On Sat, Oct 2, 2010 at 5:06 PM, Carl-Daniel Hailfinger c-d.hailfinger.devel.2006@gmx.net wrote:
On 03.10.2010 01:28, Warren Turkal wrote:
I think that a base coreboot certification should basically state that all the hardware on the board is usable with a major free OS (e.g. Linux-based OSes like Debian, Ubuntu, and Redhat maybe).
We could maybe have extended certifications for things like non-free OS and driver compatibility.
Well, if we only care about Linux, you can avoid most (if not all) of ACPI on many machines, and you can avoid SeaBIOS as well. Heck, you could even avoid FILO and require a Linux kernel in flash. Whether such ab board would be usable for end users is a totally different question.
Booting into a certain OS is clearly not the only bar that we should be looking for. It would need to comply with various standards like ACPI, etc.
IMHO being able to install a Linux distribution from CD is an absolute must even if you only target professional users in clusters etc.
I agree.
Windows support means a board is usable by the general population, and this is something vendors care about deeply.
Out of curiosity, are there actually non-niche vendors that ship Coreboot as their system firmware?
We renamed LinuxBIOS to coreboot exactly because people said all the time "I don't want Linux", and EFI marketing would love to make fun of us if we ever said "Linux only is good enough for certification".
I believe that most of the folks who care enough to use Coreboot at this time are probably running Linux or some other free OS. I also believe that most developers who have access to systems are probably running on or have easy access to some free OS. I also know that non-free OSes are not easily available to everyone. It's not that I believe that we can't test for non-free software support. It's just that I believe that certifying the boards for Coreboot should not gate on that having been done.
Ideally, having some OS independent test suite would be the best approach I think, but I don't see that getting developed overnight. :)
Maybe we should have multiple compliance levels like "Coreboot minimal" and "Coreboot standard"? The minimal could be some set of requirements. The standard could be the minimal requirements plus some extra set of requirements. We could also have tags for the compliance. These could be used to indicate special support like Windows or something like that.
Think of "Coreboot minimal" as enough to conveniently develop coreboot further without expensive and specialized hardware. Its requirement might look like the following: * Full initialization of the cpu and supporting chipset - Intialized RAM - Cooling systems should work enough to protect the system from burning itself up * At least one of the following: - Outputs reasonable POST codes to a POST card if one can be plugged in - USB debug ports enabled, if available on the board - RS232 port enabled, if available on board - Some other agreed method for debugging the system during Coreboot execution (e.g. debug LEDs on mobo, etc.) * At least one of the following - At least one usb hub usable for keyboard, if one exists - PS/2 port usable for keyboard, if one exists * All expansion slots initialized and usable * ACPI - DSDT should exist * Able to load VGA option ROM * Able to boot into an OS not stored as a coreboot payload (e.g. boot from CD, USB, or SATA drive). For the minimal certfication, I believe that OS should be Debian since it's a well supported environment for developing coreboot.
Think of "Coreboot Standard" as basically fully tested with free OSes. Its requirements might look like the following: * all requirements of the minimal certification * Any requirement of the minimal certification that allows a subset to be implemented should implement all the sub-items. For example, both rs232 and usb debug ports should be enabled if they exists on the same board instead of implementing one or the other. * Uses tiny bootblock * Uses cache-as-ram if the system is capable * Soft poweroff works * Extended initialization of supporting chipset - Fan uses vary speed based on temperature of CPU * All legacy io ports (i.e. rs232, parallel, etc.) usable * All USB ports usable * All other external (including header) ports usable (e.g. firewire, sound, etc.) * better ACPI support - OSPM (e.g. G, S, D, and C states, etc.) fully exposed and usable * CPU frequency scaling works * After booting into Debian Linux, the OS should be able to rely on any info provided by the Coreboot and any intervening payloads should allow the system to be fully enumerated and configured as much as the hardware will allow.
Tags could be used to identify specific extended support. For instance, a system certified to boot Windows 7 could be "Coreboot minimal+MSWin7" certified or "Coreboot standard+MSWin7" certified or something like that. If anyone wanted to display the certification, they could display it with or without the tags.
Possible tags: MSWin{XP,7} ReactOS GPXE AOE ISCSI ...
We should probably note the known revisions that are compliant for each board. This doesn't mean that every revision that would pass the requirements needs to be listed. Only tested versions should be listed. Those revisions would be the recommended revisions with users using other revisions at their own risk. I could imagine something like the following for a mobo: minimal: 11 99 103 150 standard: 50 75 135 155 minimal+Win7: 77
The numbers are the Coreboot svn revisions that are tested. These revisions should probably be listed in the wiki somewhere. There is probably some better way to visualize the data. We may also need to list SeaBIOS hashes for the certified builds.
*snip*
Frankly, I think that ability to use the free drivers should be good enough. We shouldn't be hold out any kind of coreboot certification on the condition that non-free drivers work.
There are two aspects of the problem:
- We can't test everything (fact of life)
- Closed-source drivers have a huge market share, and won't go away any
time soon
Agreed. I also think that some developers won't be able to test certain non-free software. For instance, I wouldn't be able to test that Windows 7 boots. I also don't have an Nvidia card to test their drivers.
Given the fact that we can't test everything, we should make a minimal amount of compliance only include things that can be tested by most folks. I think that minimally certified hardware should really be capable enough to develop Coreboot further on the hardware conveniently. Standard certified hardware should work well for free OSes if the OSes have appropriate driver support.
Also, minimally certified hardware can support more features. For example, a minimally certified board could have logic for adjusting fan speed intelligently. It would still be minimally certified until someone developed the rest of the functionality for the standard level certification and tested it.
*snip*
Fans can be loud. If all fans run at 100% non-stop, machines can be essentially unusable for noise reasons.
While I agree with this, I think that a minimally certified piece of hardware should not need working fan logic.
*snip*
Indeed, but given that vendor code may not always be suitable for merging, do we want to withhold certification if code is available but not merged? And what happens if Kevin is on vacation?
If the code's not merged into SeaBIOS, we shouldn't certify the build. If Kevin is the only person that can merge push the canonical SeaBIOS tree, then I suppose it would need to wait on him. It's the same way that a vendor can't really claim that their hardware with compliant with Microsoft's certifications until MS signs off.
If it really becomes a problem that vendors want the certification during a time when Kevin isn't around, that would be a good problem for the project to have. :)
*snip*
Comments would be appreciated, especially on the minimal and standard certification requirements above. Also, please take a look at the tags and see if you can think of any additions.
Thanks, wt