Skip to contentSkip to conversation

ethdev: missing typecast from void in eth_dev_pci_specific_init

PATCH In review David Christensen 7 participants 19 messages Apr 10, 2019 – Apr 16, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ethdev: missing typecast from void in eth_dev_pci_specific_init

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device.

---
 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
index 23257e9..a325311 100644
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 
 	if (!pci_dev)
 		return -ENODEV;
Signature · 1 line
1.8.3.1
#1 lore raw

10/04/2019 22:16, David Christensen:

Quoted 7 lines

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device.

This is a duplicate of this patch:

https://patches.dpdk.org/patch/52505/

You are from 2 different Linux teams at IBM,
you hit the same issue at the same time,
you both miss to give an explanation,
funny.

About collaborating inside IBM, please check this email:
http://mails.dpdk.org/archives/dev/2019-April/129915.html

#2 lore raw
Quoted 8 lines

This is a duplicate of this patch:

https://patches.dpdk.org/patch/52505/

You are from 2 different Linux teams at IBM,
you hit the same issue at the same time,
you both miss to give an explanation,
funny.

We both work for infrastructure teams who are supporting a project team that is developing a DPDK application that runs across multiple CPU architectures and encountered the error.

The error comes from g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0:

[CXX] g++ -o utils/pcap_handle.o -c utils/pcap_handle.cc -g3 -ggdb3 -mcpu=native -mtune=native -isystem /home/rchirra/gen/io/ppc64include -std=gnu++11 -flax-vector-conversions -Werror -isystem /home/rchirra/gen/io

In file included from drivers/pmd.cc:7:0:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:
In function ‘int eth_dev_pci_specific_init(rte_eth_dev*, void*)’:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:75:35: error: invalid conversion from ‘void*’ to ‘rte_pci_device*’ [-fpermissive]

  struct rte_pci_device *pci_dev = bus_device;
                                   ^~~~~~~~~~

make[1]: *** [drivers/pmd.o] Error 1

Do you believe that more permissive compiler flags are the better answer?

Dave

#3 lore raw

10/04/2019 22:58, David Christensen:

Quoted 25 lines
Quoted 8 lines

This is a duplicate of this patch:

https://patches.dpdk.org/patch/52505/

You are from 2 different Linux teams at IBM,
you hit the same issue at the same time,
you both miss to give an explanation,
funny.

We both work for infrastructure teams who are supporting a project team that is developing a DPDK application that runs across multiple CPU architectures and encountered the error.

The error comes from g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0:

[CXX] g++ -o utils/pcap_handle.o -c utils/pcap_handle.cc -g3
-ggdb3 -mcpu=native -mtune=native -isystem /home/rchirra/gen/io/ppc64include -std=gnu++11 -flax-vector-conversions -Werror -isystem /home/rchirra/gen/io

In file included from drivers/pmd.cc:7:0:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:
In function ‘int eth_dev_pci_specific_init(rte_eth_dev*, void*)’:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:75:35:
error: invalid conversion from ‘void*’ to ‘rte_pci_device*’ [-fpermissive]

Quoted 5 lines
   struct rte_pci_device *pci_dev = bus_device;
                                    ^~~~~~~~~~

make[1]: *** [drivers/pmd.o] Error 1

Do you believe that more permissive compiler flags are the better answer?

No, we should not force any compiler flag of the application. We are even trying to support pedantic flags in the app. So it seems we must fix it.

We just need to change the title to mention it fixes the build for C++, and add the error message in the body.

#4 lore raw
Quoted 30 lines · Thomas Monjalon

10/04/2019 22:58, David Christensen:

Quoted 29 lines
Quoted 8 lines

This is a duplicate of this patch:

https://patches.dpdk.org/patch/52505/

You are from 2 different Linux teams at IBM,
you hit the same issue at the same time,
you both miss to give an explanation,
funny.

We both work for infrastructure teams who are supporting a project team that is developing a DPDK application that runs across multiple CPU architectures and encountered the error.

