Skip to contentSkip to conversation

doc: announce renaming of rte_ether_hdr fields

PATCH In review Dmitry Kozlyuk 8 participants 31 messages Mar 3, 2021 – Aug 2, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 doc: announce renaming of rte_ether_hdr fields

It is proposed to rename fields of struct rte_ether_hdr, s_addr tp src_addr and d_addr to dst_addr, due to the clash with system macro on Windows. Until remaining is done in 21.11, a workaround can be used.

Windows Sockets headers contain #define s_addr S_un.S_addr, which conflicts with s_addr field of struct rte_ether_hdr. Undefining this macro in <rte_ether.h> is breaking some usages of DPDK and Windows headers in one file.

Example 1:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    /* addr.s_addr = 0;     ERROR: s_addr undefined by DPDK */

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;     ERROR: `addr_s` is a macro */

It is not mandatory to rename d_addr, this is for consistency only. Naming in rte_ether_hdr will also resemble rte_ipv4/6_hdr.

Workaround is to define struct rte_ether_hdr in such a away that it can be used with or without s_addr macro (as defined on Windows) This can be done for Windows only or for all platforms to save space.

    #pragma push_macro("s_addr")
    #ifdef s_addr
    #undef s_addr
    #endif

    struct rte_ether_hdr {
        struct rte_ether_addr d_addr; /**< Destination address. */
        RTE_STD_C11
        union {
            struct rte_ether_addr s_addr; /**< Source address. */
            struct {
                struct rte_ether_addr S_un;
                /**< MUST NOT be used directly, only via s_addr */
            } S_addr;
            /*< MUST NOT be used directly, only via s_addr */
        };
        uint16_t ether_type; /**< Frame type. */
    } __rte_aligned(2);

    #pragma pop_macro("s_addr")
---
 doc/guides/rel_notes/deprecation.rst | 4 ++++
 1 file changed, 4 insertions(+)
doc/guides/rel_notes/deprecation.rst +4 -0
index 82c1a90a3..f7be10543 100644
@@ -125,3 +125,7 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed to ``src_addr`` and ``dst_addr`` respectively in DPDK 20.11
+  in order to avoid conflict with Windows Sockets headers.
Signature · 1 line
2.29.2
#1 lore raw

doc/guides/rel_notes/deprecation.rst Deprecation Notices · line 131

Quoted 5 lines
doc/guides/rel_notes/deprecation.rst +4 -0
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed to ``src_addr`` and ``dst_addr`` respectively in DPDK 20.11
+  in order to avoid conflict with Windows Sockets headers.

If those fields were a problem now, there might be others in future. Don't use src_addr and dst_addr because those are already used in rte_ipv4_hdr.

Linux and FreeBSD use:

struct ether_header
{
uint8_t ether_dhost[ETH_ALEN]; /* destination eth addr */
uint8_t ether_shost[ETH_ALEN]; /* source ether addr */
uint16_t ether_type; /* packet type ID field */
} __attribute__ ((__packed__));

So why not ether_dhost/ether_shost?

#2 lore raw

2021-03-03 15:54, Stephen Hemminger:

Quoted 6 lines

doc/guides/rel_notes/deprecation.rst Deprecation Notices · line 131

Quoted 4 lines
doc/guides/rel_notes/deprecation.rst +4 -0
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed to ``src_addr`` and ``dst_addr`` respectively in DPDK 20.11
+  in order to avoid conflict with Windows Sockets headers.  

If those fields were a problem now, there might be others in future.

One I can think of is min and max macros from windows.h: they are used as field names in rte_compressdev.h and rte_cryptodev.h (and more than once have they been worked around in PMD code, see i40e and ice patches). Do you prefer a single notice for all such conflicts we can identify now?

Quoted 1 line

Don't use src_addr and dst_addr because those are already used in rte_ipv4_hdr.

Not sure what DPDK policy is: rte_ipv4/6_hdr use completely custom names, while rte_arp_hdr uses traditional names with arp_ prefix. Coming from C++, I chose the former approach, but it's not a strong opinion.

Quoted 10 lines

Linux and FreeBSD use:

struct ether_header
{
uint8_t ether_dhost[ETH_ALEN]; /* destination eth addr */
uint8_t ether_shost[ETH_ALEN]; /* source ether addr */
uint16_t ether_type; /* packet type ID field */
} __attribute__ ((__packed__));

So why not ether_dhost/ether_shost?

Works for me, let's see what others think.

#3 lore raw
Quoted 8 lines · Dmitry Kozlyuk

2021-03-03 15:54, Stephen Hemminger:

Quoted 6 lines

doc/guides/rel_notes/deprecation.rst Deprecation Notices · line 131

Quoted 4 lines
doc/guides/rel_notes/deprecation.rst +4 -0
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed to ``src_addr`` and ``dst_addr`` respectively in DPDK 20.11
+  in order to avoid conflict with Windows Sockets headers.  

If those fields were a problem now, there might be others in future.

Quoted 7 lines

One I can think of is min and max macros from windows.h: they are used as field names in rte_compressdev.h and rte_cryptodev.h (and more than once have they been worked around in PMD code, see i40e and ice patches). Do you prefer a single notice for all such conflicts we can identify now?

Quoted 1 line

Don't use src_addr and dst_addr because those are already used in rte_ipv4_hdr.

Why src_addr being used in rte_ipv4_hdr is a problem?

Quoted 4 lines

Not sure what DPDK policy is: rte_ipv4/6_hdr use completely custom names, while rte_arp_hdr uses traditional names with arp_ prefix. Coming from C++, I chose the former approach, but it's not a strong opinion.

I am now aware of a DPDK policy for it, but +1 to former approach.

Quoted 13 lines
Quoted 10 lines

Linux and FreeBSD use:

struct ether_header
{
uint8_t ether_dhost[ETH_ALEN]; /* destination eth addr */
uint8_t ether_shost[ETH_ALEN]; /* source ether addr */
uint16_t ether_type; /* packet type ID field */
} __attribute__ ((__packed__));

So why not ether_dhost/ether_shost?

Works for me, let's see what others think.

#4 lore raw

PATCH v2 1/1 doc: announce API changes for Windows compatibility

In reply to Dmitry Kozlyuk

Windows socket headers define s_addr, min, and max macros which break structure definitions containing fields with one of these names. Undefining those macros would break some consumers as well.

Example 1:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    /* addr.s_addr = 0;     ERROR: s_addr undefined by DPDK */

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;     ERROR: `addr_s` is a macro */

It is proposed to rename the conflicting fields on DPDK side, because Win32 API has wider use and is slower and harder to change.

New names are left unspecified, open suggestions:

* struct rte_ether_hdr (by Stephen Hemminger):

* s_addr -> ether_shost
* d_addr -> ether_dhost (for consistency)

* struct rte_param_log2_range, struct rte_crypto_param_range:

* min -> minimum
* max -> maximum

For s_addr, a workaround is possible to use it until 21.11. For min and max, no workaround seems available. If cryptodev or compressdev is going to be enabled on Windows before 21.11, the only option seems to use a new name on Windows (using #ifdef).

