| ★ wanayoo — archive 1999 http://kernelnotes.org/kernel/v2.1/0-9.html | Nouvelle recherche | Portail wanayoo |
Linux 2.1 Kernel Patches |
[Home] [Linux 2.0] [Linux 2.1] [Information] [Software] [Distributions] [Links] |
|
About the Kernel Patch Summaries
These summaries are written after each kernel release by examining the chunks of each patch file and figuring out what has changed. Occasionally, they are supplemented by notes that Linus (and others) posted to the linux-kernel mailing list.
The summaries from 2.1.0 to 2.1.51 were put together by Mark Evans <mevans@ecsnet.com>.
The summaries from 2.1.54 onwards were put together by Myrdraal@deathsdoor.com -- originally appearing on the Cutting Edge Linux website.
Yaroslav Rosomakho is now also writing some kernel summaries, including pre-patches!
|
5 Latest 0-9 10-19 20-29 30-39 40-49 50-59 60-69 70-79 80-89 90-99 100-109 110-119 120-129 130-132 2.2.0-pre |
Newer >> 10-19 |
Linux 2.1.9 patch
-
patch-2.1.9.gz (Nov 12, 1996)
Updated Display adapter and video mode setup code. Sparc Architecture Updates Additional IDE CD-ROM Updates - Serveral Changes Keyboard update - SAK should now work in the raw mode. G8BPQ Compatible "AX.25 via eithernet" Driver Included TULIP Ethernet driver update = Include AUI in dc21041 media printout. MKISS Driver Added - This module implements the AX.25 protocol for kernel-based devices like TTYs. PI2 Interface can now be used as a module. NCR53c8xx Driver update to v1.14c AF Filesystem Updates IPv6 Updates NET/ROM Update - Release 004 - Supports use as a module. Addition of ROSE Driver Release 001 Linux 2.1.8 patch - patch-2.1.8.gz (Nov 9, 1996)
The 2.1.8 release is more "interesting", and the big thing here is the new IPv6 code together with quite a number of cleanups for IPv4 (TCP retransmissions are really done by TCP, not by the IP layer like they used to be done). Thanks to Pedro for all the work on this.The new networking code can certainly have some problems, but hey, that's the whole point with a development kernel. At least you shouldn't be bored by testing it out (but you might: it's not exactly _unstable_ either, and it might just turn out to work perfectly for everybody, who knows?).
2.1.8 also contains various other patches:
- ide code cleanups
- three instructions less in system calls ;)
- preliminary module exception handling stuff (but your insmod needs to know about this all)
- exception 15 handling for those buggy PPro's (this is in 2.0.x too)
- AMD k5 divide overflow fix (this is in the 2.0 tree too)
- dead-Cedilla keyboard support ("the only good cedilla is a dead cedilla"). This needs a new version of kbdtools to be useful. Also in 2.0.x
- SMBFS inode time fix - now sets file modification time correctly on SMB file systems.
- NCR53C8xx SCSI Driver Update
Tell me if you have problems,
Linus
Linux 2.1.7 patch - patch-2.1.7.gz (Nov 1, 1996)
Ok, 2.1.7 is out there in ftp-land, and has the new clever exception handling implementation done by Richard Henderson for both x86 and alpha (I think the idea was originally Ingo Molnars, but I could be mistaken, and Richard certainly took it to new heights of elegance). Anyway, the new stuff uses some clever linker tricks to essentially do zero-cost exception handling rather than having to waste precious instructions to update the per-task exception structure. Very nifty indeed.The interface is still exactly the same, only some low-level architecture- specific implementation details changed. It's all so clever that it's scary. People interested in low-level scary stuff should take a look at the uaccess.h files for x86 or alpha, and be ready to spend some time just figuring out what it all does ;)
Anyway, the new exceptions do have a down-side: the exception tables do not currently get exported from kernel loadable modules. This is not something fundamentally hard: it's just something that needs to be handled by the insmod &co people, and which hasn't been done yet. Anybody who feels he can be useful in making insmod talk to the kernel about the exception tables (Bjvrn? Jacques?), please contact Richard
about this. Note that the lack of exception data in modules does _not_ mean that modules don't work. It only means that modules can't handle user space address exceptions, and thus can't recover gracefully from EFAULT errors. As such you can still load and use the modules (assuming you have the other 2.1.x fixes for insmod module loading), but I'd suggest against it on any system with untrusted users that you expect to do nasty things just for fun.
