[dpdk-dev] Compilation failure when disabling Cavium OCTEONTX network PMD driver

[dpdk-dev] Compilation failure when disabling Cavium OCTEONTX network PMD driver

[Daniel Pharos]

Hi,

If I disable the Cavium OCTEONTX network PMD driver in the config file (so: "CONFIG_RTE_LIBRTE_OCTEONTX_PMD=n"), compilation of DPDK 19.05 fails with:
/home/ubuntu/Downloads/dpdk-19.05/build/lib/librte_pmd_octeontx_ssovf.a(ssovf_worker.o): In function `ssows_flush_events':
ssovf_worker.c:(.text+0x102): undefined reference to `rte_octeontx_pchan_map'
/home/ubuntu/Downloads/dpdk-19.05/build/lib/librte_pmd_octeontx_ssovf.a(ssovf_worker.o): In function `ssows_deq':
ssovf_worker.c:(.text.hot+0x138): undefined reference to `rte_octeontx_pchan_map'
/home/ubuntu/Downloads/dpdk-19.05/build/lib/librte_pmd_octeontx_ssovf.a(ssovf_worker.o): In function `ssows_deq_timeout':
ssovf_worker.c:(.text.hot+0x301): undefined reference to `rte_octeontx_pchan_map'
/home/ubuntu/Downloads/dpdk-19.05/build/lib/librte_pmd_octeontx_ssovf.a(ssovf_worker.o): In function `ssows_deq_burst':
ssovf_worker.c:(.text.hot+0x458): undefined reference to `rte_octeontx_pchan_map'
/home/ubuntu/Downloads/dpdk-19.05/build/lib/librte_pmd_octeontx_ssovf.a(ssovf_worker.o): In function `ssows_deq_timeout_burst':
ssovf_worker.c:(.text.hot+0x61c): undefined reference to `rte_octeontx_pchan_map'
collect2: error: ld returned 1 exit status
/home/ubuntu/Downloads/dpdk-19.05/mk/rte.app.mk:404: recipe for target 'test' failed
make[3]: *** [test] Error 1
/home/ubuntu/Downloads/dpdk-19.05/mk/rte.subdir.mk:35: recipe for target 'test' failed
make[2]: *** [test] Error 2
/home/ubuntu/Downloads/dpdk-19.05/mk/rte.sdkbuild.mk:46: recipe for target 'app' failed
make[1]: *** [app] Error 2
/home/ubuntu/Downloads/dpdk-19.05/mk/rte.sdkroot.mk:98: recipe for target 'all' failed
make: *** [all] Error 2

This is on x86_64, Ubuntu 18.04 (up-to-date), using the DPDK 19.05 tarball, with only the mentioned config-line changed.


Kind regards,
DanielPharos

Re: [dpdk-dev] Compilation failure when disabling Cavium OCTEONTX network PMD driver

[Jerin Jacob Kollanukkaran]

> -----Original Message-----
> From: dev <dev-bounces@dpdk.org> On Behalf Of Daniel Pharos 
> Sent: Saturday, July 6, 2019 8:41 PM
> To: dev@dpdk.org
> Subject: [dpdk-dev] Compilation failure when disabling Cavium OCTEONTX
> network PMD driver
> 
> Hi,
> 
> If I disable the Cavium OCTEONTX network PMD driver in the config file (so:
> "CONFIG_RTE_LIBRTE_OCTEONTX_PMD=n"), compilation of DPDK 19.05 fails
> with:

CONFIG_RTE_LIBRTE_PMD_OCTEONTX_SSOVF is depended on 
CONFIG_RTE_LIBRTE_OCTEONTX_PMD.

Please disable CONFIG_RTE_LIBRTE_PMD_OCTEONTX_SSOVF as well.

Re: [dpdk-dev] Compilation failure when disabling Cavium OCTEONTX network PMD driver

[Daniel Pharos]

Huh, looks like I've been hitting "Reply" instead of the required "Reply all", and the mailing list got dropped from the CC. Below is our conversation so far (might be incomplete).


