[dpdk-dev] test setup for l3fwd application running on VM

2014-04-21 Thread B Gopikrishna 


Hi
I have the below setup and would like to check the l3fwd application running on 
VM 

VM1 VM2(DPDK-L3fwd) 
   VM3
vnic1 --bridge1 vnic1 vnic2--bridge2vnic1


On VM1 -TCP client and, on VM3 -TCP server applications are running
The TCP client and server should communicate via the DPDK L3fwd application 
running on VM2.

I created a static arp entry on VM1 in order that the packets reach VM2 vnic1, 
and then the DPDK-L3fwd app would take care of forwarding the packets to VM2 
vnic2, and further the packets reach VM3 vnic1

This is not working for me, please let me know if I need to do some additional 
steps in order to complete the flow.

All the vnic's are configured as e1000 drivers.

Regards
Gopi Krishna B
=-=-=
Notice: The information contained in this e-mail
message and/or attachments to it may contain 
confidential or privileged information. If you are 
not the intended recipient, any dissemination, use, 
review, distribution, printing or copying of the 
information contained in this e-mail message 
and/or attachments to it are strictly prohibited. If 
you have received this communication in error, 
please notify us by reply e-mail or telephone and 
immediately and permanently delete the message 
and any attachments. Thank you

[dpdk-dev] [PATCH 01/15 v2] makefiles: Fixed -share command line option error

2014-04-21 Thread Neil Horman 
On Fri, Apr 18, 2014 at 03:29:01PM +0200, Thomas Monjalon wrote:
> 2014-04-18 09:18, Neil Horman:
> > On Fri, Apr 18, 2014 at 01:23:19PM +0200, Thomas Monjalon wrote:
> > > I think that CPU_LDFLAGS should be prefixed with -Wl, in case of CC
> > > linking. So blindly assigning CC to LD variable seems a bad idea.
> > > Other makefiles have different O_TO_S commands depending of LINK_USING_CC.
> > 
> > I'm not so sure about that.  Or more specifically, I wonder if some more
> > rework isn't needed here.  I say that because, while what you say makes
> > sense in terms of formatting the CPU_FLAGS variable for use with CC, the
> > only current use of CPU_LDFLAGS set -melf_i386, which IIRC is a gcc flag,
> > not meant to be passed to LD.  I can change the makefile to completely
> > rewrite the comand based on LINK_USING_CC, but it seems to me that
> > CPU_LDFLAGS should not be passed in the use of the LD case.
> 
> Right, -melf_i386 shouldn't be a LDFLAG.
> Feel free to fix it.
> By the way, It's cleaner to prepare -Wl prefixing and keep an empty LDFLAGS.
> 
Actually, I looked into this a bit more, turns out -melf_i386 is an LDFLAG.  Its
a bit silly to have it there because its passed automatically from the compiler
to the linker depending on the machine that gcc is building for, or ld does it
natively depending on the input file format.  But perhaps some compiler or
linker version has a bug where that doesn't happen properly.  Either way, it can
be prefixed with a -Wl, so I'm just going to leave it alone
Neil

> -- 
> Thomas
>

[dpdk-dev] [PATCH v5 00/14] dpdk: Separate compile time linkage between eal lib and pmd's

2014-04-21 Thread Neil Horman 
Disconnect compile time linkage between eal library / applications and pmd's

I noticed that, while tinkering with dpdk, building for shared libraries still
resulted in all the test applications linking to all the built pmd's, despite
not actually needing them all.  We are able to tell an application at run time
(via the -d/--blacklist/--whitelist/--vdev options) which pmd's we want to use,
and so have no need to link them at all. The only reason they get pulled in is
because rte_eal_non_pci_init_etherdev and rte_pmd_init_all contain static lists
to the individual pmd init functions. The result is that, even when building as
DSO's, we have to load all the pmd libraries, which is space inefficient and
defeating of some of the purpose of shared objects.

To correct this, I developed this patch series, which introduces a new macro,
PMD_REGISTER_DRIVER, which wraps up Oliviers work using constructors on the
virtual device pmds, then expands on it to include the physical device pmds,
allowing us to break linkages between dpdk applications and pmd's almost
entirely (save for the ring and xenvirt drivers, which have additional api's
outside of the standard dpdk code that we need to further fix).  This also
allows us to completely remove the rte_pmd_init_all routine, hiding its function
internally to the rte_eal_init path.

I've tested this feature using the igb and pcap pmd's, both statically and
dynamically linked with the test and testpmd sample applications, and it seems
to work well.

Note, I encountered  a few bugs along the way, which I fixed and noted in the
series.

Regards
Neil

Change Notes:

1) Reposted the entire series.  Recent changes permeated accross several
patches, and since a few patches already had multiple versions, I've reposted
the entire series and bumped the version number to 5, whcih skips a few
versions, but ensures this series is seen as the most recent.

2) Merged the PMD_REGISTER_DRIVER macro into rte_dev.h, which is a better
location for it.  Required removing include of rte_pmd.h across most of the
patches in the series.

3) Cleaned up various leftover "virtual" comments in the driver registration api
that no longer belong there after making it generic

4) Fixed the LINK_USING_CC check in rte.lib.mk so it works like other uses

5) Fixed CPU_LDFLAGS conversion to use -Wl in case we link with gcc.

[dpdk-dev] [PATCH 0/X v5 01/14] makefiles: Fixed -share command line option error

2014-04-21 Thread Neil Horman 
The shared libraries built with the current makefile set produce static
libraries rather than actual shared objects.  This is due to several missing
options that are required to correctly build shared objects using ld, as well as
a mis-specified -share option (which should be -shared). Switching to the use of
CC rather than LD and fixing the -shared option corrects these problems and
builds the DSOs correctly.

Signed-off-by: Neil Horman 

---
Change notes:

v2) Added test for LINK_USING_CC to convert to link with gcc
---
 mk/rte.lib.mk  | 14 --
 mk/rte.sharelib.mk |  2 +-
 2 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/mk/rte.lib.mk b/mk/rte.lib.mk
index f75ca92..fc8fd01 100644
--- a/mk/rte.lib.mk
+++ b/mk/rte.lib.mk
@@ -59,6 +59,18 @@ build: _postbuild

 exe2cmd = $(strip $(call dotfile,$(patsubst %,%.cmd,$(1

+ifeq ($(LINK_USING_CC),1)
+comma := ,
+CPU_LDFLAGS := $(addprefix -Wl$(comma),$(CPU_LDFLAGS))
+O_TO_S = $(CC) $(CPU_LDFLAGS) -Wl$(comma)-z$(comma)muldefs -shared $(OBJS-y) 
-o $(LIB)
+
+ifeq ($(RTE_BUILD_SHARED_LIB),y)
+O_TO_C = $(CC) -z muldefs -shared $(OBJS-y) -o $(LIB_ONE)
+endif
+else
+O_TO_S = $(LD) $(CPU_LDFLAGS) -z muldefs -shared $(OBJS-y) -o $(LIB)
+endif
+
 O_TO_A = $(AR) crus $(LIB) $(OBJS-y)
 O_TO_A_STR = $(subst ','\'',$(O_TO_A)) #'# fix syntax highlight
 O_TO_A_DISP = $(if $(V),"$(O_TO_A_STR)","  AR $(@)")
@@ -68,7 +80,6 @@ O_TO_A_DO = @set -e; \
$(O_TO_A) && \
echo $(O_TO_A_CMD) > $(call exe2cmd,$(@))

-O_TO_S = $(LD) $(CPU_LDFLAGS) -z muldefs -share $(OBJS-y) -o $(LIB)
 O_TO_S_STR = $(subst ','\'',$(O_TO_S)) #'# fix syntax highlight
 O_TO_S_DISP = $(if $(V),"$(O_TO_S_STR)","  LD $(@)")
 O_TO_S_DO = @set -e; \
@@ -84,7 +95,6 @@ O_TO_C_DO = @set -e; \
$(lib_dir) \
$(copy_obj)
 else
-O_TO_C = $(LD) -z muldefs -share $(OBJS-y) -o $(LIB_ONE)
 O_TO_C_STR = $(subst ','\'',$(O_TO_C)) #'# fix syntax highlight
 O_TO_C_DISP = $(if $(V),"$(O_TO_C_STR)","  LD_C $(@)")
 O_TO_C_DO = @set -e; \
diff --git a/mk/rte.sharelib.mk b/mk/rte.sharelib.mk
index 3967b51..429309f 100644
--- a/mk/rte.sharelib.mk
+++ b/mk/rte.sharelib.mk
@@ -45,7 +45,7 @@ sharelib: $(LIB_ONE) FORCE

 OBJS = $(wildcard $(RTE_OUTPUT)/build/lib/*.o)

-O_TO_S = $(LD) $(CPU_LDFLAGS) -share $(OBJS) -o $(RTE_OUTPUT)/lib/$(LIB_ONE)
+O_TO_S = $(LD) $(CPU_LDFLAGS) -shared $(OBJS) -o $(RTE_OUTPUT)/lib/$(LIB_ONE)
 O_TO_S_STR = $(subst ','\'',$(O_TO_S)) #'# fix syntax highlight
 O_TO_S_DISP = $(if $(V),"$(O_TO_S_STR)","  LD $(@)")
 O_TO_S_CMD = "cmd_$@ = $(O_TO_S_STR)"
-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 02/14] pmd: Add PMD_REGISTER_DRIVER macro

2014-04-21 Thread Neil Horman 
Rather than have each driver have to remember to add a constructor to it to make
sure its gets registered properly, wrap that process up in a macro to make
registration a one line affair.  This also sets the stage for us to make
registration of vdev pmds and physical pmds a uniform process

