The Linux Ultra-DMA Mini-Howto Brion Vibber, brion@pobox.com v1.3, 5 March 1998 This document is intended to explain how to use Ultra-DMA aka Ultra- ATA aka Ultra33 hard drives and controllers with Linux. The most recent version of this mini-Howto can be obtained in HTML format at http://pobox.com/~brion/linux/Ultra-DMA.html. 1. Introduction This document is intended to explain how to use Ultra-DMA aka Ultra- ATA aka Ultra33 hard drives and controllers with Linux. In some cases there is no difficulty in using them, but some tweaking can increase performance. In other cases, you need to go to extraordinary lengths simply to access your hard drives. 1.1. Disclaimer The information in this is document is, to the best of my knowledge, correct, and should work. However, there may be typos, there may be mysterious transmission errors, and there may be strange incompatibilities within your own system that prevent the techniques described herein from working properly. So... before you go fiddling around with you hard drive, BACK UP ANY DATA YOU WANT TO KEEP! 1.2. Credits Brion Vibber (brion@pobox.com) - The document itself Gadi Oxman (gadio@netvision.net.il) - The Promise Ultra33 patch & finding the secret numbers for the workaround John G. (prefect@ipass.net) - VIA VP2 patch & info Giovanni (giovanni@sudfr.com) - UDMA-enabled VIA-related patch & more info Martin Gaitan - Promise ide0/ide1 workaround Norman Jacobowitz - Bugged me to add info on the VP3 1.3. Document History v1.3, 5 March 1998: VIA VP3 info, better patching instructions, pointer to more recent Promise patch. v1.2, 27 January 1998: Additional Promise workaround info. v1.1, 21 January 1998: New info about VIA chipset, installing around the Promise Ultra33, and enabling Bus Master & UDMA transfer modes. v1.0, 19 January 1998: More or less complete, first version done in SGML. 2. What is Ultra-DMA and why do I want it? Here's a brief overview of IDE-based drive technologies: 2.1. IDE, EIDE, & ATAPI These are older drive technologies. Most non-SCSI hard drives and drive controllers that you can buy today or are likely to be using are EIDE, although many of the larger drives now available are UDMA. 2.2. Bus Master DMA Bus Master DMA is a technology for increasing the speed of hard disk data transfers which requires support from the motherboard and the BIOS, and at least some support from the drive. You can learn more at http://developer.intel.com/design/pcisets/busmastr/FAQs.htm. 2.3. Ultra-DMA aka Ultra-ATA aka Ultra33 aka... Ultra-DMA has many names, but we'll just call it UDMA in here. UDMA is a more advanced technology which provides for even faster throughput, up to 33.3 MB/s, twice that of EIDE, for much lower prices than SCSI. Many new computers come with large UDMA drives and UDMA controllers, and it's possible to add a UDMA controller card (such as the Promise Ultra33) to an existing system to boost speed, even on older non-UDMA drives. You can learn great details about UDMA at http://www.quantum.com/src/whitepapers/ultraata/ 3. Using your UDMA Hard Drive with an EIDE Controller This is easy to do. Since all UDMA drives are fully EIDE backward- compatible, just plunk your drive on your EIDE controller like it was any old hard drive & Linux should have no problems detecting or using the drive. However, you will of course be limited to the 16.7 MB/s bandwidth of EIDE. 4. Using your hard drives with a UDMA Controller Well, there's good news and there's bad news. The good news is that a UDMA controller can be used with both UDMA hard drives and legacy EIDE hard drives, and will be a lot faster than an EIDE controller. The bad news is that they're not technically supported by the current stable releases of Linux (2.0.x kernels). In some cases the controllers can be used in legacy (EIDE-compatible mode) with no difficulty, in others simply using the controller requires a bit of effort. In all cases, in order to use UDMA transfer speeds, an appropriate patch is required. The following sections will each pertain to a specific type of UDMA controller. 5. The Promise Ultra33 IDE This is a PCI card that has two UDMA channels on it, supporting up to four drives. You can look up specs & pricing & such at http://www.promise.com. This card shipped in early model Gateway 2000 Pentium II systems, and may or may not be in more recent models. The more recent development kernels (current is 2.1.88) have generic PCI IDE controller support, which ought to automatically detect the Promise Ultra33. However, the stable kernels (current is 2.0.33) require a patch, and it can be a little difficult to get Linux installed because of this. 5.1. Installing Linux with the Promise Although there is a patch for the Promise controller, it's not very easy to apply a patch and recompile your kernel if you haven't installed Linux yet! So, here's a workaround which may or may not work for you. Thanks to Gadi Oxman for the following information on getting the interface settings: If we can access the console with the installation disk, we can also use "cat /proc/pci" to display the Promise interface settings: RAID bus controller: Promise Technology Unknown device (rev 1). Vendor id=105a. Device id=4d33. Medium devsel. IRQ 12. Master Capable. Latency=32. I/O at 