Workaround for s_addr is to define struct rte_ether_hdr in such a away that it can be used with or without s_addr macro (as defined on Windows):

    #pragma push_macro("s_addr")
    #ifdef s_addr
    #undef s_addr
    #endif

    struct rte_ether_hdr {
        struct rte_ether_addr d_addr; /**< Destination address. */
        RTE_STD_C11
        union {
            struct rte_ether_addr s_addr; /**< Source address. */
            struct {
                struct rte_ether_addr S_un;
                /**< MUST NOT be used directly, only via s_addr */
            } S_addr;
            /**< MUST NOT be used directly, only via s_addr */
        };
        uint16_t ether_type; /**< Frame type. */
    } __rte_aligned(2);

    #pragma pop_macro("s_addr")

---
v2: * Propose to rename all problematic fields identified so far.
* Leave future names unspecified, no need to promise now.
* Propose better names for MAC addresses (Stephen Hemminger).

Thread about s_addr workaround:
https://mails.dpdk.org/archives/dev/2021-March/200700.html
Tyler Retzlaff confirmed offline that Microsoft took similar approach.

 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)
doc/guides/rel_notes/deprecation.rst +9 -0
index 64629e0641..854618f091 100644
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed in DPDK 20.11 to avoid conflict with Windows Sockets headers.
+
+* compressdev: ``min`` and ``max`` fields of ``rte_param_log2_range`` structure
+  will be renamed in DPDK 20.11 to avoid conflict with Windows Sockets headers.
+
+* cryptodev: ``min`` and ``max`` fields of ``rte_crypto_param_range`` structure
+  will be renamed in DPDK 20.11 to avoid conflict with Windows Sockets headers.
Signature · 1 line
2.29.2
#5 lore raw

Re: [dpdk-dev] [PATCH v2] doc: announce API changes for Windows compatibility

Quoted 46 lines

-----Original Message-----
From: dev <[email protected]> On Behalf Of Dmitry Kozlyuk
Sent: 10 March 2021 23:54
To: [email protected]
Cc: Ferruh Yigit <[email protected]>; Pallavi Kadam <[email protected]>; Dmitry Malloy <[email protected]>; Narcisa Ana Maria Vasile <[email protected]>; Stephen Hemminger <[email protected]>; Tyler Retzlaff <[email protected]>; Declan Doherty <[email protected]>; Fiona Trahe <[email protected]>; Ashish Gupta <[email protected]>; Dmitry Kozlyuk <[email protected]>; Ray Kinsella <[email protected]>; Neil Horman <[email protected]>
Subject: [dpdk-dev] [PATCH v2] doc: announce API changes for Windows compatibility

CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.

Windows socket headers define s_addr, min, and max macros which break structure definitions containing fields with one of these names. Undefining those macros would break some consumers as well.

Example 1:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    /* addr.s_addr = 0;     ERROR: s_addr undefined by DPDK */

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;     ERROR: `addr_s` is a macro */

It is proposed to rename the conflicting fields on DPDK side, because Win32 API has wider use and is slower and harder to change.

New names are left unspecified, open suggestions:

* struct rte_ether_hdr (by Stephen Hemminger):

Quoted 7 lines

* s_addr -> ether_shost
* d_addr -> ether_dhost (for consistency)

* struct rte_param_log2_range, struct rte_crypto_param_range:

* min -> minimum
* max -> maximum

The min/max macros in the Windows headers cause issues with C++ projects also (breaks std::min/std::max). The fix there is to "#define NOMINMAX" prior to including windows.h, maybe that's appropriate here too?

Quoted 69 lines

For s_addr, a workaround is possible to use it until 21.11. For min and max, no workaround seems available. If cryptodev or compressdev is going to be enabled on Windows before 21.11, the only option seems to use a new name on Windows (using #ifdef).

Workaround for s_addr is to define struct rte_ether_hdr in such a away that it can be used with or without s_addr macro (as defined on Windows):

    #pragma push_macro("s_addr")
    #ifdef s_addr
    #undef s_addr
    #endif

    struct rte_ether_hdr {
        struct rte_ether_addr d_addr; /**< Destination address. */
        RTE_STD_C11
        union {
            struct rte_ether_addr s_addr; /**< Source address. */
            struct {
                struct rte_ether_addr S_un;
                /**< MUST NOT be used directly, only via s_addr */
            } S_addr;
            /**< MUST NOT be used directly, only via s_addr */
        };
        uint16_t ether_type; /**< Frame type. */
    } __rte_aligned(2);

    #pragma pop_macro("s_addr")

---
v2: * Propose to rename all problematic fields identified so far.
* Leave future names unspecified, no need to promise now.
* Propose better names for MAC addresses (Stephen Hemminger).

Thread about s_addr workaround:
https://mails.dpdk.org/archives/dev/2021-March/200700.html
Tyler Retzlaff confirmed offline that Microsoft took similar approach.

 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)
doc/guides/rel_notes/deprecation.rst +0 -0
index 64629e0641..854618f091 100644
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-

specific
content. On Linux and FreeBSD, supported prior to DPDK 20.11, original structure will be kept until DPDK 21.11.

doc/guides/rel_notes/deprecation.rst +3 -0
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed in DPDK 20.11 to avoid conflict with Windows Sockets

headers.

doc/guides/rel_notes/deprecation.rst +4 -0
+
+* compressdev: ``min`` and ``max`` fields of ``rte_param_log2_range``
+structure
+  will be renamed in DPDK 20.11 to avoid conflict with Windows Sockets

headers.

doc/guides/rel_notes/deprecation.rst +4 -0
+
+* cryptodev: ``min`` and ``max`` fields of ``rte_crypto_param_range``
+structure
+  will be renamed in DPDK 20.11 to avoid conflict with Windows Sockets

headers.
--
2.29.2

#6 lore raw

Re: [dpdk-dev] [PATCH v2] doc: announce API changes for Windows compatibility

2021-03-11 16:19 (UTC+0000), John Alexander:
[...]

Quoted 7 lines
Quoted 4 lines

* struct rte_param_log2_range, struct rte_crypto_param_range:

* min -> minimum
* max -> maximum

The min/max macros in the Windows headers cause issues with C++ projects also (breaks std::min/std::max). The fix there is to "#define NOMINMAX" prior to including windows.h, maybe that's appropriate here too?

We don't control include order in user code and we shouldn't #undef system macros in public headers. We could push_macro/pop_macro around structure definition and have min/max undefined in DPDK internal code, so that including this header always works. Then, if user wants to access the fields, they should take care of macros themselves.

#7 lore raw

Re: [dpdk-dev] [PATCH v2] doc: announce API changes for Windows compatibility

Quoted 8 lines · Dmitry Kozlyuk

2021-03-11 16:19 (UTC+0000), John Alexander:
[...]

Quoted 6 lines
Quoted 4 lines

* struct rte_param_log2_range, struct rte_crypto_param_range:

* min -> minimum
* max -> maximum

Quoted 7 lines
Quoted 1 line

The min/max macros in the Windows headers cause issues with C++ projects also (breaks std::min/std::max). The fix there is to "#define NOMINMAX" prior to including windows.h, maybe that's appropriate here too?

We don't control include order in user code and we shouldn't #undef system macros in public headers. We could push_macro/pop_macro around structure definition and have min/max undefined in DPDK internal code, so that including this header always works. Then, if user wants to access the fields, they should take care of macros themselves.

agreed, nothing stops an application from including windows.h before dpdk includes windows.h and similarly the application shouldn't have visibility of min/max (broken as they are) disappear from the application namespace after including dpdk headers.

the approach of altering what appears in the namespace for the scope of the dpdk header preprocessing is about the best that can be achieved.

#8 lore raw

Re: [dpdk-dev] [PATCH v2] doc: announce API changes for Windows compatibility

In reply to Dmitry Kozlyuk

11/03/2021 00:54, Dmitry Kozlyuk:

Quoted 21 lines

Windows socket headers define s_addr, min, and max macros which break structure definitions containing fields with one of these names. Undefining those macros would break some consumers as well.

Example 1:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    /* addr.s_addr = 0;     ERROR: s_addr undefined by DPDK */

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;     ERROR: `addr_s` is a macro */

It is proposed to rename the conflicting fields on DPDK side, because Win32 API has wider use and is slower and harder to change.

Quoted 6 lines

New names are left unspecified, open suggestions:

* struct rte_ether_hdr (by Stephen Hemminger):

* s_addr -> ether_shost
* d_addr -> ether_dhost (for consistency)

I prefer "addr" over "host".

Quoted 9 lines

* struct rte_param_log2_range, struct rte_crypto_param_range:

* min -> minimum
* max -> maximum

For s_addr, a workaround is possible to use it until 21.11. For min and max, no workaround seems available. If cryptodev or compressdev is going to be enabled on Windows before 21.11, the only option seems to use a new name on Windows (using #ifdef).

I think we have enough work on ethdev until 21.11.

#9 lore raw

PATCH v3 1/1 doc: announce API changes for Windows compatibility

In reply to Dmitry Kozlyuk

Windows system headers define s_addr, min, and max macros which break structure definitions containing fields with one of these names. Undefining those macros would break consumer code that relies on them.

Example 1:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    /* addr.s_addr = 0;     ERROR: s_addr undefined by DPDK */

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;     ERROR: `addr_s` is a macro */

Commit 6c068dbd9fea ("net: work around s_addr macro on Windows") modified definition of struct rte_ether_hdr to avoid the issue. However, the workaround assumes #define s_addr S_addr.S_un in Windows headers, which is not a part of official API. It also complicates the definition of struct rte_ether_hdr.

For min and max, no workaround seems available. If cryptodev or compressdev is going to be enabled on Windows before 21.11, the only option seems to use a new name on Windows (using #ifdef).

It is proposed to rename the conflicting fields on DPDK side, because Win32 API has wider use and is slower and harder to change. Exact new names are left for further discussion.

---
v3: fix typos (Ferruh), remove naming speculation,
replace workaround snippet with commit reference.

 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)
doc/guides/rel_notes/deprecation.rst +9 -0
index 9584d6bfd7..cc6e8db92c 100644
@@ -147,3 +147,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
+
+* compressdev: ``min`` and ``max`` fields of ``rte_param_log2_range`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
+
+* cryptodev: ``min`` and ``max`` fields of ``rte_crypto_param_range`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
Signature · 1 line
2.29.3
#10 lore raw

Re: [dpdk-dev] [EXT] [PATCH v3] doc: announce API changes for Windows compatibility

Quoted 4 lines

Windows system headers define s_addr, min, and max macros which break structure definitions containing fields with one of these names. Undefining those macros would break consumer code that relies on them.

From the commit message the requirement for changing the structure definitions Is not clear. Please note that 'min' - 'max' are not macros. These are variables of a structure which should not break any other structure/Macro in windows.

Quoted 63 lines

Example 1:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    /* addr.s_addr = 0;     ERROR: s_addr undefined by DPDK */

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;     ERROR: `addr_s` is a macro */

Commit 6c068dbd9fea ("net: work around s_addr macro on Windows") modified definition of struct rte_ether_hdr to avoid the issue. However, the workaround assumes #define s_addr S_addr.S_un in Windows headers, which is not a part of official API. It also complicates the definition of struct rte_ether_hdr.

For min and max, no workaround seems available. If cryptodev or compressdev is going to be enabled on Windows before 21.11, the only option seems to use a new name on Windows (using #ifdef).

It is proposed to rename the conflicting fields on DPDK side, because Win32 API has wider use and is slower and harder to change. Exact new names are left for further discussion.

---
v3: fix typos (Ferruh), remove naming speculation,
replace workaround snippet with commit reference.

 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)
doc/guides/rel_notes/deprecation.rst +0 -0
index 9584d6bfd7..cc6e8db92c 100644
@@ -147,3 +147,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-

specific
content. On Linux and FreeBSD, supported prior to DPDK 20.11, original structure will be kept until DPDK 21.11.

doc/guides/rel_notes/deprecation.rst +3 -0
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets

headers.

doc/guides/rel_notes/deprecation.rst +2 -0
+
+* compressdev: ``min`` and ``max`` fields of ``rte_param_log2_range``

structure
+ will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.

doc/guides/rel_notes/deprecation.rst +2 -0
+
+* cryptodev: ``min`` and ``max`` fields of ``rte_crypto_param_range``

structure
+ will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
--
2.29.3

#11 lore raw

Re: [dpdk-dev] [EXT] [PATCH v3] doc: announce API changes for Windows compatibility

2021-05-20 18:59 (UTC+0000), Akhil Goyal:

Quoted 8 lines
Quoted 4 lines

Windows system headers define s_addr, min, and max macros which break structure definitions containing fields with one of these names. Undefining those macros would break consumer code that relies on them.

From the commit message the requirement for changing the structure definitions Is not clear. Please note that 'min' - 'max' are not macros. These are variables of a structure which should not break any other structure/Macro in windows.

Err, yes, that's what the commit message says.
Structure fields of course break nothing; they are broken by Windows macros. Would this make more sense?

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken. If DPDK headers undefined these macros, it could break consumer code relying on these macros. It is proposed to rename structure fields in DPDK, because Win32 headers are more widely used and harder to fix.

#12 lore raw

Re: [dpdk-dev] [EXT] [PATCH v3] doc: announce API changes for Windows compatibility

Quoted 20 lines

2021-05-20 18:59 (UTC+0000), Akhil Goyal:

Quoted 3 lines
Quoted 3 lines

Windows system headers define s_addr, min, and max macros which break structure definitions containing fields with one of these names. Undefining those macros would break consumer code that relies on

them.

Quoted 3 lines
Quoted 1 line

From the commit message the requirement for changing the structure

definitions

Quoted 1 line

Is not clear. Please note that 'min' - 'max' are not macros. These are

variables of a

Quoted 1 line

structure which should not break any other structure/Macro in windows.

Err, yes, that's what the commit message says.
Structure fields of course break nothing; they are broken by Windows macros.
Would this make more sense?

Quoted 8 lines

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken. If DPDK headers undefined these macros, it could break consumer
code
relying on these macros. It is proposed to rename structure fields in DPDK, because Win32 headers are more widely used and harder
to fix.

Yes it makes more sense now. But ideally it should be fixed in windows. This may be just one such issue, there may be many more. Will this also mean that nobody can define a local variable 'min'? Is this acceptable?

Any macro definition in a subsystem should have a prefix to denote that, Just like in DPDK 'RTE_' is added.
Macros with generic names should be avoided so that we do not get into these issues.

Adding more people for comments. I don't have a good feeling about this change.

#13 lore raw

Re: [dpdk-dev] [EXT] [PATCH v3] doc: announce API changes for Windows compatibility

2021-05-20 20:17 (UTC+0000), Akhil Goyal:

Quoted 29 lines
Quoted 20 lines

2021-05-20 18:59 (UTC+0000), Akhil Goyal:

Quoted 3 lines
Quoted 3 lines

Windows system headers define s_addr, min, and max macros which break structure definitions containing fields with one of these names. Undefining those macros would break consumer code that relies on

them.

Quoted 3 lines
Quoted 1 line

From the commit message the requirement for changing the structure

definitions

Quoted 1 line

Is not clear. Please note that 'min' - 'max' are not macros. These are

variables of a

Quoted 1 line

structure which should not break any other structure/Macro in windows.

Err, yes, that's what the commit message says.
Structure fields of course break nothing; they are broken by Windows macros.
Would this make more sense?

Quoted 8 lines

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken. If DPDK headers undefined these macros, it could break consumer
code
relying on these macros. It is proposed to rename structure fields in DPDK, because Win32 headers are more widely used and harder
to fix.

Quoted 4 lines

Yes it makes more sense now. But ideally it should be fixed in windows. This may be just one such issue, there may be many more. Will this also mean that nobody can define a local variable 'min'? Is this acceptable?

Only in public headers. There happens to be one such, rte_lru_x86.h.

Quoted 7 lines

Any macro definition in a subsystem should have a prefix to denote that, Just like in DPDK 'RTE_' is added.
Macros with generic names should be avoided so that we do not get into these issues.

Adding more people for comments. I don't have a good feeling about this change.

Friendly ping to everyone Akhil cc'ed.
As far as I understand, if we want to fix it in 21.11, deprecation notice should make it into 21.08.

#14 lore raw

Re: [dpdk-dev] [EXT] [PATCH v3] doc: announce API changes for Windows compatibility

2021-06-09 18:52 (UTC+0300), Dmitry Kozlyuk:

Quoted 42 lines

2021-05-20 20:17 (UTC+0000), Akhil Goyal:

Quoted 29 lines
Quoted 20 lines

2021-05-20 18:59 (UTC+0000), Akhil Goyal:

Quoted 3 lines
Quoted 3 lines

Windows system headers define s_addr, min, and max macros which break structure definitions containing fields with one of these names. Undefining those macros would break consumer code that relies on

them.

Quoted 3 lines
Quoted 1 line

From the commit message the requirement for changing the structure

definitions

Quoted 1 line

Is not clear. Please note that 'min' - 'max' are not macros. These are

variables of a

Quoted 1 line

structure which should not break any other structure/Macro in windows.

Err, yes, that's what the commit message says.
Structure fields of course break nothing; they are broken by Windows macros.
Would this make more sense?

Quoted 8 lines

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken. If DPDK headers undefined these macros, it could break consumer
code
relying on these macros. It is proposed to rename structure fields in DPDK, because Win32 headers are more widely used and harder
to fix.

Quoted 4 lines

Yes it makes more sense now. But ideally it should be fixed in windows. This may be just one such issue, there may be many more. Will this also mean that nobody can define a local variable 'min'? Is this acceptable?

Only in public headers. There happens to be one such, rte_lru_x86.h.

Quoted 5 lines

Any macro definition in a subsystem should have a prefix to denote that, Just like in DPDK 'RTE_' is added.
Macros with generic names should be avoided so that we do not get into these issues.

Quoted 6 lines
Quoted 2 lines

Adding more people for comments. I don't have a good feeling about this change.

Friendly ping to everyone Akhil cc'ed.
As far as I understand, if we want to fix it in 21.11, deprecation notice should make it into 21.08.

Friendly ping v2.
Hopefully Tyler's answer will help with the decision.

#15 lore raw

Re: [dpdk-dev] [EXT] [PATCH v3] doc: announce API changes for Windows compatibility

In reply to Akhil Goyal
Quoted 3 lines · Akhil Goyal
Quoted 1 line

Yes it makes more sense now. But ideally it should be fixed in windows.

they won't be fixed in windows. it is extremely rare to withdraw something from a platform headers namespace and is avoided to maintain API compatibility no matter how horrible these macros are.

Quoted 2 lines

This may be just one such issue, there may be many more. Will this also mean that nobody can define a local variable 'min'?

that is correct. As well as a long list of other commonly used names not only from windows but other platform headers, standard library headers and keywords.

Quoted 2 lines

Any macro definition in a subsystem should have a prefix to denote that, Just like in DPDK 'RTE_' is added.

completely agree, best practice would be to avoid contaminating the applications namespace.

Quoted 5 lines

Macros with generic names should be avoided so that we do not get into these issues.

Adding more people for comments. I don't have a good feeling about this change.

there is no real choice, the platform header won't be "fixed" so it has to be dealt with and even thinking you can avoid it by just not including windows.h here keep in mind the consuming application will have to anyway so it's difficult to avoid.

#16 lore raw

PATCH v3 1/1 doc: announce API changes for Windows compatibility

In reply to Dmitry Kozlyuk

Windows system headers define s_addr, min, and max macros which break structure definitions containing fields with one of these names. Undefining those macros would break consumer code that relies on them.

Example 1:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    /* addr.s_addr = 0;     ERROR: s_addr undefined by DPDK */

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;     ERROR: `addr_s` is a macro */

Commit 6c068dbd9fea ("net: work around s_addr macro on Windows") modified definition of struct rte_ether_hdr to avoid the issue. However, the workaround assumes #define s_addr S_addr.S_un in Windows headers, which is not a part of official API. It also complicates the definition of struct rte_ether_hdr.

For min and max, no workaround seems available. If cryptodev or compressdev is going to be enabled on Windows before 21.11, the only option seems to use a new name on Windows (using #ifdef).

It is proposed to rename the conflicting fields on DPDK side, because Win32 API has wider use and is slower and harder to change. Exact new names are left for further discussion.

---
v3: fix typos (Ferruh), remove naming speculation,
replace workaround snippet with commit reference.

 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)
doc/guides/rel_notes/deprecation.rst +9 -0
index 9584d6bfd7..cc6e8db92c 100644
@@ -147,3 +147,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
+
+* compressdev: ``min`` and ``max`` fields of ``rte_param_log2_range`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
+
+* cryptodev: ``min`` and ``max`` fields of ``rte_crypto_param_range`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
Signature · 1 line
2.29.3
#17 lore raw

PATCH v4 1/1 doc: announce API changes for Windows compatibility

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken (example 1), as well as any usage of such fields (example 2). If DPDK headers undefined these macros, it could break consumer code (example 3). It is proposed to rename structure fields in DPDK, because Win32 headers are used more widely than DPDK, as a general-purpose platform compared to domain-specific kit, and are harder to fix because of that. Exact new names are left for further discussion.

Example 1:

    /* in DPDK public header included after windows.h */
    struct rte_type {
        int min;    /* ERROR: `min` is a macro */
    };

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    eh.s_addr.addr_bytes[0] = 0;    /* ERROR: `addr_s` is a macro */

Example 3:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    addr.s_addr = 0;      /* ERROR: there is no `s_addr` field,
                             and `s_addr` macro is undefined by DPDK. */

Commit 6c068dbd9fea ("net: work around s_addr macro on Windows") modified definition of struct rte_ether_hdr to avoid the issue. However, the workaround assumes #define s_addr S_addr.S_un in Windows headers, which is not a part of official API. It also complicates the definition of struct rte_ether_hdr.

---
v4: improve wording (Akhil).
v3: fix typos (Ferruh), remove naming speculation,
replace workaround snippet with commit reference.

 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)
doc/guides/rel_notes/deprecation.rst +9 -0
index 9584d6bfd7..cc6e8db92c 100644
@@ -147,3 +147,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
+
+* compressdev: ``min`` and ``max`` fields of ``rte_param_log2_range`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
+
+* cryptodev: ``min`` and ``max`` fields of ``rte_crypto_param_range`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
Signature · 1 line
2.29.3
Acked-by on v4 1/1
#18 lore raw

Re: [dpdk-dev] [PATCH v4] doc: announce API changes for Windows compatibility

21/07/2021 21:55, Dmitry Kozlyuk:

Quoted 41 lines

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken (example 1), as well as any usage of such fields (example 2). If DPDK headers undefined these macros, it could break consumer code (example 3). It is proposed to rename structure fields in DPDK, because Win32 headers are used more widely than DPDK, as a general-purpose platform compared to domain-specific kit, and are harder to fix because of that. Exact new names are left for further discussion.

Example 1:

    /* in DPDK public header included after windows.h */
    struct rte_type {
        int min;    /* ERROR: `min` is a macro */
    };

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    eh.s_addr.addr_bytes[0] = 0;    /* ERROR: `addr_s` is a macro */

Example 3:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    addr.s_addr = 0;      /* ERROR: there is no `s_addr` field,
                             and `s_addr` macro is undefined by DPDK. */

Commit 6c068dbd9fea ("net: work around s_addr macro on Windows") modified definition of struct rte_ether_hdr to avoid the issue. However, the workaround assumes #define s_addr S_addr.S_un in Windows headers, which is not a part of official API. It also complicates the definition of struct rte_ether_hdr.

---

doc/guides/rel_notes/deprecation.rst Deprecation Notices · line 155

Quoted 5 lines
doc/guides/rel_notes/deprecation.rst +5 -0
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
+
+* compressdev: ``min`` and ``max`` fields of ``rte_param_log2_range`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.

The struct rte_param_log2_range should also be renamed to include "compress" prefix. But as we break the struct API, it is not an issue I guess.

doc/guides/rel_notes/deprecation.rst Deprecation Notices · line 158

Quoted 2 lines
doc/guides/rel_notes/deprecation.rst +2 -0
+* cryptodev: ``min`` and ``max`` fields of ``rte_crypto_param_range`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets headers.
#19 lore raw

Re: [dpdk-dev] [EXT] Re: [PATCH v4] doc: announce API changes for Windows compatibility

Quoted 55 lines

21/07/2021 21:55, Dmitry Kozlyuk:

Quoted 41 lines

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken (example 1), as well as any usage of such fields (example 2). If DPDK headers undefined these macros, it could break consumer code (example 3). It is proposed to rename structure fields in DPDK, because Win32 headers are used more widely than DPDK, as a general-purpose platform compared to domain-specific kit, and are harder to fix because of that. Exact new names are left for further discussion.

Example 1:

    /* in DPDK public header included after windows.h */
    struct rte_type {
        int min;    /* ERROR: `min` is a macro */
    };

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    eh.s_addr.addr_bytes[0] = 0;    /* ERROR: `addr_s` is a macro */

Example 3:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    addr.s_addr = 0;      /* ERROR: there is no `s_addr` field,
                             and `s_addr` macro is undefined by DPDK. */

Commit 6c068dbd9fea ("net: work around s_addr macro on Windows") modified definition of struct rte_ether_hdr to avoid the issue. However, the workaround assumes #define s_addr S_addr.S_un in Windows headers, which is not a part of official API. It also complicates the definition of struct rte_ether_hdr.

---

Review

Quoted 2 lines
(quoted diff) +2 -0
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets

headers.

Review

Quoted 2 lines
(quoted diff) +2 -0
+
+* compressdev: ``min`` and ``max`` fields of ``rte_param_log2_range``

structure

Quoted 1 line

+ will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets

headers.

The struct rte_param_log2_range should also be renamed to include "compress" prefix.
But as we break the struct API, it is not an issue I guess.

Quoted 7 lines
Quoted 1 line

+* cryptodev: ``min`` and ``max`` fields of ``rte_crypto_param_range``

structure

Quoted 1 line

+ will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets

headers.

Can we have a local variable named as min/max?
If not, then I believe it is not a good idea.

#20 lore raw

Re: [dpdk-dev] [EXT] Re: [PATCH v4] doc: announce API changes for Windows compatibility

2021-08-02 12:45 (UTC+0000), Akhil Goyal:

Quoted 56 lines
Quoted 56 lines

21/07/2021 21:55, Dmitry Kozlyuk:

Quoted 41 lines

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken (example 1), as well as any usage of such fields (example 2). If DPDK headers undefined these macros, it could break consumer code (example 3). It is proposed to rename structure fields in DPDK, because Win32 headers are used more widely than DPDK, as a general-purpose platform compared to domain-specific kit, and are harder to fix because of that. Exact new names are left for further discussion.

Example 1:

    /* in DPDK public header included after windows.h */
    struct rte_type {
        int min;    /* ERROR: `min` is a macro */
    };

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    eh.s_addr.addr_bytes[0] = 0;    /* ERROR: `addr_s` is a macro */

Example 3:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    addr.s_addr = 0;      /* ERROR: there is no `s_addr` field,
                             and `s_addr` macro is undefined by DPDK. */

Commit 6c068dbd9fea ("net: work around s_addr macro on Windows") modified definition of struct rte_ether_hdr to avoid the issue. However, the workaround assumes #define s_addr S_addr.S_un in Windows headers, which is not a part of official API. It also complicates the definition of struct rte_ether_hdr.

---

Review

Quoted 2 lines
(quoted diff) +2 -0
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets  

headers.

Review

Quoted 2 lines
(quoted diff) +2 -0
+
+* compressdev: ``min`` and ``max`` fields of ``rte_param_log2_range``  

structure

Quoted 1 line

+ will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets

headers.

The struct rte_param_log2_range should also be renamed to include "compress" prefix.
But as we break the struct API, it is not an issue I guess.

Quoted 1 line

+* cryptodev: ``min`` and ``max`` fields of ``rte_crypto_param_range``

Quoted 8 lines
Quoted 6 lines

structure

Quoted 1 line

+ will be renamed in DPDK 21.11 to avoid conflict with Windows Sockets

headers.

Can we have a local variable named as min/max?
If not, then I believe it is not a good idea.

Yes, except for inline functions in public headers. The only problematic one I know of is this (rte_lru_x86.h):

static inline int
f_lru_pos(uint64_t lru_list)

{
	__m128i lst = _mm_set_epi64x((uint64_t)-1, lru_list);
	__m128i min = _mm_minpos_epu16(lst); /* <<< */
	return _mm_extract_epi16(min, 1);
}

Fixing it breaks neither API nor ABI, thus no explicit deprecation notice.

Acked-by
#21 lore raw

Re: [dpdk-dev] [EXT] Re: [PATCH v4] doc: announce API changes for Windows compatibility

Quoted 76 lines

2021-08-02 12:45 (UTC+0000), Akhil Goyal:

Quoted 7 lines
Quoted 7 lines

21/07/2021 21:55, Dmitry Kozlyuk:

Quoted 6 lines

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken (example 1), as well as any usage of such fields (example 2). If DPDK headers undefined these macros, it could break consumer code (example 3). It is proposed to rename structure fields in DPDK, because Win32

headers

Quoted 1 line
Quoted 1 line
Quoted 1 line

are used more widely than DPDK, as a general-purpose platform

compared

Quoted 31 lines
Quoted 31 lines
Quoted 31 lines

to domain-specific kit, and are harder to fix because of that. Exact new names are left for further discussion.

Example 1:

    /* in DPDK public header included after windows.h */
    struct rte_type {
        int min;    /* ERROR: `min` is a macro */
    };

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    eh.s_addr.addr_bytes[0] = 0;    /* ERROR: `addr_s` is a macro */

Example 3:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    addr.s_addr = 0;      /* ERROR: there is no `s_addr` field,
                             and `s_addr` macro is undefined by DPDK. */

Commit 6c068dbd9fea ("net: work around s_addr macro on Windows") modified definition of struct rte_ether_hdr to avoid the issue. However, the workaround assumes #define s_addr S_addr.S_un in Windows headers, which is not a part of official API. It also complicates the definition of struct rte_ether_hdr.