Kind regards,
DanielPharos

From: Daniel Pharos
Sent: 08 July 2019 22:50
To: Jerin Jacob Kollanukkaran <jerinj@marvell.com>
Subject: RE: [dpdk-dev] Compilation failure when disabling Cavium OCTEONTX network PMD driver

> Looks at the decencies, You cannot document this as it is not scale to document this for every driver.
> LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring -lrte_ethdev -lrte_net -lrte_kvargs -lrte_common_octeontx -lrte_eventdev -lrte_bus_pci -lrte_bus_vdev
If it's too complex to even document, that's basically an admission one can't do it by hand either. All the more reason to have some automated script do the work.

> I know you cannot create nice patch with that approach.
I've seen configure+makefiles that do just that. It's a shame the way you've chosen to organize things cannot support this.

> use meson build if care about build time. It is super fast.
If your compile-process is spending more time in the makefile than actually compiling, you are doing it wrong. But hey, maybe you are right. Please tell me how to have meson build only the drivers that I need, with automatic dependency-resolving.


Kind regards,
DanielPharos


From: Jerin Jacob Kollanukkaran <jerinj@marvell.com<mailto:jerinj@marvell.com>>
Sent: 08 July 2019 21:17
To: Daniel Pharos <danielpharos@hotmail.com<mailto:danielpharos@hotmail.com>>
Subject: RE: [dpdk-dev] Compilation failure when disabling Cavium OCTEONTX network PMD driver



From: Daniel Pharos <danielpharos@hotmail.com<mailto:danielpharos@hotmail.com>>
Sent: Monday, July 8, 2019 11:17 PM
To: Jerin Jacob Kollanukkaran <jerinj@marvell.com<mailto:jerinj@marvell.com>>
Subject: [EXT] RE: [dpdk-dev] Compilation failure when disabling Cavium OCTEONTX network PMD driver

External Email
________________________________
> What is the difference between comment and expressing in makefile. It is already depected in makefile.
When toggling the config options, I'm already in the config file. I shouldn't have to open the makefile as well to manually read all the various LDLIBS lines, figure out which configuration option belongs to the library name it references, then go back to the config file to modify it, and then repeat the process for its dependencies as well.

[Jerin] Looks at the decencies, You cannot document this as it is not scale to document this for every driver.
LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring -lrte_ethdev -lrte_net -lrte_kvargs -lrte_common_octeontx -lrte_eventdev -lrte_bus_pci -lrte_bus_vdev

> Send a patch if you think there is change required.
Since this would be something that would need to be enforced, it would need broad support to be maintained, something that appears to be lacking at the moment, even though we've just demonstrated its added value. So I'm not going to invest my time to make a nice patch; I'll just hack something together privately.

[Jerin] I know you cannot create nice patch with that approach.

> BTW, Any reason for disabling the config in first place, Is it to reduce the space?
It's mostly to reduce compile time. Why spend lots of CPU cycles (re-)compiling drivers and libraries I can't even use, because I don't have the hardware? Also, I'm of the "only compile what you need, the rest is a (security?) incident waiting to happen"-school. And finally, when building high-performance systems, I'd like to reduce the amount of bloatware it contains (yes, even if it doesn't get loaded). So personally, I'd ask the question the other way around: why are you (not you personally!) making me compile stuff I'm never going to need or use?

[Jerin] use meson build if care about build time. It is super fast.


Kind regards,
DanielPharos

From: Jerin Jacob Kollanukkaran <jerinj@marvell.com<mailto:jerinj@marvell.com>>
Sent: 07 July 2019 19:18
To: Daniel Pharos <danielpharos@hotmail.com<mailto:danielpharos@hotmail.com>>
Subject: Re: [dpdk-dev] Compilation failure when disabling Cavium OCTEONTX network PMD driver

What is the difference between comment and expressing in makefile. It is already depected in makefile. Send a patch if you think there is change required. BTW, Any reason for disabling the config in first place, Is it to reduce the space?
________________________________
From: Daniel Pharos <danielpharos@hotmail.com<mailto:danielpharos@hotmail.com>>
Sent: Sunday, July 7, 2019 10:35:34 PM
To: Jerin Jacob Kollanukkaran
Subject: [EXT] RE: [dpdk-dev] Compilation failure when disabling Cavium OCTEONTX network PMD driver

External Email

----------------------------------------------------------------------
> Like linux kernel Kconfig we don't have means to express the dependency In Kconfig.
Yes you do. You can add a comment explaining the dependencies, so I don't have to go through multiple "compile -> fail -> guess which option I need to turn off -> retry" cycles. Heck, the very fact that your first answer got it wrong (as in: it was incomplete) strongly suggests this needs to be documented in the one place where people messing with the options will see it: as a comment line right next to the relevant option.