0xe000. (a) I/O at 0xd804. (b) I/O at 0xd400. (c) I/O at 0xd004. (d) I/O at 0xc800. (e) and pass "ide2=a,b+2 ide3=c,d+2" as a command line parameter to the kernel. Note that the numbers probably aren't the same as what you'll have. Just as an example, the parameters to use for the above set of numbers would be ``ide2=0xe000,0xd806 ide3=0xd400,0xd006''. You may or may not also need to include the IRQ, which would make it ``ide2=0xe000,0xd806,12 ide3=0xd400,0xd006,12''. If you're only using the first channel on the Promise controller (for instance, if you only have one drive, or two if they're master and slave on the same channel, on the Promise), then you won't need to specify ide3. Note that you can definately get to a command prompt and do ``cat/proc/pci'' using Red Hat 5.0 or Slackware 3.4; it's almost certainly possible with other distributions but I haven't worked with them and don't know how. Under Red Hat 5.0, use the installation boot floppy, and at the boot prompt type ``rescue''. It'll load some stuff, prompt you for the supplemental disk, ask you about your monitor and keyboard, and finally put you into a command prompt. Then, do ``cat /proc/pci'', write down the numbers as above, and reboot from the boot disk. This time, type ``linux ide2=(this is where you put the numbers like shown above) ide3=(more numbers)''. It should now be able to install onto your hard disk without difficulty. Under Slackware 3.4, the process is very similar. Boot with the boot disk of your choice, and at the boot prompt, just hit Enter for now. It'll load the kernel and prompt for the root disk. Insert the root disk, wait for it to load, and log in as root. Do the ``cat /proc/pci'' thing and write down those numbers somewhere safe! Reboot with the boot floppy and at the boot prompt type ``ramdisk ide2=(this is where you put the numbers) ide3=(some more numbers)''. You should then be able to install onto your hard disk as normal. IMPORTANT: Without the patch (discussed in the next section), the kernel needs these boot parameters in order to access your hard disk! Therefore it is very important that when you configure LILO, either on the hard disk or on a boot floppy, that you give it the exact same parameters that you gave when installing. Otherwise your system won't boot! It should be possible to give them to LILO when you boot (ie, press Shift, type in ``linux ide2=.....'' each time you boot), but only if you kept the numbers! It is recommended that you patch your kernel as soon as possible so you won't have to worry about that anymore; once you're booting off a patched kernel, you can get rid of the boot parameters. Also, as far as I know there is no way to pass boot parameters to a plain kernel boot floppy (as made with ``make zdisk''), you must use LILO or another loader that lets you pass boot parameters. However, unpatched kernels often have a difficult time actually using ide2 and ide3, even if the drives are detected properly. So if you can't get Linux to install using the above technique, try specifying ide0 or ide1 instead of ide2 or ide3 (thanks to Martin Gaitan for this technique). This essentially replaces the on-board controller (if any) with the Promise Ultra33 as far as the kernel is concerned, and you can follow the directions in the next section as if you had physically moved it. Note that if you're using an IDE CD-ROM drive connected to your on-board controller to install from, you'll want to make sure that you don't take over the controller that the CD is on or you won't be able to install! If the CD is hda or hdb, use ide1 for your hard drive, and if it's hdc or hdd, then use ide0. 5.2. Installing Linux Around the Promise If you can't get the software workaround to work, you'll have to try a more brute force approach. Here's an alternative method mthat is virtually guaranteed to work, but will require you to open up your computer and mess about in it. NOTE: If you are not familiar with the process of connecting and disconnecting IDE drives, read the manuals that came with your computer, your hard drive, and/or the Promise Ultra33 before attempting this! If you screw something up and don't know how to put it back, you could end up being sorry! That being said, it's all really quite simple. Most motherboards these days have built-in EIDE controllers. Disconnect your hard drive from the Ultra33 and connect it to the onboard controller. If you have other IDE devices, such as a CD-ROM, tape, or ZIP drive, on your oboard controller, it's easiest if you either add the hard drive on an unused channel (the secondary instead of the primary) or temporarily displace a device that you don't need immediately (such as ZIP or tape). Install Linux. Download and apply the Promise UDMA patch (see next section). Now you're ready to move the drive back onto the Promise... almost. To be safe, make a kernel-image boot floppy (cd /usr/src/linux ; make zdisk), which you will be able to use to boot your system in case LILO doesn't work. Actually, to be very safe, make two and put one away for now. Okay, now it's time to think a little... if you have just one hard drive and it's going to be on the Promise, then it will most likely be /dev/hde (a and b are for the primary onboard controller, c and d