The error comes from g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0:

[CXX] g++ -o utils/pcap_handle.o -c utils/pcap_handle.cc -g3
-ggdb3 -mcpu=native -mtune=native -isystem /home/rchirra/gen/io/ppc64include -std=gnu++11 -flax-vector-conversions -Werror -isystem /home/rchirra/gen/io

In file included from drivers/pmd.cc:7:0:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:
In function ‘int eth_dev_pci_specific_init(rte_eth_dev*, void*)’:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:75:35:
error: invalid conversion from ‘void*’ to ‘rte_pci_device*’ [-fpermissive]

   struct rte_pci_device *pci_dev = bus_device;
                                    ^~~~~~~~~~

make[1]: *** [drivers/pmd.o] Error 1

Quoted 9 lines
Quoted 1 line

Do you believe that more permissive compiler flags are the better answer?

No, we should not force any compiler flag of the application. We are even trying to support pedantic flags in the app. So it seems we must fix it.

We just need to change the title to mention it fixes the build for C++, and add the error message in the body.

I agree our public headers should be usable from c++ applications, but this header is not a public header.

Is the DPDK driver compiled with c++ in this case?
If so do we support (or have a target for support) that DPDK library itself can be compiled by c++ compiler?

#5 lore raw
In reply to David Christensen
Quoted 16 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device.

---
 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +0 -0
index 23257e9..a325311 100644

lib/librte_ethdev/rte_ethdev_pci.h line 76

Quoted 6 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 

Is this needed for building some C++ apps that are including the header file (directly, or indirectly), because for pure C, "void *" types should be assignable to any other pointer type without casting?

/Bruce

#6 lore raw
Quoted 28 lines · On Wed, 10 Apr 2019 22:00:18 +0100 Bruce Richardson…
Quoted 16 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device.

---
 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +0 -0
index 23257e9..a325311 100644

lib/librte_ethdev/rte_ethdev_pci.h line 76

Quoted 6 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 

Is this needed for building some C++ apps that are including the header file (directly, or indirectly), because for pure C, "void *" types should be assignable to any other pointer type without casting?

Quoted 1 line

/Bruce

Another example of Why the Hell is this inline?

#7 lore raw
Quoted 31 lines · Stephen Hemminger
Quoted 28 lines · On Wed, 10 Apr 2019 22:00:18 +0100 Bruce Richardson…
Quoted 16 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device.

---
 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +0 -0
index 23257e9..a325311 100644

lib/librte_ethdev/rte_ethdev_pci.h line 76

Quoted 6 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 

Is this needed for building some C++ apps that are including the header file (directly, or indirectly), because for pure C, "void *" types should be assignable to any other pointer type without casting?

Quoted 4 lines
Quoted 1 line

/Bruce

Another example of Why the Hell is this inline?

It has been done inline intentionally at the time as far as remember, this header is for drivers not for applications, it has helper functions.

The common code from drivers related to the bus put into header files, so the code itself belongs to drivers not ethdev and reduces duplicates in them.

#8 lore raw
Quoted 44 lines

-----Original Message-----
From: dev [mailto:[email protected]] On Behalf Of Ferruh Yigit
Sent: Friday, April 12, 2019 6:09 PM
To: Stephen Hemminger <[email protected]>; Richardson, Bruce <[email protected]>
Cc: David Christensen <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [dpdk-dev] [PATCH] ethdev: missing typecast from void in eth_dev_pci_specific_init

Quoted 33 lines · Stephen Hemminger
Quoted 29 lines · On Wed, 10 Apr 2019 22:00:18 +0100 Bruce Richardson…
Quoted 16 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device.

---
 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +0 -0
index 23257e9..a325311 100644

lib/librte_ethdev/rte_ethdev_pci.h line 76

Quoted 6 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 

Is this needed for building some C++ apps that are including the header file (directly, or indirectly), because for pure C, "void *" types should be assignable to any other pointer type without casting?

/Bruce

Quoted 8 lines
Quoted 2 lines

Another example of Why the Hell is this inline?

