Kernel Traffic #248 For 20 Jan 2004

I was going through KT247 and noticed the summary of the real-time maintainership discussion. Most of the summary is right on the spot and I have nothing to add about it. The only thing I'd like to bring to your attention is Linus' statement on the issue.

As I pointed out in my response to him (http://marc.theaimsgroup.com/?l=linux-kernel&m=107208752427058&w=2), Linus is making major mistakes on this issue. For example, the mail you quote has him saying: "It's doubly discgusting with some of the people who were trying to spread all the FUD and mis-information were doing so because they were themselves doing a non-GPL microkernel, and they complained about how the patents were somehow against the GPL and wanted to get community support by trying to make out the situation to be somehow different from what it was."

Yet, neither I or anyone involved in this debate has ever asked for a non-GPL microkernel. In fact, I was quoted by KT in June 2002 (http://kt.zork.net/kernel-traffic/kt20020609_170.html#1) as saying: "We don't want a non-GPL real-time executive or a non-GPL OS. All we want is the right to develop applications using our licenses as others are for Linux."

In addition, Linus' claim that I'm somehow spreading FUD is contradicted by Victor Yodaiken's own statements on the issues, as I explained in my reply to Linus (see above URL.)

I do understand that given Linus' community stance, his word is worth reporting. That being said, Linus' stance doesn't mean he's infallible and I think KT readers would profit from having the full facts presented to them, even when such facts show that a prominent figure of the community is mistaken.

Good eyes - I only analysed the ctype.h thing, and didn't look up errno.h in the original sources. errno.h has a _big_ comment saying where the numbers came from (and some swearwords about POSIX ;)

Looking at signal.h, those numbers also seem to largely match minix. Which makes sense - I actually had access to them.

In both cases it's only the numbers that got copied, though. And not all of them either - for some reason I tried to make the signal numbers match (probably lazyness - not so much that I cared about the numbers themselves, but about the list of signal names), but for example the SA_xxxx macros - in the very same file - bear no relation to the minix ones.

Just to shore up what jw was talking about:

http://www.opengroup.org/onlinepubs/007904975/basedefs/signal.h.html
http://www.opengroup.org/onlinepubs/007904975/basedefs/errno.h.html
http://www.opengroup.org/onlinepubs/007904975/basedefs/ctype.h.html

and for a listing of all the defines the open group has (dir listing):

http://www.opengroup.org/onlinepubs/007904975/basedefs/

The comments were added in Linux 0.99.1, and I'm not sure what was the source. For example, Linux has:

#define ENOTTY          25      /* Not a typewriter */

Solaris:

#define ENOTTY  25      /* Inappropriate ioctl for device       */

Current POSIX:

    [ENOTTY]
        Inappropriate I/O control operation.

I couldn't find any historic Minix header files. Minix 2 has:

#define ENOTTY        (_SIGN 25)  /* inappropriate I/O control operation */

In http://www.opensource.apple.com/darwinsource/DevToolsMay2002/gcc-937.2/libiberty/strerror.c

/* Extended support for using errno values.
   Written by Fred Fish.  fnf@cygnus.com
   This file is in the public domain.  --Per Bothner.  */
(...)
#if defined (ENOTTY)
  ENTRY(ENOTTY, "ENOTTY", "Not a typewriter"),
#endif

FYI there was a proposed patch to change "Not a typewriter" to "Inappropriate ioctl for device". Check the interesting thread of lkml: http://www.ussg.iu.edu/hypermail/linux/kernel/0105.1/0471.html

Something like that may well be the source of the string. Fred Fish was active long before this timeframe (if it's the same Fred Fish - he used to do freeware collections for the Amiga in the '80's).

But there were multiple libc's around (estdio, libc5, glibc..), and it could be any of them.

Trying to find the kernel list archives from that timeframe would likely clarify the issue. There were several lists back then: "linux-activists" mailing list, and of course the "comp.os.linux" newsgroup (this was before it split into multiple newsgroups).

I've found some archives for linux-activists, but no newsgroup archives going that far back.. Anybody?

http://groups.google.com/advanced_group_search

They'd done what DN had promised and never did - merged old usenet archives into their database, all way back to '81.

The earliest they have on comp.os.linux is March 1992 and results of voting on newgroup had been posted by tytso on Mar 25 1992. There's a bogus crosspost into (still not existing) c.o.l on Mar 21 and regular postings starting on Mar 31. IOW, archive goes all way back to the group creation.

alt.os.linux archives there start on Jan 19 1992 (newgrouped somewhere around Jan 17, took several days to propagate). Before that it's c.o.minix, but by the time you are looking for migration from c.o.m should've been over.

the linux-activists messages for that period would be in:

http://www.ibiblio.org/pub/historic-linux/ftp-archives/tsx-11.mit.edu/
Oct-07-1996/mail-archive/linux-activists/Volume2/Linux-V2-5XX.tar.z

Specifically, the files "digest51[789] digest52[012]" are the neighborhood around july 25th. I did a quick grep for "errno" and a few strings from the errno.h file and didn't see anything relevant pop up, though.

