2.4.20-ben1 : Initial release, usual pmac stuff not yet merged
              upstream like UniNorth AGP, updated dmasound, ...

2.4.20-ben2 : - Reset sound chip during sleep on TAS based machines,
                should avoid the occasionnal annoying hiss on wakeup
              - Tweak radeonfb sleep code to not put the chip to D2
	        state with Radeon M7. This helps sleep to work with
		some models, but definitely doesn't work if you use DRI.
		Still waiting for a real solution...
              - Fix a bug with IO resource allocation on some
                PCI<->PCI bridges
              - Fix a problem with radeonfb on some DVI cards
              - Fix briQ serial port setup & compile with CONFIG_VT
              - Ask OpenFirmware to open all "display" devices on boot
                (should help with getting multi head cards fully POSTed).
              - Include Anton's strace fix (don't clobber ptreg->grp0 on
	        syscall return).

2.4.20-ben3 : - Make atyfb sleep code more robust (prevents any access
                to the hardware when it is asleep)
              - Don't allow growing stack on reads
	        this fixes a problem with mount() that was typically
		triggered when using prelink
              - Apply -aa patch "silent-stack-overflow" that puts a guard
	        page between last core VMA and stack to catch stack
		overflows
              - Slight changes to try to make OHCI USB sleep code more
	        robust.
              - Fix off-by-one error in battery percentage returned by
	        apm_emu (Andreas Schwab)
              - Fix a problem with one shot interrupt URBs in usb OHCI
	        driver (Joachim Achtzehnter)

2.4.20-ben4 : - Real USB fix for one short interrupts (was correct in
                rsync tree, not on kernel.org)
	      - Add support for machines based on the "Intrepid" chipset
	        (apparently a U2 + KL "combo", kind of new Pangea).

2.4.20-ben5 : - More fixes to IDE + highmem (specifically fixes issues
                with ide-scsi).
              - Include fix for ext3 scheduling storm (from 2.4.21-pre4)
	      - Include fix for fs/namei.c memory leak from 2.4.21-pre3
              - Include for for scsi duplicate IDs from 2.4.21-pre3
	      - Include fix for incorrect handling of error code in SCSI
	        from 2.4.21-pre
              - Include fix for ext3 free-after-use bug from 2.4.21-pre

2.4.20-ben6 : - Add a mecanism for userland to retreive ADB original
                address of devices
              - Map Fn-Cmd to something else than Power on PowerBooks
	        and properly remap Power key on 2400/3400/3500 models
              - Fix unistd.h syscall macros for clobber & gcc3
              - Strip FCS on incoming frames in sungem
              - Fix "advertising" setting returned to userland with
	        ethtool ioctls. Ethtool itself used that to setup
		advertising when enabling autoneg.
              - More fixes to sungem link stuff, better handle parameters
	        passed by ethtool, don't lose them on link change,
		better validity checking based on PHY type
	      - Improve code that copies CPU settings from CPU 0 to
	        CPU 1 on pmac SMP boot
	      - Add support for cpufreq on "intrepid" machines