It has been done inline intentionally at the time as far as remember, this header is for drivers not for applications, it has helper functions.

The common code from drivers related to the bus put into header files, so the code itself belongs to drivers not ethdev and reduces duplicates in them.

Ok that's the common code used by the drivers...
But why it still can't be in .c file?
Konstantin

#9 lore raw
Quoted 47 lines · Konstantin
Quoted 47 lines

-----Original Message-----
From: dev [mailto:[email protected]] On Behalf Of Ferruh Yigit
Sent: Friday, April 12, 2019 6:09 PM
To: Stephen Hemminger <[email protected]>; Richardson, Bruce <[email protected]>
Cc: David Christensen <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [dpdk-dev] [PATCH] ethdev: missing typecast from void in eth_dev_pci_specific_init

Quoted 35 lines · Stephen Hemminger
Quoted 28 lines · On Wed, 10 Apr 2019 22:00:18 +0100 Bruce Richardson…
Quoted 16 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device.

---
 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +0 -0
index 23257e9..a325311 100644

lib/librte_ethdev/rte_ethdev_pci.h line 76

Quoted 6 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 

Is this needed for building some C++ apps that are including the header file (directly, or indirectly), because for pure C, "void *" types should be assignable to any other pointer type without casting?

Quoted 1 line

/Bruce

Another example of Why the Hell is this inline?

Quoted 8 lines
Quoted 5 lines

It has been done inline intentionally at the time as far as remember, this header is for drivers not for applications, it has helper functions.

The common code from drivers related to the bus put into header files, so the code itself belongs to drivers not ethdev and reduces duplicates in them.

Ok that's the common code used by the drivers...
But why it still can't be in .c file?

When it is in .c file, it will be either in ethdev library, single location in .c file and binary file, but location is not exactly right, because code belongs to drivers.
Or code should be in .c files of each drivers, this will be code duplication.

Having in .h file makes code in single place, but when compiled code will be in each driver object file/ library.

Of course it works when put into a .c file in ehtdev, but bus (pci and vdev) related code are not belongs to ethdev library and I believe shouldn't be part of ethdev binary. And those bus helper headers are only for drivers to include, so having inline shouldn't be a problem at all because there is not stability concern in that interface.

#10 lore raw
Quoted 62 lines · Ferruh Yigit
Quoted 47 lines · Konstantin
Quoted 47 lines

-----Original Message-----
From: dev [mailto:[email protected]] On Behalf Of Ferruh Yigit
Sent: Friday, April 12, 2019 6:09 PM
To: Stephen Hemminger <[email protected]>; Richardson, Bruce <[email protected]>
Cc: David Christensen <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [dpdk-dev] [PATCH] ethdev: missing typecast from void in eth_dev_pci_specific_init

Quoted 35 lines · Stephen Hemminger
Quoted 28 lines · On Wed, 10 Apr 2019 22:00:18 +0100 Bruce Richardson…
Quoted 16 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device.

---
 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +0 -0
index 23257e9..a325311 100644

lib/librte_ethdev/rte_ethdev_pci.h line 76

Quoted 6 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 

Is this needed for building some C++ apps that are including the header file (directly, or indirectly), because for pure C, "void *" types should be assignable to any other pointer type without casting?

Quoted 1 line

/Bruce

Another example of Why the Hell is this inline?

Quoted 8 lines
Quoted 5 lines

It has been done inline intentionally at the time as far as remember, this header is for drivers not for applications, it has helper functions.

The common code from drivers related to the bus put into header files, so the code itself belongs to drivers not ethdev and reduces duplicates in them.

Ok that's the common code used by the drivers...
But why it still can't be in .c file?

When it is in .c file, it will be either in ethdev library, single location in .c file and binary file, but location is not exactly right, because code belongs to drivers.
Or code should be in .c files of each drivers, this will be code duplication.

Quoted 9 lines

Having in .h file makes code in single place, but when compiled code will be in each driver object file/ library.

