Soeren Schmidt announced this weekend that ATAng has been merged into the FreeBSD -current kernel tree. Described by the author as "rather radical changes to the ATA driver", ATAng offers a number of impovements over the old ATA driver. This includes removal of the 'GIANT' lock, an improved framework that supports newer ATA controllers, merging of ATA and ATAPI code, and the removal of numerous bugs.
From: Soeren Schmidt [email blocked] To: current Subject: TESTERS WANTED for ATAng preview 1 Date: Fri, 8 Aug 2003 21:55:11 +0200 (CEST) The first preview release of ATAng is now available on: ftp://ftp.deepcore.dk/pub/ATAng >From the README there: ATAng preview 1 release Before these rather radical changes to the ATA driver hits the tree, here is the opportunity to test them out, give usefull feedback and for the depending subsytems to adjust to the new ways of things (burncd & atapicam are good examples). Now why all these changes suddenly ? 1. Get GIANT out of the ATA driver. This was pretty tedious in the old version (I tried that long ago and found it way too painfull). With this new structure it was a walk in the park since handling of ATA commands/requests are now split up in a "queue" part and a "lowlevel" part. This makes the places where locks are needed a lot more obvious to locate, making it easier to implement and maintain. 2. Provide the framework for supporting new ATA controllers that has facilities for chaining commands and HW XOR's etc. This called for some radical changes to the basic framework which fitted nicely with item 1 above. Most of the hooks needed for "exotic" HW are now present, and the next step will be for drivers to use this to get an edge over conventional ATA controllers. Promise Inc. has kindly provided HW and documentation for their latests chips that has lots of interesting features that we can now start to use. Much more on this when these basic bits has settled.. 3. Merge of ATA and ATAPI code. As an effect of the new framework there is now a single entrypoint for ATA/ATAPI commands. This will extend to userland where it will be possible to issue both ATA and ATAPI commands just as before with just ATAPI ones. This will add the ability to do all kinds of stuff from userland, like SMART etc, however it will also provide a nice way of trashing your system :) The code for this is not included in this preview as I still need to settle on the API for this. 4. During all this lots of bugs and problems in the old code has been found and fixed. 5. The way ATA RAID support works today need to change as well to make full use of fx Promise's new chips. This is NOT done yet, in fact the ata-raid.? driver is still the same, bugs and all. To try this out you need to apply the "conf-patch" file to your sys tree then delete the contents of dev/ata and put the contents of the ATAng-<timestamp>.tgz in there instead. You might want to include "device ataraid" in your kernel config file if you use ATA RAID's. Config & make your kernel as usual. Now, some things are not completly done yet, some things will not be done at all and there are bound to be bugs lurking, so use protective measures for safe engagement, you have been warned 8) Constructive feedback as always very welcome!! Enjoy! -Søren
From: Soren Schmidt Subject: HEADS UP! ATAng committed Date: Sun, 24 Aug 2003 11:27:05 +0200 (CEST) ATAng has just been committed. You need to make world after this update as atacontrol etc needs to pick up the changes. -Søren
From: Roderick van Domburg [email blocked] Subject: Re: HEADS UP! ATAng committed Date: Sun, 24 Aug 2003 14:49:23 +0200 > ATAng has just been committed. You need to make world after this > update as atacontrol etc needs to pick up the changes. Coolness. Judging from the CVS log, it also gets ATA out under Giant? That should make ATA-run SMP boxes eligible for an impressive increase in I/O performance... has anyone noticed? Regards, Roderick
From: Soren Schmidt [email blocked] Subject: Re: HEADS UP! ATAng committed Date: Sun, 24 Aug 2003 15:28:45 +0200 (CEST) It seems Roderick van Domburg wrote: > > ATAng has just been committed. You need to make world after this > > update as atacontrol etc needs to pick up the changes. > > Coolness. Judging from the CVS log, it also gets ATA out under Giant? Yes it does. > That should make ATA-run SMP boxes eligible for an impressive increase > in I/O performance... has anyone noticed? Depends alot on what you are doing.. -Søren