Ok, this is the source.

In particular, I can re-create _exactly_ the linux-0.97 "errno.h" file by using the "sys_errlist[]" contents from "libc-2.2.2". In particular, this trivial loop will generate the exact (byte-for-byte) list that is in the kernel:

        int i;

        for (i = 1; i < 122; i++) {
                const char *name = names[i];
                int n = strlen(name);
                char *tabs = "\t\t"+(n > 7);
                const char *expl = libc222_errlist[i];
                printf("#define\t%s%s%2d\t/* %s */\n",
                        name, tabs, i, expl);
        }

here, the "names[]" array was filled in with the error names, ie

        const char *names[] = { "none",
        "EPERM", "ENOENT", "ESRCH", "EINTR", "EIO", "ENXIO", "E2BIG",
        ...

and the "libc222_errlist[]" array was filled in with the strings found by just downloading the old "libc-2.2.2" binary that can still be found at

http://www.ibiblio.org/pub/Linux/libs/oldlibs/libc-2.2.2/

and then just doing a "strings - libc-2.2.2" and "sys_errlist[]" will be obvious:

        static char *libc222_errlist[] = {
                "Unknown error",
                "Operation not permitted",
        ...

This was literally a five-minute hack (I wrote the silly loop yesterday to see what it does with the current "strerror()" - there is very good correlation even today, but using the libc-2.2.2 sys_nerrlist[] you get _exactly_ the same result).

So this is definitely the source of the kernel error header. It's either a file from the libc sources, or it is literally auto-generated like the above (I actually suspect the latter - now that I did the auto-generation it all felt very familiar, but that may just be my brain rationalizing things. Humans are good at rationalizing reality.).

Can anybody find the actual libc _sources_? Not the kernel headers that hjl mentions (those are the old ones from _before_ the change), but the file "libc-2.2.2.tar.Z"?

Anyway, we know where the kernel header comes from. Let's figure out where the libc data comes from.

BTW, a few more details on this file - the linux GCC 2.2.2 release was originally announced 28-Jun-1992. The 0.96bp2inc.tar.Z file originally lived on the then-primary linux ftp site banjo.concert.net in directory pub/Linux/GCC.

banjo stopped being an FTP server a couple months later - however, Jonathan Magid announced on 13-Aug-1992 that the entire banjo site was being reincarnated at host reggae.oit.unc.edu in directory ftp/pub/pc-stuff/Linux. Here's a copy of the announcement: http://www.kclug.org/old_archives/linux-activists/1992/aug/1/0708.shtml

My understanding is that reggae.oit morphed at some point into sunsite.unc.edu (which is now, of course, ibiblio.org) Jonathan still appears to be there, so I'm cc:ing him on this (apologies in advance if its an intrusion, Jonathan) on the off-chance that there might still be a 1992-era archive of the linux files once hosted by banjo.

The only other person likely to have access to a copy is H J Lu himself (also cc:'ed although I'm 99% sure he's still on lkml :-)

Note that we really don't care about that "0.96bp2inc.tar.Z" file: that's just the kernel headers, and 0.96b-pl2 did _not_ contain the comments yet. But libc used to use the kernel headers for other things (for things like system call numbers etc).

It's almost certainly the "libc-2.2.2.tar.Z" file that we want - that's the one that is going to contain the sys_errlist[] lists etc. Note how this libc-2.2.2 announcement predates the merging of the kernel header by almost a month - the kernel header information came from libc, not the other way around.

Does anybody have old CD-ROM's lying around?

In particular, the Yggdrasil Linux/GNU/X alpha CD-ROM was apparently released just a few months later. It would quite possibly contain the libc-2.2.2 sources... Adam Richter is still active, and I added him to the cc..

Who else was doing CD's back then? SLS? If nobody has the thing on a web-site any more, maybe they exist in physical format on somebodys bookshelf? The only reason that the really historic kernel archives still exist is that people saved them, and even so we're missing versions 0.02 and 0.03, but by the latter half of -92 there were already CD-ROMs being manufactured...

Of course, maybe the CD's are unreadable by now.

Yup, and I can confirm two things:

• the strings match 100% (well, duh, we already saw that from the binary)
• it doesn't even have an "errno.h"

that, together with the timing, pretty much proves that the kernel header was indeed just auto-generated from sys_errlist[] of that timeframe, with a program very much like the one I already posted.

Now, the libc file just says

        /* This is a list of all known signal numbers.  */

(which is obviously just a cut-and-paste from siglist.c n the same directory). But it shouldn't much matter, since I don't think SCO really is going to try to claim copyright ownership of the result of standard C library interactions like using "sys_errlist[]".

