52x and BF537.
http://docs.blackfin.uclinux.org/doku.php?id=bfin:sdram
-Michael