[Qemu-devel] Commit 5632ae46 broke the network on mips.

[Qemu-devel] Commit 5632ae46 broke the network on mips.

[Rob Landley]

commit 5632ae46: "mips_malta: move i8259 initialization after piix4
initialization" broke the network on mips.  It still comes up, but
doesn't pass packets.

Try this:

  wget http://landley.net/aboriginal/bin/system-image-mips.tar.bz2
  tar xvjf system-image-mips.tar.bz2
  cd system-image-mips
  ./run-emulator.sh

  [ wait for shell prompt ]

  wget http://google.com

Under 0.15.1 it works, under 1.0 it doesn't.  Identical binary.  I
bisected it to the commit in question, it's still broken in -master.

Rob

Re: [Qemu-devel] Commit 5632ae46 broke the network on mips.

[Stefan Weil]

Am 21.01.2012 20:04, schrieb Rob Landley:
> commit 5632ae46: "mips_malta: move i8259 initialization after piix4
> initialization" broke the network on mips.  It still comes up, but
> doesn't pass packets.
>
> Try this:
>
>    wget http://landley.net/aboriginal/bin/system-image-mips.tar.bz2
>    tar xvjf system-image-mips.tar.bz2
>    cd system-image-mips
>    ./run-emulator.sh
>
>    [ wait for shell prompt ]
>
>    wget http://google.com
>
> Under 0.15.1 it works, under 1.0 it doesn't.  Identical binary.  I
> bisected it to the commit in question, it's still broken in -master.
>
> Rob

This was fixed with commits e9b40fd34ceb23461083d505a444a389c094455b
and 0b23c5d40ea933cfece3b4f69427f79c8a23256d in master and stable-1.0.

My latest QEMU works with MIPS Malta running from an NFS root.

Regards,
Stefan

Re: [Qemu-devel] Commit 5632ae46 broke the network on mips.

[Rob Landley]

On 01/22/2012 01:06 AM, Stefan Weil wrote:
> Am 21.01.2012 20:04, schrieb Rob Landley:
>> commit 5632ae46: "mips_malta: move i8259 initialization after piix4
>> initialization" broke the network on mips.  It still comes up, but
>> doesn't pass packets.
>>
>> Try this:
>>
>>    wget http://landley.net/aboriginal/bin/system-image-mips.tar.bz2
>>    tar xvjf system-image-mips.tar.bz2
>>    cd system-image-mips
>>    ./run-emulator.sh
>>
>>    [ wait for shell prompt ]
>>
>>    wget http://google.com
>>
>> Under 0.15.1 it works, under 1.0 it doesn't.  Identical binary.  I
>> bisected it to the commit in question, it's still broken in -master.
>>
>> Rob
> 
> This was fixed with commits e9b40fd34ceb23461083d505a444a389c094455b
> and 0b23c5d40ea933cfece3b4f69427f79c8a23256d in master and stable-1.0.

The git tag for v1.0 does not include that commit.  Apparently the
current 1.0 tarball on the website does, which is odd since e9b4 was
only applied to the git tree _after_ a January 6th commit.  So either
"git log" is being weird, or you guys retroactively re-released the 1.0
tarball without renumbering it?

I'm confused...

> My latest QEMU works with MIPS Malta running from an NFS root.

I can confirm that current -master works for me.

The 1.0 tarball includes the fix, the 1.0 git tag doesn't.  I'll include
a note to that effect in my release notes.

Thanks,

Rob

Re: [Qemu-devel] Commit 5632ae46 broke the network on mips.

[Anthony Liguori]

On 01/22/2012 05:42 PM, Rob Landley wrote:
> On 01/22/2012 01:06 AM, Stefan Weil wrote:
>> This was fixed with commits e9b40fd34ceb23461083d505a444a389c094455b
>> and 0b23c5d40ea933cfece3b4f69427f79c8a23256d in master and stable-1.0.
>
> The git tag for v1.0 does not include that commit.  Apparently the
> current 1.0 tarball on the website does,

Hi Rob,

I've verified this three different ways..  You must have something messed up on 
your end.

v1.0 tag is:

commit 1c8a881daaca6fe0646a425b0970fb3ad25f6732
Author: Anthony Liguori <aliguori@us.ibm.com>
Date:   Thu Dec 1 14:04:21 2011 -0600

     Update version for 1.0 release

     Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

I've verified this locally and on qemu.org.

md5sum of the tarball is:

a64b36067a191451323b0d34ebb44954  qemu-1.0.tar.gz

I've verified that the contents of this tarball match exactly the above commit 
after the following commands:

$ git submodule update --init
$ rm -rf .git roms/*/.git

> which is odd since e9b4 was
> only applied to the git tree _after_ a January 6th commit.  So either
> "git log" is being weird, or you guys retroactively re-released the 1.0
> tarball without renumbering it?
>
> I'm confused...

