Official thread: "zen-sources" - Part 7

Message

Waninkoko · Post by **Waninkoko** » Tue Sep 08, 2009 5:10 pm

I merged airlied's drm-next branch (with an extra fix) so now we can use in-kernel radeon drm to get 2d/3d acceleration and kms with r600/r700 cards

It works fine here (RV770) but I can't get KMS and X working simultaneously (I have to disable KMS).

kernelOfTruth · Post by **kernelOfTruth** » Tue Sep 08, 2009 5:20 pm

Waninkoko wrote:I merged airlied's drm-next branch (with an extra fix) so now we can use in-kernel radeon drm to get 2d/3d acceleration and kms with r600/r700 cards It works fine here (RV770) but I can't get KMS and X working simultaneously (I have to disable KMS).

that means dri2 works with that card ?

if yes - how so and which settings do I need to enable additionally to the old ones ?

<M> Direct Rendering Manager (XFree86 4.1.0 and higher DRI support
-> <M> ATI Radeon

edit:

I see: additionally enabling staging drivers then

enable "Enable modesetting on radeon by default"

edit2:

Fran got it working:

http://phoronix.com/forums/showthread.php?t=18959

kernelOfTruth · Post by **kernelOfTruth** » Tue Sep 08, 2009 9:37 pm

@Waninkoko:

I got it working - unfortunately it was pretty unstable and hence unusable for me right now

thanks for merging !

I guess I'll still have to keep using fglrx for some time

Naib · Post by **Naib** » Wed Sep 09, 2009 9:24 am

Very nice kernelset
Last customset I was impressed with was the love sources

One thing:
The BFS, that shouldn't be under the I/O sched.
CFS and BFS are process sched and not I/O sched. We have been able to choose between I/O for a long time (DEADLINE, CFQ,...) but never process due to the kernel dev saying there should be no choice.

So

1) shouldn't CFS and BFS as a choice come under its own choice
2) As it stands I pick BFS, so what is the I/O schedular since you can't choose that (due to process sched in the same radio list as i/o sched)

rahulthewall · Post by **rahulthewall** » Wed Sep 09, 2009 11:18 am

Naib wrote:Very nice kernelset
Last customset I was impressed with was the love sources

One thing:
The BFS, that shouldn't be under the I/O sched.
CFS and BFS are process sched and not I/O sched. We have been able to choose between I/O for a long time (DEADLINE, CFQ,...) but never process due to the kernel dev saying there should be no choice.

So

1) shouldn't CFS and BFS as a choice come under its own choice
2) As it stands I pick BFS, so what is the I/O schedular since you can't choose that (due to process sched in the same radio list as i/o sched)

Last I chekced, they were separate.

Naib · Post by **Naib** » Wed Sep 09, 2009 11:32 am

mmm my mmistake, so you can't choose between BFS and CFS, its BFS by default
ok

aTan · Post by **aTan** » Wed Sep 09, 2009 12:21 pm

Naib wrote:mmm my mmistake, so you can't choose between BFS and CFS, its BFS by default
ok

You can choose between BFS and CFS in Zen Sources. Also you can enable/disable AUTOISO for X.

Roman_Gruber · Post by **Roman_Gruber** » Wed Sep 09, 2009 12:28 pm

Hello, short question. If I want to use these sources, is tuxonice included, lvm support and luks support? thx.

And is the machine, sorry for saying it like a nuub, more efficent?

kernelOfTruth · Post by **kernelOfTruth** » Wed Sep 09, 2009 12:30 pm

Naib wrote:Very nice kernelset
Last customset I was impressed with was the love sources

One thing:
The BFS, that shouldn't be under the I/O sched.
CFS and BFS are process sched and not I/O sched. We have been able to choose between I/O for a long time (DEADLINE, CFQ,...) but never process due to the kernel dev saying there should be no choice.