Of course it works when put into a .c file in ehtdev, but bus (pci and vdev) related code are not belongs to ethdev library and I believe shouldn't be part of ethdev binary. And those bus helper headers are only for drivers to include, so having inline shouldn't be a problem at all because there is not stability concern in that interface.

btw, if you put those into .c file in ethdev, you will be creating a dependency from ethdev to bus code, to all available buses which will make impossible to disable any bus type if you use ethdev.

#11 lore raw
Quoted 73 lines · On Fri, 12 Apr 2019 18:29:46 +0100 Ferruh Yigit <[email protected]> wrote:
Quoted 62 lines · Ferruh Yigit
Quoted 47 lines · Konstantin
Quoted 47 lines

-----Original Message-----
From: dev [mailto:[email protected]] On Behalf Of Ferruh Yigit
Sent: Friday, April 12, 2019 6:09 PM
To: Stephen Hemminger <[email protected]>; Richardson, Bruce <[email protected]>
Cc: David Christensen <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [dpdk-dev] [PATCH] ethdev: missing typecast from void in eth_dev_pci_specific_init

Quoted 35 lines · Stephen Hemminger
Quoted 28 lines · On Wed, 10 Apr 2019 22:00:18 +0100 Bruce Richardson…
Quoted 16 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device.

---
 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +0 -0
index 23257e9..a325311 100644

lib/librte_ethdev/rte_ethdev_pci.h line 76

Quoted 6 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 

Is this needed for building some C++ apps that are including the header file (directly, or indirectly), because for pure C, "void *" types should be assignable to any other pointer type without casting?

Quoted 1 line

/Bruce

Another example of Why the Hell is this inline?

Quoted 8 lines
Quoted 5 lines

It has been done inline intentionally at the time as far as remember, this header is for drivers not for applications, it has helper functions.

The common code from drivers related to the bus put into header files, so the code itself belongs to drivers not ethdev and reduces duplicates in them.

Ok that's the common code used by the drivers...
But why it still can't be in .c file?

When it is in .c file, it will be either in ethdev library, single location in .c file and binary file, but location is not exactly right, because code belongs to drivers.
Or code should be in .c files of each drivers, this will be code duplication.

Quoted 9 lines

Having in .h file makes code in single place, but when compiled code will be in each driver object file/ library.

Of course it works when put into a .c file in ehtdev, but bus (pci and vdev) related code are not belongs to ethdev library and I believe shouldn't be part of ethdev binary. And those bus helper headers are only for drivers to include, so having inline shouldn't be a problem at all because there is not stability concern in that interface.

Quoted 3 lines

btw, if you put those into .c file in ethdev, you will be creating a dependency from ethdev to bus code, to all available buses which will make impossible to disable any bus type if you use ethdev.

The problem I see is rte_ethdev_pci.h, it should be headers only and then put code rte_ethdev_pci.c

#12 lore raw
Quoted 76 lines · Stephen Hemminger
Quoted 73 lines · On Fri, 12 Apr 2019 18:29:46 +0100 Ferruh Yigit <[email protected]> wrote:
Quoted 71 lines · Ferruh Yigit
Quoted 47 lines · Konstantin
Quoted 47 lines

-----Original Message-----
From: dev [mailto:[email protected]] On Behalf Of Ferruh Yigit
Sent: Friday, April 12, 2019 6:09 PM
To: Stephen Hemminger <[email protected]>; Richardson, Bruce <[email protected]>
Cc: David Christensen <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [dpdk-dev] [PATCH] ethdev: missing typecast from void in eth_dev_pci_specific_init

Quoted 35 lines · Stephen Hemminger
Quoted 28 lines · On Wed, 10 Apr 2019 22:00:18 +0100 Bruce Richardson…
Quoted 16 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device.

---
 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +0 -0
index 23257e9..a325311 100644

lib/librte_ethdev/rte_ethdev_pci.h line 76

Quoted 6 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 

Is this needed for building some C++ apps that are including the header file (directly, or indirectly), because for pure C, "void *" types should be assignable to any other pointer type without casting?

