http://os1a.cs.columbia.edu/lxr/source/drivers/ata/pata_mpc52xx.c

O, czyli co, jednak da się coś lepszego z 5200B wyciągnąć niż PIO-0?

Wyciągnąć się da, ale czy będzie działać bez corruptowania danych to inna sprawa. Po pierwsze w MPC5200B jest bug, o którym z resztą mowa nawet w komentarzach tego sterownika, po drugie design kontrolera DMA BestComm jest zaprawdę szalony.

Ostatnia aktualizacja: 13.07.2012 18:30:53 przez strim

Czyli teraz zamiast 2,5MB/s będzie jakieś 3?

Czyli teraz zamiast 2,5MB/s będzie jakieś 3?

Test RAW-READ z dysku na Efice w PIO-0 daje wynik 5,2 MB/s. Oczywiście niestety obciąża to procesor.

O, czyli co, jednak da się coś lepszego z 5200B wyciągnąć niż PIO-0?

wg tego tekst pod linkiem co tam pisze, transfer do 33 MB/s


Ostatnia aktualizacja: 14.07.2012 00:19:30 przez ] SKOLMAN_MWS ˇ agrEssOr [

niż PIO-0?

"Actually the first drivers out there were written by bplan, and were as fast as they get. That is to say, actually pretty slow (PIO4)" - tekst pod linkiem

no właśnie, PIO-0 nie bardzo pasowało, bo na FastATA z BlizzardPPC jest podobny transfer z dysku.

Wyciągnąć się da, ale czy będzie działać bez corruptowania danych to inna sprawa.

"The drivers work fine."
Matt Sealey, Genesi USA Inc.
http://www.powerdeveloper.org/forums/viewtopic.php?p=13498#p13498

Po pierwsze w MPC5200B jest bug, o którym z resztą mowa nawet w komentarzach tego sterownika,

Na temat komentarzy w sterowniku.

"Not entirely true.

1) The Efika has the signals routed. DMA "works" if you turn it on

2) LocalPlus and ATA aren't used together on the Efika (LocalPlus and PCI are mutually exclusive, and PCI doesn't cause the problem) so this is not causing broken DMA.

3) the ATA state engine bug is NOT a documented errata. Freescale's semi-official stance on it is, yes, it is broken, but who cares, this is an automotive chip and you probably won't be using ATA drives on it anyway. If you get it to work, good luck!

Remember ATA DMA is not there to improve the speed of the transfer but to work around the IDE cable being badly designed (basically UDMA is DDR MWDMA), and reduce CPU usage which helps to keep a reliable speed under load. MWDMA is not any faster than PIO, and UDMA2 is 33MB/s which is actually faster than most 2.5" magnetic media operates. If you factor in access latency on the drive and random access, you will not see much improved performance in these modes whether you use DMA-assisted-PIO (bplan style driver) or UDMA."

Matt Sealey, Genesi USA Inc.
http://www.powerdeveloper.org/forums/viewtopic.php?p=13494#p13494

po drugie design kontrolera DMA BestComm jest zaprawdę szalony.

Na temat kontrolera DMA.

"The DMA controller - BestComm - is actually pretty powerful. It is also the ONLY way you can get DMA involved in disk access reliably. The MPC5200B ATA state engine has a horrible bug in it that means that as a buffer underrun or overrun (and the FIFO is only 512 bytes so, one disk block) it does not inform the ATA DMA engine, which will simply keep pulling garbage data out of the FIFO. What it should do is insert a wait state until the underrun condition is over. Added to that the FIFO does not trigger alarms on those conditions to userspace - basically, the mechanisms to support making sure the data in the FIFO is valid and not being trampled with garbage are broken on both sides.

What BestComm allows you to do is submit a PIO job (not much CPU time needed to do this, a few IO register writes) and then use BestComm to fill and empty the FIFO in 512-byte blocks.

Yes there is ATA DMA support in the current Linux driver, but good luck getting it to work."

Matt Sealey, Genesi USA Inc.
Product Development Analyst
http://www.powerdeveloper.org/forums/viewtopic.php?p=13483#p13483