The last time I messed with the graphic drivers in this ATI RS780-based laptop, things didn't go so well with fglrx and I had to go back to the X.org radeon module in order to protect whatever piece of hardware made a buzzing sound with ATI's proprietary drivers.

I knew that radeon occasionally failed to restore non-X framebuffer consoles and they become black screens. Now I've unfortunately confirmed that this does not affect just the framebuffer (fbcon/vesafb) consoles, but also plain text-mode consoles. It's probably a known bug that this happens under some conditions, however this snippet of code in radeon's source code (src/radeon_driver.c) does not give me much hope since it's been around since 2003 according to the commit logs:

#if 1
    /* Temp fix to "solve" VT switch problems.  When switching VTs on
     * some systems, the console can either hang or the fonts can be
     * corrupted.  This hack solves the problem 99% of the time.  A
     * correct fix is being worked on.
     */
    usleep(100000);
#endif

Interestingly radeon inserts long extra delays (thousands of µsec) for many operations “just in case”, whereas radeonhd uses delays of less than 10 µsec in all but one case.

Nevertheless, I've not stumbled upon any race condition when resuming X.org from hibernation with this driver, although it's only had chances for acting up during 5 days — that is, 5 hibernation cycles. When it was the time for the 6th cycle, I upgraded Squeeze's X.org synaptics driver and restarted the X server, and it mysteriously locked up forever, forcing me to reboot the whole system — although there was still a process listening to the laptop's power button event so it was a safe shutdown. Since X stopped logging when it locked up, I cannot really prove that radeon wasn't at fault there, but I strongly suspect it was the synaptics driver upgrade what caused it. The kernel logs don't indicate anything unusual at that time either.

And in other news, the Linux kernel version 2.6.32 has been released, and one of the many changes in this version includes Kernel Mode-setting for the ATI R7xx chipsets. I guess I'll check it out once there's a Tux-On-Ice patch for it and I feel like replacing Debian's libdrm with an upstream version.

UPDATE: there's a Tux-On-Ice patch (3.0.99.32) for this version now. You know what that means. Let's rock!