Quoted 1 line

/Bruce

Another example of Why the Hell is this inline?

Quoted 8 lines
Quoted 5 lines

It has been done inline intentionally at the time as far as remember, this header is for drivers not for applications, it has helper functions.

The common code from drivers related to the bus put into header files, so the code itself belongs to drivers not ethdev and reduces duplicates in them.

Ok that's the common code used by the drivers...
But why it still can't be in .c file?

When it is in .c file, it will be either in ethdev library, single location in .c file and binary file, but location is not exactly right, because code belongs to drivers.
Or code should be in .c files of each drivers, this will be code duplication.

Having in .h file makes code in single place, but when compiled code will be in each driver object file/ library.

Of course it works when put into a .c file in ehtdev, but bus (pci and vdev) related code are not belongs to ethdev library and I believe shouldn't be part of ethdev binary. And those bus helper headers are only for drivers to include, so having inline shouldn't be a problem at all because there is not stability concern in that interface.

Quoted 7 lines
Quoted 3 lines

btw, if you put those into .c file in ethdev, you will be creating a dependency from ethdev to bus code, to all available buses which will make impossible to disable any bus type if you use ethdev.

The problem I see is rte_ethdev_pci.h, it should be headers only and then put code rte_ethdev_pci.c

Where this 'rte_ethdev_pci.c' should be? Because of reasons explained above, ehtdev is not good place.
Perhaps a 'common' folder for net drivers may work, create a 'rte_ethdev_pci.o' and link it with relevant drivers.?

#13 lore raw

PATCH v2 1/1 ethdev: missing typecast causes C++ build error

In reply to David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device. This causes build issues in the GNU C++ compiler.

[CXX] g++ -o utils/pcap_handle.o -c utils/pcap_handle.cc -g3·
-ggdb3 -mcpu=native -mtune=native -isystem· /home/rchirra/gen/io/ppc64include -std=gnu++11 -flax-vector-conversions· -Werror -isystem /home/rchirra/gen/io

In file included from drivers/pmd.cc:7:0:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:
In function ‘int eth_dev_pci_specific_init(rte_eth_dev*, void*)’:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:75:35:·
error: invalid conversion from ‘void*’ to ‘rte_pci_device*’ [-fpermissive]

  struct rte_pci_device *pci_dev = bus_device;
                                   ^~~~~~~~~~

make[1]: *** [drivers/pmd.o] Error 1

$ g++ --version
g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0

---
v2:
* Mmodified subject line to indicate a C++ compiler issue
* Added error output from the compiler

 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
index 23257e9..a325311 100644
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 
 	if (!pci_dev)
 		return -ENODEV;
Signature · 1 line
1.8.3.1
#14 lore raw

Re: [dpdk-dev] [dpdk-stable] [PATCH v2] ethdev: missing typecast causes C++ build error

Quoted 15 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device. This causes build issues in the GNU C++ compiler.

[CXX] g++ -o utils/pcap_handle.o -c utils/pcap_handle.cc -g3·
-ggdb3 -mcpu=native -mtune=native -isystem· /home/rchirra/gen/io/ppc64include -std=gnu++11 -flax-vector-conversions· -Werror -isystem /home/rchirra/gen/io

In file included from drivers/pmd.cc:7:0:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:
In function ‘int eth_dev_pci_specific_init(rte_eth_dev*, void*)’:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:75:35:·
error: invalid conversion from ‘void*’ to ‘rte_pci_device*’ [-fpermissive]

Quoted 6 lines
  struct rte_pci_device *pci_dev = bus_device;
                                   ^~~~~~~~~~

make[1]: *** [drivers/pmd.o] Error 1

$ g++ --version
g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0

This problem is while building a driver with c++ compiler, but since the solution is a simple casting, which doesn't have any side affect, I am for getting the patch, any objection?