(I take that back - _of_course_ they are going to try to claim ownership. After all, they already claimed ownership of code I provably wrote).

I've done a resync and update of my batch scheduling that is also hyper-thread aware.

What is batch scheduling? Specifying a task as batch allows it to only use cpu time if there is idle time available, rather than having a proportion of the cpu time based on niceness.

Why do I need hyper-thread aware batch scheduling?

If you have a hyperthread (P4HT) processor and run it as two logical cpus you can have a very low priority task running that can consume 50% of your physical cpu's capacity no matter how high priority tasks you are running. For example if you use the distributed computing client setiathome you will be effectively be running at half your cpu's speed even if you run setiathome at nice 20. Batch scheduling for normal cpus allows only idle time to be used for batch tasks, and for HT cpus only allows idle time when both logical cpus are idle.

This is not being pushed for mainline kernel inclusion, but the issue of how to prevent low priority tasks slowing down HT cpus needs to be considered for the mainline HT scheduler if it ever gets included. This patch provides a temporising measure for those with HT processors, and a demonstrative way to handle them in mainline.

Patch available here:
http://ck.kolivas.org/patches/2.6/2.6.0/

I wonder how does Intel suggest we handle this problem? Batch scheduling aside, I wonder how to do any sort of priorities at all? I think POWER5 can do priorities in hardware, that is the only sane way I can think of doing it.

I think this patch is much too ugly to get into such an elegant scheduler. No fault to you Con because its an ugly problem.

How about this: if a task is "delta" priority points below a task running on another sibling, move it to that sibling (so priorities via timeslice start working). I call it active unbalancing! I might be able to make it fit if there is interest. Other suggestions?

Today utilization of execution resources of a logical processor is around 60% as you can find in public papers, and it's dependent on the processor implementation and the workload. It could be higher in the future, and their relative priority could be much higher then. So I don't think it's a good idea to hard code such a implementation-specific factor into the generic scheduler code.

Regarding H/W-based priority, I'm not sure it's very useful especially because so many events happen inside the processor and a set of the execution resources required changes very rapidly at runtime, i.e. the H/W knows what it should do to run faster at runtime, and imposing priority on those logical processor could make them run slower.

There are two goals here. Not having a batch process on one siling makes sense, and I'm going to try Con's patch after I try Nick's latest. Actually, if they play nicely I would use both, batch would be very useful for nightly report generation on servers.

But WRT the whole HT scheduling, it would seem that ideally you want to schedule the two (or N) processes which have the lowest aggregate cache thrash, if you had a way to determine that. I suspect that a process which had a small itterative inner loop with a code+data footprint of 2-3k would coexist well with almost anything else. Minimizing the FPU contention also would improve performance, no doubt. I don't know that there are the tools at the moment to get this information, but it seems as though until it's available any scheduling will be working in the dark to some extent.

Here are the sysfs class patches reworked against a clean 2.6.0 tree. I've created a class_simple.c file that contains a "simple" class device interface. I've then converted the tty core to use this interface (the combo of these two patches makes for no extra code added).

Then there are 3 patches, adding class support for misc, mem, and vc class devices. As the interface to add simple class support for devices is now so low, I feel that we do need to have mem class support as to not special case any char device.

With these patches, it's now much easier for others to implement class support for remaining char drivers/subsystems that do not have it yet.

I've released the 011 version of udev. It can be found at: kernel.org/pub/linux/utils/kernel/hotplug/udev-011.tar.gz

rpms built against Red Hat FC1 are available at: kernel.org/pub/linux/utils/kernel/hotplug/udev-011-1.i386.rpm with the source rpm at: kernel.org/pub/linux/utils/kernel/hotplug/udev-011-1.src.rpm

udev is a implementation of devfs in userspace using sysfs and /sbin/hotplug. It requires a 2.6 kernel to run. Please see the udev FAQ for any questions about it: kernel.org/pub/linux/utils/kernel/hotplug/udev-FAQ

The major changes since the 010 release are:

• The _long_ time delay on startup if you used the udev init script is now gone. We now fork a new process of udev for every device in the sysfs tree, causing any slow udev proceses (due to the type of device) to be not noticeable by the user.
• The RPM package is built against klibc. If you have any problems with this image, please let me know.
• The big "always wait 1 second" problem with the 010 release is now pretty much fixed. I still need some libsysfs changes to work around some of the delays I had to hard code in, but for any device that is not a partition, and has a "device" symlink, udev will not sleep at all. If a device does not have a "device" symlink, we will wait around for a few seconds to see if one is going to be created or not, and then move on. It's the only sane way of handling the issue of not knowing what kinds of devices have symlinks and which ones do not.
• fixed a few bugs when devices did not have a "device" symlink. Rules were getting applied when they should not have been.
• LABEL and CALLOUT rules now no longer require the BUS key. If the BUS key is not present, the LABEL or CALLOUT rule will be always be checked. This allows us to use these types of rules for devices without a "device" symlink.
• a few other minor tweaks and bug fixes too.

