Hello
I pointed out this in one or two or my previous topics as an additional information but this time it became a real problem. The situation is as follows:
I'm using McASP as a PCM transmitter and receiver on the DSP side of the OMAP-L138 system. EDMA configuration for McASP is done through EDMA3 LLD. In general I'm using 4 DMA channels and a significant number of Linking PaRam sets on the DSP to accomplish some data sorting for my PCM TX/RX. When I print what channels and their PaRAM set addresses I'm using on the DSP I see an inconsistency:
For DMA Channels all is in order: e.g. for one of McASP event channels numbered '1' I get an address of 0x1c04020 (base is 0x1c04000 and offset 0x20 since it is channel number '1' not '0').
For Linking Channels on the other hand I see something like this: channel number 64 (first one given to me by EDMA3_DRV_LINK_CHANNEL address is 0x1c04400 and should be 0x1c04800, since 64*0x20 is 0x800).
Normally I wouldn't bother, but when Linux uses EDMA for a transfer to/from MMC card it uses first free channels it sees - from 32 and on for linking purposes. As you probably have guessed those channels have PaRAM set addresses 0x1c04400 and on causing my PCM transfer to fail.
Is it because:
1. EDMA3 LLD for some reason numbers Linking Channels in a different way than Linux does (by a margin of 32 for OMAP-L138)? Moreover EDMA3 was not designed to work in parallel with Linux driver (or the other way around - Linux driver was not written to work in parallel with EMDA3)?
To support this I examined the source code of EDMA3 LLD and concluded that I'm getting the channel numbering from 64 since EDMA3_DRV_LINK_CH_MIN_VAL is set to 64, as well as EDMA3_MAX_DMA_CH (even though OMAP-L138 has 32 DMA channels for one EDMA instance). 64 is used as a channel number for the first linking channel for compability, but in fact PaRAM set addresses are counted as for a 32 DMA system?
2. There is a bug in EDMA3 LLD in assigning PaRAM set addresses for Linking Channels (since Linux does it really simple through a macro #define PARM_OFFSET(param_no) (EDMA_PARM + ((param_no) << 5))?
Versions of software:
Linux version 2.6.32-rc6 (szymonk@szymonk) (gcc version 4.0.0 (DENX ELDK 4.1 4.0.0)) #66 PREEMPT Tue Jun 8 12:13:36 CEST 2010
edma3_lld_01_10_00_01
I would appreciate your thoughts on that.
Regards
Szymon