Signed-off-by: Neil Horman 
---
 lib/librte_eal/common/Makefile   |  2 +-
 lib/librte_eal/common/include/rte_vdev.h | 18 ++
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/lib/librte_eal/common/Makefile b/lib/librte_eal/common/Makefile
index 2f99bf4..d3e7e39 100644
--- a/lib/librte_eal/common/Makefile
+++ b/lib/librte_eal/common/Makefile
@@ -37,7 +37,7 @@ INC += rte_log.h rte_memcpy.h rte_memory.h rte_memzone.h 
rte_pci.h
 INC += rte_pci_dev_ids.h rte_per_lcore.h rte_prefetch.h rte_random.h
 INC += rte_rwlock.h rte_spinlock.h rte_tailq.h rte_interrupts.h rte_alarm.h
 INC += rte_string_fns.h rte_cpuflags.h rte_version.h rte_tailq_elem.h
-INC += rte_eal_memconfig.h rte_malloc_heap.h
+INC += rte_eal_memconfig.h rte_malloc_heap.h 
 INC += rte_hexdump.h rte_devargs.h rte_vdev.h

 ifeq ($(CONFIG_RTE_INSECURE_FUNCTION_WARNING),y)
diff --git a/lib/librte_eal/common/include/rte_vdev.h 
b/lib/librte_eal/common/include/rte_vdev.h
index 48f71b7..c7aa795 100644
--- a/lib/librte_eal/common/include/rte_vdev.h
+++ b/lib/librte_eal/common/include/rte_vdev.h
@@ -83,6 +83,24 @@ void rte_eal_vdev_driver_register(struct rte_vdev_driver 
*driver);
  */
 void rte_eal_vdev_driver_unregister(struct rte_vdev_driver *driver);

+enum rte_pmd_driver_type {
+   PMD_VDEV = 1
+};
+
+extern void rte_eal_nonpci_dev_init_register(const char *name, int 
(*dev_initfn)(const char *, const char *));
+#define PMD_REGISTER_DRIVER(d, t)\
+void devinitfn_ ##d(void);\
+void __attribute__((constructor, used)) devinitfn_ ##d(void)\
+{\
+   enum rte_pmd_driver_type _t = (t);\
+   switch(_t)\
+   {\
+   case PMD_VDEV:\
+   rte_eal_vdev_driver_register(&d);\
+   break;\
+   };\
+}
+
 #ifdef __cplusplus
 }
 #endif
-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 03/14] pcap: Convert to use of PMD_REGISTER_DRIVER and fix linking

2014-04-21 Thread Neil Horman 
convert the pcap driver to use the PMD_REGISTER_DRIVER macro and fix up the
Makefile so that its linkage is only done if we are building static libraries.
This means that the test applications now have no reference to the pcap library
when building DSO's and must specify its use on the command line with the -d
option.  Static linking will still initalize the driver automatically.

Signed-off-by: Neil Horman 
---
 lib/librte_pmd_pcap/rte_eth_pcap.c | 7 +--
 mk/rte.app.mk  | 2 ++
 2 files changed, 3 insertions(+), 6 deletions(-)

diff --git a/lib/librte_pmd_pcap/rte_eth_pcap.c 
b/lib/librte_pmd_pcap/rte_eth_pcap.c
index 680dfdc..c987940 100644
--- a/lib/librte_pmd_pcap/rte_eth_pcap.c
+++ b/lib/librte_pmd_pcap/rte_eth_pcap.c
@@ -772,9 +772,4 @@ static struct rte_vdev_driver pmd_pcap_drv = {
.init = rte_pmd_pcap_devinit,
 };

-__attribute__((constructor))
-static void
-rte_pmd_pcap_init(void)
-{
-   rte_eal_vdev_driver_register(&pmd_pcap_drv);
-}
+PMD_REGISTER_DRIVER(pmd_pcap_drv, PMD_VDEV);
diff --git a/mk/rte.app.mk b/mk/rte.app.mk
index 072718a..83f0867 100644
--- a/mk/rte.app.mk
+++ b/mk/rte.app.mk
@@ -172,9 +172,11 @@ ifeq ($(CONFIG_RTE_LIBRTE_CMDLINE),y)
 LDLIBS += -lrte_cmdline
 endif

+ifeq ($(RTE_BUILD_SHARED_LIB),n)
 ifeq ($(CONFIG_RTE_LIBRTE_PMD_PCAP),y)
 LDLIBS += -lrte_pmd_pcap -lpcap
 endif
+endif

 LDLIBS += $(EXECENV_LDLIBS)

-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 04/14] ring: Convert to use of PMD_REGISTER_DRIVER and fix linking

2014-04-21 Thread Neil Horman 
convert the ring driver to use the PMD_REGISTER_DRIVER macro and fix up the
Makefile so that its linkage is only done if we are building static libraries.
This means that the test applications now have no reference to the ring library
when building DSO's and must specify its use on the command line with the -d
option.  Static linking will still initalize the driver automatically.

Note that the ring driver was also written in such a way that it violated some
general layering principles, several functions were contained in the pmd which
were being called by example from the test application in the app/test
directory.  Specifically it was calling eth_ring_pair_attach,
eth_ring_pair_create and rte_eth_ring_devinit, which should only be called
internally to the dpdk core library.  To correct this I've removed those
functions, and instead allowed them to be called indirectly at initalization
time using the vdev command line argument key nodeaction=::
where action is one of ATTACH or CREATE.  I've tested out the functionality of
the command line with the testpmd utility, with success, and have removed the
called functions from the test utility.  This will affect how the test utility
is invoked (the -d and --vdev option will need to be specified on the command
line now), but honestly, given the way it was coded, I think the testing of the
ring pmd was not the best example of how to code with dpdk to begin with.  I
have also left the two layer violating functions in place, so as not to break
existing applications, but added deprecation warnings to them so that apps can
migrate off them.

Signed-off-by: Neil Horman 

---
Change notes

v2) fixed DEPDIR specifcation, should depend on RING not PCAP

v3) Cleaned up strcmp error checking, printfs, and other parsing issues as
pointed out by Konstantin Ananyev 
---
 app/test/test_pmd_ring.c   |  95 -
 lib/librte_pmd_ring/Makefile   |   1 +
 lib/librte_pmd_ring/rte_eth_ring.c | 140 ++---
 mk/rte.app.mk  |  10 +--
 4 files changed, 137 insertions(+), 109 deletions(-)

diff --git a/app/test/test_pmd_ring.c b/app/test/test_pmd_ring.c
index 4d9c2ba..1fe38fa 100644
--- a/app/test/test_pmd_ring.c
+++ b/app/test/test_pmd_ring.c
@@ -42,7 +42,6 @@
 /* two test rings, r1 is used by two ports, r2 just by one */
 static struct rte_ring *r1[2], *r2;

-static struct rte_ring *nullring = NULL;
 static struct rte_mempool *mp;
 static uint8_t start_idx; /* will store the port id of the first of our new 
ports */

@@ -59,58 +58,6 @@ static uint8_t start_idx; /* will store the port id of the 
first of our new port
 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 #define NB_MBUF   512

-
-static int
-test_ring_ethdev_create(void)
-{
-   int retval;
-   printf("Testing ring pmd create\n");
-
-   retval = rte_eth_from_rings(NULL, 0, NULL, 0, SOCKET0);
-   if (retval < 0) {
-   printf("Failure, failed to create zero-sized RXTX ring pmd\n");
-   return -1;
-   }
-
-   retval = rte_eth_from_rings(NULL, 0, NULL, 0, RTE_MAX_NUMA_NODES);
-   if (retval >= 0) {
-   printf("Failure, can create ring pmd on socket %d\n", 
RTE_MAX_NUMA_NODES);
-   return -1;
-   }
-
-   retval = rte_eth_from_rings(NULL, 1, &r2, 1, SOCKET0);
-   if (retval >= 0) {
-   printf("Failure, can create pmd with null rx rings\n");
-   return -1;
-   }
-
-   retval = rte_eth_from_rings(r1, 1, NULL, 1, SOCKET0);
-   if (retval >= 0) {
-   printf("Failure, can create pmd with null tx rings\n");
-   return -1;
-   }
-
-   retval = rte_eth_from_rings(&nullring, 1, r1, 2, SOCKET0);
-   if (retval < 0) {
-   printf("Failure, failed to create TX-only ring pmd\n");
-   return -1;
-   }
-
-   retval = rte_eth_from_rings(r1, 1, &nullring, 1, SOCKET0);
-   if (retval < 0) {
-   printf("Failure, failed to create RX-only ring pmd\n");
-   return -1;
-   }
-
-   retval = rte_eth_from_rings(&r2, 1, &r2, 1, SOCKET0);
-   if (retval < 0) {
-   printf("Failure, failed to create RXTX ring pmd\n");
-   return -1;
-   }
-
-   return 0;
-}
-
 static int
 test_ethdev_configure(void)
 {
@@ -305,26 +252,12 @@ test_stats_reset(void)
 static int 
 test_pmd_ring_init(void)
 {
-   const char * name1 = "R3";
-   const char * name2 = "R4";
-   const char * params_null = NULL;
-   const char * params = "PARAMS";
struct rte_eth_stats stats;
struct rte_mbuf buf, *pbuf = &buf;
struct rte_eth_conf null_conf;

printf("Testing ring pmd init\n");

-   if (rte_pmd_ring_devinit(name1, params_null) < 0) {
-   printf("Testing ring pmd init fail\n");
-   return -1;
-   }
-
-   if (rte_pmd_ring_devinit(name2, params) < 0)

[dpdk-dev] [PATCH 0/X v5 05/14] xenvirt: Convert to use of PMD_REGISTER_DRIVER and fix linking

2014-04-21 Thread Neil Horman 
convert the xenvirt driver to use the PMD_REGISTER_DRIVER macro.
This means that the test applications now have no reference to the xenvirt 
library
when building DSO's and must specify its use on the command line with the -d
option.  Static linking will still initalize the driver automatically.

