diff mbox

[dpdk-dev,2/2] mk: reduce scope of whole-archive to pmd libraries

Message ID 1464367686-3475-2-git-send-email-ferruh.yigit@intel.com (mailing list archive)
State Superseded, archived
Delegated to: Thomas Monjalon
Headers show

Commit Message

Ferruh Yigit May 27, 2016, 4:48 p.m. UTC
--whole-archive argument only required for pmd libraries, and currently
it covers more libraries. Reducing scope of the argument to pmd
libraries slightly reduce final application size.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
---
 mk/rte.app.mk | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

Comments

Thomas Monjalon June 10, 2016, 9:03 a.m. UTC | #1
2016-05-27 17:48, Ferruh Yigit:
> --whole-archive argument only required for pmd libraries, and currently
> it covers more libraries. Reducing scope of the argument to pmd
> libraries slightly reduce final application size.

In my understanding, --whole-archive is required for static libraries used
by plugins:
	http://dpdk.org/commit/20afd76a

If we want to restrict it, I would say it must be around EAL, ring,
mbuf, mempool, ethdev, cryptodev, etc.
Ferruh Yigit June 10, 2016, 9:57 a.m. UTC | #2
On 6/10/2016 10:03 AM, Thomas Monjalon wrote:
> 2016-05-27 17:48, Ferruh Yigit:
>> --whole-archive argument only required for pmd libraries, and currently
>> it covers more libraries. Reducing scope of the argument to pmd
>> libraries slightly reduce final application size.
> 
> In my understanding, --whole-archive is required for static libraries used
> by plugins:
> 	http://dpdk.org/commit/20afd76a
Right, required for static libraries. But more than used by plugins,
required for plugin to work.

> 
> If we want to restrict it, I would say it must be around EAL, ring,
> mbuf, mempool, ethdev, cryptodev, etc.
> 

We should restrict to plugins. What happens is, since there is no direct
call to PMDs from application, PMD code is not included into final
application. To force PMDs to be included into final binary,
--whole-archive is required.

But --whole-archive is not required for other libraries, and just cause
unnecessary increase in binary size.

Sample logs, with same setup, [1] compiled without --whole-archive, [2]
is current code:

[1]:
# ./testpmd -- -i
EAL: Detected 32 lcore(s)
EAL: Probing VFIO support...
EAL: No probed ethernet devices
Interactive-mode selected
USER1: create a new mbuf pool <mbuf_pool_socket_0>: n=395456, size=2176,
socket=0
Done
testpmd> show port info all
testpmd> quit


[2]
# ./testpmd -- -i
EAL: Detected 32 lcore(s)
EAL: Probing VFIO support...
EAL: PCI device 0000:04:00.0 on NUMA socket 0
EAL:   probe driver: 8086:1521 rte_igb_pmd
EAL: PCI device 0000:08:00.0 on NUMA socket 0
EAL:   probe driver: 8086:10fb rte_ixgbe_pmd
EAL: PCI device 0000:08:00.1 on NUMA socket 0
EAL:   probe driver: 8086:10fb rte_ixgbe_pmd
EAL: PCI device 0000:0a:00.0 on NUMA socket 0
EAL:   probe driver: 8086:1584 rte_i40e_pmd
EAL: PCI device 0000:81:00.0 on NUMA socket 1
EAL:   probe driver: 8086:10fb rte_ixgbe_pmd
EAL: PCI device 0000:81:00.1 on NUMA socket 1
EAL:   probe driver: 8086:10fb rte_ixgbe_pmd
EAL: PCI device 0000:86:00.0 on NUMA socket 1
EAL:   probe driver: 8086:15a4 rte_pmd_fm10k
EAL: PCI device 0000:87:00.0 on NUMA socket 1
EAL:   probe driver: 8086:15a4 rte_pmd_fm10k
EAL: PCI device 0000:88:00.0 on NUMA socket 1
EAL:   probe driver: 8086:1572 rte_i40e_pmd
EAL: PCI device 0000:88:00.1 on NUMA socket 1
EAL:   probe driver: 8086:1572 rte_i40e_pmd
EAL: PCI device 0000:88:00.2 on NUMA socket 1
EAL:   probe driver: 8086:1572 rte_i40e_pmd
EAL: PCI device 0000:88:00.3 on NUMA socket 1
EAL:   probe driver: 8086:1572 rte_i40e_pmd
Interactive-mode selected
USER1: create a new mbuf pool <mbuf_pool_socket_0>: n=395456, size=2176,
socket=0
Configuring Port 0 (socket 0)
Port 0: 90:E2:BA:0E:49:B9
Configuring Port 1 (socket 0)
Port 1: 00:1B:21:76:FA:21
Checking link statuses...
Port 0 Link Up - speed 10000 Mbps - full-duplex
Port 1 Link Up - speed 10000 Mbps - full-duplex
Done
testpmd>
Thomas Monjalon June 10, 2016, 10:18 a.m. UTC | #3
2016-06-10 10:57, Ferruh Yigit:
> On 6/10/2016 10:03 AM, Thomas Monjalon wrote:
> > 2016-05-27 17:48, Ferruh Yigit:
> >> --whole-archive argument only required for pmd libraries, and currently
> >> it covers more libraries. Reducing scope of the argument to pmd
> >> libraries slightly reduce final application size.
> > 
> > In my understanding, --whole-archive is required for static libraries used
> > by plugins:
> > 	http://dpdk.org/commit/20afd76a
> Right, required for static libraries. But more than used by plugins,
> required for plugin to work.
> 
> > If we want to restrict it, I would say it must be around EAL, ring,
> > mbuf, mempool, ethdev, cryptodev, etc.
> > 
> 
> We should restrict to plugins. What happens is, since there is no direct
> call to PMDs from application, PMD code is not included into final
> application. To force PMDs to be included into final binary,
> --whole-archive is required.
> 
> But --whole-archive is not required for other libraries, and just cause
> unnecessary increase in binary size.

