Skip to contentSkip to conversation

RFC: i386: kill !4KSTACKS

DISCUSSION Alex Davis 16 participants 19 messages Sep 2, 2005 – Sep 9, 2005 5 messages not in the archive

lore
Share & tools
lore

Alex Davis

author
#0 lore raw

ndiswrapper and driverloader will not work reliably with 4k stacks. This is because of the Windoze drivers they use, to which, obviously, they do not have the source. Since quite a few laptops have built-in wireless cards by companies who will not release an open-source driver, or won't release specs, ndiswrapper and driverloader are the only way to get these cards to work.
Please don't tell me to "get a linux-supported wireless card". I don't
want the clutter of an external wireless adapter sticking out of my laptop, nor do I want to spend money on a card when I have a free and working solution.

Thank you.

-Alex

I code, therefore I am

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com

#1 lore raw
Quoted 9 lines · Alex Davis

ndiswrapper and driverloader will not work reliably with 4k stacks. This is because of the Windoze drivers they use, to which, obviously, they do not have the source. Since quite a few laptops have built-in wireless cards by companies who will not release an open-source driver, or won't release specs, ndiswrapper and driverloader are the only way to get these cards to work.
Please don't tell me to "get a linux-supported wireless card". I don't
want the clutter of an external wireless adapter sticking out of my laptop, nor do I want to spend money on a card when I have a free and working solution.

Please don't tell me to "care for closed-source drivers". I don't want the pain of debugging crashes on the machines which run unknown code in kernel space.

IOW, if you run closed source modules - it's _your_ problem, not ours. --
vda

#2 lore raw
Quoted 11 lines · Denis Vlasenko
Quoted 9 lines · Alex Davis

ndiswrapper and driverloader will not work reliably with 4k stacks. This is because of the Windoze drivers they use, to which, obviously, they do not have the source. Since quite a few laptops have built-in wireless cards by companies who will not release an open-source driver, or won't release specs, ndiswrapper and driverloader are the only way to get these cards to work.
Please don't tell me to "get a linux-supported wireless card". I don't
want the clutter of an external wireless adapter sticking out of my laptop, nor do I want to spend money on a card when I have a free and working solution.

Quoted 6 lines

Please don't tell me to "care for closed-source drivers". I don't want the pain of debugging crashes on the machines which run unknown code in kernel space.

IOW, if you run closed source modules - it's _your_ problem, not ours. --

There is no logic to the above statement. We know when a kernel is tainted and do not try hard to debug problems when tainted . We also know that ndiswrapper and friends _currently_ let people use hardware that otherwise would have to run MS stuff. We know that 4K stacks hurt the above. Do we really want to break working configs just to enforce 4K stacks? How does it hurt to make 4K the default and allow 8K? What _might_ make sense is to make 8K a reason to taint the kernel.

Ed Tomlinson

#3 lore raw
Quoted 3 lines · Ed Tomlinson

MS stuff. We know that 4K stacks hurt the above. Do we really want to break working configs just to enforce 4K stacks? How does it hurt to make 4K the default and allow 8K? What _might_ make sense is to make 8K a reason to taint the kernel.

The question is whether ndiswrapper can do stack switching itself. Since as I understand it the NT stack is way more than 8K. Is there anything else needed so it (and perhaps in future other 'hard cases') can handle stacks themselves. We have seperate IRQ stack handling already which should also help this.

So what is needed to make it go away - specific technical items or just the persuasive effect of having to fix it ?

#4 lore raw
In reply to Denis Vlasenko
Quoted 12 lines · Denis Vlasenko
Quoted 10 lines · Alex Davis

ndiswrapper and driverloader will not work reliably with 4k stacks. This is because of the Windoze drivers they use, to which, obviously, they do not have the source. Since quite a few laptops have built-in wireless cards by companies who will not release an open-source driver, or won't release specs, ndiswrapper and driverloader are the only way to get these cards to work.
Please don't tell me to "get a linux-supported wireless card". I don't
want the clutter of an external wireless adapter sticking out of my laptop, nor do I want to spend money on a card when I have a free and working solution.

Quoted 8 lines

Please don't tell me to "care for closed-source drivers". I don't want the pain of debugging crashes on the machines which run unknown code in kernel space.

