On Fri, 12 Sep 2008 21:19:20 +0400, Sergei Shtylyov <sshtylyov@xxxxxxxxxxxxx> wrote: > > Well, let me explain a bit. The datasheed say I should wait _both_ > > XFERINT and HOST interrupt. So, if only one of them was asserted, I > > mask it and wait another one. But on the error case, only HOST was > > asserted and XFERINT was never asserted. Then I could not exit from > > "waiting another one" state, until timeout. > > Hmm, I got it: you decide whether it's worth waiting more for XFEREND > interrupt based on whether ERR is set or not. I suppose IDE_INT doesn't get > set in case the command gets endede with ERR set? IIRC, yes. And anyway, the interrupt signal from this controller to CPU is not asserted because HOSTINT was masked by int_ctl register to wait for XFERINT interrupt. So, regardless of IDE_INT was set or not, no more interrupt raised to CPU. Many of strangeness of interrupt handling in this driver is based on the fact that the IDE_INT bit in DMA status register does not refrect the controllers interrupt status directly. And the implementation of the IDE_INT bit is actually broken. Claring the IDE_INT bit also clears all mask bits in int_ctl registers. Usually this sort of behaviour is called "bug". ;) --- Atsushi Nemoto
