Date: Thu, 25 Mar 2004 15:28:03 -0800 (PST) From: Doug Ambrisko <ambrisko@ambrisko.com> To: freebsd-stable@freebsd.org Subject: Announce FreeBSD -stable Promise SATA and Intel ICH SATA patches Message-ID: <200403252328.i2PNS3cY093322@ambrisko.com>
next in thread | raw e-mail | index | archive | help
I've added support for the Promise SATA TX2/TX4 cards and Intel ICH5/6300ESB SATA chips. The big challenges that SATA errors tend to lock-up the system unless you read the SATA registers or deal with the controller errors via their error registers. If you ignore them and just continue with normal ATA controller commands the system will lock up. Some functions need to be renamed. The patch set it at: http://www.ambrisko.com/doug/ata/ata_stable_sata_2.patch and applies src. A make world is a good idea. Here is the README at the top of the patch set: The basic Promise support came from -current. Everything else was developed in -stable and trying not to change the core -stable code to reduce maintainance burdens. The goal was to make the most minimal changes to -stable as possible and at the same time create a robust system to deal with disk failures. Thanks to Promise for programing information. ICH SATA support came from the Intel doc's. Features added/bug fixes: - Support for Promise SATA TX2/TX4 controllers mostly from -current - Support for reading hot-swap evens and tearing down the device when removed. A few kernel panics fixes when things disappear. - Changing the ata system so it mostly runs out from under splbio so that we can get interrupts to tell us a drive is gone. If that is ignored then the controller can lock up the system. This required a work thread to run through channel requests. A shim to ata_start was used to implement this. This gets the code-excution path away from splbio. - Clamping the size of the disk when making a RAID. - The ATA_IN/ATA_OUT macros were re-done so that the existing code would not need to be modified. It was was more space in the softc to hold the arrays for the memory regions. This is better to reduce maintainance burden. Minor code changes in allocation clean-up had to be changed. - MFC of addspare. - add support for SATA error handling of ICH5 & 6300ESB (we need to read the SATA registers. Note we could detect hot swap if we polled on the sstatus register. Map SATA drives into which parts of the controller they are on. - the Promise channel needs to be reset on error (such as a media error) or it will hang on further I/O (ie. in ata_intr) - Fixed a PIO bug in which there is an under-flow & buffer over-flow bug. - On media error don't fall back to PIO on some the Intel SATA or Promise SATA devices. This doesn't help much and kill performance. Fail out I/O straight from DMA mode gracefully. - Add some more ICH device ID's. Update to new ICH UDMA flags. Potential Issues/brokeness - kernel core dumps should work not tested - Tagged Queuing should work not tested Known issues: - The RAID doesn't automatically rebuild need to run "atacontrol rebuild ar0" sort-of works. Let me know of any success or failures using this patches. I will be working on bringing these features into -current. My target deployment for this was -stable. Thanks, Doug A.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200403252328.i2PNS3cY093322>