A few notes:

xenvirt was unbuildable as of commit 4c39baf297d10c217e7d3e7370f26a1fede58308..
That commit neglected to include the rte_vdev.h header, so several structs were
left undefined.  This patch includes a fix for that as well.

Also, The linkage for xenvirt is broken in much the same way pmd_ring was, in
that the xenvirt pmd has a function that is called directly from applications
(the example being the testpmd application).  The function is
rte_mempool_gntalloc_create, and should clearly be moved into the rte_mempool
library, with the supporting code in the function implementation moved to a new
xenvirt library separate from the pmd.  This is a large undertaking that
detracts from the purpose of this series however, and so for now, I'm leaving
the linkage to the application in place, and will address this issue in a later
series

Signed-off-by: Neil Horman 
---
 lib/librte_pmd_xenvirt/rte_eth_xenvirt.c | 8 ++--
 mk/rte.app.mk| 7 +++
 2 files changed, 5 insertions(+), 10 deletions(-)

diff --git a/lib/librte_pmd_xenvirt/rte_eth_xenvirt.c 
b/lib/librte_pmd_xenvirt/rte_eth_xenvirt.c
index 533aa76..ecafe7a 100644
--- a/lib/librte_pmd_xenvirt/rte_eth_xenvirt.c
+++ b/lib/librte_pmd_xenvirt/rte_eth_xenvirt.c
@@ -53,6 +53,7 @@
 #include 
 #include 
 #include 
+#include 
 #include 
 #include 

@@ -710,9 +711,4 @@ static struct rte_vdev_driver pmd_xenvirt_drv = {
.init = rte_pmd_xenvirt_devinit,
 };

-__attribute__((constructor))
-static void
-rte_pmd_xenvirt_init(void)
-{
-   rte_eal_vdev_driver_register(&pmd_xenvirt_drv);
-}
+PMD_REGISTER_DRIVER(pmd_xenvirt_drv, PMD_VDEV);
diff --git a/mk/rte.app.mk b/mk/rte.app.mk
index 16edbce..aabe75e 100644
--- a/mk/rte.app.mk
+++ b/mk/rte.app.mk
@@ -158,16 +158,15 @@ ifeq ($(CONFIG_RTE_LIBRTE_EAL),y)
 LDLIBS += -lrte_eal
 endif

+ifeq ($(CONFIG_RTE_LIBRTE_CMDLINE),y)
+LDLIBS += -lrte_cmdline
+endif

 ifeq ($(CONFIG_RTE_LIBRTE_PMD_XENVIRT),y)
 LDLIBS += -lrte_pmd_xenvirt
 LDLIBS += -lxenstore
 endif

-ifeq ($(CONFIG_RTE_LIBRTE_CMDLINE),y)
-LDLIBS += -lrte_cmdline
-endif
-
 ifeq ($(RTE_BUILD_SHARED_LIB),n)

 ifeq ($(CONFIG_RTE_LIBRTE_PMD_RING),y)
-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 06/14] eal: Make vdev init path generic for both virtual and physcial devices

2014-04-21 Thread Neil Horman 
Currently, physical device pmds use a separate initalization path
(rte_pmd_init_all) while virtual devices use a constructor registration and
rte_eal_dev_init.  Theres no reason to have them be separate.  This patch
removes the vdev specific nomenclature from the vdev init path and makes it more
generic for use with all pmds.  This is the first step in converting the
physical device pmds to using the same constructor based registration path that
the virtual devices use

Signed-off-by: Neil Horman 
---
 lib/librte_eal/bsdapp/eal/Makefile  |   2 +-
 lib/librte_eal/bsdapp/eal/eal.c |   4 +-
 lib/librte_eal/common/Makefile  |   4 +-
 lib/librte_eal/common/eal_common_dev.c  | 109 +++
 lib/librte_eal/common/eal_common_vdev.c |  97 
 lib/librte_eal/common/include/eal_private.h |   2 +-
 lib/librte_eal/common/include/rte_dev.h | 111 
 lib/librte_eal/common/include/rte_vdev.h| 108 ---
 lib/librte_eal/linuxapp/eal/Makefile|   2 +-
 lib/librte_eal/linuxapp/eal/eal.c   |   4 +-
 lib/librte_pmd_pcap/rte_eth_pcap.c  |   7 +-
 lib/librte_pmd_ring/rte_eth_ring.c  |   7 +-
 lib/librte_pmd_xenvirt/rte_eth_xenvirt.c|   7 +-
 13 files changed, 241 insertions(+), 223 deletions(-)
 create mode 100644 lib/librte_eal/common/eal_common_dev.c
 delete mode 100644 lib/librte_eal/common/eal_common_vdev.c
 create mode 100644 lib/librte_eal/common/include/rte_dev.h
 delete mode 100644 lib/librte_eal/common/include/rte_vdev.h

diff --git a/lib/librte_eal/bsdapp/eal/Makefile 
b/lib/librte_eal/bsdapp/eal/Makefile
index 4c2a4f1..abf1ad2 100644
--- a/lib/librte_eal/bsdapp/eal/Makefile
+++ b/lib/librte_eal/bsdapp/eal/Makefile
@@ -69,7 +69,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_EAL_BSDAPP) += eal_common_errno.c
 SRCS-$(CONFIG_RTE_LIBRTE_EAL_BSDAPP) += eal_common_cpuflags.c
 SRCS-$(CONFIG_RTE_LIBRTE_EAL_BSDAPP) += eal_common_hexdump.c
 SRCS-$(CONFIG_RTE_LIBRTE_EAL_BSDAPP) += eal_common_whitelist.c
-SRCS-$(CONFIG_RTE_LIBRTE_EAL_BSDAPP) += eal_common_vdev.c
+SRCS-$(CONFIG_RTE_LIBRTE_EAL_BSDAPP) += eal_common_dev.c

 CFLAGS_eal.o := -D_GNU_SOURCE
 #CFLAGS_eal_thread.o := -D_GNU_SOURCE
diff --git a/lib/librte_eal/bsdapp/eal/eal.c b/lib/librte_eal/bsdapp/eal/eal.c
index e944aba..8de3fad 100644
--- a/lib/librte_eal/bsdapp/eal/eal.c
+++ b/lib/librte_eal/bsdapp/eal/eal.c
@@ -854,8 +854,8 @@ rte_eal_init(int argc, char **argv)

rte_eal_mcfg_complete();

-   if (rte_eal_vdev_init() < 0)
-   rte_panic("Cannot init virtual devices\n");
+   if (rte_eal_dev_init() < 0)
+   rte_panic("Cannot init pmd devices\n");

RTE_LCORE_FOREACH_SLAVE(i) {

diff --git a/lib/librte_eal/common/Makefile b/lib/librte_eal/common/Makefile
index d3e7e39..0016fc5 100644
--- a/lib/librte_eal/common/Makefile
+++ b/lib/librte_eal/common/Makefile
@@ -37,8 +37,8 @@ INC += rte_log.h rte_memcpy.h rte_memory.h rte_memzone.h 
rte_pci.h
 INC += rte_pci_dev_ids.h rte_per_lcore.h rte_prefetch.h rte_random.h
 INC += rte_rwlock.h rte_spinlock.h rte_tailq.h rte_interrupts.h rte_alarm.h
 INC += rte_string_fns.h rte_cpuflags.h rte_version.h rte_tailq_elem.h
-INC += rte_eal_memconfig.h rte_malloc_heap.h 
-INC += rte_hexdump.h rte_devargs.h rte_vdev.h
+INC += rte_eal_memconfig.h rte_malloc_heap.h
+INC += rte_hexdump.h rte_devargs.h rte_dev.h

 ifeq ($(CONFIG_RTE_INSECURE_FUNCTION_WARNING),y)
 INC += rte_warnings.h
diff --git a/lib/librte_eal/common/eal_common_dev.c 
b/lib/librte_eal/common/eal_common_dev.c
new file mode 100644
index 000..8c42738
--- /dev/null
+++ b/lib/librte_eal/common/eal_common_dev.c
@@ -0,0 +1,109 @@
+/*-
+ *   BSD LICENSE
+ * 
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2014 6WIND S.A.
+ *   All rights reserved.
+ * 
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ * 
+ * * Redistributions of source code must retain the above copyright
+ *   notice, this list of conditions and the following disclaimer.
+ * * Redistributions in binary form must reproduce the above copyright
+ *   notice, this list of conditions and the following disclaimer in
+ *   the documentation and/or other materials provided with the
+ *   distribution.
+ * * Neither the name of Intel Corporation nor the names of its
+ *   contributors may be used to endorse or promote products derived
+ *   from this software without specific prior written permission.
+ * 
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY

[dpdk-dev] [PATCH 0/X v5 07/14] igb: Convert to use of PMD_REGISTER_DRIVER and fix linking

2014-04-21 Thread Neil Horman 
convert the igb pmd driver to use the PMD_REGISTER_DRIVER macro.
This means that the test applications now have no reference to the igb library
when building DSO's and must specify its use on the command line with the -d
option.  Static linking will still initalize the driver automatically.