for the secondary onboard controller). If you're going to put any other drives on it, then the slave of the Promise's first channel will be /dev/hdf, the master of the second will be /dev/hdg, and the slave of the second will be /dev/hdh. Edit /etc/fstab, and change all the partitions of the hard drives you're moving from the onboard drives (/dev/hda, hdb, etc) to their new locations on the Promise (/dev/hde, hdf, etc). If you had to displace any devices (such as a CD-ROM or ZIP drive) that you want to leave on the onboard controller, then change them to their new locations as well. For instance, if your CD-ROM was originally the master on the primary channel (/dev/hda), but you put your hard disk there and had to bump the CD to the slave (/dev/hdb) or to the secondary channel (/dev/hdc), and now you want to put it back, then change it to /dev/hda. If you're using LILO, reconfigure LILO to use the new location of the drive (LILO configuration is beyond the scope of this document, if you don't know how, read the LILO mini-HOWTO ), or else it probably won't be able to boot unless you use that boot floppy I had you make, which you will also want to configure to boot off the new partition. This is done using the rdev command. Put the floppy in the drive and type ``rdev /dev/fd0 /dev/hde1''. Of course that's assuming your root partition is the first on your first UDMA drive. If not (mine is /dev/hde7, for instance), then obviously use the appropriate partition number! Reboot. Your system should now work fine. 5.3. Patching for the Promise There is a kernel patch available for the Promise Ultra33, written by Gadi Oxman (gadio@netvision.net.il). The current version of the patch works on kernel version 2.0.33 and can be obtained from Linux Mama . Note that you will need to use bzip2 to decompress this patch rather than gzip. You can also download a version of the patch for kernel 2.0.30, compressed with gzip, from http://pobox.com/~brion/linux/promise_update.gz. This version will work on 2.0.33 but will result in ``unknown PCI device'' messages because of a change in pci.c between 2.0.30 and 2.0.33. For instructions on how to patch and compile the kernel, make sure you have the kernel source and gcc installed and read /usr/src/linux/README. Wondering why you should bother with the patch, aside from saving yourself the trouble of the boot parameters from the workaround? Well, according to the author: One advantage of using the kernel patch is that we will be able to use bus mastering DMA transfers ("hdparm -d1 /dev/hdx") in case the triton driver is compiled into the kernel. In other words, you'll not only be able to use your hard drive, you'll be able to use it at the high speeds that UDMA is supposed to bring! 6. Intel TX motherboard onboard UDMA controller Thanks again to Gadi for this info: Bus mastering DMA support for the Intel TX chipset is available in 2.0.31 and above. In older kernels (such as Slackware 3.4's 2.0.30), the controller will be used in the slower EIDE mode. In either case the controller will be automatically detected by the kernel and you should have no trouble using it. As far as I know there is not a patch to enable full UDMA transfers, but having Bus Mastering is still better than not. 7. The VIA VPx and Related Chipsets This controller also can be autodetected and used in EIDE mode by an unpatched kernel, but if you've got one of these, you'll want to grab a patch so you can get faster throughput and do away with annoying "unkown PCI device" messages. One is available at http://www.ipass.net/~prefect/; it is designed for the VIA VP2/97 chipset, found on FIC's PA-2007 and PA-2011 motherboards, but may work on related chipsets. It has been reported that it functions on the newer VIA VP3 chipset, your mileage may vary. Note that this patch only supports Bus Mastering mode, not full UDMA mode, but it's still better than plain-vanilla EIDE mode. Follow the directions at the patch's site for enabling BMDMA mode. There is another patch that supports full UDMA mode at http://www.pyreneesweb.com/Udma/udma.html, designed for the VIA VT82C586B, and it ought to work on the VP2, VP3, VPX, P6 and AGP Apollo chipsets. Follow the directions for installation and UDMA enabling there, but it is recommended that you back up any data you want to keep, as there are potential problems with incompatible motherboards. But, if it does work, it should work without problems. 8. If you still can't get it to work! If nothing in this document proved helpful, or at least not helpful enough to get your machine working, your best bet is to write up a message that fully describes your difficulty, what type of UDMA controller you have, whether it's onboard or on a card, if your drive is actually UDMA or plain EIDE, exactly what configuration of drives you have, what version (distribution & kernel versions if possible) of Linux you're using, and anything else that sounds useful, and post it to the newsgroup comp.os.linux.hardware. You'll probably get some helpful suggestions soon. 9. If you have some information about UDMA stuff that's not in this mini-howto... Great! If you know something I don't, by all means send it to me (brion@pobox.com) and I'll put it in this document and update it fairly soon.