Apart from the new clever exceptions, the other changes in 2.1.7 are:
- baycom driver is a real network driver instead of a strange character driver.
- soundmodem driver. You don't want to know what perversions people are up to ;) (If you _do_ want to know, read the docs)
- various drivers which had boo-boo's with the new exception code fixed up (ie ftape/md/cdrom/eql should work again)
- isdn silly problems should be fixed
- scc driver update
- tulip driver multicast lists
- aha1740 should work again (virtual/physical address probs on x86, funnily it worked on alpha)
- quota unmount bug should be fixed in 2.1.x too
Anyway, I finally think the exception stuff is starting to be stable (apart from the issues on module loading), and I think I'll try to start integrating some of the other patches in the next few weeks. So expect to start seeing ipv6/PPC/Sparc/m68k/ARM etc patches any week now (and my apologies to the developers of those things for being slow to react - you're all in my queue of things to handle, I've just been slow to get everything else cleared up).
Linus
Linux 2.1.6 patch - patch-2.1.6.gz (Oct 29, 1996)
Ok, I'm back, and I made a 2.1.6 release. I'll try to make a 2.0.24 release later today, but quite frankly I had 1500+ emails to go through, and I want to calm down a bit before making that release..2.1.6 does:
- changed some implementation details of the user access macros on x86: it seems my old inline assembly wasn't really 100% stable. That might have led to unexpected problems (incorrect copying of memory etc). The new code also does a first version of constant-size optimizations.
- new sound driver. I haven't actually tested whether it even compiles, but hey, it can't be worse than 2.1.5 ;)
- updated ISDN code. Some of the code has been updated by yours truly (the readb/membase stuff for the TELES driver) without having access to the hardware or even compiling it, but as you all know I never make mistaeks, so it will obviously work on the first try. It _might_ even work on alphas.
- math emulation should work again on 386's..
- SCC character driver rewritten to be a real network driver instead.
- getting rid of "segment.h" eventually..
- cdu31a should work with sound again
- aha2940 driver temporary disable to make it boot for some people (please tell me if it still doesn't work for you)
- eata driver update
- and the big-ping patch, of course..
For obvious reasons (see about email above), I've been kind of busy and haven't had time to check all the changes I made on an alpha yet. Testers appreciated.
Also, if you have sent me patches and don't see them in 2.1.6, you can more-or-less assume that I don't have them pending, but am hoping to have them re-sent in the near future (but don't all flood me immediately, I'd rather avoid having to go through 50 patches in one day).
Linux 2.1.5 patch - patch-2.1.5.gz (Oct 18, 1996)
This is a compilation of patch submissions from 2.1.4. It should fix most problems with PPP and other modules.
- Virtual Memory Fixes
- Enable 4MB page tables for Intel Pentium CPU
- Floppy Driver Updates
- Block Driver Fixes
- Sony CDU31A CD-ROM Fixes - fix bug when returning track numbers
- FTAPE Support Updates / Fixes
- QIC02 updates / fixes
- Virutal Screen Code Fix
- PPP/SLIP Driver Updates
- ELF Fix for bad binaries
- EXT2 Update
- Serveral update for ALPHA code tree
Linux 2.1.4 Patch - patch-2.1.4.gz (Oct 15, 1996)
Ok, I just made Linux-2.1.4 available. The patch is _huge_, as this kernel renames some of the rather basic user access primitives due to the new semantics of those primitives.This patch does:
- mandatory locks as a mount option
- 3c505 driver update
- aha2940 driver update (yes, it should hopefully work)
- BSD ncr driver update
- in2000 driver update
- FPU emulation should work again
- user-space memory access update
It's the last one which is the large one. The changes are essentially:
x = get_user(addr) => err = get_user(x,addr); put_user(x,addr) => err = put_user(x,addr); memcpy_fromfs(...) => bytes_left = copy_from_user(...); memcpy_tofs(...) => bytes_left = copy_to_user(...);Note especially "get_user()" - it actually has two return codes (one of which is the first argument that gets changed: no pointer dereferencing, but done with some rather ugly macros). Note how all of these now return a "status" indication: the get/put functions return an error (0 for success, -EFAULT for bad access), and the copy_xxx_user() functions return how many bytes they _failed_ to copy (so again zero is a successful return).
Now, the _really_ interesting thing is that none of the above need the old "verify_area()" call any more, as they do all the area verification on their own. Less chance of missing address verification that way.. Also, the new address verification is a lot less expensive than the old scheme was.
