On Tue, Jul 23, 2013 at 4:08 PM, Michael Schuster michaelschuster@email.de wrote:
Dear Mr. Deucher,
I received a reply from Mr. Airlie with your email as developer of the radeon kernel driver. I don't know if he send you the both emails and (patch) attachements, if not, please send me a short note than I would like to send you some information about me why I decided to try to develope a patch...
I only saw the message he cc'ed me on. I haven't seen any of the previous patches or emails.
Attached is the actual version of the patche, changed some lines e.g. in atombios.c and radeon_pm.c (and corrected some off/on typos in evergreen.c) to start he asic with AUTO/BALANCED power mode using LOW clocks if some power modes are available. By now I unsuccesfully tried to autosuspend my muxless (was my real initial intention to develope the patch) from the start but did not find an succesful way to do it...
You can use vgaswitcheroo to disable the dGPU on your system if you aren't using it.
Hope he patch will do some good (almost byte-wise) parsing the PowerPlayTable used for radeon_atombios_parse_power_table_4_5 (and probably assembling missing power modes) for proper use in 'static mode' power management.
It's not entirely clear to me what your patch is doing. I think you might be overcomplicating things. The power tables are designed with the DPM (Dynamic Power Management) hardware in mind, so they don't match perfectly to the the older static levels from older asics. Now that we have DPM support (as of 3.11), I'd like to eventually deprecate the the old static profiles on asics that support DPM so I'd prefer not make any major changes to that code to avoid any regressions.
By the way, the 'rom' file size from (debian 7, ext4) is 128k, 'rom' is not readable after echo 1 > rom may be it depends of my Samsung 700Z3A having an intel integrated graphics card and a (muxless) HD6490M as a second graphics card.
In hybrid laptops the vbios for the discrete card is stored in ACPI, you can't just dump it from the pci rom. You'd need to dump it from the driver itself or provide an interface in the driver to access it via debugfs or sysfs.
Corrections und hints for learning C / doing kernel module programming the right/code safe way are highly appreciated, thanks a lot in advance, best wishes and regards,
Best advice is to look at other kernel drivers and see what they are doing and how the code is written for pointers on style.
Alex
Michael Schuster
post scriptum: the patch is 66k plain text, I attached it because of the size, hoping this will be okay for you ... - M.