Re: [PATCH 00/12] IO-less dirty throttling v7

[Sedat Dilek <sedat.dilek@googlemail.com>]

[-- Attachment #1: Type: text/plain, Size: 5632 bytes --]

On Sun, Apr 17, 2011 at 6:46 AM, Sedat Dilek <sedat.dilek@googlemail.com> wrote:
> On Sun, Apr 17, 2011 at 6:10 AM, Wu Fengguang <fengguang.wu@intel.com> wrote:
>> On Sun, Apr 17, 2011 at 11:18:43AM +0800, Sedat Dilek wrote:
>>> On Sun, Apr 17, 2011 at 3:44 AM, Wu Fengguang <fengguang.wu@intel.com> wrote:
>>> > Hi Sedat,
>>> >
>>> > On Sun, Apr 17, 2011 at 12:27:58AM +0800, Sedat Dilek wrote:
>>> >
>>> >> I pulled your tree into linux-next (next-20110415) on an i386 Debian host.
>>> >>
>>> >> My build breaks here:
>>> >> ...
>>> >>   MODPOST vmlinux.o
>>> >>   GEN     .version
>>> >>   CHK     include/generated/compile.h
>>> >>   UPD     include/generated/compile.h
>>> >>   CC      init/version.o
>>> >>   LD      init/built-in.o
>>> >>   LD      .tmp_vmlinux1
>>> >> mm/built-in.o: In function `bdi_position_ratio':
>>> >> page-writeback.c:(.text+0x5c83): undefined reference to `__udivdi3'
>>> >
>>> > Yes it can be fixed by the attached patch.
>>> >
>>> >> mm/built-in.o: In function `calc_period_shift.part.10':
>>> >> page-writeback.c:(.text+0x6446): undefined reference to `____ilog2_NaN'
>>> >
>>> > I cannot reproduce this error. In the git tree, calc_period_shift() is
>>> > actually quite simple:
>>> >
>>> > static int calc_period_shift(void)
>>> > {
>>> >        return 2 + ilog2(default_backing_dev_info.avg_write_bandwidth);
>>> > }
>>> >
>>> > where avg_write_bandwidth is of type "unsigned long".
>>> >
>>> >> make[4]: *** [.tmp_vmlinux1] Error
>>> >>
>>> >> BTW, which kernel-config options have to be set besides
>>> >> CONFIG_BLK_DEV_THROTTLING=y?
>>> >
>>> > No. I used your kconfig on 2.6.39-rc3 and it compiles OK for i386.
>>> >
>>> > I've pushed two patches into the git tree fixing the compile errors.
>>> > Thank you for trying it out and report!
>>> >
>>> > Thanks,
>>> > Fengguang
>>> >
>>>
>>> Thanks for your patch.
>>>
>>> The 1st part of the build-error is gone, but 2nd part still remains:
>>>
>>>   LD      .tmp_vmlinux1
>>> mm/built-in.o: In function `calc_period_shift.part.10':
>>> page-writeback.c:(.text+0x6458): undefined reference to `____ilog2_NaN'
>>> make[4]: *** [.tmp_vmlinux1] Error 1
>>>
>>> I have attached some disasm-ed files.
>>
>> OK. I tried next-20110415 and your kconfig and still got no error.
>>
>> Please revert the last commit. It's not necessary anyway.
>>
>> commit 84a9890ddef487d9c6d70934c0a2addc65923bcf
>> Author: Wu Fengguang <fengguang.wu@intel.com>
>> Date:   Sat Apr 16 18:38:41 2011 -0600
>>
>>    writeback: scale dirty proportions period with writeout bandwidth
>>
>>    CC: Peter Zijlstra <a.p.zijlstra@chello.nl>
>>    Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
>>
>>> Unfortunately, I don't see any new commits in your GIT tree.
>>
>> Yeah I cannot see it in the web interface, too:
>>
>> http://git.kernel.org/?p=linux/kernel/git/wfg/writeback.git;a=shortlog;h=refs/heads/dirty-throttling-v7
>>
>> But they are in the dirty-throttling-v7 branch at kernel.org:
>>
>> commit d0e30163e390d87387ec13e3b1c2168238c26793
>> Author: Wu Fengguang <fengguang.wu@intel.com>
>> Date:   Sun Apr 17 11:59:12 2011 +0800
>>
>>    Revert "writeback: scale dirty proportions period with writeout bandwidth"
>>
>>    This reverts commit 84a9890ddef487d9c6d70934c0a2addc65923bcf.
>>
>>    sedat.dilek@gmail.com:
>>
>>        LD      .tmp_vmlinux1
>>      mm/built-in.o: In function `calc_period_shift.part.10':
>>      page-writeback.c:(.text+0x6458): undefined reference to `____ilog2_NaN'
>>      make[4]: *** [.tmp_vmlinux1] Error 1
>>
>> commit fc5c8b04119a5bcc46865e66eec3e6133ecb56e9
>> Author: Wu Fengguang <fengguang.wu@intel.com>
>> Date:   Sun Apr 17 09:22:41 2011 -0600
>>
>>    writeback: quick CONFIG_BLK_DEV_THROTTLING=n compile fix
>>
>>    Reported-by: Sedat Dilek <sedat.dilek@googlemail.com>
>>    Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
>>
>> commit c4a7e3f48dcfae71d0e3d2c55dcc381b3def1919
>> Author: Wu Fengguang <fengguang.wu@intel.com>
>> Date:   Sun Apr 17 09:04:44 2011 -0600
>>
>>    writeback: i386 compile fix
>>
>>    mm/built-in.o: In function `bdi_position_ratio':
>>    page-writeback.c:(.text+0x5c83): undefined reference to `__udivdi3'
>>    mm/built-in.o: In function `calc_period_shift.part.10':
>>    page-writeback.c:(.text+0x6446): undefined reference to `____ilog2_NaN'
>>    make[4]: *** [.tmp_vmlinux1] Error
>>
>>    Reported-by: Sedat Dilek <sedat.dilek@googlemail.com>
>>    Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
>>
>>
>> Thanks,
>> Fengguang
>>
>
> The 2nd part disappears here, when I switch from gcc-4.6 to gcc-4.5.
>
> - Sedat -
>

Just FYI:

Build with gcc-4.6 is OK, now.

  (+) OK   writeback-dirty-throttling-v7-fix/writeback-i386-compile-fix.patch
  (+) OK   writeback-dirty-throttling-v7-fix/0001-Revert-writeback-scale-dirty-proportions-period-with.patch

In case of the ilog2 error I have g00gled a bit and found gcc-bug
#36359 and looked also at include/linux/log2.h [2].
Not sure if this is a gcc-4.6 bug and I shall open a ticket (or a
problem in your code?).
I have tried the testcase from [3], it gives same output for nm.
[4] I could not follow.

- Sedat -

[1] http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36359
[2] http://git.us.kernel.org/?p=linux/kernel/git/wfg/writeback.git;a=blob;f=include/linux/log2.h#l18
[3] http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36359#c12
[4] http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36359#c17

[-- Attachment #2: testcase_gcc-bug-36359.tar.xz --]
[-- Type: application/octet-stream, Size: 948 bytes --]

[-- Attachment #3: testcase_gcc-bug-36359.tar.xz.sha256sum --]
[-- Type: application/octet-stream, Size: 96 bytes --]

c5ab2c29420772d49e4942ec72999f42b8393d14c7762427ac385edaa8936af2  testcase_gcc-bug-36359.tar.xz