Quoted 5 lines
Quoted 5 lines

Review

Quoted 5 lines

---

(quoted diff) +2 -0
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows

Sockets

Quoted 5 lines
Quoted 5 lines

headers.

Review

Quoted 2 lines
(quoted diff) +2 -0
+
+* compressdev: ``min`` and ``max`` fields of ``rte_param_log2_range``

structure

Quoted 1 line

+ will be renamed in DPDK 21.11 to avoid conflict with Windows

Sockets

Quoted 9 lines
Quoted 9 lines

headers.

The struct rte_param_log2_range should also be renamed to include "compress" prefix.
But as we break the struct API, it is not an issue I guess.

Quoted 1 line

+* cryptodev: ``min`` and ``max`` fields of ``rte_crypto_param_range``

structure

Quoted 1 line

+ will be renamed in DPDK 21.11 to avoid conflict with Windows

Sockets

Quoted 6 lines
Quoted 4 lines

headers.

Can we have a local variable named as min/max?
If not, then I believe it is not a good idea.

Yes, except for inline functions in public headers. The only problematic one I know of is this (rte_lru_x86.h):

static inline int
f_lru_pos(uint64_t lru_list)

Quoted 7 lines
{
	__m128i lst = _mm_set_epi64x((uint64_t)-1, lru_list);
	__m128i min = _mm_minpos_epu16(lst); /* <<< */
	return _mm_extract_epi16(min, 1);
}

Fixing it breaks neither API nor ABI, thus no explicit deprecation notice.

OK,

I hope when windows compilation is enabled, it will be part of CI and it will run on each patch which goes to patchworks.

#22 lore raw

Re: [dpdk-dev] [EXT] Re: [PATCH v4] doc: announce API changes for Windows compatibility

Quoted 93 lines

Subject: Re: [dpdk-dev] [EXT] Re: [PATCH v4] doc: announce API changes for Windows compatibility

External email: Use caution opening links or attachments

Quoted 7 lines

2021-08-02 12:45 (UTC+0000), Akhil Goyal:

Quoted 6 lines
Quoted 6 lines

21/07/2021 21:55, Dmitry Kozlyuk:

Quoted 5 lines

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken (example 1), as well as any usage of such fields (example 2). If DPDK headers undefined these macros, it could break consumer code

(example 3).

Quoted 31 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

It is proposed to rename structure fields in DPDK, because Win32

headers

Quoted 1 line
Quoted 1 line
Quoted 1 line

are used more widely than DPDK, as a general-purpose platform

compared

Quoted 27 lines
Quoted 27 lines
Quoted 27 lines

to domain-specific kit, and are harder to fix because of that. Exact new names are left for further discussion.

Example 1:

    /* in DPDK public header included after windows.h */
    struct rte_type {
        int min;    /* ERROR: `min` is a macro */
    };

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    eh.s_addr.addr_bytes[0] = 0;    /* ERROR: `addr_s` is a macro */

Example 3:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    addr.s_addr = 0;      /* ERROR: there is no `s_addr` field,
                             and `s_addr` macro is undefined by

DPDK. */

Commit 6c068dbd9fea ("net: work around s_addr macro on

Windows")

Quoted 47 lines
Quoted 11 lines
Quoted 11 lines

Review

Quoted 11 lines

modified definition of struct rte_ether_hdr to avoid the issue. However, the workaround assumes #define s_addr S_addr.S_un in Windows headers, which is not a part of official API. It also complicates the definition of struct rte_ether_hdr.

---

(quoted diff) +3 -0
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr``
+structure
+  will be renamed in DPDK 21.11 to avoid conflict with Windows

Sockets

Quoted 6 lines
Quoted 6 lines

headers.

Review

Quoted 3 lines
(quoted diff) +3 -0
+
+* compressdev: ``min`` and ``max`` fields of
+``rte_param_log2_range``

structure

Quoted 1 line

+ will be renamed in DPDK 21.11 to avoid conflict with Windows

Sockets

Quoted 10 lines
Quoted 10 lines

headers.

The struct rte_param_log2_range should also be renamed to include "compress" prefix.
But as we break the struct API, it is not an issue I guess.

Quoted 2 lines

+* cryptodev: ``min`` and ``max`` fields of
+``rte_crypto_param_range``

structure

Quoted 1 line

+ will be renamed in DPDK 21.11 to avoid conflict with Windows

Sockets

Quoted 6 lines
Quoted 4 lines

headers.

Can we have a local variable named as min/max?
If not, then I believe it is not a good idea.

Yes, except for inline functions in public headers. The only problematic one I know of is this (rte_lru_x86.h):

static inline int
f_lru_pos(uint64_t lru_list)

{
      __m128i lst = _mm_set_epi64x((uint64_t)-1, lru_list);
      __m128i min = _mm_minpos_epu16(lst); /* <<< */
      return _mm_extract_epi16(min, 1); }
Quoted 6 lines
Quoted 1 line

Fixing it breaks neither API nor ABI, thus no explicit deprecation notice.

OK,

I hope when windows compilation is enabled, it will be part of CI and it will run on each patch which goes to patchworks.

Windows compilation is already part of CI in ci/iol-testing and ci/Intel-compilation.

#23 lore raw

Re: [dpdk-dev] [EXT] Re: [PATCH v4] doc: announce API changes for Windows compatibility

In reply to Akhil Goyal

02/08/2021 15:48, Akhil Goyal:

Quoted 44 lines
Quoted 44 lines

2021-08-02 12:45 (UTC+0000), Akhil Goyal:

Quoted 7 lines
Quoted 7 lines

21/07/2021 21:55, Dmitry Kozlyuk:

Quoted 6 lines

Windows headers define s_addr, min, and max as macros. If DPDK headers are included after Windows ones, DPDK structure definitions containing fields with these names get broken (example 1), as well as any usage of such fields (example 2). If DPDK headers undefined these macros, it could break consumer code (example 3). It is proposed to rename structure fields in DPDK, because Win32

headers

Quoted 1 line
Quoted 1 line
Quoted 1 line

are used more widely than DPDK, as a general-purpose platform

compared

Quoted 33 lines
Quoted 33 lines
Quoted 33 lines

to domain-specific kit, and are harder to fix because of that. Exact new names are left for further discussion.

Example 1:

    /* in DPDK public header included after windows.h */
    struct rte_type {
        int min;    /* ERROR: `min` is a macro */
    };

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    eh.s_addr.addr_bytes[0] = 0;    /* ERROR: `addr_s` is a macro */

Example 3:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    addr.s_addr = 0;      /* ERROR: there is no `s_addr` field,
                             and `s_addr` macro is undefined by DPDK. */

Commit 6c068dbd9fea ("net: work around s_addr macro on Windows") modified definition of struct rte_ether_hdr to avoid the issue. However, the workaround assumes #define s_addr S_addr.S_un in Windows headers, which is not a part of official API. It also complicates the definition of struct rte_ether_hdr.

[...]

Quoted 16 lines
Quoted 16 lines
Quoted 4 lines
Quoted 2 lines

Can we have a local variable named as min/max?
If not, then I believe it is not a good idea.