You somehow messed up your repository or have a bogus 1.0 tarball.  Please 
confirm the md5sum.

>
>> My latest QEMU works with MIPS Malta running from an NFS root.
>
> I can confirm that current -master works for me.
>
> The 1.0 tarball includes the fix, the 1.0 git tag doesn't.  I'll include
> a note to that effect in my release notes.

That's definitely not the case.

Regards,

Anthony Liguori	

> Thanks,
>
> Rob
>

Re: [Qemu-devel] Commit 5632ae46 broke the network on mips.

[Rob Landley]

On 01/22/2012 07:44 PM, Anthony Liguori wrote:
> On 01/22/2012 05:42 PM, Rob Landley wrote:
>> On 01/22/2012 01:06 AM, Stefan Weil wrote:
>>> This was fixed with commits e9b40fd34ceb23461083d505a444a389c094455b
>>> and 0b23c5d40ea933cfece3b4f69427f79c8a23256d in master and stable-1.0.
>>
>> The git tag for v1.0 does not include that commit.  Apparently the
>> current 1.0 tarball on the website does,
> 
> Hi Rob,
> 
> I've verified this three different ways..  You must have something
> messed up on your end.

Entirely possible.

> v1.0 tag is:
> 
> commit 1c8a881daaca6fe0646a425b0970fb3ad25f6732
> Author: Anthony Liguori <aliguori@us.ibm.com>
> Date:   Thu Dec 1 14:04:21 2011 -0600
> 
>     Update version for 1.0 release

Yup, that's the same here. Both of the above mentioned commits are
_after_ that tag.  (The clue is that there are _two_ of the same commit,
one applied to the 1.0-stable branch and one applied to the development
branch.  If it was _before_ the tag the same commit wouldn't have to be
applied twice with two different revision IDs.)

So when I check out the above commit, I don't get either of those fix
commits.  If I go "git log v1.0 | grep" for either of those commit
numbers, it doesn't find it in the log of what came before that commit.

I just did a fresh pull of git://git.qemu.org/qemu.git into a fresh
empty directory, and tested that: the result was the same.  On that
fresh checkout, if I do a git log of the 0b23 commit it shows the 1c8a
commit as its parent.

I tried this on my xubuntu 10.04 netbook and on my gentoo server: same
result.

I'm not a git expert, I just did this:

  mkdir temp
  cd temp
  git clone git://git.qemu.org/qemu.git
  cd qemu
  git log v1.0 | grep 0b23c5d40ea933cfece3b4f69427f79c8a23256d
  git log 0b23c5d40ea933cfece3b4f69427f79c8a23256d

It's possible I'm misunderstanding the results.  I do know that when I
"git checkout v1.0" and build against that, mips network doesn't work.
And when I go:

  git checkout v1.0
  vi hw/mips_malta.c

I can visually confirm that #include "exec-memory.h" is _not_ followed
by sysbus.h on the next line, which is the first change made by the fix
commit.

What am I doing wrong?

>     Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
> 
> I've verified this locally and on qemu.org.
> 
> md5sum of the tarball is:
> 
> a64b36067a191451323b0d34ebb44954  qemu-1.0.tar.gz

git archive v1.0 --prefix=qemu-1.0/ | gzip | md5sum
aa7443384602514c31c0e8ef2dab245e  -

Again, fresh checkout, 2 different machines running 2 different distros.

> I've verified that the contents of this tarball match exactly the above
> commit after the following commands:
> 
> $ git submodule update --init
> $ rm -rf .git roms/*/.git

What does "git submodule" do?

>> which is odd since e9b4 was
>> only applied to the git tree _after_ a January 6th commit.  So either
>> "git log" is being weird, or you guys retroactively re-released the 1.0
>> tarball without renumbering it?
>>
>> I'm confused...
> 
> You somehow messed up your repository or have a bogus 1.0 tarball. 
> Please confirm the md5sum.

