Kernel 2.6.38 really impresses!

Message

Post by **asturm** » Thu Mar 31, 2011 6:00 pm

For me, 2.6.38 currently mostly impresses with silence. Even 2.6.38.2 didn't fix sound and there are no fixes currently in stable-queue for 2.6.38, so I'm trying to revert the sound tree back to 2.6.37...

VinzC · Post by **VinzC** » Fri Apr 01, 2011 11:39 am

genstorm wrote:For me, 2.6.38 currently mostly impresses with silence. Even 2.6.38.2 didn't fix sound and there are no fixes currently in stable-queue for 2.6.38, so I'm trying to revert the sound tree back to 2.6.37...

So what is your sound chipset?

dennisn · Post by **dennisn** » Fri Apr 01, 2011 6:56 pm

I just tried "make oldconfig"ing from 2.6.37, to 2.6.38.2, as I've done countless times before, except this time something is seriously wrong -- the second grub loads the new kernel, there is an endless stream of backtrace-type debug messages that flood my screen -- I have no way of reading what's going on. What the heck happened?!?

VinzC · Post by **VinzC** » Sat Apr 02, 2011 11:20 pm

dennisn wrote:I just tried "make oldconfig"ing from 2.6.37, to 2.6.38.2, as I've done countless times before, except this time something is seriously wrong -- the second grub loads the new kernel, there is an endless stream of backtrace-type debug messages that flood my screen -- I have no way of reading what's going on. What the heck happened?!?

Weird. That's how I usually configure/install new kernels too...

jbouzan · Post by **jbouzan** » Sat Apr 02, 2011 11:22 pm

VinzC wrote:
dennisn wrote:I just tried "make oldconfig"ing from 2.6.37, to 2.6.38.2, as I've done countless times before, except this time something is seriously wrong -- the second grub loads the new kernel, there is an endless stream of backtrace-type debug messages that flood my screen -- I have no way of reading what's going on. What the heck happened?!?
Weird. That's how I usually configure/install new kernels too...

I did that and it works fine. Bad compile perhaps? I once had a kernel that didn't work, and with no changes to its .config it worked after a recompile. File corruption or something. Or perhaps support for some hardware was dropped, which your computer needed.

wuzzerd · Post by **wuzzerd** » Sun Apr 03, 2011 12:02 am

In a new install using the exact same 2.6.38.2 there was no sound although it worked in the old borked install. The sound card(s) , mixers etc. all showed up in /dev. Yesterday it started working after an emerge -uDN world. Go figure. I did switch from the default profile to desktop.

VinzC · Post by **VinzC** » Mon Apr 04, 2011 7:16 am

The only serious glitch I'm experiencing so far is a kernel Oops in TTM. I have noted down the message partly but it always occurs while the screensaver is active. It looks like a threading issue. Will post what I have noted down later in this thread. Since I've installed 2.6.38-gentoo-r1 I get this panic message when the screensaver is active. Until then the panic message only occurred when unlocking the screen (or cancelling the saver).

dennisn · Post by **dennisn** » Tue Apr 05, 2011 4:49 pm

2.6.38.2 also broke my resuming from hibernation, which was working just fine in 2.6.36. On my x86_64 Acer Ferrari laptop, I get a "general protection fault" and kernel panic (in swapper / swsusp_arch_resume) whenever I try resuming. No good.

PaulBredbury · Post by **PaulBredbury** » Tue Apr 05, 2011 7:14 pm

dennisn wrote:broke my resuming from hibernation

Known upstream apparently.

Yamakuzure · Post by **Yamakuzure** » Wed Apr 06, 2011 8:26 am

DestroyFX wrote:Here from the timed kernel compilation test I made, with my X6 CPU:
(... snip ...)
The optimal compilation speed was with -jCore. The performance go down with -jCore+1 and more.

I remember that with the 2.6.36-, -jCore+1 was better except with BFS who required -jCore (and was more fast than without BFS with -jCore+1).