Signed-off-by: Neil Horman 
---
 examples/dpdk_qat/main.c  |  4 
 lib/librte_ether/rte_ethdev.h | 13 -
 lib/librte_pmd_e1000/igb_ethdev.c | 12 ++--
 3 files changed, 10 insertions(+), 19 deletions(-)

diff --git a/examples/dpdk_qat/main.c b/examples/dpdk_qat/main.c
index cdf6832..64cb610 100644
--- a/examples/dpdk_qat/main.c
+++ b/examples/dpdk_qat/main.c
@@ -697,10 +697,6 @@ MAIN(int argc, char **argv)
return -1;

/* init driver */
-#ifdef RTE_LIBRTE_IGB_PMD
-   if (rte_igb_pmd_init() < 0)
-   rte_panic("Cannot init igb pmd\n");
-#endif
 #ifdef RTE_LIBRTE_IXGBE_PMD
if (rte_ixgbe_pmd_init() < 0)
rte_panic("Cannot init ixgbe pmd\n");
diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index dea7471..e50ed48 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h
@@ -1317,15 +1317,6 @@ extern void rte_eth_driver_register(struct eth_driver 
*eth_drv);

 /**
  * The initialization function of the driver for
- * Intel(r) IGB Gigabit Ethernet Controller devices.
- * This function is invoked once at EAL start time.
- * @return
- *   0 on success
- */
-extern int rte_igb_pmd_init(void);
-
-/**
- * The initialization function of the driver for
  * Intel(r) EM Gigabit Ethernet Controller devices.
  * This function is invoked once at EAL start time.
  * @return
@@ -1402,10 +1393,6 @@ int rte_pmd_init_all(void)
int ret = -ENODEV;

 #ifdef RTE_LIBRTE_IGB_PMD
-   if ((ret = rte_igb_pmd_init()) != 0) {
-   RTE_LOG(ERR, PMD, "Cannot init igb PMD\n");
-   return (ret);
-   }
if ((ret = rte_igbvf_pmd_init()) != 0) {
RTE_LOG(ERR, PMD, "Cannot init igbvf PMD\n");
return (ret);
diff --git a/lib/librte_pmd_e1000/igb_ethdev.c 
b/lib/librte_pmd_e1000/igb_ethdev.c
index 673b4de..5e578c5 100644
--- a/lib/librte_pmd_e1000/igb_ethdev.c
+++ b/lib/librte_pmd_e1000/igb_ethdev.c
@@ -51,6 +51,7 @@
 #include 
 #include 
 #include 
+#include 

 #include "e1000_logs.h"
 #include "e1000/e1000_api.h"
@@ -619,8 +620,8 @@ static struct eth_driver rte_igbvf_pmd = {
.dev_private_size = sizeof(struct e1000_adapter),
 };

-int
-rte_igb_pmd_init(void)
+static int
+rte_igb_pmd_init(const char *name __rte_unused, const char *params 
__rte_unused)
 {
rte_eth_driver_register(&rte_igb_pmd);
return 0;
@@ -2182,3 +2183,10 @@ eth_igb_rss_reta_query(struct rte_eth_dev *dev,

return 0;
 }
+
+static struct rte_driver pmd_igb_drv = {
+.type = PMD_PDEV,
+   .init = rte_igb_pmd_init,
+};
+
+PMD_REGISTER_DRIVER(pmd_igb_drv);
-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 08/14] igbvf: Convert to use of PMD_REGISTER_DRIVER and fix linking

2014-04-21 Thread Neil Horman 
convert the igbvf pmd driver to use the PMD_REGISTER_DRIVER macro.
This means that the test applications now have no reference to the igbvf library
when building DSO's and must specify its use on the command line with the -d
option.  Static linking will still initalize the driver automatically.

Signed-off-by: Neil Horman 
---
 lib/librte_ether/rte_ethdev.h | 16 
 lib/librte_pmd_e1000/igb_ethdev.c | 10 --
 2 files changed, 8 insertions(+), 18 deletions(-)

diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index e50ed48..d13cc4f 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h
@@ -1325,15 +1325,6 @@ extern void rte_eth_driver_register(struct eth_driver 
*eth_drv);
 extern int rte_em_pmd_init(void);

 /**
- * The initialization function of the driver for 1Gbps Intel IGB_VF
- * Ethernet devices.
- * Invoked once at EAL start time.
- * @return
- *   0 on success
- */
-extern int rte_igbvf_pmd_init(void);
-
-/**
  * The initialization function of the driver for 10Gbps Intel IXGBE
  * Ethernet devices.
  * Invoked once at EAL start time.
@@ -1392,13 +1383,6 @@ int rte_pmd_init_all(void)
 {
int ret = -ENODEV;

-#ifdef RTE_LIBRTE_IGB_PMD
-   if ((ret = rte_igbvf_pmd_init()) != 0) {
-   RTE_LOG(ERR, PMD, "Cannot init igbvf PMD\n");
-   return (ret);
-   }
-#endif /* RTE_LIBRTE_IGB_PMD */
-
 #ifdef RTE_LIBRTE_EM_PMD
if ((ret = rte_em_pmd_init()) != 0) {
RTE_LOG(ERR, PMD, "Cannot init em PMD\n");
diff --git a/lib/librte_pmd_e1000/igb_ethdev.c 
b/lib/librte_pmd_e1000/igb_ethdev.c
index 5e578c5..6aaded0 100644
--- a/lib/librte_pmd_e1000/igb_ethdev.c
+++ b/lib/librte_pmd_e1000/igb_ethdev.c
@@ -643,8 +643,8 @@ igb_vmdq_vlan_hw_filter_enable(struct rte_eth_dev *dev)
  * Invoked one at EAL init time.
  * Register itself as the [Virtual Poll Mode] Driver of PCI IGB devices.
  */
-int
-rte_igbvf_pmd_init(void)
+static int
+rte_igbvf_pmd_init(const char *name __rte_unused, const char *params 
__rte_unused)
 {
DEBUGFUNC("rte_igbvf_pmd_init");

@@ -2189,4 +2189,10 @@ static struct rte_driver pmd_igb_drv = {
.init = rte_igb_pmd_init,
 };

+static struct rte_driver pmd_igbvf_drv = {
+   .type = PMD_PDEV,
+   .init = rte_igbvf_pmd_init,
+};
+
 PMD_REGISTER_DRIVER(pmd_igb_drv);
+PMD_REGISTER_DRIVER(pmd_igbvf_drv);
-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 09/14] e1000: Convert to use of PMD_REGISTER_DRIVER and fix linking

2014-04-21 Thread Neil Horman 
convert the e1000 pmd driver to use the PMD_REGISTER_DRIVER macro.
This means that the test applications now have no reference to the e1000 library
when building DSO's and must specify its use on the command line with the -d
option.  Static linking will still initalize the driver automatically.

Signed-off-by: Neil Horman 
---
 lib/librte_ether/rte_ethdev.h| 16 
 lib/librte_pmd_e1000/em_ethdev.c | 12 ++--
 mk/rte.app.mk|  8 
 3 files changed, 14 insertions(+), 22 deletions(-)

diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index d13cc4f..9e32f56 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h
@@ -1316,15 +1316,6 @@ struct eth_driver {
 extern void rte_eth_driver_register(struct eth_driver *eth_drv);

 /**
- * The initialization function of the driver for
- * Intel(r) EM Gigabit Ethernet Controller devices.
- * This function is invoked once at EAL start time.
- * @return
- *   0 on success
- */
-extern int rte_em_pmd_init(void);
-
-/**
  * The initialization function of the driver for 10Gbps Intel IXGBE
  * Ethernet devices.
  * Invoked once at EAL start time.
@@ -1383,13 +1374,6 @@ int rte_pmd_init_all(void)
 {
int ret = -ENODEV;

-#ifdef RTE_LIBRTE_EM_PMD
-   if ((ret = rte_em_pmd_init()) != 0) {
-   RTE_LOG(ERR, PMD, "Cannot init em PMD\n");
-   return (ret);
-   }
-#endif /* RTE_LIBRTE_EM_PMD */
-
 #ifdef RTE_LIBRTE_IXGBE_PMD
if ((ret = rte_ixgbe_pmd_init()) != 0) {
RTE_LOG(ERR, PMD, "Cannot init ixgbe PMD\n");
diff --git a/lib/librte_pmd_e1000/em_ethdev.c b/lib/librte_pmd_e1000/em_ethdev.c
index d8c9a9b..000feef 100644
--- a/lib/librte_pmd_e1000/em_ethdev.c
+++ b/lib/librte_pmd_e1000/em_ethdev.c
@@ -51,6 +51,7 @@
 #include 
 #include 
 #include 
+#include 

 #include "e1000_logs.h"
 #include "e1000/e1000_api.h"
@@ -285,8 +286,8 @@ static struct eth_driver rte_em_pmd = {
.dev_private_size = sizeof(struct e1000_adapter),
 };

-int
-rte_em_pmd_init(void)
+static int
+rte_em_pmd_init(const char *name __rte_unused, const char *params __rte_unused)
 {
rte_eth_driver_register(&rte_em_pmd);
return 0;
@@ -1433,3 +1434,10 @@ eth_em_rar_clear(struct rte_eth_dev *dev, uint32_t index)

e1000_rar_set(hw, addr, index);
 }
+
+struct rte_driver em_pmd_drv = {
+   .type = PMD_PDEV,
+   .init = rte_em_pmd_init,
+};
+
+PMD_REGISTER_DRIVER(em_pmd_drv);
diff --git a/mk/rte.app.mk b/mk/rte.app.mk
index aabe75e..a94830e 100644
--- a/mk/rte.app.mk
+++ b/mk/rte.app.mk
@@ -73,10 +73,6 @@ LDLIBS += -lrte_ivshmem
 endif
 endif

-ifeq ($(CONFIG_RTE_LIBRTE_E1000_PMD),y)
-LDLIBS += -lrte_pmd_e1000
-endif
-
 ifeq ($(CONFIG_RTE_LIBRTE_IXGBE_PMD),y)
 LDLIBS += -lrte_pmd_ixgbe
 endif
@@ -169,6 +165,10 @@ endif

 ifeq ($(RTE_BUILD_SHARED_LIB),n)

+ifeq ($(CONFIG_RTE_LIBRTE_E1000_PMD),y)
+LDLIBS += -lrte_pmd_e1000
+endif
+
 ifeq ($(CONFIG_RTE_LIBRTE_PMD_RING),y)
 LDLIBS += -lrte_pmd_ring
 endif
-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 10/14] ixgbe: Convert to use of PMD_REGISTER_DRIVER and fix linking

2014-04-21 Thread Neil Horman 
convert the ixgbe pmd driver to use the PMD_REGISTER_DRIVER macro.
This means that the test applications now have no reference to the ixgbe library
when building DSO's and must specify its use on the command line with the -d
option.  Static linking will still initalize the driver automatically.