Thanks again to everyone who has send me patches for this release, a full list of everyone, and their changes is below.

udev development is done in a BitKeeper repository located at: bk://linuxusb.bkbits.net/udev

Daily snapshots of this tree used to be found at: http://www.codemonkey.org.uk/projects/bitkeeper/udev/. But that box seems to be down now. Hopefully it will be restored someday. If anyone ever wants a tarball of the current bk tree, just email me.

Yes. This is something I actually want to do anyway for 2.7.x. Dan Phillips had some patches for this six months ago.

You have to be careful, since you have to be able to mmap "partial pages", which is what makes it less than trivial, but there are tons of reasons to want to do this, and cache coloring is actually very much a secondary concern.

I didn't see the patch itself, but I spent some time talking to Daniel after your talk at the kernel summit. At least I _think_ it was him I was talking to - my memory for names and faces is basically zero.

Daniel claimed to have it working back then, and that it actually shrank the kernel source code. The basic approach is to just make PAGE_SIZE larger, and handle temporary needs for smaller subpages by just dynamically allocating "struct page" entries for them. The size reduction came from getting rid of the "struct buffer_head", because it ends up being just another "small page".

Your description is accurate. Another reason for code size shrinkage is getting rid of the loops across buffers in the block IO library, e.g., block_read_full_page.

Subpages only make sense for file-backed memory, which conveniently lets the page cache keep track of subpages. Each address_space has pages of all the same size, which may be smaller, larger or the same as PAGE_CACHE_SIZE. The first case, "subpages", is the interesting one.

An address_space with subpages has base pages of PAGE_CACHE_SIZE for its "even" entries and up to N-1 dynamically allocated struct pages for the "odd" entries where N is PAGE_CACHE_SIZE divided by the subpage size. Base pages are normal members of mem_map. Subpages are not referenced by mem_map, but only by the page cache. They are created by operations such as find_or_create_page, which first creates a base page if necessary. A counter field in the page flags of the base page keeps track of how many subpages share a base page's physical memory; when this field goes to zero the base page may be removed from the page cache.

Subpages always have a ->virtual field regardless of whether mem_map pages do. This is used for virt_to_phys and to locate the base page when a subpage is freed.

Page fault handling doesn't change much if at all, since the faulting address is rounded down to a physical page, which will be a base page.

Most of the changes for subpages are in the buffer.c page cache operations and are largely transparent to the VMM, though PAGE_CACHE_SHIFT becomes mapping->page_shift, which touches a lot of files. As you noted, buffer_head functionality can be taken over by struct page and buffers become expendible. However it is not necessary to cross that bridge immediately; page buffer lists continue to work though the buffer list is never longer than one.

With a little more work, subpages can be used to shrink mem_map: implement a larger PAGE_CACHE_SIZE then use subpages to handle ABI problems. In this case faults on subpages are possible and the fault path probably needs to know something about it. With a larger-than-physical PAGE_CACHE_SIZE we can finally have large buffers, though the kernel would have to be compiled for it. Some more work to allowing mapping->page_shift to be larger than PAGE_CACHE_SIZE would complete the process of generalizing the page size. My impression is, this isn't too messy, most of the impact is on faulting. Bill and others are already familiar with this I think. The work should dovetail.

I took a stab at implementing subpages some time ago in 2.4 and got it mostly working but not quite bootable. I did find out roughly how invasive the patch is, which is: not very, unless I've overlooked something major. I'll get busy on a 2.6 prototype, and of course I'll listen attentively for reasons why this plan won't work.

I haven't even _looked_ at any 2.7.x timeframe patches, and I'm not even going to for the next few months.

I don't care what does it, I want a bigger PAGE_CACHE_SIZE, and working patches are the only thing that matters. But for now, I have my 2.6.x blinders on.

Well, I don't even know what your approach is - mind giving an overview?

My original plan (and you can see some of it in the fact that PAGE_CACHE_SIZE is separate from PAGE_SIZE), was to just have the page cache be able to use bigger pages than the "normal" pages, and the normal pages would continue to be the hardware page size.

However, especially with mem_map[] becoming something of a problem, and all the problems we'd have if PAGE_SIZE and PAGE_CACHE_SIZE were different, I suspect I'd just be happier with increasing PAGE_SIZE altogether (and PAGE_CACHE_SIZE with it), and then just teaching the VM mapping about "fractional pages".

What's your approach?