So

1) shouldn't CFS and BFS as a choice come under its own choice
2) As it stands I pick BFS, so what is the I/O schedular since you can't choose that (due to process sched in the same radio list as i/o sched)

you mean BFQ ?:

Symbol: IOSCHED_BFQ [=y] │
│ Prompt: BFQ I/O scheduler │
│ Defined at block/Kconfig.iosched:49 │
│ Depends on: BLOCK && EXPERIMENTAL │
│ Location: │
│ -> Enable the block layer (BLOCK [=y]) │
│ -> IO Schedulers │
│ │
│ │
│ Symbol: DEFAULT_BFQ [=n] │
│ Prompt: BFQ │
│ Defined at block/Kconfig.iosched:94 │
│ Depends on: <choice> && IOSCHED_BFQ=y │
│ Location: │
│ -> Enable the block layer (BLOCK [=y]) │
│ -> IO Schedulers

it's an improved version of CFQ

Naib wrote:mmm my mmistake, so you can't choose between BFS and CFS, its BFS by default
ok

nope, it's CFS or BFS and they are to be set under "General setup"

see:

[*] Prompt for development and/or incomplete code/drivers │ │
│ │ () Local version - append to kernel release │ │
│ │ [ ] Automatically append version information to the version strin│ │
│ │ ZenSched CPU Scheduler Choice (Completely Fair Scheduler (CFS

ZenSched CPU Scheduler Choice ───────────────────┐
│ Use the arrow keys to navigate this window or press the hotkey of │
│ the item you wish to select followed by the <SPACE BAR>. Press │
│ <?> for additional information about this option. │
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │ (X) Completely Fair Scheduler (CFS) │ │
│ │ ( ) Brainfuck Scheduler (BFS)

(that's from 2.6.31-rc9-zen1)

kernelOfTruth · Post by **kernelOfTruth** » Wed Sep 09, 2009 12:38 pm

tw04l124 wrote:Hello, short question. If I want to use these sources, is tuxonice included, lvm support and luks support? thx.

And is the machine, sorry for saying it like a nuub, more efficent?

tuxonice ???

I dunno

you get lvm and luks-support via initrd / genkernel

it's way more "efficient", provided you use BFS and BFQ (BFS reveals some mainline race conditions and bugs and therefore isn't that stable yet, for me it just stalls at boot and doesn't continue

)

for now you could continue using CFS and disable some unneeded stuff such as group scheduling,
and the following set

CONFIG_SCHED_OMIT_FRAME_POINTER=y
# CONFIG_SCHED_SMT is not set
# CONFIG_SCHED_MC is not set

it depends if you need CONFIG_SCHED_MC or not ...

rahulthewall · Post by **rahulthewall** » Wed Sep 09, 2009 2:02 pm

TuxOnIce is not there.

ponciarello · Post by **ponciarello** » Wed Sep 09, 2009 3:42 pm

http://zen-sources.org/content/tuxonice-gone

Naib · Post by **Naib** » Wed Sep 09, 2009 6:22 pm

kernelOfTruth wrote:
Naib wrote:Very nice kernelset
Last customset I was impressed with was the love sources

One thing:
The BFS, that shouldn't be under the I/O sched.
CFS and BFS are process sched and not I/O sched. We have been able to choose between I/O for a long time (DEADLINE, CFQ,...) but never process due to the kernel dev saying there should be no choice.

So

1) shouldn't CFS and BFS as a choice come under its own choice
2) As it stands I pick BFS, so what is the I/O schedular since you can't choose that (due to process sched in the same radio list as i/o sched)
you mean BFQ ?:

Symbol: IOSCHED_BFQ [=y] │
│ Prompt: BFQ I/O scheduler │
│ Defined at block/Kconfig.iosched:49 │
│ Depends on: BLOCK && EXPERIMENTAL │
│ Location: │
│ -> Enable the block layer (BLOCK [=y]) │
│ -> IO Schedulers │
│ │
│ │
│ Symbol: DEFAULT_BFQ [=n] │
│ Prompt: BFQ │
│ Defined at block/Kconfig.iosched:94 │
│ Depends on: <choice> && IOSCHED_BFQ=y │
│ Location: │
│ -> Enable the block layer (BLOCK [=y]) │
│ -> IO Schedulers
it's an improved version of CFQ