As this is something seldom used, I tried with emerging geany, nano, mlview, kile and hexedit with different MAKEOPTS:

With MAKEOPTS="-j4" (I leave the list in once for reference)

Code: Select all

 # time emerge --oneshot geany nano mlview kile hexedit

>>> Emerging (1 of 5) app-editors/hexedit-1.2.12
>>> Emerging (2 of 5) dev-util/geany-0.20
>>> Emerging (3 of 5) app-editors/nano-2.2.5
>>> Installing (1 of 5) app-editors/hexedit-1.2.12
>>> Emerging (4 of 5) app-editors/mlview-0.9.0
>>> Emerging (5 of 5) app-editors/kile-2.1_beta5
>>> Installing (3 of 5) app-editors/nano-2.2.5
>>> Installing (2 of 5) dev-util/geany-0.20
>>> Installing (5 of 5) app-editors/kile-2.1_beta5
>>> Installing (4 of 5) app-editors/mlview-0.9.0
>>> Jobs: 5 of 5 complete                           Load avg: 5.73, 4.35, 3.05

real    3m59.929s
user    8m1.822s
sys     1m5.482s

With MAKEOPTS="-j7"

Code: Select all

real    2m10.122s
user    2m23.061s
sys     0m36.148s

With MAKEOPTS="-j11"

Code: Select all

real    1m37.223s
user    2m25.297s
sys     0m34.484s

With MAKEOPTS="-j15"

Code: Select all

real    1m36.465s
user    2m24.854s
sys     0m34.618s

So for emerging the gain in speed stalls, but doesn't get worse with more parallel jobs.

I am testing kernel compilation next, but I reckon the results will be the same as yours.

=== Edit ===

And here are the tests with the kernel:

Code: Select all

 # make clean && time make -j 4
real    4m18.876s
user    13m24.801s
sys     1m9.835s

Code: Select all

 # make clean && time make -j 7
real    4m35.801s
user    14m7.941s
sys     1m11.109s

Code: Select all

 # make clean && time make -j 11
real    4m16.216s
user    14m7.978s
sys     1m9.202s

Code: Select all

 # make clean && time make -j 15
real    4m23.011s
user    14m11.338s
sys     1m9.929s

This certainly is weird. While using -j7 performs worse than -j4 (like suggested by your tests), -j15 performs better than -j7 and -j11 performs better than -j4. It's not much, but it is strange... (Or not so strange at all if I understood more of the internal mechanics

)

VinzC · Post by **VinzC** » Wed Apr 06, 2011 11:40 am

Yamakuzure wrote:This certainly is weird. While using -j7 performs worse than -j4 (like suggested by your tests), -j15 performs better than -j7 and -j11 performs better than -j4. It's not much, but it is strange... (Or not so strange at all if I understood more of the internal mechanics )

Isn't to me. "Performs" is not the appropriate term. The patch doesn't improve compile times, it just improves responsiveness, which is totally different. It means, for instance, at equal load, groups of interactive processes in the same session will have better chances to catch a keyboard/mouse events on time and actually *do* the requested operation in a timely fashion. But globally I expect compile times to be just slightly bigger than before.

It's only when the amount of I/O grows such as there's a latency due to excess of I/O operations. So compile times aren't a good measure of the patch efficiency. Don't expect compile times to vary significantly then.

Yamakuzure · Post by **Yamakuzure** » Thu Apr 07, 2011 7:36 am

Well, yes, during all these time tests my laptop was perfectly responsive all the time. I just did the tests because someone stated that it would be stupid to use a different value for the -j option than the number of logical CPUs. That has been proven wrong, now.

cach0rr0 · Post by **cach0rr0** » Thu Apr 07, 2011 8:48 am

just to be a pest and BFS fanboy
test during which my desktop (KDE-4.6.2 with compositing enabled, bluray rip playing, and music playing just for grins) responds as though nothing else is going on

quad core phenom

Code: Select all

#make clean && time make -j4
real    2m53.167s
user    8m58.095s
sys     0m43.128s
ricker linux # ls -alh arch/x86/boot/bzImage 
-rw-r--r-- 1 root root 5.9M Apr  7 03:17 arch/x86/boot/bzImage #roughly half of this is embedded initramfs

Anon-E-moose · Post by **Anon-E-moose** » Thu Apr 07, 2011 11:51 am

Yamakuzure wrote:Well, yes, during all these time tests my laptop was perfectly responsive all the time. I just did the tests because someone stated that it would be stupid to use a different value for the -j option than the number of logical CPUs. That has been proven wrong, now.