Quoted 27 lines

---
v2:
* Mmodified subject line to indicate a C++ compiler issue
* Added error output from the compiler

 lib/librte_ethdev/rte_ethdev_pci.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
index 23257e9..a325311 100644
@@ -72,7 +72,7 @@
 
 static inline int
 eth_dev_pci_specific_init(struct rte_eth_dev *eth_dev, void *bus_device) {
-	struct rte_pci_device *pci_dev = bus_device;
+	struct rte_pci_device *pci_dev = (struct rte_pci_device *)bus_device;
 
 	if (!pci_dev)
 		return -ENODEV;
#15 lore raw

Re: [dpdk-dev] [dpdk-stable] [PATCH v2] ethdev: missing typecast causes C++ build error

Quoted 20 lines · Ferruh Yigit
Quoted 19 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device. This causes build issues in the GNU C++ compiler.

[CXX] g++ -o utils/pcap_handle.o -c utils/pcap_handle.cc -g3·
-ggdb3 -mcpu=native -mtune=native -isystem· /home/rchirra/gen/io/ppc64include -std=gnu++11 -flax-vector-conversions· -Werror -isystem /home/rchirra/gen/io

In file included from drivers/pmd.cc:7:0:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:
In function ‘int eth_dev_pci_specific_init(rte_eth_dev*, void*)’:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:75:35:·
error: invalid conversion from ‘void*’ to ‘rte_pci_device*’ [-fpermissive]

   struct rte_pci_device *pci_dev = bus_device;
                                    ^~~~~~~~~~

make[1]: *** [drivers/pmd.o] Error 1

Quoted 5 lines
Quoted 2 lines
$ g++ --version
g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0

This problem is while building a driver with c++ compiler, but since the solution is a simple casting, which doesn't have any side affect, I am for getting the patch, any objection?

It should be a decision that we support PMDs in C++ (if I understand the usage correctly) and it will require all headers to be compatible and ideally it should be checked by build automation (otherwise we will break and fix it pretty often).

#16 lore raw

Re: [dpdk-dev] [dpdk-stable] [PATCH v2] ethdev: missing typecast causes C++ build error

Quoted 36 lines

-----Original Message-----
From: dev [mailto:[email protected]] On Behalf Of Andrew Rybchenko
Sent: Tuesday, April 16, 2019 5:40 PM
To: Yigit, Ferruh <[email protected]>; David Christensen <[email protected]>; [email protected]
Cc: [email protected]; [email protected]; [email protected]
Subject: Re: [dpdk-dev] [dpdk-stable] [PATCH v2] ethdev: missing typecast causes C++ build error

Quoted 20 lines · Ferruh Yigit
Quoted 19 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device. This causes build issues in the GNU C++ compiler.

[CXX] g++ -o utils/pcap_handle.o -c utils/pcap_handle.cc -g3·
-ggdb3 -mcpu=native -mtune=native -isystem· /home/rchirra/gen/io/ppc64include -std=gnu++11 -flax-vector-conversions· -Werror -isystem /home/rchirra/gen/io

In file included from drivers/pmd.cc:7:0:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:
In function ‘int eth_dev_pci_specific_init(rte_eth_dev*, void*)’:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:75:35:·
error: invalid conversion from ‘void*’ to ‘rte_pci_device*’ [-fpermissive]

   struct rte_pci_device *pci_dev = bus_device;
                                    ^~~~~~~~~~

make[1]: *** [drivers/pmd.o] Error 1

Quoted 5 lines
Quoted 2 lines
$ g++ --version
g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0

This problem is while building a driver with c++ compiler, but since the solution is a simple casting, which doesn't have any side affect, I am for getting the patch, any objection?

Quoted 4 lines

It should be a decision that we support PMDs in C++ (if I understand the usage correctly) and it will require all headers to be compatible and ideally it should be checked by build automation (otherwise we will break and fix it pretty often).