Kind regards,
DanielPharos


-----Original Message-----
From: Jerin Jacob Kollanukkaran <jerinj@marvell.com<mailto:jerinj@marvell.com>>
Sent: 07 July 2019 17:00
To: Daniel Pharos <danielpharos@hotmail.com<mailto:danielpharos@hotmail.com>>
Subject: RE: [dpdk-dev] Compilation failure when disabling Cavium OCTEONTX network PMD driver

> -----Original Message-----
> From: Daniel Pharos <danielpharos@hotmail.com<mailto:danielpharos@hotmail.com>>
> Sent: Sunday, July 7, 2019 8:03 PM
> To: Jerin Jacob Kollanukkaran <jerinj@marvell.com<mailto:jerinj@marvell.com>>
> Subject: [EXT] RE: [dpdk-dev] Compilation failure when disabling
> Cavium OCTEONTX network PMD driver
>
> External Email
>
> ----------------------------------------------------------------------
> Hi,
>
> It would be nice if the config/make file complained about such
> incompatible settings. Also, that still doesn't work:

The makefile already has dependency[1] for the specific driver.
Like linux kernel Kconfig we don't have means to express the dependency In Kconfig.

[1]
drivers/net/octeontx/Makefile
LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring LDLIBS += -lrte_ethdev -lrte_net -lrte_kvargs -lrte_common_octeontx LDLIBS += -lrte_mempool_octeontx LDLIBS += -lrte_eventdev LDLIBS += -lrte_bus_pci LDLIBS += -lrte_bus_vdev






It would be nice if the config/make file complained about such incompatible settings. Also, that still doesn't work:



/home/ubuntu/Downloads/dpdk-19.05/build/lib/librte_mempool_octeontx.a(octeontx_fpavf.o): In function `octeontx_fpapf_pool_setup':

octeontx_fpavf.c:(.text+0x318): undefined reference to `octeontx_mbox_send'

/home/ubuntu/Downloads/dpdk-19.05/build/lib/librte_mempool_octeontx.a(octeontx_fpavf.o): In function `octeontx_fpapf_pool_destroy':

octeontx_fpavf.c:(.text+0x465): undefined reference to `octeontx_mbox_send'

/home/ubuntu/Downloads/dpdk-19.05/build/lib/librte_mempool_octeontx.a(octeontx_fpavf.o): In function `octeontx_fpapf_aura_attach':

octeontx_fpavf.c:(.text+0x538): undefined reference to `octeontx_mbox_send'

/home/ubuntu/Downloads/dpdk-19.05/build/lib/librte_mempool_octeontx.a(octeontx_fpavf.o): In function `octeontx_fpapf_aura_detach':

octeontx_fpavf.c:(.text+0x681): undefined reference to `octeontx_mbox_send'

/home/ubuntu/Downloads/dpdk-19.05/build/lib/librte_mempool_octeontx.a(octeontx_fpavf.o): In function `octeontx_fpapf_start_count':

octeontx_fpavf.c:(.text+0x7f8): undefined reference to `octeontx_mbox_send'

collect2: error: ld returned 1 exit status

/home/ubuntu/Downloads/dpdk-19.05/mk/rte.app.mk:404: recipe for target 'test' failed

make[3]: *** [test] Error 1

/home/ubuntu/Downloads/dpdk-19.05/mk/rte.subdir.mk:35: recipe for target 'test' failed

make[2]: *** [test] Error 2

/home/ubuntu/Downloads/dpdk-19.05/mk/rte.sdkbuild.mk:46: recipe for target 'app' failed

make[1]: *** [app] Error 2

/home/ubuntu/Downloads/dpdk-19.05/mk/rte.sdkroot.mk:98: recipe for target 'all' failed

make: *** [all] Error 2



(And yes, I did start with a clean extract from the tarball.)



I suspect there are more hidden dependencies than just that one. This compilation failure seems to suggest CONFIG_RTE_LIBRTE_OCTEONTX_MEMPOOL is also involved...





Kind regards,

DanielPharos