Yes, except for inline functions in public headers. The only problematic one I know of is this (rte_lru_x86.h):

static inline int
f_lru_pos(uint64_t lru_list)

{
	__m128i lst = _mm_set_epi64x((uint64_t)-1, lru_list);
	__m128i min = _mm_minpos_epu16(lst); /* <<< */
	return _mm_extract_epi16(min, 1);
}
Quoted 3 lines
Quoted 1 line

Fixing it breaks neither API nor ABI, thus no explicit deprecation notice.

OK,

Applied, thanks.

#24 lore raw
In reply to Dmitry Kozlyuk
Quoted 2 lines · Dmitry Kozlyuk

It is proposed to rename fields of struct rte_ether_hdr, s_addr tp src_addr and d_addr to dst_addr,

s/tp/to/

Quoted 2 lines

due to the clash with system macro on Windows. Until remaining is done in 21.11, a workaround can be used.

s/remaining/renaming/ ??

Quoted 13 lines

Windows Sockets headers contain #define s_addr S_un.S_addr, which conflicts with s_addr field of struct rte_ether_hdr. Undefining this macro in <rte_ether.h> is breaking some usages of DPDK and Windows headers in one file.

Example 1:

    #include <winsock2.h>
    #include <rte_ether.h>
    struct in_addr addr;
    /* addr.s_addr = 0;     ERROR: s_addr undefined by DPDK */
Quoted 6 lines

Example 2:

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;     ERROR: `addr_s` is a macro */

s/addr_s/s_addr/ ??

Quoted 20 lines

It is not mandatory to rename d_addr, this is for consistency only. Naming in rte_ether_hdr will also resemble rte_ipv4/6_hdr.

Workaround is to define struct rte_ether_hdr in such a away that it can be used with or without s_addr macro (as defined on Windows) This can be done for Windows only or for all platforms to save space.

    #pragma push_macro("s_addr")
    #ifdef s_addr
    #undef s_addr
    #endif

    struct rte_ether_hdr {
        struct rte_ether_addr d_addr; /**< Destination address. */
        RTE_STD_C11
        union {
            struct rte_ether_addr s_addr; /**< Source address. */
            struct {
                struct rte_ether_addr S_un;
Quoted 9 lines
                /**< MUST NOT be used directly, only via s_addr */
            } S_addr;
            /*< MUST NOT be used directly, only via s_addr */
        };
        uint16_t ether_type; /**< Frame type. */
    } __rte_aligned(2);

    #pragma pop_macro("s_addr")

What is the problem with the workaround, why we can't live with it?

It requires an order in include files, right?

Quoted 10 lines
---
 doc/guides/rel_notes/deprecation.rst | 4 ++++
 1 file changed, 4 insertions(+)
doc/guides/rel_notes/deprecation.rst +0 -0
index 82c1a90a3..f7be10543 100644

doc/guides/rel_notes/deprecation.rst Deprecation Notices · line 130

Quoted 6 lines
doc/guides/rel_notes/deprecation.rst +3 -0
@@ -125,3 +125,7 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* net: ``s_addr`` and ``d_addr`` fields of ``rte_ether_hdr`` structure
+  will be renamed to ``src_addr`` and ``dst_addr`` respectively in DPDK 20.11

v21.11

Quoted 2 lines
doc/guides/rel_notes/deprecation.rst +1 -0
+  in order to avoid conflict with Windows Sockets headers.
#25 lore raw

2021-05-20 15:24 (UTC+0100), Ferruh Yigit:

Quoted 1 line · Dmitry Kozlyuk

[...]

Quoted 27 lines
Quoted 27 lines

It is not mandatory to rename d_addr, this is for consistency only. Naming in rte_ether_hdr will also resemble rte_ipv4/6_hdr.

Workaround is to define struct rte_ether_hdr in such a away that it can be used with or without s_addr macro (as defined on Windows) This can be done for Windows only or for all platforms to save space.

    #pragma push_macro("s_addr")
    #ifdef s_addr
    #undef s_addr
    #endif

    struct rte_ether_hdr {
        struct rte_ether_addr d_addr; /**< Destination address. */
        RTE_STD_C11
        union {
            struct rte_ether_addr s_addr; /**< Source address. */
            struct {
                struct rte_ether_addr S_un;
                /**< MUST NOT be used directly, only via s_addr */
            } S_addr;
            /*< MUST NOT be used directly, only via s_addr */
        };
        uint16_t ether_type; /**< Frame type. */
    } __rte_aligned(2);
Quoted 6 lines
Quoted 2 lines
    #pragma pop_macro("s_addr")

What is the problem with the workaround, why we can't live with it?

It requires an order in include files, right?

There's no problem except a tricky structure definition with fields that violate DPDK coding rules. It works with any include order.

Will fix typos in v3, thanks.

#26 lore raw
Quoted 33 lines · Dmitry Kozlyuk

2021-05-20 15:24 (UTC+0100), Ferruh Yigit:

Quoted 1 line · Dmitry Kozlyuk

[...]

Quoted 30 lines
Quoted 29 lines

It is not mandatory to rename d_addr, this is for consistency only. Naming in rte_ether_hdr will also resemble rte_ipv4/6_hdr.

Workaround is to define struct rte_ether_hdr in such a away that it can be used with or without s_addr macro (as defined on Windows) This can be done for Windows only or for all platforms to save space.

    #pragma push_macro("s_addr")
    #ifdef s_addr
    #undef s_addr
    #endif

    struct rte_ether_hdr {
        struct rte_ether_addr d_addr; /**< Destination address. */
        RTE_STD_C11
        union {
            struct rte_ether_addr s_addr; /**< Source address. */
            struct {
                struct rte_ether_addr S_un;
                /**< MUST NOT be used directly, only via s_addr */
            } S_addr;
            /*< MUST NOT be used directly, only via s_addr */
        };
        uint16_t ether_type; /**< Frame type. */
    } __rte_aligned(2);

    #pragma pop_macro("s_addr")
Quoted 9 lines
Quoted 3 lines

What is the problem with the workaround, why we can't live with it?

It requires an order in include files, right?

There's no problem except a tricky structure definition with fields that violate DPDK coding rules. It works with any include order.

Will fix typos in v3, thanks.

For following case, won't compiler take 's_addr' as macro?

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;
#27 lore raw

2021-05-20 16:27 (UTC+0100), Ferruh Yigit:

Quoted 44 lines · Dmitry Kozlyuk
Quoted 33 lines

2021-05-20 15:24 (UTC+0100), Ferruh Yigit:

Quoted 1 line · Dmitry Kozlyuk

[...]

Quoted 30 lines
Quoted 29 lines

It is not mandatory to rename d_addr, this is for consistency only. Naming in rte_ether_hdr will also resemble rte_ipv4/6_hdr.

Workaround is to define struct rte_ether_hdr in such a away that it can be used with or without s_addr macro (as defined on Windows) This can be done for Windows only or for all platforms to save space.

    #pragma push_macro("s_addr")
    #ifdef s_addr
    #undef s_addr
    #endif

    struct rte_ether_hdr {
        struct rte_ether_addr d_addr; /**< Destination address. */
        RTE_STD_C11
        union {
            struct rte_ether_addr s_addr; /**< Source address. */
            struct {
                struct rte_ether_addr S_un;
                /**< MUST NOT be used directly, only via s_addr */
            } S_addr;
            /*< MUST NOT be used directly, only via s_addr */
        };
        uint16_t ether_type; /**< Frame type. */
    } __rte_aligned(2);

    #pragma pop_macro("s_addr")
Quoted 9 lines
Quoted 3 lines

What is the problem with the workaround, why we can't live with it?