Naib wrote:mmm my mmistake, so you can't choose between BFS and CFS, its BFS by default
ok
nope, it's CFS or BFS and they are to be set under "General setup"

see:
[*] Prompt for development and/or incomplete code/drivers │ │
│ │ () Local version - append to kernel release │ │
│ │ [ ] Automatically append version information to the version strin│ │
│ │ ZenSched CPU Scheduler Choice (Completely Fair Scheduler (CFS

ZenSched CPU Scheduler Choice ───────────────────┐
│ Use the arrow keys to navigate this window or press the hotkey of │
│ the item you wish to select followed by the <SPACE BAR>. Press │
│ <?> for additional information about this option. │
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │ (X) Completely Fair Scheduler (CFS) │ │
│ │ ( ) Brainfuck Scheduler (BFS)
(that's from 2.6.31-rc9-zen1)

mmm ok I don't have those options o.O

[ebuild R ] sys-kernel/zen-sources-2.6.30-r1 USE="-build -symlink" 0 kB [1]

Code: Select all

 [*] Prompt for development and/or incomplete code/drivers                                       │ │  
  │ │                         ()  Local version - append to kernel release                                                    │ │  
  │ │                         [*] Automatically append version information to the version string                              │ │  
  │ │                             Kernel compression mode (LZMA)  --->                                                        │ │  
  │ │                         [*] Support for paging of anonymous memory (swap)                                               │ │  
  │ │                         [*] System V IPC                                                                                │ │  
  │ │                         [*] POSIX Message Queues                                                                        │ │  
  │ │                         [ ] BSD Process Accounting                                                                      │ │  
  │ │                         [ ] Export task/process statistics through netlink (EXPERIMENTAL)                               │ │  
  │ │                         [ ] Auditing support                                                                            │ │  
  │ │                             RCU Subsystem  --->                                                                         │ │  
  │ │                         <*> Kernel .config support                                                                      │ │  
  │ │                         [*]   Enable access to .config through /proc/config.gz                                          │ │  
  │ │                         (18) Kernel log buffer size (16 => 64KB, 17 => 128KB)                                           │ │  
  │ │                         [*] Group CPU scheduler                                                                         │ │  
  │ │                         [*]   Group scheduling for SCHED_OTHER                                                          │ │  
  │ │                         [*]   Group scheduling for SCHED_RR/FIFO                                                        │ │  
  │ │                               Basis for grouping tasks (user id)  --->                                                  │ │  
  │ │                         [ ] Control Group support  --->                                                                 │ │  
  │ │                         [ ] Create deprecated sysfs layout for older userspace tools                                    │ │  
  │ │                         [ ] Kernel->user space relay support (formerly relayfs)                                         │ │  
  │ │                         -*- Namespaces support

Naib · Post by **Naib** » Wed Sep 09, 2009 6:37 pm

so the ebuilds in the overlay havn't been updated then

kernelOfTruth · Post by **kernelOfTruth** » Wed Sep 09, 2009 8:52 pm

Naib wrote:so the ebuilds in the overlay havn't been updated then

exactly,

you need at least 2.6.30-zen5 for which a patch does't exist (yet)

so the best practice would be to create a local git-repository

see first page of this thread for more information

pappy_mcfae · Post by **pappy_mcfae** » Wed Sep 09, 2009 11:51 pm

The second uniprocessor machine I have has finished it's emerge -aev world. It continues to run with the .31-rc with bfs. .31-rc would not boot on my amd64 after six attempts.

Blessed be!
Pappy

darklegion · Post by **darklegion** » Thu Sep 10, 2009 4:11 am

pappy_mcfae wrote:The second uniprocessor machine I have has finished it's emerge -aev world. It continues to run with the .31-rc with bfs. .31-rc would not boot on my amd64 after six attempts.