I presented on this at KS. Basically, it's identical to Hugh Dickins' approach from 2000. The only difference is really that it had to be forward ported (or unfortunately in too many cases reimplemented) to mix with current code and features.

Basically, elevate PAGE_SIZE, introduce MMUPAGE_SIZE to be a nice macro representing the hardware pagesize, and the fault handling is done with some relatively localized complexity. Numerous s/PAGE_SIZE/MMUPAGE_SIZE/ bits are sprinkled around, along with a few more involved changes because a large number of distributed changes are required to handle oddities that occur when PAGE_SIZE changes from 4KB. The more involved changes are often for things such as the only reason it uses PAGE_SIZE is really that it just expects 4KB and says PAGE_SIZE, or that it wants some fixed (even across compiles) size and needs updating for more general PAGE_SIZE numbers, or sometimes that it expects PAGE_SIZE to be what a pte maps when this is now represented by MMUPAGE_SIZE. I have a bad feeling the diligence of the original code audit could be bearing against me (and though I'm trying to be equally diligent, I'm not hugh).

The fact merely elevating PAGE_SIZE breaks numerous things makes me rather suspicious of claims that minimalistic patches can do likewise.

The only new infrastructures introduced are the MMUPAGE_SIZE and a couple of related macros (defining numbers, not structures or code) and the fault handler implementations. The diff size is not small. The memory footprint is, and demonstrably so (c.f. March 27 2003).

My 2.6 code has been heavily leveraging the pfn abstraction in its favor to represent physical addresses measured in units of the hardware pagesize. Generally, my maintenance approach has been incrementally advancing the state of the thing while keeping it working on as broad a cross section of i386 systems as I can test or get testers on. It has been verified to run userspace on Thinkpad T21's and 16x/32GB and 32x/64GB NUMA-Q's at every point release it's been ported to, which since 2.5.68 or so has been every point release coming out of kernel.org.

Apps must, of course, be relinked for that, but that's userspace. This ABI change is largely out of the picture due to legacy binaries, user virtualspace fragmentation (most likely an issue for 32-bit threading), and so on. The choice of PAGE_SIZE in such schemes is also restricted to no larger than whatever choice used for userspace linking, which is a relatively ugly dependency. There's also a question of "smooth transition": the only way to "incrementally deploy" it on a mixture "ready" userspace and "unready" userspace is to turn it off. I suppose it has the minor advantage of being trivial to program.

I had in mind pure kernel internal issues, not ABI.

The issues from raising PAGE_SIZE alone are things like interpreting hardware descriptions in arch code, some shifts underflowing for things like hashtables, certain drivers doing ioremap() and the like either filling up vmallocspace or getting their math wrong, and some other drivers doing calculations on physical addresses getting them wrong, or using PAGE_SIZE to represent some 4KB or other fixed-size memory area interpreted by hardware, and filesystems that assume blocksize == PAGE_SIZE or assume PAGE_SIZE is less than some particular value (e.g. short offsets into pages, worst of all being signed shorts), and tripping BUG()'s in ll_rw_blk.c when 512*q->max_sectors < PAGE_SIZE.

These issues are the bulk of the work needing to be done for the driver and fs sweeps. Actual concerns about MMUPAGE_SIZE in drivers/ and fs/ are rather limited in scope, though drivers/char/drm/ was somewhat painful to get going (Zwane actually did most of this for me, as I have no DRM/DRI -capable graphics cards at my disposal).

I talked about this for a little while. Basically, there is only one concept in the entire patch, despite its large size. The vast bulk of the "distributed changes" are s/PAGE_SIZE/MMUPAGE_SIZE/.

At some point I was told to keep the whole shebang rolling out of tree or otherwise not answered by akpm and/or Linus, after I sent in what a split up (this is actually very easy to split up file-by-file) version of what just some of the totally trivial arch/i386/ changes would look like. The nontrivial changes are stupid in nature, but touch "fragile" or otherwise "scary to touch" code, and so sort of relegate them to 2.7. This is not entirely unjustified, as changes of a similar code impact wrt. the GDT appear to have affected some APM systems' suspend ability (I know for a fact my changes do not have impacts on APM suspend, but other, analogous support issues could arise after broader testing.)

Basically, the MMUPAGE_SIZE introductions didn't interest anyone a while ago, and I suspect people probably just want them all at once, since it's unlikely people want to repeat the pain analogous to PAGE_CACHE_SIZE (I should clarify later how this is different) where the incremental introduction never culminated in the introduction of functionality.

This is my patch. http://pkgcvs.turbolinux.co.jp/~go/patch-2.6/dpt_i2o.patch

Worked fine for me on quad xeon with 4G mem and 64bit PCI. It include.

• Support 2.6 kernel and DMA-mapping
• ioctl fix for raid tools
• use schedule_timeout in long long loop
• not support 64bit CPU yet.

