I spent a while this weekend looking in to this; it seems that the problem is related to some AGP writes either writing corrupt data or otherwise causing an instability. The bad mode problem seems to be one symptom of this -- the modeline registers are probably receiving corrupt data. I added some calls to do a SaveMode() right after a RestoreMode() to dump both register sets, and on both a case where the X server came up fine, and where it crashed, the registers written/read were identical (between the two cases). No modeline corruption was displayed during the lockup -- the monitor simply didn't receive any signal. I've also yet to experience the lockup in LoadCursor that sct mentioned -- it gets through that fine for me. It's hard to figure out exactly where the crash is occuring, since any debugging output from either the drm module or X itself get stuck in a buffer at that point. All I know is that, for me, it's consistently after the mouse initialization step in X. If the server does come up, it is very prone to crashing upon ctrl-alt-backspace; only once when the server has come up did it successfully restore the text mode and exit cleanly on ctrl-alt-backspace. I've disabled AGP Fast Writes and dropped down to AGP Mode 1 in the BIOS (incidentally, linux didn't pick up the AGP aperture size and mode settings from the bios -- it claimed that it was 8MB and 1, when the settings were 64MB and 4; I manually set these in XF86Config. Is this indicative of another problem?) Anyway, I'll keep plugging away at this -- any thoughts or ideas would be much appreciated. - Vlad _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/dri-devel