Refactor RDMA IOCTL declarations

PATCH In review Leon Romanovsky 5 participants 16 messages Sep 4, 2016 – Sep 23, 2016 7 messages not in the archive InfiniBand

lore

Share & tools

lore

Leon Romanovsky

author

2016-09-04 11:30 UTC #0 lore raw

v3 cover Refactor RDMA IOCTL declarations

Hi Doug,

This patch set refactors RDMA IOCTL declarations, unifies their location, moves all current IOCTL exporters to one place and renames them to fix the differences in coding style.

IOCTLs commands need to export the size of their data for their transfer which is usually declared as a struct defined in relevant submodule. These structures should be visible at the linkage stage.

There are number of possible solutions to overcome this limitations.

One of the options is to declare anonymous variable which will be visible during the linkage. As a downside, such option won't allow direct use of this header by other parts of code without their specific includes of other logically unrelated submodules.

Another possible option will be to move declarations of that structures to that common header file. Such move will cause to bloat this file with different submodules declarations.

Our decision was to choose direct include of other exported files with declarations of such structures to simplify future usage of this file. There is no change for legacy applications and libraries.

Changes v2->v3:
* Added Jason's Reviewed-by
* Open-ceded ioctl indexes

Changes v0->v2 (v1 was skipped):
* Update commit messages of patch number 3 and 6.

Available in the "topic/ioctl-header" topic branch of this git repo:
git://git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git

Or for browsing:
https://git.kernel.org/cgit/linux/kernel/git/leon/linux-rdma.git/log/?h=topic/ioctl-header

Thanks

Leon Romanovsky (6):
RDMA/core: Commonize RDMA IOCTL declarations location RDMA/core: Move legacy MAD IOCTL declarations to common file RDMA/hfi1: Avoid redeclaration error
RDMA/core: Move HFI1 IOCTL declarations to common file RDMA/core: Rename RDMA magic number
RDMA/core: Unify style of IOCTL commands

 include/uapi/rdma/Kbuild            |   1 +
 include/uapi/rdma/hfi/Kbuild        |   1 +
 include/uapi/rdma/hfi/hfi1_ioctl.h  | 173 +++++++++++++++++++++++++++++++++++
 include/uapi/rdma/hfi/hfi1_user.h   | 175 +-----------------------------------
 include/uapi/rdma/ib_user_mad.h     |  14 +--
 include/uapi/rdma/rdma_user_ioctl.h |  87 ++++++++++++++++++
 6 files changed, 264 insertions(+), 187 deletions(-)
 create mode 100644 include/uapi/rdma/hfi/hfi1_ioctl.h
 create mode 100644 include/uapi/rdma/rdma_user_ioctl.h

--
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 1472988635-31463-1-git-send-email-leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org

Leon Romanovsky

author

2016-09-04 11:30 UTC #2 lore raw

PATCH v3 1/6 RDMA/core: Commonize RDMA IOCTL declarations location

From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

This patch provides one common file (rdma_user_ioctl.h) for all RDMA UAPI IOCTLs.

Signed-off-by Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Haggai Eran <haggaie-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

Reviewed-by Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>

---
 include/uapi/rdma/Kbuild            |  1 +
 include/uapi/rdma/hfi/hfi1_user.h   |  2 +-
 include/uapi/rdma/ib_user_mad.h     |  4 +---
 include/uapi/rdma/rdma_user_ioctl.h | 43 +++++++++++++++++++++++++++++++++++++
 4 files changed, 46 insertions(+), 4 deletions(-)
 create mode 100644 include/uapi/rdma/rdma_user_ioctl.h

include/uapi/rdma/Kbuild +1 -0

index 4edb0f2..ac62908 100644
@@ -1,5 +1,6 @@
 # UAPI Header export list
 header-y += ib_user_cm.h
+header-y += rdma_user_ioctl.h
 header-y += ib_user_mad.h
 header-y += ib_user_sa.h
 header-y += ib_user_verbs.h

include/uapi/rdma/hfi/hfi1_user.h +1 -1

index d15e728..a404919 100644
@@ -57,6 +57,7 @@
 #define _LINUX__HFI1_USER_H
 
 #include <linux/types.h>
+#include <rdma/rdma_user_ioctl.h>
 
 /*
  * This version number is given to the driver by the user code during

@@ -132,7 +133,6 @@
  * User IOCTLs can not go above 128 if they do then see common.h and change the
  * base for the snoop ioctl
  */
-#define IB_IOCTL_MAGIC 0x1b /* See Documentation/ioctl/ioctl-number.txt */
 
 /*
  * Make the ioctls occupy the last 0xf0-0xff portion of the IB range

include/uapi/rdma/ib_user_mad.h +1 -3

index 09f809f..9de7a2b 100644
@@ -35,7 +35,7 @@
 #define IB_USER_MAD_H
 
 #include <linux/types.h>
-#include <linux/ioctl.h>
+#include <rdma/rdma_user_ioctl.h>
 
 /*
  * Increment this value if any changes that break userspace ABI

@@ -230,8 +230,6 @@ struct ib_user_mad_reg_req2 {
 	__u8	reserved[3];
 };
 
-#define IB_IOCTL_MAGIC		0x1b
-
 #define IB_USER_MAD_REGISTER_AGENT	_IOWR(IB_IOCTL_MAGIC, 1, \
 					      struct ib_user_mad_reg_req)

include/uapi/rdma/rdma_user_ioctl.h new file +43 -0

43 lines not on this page · raw message

--
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Leon Romanovsky

author

2016-09-04 11:30 UTC #3 lore raw

PATCH v3 2/6 RDMA/core: Move legacy MAD IOCTL declarations to common file

In reply to a message not in the archive

From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

Move legacy MAD IOCTL declarations to rdma_user_ioctl.h file.