However, It may differ from the Adaptec policy (linux-scsi ML).

well if you currently have 1tb in 8 non-redundant drives then you using 160GB disks... no?

the biggest p-ata disks right now are ~320GB so you can do a ~1TB software raid 5 stripe on a single 4 port ata controller such as a promise tx4000 using regular software raid rather than the promise raid. that would end up being fairly inexpensive and buy you more protection.

linux software raid has been as releiable as anything else we've used over the years, the lack of reliabilitiy in your situation will come entirely from failing disks, lose one and your filesystem is toast.

For the cost differential between linux native RAID and an external device of similar capabilities, outfit yourself with an additional CPU. I don't use RAID5 a lot but to a modern CPU, checksumming dozens of megabytes/second is child's play:

raid5: measuring checksumming speed
   8regs     :  1479.600 MB/sec
   32regs    :   744.400 MB/sec
   pIII_sse  :  1649.200 MB/sec
   pII_mmx   :  1806.000 MB/sec
   p5_mmx    :  1915.200 MB/sec
raid5: using function: pIII_sse (1649.200 MB/sec)

This is on a 800MHz Celeron, so a recent >2Ghz system will do lots better still.

Speaking of which, since most DIE^WIDE RAID controllers are really software driven, it may be possible to get the sw module to read the disks off of any controller

My machine at work has an onboard promise raid controller but I run linux sw raid0 on 2 of the disks (4 total, other 2 are not in a raid)

I'm not sure about real hardware raid. Like the Mylex or Adaptec (SCSI)

It really depends on what you mean by low cost? The ony ide raid controller that does 8 PATA drives well under linux is the 3ware controller. For SATA drives you have the 3ware, and adaptec controller.

In theroy the highpoint 8 port sata card would be a good canidate for software raid, but highpoint has yet to cough up an open source drive yet.

If you want to go the software raid route and have 2 spare pci solts. You can go with either the high point rocket raid 454 (PATA), or the promise SATA150 TX4.

I really don't recommend any of promise's cards that use use the i2o driver, or any sort of binary only driver.

PS- Why not at least run software raid 5? It takes far less cpu than you'd think, and can save your ass.

I think the reason is to avoid unnecessary code duplication. device-mapper provides a generic method to do such things. Also the developers are heading towards removing code from the kernel that can be done in userspace. There are plans to remove partition detection from the kernel in 2.7 and move the detection and setup code to a userspace program (using device-mapper) which can be placed in the initramfs (so that the user won't notice any difference). Ataraid detection and setup could also be placed there later.

If someone writes an ataraid detection and setup program in userspace it could be placed on an initrd.

You can find the dmsetup tool in Sistina's device-mapper package.

Accusys is a stolen technology from Dupli-Disk, there are law suits pending the last time I heard.

If it was based on the original firmware of the Dupli-Disk, it has totally wrong state machines for executing hdparm style callers.

I know the FSM's are wrong because I fixed them for Dupli-Disk. How they operate, I can not disclose. But Accusys can not handle correct settings for FSM to Taskfile.

Dupli-Disk requires taskfile access and will correctly operate.

Accusys is total CRAP, unless they licensed the technology proper.

ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.0/2.6.0-mm2/

• Added the adaptec SCSI driver update from Justin. There's some disagreement over whether the changes in this patch are appropriate, but we may as well get some external testing underway.
• Included an update from the ieee1394 team. Anyone who was having firewire problems, please retest.
• A couple of patches here should fix the CDROM-related problems which some people saw in 2.6.0-mm1.

I think Linus will be releasing the 2.6-pre kernels, and things will continue like that until 2.7 opens up.

I think Andrew is trying to get all of the after 2.6.0 fixes in -mm and tested before syncing up with Linus.

Indeed. The fact is, we do need a "testing ground" for some experimental fixes to stuff that needs to be fixed, and that's one of the things the -mm kernels used to do for 2.5.x.

Since everybody was pretty comfy with that setup, we'll just continue that way. By the time 2.7.x opens up, things should have calmed down, and commercial vendors have their support trees in shape etc, but for now the -mm tree is a testing ground, and I'll make -pre trees that should be fairly stable, and then we'll do the 2.6.1 etc "real releases" based off them.

When doing things like this, can you split up the patches into two separate things: one that _only_ does whitespace changes, and that is guaranteed not to change anything else, and another that does the rest.

It's a total b*tch to try to figure out which change resulted in some difference, if the changes are intermixed with large whitespace cleanups.

Yes and no.

Thirty separate patches make sense if they are independent and really do conceptually different things. Then it makes sense to have them as separate checkins, and be able to tell people "ok, try undoing that one, maybe that's the problem".

However, if they are all just "fix silly bugs in xxx", then I'd much rather see it as one big patch. Having it split up into "fix bug on line 50" and "fix bug on line 75" just doesn't make any sense - it only makes the patch history harder to follow.

