Re: 19.11.11 patches review and test

[Ferruh Yigit <ferruh.yigit@intel.com>]

On 12/14/2021 2:46 PM, Christian Ehrhardt wrote:
> On Tue, Dec 14, 2021 at 2:58 PM Christian Ehrhardt
> <christian.ehrhardt@canonical.com> wrote:
>>
>> On Tue, Dec 14, 2021 at 2:10 PM Ferruh Yigit <ferruh.yigit@intel.com> wrote:
>>>
>>> On 12/14/2021 11:39 AM, Christian Ehrhardt wrote:
>>>> On Tue, Dec 14, 2021 at 11:13 AM Ferruh Yigit <ferruh.yigit@intel.com> wrote:
>>>>>
>>>>> On 12/14/2021 7:44 AM, Christian Ehrhardt wrote:
>>>>>> On Tue, Dec 14, 2021 at 6:49 AM Kalesh Anakkur Purayil
>>>>>> <kalesh-anakkur.purayil@broadcom.com> wrote:
>>>>>>
>>>>>> [snip]
>>>>>>
>>>>>>>>> [Kalesh] Yes, i am seeing the same error. I used make command to build dpdk, not meson.
>>>>>>>>> The back ported commit you mentioned takes care of meson build only I think.
>>>>>>>>>
>>>>>>>>
>>>>>>>> I see, make build is failing, and yes the fix is only for the meson.
>>>>>>>> I will check the make build and will send a fix for it.
>>>>>>>
>>>>>>> [Kalesh]: looks like the below changes fixes the issue. I tried only on SLES15 SP3 and not on other SLES flavors.
>>>>>>>
>>>>>>> diff --git a/kernel/linux/kni/Makefile b/kernel/linux/kni/Makefile
>>>>>>> index 595bac2..bf0efab 100644
>>>>>>> --- a/kernel/linux/kni/Makefile
>>>>>>> +++ b/kernel/linux/kni/Makefile
>>>>>>> @@ -16,6 +16,16 @@ MODULE_CFLAGS += -I$(RTE_OUTPUT)/include
>>>>>>>     MODULE_CFLAGS += -include $(RTE_OUTPUT)/include/rte_config.h
>>>>>>>     MODULE_CFLAGS += -Wall -Werror
>>>>>>>
>>>>>>> +#
>>>>>>> +# Use explicit 'source' folder for header path. In SUSE 'source' is not linked to 'build' folder.
>>>>>>> +#
>>>>>>> +ifdef CONFIG_SUSE_KERNEL
>>>>>>> +   KSRC = /lib/modules/$(shell uname -r)/source
>>>>>>> +   ifneq ($(shell grep -A 1 "ndo_tx_timeout" $(KSRC)/include/linux/netdevice.h | grep -o txqueue),)
>>>>>>> +      MODULE_CFLAGS += -DHAVE_TX_TIMEOUT_TXQUEUE
>>>>>>> +   endif
>>>>>>> +endif
>>>>>>
>>>>>> Back in the day we tried various "is Suse and kernel version x.y"
>>>>>> approaches, but they failed as there was no clear version throughout
>>>>>> all of the Suse streams (leap, tumbleweed, sles) that worked well for
>>>>>> all.
>>>>>> This change here follows the upstream approach of "just check if it is there".
>>>>>>
>>>>>> I've applied this to 19.11 and did test builds across various distributions:
>>>>>> 1. no non-suse build changed
>>>>>> 2. suse builds stayed as-is or improved
>>>>>> Formerly failing:
>>>>>>       openSUSE_Factory_ARM aarch64
>>>>>>       SLE_15  x86_64 -> now working
>>>>>>       openSUSE_Leap_15.3 x86_64 -> now working
>>>>>>       openSUSE_Tumbleweed  x86_64 -> still failing
>>>>>> Formerly working:
>>>>>>       SLE_12_SP4 x86_64 ppc64le -> still fine
>>>>>>       openSUSE_Factory_ARM armv7l  -> still fine
>>>>>>       openSUSE_Leap_15.2 x86_64  -> still fine
>>>>>>
>>>>>
>>>>> Thanks Kalesh for the fix, and thanks Christian for testing.
>>>>>
>>>>> I was expecting this approach will fix all builds, after patch only
>>>>> 'openSUSE_Tumbleweed' is failing, right? I will check it.
>>>>
>>>> As just discussed on IRC, yes and the log for that is at
>>>> https://build.opensuse.org/package/live_build_log/home:cpaelzer:branches:home:bluca:dpdk/dpdk-19.11/openSUSE_Tumbleweed/x86_64
>>>>
>>>> It also is affected by an issue around  -Werror=implicit-fallthrough,
>>>> so even with KNI fixed it likely is going to fail.
>>>>
>>>>> And I think you need the fix as a patch anyway, @Kalesh are you
>>>>> planning to send the patch?
>>>>
>>>> I don't need it, as I have already grabbed and preliminary added it:
>>>> https://github.com/cpaelzer/dpdk-stable-queue/commit/d43fa3e198c08a3a76d70f4565b31ad3ab5f29c4
>>>>
>>>> But surely, once/If you come up with a full patch that also includes
>>>> tumbleweed I can replace it with yours.
>>>>
>>>
>>> 'tumbleweed' error is odd, it complains about macro being redefined,
>>> not sure why only in this platform we are getting an error.
>>>
>>> Macro is only defined in one place, but indeed header file included
>>> multiple times, one direct and one indirect, so macro defined multiple
>>> times but without value, so it should be OK and it is OK for other
>>> platforms, it is defined as:
>>> #define HAVE_TX_TIMEOUT_TXQUEUE
>>>
>>> Another option is that macro is defined in some other header file,
>>> although I think that is very unlikely, can you please test with
>>> below change to rule out that option:
>>
>> I'm testing that and will let you know in a bit ...
> 
> Hi Ferruh,
> with your change the build now works.
> So indeed the symbol might have been defined elsewhere.
> 