NOTE NOTE NOTE! I have done all the conversions that _I_ personally need to get it to compile and work on both alpha and x86, and it certainly helps performance on some benchmarks. HOWEVER, I haven't compiled everything, and I might have missed a few places, so if 2.1.4 doesn't compile for you, please just report it and if possible you can try to fix it with the above mapping in mind.
ALSO, this is not the final optimized version yet:
- there are some functions that essentially think that user accesses are "cheap". They do loops like this:
while (count) { count--; put_user(0,buff++); }The above is pretty bad for performance (it wasn't very good before either, but now it's even worse), because now we have the small verification overhead for _each_ byte. The overhead is not very high, but it's there. So the process creation benchmark performance is down until the above gets fixed (one of the places doing things like that is "execve()" and the ELF loader). Todo: use "set_user(buf, 0, len)" instead.
- I changed _some_ of the code to take advantage of the new setup, but most of it has been only minimally changed to make it work. As such, a lot of "verify_area()" calls still exist, and the code isn't very optimal.
- Some of the user access functions aren't optimized yet (the user memory copy functions do not optimize constant length area accesses etc).
Now, the above makes it sound like 2.1.4 is a slow kernel. It isn't. It's just that there are still some things to optimize before we see the _real_ performance increase from this all. Any comments and patches are welcome (I've essentially spent the last 5 days getting it all to work the way I wanted, I haven't really bothered about optimizations yet).
Linus
Linux 2.1.3 Patch - patch-2.1.3.gz (Oct 10, 1996)
You know what this means by now.. New kernels faster than you can pronounce "Linux". I decided that I'd be better off releasing a 2.1.3 with the TCP fixes and the other stuff that has come in (FPU emulator should compile again etc), so that people can play with it without having to know how to hack it together..Fixes:
- various structure initializations moved to real variables. This is simply because it turns out that gcc puts the structures into the code section otherwise, which doesn't necessarily work too well on all platforms.
- The TCP fixes you've all already seen at least twice.
- some cleanups wrt "wp_works_ok" which is really rather x86-specific problem in the first place.
- qic tape and x87 emulation support should compile
- ni52 and ni65 network driver updates for the new memory model
- tulip driver multicast bugfix
- fs/buffer.c cleaned up (and potential bugfix that I suspect will have to percolate back to 2.0.x)
- getname() uses exceptions instead of checking the memory map by hand
- IP forward memory leak fix
- SYN sequence fix against KA9Q-based TCP stacks.
- URG handling fix
This kernel is obviously bug-free, so it's entirely safe to test. No need to worry at all. No sirree. Show your children. Impress your neighbours. Take your cat off your keyboard. Because this is what you've been waiting for!
Linus
Linux 2.1.2 Patch - patch-2.1.2.gz (Oct 8, 1996)
I made both a new 2.0.x and a 2.1.x release, and they are available on the usual sites. The 2.0.x is mainly a few driver updates, but also has the pty bug-fix that people in the 2.1.x tree have already been testing out.2.1.2 is the much more interestign one, at least as far as I'm concerned. This release does:
- goodbye old inefficient "verify_area()", hello run-time exceptions.
Essentially, the old "verify_area()" was a very straightforward system that checked that any pointer given to the kernel was valid. It verified that the area was mapped, and returned either 0 or -EFAULT depending on that. It was reliable, boringly so.
However, "verify_area()" did have a few problems. Notably: it was slow. Not "horribly and inexplicably slow", but it was simply called a lot, and it had to do a lot of checks. End result: it shows up on kernel profiles pretty well.
Now, I tried to optimize it, but it was basically pretty optimized already. It used avl trees to find the memory areas, and it was just plain a good implementation. But the bad thing about it all was that it never found any errors. The things it checks against are things that never happen for normal programs, and as such it is very irritating to see it taking up almost 20% of the kernel time under some circumstances.
Now, "verify_area()" had another more subtle problem, and that was the fact that the memory mapping that it checked could _change_ after it had been checked when people used kernel threads (another thread did a mmap or a munmap). That wasn't originally a problem, as Linux didn't have threads originally, but it's been nagging me for the last year or two.
So... Enter the all-new-and-improved Linux-2.1.2. It gets rid of "verify_area()" altogether. Or rather, the function still exists, but it doesn't really do much any more. In fact, it does just about the same thing that it did in Linux-0.01: it only tests that the pointer is in user space (and it does _not_ check whether the area it points to is actually mapped).
Voila, instant speedup: the function that used to take 20% of kernel time under some circumstances is now a inline function that expands to just a few assembly instructions.
The downside? Bad programs who mess with the kernel. They pass invalid addresses to system calls, and nothing checks it any more. Bummer. But don't despair: we do actually have something that checks the invalid addresses. And what's nice about it is that it's actually built-in in the CPU hardware. The paging unit.
Anyway, Linux-2.1.2 now uses exceptions to trap bad system call buffers. The new scheme is a bit more complex to actually use, but it's a _lot_ faster, and it actually works correctly in the presense of kernel threads (the old one worked fine too usually, but it _could_ make mistakes, and making mistakes isn't really acceptable when the whole and only idea of the function is to verify that no mistakes happen in the first place).
However, I didn't have the stamina to add the extra exception handling to all of the system calls, so Linux-2.1.2 is a work in progress in that sense. It's not hard to add the necessary "exception()/end_exception()" pairs, but it _is_ tedious, and it does require some care (to make sure everything is "rolled back" correctly when an exception happens). So I'll be doing that over a period of kernels, but I wanted to make the basic stuff available now.
In addition to the exception stuff, 2.1.2 also does:
- updates some alpha handling. This should contain all the important things from the axp-diffs that David M-T has been making.
- alpha and x86 system call entry optimizations. I now get less than 1us system call latencies on my ev5/333 according to lmbench.
- various minor changes ("a" -> "an" fixes etc)
David: don't expect all of your patches to be in. I didn't actually use much more than you machine check patches as-is: everything else is essentially re-done the way I perferred it.
Linus
Linux 2.1.1 Patch - patch-2.1.1.gz (Oct 3, 1996)
Ok, 2.1.1 is available at the usual places. This kernel fixes:
- vm86 mode hopefully works, so dosemu people please try it out again..
- initrd should compile. Whether it works needs to be tested by somebody else.
- eata SCSI driver updates for 2.1 mm handling
- fdomain SCSI driver update for 2.1 mm handling
- SCSI tape bugfix
- UltraStor 14F/34F SCSI driver update for 2.1 mm handling
- sound should work
- file locking update
- NCP filesystem should work again.
- UFS update
- umsdos updates
- asm-i386/unistd.h error detection update
- arp NULL device fixes
NOTE! The problem with module loading is _not_ a bug in the kernel, it's a problem with the error detection in user space. Don't send me kernel patches for this: you need to recompile insmod with patches that have been floating around here on the kernel mailing list.
The arp NULL pointer fixes and the SCSI tape bugfix is also in my personal 2.0.x tree for future release (probably 2.0.22 early next week).
Note that the SCSI driver changes should _also_ mean that the drivers now work on Linux/alpha. If there are people who have any of the SCSI cards listed above, I'd be very interested in hearing about it (and I'm sure the people keeping up the alpha FAQ would also want to know about new supported hardware). Maybe someone could try out a card for quick testing?
Linus
Linux 2.1.0 Patch from 2.0.21 - patch-2.0.21-2.1.0.gz (Sep 30, 1996)
Ok, I released the final 2.1.0 (both as full source and as patches against 2.0.21), and as always I'm interested in any comments on it.NOTE! The driver situation has improved, but is by no means perfect. Most notably, there is no working aic7xx0 driver (aha2940 based cards) available right now. Quite frankly, I can't fix this on my own, so I would really be very grateful if somebody who has the hardware and the inclination would look into it.
(I looked into it just enough to see that it's ugly: the scatter-gather lists used by the SCSI code use virtual addresses, and the aic7xx0 controller code uses those scatter-gather lists directly. The high-level SCSI code would either have to use bus addresses instead of virtual addresses, or the aic7xx0 driver should do the translation either on the CPU or in the aic7xx0 assembly).
Most other drivers would seem to work, although some of them (like the xd driver) haven't been tested.
Other than the virtual kernel address setup changes on x86, the 2.1.0 release also changes the VFS interface (lseek/read/write get longer types). There are also quite a lot of m68k-specific changes (but I'm waiting with the generic m68k changes until the driver situation has calmed down a bit). There is also a tentative patch for the pty problems people have reported (that will be fed back to the 2.0.x tree too, but you can test it out in the 2.1.x series first)
Linus
[Home] [Up] [Search] [FeedBack]
Copyleft © The LinuxHQ Project. Licensed under the GNU Public License.