IOW, if you run closed source modules - it's _your_ problem, not ours. --
vda
-

No one is asking you to 'care' about our problems running a notebook with a closed source driver under ndiswrapper. We aren't asking you to debug problems with them either. All we're asking is for you to not go out of your way to break existing working machines, and make it difficult to run Linux on them. You are talking about knowingly removing an option that allows many machines to currently run without problems, some of them for reasons other than closed source code.

If you want 4k stacks to be the default, I have no problem with that. If you want to rip out the provision for 8k stacks to be selectable at build time, that is a different issue entirely.

Paul

#5 lore raw
Quoted 2 lines · Paul Misner

No one is asking you to 'care' about our problems running a notebook with a closed source driver under ndiswrapper.

Yes you are. You're asking for 4KSTACKS config option to maintained and it is not something you get for free. Besides, if it is indeed ripped out of mainline kernel, you can always keep it a separate patch for ndiswrapper.

Pekka

#6 lore raw
Quoted 4 lines

Yes you are. You're asking for 4KSTACKS config option to maintained and it is not something you get for free. Besides, if it is indeed ripped out of mainline kernel, you can always keep it a separate patch for ndiswrapper.

Though 4K stacks are used a lot, they probably aren't used on all configurations yet. Other situations may arise where 8K stacks may be preferred. It is too early to kill 8K stacks imho.

Signature · 3 lines
Bas Westerbaan
http://blog.w-nz.com/
GPG Public Keys: http://w-nz.com/keys/bas.westerbaan.asc
#7 lore raw
Quoted 8 lines · Bas Westerbaan
Quoted 4 lines

Yes you are. You're asking for 4KSTACKS config option to maintained and it is not something you get for free. Besides, if it is indeed ripped out of mainline kernel, you can always keep it a separate patch for ndiswrapper.

Though 4K stacks are used a lot, they probably aren't used on all configurations yet. Other situations may arise where 8K stacks may be preferred. It is too early to kill 8K stacks imho.

Please name situations where 8K stacks may be preferred that do not involve binary-only modules.

Quoted 1 line

Bas Westerbaan

cu
Adrian

Signature · 4 lines
       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed
#8 lore raw
Quoted 6 lines
Quoted 3 lines

Though 4K stacks are used a lot, they probably aren't used on all configurations yet. Other situations may arise where 8K stacks may be preferred. It is too early to kill 8K stacks imho.

Please name situations where 8K stacks may be preferred that do not involve binary-only modules.

I meant that there could be situations, which have not yet been found, where it could be preferred to use 8K stacks instead of 4K. When you switch from having 8K stacks as default to 4K stacks without possibility for 8K stacks you'd possibly encounter these yet to be found situations.

When on the other hand the 4K stacks are set as default, leaving the option in, instead of removing it, these possible situations, when found, could be resolved (temporarilly) by switching back to 8K stacks.

After a while having 4K stacks as default would be a better time to decide whether to remove the option or not instead of now.

Signature · 3 lines
Bas Westerbaan
http://blog.w-nz.com/
GPG Public Keys: http://w-nz.com/keys/bas.westerbaan.asc
#9 lore raw
Quoted 8 lines · Bas Westerbaan
Quoted 6 lines
Quoted 3 lines

Though 4K stacks are used a lot, they probably aren't used on all configurations yet. Other situations may arise where 8K stacks may be preferred. It is too early to kill 8K stacks imho.

Please name situations where 8K stacks may be preferred that do not involve binary-only modules.

I meant that there could be situations, which have not yet been found,

And the boogeyman might really exist too.
This is just hypotetical hand-waving.

Quoted 4 lines

where it could be preferred to use 8K stacks instead of 4K. When you switch from having 8K stacks as default to 4K stacks without possibility for 8K stacks you'd possibly encounter these yet to be found situations.

Fedora kernels have been built with 4K stacks for a long time. (Since even before the option went upstream). The only things that have been reported to have problems with 4KB stacks are..

- NDISwrapper / driverloader.
(Shock, horror - no-one cares).
- XFS when used in conjunction with RAID
Fixed now ? (Though Neil Brown does have a pending patch for md to make that use less stack, which will also help).
- Reiser4
Fixed 'soon'.

Quoted 7 lines

When on the other hand the 4K stacks are set as default, leaving the option in, instead of removing it, these possible situations, when found, could be resolved (temporarilly) by switching back to 8K stacks.

After a while having 4K stacks as default would be a better time to decide whether to remove the option or not instead of now.

