[meta-intel] [PATCH 2/4 dizzy] meta-crystalforest: Intel QAT v1.5 recipe under recipes-extended

Ong, Boon Leong boon.leong.ong at intel.com
Mon Jun 22 18:23:10 PDT 2015


>Hi,
>
>When I build this with dizzy I get a build failure that:
>
>> | ../../common/cpa_sample_utils.c
>> | cpa_dc_stateless_multi_op_checksum_sample.c
>> | ../../../performance/qae/linux/user_space/qae_mem_utils.c
>> | ../stateless_sample/cpa_dc_sample_user.c
>> | /srv/ssd/builds/world/tmp/work/crystalforest-poky-linux/qat/2.2.0-30
>> | -r0/QAT1.5/build/libicp_qa_al_s.so -L/usr/Lib -lpthread -lcrypto -o
>> | dc_stateless_multi_op_sample
>> | /usr/bin/ld: cannot find -lcrypto
>> | collect2: error: ld returned 1 exit status
>> | Makefile:111: recipe for target 'default' failed
>
>This sysroot does contain a /lib/libcrypto.so, and you can see the -L/usr/lib, but
>no -L/lib, Did you do a compile check?

Interesting. Perhaps, it is due to sysroot has the libcrypto from earlier build.
We did bitbake qat -f -c cleanall before bitbake qat.  Will check back again.

>
>
>Also, I am confused with this version-ing of 1.5 and 1.6, but you set PV
>in those recipes to 2.2.0-30, which could cause problems with upgrade path.
>
>Have you tested building qat 1.6 version?  I think it will have the same
>crypto problem.
>
>I thing a better thing would be to name them qat15_2.2.0-30 and
>qat16_2.2.0-30, that way the package names are distinct. You can still
>control which QAT is installed in an image using a virtual/qat set up
>similar to the kernel.

Well, this is definitely a clean-up needed on QAT source package.
The tarball package is as follow

qatmux.l.2.2.0-30.tar_.gz
which untar into two folders:
QAT1.5/QAT1.5.L.1.7.0-30.tar.gz
QAT1.6/QAT1.6.L.2.2.0-30.tar.gz

I think that to be more specific, we can have the following:
qat15_1.7.0-30.bb
qat16_2.2.0-30.bb

Then as suggested by Saul, use the following 

PREFERRED_PROVIDER_virtual/qat ?= "qat15"
PREFERRED_VERSION_qat15 ?= "1.7.0-%"

PREFERRED_PROVIDER_virtual/qat ?= "qat16"
PREFERRED_VERSION_qat16 ?= "2.2.0-%"




More information about the meta-intel mailing list