Found these by inspection: Command issuing goes via ata_qc_reinit() which sets up the device bits for the command to include the device select bit. If we are using NCQ then the code in ata_build_rw_tf sets bit 6 directly without using |= which clears the device select bit and means any NCQ command will go to the first device regardless. The non NCQ path thankfully doesn't blat the other bits and right now we have no slave devices on NCQ supporting hardware I believe. The same bug is present in the libata patches for HPA, and it pulls the primary HPA blindly in each case. I've not found any others but they may be there and the code is a bit fragile around here. I suspect this is the Macintosh problem. Is there a reason we can't make exec_internel and/or qc_issue BUG() if the passed device and tf device bit disagree ? There is another ugly here too ata_tf_init sets the device bits but doesn't set TFLAG_DEVICE. Scarily in fact the qc_reinit path that produces new qc structures doesn't touch tf->flags at all but leaves them as they were previously (with or without device). Any command issued with a device set but the flag forgotten is going to work *until* we issue a command to one device and forget the flag, after a command to another device, in which case the wrong device will get the command. There are other flags that will cause lunacy to occur as well - LBA48 being an obvious one that will blow up in our faces. Follow the path of a start_stop command - we set the device flag but we assume the newly allocated command flags started clear, not random. We don't clear the LBA48 flag either.... Alan - To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