This is what was proposed not long after the option got merged. "After a while" has passed by quite a stretch.

Dave

#10 lore raw
Quoted 18 lines · Dave Jones
Quoted 8 lines · Bas Westerbaan
Quoted 6 lines
Quoted 3 lines

Though 4K stacks are used a lot, they probably aren't used on all configurations yet. Other situations may arise where 8K stacks may be preferred. It is too early to kill 8K stacks imho.

Please name situations where 8K stacks may be preferred that do not involve binary-only modules.

I meant that there could be situations, which have not yet been found,

And the boogeyman might really exist too.
This is just hypotetical hand-waving.

Quoted 4 lines

where it could be preferred to use 8K stacks instead of 4K. When you switch from having 8K stacks as default to 4K stacks without possibility for 8K stacks you'd possibly encounter these yet to be found situations.

Quoted 6 lines

Fedora kernels have been built with 4K stacks for a long time. (Since even before the option went upstream). The only things that have been reported to have problems with 4KB stacks are..

- NDISwrapper / driverloader.
(Shock, horror - no-one cares).

Actually, people who want to run Linux on laptops instead of MS care a whole bunch! And not everyone has a committment from their employer to provide Linux compatible hardware, or the personal funds to spend extra to buy their own instead of getting a bargain laptop which may not be fully supported. 8KSTACKS is in and working, you are proposing to break Linux on a number of machines just to satisfy some personal distaste for the code, not because there is neat new code which fails to work with 8K stacks.

You must have something more useful to work on, which would ADD value to the kernel instead of breaking existing installations. Ripping out petty stuff which works is a waste of your time and talent, please find something better to do. Perhaps devise a way for programs like ndiswrapper to provide their own stack, for instance.

Signature · 3 lines
   -bill davidsen ([email protected])
"The secret to procrastination is to put things off until the
 last possible moment - but no longer"  -me
#11 lore raw
Quoted 1 line · Bill Davidsen

You must have something more useful to work on, which would ADD value to the kernel instead of breaking existing installations. Ripping out petty stuff which works is a waste of your time and talent, please find something better to do.

Ahem. Please...

Quoted 1 line

Perhaps devise a way for programs like ndiswrapper to provide their own stack, for instance.

...follow your own suggestion instead of hammering someone else.

I've seen some discussion. More of that, and less of this please.

-Mike

#12 lore raw
Quoted 5 lines · Mike Galbraith
Quoted 1 line · Bill Davidsen

You must have something more useful to work on, which would ADD value to the kernel instead of breaking existing installations. Ripping out petty stuff which works is a waste of your time and talent, please find something better to do.

Quoted 8 lines

Ahem. Please...

Quoted 1 line

Perhaps devise a way for programs like ndiswrapper to provide their own stack, for instance.

...follow your own suggestion instead of hammering someone else.

I've seen some discussion. More of that, and less of this please.

Frankly this should be done by someone who really understands the code, and considering the time it's likely to take it would probably be (a) someone with a desparate need, (b) someone rich or retired who doesn't work for a living and has the time, or (c) someone who works for a company which sells Linux distributions and therefore could get paid to do this. That lets me out on all counts, I would resent wasting the time to patch 8KSTACKS back in as a patch, but I could do that to make laptops useful. As Andi pointed out some architectures can't run 4k stacks, and at the memory sizes people typically use there would probably be a performance gain to do memory in 8k or larger blocks anyway.

I just see this as a large hassle for many laptop users and people with unconverted drivers, and no significant gain for most. 4k stacks work fine on most machines, but some people just can't use them.

Signature · 3 lines
   -bill davidsen ([email protected])
"The secret to procrastination is to put things off until the
 last possible moment - but no longer"  -me
#13 lore raw

Large File Support in Kernel 2.6

I have a question about the Large File Support using Linux and glibc 2.3 on a 32-Bit machine. What's the correct limit for the file size and the file system using LFS (just for the kernel, not to mention filesystem limits etc)?

I found two references:

"The 2.6 kernel imposes its own limits on the size of files and file systems handled by it. These are as follows:
- file size: On 32-bit systems, files may not exceed the size of 2 TB.
- file system size: File systems may be up to 2e73 bytes large. However, this limit is still out of reach for the currently available hardware."