So "many small patches" aren't automatically any better than one big one. The thing that matters is to keep things _conceptually_ separated. If one patch fixes whitespace, and another one fixes bugs, then that's good.

This tree has 38 deltas, all just merges.

The end result is a horribly messy revision tree, for a few one-liners.

I'm going to take the patch as a patch instead, and hope that you'll throw your BK tree away.

Please don't follow the release tree in your development trees, it makes it impossible to see how the revision history happened.

Agreed. Several BK developers do this, forgetting that one of things that makes BK so useful is its merge technology.

I recommend (assuming no patches outstanding),

• clone latest tree
• do development
• only 'bk pull' from latest tree iff (a) you are about to submit to Linus/Andrew or (b) you know there is a conflicting change in upstream

Pulling the latest, just to be up-to-date, just obfuscates things and needlessly increases the size of the master ChangeSet file.

Another possibility I like is to recreate my changes (what few so far ...) against a clean bk tree, before sending. Hide all my internal iterations and changes from others.

I will pull frequently and liberally into the bk clones that I use to track 2.6, 2.6-mm and whatever else I am based on. These in turn I pull into my main working bk tree, along with pulling in the various changes I have in progress, each from their own bk clone.

Then when it comes time to send out a patch, I:

1. Generate an old fashioned patch (bk export -tpatch), containing just the revisions relevant to what I will send.
2. Clone a fresh bk tree that is closest to whatever the recipient of my patch would like to work with
3. Apply the patch to the fresh clone, generating a clean history of one change for just that patch.
4. Double check that that builds and boots.
5. Then send that change out, usually by exporting it as a -second- old fashioned patch, since for reasons not relevant here, I end up sending patches, not bk pulls, down stream.

The objective being:

My final "published work" is that patch - it should be as clean as practical.

By going into and back out of old fashioned patches, I isolate the anal history that bk kept of all my interim changes from the rest of the world.

Yeah, I know. It's one of the downsides of having anal revision control, and BK is more anal than most.

I do end up taking patches that have this syndrome if it looks like the pain of not taking the messy revision history is larger than the pain of just fixing it. Sometimes it's hard to avoid.

But most of the time the proper thing to do is to just not merge unnecessarily - if something is pending for a while, Bk does the merge correctly anyway, so you can just leave it pending and have me pull from an old tree (after you have verified in your own tree that the pull will succeed and do the right thing).

That way it ends up being trivial to see where/when the changes happened.

Not being very used to BK, does that mean I have several trees around:

1. the official release tree
2. an "old tree" with my local change that I'm forwarding
3. a temporary test tree to see if the merge would succeed, which I'll get by cloning (1) and then pulling from (2)?

Well, talk about FAAAAAAST drives (10,025/min SCSI kind) unless you have time to waste on all those BK consistency checks (which are, of course, what #3 is all about).

Or am I missing some obvious short cut?

The answer to that question is always a resounding "yes".

BK really thrives on having several trees around. You don't _have_ to have them, but basically the default rule should be: use separate trees for everything you can think of that isn't directly dependent on stuff in another tree.

That does _not_ mean that you should necessarily create "temporary trees". I actually do that a lot of the time, because I tend to create a totally new clone when I start applying long series of patches or do anything even half-way strange: it's just a lot easier to throw failures away, than it is to try to sort it out later.

But most people probably do _not_ want to have that kind of "temporary tree" mentality in general. People should realize that it's ok, and in particular that if you're doing something experimental it's fine to just create a new tree and later on decide that it was crap and just do a full "rm -rf" on it, and realize that the only thing you lost was some time.

Basically, the obvious shortcut is to keep your work-tree around, so you don't have to clone and re-pull it all the time.

After a while, your work-tree is really messy (especially if you pull from multiple different development trees), but the point should be that no actual development gets done there, so you don't care: you can always just flush it entirely, and re-create it anew.

But you don't have to flush and re-create it _all_ the time. That would just be wasteful. Although if you have the hardware, it isn't that painful..

Updated my snapshot script for 2.6.0 release, and snapshots are once again flowing into

http://www.kernel.org/pub/linux/kernel/v2.6/snapshots/

The snapshots for the 2.6.0-testX series were moved to

http://www.kernel.org/pub/linux/kernel/v2.6/snapshots/old/

Here are some USB patches for 2.6.0. There are a number of different fixes and a few new drivers added. Some of the highlights are:

• lots of usb-storage fixes
• lots of usb-storage unusual-devs updates
• added new w9968cf driver
• added new lego tower driver
• core tweaks for non-compliant USB devices.
• lots of other little fixes

Please pull from: bk://linuxusb.bkbits.net/usb-devel-2.6

Patches will be posted to linux-usb-devel as a follow-up thread for those who want to see them.