Interesting, this is self note to prefix 'RTE_' future macros.

> https://build.opensuse.org/package/live_build_log/home:cpaelzer:branches:home:bluca:dpdk/dpdk-19.11/openSUSE_Tumbleweed/x86_64
> 
> It still fails later with the "-Werror=implicit-fallthrough=" but that
> is a different problem
> => https://bugs.dpdk.org/show_bug.cgi?id=907
> 

Yep, this is igb_uio error, I assigned the bug to myself and will look at it.

> Ferruh - are you ok if I merge your suggestion with the backport I got
> from Kalesh?
> 

Sure.
But would you mind sending the final patch to the stable mail list as record?
Or I can do the same if you prefer?

>>> diff --git a/kernel/linux/kni/compat.h b/kernel/linux/kni/compat.h
>>> index 664785674ff1..71846419f437 100644
>>> --- a/kernel/linux/kni/compat.h
>>> +++ b/kernel/linux/kni/compat.h
>>> @@ -135,7 +135,7 @@
>>>           (defined(RHEL_RELEASE_CODE) && \
>>>            RHEL_RELEASE_VERSION(8, 3) <= RHEL_RELEASE_CODE) || \
>>>            (defined(CONFIG_SUSE_KERNEL) && defined(HAVE_ARG_TX_QUEUE))
>>> -#define HAVE_TX_TIMEOUT_TXQUEUE
>>> +#define RTE_HAVE_TX_TIMEOUT_TXQUEUE
>>>    #endif
>>>
>>>    #if KERNEL_VERSION(5, 9, 0) > LINUX_VERSION_CODE
>>> diff --git a/kernel/linux/kni/kni_net.c b/kernel/linux/kni/kni_net.c
>>> index c8bad5f197ca..7397de4659b2 100644
>>> --- a/kernel/linux/kni/kni_net.c
>>> +++ b/kernel/linux/kni/kni_net.c
>>> @@ -623,7 +623,7 @@ kni_net_rx(struct kni_dev *kni)
>>>    /*
>>>     * Deal with a transmit timeout.
>>>     */
>>> -#ifdef HAVE_TX_TIMEOUT_TXQUEUE
>>> +#ifdef RTE_HAVE_TX_TIMEOUT_TXQUEUE
>>>    static void
>>>    kni_net_tx_timeout(struct net_device *dev, unsigned int txqueue)
>>>    #else
>>>
>>>
>>>>>> Past fixes always "inverted" the result, by fixing some but breaking others.
>>>>>> This new patch works in "not breaking any formerly working build" but
>>>>>> at the same time fixing a few builds.
>>>>>> Therefore -> applied & thanks!
>>>>>>
>>>>>> I'll likely tag -rc2 before the end of the week.
>>>>>> The good thing is that (so far) we have:
>>>>>> 1. a non functional change
>>>>>> 2. a change fixing clang-13 builds (TBH only one of many needed clang13 issues)
>>>>>> 3. a change fixing sles15SP3 builds
>>>>>>
>>>>>> Due to those, no current ongoing tests will have to be restarted.
>>>>>> Whoever was able to build, can continue the current tests.
>>>>>> Whoever was blocked by SLES15SP3 or clang-13 had no tests other than a
>>>>>> failing build and can work with -rc2 then.
>>>>>> I'll explain the same in the mail about -rc2.
>>>>>>
>>>>>>>     -include /etc/lsb-release
>>>>>>>
>>>>>>>     ifeq ($(DISTRIB_ID),Ubuntu)
>>>>>>>
>>>>>>> Regards,
>>>>>>> Kalesh
>>>>>>
>>>>>> [snip]
>>>>>>
>>>>>
>>>>
>>>>
>>>
>>
>>
>> --
>> Christian Ehrhardt
>> Staff Engineer, Ubuntu Server
>> Canonical Ltd
> 
> 
>