Signed-off-by Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Haggai Eran <haggaie-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

Reviewed-by Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>

---
 include/uapi/rdma/ib_user_mad.h     | 10 ----------
 include/uapi/rdma/rdma_user_ioctl.h | 11 +++++++++++
 2 files changed, 11 insertions(+), 10 deletions(-)

include/uapi/rdma/ib_user_mad.h +0 ~~-10~~

index 9de7a2b..5c7abd8 100644
@@ -230,14 +230,4 @@ struct ib_user_mad_reg_req2 {
 	__u8	reserved[3];
 };
 
-#define IB_USER_MAD_REGISTER_AGENT	_IOWR(IB_IOCTL_MAGIC, 1, \
-					      struct ib_user_mad_reg_req)
-
-#define IB_USER_MAD_UNREGISTER_AGENT	_IOW(IB_IOCTL_MAGIC, 2, __u32)
-
-#define IB_USER_MAD_ENABLE_PKEY		_IO(IB_IOCTL_MAGIC, 3)
-
-#define IB_USER_MAD_REGISTER_AGENT2     _IOWR(IB_IOCTL_MAGIC, 4, \
-					      struct ib_user_mad_reg_req2)
-
 #endif /* IB_USER_MAD_H */

include/uapi/rdma/rdma_user_ioctl.h +11 -0

index ba1bcdd..820bf34 100644
@@ -35,9 +35,20 @@
 
 #include <linux/types.h>
 #include <linux/ioctl.h>
+#include <rdma/ib_user_mad.h>
 
 /* Documentation/ioctl/ioctl-number.txt */
 #define RDMA_IOCTL_MAGIC		0x1b
 #define IB_IOCTL_MAGIC			RDMA_IOCTL_MAGIC
 
+#define IB_USER_MAD_REGISTER_AGENT	_IOWR(IB_IOCTL_MAGIC, 1, \
+					      struct ib_user_mad_reg_req)
+
+#define IB_USER_MAD_UNREGISTER_AGENT	_IOW(IB_IOCTL_MAGIC, 2, __u32)
+
+#define IB_USER_MAD_ENABLE_PKEY		_IO(IB_IOCTL_MAGIC, 3)
+
+#define IB_USER_MAD_REGISTER_AGENT2     _IOWR(IB_IOCTL_MAGIC, 4, \
+					      struct ib_user_mad_reg_req2)
+
 #endif /* RDMA_USER_IOCTL_H */

--
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Leon Romanovsky

author

2016-09-04 11:30 UTC #4 lore raw

PATCH v3 3/6 RDMA/hfi1: Avoid redeclaration error

In reply to a message not in the archive

From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

Move hfi1 ioctl definitions to a new header which can be included by both the hfi1 and qib drivers to avoid a duplicate enum definition as shown in this build error for qib:

CC [M] drivers/infiniband/hw/qib/qib_sysfs.o
In file included from ./include/uapi/rdma/rdma_user_ioctl.h:39:0,

		 from include/uapi/rdma/ib_user_mad.h:38,
		 from include/rdma/ib_mad.h:43,
		 from include/rdma/ib_pma.h:38,
		 from drivers/infiniband/hw/qib/qib_mad.h:37,
		 from drivers/infiniband/hw/qib/qib_init.c:49:

./include/uapi/rdma/hfi/hfi1_user.h:370:2: error: redeclaration of enumerator ‘ur_rcvhdrtail’
ur_rcvhdrtail = 0,

Move hfi1 structures to separate file to avoid this failure.

The actual move of the ioctl definitions comes in a follow on patch.

Signed-off-by Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Haggai Eran <haggaie-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

Reviewed-by Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>

---
 include/uapi/rdma/hfi/Kbuild       |   1 +
 include/uapi/rdma/hfi/hfi1_ioctl.h | 173 +++++++++++++++++++++++++++++++++++++
 include/uapi/rdma/hfi/hfi1_user.h  | 120 +------------------------
 3 files changed, 175 insertions(+), 119 deletions(-)
 create mode 100644 include/uapi/rdma/hfi/hfi1_ioctl.h

include/uapi/rdma/hfi/Kbuild +1 -0

index ef23c29..b65b0b3 100644
@@ -1,2 +1,3 @@
 # UAPI Header export list
 header-y += hfi1_user.h
+header-y += hfi1_ioctl.h

include/uapi/rdma/hfi/hfi1_ioctl.h new file +173 -0

173 lines not on this page · raw message

include/uapi/rdma/hfi/hfi1_user.h +1 ~~-119~~

138 lines not on this page · raw message

--
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Leon Romanovsky

author

2016-09-04 11:30 UTC #5 lore raw

PATCH v3 4/6 RDMA/core: Move HFI1 IOCTL declarations to common file

In reply to a message not in the archive

From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

Move HFI1 IOCTL declarations to rdma_user_ioctl.h file.