Signed-off-by: Neil Horman 
---
 examples/dpdk_qat/main.c|  6 --
 examples/vmdq_dcb/main.c|  4 
 lib/librte_ether/rte_ethdev.h   | 13 -
 lib/librte_pmd_ixgbe/ixgbe_ethdev.c | 12 ++--
 4 files changed, 10 insertions(+), 25 deletions(-)

diff --git a/examples/dpdk_qat/main.c b/examples/dpdk_qat/main.c
index 64cb610..741d3f3 100644
--- a/examples/dpdk_qat/main.c
+++ b/examples/dpdk_qat/main.c
@@ -696,12 +696,6 @@ MAIN(int argc, char **argv)
if (ret < 0)
return -1;

-   /* init driver */
-#ifdef RTE_LIBRTE_IXGBE_PMD
-   if (rte_ixgbe_pmd_init() < 0)
-   rte_panic("Cannot init ixgbe pmd\n");
-#endif
-
if (rte_eal_pci_probe() < 0)
rte_panic("Cannot probe PCI\n");

diff --git a/examples/vmdq_dcb/main.c b/examples/vmdq_dcb/main.c
index 844f334..1626ba5 100644
--- a/examples/vmdq_dcb/main.c
+++ b/examples/vmdq_dcb/main.c
@@ -455,10 +455,6 @@ MAIN(int argc, char *argv[])
if (ret < 0)
rte_exit(EXIT_FAILURE, "Invalid VMDQ argument\n");

-   if (rte_ixgbe_pmd_init() != 0 ||
-   rte_eal_pci_probe() != 0)
-   rte_exit(EXIT_FAILURE, "Error with NIC driver 
initialization\n");
-   
cores = rte_lcore_count();
if ((cores & (cores - 1)) != 0 || cores > 128) {
rte_exit(EXIT_FAILURE,"This program can only run on an even"
diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index 9e32f56..9621be7 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h
@@ -1316,15 +1316,6 @@ struct eth_driver {
 extern void rte_eth_driver_register(struct eth_driver *eth_drv);

 /**
- * The initialization function of the driver for 10Gbps Intel IXGBE
- * Ethernet devices.
- * Invoked once at EAL start time.
- * @return
- *   0 on success
- */
-extern int rte_ixgbe_pmd_init(void);
-
-/**
  * The initialization function of the driver for 10Gbps Intel IXGBE_VF
  * Ethernet devices.
  * Invoked once at EAL start time.
@@ -1375,10 +1366,6 @@ int rte_pmd_init_all(void)
int ret = -ENODEV;

 #ifdef RTE_LIBRTE_IXGBE_PMD
-   if ((ret = rte_ixgbe_pmd_init()) != 0) {
-   RTE_LOG(ERR, PMD, "Cannot init ixgbe PMD\n");
-   return (ret);
-   }
if ((ret = rte_ixgbevf_pmd_init()) != 0) {
RTE_LOG(ERR, PMD, "Cannot init ixgbevf PMD\n");
return (ret);
diff --git a/lib/librte_pmd_ixgbe/ixgbe_ethdev.c 
b/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
index a5a7f9a..efe899a 100644
--- a/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
+++ b/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
@@ -58,6 +58,7 @@
 #include 
 #include 
 #include 
+#include 

 #include "ixgbe_logs.h"
 #include "ixgbe/ixgbe_api.h"
@@ -955,8 +956,8 @@ static struct eth_driver rte_ixgbevf_pmd = {
  * Invoked once at EAL init time.
  * Register itself as the [Poll Mode] Driver of PCI IXGBE devices.
  */
-int
-rte_ixgbe_pmd_init(void)
+static int
+rte_ixgbe_pmd_init(const char *name __rte_unused, const char *params 
__rte_unused)
 {
PMD_INIT_FUNC_TRACE();

@@ -3060,3 +3061,10 @@ ixgbe_mirror_rule_reset(struct rte_eth_dev *dev, uint8_t 
rule_id)

return 0;
 }
+
+struct rte_driver rte_ixgbe_driver = {
+   .type = PMD_PDEV,
+   .init = rte_ixgbe_pmd_init,
+};
+
+PMD_REGISTER_DRIVER(rte_ixgbe_driver);
-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 11/14] ixgbevf: Convert to use of PMD_REGISTER_DRIVER and fix linking

2014-04-21 Thread Neil Horman 
convert the ixgbevf pmd driver to use the PMD_REGISTER_DRIVER macro.
This means that the test applications now have no reference to the ixgbevf 
library
when building DSO's and must specify its use on the command line with the -d
option.  Static linking will still initalize the driver automatically.

Signed-off-by: Neil Horman 
---
 lib/librte_ether/rte_ethdev.h   | 16 
 lib/librte_pmd_ixgbe/ixgbe_ethdev.c | 12 +---
 mk/rte.app.mk   |  8 
 3 files changed, 13 insertions(+), 23 deletions(-)

diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index 9621be7..ba535dc 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h
@@ -1316,15 +1316,6 @@ struct eth_driver {
 extern void rte_eth_driver_register(struct eth_driver *eth_drv);

 /**
- * The initialization function of the driver for 10Gbps Intel IXGBE_VF
- * Ethernet devices.
- * Invoked once at EAL start time.
- * @return
- *   0 on success
- */
-extern int rte_ixgbevf_pmd_init(void);
-
-/**
  * The initialization function of the driver for Qumranet virtio-net
  * Ethernet devices.
  * Invoked once at EAL start time.
@@ -1365,13 +1356,6 @@ int rte_pmd_init_all(void)
 {
int ret = -ENODEV;

-#ifdef RTE_LIBRTE_IXGBE_PMD
-   if ((ret = rte_ixgbevf_pmd_init()) != 0) {
-   RTE_LOG(ERR, PMD, "Cannot init ixgbevf PMD\n");
-   return (ret);
-   }
-#endif /* RTE_LIBRTE_IXGBE_PMD */
-
 #ifdef RTE_LIBRTE_VIRTIO_PMD
if ((ret = rte_virtio_pmd_init()) != 0) {
RTE_LOG(ERR, PMD, "Cannot init virtio PMD\n");
diff --git a/lib/librte_pmd_ixgbe/ixgbe_ethdev.c 
b/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
index efe899a..10dde5f 100644
--- a/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
+++ b/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
@@ -970,8 +970,8 @@ rte_ixgbe_pmd_init(const char *name __rte_unused, const 
char *params __rte_unuse
  * Invoked one at EAL init time.
  * Register itself as the [Virtual Poll Mode] Driver of PCI niantic devices.
  */
-int
-rte_ixgbevf_pmd_init(void)
+static int
+rte_ixgbevf_pmd_init(const char *name __rte_unused, const char *param 
__rte_unused)
 {
DEBUGFUNC("rte_ixgbevf_pmd_init");

@@ -3062,9 +3062,15 @@ ixgbe_mirror_rule_reset(struct rte_eth_dev *dev, uint8_t 
rule_id)
return 0;
 }

-struct rte_driver rte_ixgbe_driver = {
+static struct rte_driver rte_ixgbe_driver = {
.type = PMD_PDEV,
.init = rte_ixgbe_pmd_init,
 };

+static struct rte_driver rte_ixgbevf_driver = {
+   .type = PMD_PDEV,
+   .init = rte_ixgbevf_pmd_init,
+};
+
 PMD_REGISTER_DRIVER(rte_ixgbe_driver);
+PMD_REGISTER_DRIVER(rte_ixgbevf_driver);
diff --git a/mk/rte.app.mk b/mk/rte.app.mk
index a94830e..d8f7907 100644
--- a/mk/rte.app.mk
+++ b/mk/rte.app.mk
@@ -73,10 +73,6 @@ LDLIBS += -lrte_ivshmem
 endif
 endif

-ifeq ($(CONFIG_RTE_LIBRTE_IXGBE_PMD),y)
-LDLIBS += -lrte_pmd_ixgbe
-endif
-
 ifeq ($(CONFIG_RTE_LIBRTE_VIRTIO_PMD),y)
 LDLIBS += -lrte_pmd_virtio_uio
 endif
@@ -165,6 +161,10 @@ endif

 ifeq ($(RTE_BUILD_SHARED_LIB),n)

+ifeq ($(CONFIG_RTE_LIBRTE_IXGBE_PMD),y)
+LDLIBS += -lrte_pmd_ixgbe
+endif
+
 ifeq ($(CONFIG_RTE_LIBRTE_E1000_PMD),y)
 LDLIBS += -lrte_pmd_e1000
 endif
-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 12/14] virtio: Convert to use of PMD_REGISTER_DRIVER and fix linking

2014-04-21 Thread Neil Horman 
convert the virtio pmd driver to use the PMD_REGISTER_DRIVER macro.
This means that the test applications now have no reference to the virtio 
library
when building DSO's and must specify its use on the command line with the -d
option.  Static linking will still initalize the driver automatically.