Source:
http://www.novell.com/documentation/suse91/suselinux-adminguide/html/apas04.html

"Kernel 2.6: For both 32-bit systems with option CONFIG_LBD set and for 64-bit systems: The size of a file system is limited to 2e73 (far too much for today). On 32-bit systems (without CONFIG_LBD set) the size of a file is limited to 2 TiB. Note that not all filesystems and hardware drivers might handle such large filesystems."

Source: http://www.suse.de/~aj/linux_lfs.html

I think it's 2TB for the file size and 2e73 for the file system, but I don't understand the second reference and the part about the CONIFG_LBD. What is exactly the CONFIG_LBD option?

#14 lore raw

Re: Large File Support in Kernel 2.6

Quoted 4 lines · On Thu, 08 Sep 2005 21:27:42 +0200 Andreas Baer <[email protected]> wrote:

I think it's 2TB for the file size and 2e73 for the file system, but I don't understand the second reference and the part about the CONIFG_LBD. What is exactly the CONFIG_LBD option?
-

This is "Support for Large Block Devices" under Device Drivers/Block Devices:

CONFIG_LBD:

Say Y here if you want to attach large (bigger than 2TB) discs to your machine, or if you want to have a raid or loopback device bigger than 2TB. Otherwise say N.

The "2e73" refers to 2 to the exponent 73 bytes in size. Huge :-)

#15 lore raw

Re: Large File Support in Kernel 2.6

Quoted 14 lines · Mike Houston
Quoted 4 lines · On Thu, 08 Sep 2005 21:27:42 +0200 Andreas Baer <[email protected]> wrote:

I think it's 2TB for the file size and 2e73 for the file system, but I don't understand the second reference and the part about the CONIFG_LBD. What is exactly the CONFIG_LBD option?
-

This is "Support for Large Block Devices" under Device Drivers/Block Devices:

Quoted 7 lines

CONFIG_LBD:

Say Y here if you want to attach large (bigger than 2TB) discs to your machine, or if you want to have a raid or loopback device bigger than 2TB. Otherwise say N.

The "2e73" refers to 2 to the exponent 73 bytes in size. Huge :-)

So in other words, both the file size and the file system limit is 2e73 using CONFIG_LBD option, right? And 2TB are always possible?

Sorry, but I need to get pretty sure.

Quoted 7 lines

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#16 lore raw
In reply to Adrian Bunk
Quoted 2 lines · Adrian Bunk

Please name situations where 8K stacks may be preferred that do not involve binary-only modules.

How about NFS-exporting a filesystem on LVM atop md? I believe it has been mentioned before in discussions that 8k stacks are strongly recommended in this case. Are those issues solved?

Thorild Selén
Datorföreningen Update / Update Computer Club, Uppsala, SE

#17 lore raw
Quoted 7 lines · Thorild Selen
Quoted 2 lines · Adrian Bunk

Please name situations where 8K stacks may be preferred that do not involve binary-only modules.

How about NFS-exporting a filesystem on LVM atop md? I believe it has been mentioned before in discussions that 8k stacks are strongly recommended in this case. Are those issues solved?

I think the worst overflow case anyone found was nfs=>xfs=>lvm=>dm=>scsi, if someone has such a configuration, please retest with current -mm or similar. I think there are several patches in there to resolve the excessive stack usage and a few to do some sort of bio chaining (Instead of recursive calls). I don't remember what underlying hardware was behind the SCSI, but I suspect something like iSCSI or USB would push some extra stack in there for stress testing.

Cheers,
Kyle Moffett

--
I have yet to see any problem, however complicated, which, when you looked at it in the right way, did not become still more complicated.
-- Poul Anderson

#18 lore raw
In reply to Bas Westerbaan
Quoted 8 lines · Bas Westerbaan
Quoted 4 lines

Yes you are. You're asking for 4KSTACKS config option to maintained and it is not something you get for free. Besides, if it is indeed ripped out of mainline kernel, you can always keep it a separate patch for ndiswrapper.

Though 4K stacks are used a lot, they probably aren't used on all configurations yet. Other situations may arise where 8K stacks may be preferred. It is too early to kill 8K stacks imho.

At least Fedora ships 4Kstacks kernel for quite a while now. No, it is not "everywhere", but close.

Signature · 4 lines
Dr. Horst H. von Brand                   User #22616 counter.li.org
Departamento de Informatica                     Fono: +56 32 654431
Universidad Tecnica Federico Santa Maria              +56 32 654239
Casilla 110-V, Valparaiso, Chile                Fax:  +56 32 797513