Yes it is necessary for other libraries, otherwise some symbols needed by
plugins won't be available.
Example:
In function `rte_pmd_af_packet_devuninit':
rte_eth_af_packet.c:(.text+0x8bf): undefined reference to `rte_eth_dev_allocated'
rte_eth_af_packet.c:(.text+0x930): undefined reference to `rte_eth_dev_release_port'
Ferruh Yigit June 10, 2016, 11:06 a.m. UTC | #4
On 6/10/2016 11:18 AM, Thomas Monjalon wrote:
> 2016-06-10 10:57, Ferruh Yigit:
>> On 6/10/2016 10:03 AM, Thomas Monjalon wrote:
>>> 2016-05-27 17:48, Ferruh Yigit:
>>>> --whole-archive argument only required for pmd libraries, and currently
>>>> it covers more libraries. Reducing scope of the argument to pmd
>>>> libraries slightly reduce final application size.
>>>
>>> In my understanding, --whole-archive is required for static libraries used
>>> by plugins:
>>> 	http://dpdk.org/commit/20afd76a
>> Right, required for static libraries. But more than used by plugins,
>> required for plugin to work.
>>
>>> If we want to restrict it, I would say it must be around EAL, ring,
>>> mbuf, mempool, ethdev, cryptodev, etc.
>>>
>>
>> We should restrict to plugins. What happens is, since there is no direct
>> call to PMDs from application, PMD code is not included into final
>> application. To force PMDs to be included into final binary,
>> --whole-archive is required.
>>
>> But --whole-archive is not required for other libraries, and just cause
>> unnecessary increase in binary size.
> 
> Yes it is necessary for other libraries, otherwise some symbols needed by
> plugins won't be available.
> Example:
> In function `rte_pmd_af_packet_devuninit':
> rte_eth_af_packet.c:(.text+0x8bf): undefined reference to `rte_eth_dev_allocated'
> rte_eth_af_packet.c:(.text+0x930): undefined reference to `rte_eth_dev_release_port'
> 
If there is a direct call to that API, linker should include required
object file into final binary, --whole-archive shouldn't be required for
this.

What is special for PMD's is there are not directly called, but
initialized using constructors.

How are you getting above compilation error?

Thanks,
ferruh
Thomas Monjalon June 10, 2016, 12:21 p.m. UTC | #5
2016-06-10 12:06, Ferruh Yigit:
> On 6/10/2016 11:18 AM, Thomas Monjalon wrote:
> > 2016-06-10 10:57, Ferruh Yigit:
> >> On 6/10/2016 10:03 AM, Thomas Monjalon wrote:
> >>> 2016-05-27 17:48, Ferruh Yigit:
> >>>> --whole-archive argument only required for pmd libraries, and currently
> >>>> it covers more libraries. Reducing scope of the argument to pmd
> >>>> libraries slightly reduce final application size.
> >>>
> >>> In my understanding, --whole-archive is required for static libraries used
> >>> by plugins:
> >>> 	http://dpdk.org/commit/20afd76a
> >> Right, required for static libraries. But more than used by plugins,
> >> required for plugin to work.
> >>
> >>> If we want to restrict it, I would say it must be around EAL, ring,
> >>> mbuf, mempool, ethdev, cryptodev, etc.
> >>>
> >>
> >> We should restrict to plugins. What happens is, since there is no direct
> >> call to PMDs from application, PMD code is not included into final
> >> application. To force PMDs to be included into final binary,
> >> --whole-archive is required.
> >>
> >> But --whole-archive is not required for other libraries, and just cause
> >> unnecessary increase in binary size.
> > 
> > Yes it is necessary for other libraries, otherwise some symbols needed by
> > plugins won't be available.
> > Example:
> > In function `rte_pmd_af_packet_devuninit':
> > rte_eth_af_packet.c:(.text+0x8bf): undefined reference to `rte_eth_dev_allocated'
> > rte_eth_af_packet.c:(.text+0x930): undefined reference to `rte_eth_dev_release_port'
> > 
> If there is a direct call to that API, linker should include required
> object file into final binary, --whole-archive shouldn't be required for
> this.
> 
> What is special for PMD's is there are not directly called, but
> initialized using constructors.
> 
> How are you getting above compilation error?