Signed-off-by: Neil Horman 
---
 lib/librte_ether/rte_ethdev.h | 16 
 lib/librte_pmd_virtio/virtio_ethdev.c | 12 ++--
 mk/rte.app.mk |  8 
 3 files changed, 14 insertions(+), 22 deletions(-)

diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index ba535dc..7126f95 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h
@@ -1316,15 +1316,6 @@ struct eth_driver {
 extern void rte_eth_driver_register(struct eth_driver *eth_drv);

 /**
- * The initialization function of the driver for Qumranet virtio-net
- * Ethernet devices.
- * Invoked once at EAL start time.
- * @return
- *   0 on success
- */
-extern int rte_virtio_pmd_init(void);
-
-/**
  * The initialization function of the driver for VMware VMXNET3
  * Ethernet devices.
  * Invoked once at EAL start time.
@@ -1356,13 +1347,6 @@ int rte_pmd_init_all(void)
 {
int ret = -ENODEV;

-#ifdef RTE_LIBRTE_VIRTIO_PMD
-   if ((ret = rte_virtio_pmd_init()) != 0) {
-   RTE_LOG(ERR, PMD, "Cannot init virtio PMD\n");
-   return (ret);
-   }
-#endif /* RTE_LIBRTE_VIRTIO_PMD */
-
 #ifdef RTE_LIBRTE_VMXNET3_PMD
if ((ret = rte_vmxnet3_pmd_init()) != 0) {
RTE_LOG(ERR, PMD, "Cannot init vmxnet3 PMD\n");
diff --git a/lib/librte_pmd_virtio/virtio_ethdev.c 
b/lib/librte_pmd_virtio/virtio_ethdev.c
index f107161..611393e 100644
--- a/lib/librte_pmd_virtio/virtio_ethdev.c
+++ b/lib/librte_pmd_virtio/virtio_ethdev.c
@@ -50,6 +50,7 @@

 #include 
 #include 
+#include 

 #include "virtio_ethdev.h"
 #include "virtio_pci.h"
@@ -486,8 +487,8 @@ static struct eth_driver rte_virtio_pmd = {
  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
  * Returns 0 on success.
  */
-int
-rte_virtio_pmd_init(void)
+static int
+rte_virtio_pmd_init(const char *name __rte_unused, const char *param 
__rte_unused)
 {
rte_eth_driver_register(&rte_virtio_pmd);
return (0);
@@ -643,3 +644,10 @@ virtio_dev_info_get(struct rte_eth_dev *dev, struct 
rte_eth_dev_info *dev_info)
dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
 }
+
+static struct rte_driver rte_virtio_driver = {
+   .type = PMD_PDEV,
+   .init = rte_virtio_pmd_init,
+};
+
+PMD_REGISTER_DRIVER(rte_virtio_driver);
diff --git a/mk/rte.app.mk b/mk/rte.app.mk
index d8f7907..3f1386b 100644
--- a/mk/rte.app.mk
+++ b/mk/rte.app.mk
@@ -73,10 +73,6 @@ LDLIBS += -lrte_ivshmem
 endif
 endif

-ifeq ($(CONFIG_RTE_LIBRTE_VIRTIO_PMD),y)
-LDLIBS += -lrte_pmd_virtio_uio
-endif
-
 ifeq ($(CONFIG_RTE_LIBRTE_VMXNET3_PMD),y)
 LDLIBS += -lrte_pmd_vmxnet3_uio
 endif
@@ -161,6 +157,10 @@ endif

 ifeq ($(RTE_BUILD_SHARED_LIB),n)

+ifeq ($(CONFIG_RTE_LIBRTE_VIRTIO_PMD),y)
+LDLIBS += -lrte_pmd_virtio_uio
+endif
+
 ifeq ($(CONFIG_RTE_LIBRTE_IXGBE_PMD),y)
 LDLIBS += -lrte_pmd_ixgbe
 endif
-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 13/14] vmxnet3: Convert to use of PMD_REGISTER_DRIVER and fix linking

2014-04-21 Thread Neil Horman 
convert the vmxnet3 pmd driver to use the PMD_REGISTER_DRIVER macro.
This means that the test applications now have no reference to the vmxnet3 
library
when building DSO's and must specify its use on the command line with the -d
option.  Static linking will still initalize the driver automatically.

Signed-off-by: Neil Horman 
---
 lib/librte_ether/rte_ethdev.h   | 19 +--
 lib/librte_pmd_vmxnet3/vmxnet3_ethdev.c | 12 ++--
 mk/rte.app.mk   |  8 
 3 files changed, 15 insertions(+), 24 deletions(-)

diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index 7126f95..24288f8 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h
@@ -1316,16 +1316,6 @@ struct eth_driver {
 extern void rte_eth_driver_register(struct eth_driver *eth_drv);

 /**
- * The initialization function of the driver for VMware VMXNET3
- * Ethernet devices.
- * Invoked once at EAL start time.
- * @return
- *   0 on success
- */
-extern int rte_vmxnet3_pmd_init(void);
-
-
-/**
  * The initialization function of *all* supported and enabled drivers.
  * Right now, the following PMDs are supported:
  *  - igb
@@ -1345,14 +1335,7 @@ extern int rte_vmxnet3_pmd_init(void);
 static inline
 int rte_pmd_init_all(void)
 {
-   int ret = -ENODEV;
-
-#ifdef RTE_LIBRTE_VMXNET3_PMD
-   if ((ret = rte_vmxnet3_pmd_init()) != 0) {
-   RTE_LOG(ERR, PMD, "Cannot init vmxnet3 PMD\n");
-   return (ret);
-   }
-#endif /* RTE_LIBRTE_VMXNET3_PMD */
+   int ret = 0;

if (ret == -ENODEV)
RTE_LOG(ERR, PMD, "No PMD(s) are configured\n");
diff --git a/lib/librte_pmd_vmxnet3/vmxnet3_ethdev.c 
b/lib/librte_pmd_vmxnet3/vmxnet3_ethdev.c
index 8259cfe..c41032f 100644
--- a/lib/librte_pmd_vmxnet3/vmxnet3_ethdev.c
+++ b/lib/librte_pmd_vmxnet3/vmxnet3_ethdev.c
@@ -60,6 +60,7 @@
 #include 
 #include 
 #include 
+#include 

 #include "vmxnet3/vmxnet3_defs.h"

@@ -278,8 +279,8 @@ static struct eth_driver rte_vmxnet3_pmd = {
  * Invoked once at EAL init time.
  * Register itself as the [Poll Mode] Driver of Virtual PCI VMXNET3 devices.
  */
-int
-rte_vmxnet3_pmd_init(void)
+static int
+rte_vmxnet3_pmd_init(const char *name __rte_unused, const char *param 
__rte_unused)
 {
PMD_INIT_FUNC_TRACE();

@@ -763,3 +764,10 @@ vmxnet3_process_events(struct vmxnet3_hw *hw)

 }
 #endif
+
+static struct rte_driver rte_vmxnet3_driver = {
+   .type = PMD_PDEV,
+   .init = rte_vmxnet3_pmd_init,
+};
+
+PMD_REGISTER_DRIVER(rte_vmxnet3_driver);
diff --git a/mk/rte.app.mk b/mk/rte.app.mk
index 3f1386b..cfea6c6 100644
--- a/mk/rte.app.mk
+++ b/mk/rte.app.mk
@@ -73,10 +73,6 @@ LDLIBS += -lrte_ivshmem
 endif
 endif

-ifeq ($(CONFIG_RTE_LIBRTE_VMXNET3_PMD),y)
-LDLIBS += -lrte_pmd_vmxnet3_uio
-endif
-
 ifeq ($(CONFIG_RTE_LIBRTE_TIMER),y)
 LDLIBS += -lrte_timer
 endif
@@ -157,6 +153,10 @@ endif

 ifeq ($(RTE_BUILD_SHARED_LIB),n)

+ifeq ($(CONFIG_RTE_LIBRTE_VMXNET3_PMD),y)
+LDLIBS += -lrte_pmd_vmxnet3_uio
+endif
+
 ifeq ($(CONFIG_RTE_LIBRTE_VIRTIO_PMD),y)
 LDLIBS += -lrte_pmd_virtio_uio
 endif
-- 
1.8.3.1

[dpdk-dev] [PATCH 0/X v5 14/14] pmd: Remove rte_pmd_init_all

2014-04-21 Thread Neil Horman 
Now that we've converted all the pmds in dpdk to use the driver registration
macro, rte_pmd_init_all has become empty.  As theres no reason to keep it around
anymore, just remove it and fix up all the eample callers.