Signed-off-by Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Haggai Eran <haggaie-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

Reviewed-by Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>

---
 include/uapi/rdma/hfi/hfi1_user.h   | 55 -------------------------------------
 include/uapi/rdma/rdma_user_ioctl.h | 54 ++++++++++++++++++++++++++++++++++++
 2 files changed, 54 insertions(+), 55 deletions(-)

include/uapi/rdma/hfi/hfi1_user.h +0 ~~-55~~

index 8aa3867..8807f06 100644
@@ -58,7 +58,6 @@
 
 #include <linux/types.h>
 #include <rdma/rdma_user_ioctl.h>
-#include <rdma/hfi/hfi1_ioctl.h>
 
 /*
  * This version number is given to the driver by the user code during

@@ -114,60 +113,6 @@
 #define HFI1_RCVHDR_ENTSIZE_16   (1UL << 1)
 #define HFI1_RCVDHR_ENTSIZE_32   (1UL << 2)
 
-/* User commands. */
-#define HFI1_CMD_ASSIGN_CTXT     1	/* allocate HFI and context */
-#define HFI1_CMD_CTXT_INFO       2	/* find out what resources we got */
-#define HFI1_CMD_USER_INFO       3	/* set up userspace */
-#define HFI1_CMD_TID_UPDATE      4	/* update expected TID entries */
-#define HFI1_CMD_TID_FREE        5	/* free expected TID entries */
-#define HFI1_CMD_CREDIT_UPD      6	/* force an update of PIO credit */
-
-#define HFI1_CMD_RECV_CTRL       8	/* control receipt of packets */
-#define HFI1_CMD_POLL_TYPE       9	/* set the kind of polling we want */
-#define HFI1_CMD_ACK_EVENT       10	/* ack & clear user status bits */
-#define HFI1_CMD_SET_PKEY        11     /* set context's pkey */
-#define HFI1_CMD_CTXT_RESET      12     /* reset context's HW send context */
-#define HFI1_CMD_TID_INVAL_READ  13     /* read TID cache invalidations */
-#define HFI1_CMD_GET_VERS	 14	/* get the version of the user cdev */
-
-/*
- * User IOCTLs can not go above 128 if they do then see common.h and change the
- * base for the snoop ioctl
- */
-
-/*
- * Make the ioctls occupy the last 0xf0-0xff portion of the IB range
- */
-#define __NUM(cmd) (HFI1_CMD_##cmd + 0xe0)
-
-struct hfi1_cmd;
-#define HFI1_IOCTL_ASSIGN_CTXT \
-	_IOWR(IB_IOCTL_MAGIC, __NUM(ASSIGN_CTXT), struct hfi1_user_info)
-#define HFI1_IOCTL_CTXT_INFO \
-	_IOW(IB_IOCTL_MAGIC, __NUM(CTXT_INFO), struct hfi1_ctxt_info)
-#define HFI1_IOCTL_USER_INFO \
-	_IOW(IB_IOCTL_MAGIC, __NUM(USER_INFO), struct hfi1_base_info)
-#define HFI1_IOCTL_TID_UPDATE \
-	_IOWR(IB_IOCTL_MAGIC, __NUM(TID_UPDATE), struct hfi1_tid_info)
-#define HFI1_IOCTL_TID_FREE \
-	_IOWR(IB_IOCTL_MAGIC, __NUM(TID_FREE), struct hfi1_tid_info)
-#define HFI1_IOCTL_CREDIT_UPD \
-	_IO(IB_IOCTL_MAGIC, __NUM(CREDIT_UPD))
-#define HFI1_IOCTL_RECV_CTRL \
-	_IOW(IB_IOCTL_MAGIC, __NUM(RECV_CTRL), int)
-#define HFI1_IOCTL_POLL_TYPE \
-	_IOW(IB_IOCTL_MAGIC, __NUM(POLL_TYPE), int)
-#define HFI1_IOCTL_ACK_EVENT \
-	_IOW(IB_IOCTL_MAGIC, __NUM(ACK_EVENT), unsigned long)
-#define HFI1_IOCTL_SET_PKEY \
-	_IOW(IB_IOCTL_MAGIC, __NUM(SET_PKEY), __u16)
-#define HFI1_IOCTL_CTXT_RESET \
-	_IO(IB_IOCTL_MAGIC, __NUM(CTXT_RESET))
-#define HFI1_IOCTL_TID_INVAL_READ \
-	_IOWR(IB_IOCTL_MAGIC, __NUM(TID_INVAL_READ), struct hfi1_tid_info)
-#define HFI1_IOCTL_GET_VERS \
-	_IOR(IB_IOCTL_MAGIC, __NUM(GET_VERS), int)
-
 #define _HFI1_EVENT_FROZEN_BIT         0
 #define _HFI1_EVENT_LINKDOWN_BIT       1
 #define _HFI1_EVENT_LID_CHANGE_BIT     2

include/uapi/rdma/rdma_user_ioctl.h +54 -0

64 lines not on this page · raw message

--
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 1472988635-31463-5-git-send-email-leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org