Blessed be!
Pappy

Yeah, I've never been able to get a 2.6.31 kernel to boot on my AMD64 box, but most of the 2.6.30 kernels have had no issues, either vanilla + bfs or zen-sources.
The just released BFS-211 which is released for 2.6.31, doesn't work either.Maybe I'll try booting my i686-chroot, and see if it works there.

razum2um · Post by **razum2um** » Thu Sep 10, 2009 4:01 pm

Is it possible to run ati-drivers with zen-git (master)?

I've modified ebuild to skip kernel version check
It compiles successfully, but modprobe fglrx fails with "unknown symbol find_task_by_vpid"

I've tried to disctinct the difference between versions about this:

Code: Select all

grep find_task_by_vpid -R /usr/src/linux-2.6.30-zen4/ | egrep "^/" | sed 's/\/.*\///' > 30-find_task_by_vpid
grep find_task_by_vpid -R /usr/src/linux-9999-zen/ | egrep "^/" | sed 's/\/.*\///' > 31-find_task_by_vpid

and

Code: Select all

diff 30-find_task_by_vpid 31-find_task_by_vpid

shows

0a1
> sched_bfs.c: return pid ? find_task_by_vpid(pid) : current;
2d2
< pid.c:EXPORT_SYMBOL(find_task_by_vpid);
12a13
> perf_counter.c: task = find_task_by_vpid(pid);
17a19
> sched_cfs.c: return pid ? find_task_by_vpid(pid) : current;
26d27
< sched.c: return pid ? find_task_by_vpid(pid) : current;
30,32c31
< System.map:ffffffff810417a7 T find_task_by_vpid
< System.map:ffffffff814a46c0 r __ksymtab_find_task_by_vpid
< System.map:ffffffff814b6c38 r __kstrtab_find_task_by_vpid
---
> System.map:ffffffff8103dc8c T find_task_by_vpid
43d41
< Module.symvers:0x00000000 find_task_by_vpid vmlinux EXPORT_SYMBOL

and I can't imagine what to do further (
any ideas?

===
sorry for my english..

Waninkoko · Post by **Waninkoko** » Thu Sep 10, 2009 10:36 pm

In kernel/pid.c

after the function "find_task_by_vpid", add this line:

Code: Select all

EXPORT_SYMBOL(find_task_by_vpid);

Recompile the kernel (fglrx not needed) and it should work fine.

xmixahlx · Post by **xmixahlx** » Fri Sep 11, 2009 3:02 pm

tried bfs again with 2.6.31-zen0 and hasn't blown up on me yet.

does it seem to anyone else that intensive tasks are slower or latent? it took forever for avidemux to index DVD VOBs and a few other CPU intensive tasks.

ponciarello · Post by **ponciarello** » Fri Sep 11, 2009 3:24 pm

- gitted 2.6.31-zen0
- applied test2 patch
- init 1 with wireless networking and sshd

Code: Select all

Using 2130255 loops per ms, running every load for 30 seconds
Benchmarking kernel 2.6.31-zen0-test2 at datestamp 200909110944

--- Benchmarking simulated cpu of Audio in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU  % Deadlines Met
None      0.007 +/- 0.00711    0.016             100            100
Video     0.006 +/- 0.00656    0.013             100            100
X         0.007 +/- 0.0076     0.018             100            100
Burn      0.006 +/- 0.00615    0.007             100            100
Write     0.011 +/- 0.0534     0.988             100            100
Read      0.008 +/- 0.0091     0.014             100            100
Compile   0.008 +/- 0.019      0.415             100            100
Memload    0.01 +/- 0.0114     0.031             100            100

--- Benchmarking simulated cpu of Video in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU  % Deadlines Met
None      0.006 +/- 0.00707    0.028             100            100
X         0.006 +/- 0.00644    0.025             100            100
Burn       11.6 +/- 14.3        31.7             100           36.1
Write     0.012 +/- 0.0909      3.65             100            100
Read      0.007 +/- 0.00755    0.031             100            100
Compile      12 +/- 14.8        38.5            97.9           33.8
Memload   0.009 +/- 0.0295      1.15             100            100