Summary of new changes:

• bonding fixes
• e100/e1000 fixes
• other fixes
• via-rhine netpoll support

Summary of patchkit:

• new e100 driver (rewritten from scratch)
• new nVidia nForce NIC driver
• new pc200syn WAN driver
• tg3 bug fixes
• r8169 major bug fixes
• e1000 minor updates / fixes
• sk98lin vendor updates / fixes
• misc bug fixes
• 8139too NAPI support
• tulip NAPI support
• netconsole / netdump support
• net_device allocation and reference counting work

Patch:

http://www.kernel.org/pub/linux/kernel/people/jgarzik/patchkits/2.6/2.6.0-bk2-netdrvr-exp1.patch.bz2 (NOTE: _requires_ 2.6.0-bk2 snapshot, or IOW Linus-latest from BK, in order to apply successfully)

Full changelog:

http://www.kernel.org/pub/linux/kernel/people/jgarzik/patchkits/2.6/2.6.0-bk2-netdrvr-exp1.log

BK repo:

bk://gkernel.bkbits.net/net-drivers-2.5-exp

Some of you use the cset -x feature in BK (which is a way of undoing the effects of a particular changeset).

The documented behaviour of cset -x is that it only undoes content changes, not renames, creates, permissions, etc. This interface is unique in that respect, all the other interfaces in BK operate on all attributes, not just contents.

I've prototyped a version of the interface which works on contents, names, and permissions, and in addition also uses the BK merge tools to merge any conflicting changes as a result of the undo of the changeset. This is JUST A PROTOTYPE and has a lot of limitations but I'd like some feedback on whether this is a good direction and we should productize this or if this is not helpful to you.

If you use cset -x and want a better version of that interface, drop me an email and I'll send you the shell script (please make sure to send mail to me directly, not to the lists, because (a) they don't need to see all the noise and (b) I'm no longer subscribed to the Linux kernel mailing list).

Ok, I've merged a lot of pending patches into 2.6.1-rc1, and will now calm down for a while again, to make sure that the final 2.6.1 is ok.

Most of the updates is for stuff that has been in -mm for a long while and is stable, along with driver updates (SCSI, network, i2c and USB).

I've released the 012 version of udev. It can be found at:

kernel.org/pub/linux/utils/kernel/hotplug/udev-012.tar.gz

rpms built against Red Hat FC1 are available at:

kernel.org/pub/linux/utils/kernel/hotplug/udev-012-1.i386.rpm

with the source rpm at:

kernel.org/pub/linux/utils/kernel/hotplug/udev-012-1.src.rpm

udev allows users to have a dynamic /dev and provides the ability to have persistent device names. It uses sysfs and /sbin/hotplug and runs entirely in userspace. It requires a 2.6 kernel with CONFIG_HOTPLUG enabled to run. Please see the udev FAQ for any questions about it:

kernel.org/pub/linux/utils/kernel/hotplug/udev-FAQ

For any udev vs devfs questions anyone might have, please see:

kernel.org/pub/linux/utils/kernel/hotplug/udev_vs_devfs

The major changes since the 011 release are:

• updated ide devfs compatible script
• command line options are now available, allowing you to query the udev database. This is just a first cut, the query functionality will be extended later.
• lots of minor bug fixes and tweaks.

Thanks again to everyone who has send me patches for this release, a full list of everyone, and their changes is below.

udev development is done in a BitKeeper repository located at:

bk://linuxusb.bkbits.net/udev

Daily snapshots of this tree used to be found at:

http://www.codemonkey.org.uk/projects/bitkeeper/udev/

But that box seems to be down now. Hopefully it will be restored someday. If anyone ever wants a tarball of the current bk tree, just email me.

udev names are created when kernel detects corr. device. Unfortunately for removable media kernel rescans for partitions only when I try to access device. Meaning - because kernel does not know partition table it did not send hotplug event so udev did not create device nodes. But without device nodes I have no way to access device in Unix :(

specifically I have now my Jaz and I have no (reasonable) way to access partition 4 assuming device nodes are managed by udev.

devfs solved this problem by

• always exporting at least handle to the whole disk (sda as example)
• using something simple like dd if=/dev/sda count=1 on lookup for non-existing partition (/dev/sda4) that would rescan partitions and create device nodes for them.

static /dev simply has all nodes available and does not suffer from this problem at all.

unfortunately there are no lookup events in case if udev ... meaning at this moment user must manually rescan partitions after inserting new media. I do not see any way to solve this problem at all given current implementation. The closest is to blindly create nodes for all partitions as soon as block device is available.

Back in the days of kernel 2.2 there was a patch (from Rik van Riel if I recall) to the kernel so no single user could use all available cpu.

I was wondering if this feature (or something like it) has been ported or integrated in 2.4.x series ?