Jason Gunthorpe

2016-09-05 02:41 UTC #7 lore raw

Re: [PATCH rdma-next V3 4/6] RDMA/core: Move HFI1 IOCTL declarations to common file

include/uapi/rdma/rdma_user_ioctl.h line 82

Quoted 5 lines · Leon Romanovsky

include/uapi/rdma/rdma_user_ioctl.h +4 -0

+/* User commands. */
+#define HFI1_CMD_ASSIGN_CTXT     1	/* allocate HFI and context */
 
+#define HFI1_IOCTL_ASSIGN_CTXT \
+	_IOWR(IB_IOCTL_MAGIC, __NUM(ASSIGN_CTXT), struct hfi1_user_info)

I still vote for dropping the HFI1_CMD_ASSIGN_CTXT defines. Inlining the number is far more readable than that ugly __NUM macro.

If it was done consistently you might have an arugment, but just doing it in one place with no purpose is silly.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 20160905024158.GE21542-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org

Leon Romanovsky

author

2016-09-05 05:31 UTC #9 lore raw

Re: [PATCH rdma-next V3 4/6] RDMA/core: Move HFI1 IOCTL declarations to common file

Quoted 5 lines · Jason Gunthorpe

Quoted 2 lines · Leon Romanovsky
include/uapi/rdma/rdma_user_ioctl.h +2 -0
+/* User commands. */
+#define HFI1_CMD_ASSIGN_CTXT     1	/* allocate HFI and context */

Quoted 8 lines

include/uapi/rdma/rdma_user_ioctl.h line 82
Quoted 2 lines
include/uapi/rdma/rdma_user_ioctl.h +2 -0
+#define HFI1_IOCTL_ASSIGN_CTXT \
+	_IOWR(IB_IOCTL_MAGIC, __NUM(ASSIGN_CTXT), struct hfi1_user_info)
I still vote for dropping the HFI1_CMD_ASSIGN_CTXT defines. Inlining the number is far more readable than that ugly __NUM macro.
If it was done consistently you might have an arugment, but just doing it in one place with no purpose is silly.

It is done in patch 6. It is here to preserve one-change->one-patch flow.

Quoted 2 lines

Jason

Leon Romanovsky

author

2016-09-04 11:30 UTC #10 lore raw

PATCH v3 5/6 RDMA/core: Rename RDMA magic number

In reply to a message not in the archive

From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

Rename RDMA magic number to better describe IOCTLs.

Signed-off-by Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Haggai Eran <haggaie-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

Reviewed-by Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>

---
 include/uapi/rdma/rdma_user_ioctl.h | 35 ++++++++++++++++++-----------------
 1 file changed, 18 insertions(+), 17 deletions(-)

include/uapi/rdma/rdma_user_ioctl.h +18 ~~-17~~

index e9a69f0..7ecf8cd 100644
@@ -40,16 +40,17 @@
 
 /* Documentation/ioctl/ioctl-number.txt */
 #define RDMA_IOCTL_MAGIC		0x1b
+/* Legacy name, for user space application which already use it */
 #define IB_IOCTL_MAGIC			RDMA_IOCTL_MAGIC
 
-#define IB_USER_MAD_REGISTER_AGENT	_IOWR(IB_IOCTL_MAGIC, 1, \
+#define IB_USER_MAD_REGISTER_AGENT	_IOWR(RDMA_IOCTL_MAGIC, 1, \
 					      struct ib_user_mad_reg_req)
 
-#define IB_USER_MAD_UNREGISTER_AGENT	_IOW(IB_IOCTL_MAGIC, 2, __u32)
+#define IB_USER_MAD_UNREGISTER_AGENT	_IOW(RDMA_IOCTL_MAGIC, 2, __u32)
 
-#define IB_USER_MAD_ENABLE_PKEY		_IO(IB_IOCTL_MAGIC, 3)
+#define IB_USER_MAD_ENABLE_PKEY		_IO(RDMA_IOCTL_MAGIC, 3)
 
-#define IB_USER_MAD_REGISTER_AGENT2     _IOWR(IB_IOCTL_MAGIC, 4, \
+#define IB_USER_MAD_REGISTER_AGENT2     _IOWR(RDMA_IOCTL_MAGIC, 4, \
 					      struct ib_user_mad_reg_req2)
 
 /* User commands. */

@@ -79,30 +80,30 @@
 #define __NUM(cmd) (HFI1_CMD_##cmd + 0xe0)
 
