(2.4.2x) Driver for Medley software RAID (Silicon Image 3112 SATARaid, CMD680, etc?) for testing
Hi Jeff,
Quoted 1 line · Jeff Garzik
Did I miss your response to Arjan's question?
Yes you must have (I replied off-list by mistake and then forwarded to the list afterwards).
Quoted 6 lines
Since your patch removes "silraid" driver, that introduces user transition problems and potential breakage for working users.
So, why did you replace
silraid.c? Linux kernel development prefers "evolution" (changing existing code over time) to "revolution" (continually rewriting everything).
I was under the impression that silraid.c never worked at all (it certainly didn't on my system), because the magic number it looks for is incorrect (actually, part of the drive's serial number), and thus not present on all systems with this RAID).
I have since found out that some people have been using Arjan's driver with success, so my final version of the patch will certainly leave it in so people can migrate to the new driver at their own pace.
Regards,
Thomas
Quoted 4 lines · Thomas Horsten
I was under the impression that
silraid.cnever worked at all (it certainly didn't on my system), because the magic number it looks for is incorrect (actually, part of the drive's serial number), and thus not present on all systems with this RAID).
well... that's a 1 line fix to silraid.c......
Quoted 6 lines · Arjan van de Ven
Quoted 4 lines
I was under the impression that
silraid.cnever worked at all (it certainly didn't on my system), because the magic number it looks for is incorrect (actually, part of the drive's serial number), and thus not present on all systems with this RAID).well... that's a 1 line fix to
silraid.c......
Not really, it needs to look up the PCI device ID of the card to do it (of course still just a few lines).
But the reason for rewriting the driver was that I wanted to use the specification as a guideline instead of the old implementation which was really for a different RAID type (with the cutoff handling of the differing disk sizes etc).
In hindsight, after the driver is "completed" at least for RAID0, I can see that modifying the old driver could have been just as good (although I'd still advocate renaming it), but now it is done like this and I think the new driver is certainly a better starting point for Medley (e.g. if one wanted to add support for the striped modes etc).
Keep in mind that I didn't know anyone had successfully used silraid.c since on my system it a) looked in the wrong place for the superblock and
b) looked for a wrong magic number, and I didn't get any response when I provided you with a lot of details about this.
The detection of the arrays is certainly safer, and the make request routine is improved in my driver so bottom line I thik it's a better driver and will also provide a better starting point for adding stuff like RAID1 and RAID0+1 support (although one could argue that the whole ataraid method should be regarded as a temporary solution and there would be ways to integrate support for these software RAIDs better in the future).
Regards,
Thomas