Signed-off-by: Neil Horman 
---
 app/test-pmd/testpmd.c |  3 ---
 app/test/test_kni.c|  5 
 examples/exception_path/main.c |  5 
 examples/ip_reassembly/main.c  |  4 
 examples/ipv4_frag/main.c  |  4 
 examples/ipv4_multicast/main.c |  4 
 examples/kni/main.c|  5 
 examples/l2fwd-ivshmem/host/host.c |  4 
 examples/l2fwd/main.c  |  4 
 examples/l3fwd-power/main.c|  4 
 examples/l3fwd-vf/main.c   |  4 
 examples/l3fwd/main.c  |  4 
 examples/link_status_interrupt/main.c  |  4 
 examples/load_balancer/init.c  |  6 -
 .../client_server_mp/shared/init_drivers.h |  2 +-
 examples/multi_process/l2fwd_fork/main.c   |  4 
 examples/multi_process/symmetric_mp/main.c |  2 --
 examples/netmap_compat/bridge/bridge.c |  4 
 examples/qos_meter/main.c  |  4 
 examples/qos_sched/init.c  |  4 
 examples/quota_watermark/qw/init.c |  5 
 examples/vhost/main.c  |  2 +-
 examples/vhost_xen/main.c  |  2 +-
 examples/vmdq/main.c   |  2 +-
 lib/librte_ether/rte_ethdev.h  | 27 --
 25 files changed, 4 insertions(+), 114 deletions(-)

diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
index 04dca57..7f2dcde 100644
--- a/app/test-pmd/testpmd.c
+++ b/app/test-pmd/testpmd.c
@@ -1755,9 +1755,6 @@ main(int argc, char** argv)
if (diag < 0)
rte_panic("Cannot init EAL\n");

-   if (rte_pmd_init_all())
-   rte_panic("Cannot init PMD\n");
-
if (rte_eal_pci_probe())
rte_panic("Cannot probe PCI\n");

diff --git a/app/test/test_kni.c b/app/test/test_kni.c
index e0fe44e..d37758b 100644
--- a/app/test/test_kni.c
+++ b/app/test/test_kni.c
@@ -509,11 +509,6 @@ test_kni(void)
printf("fail to create mempool for kni\n");
return -1;
}
-   ret = rte_pmd_init_all();
-   if (ret < 0) {
-   printf("fail to initialize PMD\n");
-   return -1;
-   }
ret = rte_eal_pci_probe();
if (ret < 0) {
printf("fail to probe PCI devices\n");
diff --git a/examples/exception_path/main.c b/examples/exception_path/main.c
index 0bc149d..d73b413 100644
--- a/examples/exception_path/main.c
+++ b/examples/exception_path/main.c
@@ -566,11 +566,6 @@ main(int argc, char** argv)
return -1;
}

-   /* Initialise PMD driver(s) */
-   ret = rte_pmd_init_all();
-   if (ret < 0)
-   FATAL_ERROR("Could not probe PMD (%d)", ret);
-
/* Scan PCI bus for recognised devices */
ret = rte_eal_pci_probe();
if (ret < 0)
diff --git a/examples/ip_reassembly/main.c b/examples/ip_reassembly/main.c
index 4880a5f..bafa8d9 100644
--- a/examples/ip_reassembly/main.c
+++ b/examples/ip_reassembly/main.c
@@ -1553,10 +1553,6 @@ MAIN(int argc, char **argv)
rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");


-   /* init driver(s) */
-   if (rte_pmd_init_all() < 0)
-   rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
-
if (rte_eal_pci_probe() < 0)
rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");

diff --git a/examples/ipv4_frag/main.c b/examples/ipv4_frag/main.c
index 93664c8..329f2ce 100644
--- a/examples/ipv4_frag/main.c
+++ b/examples/ipv4_frag/main.c
@@ -608,10 +608,6 @@ MAIN(int argc, char **argv)
if (pool_indirect == NULL)
rte_panic("Cannot init indirect mbuf pool\n");

-   /* init driver */
-   if (rte_pmd_init_all() < 0)
-   rte_panic("Cannot init PMD\n");
-
if (rte_eal_pci_probe() < 0)
rte_panic("Cannot probe PCI\n");

diff --git a/examples/ipv4_multicast/main.c b/examples/ipv4_multicast/main.c
index 3bd37e4..da4e09d 100644
--- a/examples/ipv4_multicast/main.c
+++ b/examples/ipv4_multicast/main.c
@@ -753,10 +753,6 @@ MAIN(int argc, char **argv)
if (clone_pool == NULL)
rte_exit(EXIT_FAILURE, "Cannot init clone mbuf pool\n");

-   /* init driver */
-   if (rte_pmd_init_all() < 0)
-   rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
-
if (rte_eal_pci_probe() < 0)
rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");

diff --git a/examples/kni/main.c b/exampl

[dpdk-dev] [PATCH v5 00/14] dpdk: Separate compile time linkage between eal lib and pmd's

2014-04-21 Thread Neil Horman 
On Mon, Apr 21, 2014 at 10:59:25AM -0400, Neil Horman wrote:
> Disconnect compile time linkage between eal library / applications and pmd's
> 
> I noticed that, while tinkering with dpdk, building for shared libraries still
> resulted in all the test applications linking to all the built pmd's, despite
> not actually needing them all.  We are able to tell an application at run time
> (via the -d/--blacklist/--whitelist/--vdev options) which pmd's we want to 
> use,
> and so have no need to link them at all. The only reason they get pulled in is
> because rte_eal_non_pci_init_etherdev and rte_pmd_init_all contain static 
> lists
> to the individual pmd init functions. The result is that, even when building 
> as
> DSO's, we have to load all the pmd libraries, which is space inefficient and
> defeating of some of the purpose of shared objects.
> 
> To correct this, I developed this patch series, which introduces a new macro,
> PMD_REGISTER_DRIVER, which wraps up Oliviers work using constructors on the
> virtual device pmds, then expands on it to include the physical device pmds,
> allowing us to break linkages between dpdk applications and pmd's almost
> entirely (save for the ring and xenvirt drivers, which have additional api's
> outside of the standard dpdk code that we need to further fix).  This also
> allows us to completely remove the rte_pmd_init_all routine, hiding its 
> function
> internally to the rte_eal_init path.
> 
> I've tested this feature using the igb and pcap pmd's, both statically and
> dynamically linked with the test and testpmd sample applications, and it seems
> to work well.
> 
> Note, I encountered  a few bugs along the way, which I fixed and noted in the
> series.
> 
> Regards
> Neil
> 
> Change Notes:
> 
> 1) Reposted the entire series.  Recent changes permeated accross several
> patches, and since a few patches already had multiple versions, I've reposted
> the entire series and bumped the version number to 5, whcih skips a few
> versions, but ensures this series is seen as the most recent.
> 
> 2) Merged the PMD_REGISTER_DRIVER macro into rte_dev.h, which is a better
> location for it.  Required removing include of rte_pmd.h across most of the
> patches in the series.
> 
> 3) Cleaned up various leftover "virtual" comments in the driver registration 
> api
> that no longer belong there after making it generic
> 
> 4) Fixed the LINK_USING_CC check in rte.lib.mk so it works like other uses
> 
> 5) Fixed CPU_LDFLAGS conversion to use -Wl in case we link with gcc.
> 
> 
> 

Shoot, I fat fingered in a 0/X in all the patch prefixes.  sorry about that.
Neil

[dpdk-dev] [PATCH] malloc: fix rte_free run time in O(n) free blocks

2014-04-21 Thread Sanford, Robert 
Here is our proposed patch:

Lib rte_malloc stores free blocks in singly-linked lists.
This results in O(n), i.e., poor performance when freeing memory
to a heap that contains thousands of free blocks.

A simple solution is to store free blocks on doubly-linked lists.
- Space-wise, we add one new pointer to struct malloc_elem.
 This does not affect its size, since it is cache-aligned and
 currently has room for another pointer.
- We perform all free list adds and deletes in functions
 malloc_elem_add_to_free_list and remove_from_free_list.
- We clean up internal malloc functions that push or
 return pointers to previous blocks in the free list.


---
 lib/librte_malloc/malloc_elem.c |   82
+++---
 lib/librte_malloc/malloc_elem.h |6 ++-
 lib/librte_malloc/malloc_heap.c |   19 +++--
 3 files changed, 60 insertions(+), 47 deletions(-)

diff --git a/lib/librte_malloc/malloc_elem.c
b/lib/librte_malloc/malloc_elem.c
index f0da640..9a14e23 100644
--- a/lib/librte_malloc/malloc_elem.c
+++ b/lib/librte_malloc/malloc_elem.c
@@ -60,7 +60,7 @@ malloc_elem_init(struct malloc_elem *elem,
 {
elem->heap = heap;
elem->mz = mz;
-   elem->prev = elem->next_free = NULL;
+   elem->prev = elem->next_free = elem->prev_free = NULL;
elem->state = ELEM_FREE;
elem->size = size;
elem->pad = 0;
@@ -125,14 +125,58 @@ split_elem(struct malloc_elem *elem, struct
malloc_elem *split_pt)
 }

