[OE-core] [PATCH] nss: 3.45 -> 3.47.1

[Andreas Müller]

On Fri, Dec 13, 2019 at 2:22 PM Alexander Kanavin
<alex.kanavin at gmail.com> wrote:
>
> On Wed, 4 Dec 2019 at 19:00, Richard Purdie <richard.purdie at linuxfoundation.org> wrote:
>>
>> On Wed, 2019-12-04 at 07:09 -0800, Wang Mingyu wrote:
>> > Signed-off-by: Wang Mingyu <wangmy at cn.fujitsu.com>
>> > ---
>> >  meta/recipes-support/nss/{nss_3.45.bb => nss_3.47.1.bb} | 4 ++--
>> >  1 file changed, 2 insertions(+), 2 deletions(-)
>> >  rename meta/recipes-support/nss/{nss_3.45.bb => nss_3.47.1.bb} (98%)
>>
>> Fails on beaglebone:
>>
>> https://autobuilder.yoctoproject.org/typhoon/#/builders/65/builds/1327
>
>
> I had a look into this: the issue is that new versions of nss require hardware crypto capabilities on ARM (both 32 and 64 bit).
Do you have an link for this?
> This means we have to disable the recipe when that is not the case; a quick fix would be:
>
> COMPATIBLE_HOST_armv7a = 'null'
>
> but I would like to hear other ideas too :)
>
I think we should not do that because:

Build log says:
| cc1: error: switch '-mcpu=cortex-a8' conflicts with '-march=armv8-a'
switch [-Werror]
| cc1: all warnings being treated as errors

Nss sets '-march=armv8-a'. Checked in their sources and found that
this is done in lib/freebl source path [1]. With latests sources from
gihub-shadow they introduced 'disable_arm_hw_aes' [1]. Have no idea if
3.47.1 already supports that but we should use it as packageconfig.

I think open questions are still:
* will aarch64 accept their flags?
* isn't something wrong in their logic? They build armv8_c_lib based
on conditions like ''disable_arm_hw_aes==0 and (target_arch=="arm" or
target_arch=="arm64" or target_arch=="aarch64")'' e.g [2] Hmm arm???

[1] https://github.com/nss-dev/nss/blob/4703a817f8b90cc3423620e5c7606bdbcedbcf2e/lib/freebl/freebl.gyp#L170
[2] https://github.com/nss-dev/nss/blob/4703a817f8b90cc3423620e5c7606bdbcedbcf2e/lib/freebl/freebl.gyp#L226

Andreas