--- Benchmarking simulated cpu of X in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU  % Deadlines Met
None      0.006 +/- 0.0817         1             100           99.3
Video     0.033 +/- 0.258          2            98.7           97.7
Burn         54 +/- 75.4         164            19.6           8.88
Write     0.133 +/- 1.27          18            97.4           96.4
Read      0.023 +/- 0.208          2            99.3           98.3
Compile    57.5 +/- 80.2         196            19.3           8.51
Memload   0.106 +/- 0.688          6              98             96

--- Benchmarking simulated cpu of Gaming in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU
None      0.009 +/- 0.0499     0.452             100
Video      1.09 +/- 1.09        1.47            98.9
X          1.21 +/- 1.4          2.4            98.8
Burn        155 +/- 156          170            39.2
Write      2.25 +/- 4.68        27.7            97.8
Read      0.732 +/- 0.772       1.54            99.3
Compile     166 +/- 168          214            37.5
Memload    2.72 +/- 3.78          14            97.4

without test2 patch

Code: Select all

Using 2130255 loops per ms, running every load for 30 seconds
Benchmarking kernel 2.6.31-zen0 at datestamp 200909111918

--- Benchmarking simulated cpu of Audio in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU  % Deadlines Met
None      0.006 +/- 0.00699    0.018             100            100
Video     0.006 +/- 0.00635    0.035             100            100
X         0.006 +/- 0.00628    0.009             100            100
Burn      0.005 +/- 0.006      0.008             100            100
Write     0.013 +/- 0.0492      1.01             100            100
Read      0.009 +/- 0.00949    0.014             100            100
Compile   0.009 +/- 0.0108     0.127             100            100
Memload    0.01 +/- 0.0108     0.036             100            100

--- Benchmarking simulated cpu of Video in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU  % Deadlines Met
None      0.006 +/- 0.00705    0.021             100            100
X         0.006 +/- 0.00648    0.015             100            100
Burn       11.7 +/- 14          16.7             100           29.6
Write     0.026 +/- 0.456       17.4             100           99.9
Read      0.007 +/- 0.00741    0.016             100            100
Compile      12 +/- 14.3        33.8            98.4           28.7
Memload   0.011 +/- 0.0528      1.36             100            100

--- Benchmarking simulated cpu of X in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU  % Deadlines Met
None      0.003 +/- 0.0577         1             100           99.7
Video     0.026 +/- 0.231          2            98.7             98
Burn       54.4 +/- 75.8         164            19.3           8.69
Write     0.093 +/- 0.983         16              98             97
Read      0.023 +/- 0.208          2            99.3           98.3
Compile    56.8 +/- 78.8         182            19.2           8.44
Memload   0.159 +/- 1.52          24            97.4           96.1

--- Benchmarking simulated cpu of Gaming in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU
None      0.011 +/- 0.0475     0.443             100
Video      1.02 +/- 1.03        1.43              99
X          1.15 +/- 1.35        2.35            98.9
Burn        155 +/- 156          170            39.2
Write       1.4 +/- 3.11        23.1            98.6
Read      0.999 +/- 1.01        1.57              99
Compile     166 +/- 168          220            37.6
Memload    2.66 +/- 3.64          11            97.4

- untarred vanilla 2.6.31
- copied .config from zen
- make oldconfig
- make
- init 1 with wireless networking and sshd

Code: Select all

Using 2130255 loops per ms, running every load for 30 seconds
Benchmarking kernel 2.6.31 at datestamp 200909111517