 /*
+ * Add the specified element to its heap's free list.
+ */
+void
+malloc_elem_add_to_free_list(struct malloc_elem *elem)
+{
+   elem->state = ELEM_FREE;
+   elem->prev_free = NULL;
+   if (!!(elem->next_free = elem->heap->free_head))
+   elem->next_free->prev_free = elem;
+   elem->heap->free_head = elem;
+}
+
+/*
+ * Remove the specified element from its heap's free list.
+ */
+static inline void
+remove_from_free_list(struct malloc_elem *elem)
+{
+   struct malloc_elem *next_free_elem = elem->next_free;
+   struct malloc_elem *prev_free_elem = elem->prev_free;
+
+   if (!!next_free_elem) {
+#ifdef RTE_LIBRTE_MALLOC_DEBUG
+   RTE_VERIFY(next_free_elem->prev_free == elem);
+#endif
+   next_free_elem->prev_free = prev_free_elem;
+   }
+
+   if (!!prev_free_elem) {
+#ifdef RTE_LIBRTE_MALLOC_DEBUG
+   RTE_VERIFY(prev_free_elem->next_free == elem);
+#endif
+   prev_free_elem->next_free = next_free_elem;
+   }
+   else {
+   struct malloc_heap *heap = elem->heap;
+
+#ifdef RTE_LIBRTE_MALLOC_DEBUG
+   RTE_VERIFY(heap->free_head == elem);
+#endif
+   heap->free_head = next_free_elem;
+   }
+}
+
+/*
  * reserve a block of data in an existing malloc_elem. If the malloc_elem
  * is much larger than the data block requested, we split the element in
two.
  * This function is only called from malloc_heap_alloc so parameter
checking
  * is not done here, as it's done there previously.
  */
 struct malloc_elem *
-malloc_elem_alloc(struct malloc_elem *elem, size_t size,
-   unsigned align, struct malloc_elem *prev_free)
+malloc_elem_alloc(struct malloc_elem *elem, size_t size, unsigned align)
 {
struct malloc_elem *new_elem = elem_start_pt(elem, size, align);
const unsigned old_elem_size = (uintptr_t)new_elem - (uintptr_t)elem;
@@ -151,10 +195,7 @@ malloc_elem_alloc(struct malloc_elem *elem, size_t
size,
set_header(new_elem);
}
/* remove element from free list */
-   if (prev_free == NULL)
-   elem->heap->free_head = elem->next_free;
-   else
-   prev_free->next_free = elem->next_free;
+   remove_from_free_list(elem);

return new_elem;
}
@@ -182,25 +223,6 @@ join_elem(struct malloc_elem *elem1, struct
malloc_elem *elem2)
 }

 /*
- * scan the free list, and remove the request element from that
- * free list. (Free list to scan is got from heap pointer in element)
- */
-static inline void
-remove_from_free_list(struct malloc_elem *elem)
-{
-   if (elem == elem->heap->free_head)
-   elem->heap->free_head = elem->next_free;
-   else{
-   struct malloc_elem *prev_free = elem->heap->free_head;
-   while (prev_free && prev_free->next_free != elem)
-   prev_free = prev_free->next_free;
-   if (!prev_free)
-   rte_panic("Corrupted free list\n");
-   prev_free->next_free = elem->next_free;
-   }
-}
-
-/*
  * free a malloc_elem block by adding it to the free list. If the
  * blocks either immediately before or immediately after newly freed block
  * are also free, the blocks are merged together.
@@ -226,9 +248,7 @@ malloc_elem_free(struct malloc_elem *elem)
join_elem(elem->prev, elem);
/* otherwise add ourselves to the free list */

[dpdk-dev] [PATCH v5 00/14] dpdk: Separate compile time linkage between eal lib and pmd's

2014-04-21 Thread Stephen Hemminger 
On Mon, 21 Apr 2014 10:59:25 -0400
Neil Horman  wrote:

> Disconnect compile time linkage between eal library / applications and pmd's
> 
> I noticed that, while tinkering with dpdk, building for shared libraries still
> resulted in all the test applications linking to all the built pmd's, despite
> not actually needing them all.  We are able to tell an application at run time
> (via the -d/--blacklist/--whitelist/--vdev options) which pmd's we want to 
> use,
> and so have no need to link them at all. The only reason they get pulled in is
> because rte_eal_non_pci_init_etherdev and rte_pmd_init_all contain static 
> lists
> to the individual pmd init functions. The result is that, even when building 
> as
> DSO's, we have to load all the pmd libraries, which is space inefficient and
> defeating of some of the purpose of shared objects.
> 
> To correct this, I developed this patch series, which introduces a new macro,
> PMD_REGISTER_DRIVER, which wraps up Oliviers work using constructors on the
> virtual device pmds, then expands on it to include the physical device pmds,
> allowing us to break linkages between dpdk applications and pmd's almost
> entirely (save for the ring and xenvirt drivers, which have additional api's
> outside of the standard dpdk code that we need to further fix).  This also
> allows us to completely remove the rte_pmd_init_all routine, hiding its 
> function
> internally to the rte_eal_init path.
> 
> I've tested this feature using the igb and pcap pmd's, both statically and
> dynamically linked with the test and testpmd sample applications, and it seems
> to work well.
> 
> Note, I encountered  a few bugs along the way, which I fixed and noted in the
> series.
> 
> Regards
> Neil
> 
> Change Notes:
> 
> 1) Reposted the entire series.  Recent changes permeated accross several
> patches, and since a few patches already had multiple versions, I've reposted
> the entire series and bumped the version number to 5, whcih skips a few
> versions, but ensures this series is seen as the most recent.
> 
> 2) Merged the PMD_REGISTER_DRIVER macro into rte_dev.h, which is a better
> location for it.  Required removing include of rte_pmd.h across most of the
> patches in the series.
> 
> 3) Cleaned up various leftover "virtual" comments in the driver registration 
> api
> that no longer belong there after making it generic
> 
> 4) Fixed the LINK_USING_CC check in rte.lib.mk so it works like other uses
> 
> 5) Fixed CPU_LDFLAGS conversion to use -Wl in case we link with gcc.

I am ok with build device drivers as libraries, but there can be a
performance impact.  For performance, we link with link-time optimization
and that is not possible as shared library. Haven't measured but at least
in the kernel, modules add a performance tax because they aren't in same
memory region and therefore cause a TLB miss. Might have same impact in
user mode.

You might be able to get some of this back by compiling with 
link-time-optimization
on a device at a time, and
also use the flags to tell compiler that only certain entry points are
shared and need tob e global.

[dpdk-dev] [PATCH v5 00/14] dpdk: Separate compile time linkage between eal lib and pmd's

2014-04-21 Thread Neil Horman 
On Mon, Apr 21, 2014 at 01:10:00PM -0700, Stephen Hemminger wrote:
> On Mon, 21 Apr 2014 10:59:25 -0400
> Neil Horman  wrote:
> 
> > Disconnect compile time linkage between eal library / applications and pmd's
> > 
> > I noticed that, while tinkering with dpdk, building for shared libraries 
> > still
> > resulted in all the test applications linking to all the built pmd's, 
> > despite
> > not actually needing them all.  We are able to tell an application at run 
> > time
> > (via the -d/--blacklist/--whitelist/--vdev options) which pmd's we want to 
> > use,
> > and so have no need to link them at all. The only reason they get pulled in 
> > is
> > because rte_eal_non_pci_init_etherdev and rte_pmd_init_all contain static 
> > lists
> > to the individual pmd init functions. The result is that, even when 
> > building as
> > DSO's, we have to load all the pmd libraries, which is space inefficient and
> > defeating of some of the purpose of shared objects.
> > 
> > To correct this, I developed this patch series, which introduces a new 
> > macro,
> > PMD_REGISTER_DRIVER, which wraps up Oliviers work using constructors on the
> > virtual device pmds, then expands on it to include the physical device pmds,
> > allowing us to break linkages between dpdk applications and pmd's almost
> > entirely (save for the ring and xenvirt drivers, which have additional api's
> > outside of the standard dpdk code that we need to further fix).  This also
> > allows us to completely remove the rte_pmd_init_all routine, hiding its 
> > function
> > internally to the rte_eal_init path.
> > 
> > I've tested this feature using the igb and pcap pmd's, both statically and
> > dynamically linked with the test and testpmd sample applications, and it 
> > seems
> > to work well.
> > 
> > Note, I encountered  a few bugs along the way, which I fixed and noted in 
> > the
> > series.
> > 
> > Regards
> > Neil
> > 
> > Change Notes:
> > 
> > 1) Reposted the entire series.  Recent changes permeated accross several
> > patches, and since a few patches already had multiple versions, I've 
> > reposted
> > the entire series and bumped the version number to 5, whcih skips a few
> > versions, but ensures this series is seen as the most recent.
> > 
> > 2) Merged the PMD_REGISTER_DRIVER macro into rte_dev.h, which is a better
> > location for it.  Required removing include of rte_pmd.h across most of the
> > patches in the series.
> > 
> > 3) Cleaned up various leftover "virtual" comments in the driver 
> > registration api
> > that no longer belong there after making it generic
> > 
> > 4) Fixed the LINK_USING_CC check in rte.lib.mk so it works like other uses
> > 
> > 5) Fixed CPU_LDFLAGS conversion to use -Wl in case we link with gcc.
> 
> I am ok with build device drivers as libraries, but there can be a
> performance impact.  For performance, we link with link-time optimization
> and that is not possible as shared library. Haven't measured but at least
> in the kernel, modules add a performance tax because they aren't in same
> memory region and therefore cause a TLB miss. Might have same impact in
> user mode.
> 
Possibly, but that seems like a build time decision.  Performance numbers are
definately good for guidance here, but one way or another if you're not ok with
the impact, you configure to build statically, and go on about your day.  We're
not preventing that here, just making the DSO build case work a little more
sanely, so that the core libraries and applications don't have to link against
the DSO's directly.

> You might be able to get some of this back by compiling with 
> link-time-optimization
> on a device at a time, and
> also use the flags to tell compiler that only certain entry points are
> shared and need tob e global.
> 
That seems like a fine idea.  In fact with this patch, a typical pmd would need
no global entry points, as the constructor registers the init routine (which can
be static), and the init routine registers the ethernet device driver.  We could
probably (with this changeset) create a boilerplate linker script for the pmds
that marked all the symbols except constructors as local (xenvirt and ring are
notable exceptions to this rule, but those can likely be fixed up).

Regardless, I'd like to address that in a separate patch, as I think this series
is doing enough on its own.  It a good idea for optimization though.

Neil