The tarball works fine.  The v1.0 tag points to the same commit you
indicated.  The problem is that commit is the _parent_ of the fix I need.

The fix is included in the tarball.  The tarball is not the same as what
the v1.0 tag points to.

>>
>>> My latest QEMU works with MIPS Malta running from an NFS root.
>>
>> I can confirm that current -master works for me.
>>
>> The 1.0 tarball includes the fix, the 1.0 git tag doesn't.  I'll include
>> a note to that effect in my release notes.
> 
> That's definitely not the case.

I can reproduce it on two different machines.  What am I doing wrong?

Rob

Re: [Qemu-devel] Commit 5632ae46 broke the network on mips.

[Stefan Weil]

Hi Rob,

the patch was applied after v1.0, and there are two commits for
the stable-1.0 branch and the main development branch.
So I think that your git repository and your analysis is correct.

Where did you get your v1.0 tarball from? I just got this one from
qemu.org: http://wiki.qemu.org/download/qemu-1.0.tar.gz.
It does definitely not include my malta patch, so it is correct, too.

Any tarball with the Malta patch is not a v1.0 tarball.

Regards,
Stefan

Re: [Qemu-devel] Commit 5632ae46 broke the network on mips.

[Rob Landley]

On 01/23/2012 12:34 AM, Stefan Weil wrote:
> Hi Rob,
> 
> the patch was applied after v1.0, and there are two commits for
> the stable-1.0 branch and the main development branch.
> So I think that your git repository and your analysis is correct.
> 
> Where did you get your v1.0 tarball from?

Sigh, I know what happened.  I cd'd to the 1.0 tarball directory but
used command history to re-run the test, and didn't edit the big
$PATH=blah assignment at the start of the line, so that was still an
absolute path pointing to the git directory.

My bad, tested the wrong version.  (I've had a lot of that tracking down
this bug, the command line to make it work is very, very long.)

Rob

Re: [Qemu-devel] Commit 5632ae46 broke the network on mips.

[Anthony Liguori]

On 01/22/2012 05:42 PM, Rob Landley wrote:
> On 01/22/2012 01:06 AM, Stefan Weil wrote:
>> Am 21.01.2012 20:04, schrieb Rob Landley:
>>> commit 5632ae46: "mips_malta: move i8259 initialization after piix4
>>> initialization" broke the network on mips.  It still comes up, but
>>> doesn't pass packets.
>>>
>>> Try this:
>>>
>>>     wget http://landley.net/aboriginal/bin/system-image-mips.tar.bz2
>>>     tar xvjf system-image-mips.tar.bz2
>>>     cd system-image-mips
>>>     ./run-emulator.sh
>>>
>>>     [ wait for shell prompt ]
>>>
>>>     wget http://google.com
>>>
>>> Under 0.15.1 it works, under 1.0 it doesn't.  Identical binary.  I
>>> bisected it to the commit in question, it's still broken in -master.
>>>
>>> Rob
>>
>> This was fixed with commits e9b40fd34ceb23461083d505a444a389c094455b
>> and 0b23c5d40ea933cfece3b4f69427f79c8a23256d in master and stable-1.0.
>
> The git tag for v1.0 does not include that commit.  Apparently the
> current 1.0 tarball on the website does, which is odd since e9b4 was
> only applied to the git tree _after_ a January 6th commit.  So either
> "git log" is being weird, or you guys retroactively re-released the 1.0
> tarball without renumbering it?
>
> I'm confused...
>
>> My latest QEMU works with MIPS Malta running from an NFS root.
>
> I can confirm that current -master works for me.
>
> The 1.0 tarball includes the fix, the 1.0 git tag doesn't.  I'll include
> a note to that effect in my release notes.

BTW, since you tag doesn't have this changeset (and it should), it's your tag 
that's messed up.

Fortunately, for the 1.0 release, I started using a signed tag.

http://git.qemu.org/?p=qemu.git;a=tag;h=5e5e6d6c16e1742996d93e8f3e946cf544d0b146

So this should help you figure out where your bogus tag came from.

Regards,

Anthony Liguori

>
> Thanks,
>
> Rob
>