Re: Linux 5.17-rc5

[Guenter Roeck <linux@roeck-us.net>]

On 2/21/22 09:44, Linus Torvalds wrote:
> [ Fixed the subject line - you answered the wrong email.]
> 
> On Sun, Feb 20, 2022 at 6:47 PM Guenter Roeck <linux@roeck-us.net> wrote:
>>
>> Build results:
>>          total: 155 pass: 152 fail: 3
> 
> Grr. Things are going in the wrong direction here.
> 
> At least the two new cases look trivial, and that ppc fix is finally
> in linux-next.
> 
>> Qemu test results:
>>          total: 488 pass: 427 fail: 61
>> Failed tests:
>>          arm:orangepi-pc:multi_v7_defconfig:usb1:net,nic:sun8i-h3-orangepi-pc:rootfs
> 
> Adding some devicetree and otherwise involved people to the cc, can we
> *finally* get this fixed? It's been around for all of this release,
> and Guenter has done a fair amount of bisecting and testing, chasing
> down lockdep reports etc.
> 
> I've been ignoring this on the assumption that it will just get fixed,
> but it's starting to be pretty late in the game now..
> 
>>          <almost all mips tests>
> 
> Gaah. And what happened here? There aren't even any MIPS changes in
> rc5, so clearly something went sideways that just must have happened
> to affect your mips qemu tests. Added Thomas Bogendoerfer despite the
> lack of commits.
> 

Oh, that is "just" the result of the netfilter build problem.

Building mips:malta_defconfig ... failed
--------------
Error log:
net/netfilter/xt_socket.c: In function 'socket_mt_destroy':
net/netfilter/xt_socket.c:224:17: error: implicit declaration of function 'nf_defrag_ipv6_disable'; did you mean 'nf_defrag_ipv4_disable'?

This was introduced with commit 75063c9294fb ("netfilter: xt_socket:
fix a typo in socket_mt_destroy()"). The build error is seen whenever
NETFILTER_XT_MATCH_SOCKET is enabled and IPV6 is disabled, and mips
builds are the unfortunate victim. The patch introducing the problem
now proliferated into the 5.15 and 5.16 stable release candidates
as well, causing the same build and test failures there.

The real question is why to patch introducing the problem made it
upstream even though the problem was reported several days before
the push request was sent.

Guenter