--- Benchmarking simulated cpu of Audio in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU  % Deadlines Met
None      0.045 +/- 0.0452     0.056             100            100
Video     0.005 +/- 0.00567    0.007             100            100
X          0.03 +/- 0.0328     0.056             100            100
Burn      0.005 +/- 0.00511    0.007             100            100
Write     0.077 +/- 0.928       20.4             100            100
Read      0.011 +/- 0.0122     0.119             100            100
Compile    0.06 +/- 1.28        31.4             100            100
Memload   0.017 +/- 0.0551     0.922             100            100

--- Benchmarking simulated cpu of Video in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU  % Deadlines Met
None      0.007 +/- 0.0073     0.022             100            100
X         0.034 +/- 0.0355     0.103             100            100
Burn       8.37 +/- 13.9        37.6            99.7           79.8
Write     0.231 +/- 2.68        73.6            99.4           99.3
Read      0.009 +/- 0.0144     0.134             100            100
Compile    7.81 +/- 14.5        95.7            96.4           79.8
Memload   0.055 +/- 0.184       3.71             100            100

--- Benchmarking simulated cpu of X in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU  % Deadlines Met
None      0.006 +/- 0.0817         1             100           99.3
Video     0.033 +/- 0.258          2            98.7           97.7
Burn       49.7 +/- 76.8         224            16.9           8.87
Write       0.3 +/- 3.26          44            98.4           96.7
Read      0.023 +/- 0.208          2            99.3           98.3
Compile    53.7 +/- 83.9         225            17.9           9.43
Memload   0.053 +/- 0.44           5            98.4           97.4

--- Benchmarking simulated cpu of Gaming in the presence of simulated ---
Load    Latency +/- SD (ms)  Max Latency   % Desired CPU
None      0.004 +/- 0.0378     0.473             100
Video     0.971 +/- 0.976       1.22              99
X          1.14 +/- 1.35         2.9            98.9
Burn        135 +/- 137          209            42.6
Write      1.72 +/- 5.46        72.6            98.3
Read      0.925 +/- 0.933       1.43            99.1
Compile     173 +/- 178          251            36.7
Memload    1.48 +/- 1.9         5.37            98.5

pappy_mcfae · Post by **pappy_mcfae** » Fri Sep 11, 2009 8:07 pm

darklegion wrote:Yeah, I've never been able to get a 2.6.31 kernel to boot on my AMD64 box, but most of the 2.6.30 kernels have had no issues, either vanilla + bfs or zen-sources.
The just released BFS-211 which is released for 2.6.31, doesn't work either.Maybe I'll try booting my i686-chroot, and see if it works there.

I have tried it with multiprocessors running x86 and amd64...and it's a no go on either. It is a known bug, and apparently, it is still rather vigorous. I haven't tested with the latest patch, but reading that you and others still have issues means the bug is still alive and well.

I will hopefully get a chance to test the newest stuff later on today.

Blessed be!
Pappy

kernelOfTruth · Post by **kernelOfTruth** » Fri Sep 11, 2009 8:17 pm

anyone tried it on top of plain vanilla 2.6.31 ?

from what I read Con's system is amd64 so it at least should work for some of us

albright · Post by **albright** » Fri Sep 11, 2009 11:08 pm

well, for what it's worth, I'm running 2.6.31-zen0
with BFS enabled on amd64 and it's working fine

core i7 920 / nvidia GeForce 9600 GT ...

I can't say that things seem much "snappier" than
before but certainly no worse

Also running it on my thinkpad x300 and it
also works well (intel graphics seem definitely
better in .31) except that suspend to ram will
only work *once*. I kind of wish zen had
kept tuxonice patches

darklegion · Post by **darklegion** » Sat Sep 12, 2009 1:30 am

kernelOfTruth wrote:anyone tried it on top of plain vanilla 2.6.31 ?

from what I read Con's system is amd64 so it at least should work for some of us

No, doesn't work for me with vanilla 2.6.31.One thing I did notice, is that after making a fresh .config where I accidentally left out the correct options for my hard drives, is that it booted properly, at least as far as it would go without a root drive.Perhaps the bug is some interaction between the ATA subsystem and BFS?