Or as alternative, we probably can claim that PMDs in C++ are not supported, and if people like to do that - they have to deal with it on their own (create a C wrapper file, or so).
Konstantin

#17 lore raw

Re: [dpdk-dev] [dpdk-stable] [PATCH v2] ethdev: missing typecast causes C++ build error

Quoted 39 lines · Konstantin
Quoted 34 lines

-----Original Message-----
From: dev [mailto:[email protected]] On Behalf Of Andrew Rybchenko
Sent: Tuesday, April 16, 2019 5:40 PM
To: Yigit, Ferruh <[email protected]>; David Christensen <[email protected]>; [email protected]
Cc: [email protected]; [email protected]; [email protected]
Subject: Re: [dpdk-dev] [dpdk-stable] [PATCH v2] ethdev: missing typecast causes C++ build error

Quoted 20 lines · Ferruh Yigit
Quoted 19 lines · David Christensen

The function eth_dev_pci_specific_init is missing a typecast to (struct rte_pci_device *) for the input argument bus_device. This causes build issues in the GNU C++ compiler.

[CXX] g++ -o utils/pcap_handle.o -c utils/pcap_handle.cc -g3·
-ggdb3 -mcpu=native -mtune=native -isystem· /home/rchirra/gen/io/ppc64include -std=gnu++11 -flax-vector-conversions· -Werror -isystem /home/rchirra/gen/io

In file included from drivers/pmd.cc:7:0:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:
In function ‘int eth_dev_pci_specific_init(rte_eth_dev*, void*)’:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:75:35:·
error: invalid conversion from ‘void*’ to ‘rte_pci_device*’ [-fpermissive]

   struct rte_pci_device *pci_dev = bus_device;
                                    ^~~~~~~~~~

make[1]: *** [drivers/pmd.o] Error 1

Quoted 5 lines
Quoted 2 lines
$ g++ --version
g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0  

This problem is while building a driver with c++ compiler, but since the solution is a simple casting, which doesn't have any side affect, I am for getting the patch, any objection?

Quoted 4 lines

It should be a decision that we support PMDs in C++ (if I understand the usage correctly) and it will require all headers to be compatible and ideally it should be checked by build automation (otherwise we will break and fix it pretty often).

Quoted 5 lines

Or as alternative, we probably can claim that PMDs in C++ are not supported, and if people like to do that - they have to deal with it on their own (create a C wrapper file, or so).
Konstantin

+1 no drivers or other parts of EAL in C++

#18 lore raw

Re: [dpdk-dev] [dpdk-stable] [PATCH v2] ethdev: missing typecast causes C++ build error

Quoted 21 lines
Quoted 21 lines
Quoted 20 lines
Quoted 10 lines
Quoted 10 lines

In file included from drivers/pmd.cc:7:0:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:
In function ‘int eth_dev_pci_specific_init(rte_eth_dev*, void*)’:
/home/rchirra/gen/io/deps/dpdk/build/include/rte_ethdev_pci.h:75:35:·
error: invalid conversion from ‘void*’ to ‘rte_pci_device*’ [-fpermissive]

    struct rte_pci_device *pci_dev = bus_device;
                                     ^~~~~~~~~~

make[1]: *** [drivers/pmd.o] Error 1

Quoted 5 lines
Quoted 2 lines
$ g++ --version
g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0

This problem is while building a driver with c++ compiler, but since the solution is a simple casting, which doesn't have any side affect, I am for getting the patch, any objection?

It should be a decision that we support PMDs in C++ (if I understand the usage correctly) and it will require all headers to be compatible and ideally it should be checked by build automation (otherwise we will break and fix it pretty often).

Quoted 7 lines
Quoted 5 lines

Or as alternative, we probably can claim that PMDs in C++ are not supported, and if people like to do that - they have to deal with it on their own (create a C wrapper file, or so).
Konstantin

+1 no drivers or other parts of EAL in C++

I've learned more about the intended usage in this case and it turns out not to be a PMD at all, it's an application that's reaching into the EAL in what I consider an inappropriate way. As a result I'm withdrawing the patch request and I'll work with the developers to find an alternate solution to their problem.

Dave