The only thing you've proven is that in running a few tests that things worked well on your system

Anon-E-moose · Post by **Anon-E-moose** » Thu Apr 07, 2011 11:58 am

I've been running BFS for the last several kernel versions going back to 2.6.32 or thereabouts and I've had consistently good performance.
I'm glad that they are finally addressing the problems with performance for those not running BFS.

So once again, for those who don't want to run BFS then run whatever you want.
And for those who want to run BFS it's there.

VanFanel · Post by **VanFanel** » Fri Apr 08, 2011 11:57 am

I've been using BFS since 2.6.32: thanks for the ck-sources package, it's what makes my system so great!

But.. would you recommend automatic task grouping over BFS for emulation?
I'm trying to build a near-zero latency system for both audio & input in emulators and I don't know if BFS is the best option for it.

regards

Anon-E-moose · Post by **Anon-E-moose** » Fri Apr 08, 2011 12:09 pm

VanFanel wrote:I've been using BFS since 2.6.32: thanks for the ck-sources package, it's what makes my system so great!

But.. would you recommend automatic task grouping over BFS for emulation?
I'm trying to build a near-zero latency system for both audio & input in emulators and I don't know if BFS is the best option for it.

regards

The only way to know for sure on your system is to switch to the task grouping and rebuild the kernel.
You can have both and switch back and forth to see which gives you better performance on your system.

Ant P. · Post by **Ant P.** » Fri Apr 08, 2011 1:17 pm

Using schedtool will probably help a bit there too.

VanFanel · Post by **VanFanel** » Sat Apr 09, 2011 8:06 pm

@Ant P. : I use schedtool like this:

Code: Select all

schedtool -I -e ./mame tekken2

any other advice for the lowest possible input/output latency in emulators?

jbouzan · Post by **jbouzan** » Sat Apr 09, 2011 11:30 pm

PaulBredbury wrote:
dennisn wrote:broke my resuming from hibernation
Known upstream apparently.

Annoying, but since .38 patches a problem I was having with btrfs I can't really switch back. Any idea whether the patch will be reversed?

epsilon72 · Post by **epsilon72** » Wed Apr 13, 2011 5:48 am

PaulBredbury wrote:
dennisn wrote:broke my resuming from hibernation
Known upstream apparently.

This seems to happen with 2.6.37-r4 as well though

piedar · Post by **piedar** » Wed Apr 13, 2011 6:35 pm

jbouzan wrote:
PaulBredbury wrote:
dennisn wrote:broke my resuming from hibernation
Known upstream apparently.
Annoying, but since .38 patches a problem I was having with btrfs I can't really switch back. Any idea whether the patch will be reversed?

Just apply the patch yourself in the meantime.
https://patchwork.kernel.org/patch/690691/

pilla · Post by **pilla** » Sun Apr 17, 2011 1:22 pm

marshmallow1304 wrote:
Just apply the patch yourself in the meantime.
https://patchwork.kernel.org/patch/690691/

404 Not found

2.6.38-gentoo-r2 is still segfaulting Xorg for me.

VinzC · Post by **VinzC** » Tue Apr 19, 2011 5:15 pm

pilla wrote:2.6.38-gentoo-r2 is still segfaulting Xorg for me.

Does it crash right away or after a certain, random period of time?

Post by **asturm** » Tue Apr 19, 2011 5:22 pm

Sound is ok with unpatched 2.6.38.3 now