The error was generated with your patch on top of others I'm going to send.
I suggest to continue the discussion/review based on the v2 coming soon.
Ferruh Yigit June 10, 2016, 1:35 p.m. UTC | #6
On 6/10/2016 1:21 PM, Thomas Monjalon wrote:
> 2016-06-10 12:06, Ferruh Yigit:
>> On 6/10/2016 11:18 AM, Thomas Monjalon wrote:
>>> 2016-06-10 10:57, Ferruh Yigit:
>>>> On 6/10/2016 10:03 AM, Thomas Monjalon wrote:
>>>>> 2016-05-27 17:48, Ferruh Yigit:
>>>>>> --whole-archive argument only required for pmd libraries, and currently
>>>>>> it covers more libraries. Reducing scope of the argument to pmd
>>>>>> libraries slightly reduce final application size.
>>>>>
>>>>> In my understanding, --whole-archive is required for static libraries used
>>>>> by plugins:
>>>>> 	http://dpdk.org/commit/20afd76a
>>>> Right, required for static libraries. But more than used by plugins,
>>>> required for plugin to work.
>>>>
>>>>> If we want to restrict it, I would say it must be around EAL, ring,
>>>>> mbuf, mempool, ethdev, cryptodev, etc.
>>>>>
>>>>
>>>> We should restrict to plugins. What happens is, since there is no direct
>>>> call to PMDs from application, PMD code is not included into final
>>>> application. To force PMDs to be included into final binary,
>>>> --whole-archive is required.
>>>>
>>>> But --whole-archive is not required for other libraries, and just cause
>>>> unnecessary increase in binary size.
>>>
>>> Yes it is necessary for other libraries, otherwise some symbols needed by
>>> plugins won't be available.
>>> Example:
>>> In function `rte_pmd_af_packet_devuninit':
>>> rte_eth_af_packet.c:(.text+0x8bf): undefined reference to `rte_eth_dev_allocated'
>>> rte_eth_af_packet.c:(.text+0x930): undefined reference to `rte_eth_dev_release_port'
>>>
>> If there is a direct call to that API, linker should include required
>> object file into final binary, --whole-archive shouldn't be required for
>> this.
>>
>> What is special for PMD's is there are not directly called, but
>> initialized using constructors.
>>
>> How are you getting above compilation error?
> 
> The error was generated with your patch on top of others I'm going to send.
> I suggest to continue the discussion/review based on the v2 coming soon.
> 

I have quickly checked v2, you are getting this compiler error because
--start-group/--end-group removed, so pmd not able to resolve the
references. And you are working around this by adding all object of
library unconditionally (--whole-archive). I will comment more of v2 patch.

Thanks,
ferruh
diff mbox

Patch

diff --git a/mk/rte.app.mk b/mk/rte.app.mk
index e12226c..8bfe240 100644
--- a/mk/rte.app.mk
+++ b/mk/rte.app.mk
@@ -57,7 +57,6 @@  _LDLIBS-y += -L$(RTE_SDK_BIN)/lib
 # Order is important: from higher level to lower level
 #
 
-_LDLIBS-y += --whole-archive
 _LDLIBS-y += --start-group
 
 _LDLIBS-$(CONFIG_RTE_LIBRTE_DISTRIBUTOR)    += -lrte_distributor
@@ -129,6 +128,8 @@  _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_XENVIRT)    += -lrte_pmd_xenvirt
 ifeq ($(CONFIG_RTE_BUILD_SHARED_LIB),n)
 # plugins (link only if static libraries)
 
+_LDLIBS-y += --whole-archive
+
 _LDLIBS-$(CONFIG_RTE_LIBRTE_VMXNET3_PMD)    += -lrte_pmd_vmxnet3_uio
 _LDLIBS-$(CONFIG_RTE_LIBRTE_VIRTIO_PMD)     += -lrte_pmd_virtio
 _LDLIBS-$(CONFIG_RTE_LIBRTE_BNX2X_PMD)      += -lrte_pmd_bnx2x
@@ -174,11 +175,12 @@  _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_VHOST)      += -lrte_pmd_vhost
 
 endif # $(CONFIG_RTE_LIBRTE_VHOST)
 
+_LDLIBS-y += --no-whole-archive
+
 endif # ! $(CONFIG_RTE_BUILD_SHARED_LIB)
 
 _LDLIBS-y += $(EXECENV_LDLIBS)
 _LDLIBS-y += --end-group
-_LDLIBS-y += --no-whole-archive
 
 LDLIBS += $(_LDLIBS-y) $(CPU_LDLIBS) $(EXTRA_LDLIBS)