On 5/3/22 21:20, Gregory Rose wrote:
>
>
> On 4/26/2022 9:35 AM, Gregory Rose wrote:
>>
>>
>> On 4/22/2022 8:34 AM, David Marchand wrote:
>>> On Fri, Apr 22, 2022 at 9:24 AM Dumitru Ceara <[email protected]> wrote:
>>>>> @@ -37,14 +30,6 @@ jobs:
>>>>> - compiler: clang
>>>>> opts: --disable-ssl
>>>>>
>>>>> - - compiler: gcc
>>>>> - testsuite: test
>>>>> - kernel: 3.16
>>>>> - - compiler: clang
>>>>> - testsuite: test
>>>>> - kernel: 3.16
>>>>> - asan: asan
>>>>> -
>>>>
>>>> Maybe it's enough to just remove the 'kernel' argument here? It's
>>>> useful to have address sanitizer enabled test jobs too. Arguably, when
>>>> we added it we should've maybe not link it to a given kernel but, in any
>>>> case, I think we shouldn't remove it completely now.
>>>>
>>>> What do you think of keeping it as:
>>>>
>>>> - compiler: clang
>>>> testsuite: test
>>>> asan: asan
>>>
>>> +1
>>>
>>>>
>>>> I might be wrong but it seems to me we need something similar for afxdp
>>>> too.
>>>
>>> The Ubuntu 18.04 kernel does not have AF_XDP support iirc, so either
>>> we keep on building a 5.3 kernel and compile ovs with afxdp against
>>> this kernel.
>>> Or we may switch to Ubuntu 20.04 in GHA.
>>>
>>>
>>
>> Hi David and Dimitru,
>>
>> thanks for the input - I'll see if I can get that working.
>>
>> - Greg
>
> A quick update - I was able to reinstate the address sanitizer test by
> dropping the kernel argument but no luck with afxdp. Perhaps we could
> limit afxdp support to branch 2.16 or earlier until we get a mainstream
> Ubuntu distro that supports it working.
>
> My biggest problem to date is fixing up the debian packaging and build
> to not include the dkms driver package. They're pretty tightly coupled
> and I'm no expert in the debian rules for packaging. However, I am
> making some progress - one of the main issues is that every time I make
> a mistake I have to reclone the repo because debian seems to scatter
> little traps around the code that are not cleaned out when you run
> 'fakeroot debian/rules clean'.
>
> Stay tuned and I'll send a new series of patches after I get this debian
> packaging figured out.
Thanks for working on this!
Sorry, I was out for some time and just dug myself out of a pile of emails.
Unfortunately, I'm not familiar with debian packaging code either, so can't
really help much, but will take a look at the new version of the set once
it is available.
Small nit for the current series: 2.17 was released in February and
the module will stay there, so the removal will be part of 2.18, not
2.17.
For the afxdp, I think, we can keep the install_kernel() function, but
only remove a few things from it that are not needed for afxdp, e.g.
diff --git a/.ci/linux-build.sh b/.ci/linux-build.sh
index 6cd38ff3e..6ef629447 100755
--- a/.ci/linux-build.sh
+++ b/.ci/linux-build.sh
@@ -69,13 +69,6 @@ function install_kernel()
sed -i 's/CONFIG_STACK_VALIDATION=y/CONFIG_STACK_VALIDATION=n/' .config
make oldconfig
- # Older kernels do not include openvswitch
- if [ -d "net/openvswitch" ]; then
- make net/openvswitch/
- else
- make net/bridge/
- fi
-
if [ "$AFXDP" ]; then
sudo make headers_install INSTALL_HDR_PATH=/usr
pushd tools/lib/bpf/
@@ -91,9 +84,6 @@ function install_kernel()
sudo sed -i '/^# define __always_inline .*/i # undef __always_inline' \
/usr/include/x86_64-linux-gnu/sys/cdefs.h || true
EXTRA_OPTS="${EXTRA_OPTS} --enable-afxdp"
- else
- EXTRA_OPTS="${EXTRA_OPTS} --with-linux=$(pwd)"
- echo "Installed kernel source in $(pwd)"
fi
popd
}
---
Maybe some other parts also can be removed from the function.
This way we can keep the 'kernel: 5.3' for afxdp jobs, but remove
kernels from all other jobs. 'if [ "$AFXDP" ]' check can also be
moved outside of the function to the place where install_kernel()
is called.
Best regards, Ilya Maximets.
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev