Skip to contentSkip to conversation

eal/headers: explicitly cast void * to type *

PATCH Superseded Tyler Retzlaff 5 participants 13 messages Jan 12, 2021 – Jan 17, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 eal/headers: explicitly cast void * to type *

Explicitly cast void * to type * so that eal headers may be compiled as C or C++.

---
 lib/librte_eal/windows/include/rte_os.h | 2 +-
 lib/librte_ethdev/rte_ethdev_pci.h      | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
lib/librte_eal/windows/include/rte_os.h +1 -1
index ea3fe60e5..7ef38ff06 100644
@@ -86,7 +86,7 @@ asprintf(char **buffer, const char *format, ...)
 		return -1;
 	size++;
 
-	*buffer = malloc(size);
+	*buffer = (char *)malloc(size);
 	if (*buffer == NULL)
 		return -1;
 
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
index bf715896a..c20be29b1 100644
@@ -47,7 +47,7 @@ rte_eth_copy_pci_info(struct rte_eth_dev *eth_dev,
 
 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
2.29.0.vfs.0.0
#1 lore raw
Quoted 2 lines · Tyler Retzlaff

Explicitly cast void * to type * so that eal headers may be compiled as C or C++.

Topic should probably be "eal/windows".

Fixes e8428a9d89f1 ("eal/windows: add some basic functions and macros")
Quoted 2 lines

---

[...]

lib/librte_ethdev/rte_ethdev_pci.h rte_eth_copy_pci_info(struct rte_eth_dev *eth_dev, · line 53

Quoted 13 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
index bf715896a..c20be29b1 100644
@@ -47,7 +47,7 @@ rte_eth_copy_pci_info(struct rte_eth_dev *eth_dev,
 
 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;

This is a private header, it's never exposed---why the change is needed (not that I have a strong opinion, though)?

#2 lore raw
Quoted 5 lines · Dmitry Kozlyuk
Quoted 2 lines · Tyler Retzlaff

Explicitly cast void * to type * so that eal headers may be compiled as C or C++.

Topic should probably be "eal/windows".

i'll submit a new rev that changes this, it's not really windows specific but i guess windows is the only people crazy enough to use c++.

Quoted 3 lines
Fixes e8428a9d89f1 ("eal/windows: add some basic functions and macros")

yep, will do thanks.

Quoted 15 lines
Quoted 3 lines

---

[...]

Quoted 11 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
index bf715896a..c20be29b1 100644
@@ -47,7 +47,7 @@ rte_eth_copy_pci_info(struct rte_eth_dev *eth_dev,
 
 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;
 
Quoted 5 lines

lib/librte_ethdev/rte_ethdev_pci.h rte_eth_copy_pci_info(struct rte_eth_dev *eth_dev, · line 53

Quoted 2 lines
lib/librte_ethdev/rte_ethdev_pci.h +0 -0
 	if (!pci_dev)
 		return -ENODEV;

This is a private header, it's never exposed---why the change is needed (not that I have a strong opinion, though)?

interesting, i'll look into why/how it is being included and confirm. i suppose the question in the back of my mind is if it is private then why is the header being installed at all?

#3 lore raw
Quoted 9 lines · Tyler Retzlaff
Quoted 5 lines · Dmitry Kozlyuk
Quoted 2 lines · Tyler Retzlaff

Explicitly cast void * to type * so that eal headers may be compiled as C or C++.

Topic should probably be "eal/windows".

i'll submit a new rev that changes this, it's not really windows specific but i guess windows is the only people crazy enough to use c++.

1. Topic usually describe area of changes, rte_os.h is in windows directory.

2. It's a perfectly valid concern that public headers must be usable from C++.

Quoted 17 lines
Quoted 11 lines
Quoted 11 lines
lib/librte_ethdev/rte_ethdev_pci.h +1 -1
index bf715896a..c20be29b1 100644
@@ -47,7 +47,7 @@ rte_eth_copy_pci_info(struct rte_eth_dev *eth_dev,
 
 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;
 
Quoted 5 lines

lib/librte_ethdev/rte_ethdev_pci.h rte_eth_copy_pci_info(struct rte_eth_dev *eth_dev, · line 53

Quoted 2 lines
lib/librte_ethdev/rte_ethdev_pci.h +0 -0
 	if (!pci_dev)
 		return -ENODEV;  

This is a private header, it's never exposed---why the change is needed (not that I have a strong opinion, though)?

Quoted 3 lines

interesting, i'll look into why/how it is being included and confirm. i suppose the question in the back of my mind is if it is private then why is the header being installed at all?

+ Bruce

If it's a public header then maybe it's missing a @file?

#4 lore raw
Quoted 35 lines · Dmitry Kozlyuk
Quoted 10 lines · Tyler Retzlaff
Quoted 5 lines · Dmitry Kozlyuk
Quoted 2 lines · Tyler Retzlaff

Explicitly cast void * to type * so that eal headers may be compiled as C or C++.

Topic should probably be "eal/windows".

i'll submit a new rev that changes this, it's not really windows specific but i guess windows is the only people crazy enough to use c++.

1. Topic usually describe area of changes, rte_os.h is in windows directory.

2. It's a perfectly valid concern that public headers must be usable from C++.

Quoted 16 lines
Quoted 15 lines
Quoted 6 lines
(quoted diff) +0 -0

b/lib/librte_ethdev/rte_ethdev_pci.h index bf715896a..c20be29b1 100644 --- a/lib/librte_ethdev/rte_ethdev_pci.h +++ b/lib/librte_ethdev/rte_ethdev_pci.h @@ -47,7 +47,7 @@ rte_eth_copy_pci_info(struct rte_eth_dev *eth_dev,

Quoted 6 lines

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;

This is a private header, it's never exposed---why the change is needed (not that I have a strong opinion, though)?

Quoted 8 lines
Quoted 3 lines

interesting, i'll look into why/how it is being included and confirm. i suppose the question in the back of my mind is if it is private then why is the header being installed at all?

+ Bruce

If it's a public header then maybe it's missing a @file?

My 2c on this in general...

The use of public vs private headers is not always clear, sadly, in DPDK, for historical reasons. With the make builds, libraries picked up headers from other libraries via the "include" folder for all of DPDK, meaning that if a particular header was internal only but used by multiple other libs, it was placed in "include" for simplicity, rather than having each library using it having to have separate "-I/path/to/header" cflags specified. With the switch to meson, this common folder use is no longer be the case, but because of the old way of doing things it may be that in the transition some private headers were inadvertently kept as public (and possibly vice versa, though that is more likely to be spotted by now).

/Bruce

#5 lore raw
Quoted 15 lines
Quoted 2 lines

If it's a public header then maybe it's missing a @file?

My 2c on this in general...

The use of public vs private headers is not always clear, sadly, in DPDK, for historical reasons. With the make builds, libraries picked up headers from other libraries via the "include" folder for all of DPDK, meaning that if a particular header was internal only but used by multiple other libs, it was placed in "include" for simplicity, rather than having each library using it having to have separate "-I/path/to/header" cflags specified. With the switch to meson, this common folder use is no longer be the case, but because of the old way of doing things it may be that in the transition some private headers were inadvertently kept as public (and possibly vice versa, though that is more likely to be spotted by now).

Quoted 1 line

/Bruce

Here is why rte_ethdev_pci.h should be considered private to DPDK:

* rte_eth_copy_pci_info - intended for device init, that is, driver job
* eth_dev_pci_specific_init - wrapper for the above
* rte_eth_dev_pci_allocate - @internal, deals with private data
* rte_eth_dev_pci_generic_probe - @internal, deals with private data
* rte_eth_dev_pci_generic_remove - @internal

#6 lore raw

14/01/2021 19:27, Dmitry Kozlyuk:

Quoted 17 lines
Quoted 15 lines
Quoted 2 lines

If it's a public header then maybe it's missing a @file?

My 2c on this in general...

The use of public vs private headers is not always clear, sadly, in DPDK, for historical reasons. With the make builds, libraries picked up headers from other libraries via the "include" folder for all of DPDK, meaning that if a particular header was internal only but used by multiple other libs, it was placed in "include" for simplicity, rather than having each library using it having to have separate "-I/path/to/header" cflags specified. With the switch to meson, this common folder use is no longer be the case, but because of the old way of doing things it may be that in the transition some private headers were inadvertently kept as public (and possibly vice versa, though that is more likely to be spotted by now).

Quoted 1 line

/Bruce

Quoted 7 lines

Here is why rte_ethdev_pci.h should be considered private to DPDK:

* rte_eth_copy_pci_info - intended for device init, that is, driver job
* eth_dev_pci_specific_init - wrapper for the above
* rte_eth_dev_pci_allocate - @internal, deals with private data
* rte_eth_dev_pci_generic_probe - @internal, deals with private data
* rte_eth_dev_pci_generic_remove - @internal

Yes rte_ethdev_pci.h is a helper for ethdev drivers, it is DPDK internal.

#7 lore raw
In reply to Bruce Richardson
Quoted 6 lines · Bruce Richardson
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

This is a private header, it's never exposed---why the change is needed (not that I have a strong opinion, though)?

interesting, i'll look into why/how it is being included and confirm. i suppose the question in the back of my mind is if it is private then why is the header being installed at all?

okay, i now understand how we ended up compiling this header directly. long story short we just had it #include directly somewhere.

as you have noted it is a private header. a quick examination of the other installed headers shows that it is not included by any of them.

i will update the patch to remove the cast from rte_ethdev_pci.h since it is not necessary.

would you also like a patch submitted that stops installing the header. the change will be breaking if any other consumers have made the same mistake as we did. i'm not sure what dpdk's stance is on pulling headers back out of public space.

thanks

#9 lore raw

PATCH v2 1/1 eal/windows: explicitly cast void * to type *

In reply to Tyler Retzlaff

Explicitly cast void * to type * so that eal headers may be compiled as C or C++.

Fixes e8428a9d89f1 ("eal/windows: add some basic functions and macros")
---
 lib/librte_eal/windows/include/rte_os.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_eal/windows/include/rte_os.h +1 -1
index ea3fe60e5..7ef38ff06 100644
@@ -86,7 +86,7 @@ asprintf(char **buffer, const char *format, ...)
 		return -1;
 	size++;
 
-	*buffer = malloc(size);
+	*buffer = (char *)malloc(size);
 	if (*buffer == NULL)
 		return -1;
 
Signature · 1 line
2.29.0.vfs.0.0
#8 lore raw
In reply to Tyler Retzlaff

15/01/2021 20:21, Tyler Retzlaff:

Quoted 4 lines

would you also like a patch submitted that stops installing the header. the change will be breaking if any other consumers have made the same mistake as we did. i'm not sure what dpdk's stance is on pulling headers back out of public space.

That's a good question.
If it is described enough that it is not part of the API,
I think we can stop installing the header.
Anyway, our only commitment is on ABI compatibility, so it should be OK.

#11 lore raw

Re: [dpdk-dev] [PATCH v2] eal/windows: explicitly cast void * to type *

In reply to Tyler Retzlaff
Quoted 16 lines · On Fri, 15 Jan 2021 11:38:21 -0800 Tyler Retzlaff…

Explicitly cast void * to type * so that eal headers may be compiled as C or C++.

Fixes e8428a9d89f1 ("eal/windows: add some basic functions and macros")
---
 lib/librte_eal/windows/include/rte_os.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_eal/windows/include/rte_os.h +0 -0
index ea3fe60e5..7ef38ff06 100644

lib/librte_eal/windows/include/rte_os.h asprintf(char **buffer, const char *format, ...) · line 92

Quoted 8 lines
lib/librte_eal/windows/include/rte_os.h +1 -1
@@ -86,7 +86,7 @@ asprintf(char **buffer, const char *format, ...)
 		return -1;
 	size++;
 
-	*buffer = malloc(size);
+	*buffer = (char *)malloc(size);
 	if (*buffer == NULL)
 		return -1;
 

Why is the compiler enforcing C++ rules on code that is inside "extern C {"?

Bigger question is why is this code inlined? It is not critical path and should be a function.

#12 lore raw

Re: [dpdk-dev] [PATCH v2] eal/windows: explicitly cast void * to type *

Quoted 7 lines · Stephen Hemminger

lib/librte_eal/windows/include/rte_os.h asprintf(char **buffer, const char *format, ...) · line 91

Quoted 5 lines
lib/librte_eal/windows/include/rte_os.h +1 -1
-	*buffer = malloc(size);
+	*buffer = (char *)malloc(size);
 	if (*buffer == NULL)
 		return -1;

Why is the compiler enforcing C++ rules on code that is inside "extern C {"?

Code inside extern "C" is not compiled as C; directive only affects linkage.

Quoted 2 lines

Bigger question is why is this code inlined? It is not critical path and should be a function.

Absolutely.
There's more: windows/rte_os.h should not expose POSIX symbols at all, I'm working on a patchset to clean it up and un-inline this code.