Hi Finn, On Sun, Oct 8, 2017 at 1:09 PM, Finn Thain <[email protected]> wrote: > The "hdb: lost interrupt" messages from Linux might be important. The new
Looks like a bug in ARAnyM. Haven't seen these on actual hardware for a long time (i.e. not since your NCR5380 rewrite). And the NCR5380 driver wasn't loaded in the emulator tests (and would be pointless to have there anyway). > libata driver doesn't use the ST-DMA interrupt sharing mechanism and > Adrian's results suggest this driver works okay with Aranym's IDE CD-ROM > emulation. If the libata driver works for the installer, just use that one. The only disadvantage that I see is the lack of interrupt support and the use of polling to work out whether an IDE command has completed. This might add a little additional latency but AFAIR there was no huge impact in tests. I've been working on changes to the Falcon interrupt code that would allow use of interrupts in the libata driver - these did get some testing but are not ready for review yet. So whatever latency issue there is may be possible to address. Geert had raised concerns about libata's code size and monolithic design in the past (no way to disable features from being built into the driver that are not required on m68k - has this been addressed? Am I misremembering the discussion? Regarding the legacy IDE driver - it definitely works when compiled into the kernel. No one has evidently tested it as a module on Falcon in years, but I am quite confident it is still maintained. Conversion to a platform driver doesn't seem too difficult... Cheers, Michael