 #define HFI1_IOCTL_ASSIGN_CTXT \
-	_IOWR(IB_IOCTL_MAGIC, __NUM(ASSIGN_CTXT), struct hfi1_user_info)
+	_IOWR(RDMA_IOCTL_MAGIC, __NUM(ASSIGN_CTXT), struct hfi1_user_info)
 #define HFI1_IOCTL_CTXT_INFO \
-	_IOW(IB_IOCTL_MAGIC, __NUM(CTXT_INFO), struct hfi1_ctxt_info)
+	_IOW(RDMA_IOCTL_MAGIC, __NUM(CTXT_INFO), struct hfi1_ctxt_info)
 #define HFI1_IOCTL_USER_INFO \
-	_IOW(IB_IOCTL_MAGIC, __NUM(USER_INFO), struct hfi1_base_info)
+	_IOW(RDMA_IOCTL_MAGIC, __NUM(USER_INFO), struct hfi1_base_info)
 #define HFI1_IOCTL_TID_UPDATE \
-	_IOWR(IB_IOCTL_MAGIC, __NUM(TID_UPDATE), struct hfi1_tid_info)
+	_IOWR(RDMA_IOCTL_MAGIC, __NUM(TID_UPDATE), struct hfi1_tid_info)
 #define HFI1_IOCTL_TID_FREE \
-	_IOWR(IB_IOCTL_MAGIC, __NUM(TID_FREE), struct hfi1_tid_info)
+	_IOWR(RDMA_IOCTL_MAGIC, __NUM(TID_FREE), struct hfi1_tid_info)
 #define HFI1_IOCTL_CREDIT_UPD \
-	_IO(IB_IOCTL_MAGIC, __NUM(CREDIT_UPD))
+	_IO(RDMA_IOCTL_MAGIC, __NUM(CREDIT_UPD))
 #define HFI1_IOCTL_RECV_CTRL \
-	_IOW(IB_IOCTL_MAGIC, __NUM(RECV_CTRL), int)
+	_IOW(RDMA_IOCTL_MAGIC, __NUM(RECV_CTRL), int)
 #define HFI1_IOCTL_POLL_TYPE \
-	_IOW(IB_IOCTL_MAGIC, __NUM(POLL_TYPE), int)
+	_IOW(RDMA_IOCTL_MAGIC, __NUM(POLL_TYPE), int)
 #define HFI1_IOCTL_ACK_EVENT \
-	_IOW(IB_IOCTL_MAGIC, __NUM(ACK_EVENT), unsigned long)
+	_IOW(RDMA_IOCTL_MAGIC, __NUM(ACK_EVENT), unsigned long)
 #define HFI1_IOCTL_SET_PKEY \
-	_IOW(IB_IOCTL_MAGIC, __NUM(SET_PKEY), __u16)
+	_IOW(RDMA_IOCTL_MAGIC, __NUM(SET_PKEY), __u16)
 #define HFI1_IOCTL_CTXT_RESET \
-	_IO(IB_IOCTL_MAGIC, __NUM(CTXT_RESET))
+	_IO(RDMA_IOCTL_MAGIC, __NUM(CTXT_RESET))
 #define HFI1_IOCTL_TID_INVAL_READ \
-	_IOWR(IB_IOCTL_MAGIC, __NUM(TID_INVAL_READ), struct hfi1_tid_info)
+	_IOWR(RDMA_IOCTL_MAGIC, __NUM(TID_INVAL_READ), struct hfi1_tid_info)
 #define HFI1_IOCTL_GET_VERS \
-	_IOR(IB_IOCTL_MAGIC, __NUM(GET_VERS), int)
+	_IOR(RDMA_IOCTL_MAGIC, __NUM(GET_VERS), int)
 
 #endif /* RDMA_USER_IOCTL_H */

--
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Leon Romanovsky

author

2016-09-04 11:30 UTC #11 lore raw

PATCH v3 6/6 RDMA/core: Unify style of IOCTL commands

In reply to a message not in the archive

From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

MAD and HFI1 have different naming convention, this patch simplifies and unifies their defines and names.

As part of cleanup, the HFI1 _NUM() macro and command indexes were removed. It has a potential to break applications which use these defines directly, while shouldn't.

Signed-off-by Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Haggai Eran <haggaie-VPRAkNaXOzVWk0Htik3J/[email protected]>

Signed-off-by Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>

Reviewed-by Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>

---
 include/uapi/rdma/rdma_user_ioctl.h | 98 ++++++++++++++-----------------------
 1 file changed, 38 insertions(+), 60 deletions(-)

include/uapi/rdma/rdma_user_ioctl.h +38 ~~-60~~

index 7ecf8cd..9388125 100644
@@ -39,71 +39,49 @@
 #include <rdma/hfi/hfi1_ioctl.h>
 
 /* Documentation/ioctl/ioctl-number.txt */
-#define RDMA_IOCTL_MAGIC		0x1b
+#define RDMA_IOCTL_MAGIC	0x1b
 /* Legacy name, for user space application which already use it */
-#define IB_IOCTL_MAGIC			RDMA_IOCTL_MAGIC
-
-#define IB_USER_MAD_REGISTER_AGENT	_IOWR(RDMA_IOCTL_MAGIC, 1, \
-					      struct ib_user_mad_reg_req)
-
-#define IB_USER_MAD_UNREGISTER_AGENT	_IOW(RDMA_IOCTL_MAGIC, 2, __u32)
-
-#define IB_USER_MAD_ENABLE_PKEY		_IO(RDMA_IOCTL_MAGIC, 3)
-
-#define IB_USER_MAD_REGISTER_AGENT2     _IOWR(RDMA_IOCTL_MAGIC, 4, \
-					      struct ib_user_mad_reg_req2)
-
-/* User commands. */
-#define HFI1_CMD_ASSIGN_CTXT     1	/* allocate HFI and context */
-#define HFI1_CMD_CTXT_INFO       2	/* find out what resources we got */
-#define HFI1_CMD_USER_INFO       3	/* set up userspace */
-#define HFI1_CMD_TID_UPDATE      4	/* update expected TID entries */
-#define HFI1_CMD_TID_FREE        5	/* free expected TID entries */
-#define HFI1_CMD_CREDIT_UPD      6	/* force an update of PIO credit */
-
-#define HFI1_CMD_RECV_CTRL       8	/* control receipt of packets */
-#define HFI1_CMD_POLL_TYPE       9	/* set the kind of polling we want */
-#define HFI1_CMD_ACK_EVENT       10	/* ack & clear user status bits */
-#define HFI1_CMD_SET_PKEY        11     /* set context's pkey */
-#define HFI1_CMD_CTXT_RESET      12     /* reset context's HW send context */
-#define HFI1_CMD_TID_INVAL_READ  13     /* read TID cache invalidations */
-#define HFI1_CMD_GET_VERS	 14	/* get the version of the user cdev */
+#define IB_IOCTL_MAGIC		RDMA_IOCTL_MAGIC
 
 /*
- * User IOCTLs can not go above 128 if they do then see common.h and change the
- * base for the snoop ioctl
+ * General blocks assignments
+ * It is closed on purpose do not expose it it user space
+ * #define MAD_CMD_BASE		0x00
+ * #define HFI1_CMD_BAS		0xE0
  */
 
-/*
- * Make the ioctls occupy the last 0xf0-0xff portion of the IB range
- */
-#define __NUM(cmd) (HFI1_CMD_##cmd + 0xe0)
+/* MAD specific section */
+#define IB_USER_MAD_REGISTER_AGENT	_IOWR(RDMA_IOCTL_MAGIC, 0x01, struct ib_user_mad_reg_req)
+#define IB_USER_MAD_UNREGISTER_AGENT	_IOW(RDMA_IOCTL_MAGIC,  0x02, __u32)
+#define IB_USER_MAD_ENABLE_PKEY		_IO(RDMA_IOCTL_MAGIC,   0x03)
+#define IB_USER_MAD_REGISTER_AGENT2	_IOWR(RDMA_IOCTL_MAGIC, 0x04, struct ib_user_mad_reg_req2)
 
-#define HFI1_IOCTL_ASSIGN_CTXT \
-	_IOWR(RDMA_IOCTL_MAGIC, __NUM(ASSIGN_CTXT), struct hfi1_user_info)
-#define HFI1_IOCTL_CTXT_INFO \
-	_IOW(RDMA_IOCTL_MAGIC, __NUM(CTXT_INFO), struct hfi1_ctxt_info)
-#define HFI1_IOCTL_USER_INFO \
-	_IOW(RDMA_IOCTL_MAGIC, __NUM(USER_INFO), struct hfi1_base_info)
-#define HFI1_IOCTL_TID_UPDATE \
-	_IOWR(RDMA_IOCTL_MAGIC, __NUM(TID_UPDATE), struct hfi1_tid_info)
-#define HFI1_IOCTL_TID_FREE \
-	_IOWR(RDMA_IOCTL_MAGIC, __NUM(TID_FREE), struct hfi1_tid_info)
-#define HFI1_IOCTL_CREDIT_UPD \
-	_IO(RDMA_IOCTL_MAGIC, __NUM(CREDIT_UPD))
-#define HFI1_IOCTL_RECV_CTRL \
-	_IOW(RDMA_IOCTL_MAGIC, __NUM(RECV_CTRL), int)
-#define HFI1_IOCTL_POLL_TYPE \
-	_IOW(RDMA_IOCTL_MAGIC, __NUM(POLL_TYPE), int)
-#define HFI1_IOCTL_ACK_EVENT \
-	_IOW(RDMA_IOCTL_MAGIC, __NUM(ACK_EVENT), unsigned long)
-#define HFI1_IOCTL_SET_PKEY \
-	_IOW(RDMA_IOCTL_MAGIC, __NUM(SET_PKEY), __u16)
-#define HFI1_IOCTL_CTXT_RESET \
-	_IO(RDMA_IOCTL_MAGIC, __NUM(CTXT_RESET))
-#define HFI1_IOCTL_TID_INVAL_READ \
-	_IOWR(RDMA_IOCTL_MAGIC, __NUM(TID_INVAL_READ), struct hfi1_tid_info)
-#define HFI1_IOCTL_GET_VERS \
-	_IOR(RDMA_IOCTL_MAGIC, __NUM(GET_VERS), int)
+/* HFI specific section */
+/* allocate HFI and context */
+#define HFI1_IOCTL_ASSIGN_CTXT		_IOWR(RDMA_IOCTL_MAGIC, 0xE1, struct hfi1_user_info)
+/* find out what resources we got */
+#define HFI1_IOCTL_CTXT_INFO		_IOW(RDMA_IOCTL_MAGIC,  0xE2, struct hfi1_ctxt_info)
+/* set up userspace */
+#define HFI1_IOCTL_USER_INFO		_IOW(RDMA_IOCTL_MAGIC,  0xE3, struct hfi1_base_info)
+/* update expected TID entries */
+#define HFI1_IOCTL_TID_UPDATE		_IOWR(RDMA_IOCTL_MAGIC, 0xE4, struct hfi1_tid_info)
+/* free expected TID entries */
+#define HFI1_IOCTL_TID_FREE		_IOWR(RDMA_IOCTL_MAGIC, 0xE5, struct hfi1_tid_info)
+/* force an update of PIO credit */
+#define HFI1_IOCTL_CREDIT_UPD		_IO(RDMA_IOCTL_MAGIC,   0xE6)
+/* control receipt of packets */
+#define HFI1_IOCTL_RECV_CTRL		_IOW(RDMA_IOCTL_MAGIC,  0xE8, int)
+/* set the kind of polling we want */
+#define HFI1_IOCTL_POLL_TYPE		_IOW(RDMA_IOCTL_MAGIC,  0xE9, int)
+/* ack & clear user status bits */
+#define HFI1_IOCTL_ACK_EVENT		_IOW(RDMA_IOCTL_MAGIC,  0xEA, unsigned long)
+/* set context's pkey */
+#define HFI1_IOCTL_SET_PKEY		_IOW(RDMA_IOCTL_MAGIC,  0xEB, __u16)
+/* reset context's HW send context */
+#define HFI1_IOCTL_CTXT_RESET		_IO(RDMA_IOCTL_MAGIC,   0xEC)
+/* read TID cache invalidations */
+#define HFI1_IOCTL_TID_INVAL_READ	_IOWR(RDMA_IOCTL_MAGIC, 0xED, struct hfi1_tid_info)
+/* get the version of the user cdev */
+#define HFI1_IOCTL_GET_VERS		_IOR(RDMA_IOCTL_MAGIC,  0xEE, int)
 
 #endif /* RDMA_USER_IOCTL_H */

--
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 1472988635-31463-7-git-send-email-leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org

Jason Gunthorpe

Reviewed-by

2016-09-05 02:42 UTC #13 lore raw

Re: [PATCH rdma-next V3 6/6] RDMA/core: Unify style of IOCTL commands

Quoted 4 lines · Leon Romanovsky

From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>
MAD and HFI1 have different naming convention, this patch simplifies and unifies their defines and names.

Ah, well, nevermind then :)

Whole thing looks good to me:

Reviewed-by Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>

Thanks,
Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Leon Romanovsky

2016-09-05 05:33 UTC #14 lore raw

Re: [PATCH rdma-next V3 6/6] RDMA/core: Unify style of IOCTL commands

In reply to a message not in the archive

Quoted 17 lines · Leon Romanovsky

From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>
MAD and HFI1 have different naming convention, this patch simplifies and unifies their defines and names.
As part of cleanup, the HFI1 _NUM() macro and command indexes were removed. It has a potential to break applications which use these defines directly, while shouldn't.
Signed-off-by Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/[email protected]>
Signed-off-by Haggai Eran <haggaie-VPRAkNaXOzVWk0Htik3J/[email protected]>
Signed-off-by Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>
Reviewed-by Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
---
 include/uapi/rdma/rdma_user_ioctl.h | 98 ++++++++++++++-----------------------
 1 file changed, 38 insertions(+), 60 deletions(-)

<....>

Quoted 5 lines

/*
- * User IOCTLs can not go above 128 if they do then see common.h and change the
- * base for the snoop ioctl
+ * General blocks assignments
+ * It is closed on purpose do not expose it it user space

Doug,
Do you mind to change "it it" to be "it to" and save to everyone respin?

Thanks

A message that is not in the archive (sent off-list or not yet received): 20160905053342.GW21847-2ukJVAZIZ/[email protected]

Doug Ledford

2016-09-23 17:31 UTC #16 lore raw

Re: [PATCH rdma-next V3 6/6] RDMA/core: Unify style of IOCTL commands

Quoted 27 lines · Leon Romanovsky

Quoted 17 lines · Leon Romanovsky
From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>
MAD and HFI1 have different naming convention, this patch simplifies and unifies their defines and names.
As part of cleanup, the HFI1 _NUM() macro and command indexes were removed. It has a potential to break applications which use these defines directly, while shouldn't.
Signed-off-by Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/[email protected]>
Signed-off-by Haggai Eran <haggaie-VPRAkNaXOzVWk0Htik3J/[email protected]>
Signed-off-by Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>
Reviewed-by Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
---
 include/uapi/rdma/rdma_user_ioctl.h | 98 ++++++++++++++-----------------------
 1 file changed, 38 insertions(+), 60 deletions(-)
<....>
Quoted 5 lines
/*
- * User IOCTLs can not go above 128 if they do then see common.h and change the
- * base for the snoop ioctl
+ * General blocks assignments
+ * It is closed on purpose do not expose it it user space

Quoted 3 lines

Doug,
Do you mind to change "it it" to be "it to" and save to everyone respin?