It requires an order in include files, right?

There's no problem except a tricky structure definition with fields that violate DPDK coding rules. It works with any include order.

Will fix typos in v3, thanks.

Quoted 7 lines

For following case, won't compiler take 's_addr' as macro?

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;

Yes, it will. The macro will expand to S_addr.S_un and compile successfully. In theory, Microsoft can change the definition of s_addr, and while I doubt they will, it's a valid concern and a reason to remove workaround in 21.11.

#28 lore raw
Quoted 47 lines · Dmitry Kozlyuk

2021-05-20 16:27 (UTC+0100), Ferruh Yigit:

Quoted 46 lines · Dmitry Kozlyuk
Quoted 33 lines

2021-05-20 15:24 (UTC+0100), Ferruh Yigit:

Quoted 1 line · Dmitry Kozlyuk

[...]

Quoted 30 lines
Quoted 29 lines

It is not mandatory to rename d_addr, this is for consistency only. Naming in rte_ether_hdr will also resemble rte_ipv4/6_hdr.

Workaround is to define struct rte_ether_hdr in such a away that it can be used with or without s_addr macro (as defined on Windows) This can be done for Windows only or for all platforms to save space.

    #pragma push_macro("s_addr")
    #ifdef s_addr
    #undef s_addr
    #endif

    struct rte_ether_hdr {
        struct rte_ether_addr d_addr; /**< Destination address. */
        RTE_STD_C11
        union {
            struct rte_ether_addr s_addr; /**< Source address. */
            struct {
                struct rte_ether_addr S_un;
                /**< MUST NOT be used directly, only via s_addr */
            } S_addr;
            /*< MUST NOT be used directly, only via s_addr */
        };
        uint16_t ether_type; /**< Frame type. */
    } __rte_aligned(2);

    #pragma pop_macro("s_addr")
Quoted 9 lines
Quoted 3 lines

What is the problem with the workaround, why we can't live with it?

It requires an order in include files, right?

There's no problem except a tricky structure definition with fields that violate DPDK coding rules. It works with any include order.

Will fix typos in v3, thanks.

For following case, won't compiler take 's_addr' as macro?

Quoted 7 lines
Quoted 5 lines
    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;

Yes, it will. The macro will expand to S_addr.S_un and compile successfully.

will 'eh.S_addr.S_un.addr_bytes[0] = 0;' compile successfully?

Quoted 3 lines

In theory, Microsoft can change the definition of s_addr, and while I doubt they will, it's a valid concern and a reason to remove workaround in 21.11.

#29 lore raw

2021-05-20 17:04 (UTC+0100), Ferruh Yigit:

Quoted 54 lines · Dmitry Kozlyuk
Quoted 53 lines

2021-05-20 16:27 (UTC+0100), Ferruh Yigit:

Quoted 51 lines · Dmitry Kozlyuk
Quoted 33 lines

2021-05-20 15:24 (UTC+0100), Ferruh Yigit:

Quoted 1 line · Dmitry Kozlyuk

[...]

Quoted 30 lines
Quoted 29 lines

It is not mandatory to rename d_addr, this is for consistency only. Naming in rte_ether_hdr will also resemble rte_ipv4/6_hdr.

Workaround is to define struct rte_ether_hdr in such a away that it can be used with or without s_addr macro (as defined on Windows) This can be done for Windows only or for all platforms to save space.

    #pragma push_macro("s_addr")
    #ifdef s_addr
    #undef s_addr
    #endif

    struct rte_ether_hdr {
        struct rte_ether_addr d_addr; /**< Destination address. */
        RTE_STD_C11
        union {
            struct rte_ether_addr s_addr; /**< Source address. */
            struct {
                struct rte_ether_addr S_un;
                /**< MUST NOT be used directly, only via s_addr */
            } S_addr;
            /*< MUST NOT be used directly, only via s_addr */
        };
        uint16_t ether_type; /**< Frame type. */
    } __rte_aligned(2);

    #pragma pop_macro("s_addr")
Quoted 9 lines
Quoted 3 lines

What is the problem with the workaround, why we can't live with it?

It requires an order in include files, right?

There's no problem except a tricky structure definition with fields that violate DPDK coding rules. It works with any include order.

Will fix typos in v3, thanks.

For following case, won't compiler take 's_addr' as macro?

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;
Quoted 3 lines
Quoted 1 line

Yes, it will. The macro will expand to S_addr.S_un and compile successfully.

will 'eh.S_addr.S_un.addr_bytes[0] = 0;' compile successfully?

Yes, only it's S_un.S_addr, sorry for the typo in my explanation. Both code snippets from commit message compile successfully.

Quoted 5 lines
Quoted 4 lines

In theory, Microsoft can change the definition of s_addr, and while I doubt they will, it's a valid concern and a reason to remove workaround in 21.11.

#30 lore raw
Quoted 55 lines · Dmitry Kozlyuk

2021-05-20 17:04 (UTC+0100), Ferruh Yigit:

Quoted 54 lines · Dmitry Kozlyuk
Quoted 53 lines

2021-05-20 16:27 (UTC+0100), Ferruh Yigit:

Quoted 51 lines · Dmitry Kozlyuk
Quoted 33 lines

2021-05-20 15:24 (UTC+0100), Ferruh Yigit:

Quoted 1 line · Dmitry Kozlyuk

[...]

Quoted 30 lines
Quoted 29 lines

It is not mandatory to rename d_addr, this is for consistency only. Naming in rte_ether_hdr will also resemble rte_ipv4/6_hdr.

Workaround is to define struct rte_ether_hdr in such a away that it can be used with or without s_addr macro (as defined on Windows) This can be done for Windows only or for all platforms to save space.

    #pragma push_macro("s_addr")
    #ifdef s_addr
    #undef s_addr
    #endif

    struct rte_ether_hdr {
        struct rte_ether_addr d_addr; /**< Destination address. */
        RTE_STD_C11
        union {
            struct rte_ether_addr s_addr; /**< Source address. */
            struct {
                struct rte_ether_addr S_un;
                /**< MUST NOT be used directly, only via s_addr */
            } S_addr;
            /*< MUST NOT be used directly, only via s_addr */
        };
        uint16_t ether_type; /**< Frame type. */
    } __rte_aligned(2);

    #pragma pop_macro("s_addr")
Quoted 9 lines
Quoted 3 lines

What is the problem with the workaround, why we can't live with it?

It requires an order in include files, right?

There's no problem except a tricky structure definition with fields that violate DPDK coding rules. It works with any include order.

Will fix typos in v3, thanks.

For following case, won't compiler take 's_addr' as macro?

    #include <rte_ether.h>
    #include <winsock2.h>
    struct rte_ether_hdr eh;
    /* eh.s_addr.addr_bytes[0] = 0;
Quoted 7 lines
Quoted 3 lines
Quoted 1 line

Yes, it will. The macro will expand to S_addr.S_un and compile successfully.

will 'eh.S_addr.S_un.addr_bytes[0] = 0;' compile successfully?

Yes, only it's S_un.S_addr, sorry for the typo in my explanation. Both code snippets from commit message compile successfully.

Ah, I was missing the union on the struct, yes it will build,

And +1 to deprecation notice and clean the "struct rte_ether_hdr" whenever possible.

Quoted 6 lines
Quoted 5 lines
Quoted 4 lines

In theory, Microsoft can change the definition of s_addr, and while I doubt they will, it's a valid concern and a reason to remove workaround in 21.11.