Fixed.

Signature · 2 lines

Doug Ledford <[email protected]>
    GPG Key ID: 0E572FDD

Dalessandro, Dennis

2016-09-06 13:00 UTC #17 lore raw

Re: [PATCH rdma-next V3 6/6] RDMA/core: Unify style of IOCTL commands

In reply to a message not in the archive

T24gU3VuLCAyMDE2LTA5LTA0IGF0IDE0OjMwICswMzAwLCBMZW9uIFJvbWFub3Zza3kgd3JvdGU6 DQo+IEZyb206IExlb24gUm9tYW5vdnNreSA8bGVvbnJvQG1lbGxhbm94LmNvbT4NCj4gDQo+IE1B RCBhbmQgSEZJMSBoYXZlIGRpZmZlcmVudCBuYW1pbmcgY29udmVudGlvbiwgdGhpcyBwYXRjaA0K PiBzaW1wbGlmaWVzIGFuZCB1bmlmaWVzIHRoZWlyIGRlZmluZXMgYW5kIG5hbWVzLg0KPiANCj4g QXMgcGFydCBvZiBjbGVhbnVwLCB0aGUgSEZJMSBfTlVNKCkgbWFjcm8gYW5kIGNvbW1hbmQgaW5k ZXhlcw0KPiB3ZXJlIHJlbW92ZWQuIEl0IGhhcyBhIHBvdGVudGlhbCB0byBicmVhayBhcHBsaWNh dGlvbnMgd2hpY2ggdXNlDQo+IHRoZXNlIGRlZmluZXMgZGlyZWN0bHksIHdoaWxlIHNob3VsZG4n dC4NCg0KUGxlYXNlIGJlIGV4cGxpY2l0LiBJdCdzIG5vdCB0aGF0IGl0IGhhcyBwb3RlbnRpYWwu IEl0ICp3aWxsKiBicmVhaw0KYXBwbGljYXRpb25zIHdoaWNoIHVzZSB0aGVzZSBkZWZpbmVzIGRp cmVjdGx5Lg0KDQotRGVubnkNCg0K
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): [email protected]

Leon Romanovsky

author

2016-09-07 07:01 UTC #19 lore raw

Re: [PATCH rdma-next V3 6/6] RDMA/core: Unify style of IOCTL commands

Quoted 6 lines · Dennis

Quoted 5 lines · Leon Romanovsky
From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>
MAD and HFI1 have different naming convention, this patch simplifies and unifies their defines and names.

Quoted 6 lines

Quoted 3 lines
As part of cleanup, the HFI1 _NUM() macro and command indexes were removed. It has a potential to break applications which use these defines directly, while shouldn't.
Please be explicit. It's not that it has potential. It *will* break applications which use these defines directly.

Yes, you are right.

Doug,
What do you expect from me? Respin?

Thanks

Quoted 4 lines

-Denny
N?????r??y????b?X??ǧv?^?)޺{.n?+????{??ٚ?{ay?ʇڙ?,j??f???h???z??w??????j:+v???w?j?m????????zZ+?????ݢj"??!

A message that is not in the archive (sent off-list or not yet received): 20160907070109.GC3639-2ukJVAZIZ/[email protected]

Doug Ledford

2016-09-23 17:32 UTC #21 lore raw

Re: [PATCH rdma-next V3 6/6] RDMA/core: Unify style of IOCTL commands

Quoted 11 lines · Leon Romanovsky

Quoted 10 lines · Dennis
Quoted 8 lines · Leon Romanovsky
From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/[email protected]>
MAD and HFI1 have different naming convention, this patch simplifies and unifies their defines and names.
As part of cleanup, the HFI1 _NUM() macro and command indexes were removed. It has a potential to break applications which use these defines directly, while shouldn't.

Quoted 7 lines

Quoted 2 lines
Please be explicit. It's not that it has potential. It *will* break applications which use these defines directly.
Yes, you are right.
Doug,
What do you expect from me? Respin?

No, I reworded the commit message myself:

RDMA/core: Unify style of IOCTL commands

MAD and HFI1 have different naming convention, this patch simplifies and unifies their defines and names.

As part of cleanup, the HFI1 _NUM() macro and command indexes were removed (controversial). This will cause intentional (and arguably unnecessary) breakage to the PSM user space library.

Signature · 2 lines

Doug Ledford <[email protected]>
    GPG Key ID: 0E572FDD

Doug Ledford

2016-09-23 17:34 UTC #22 lore raw

In reply to a message not in the archive

Quoted 58 lines · Leon Romanovsky

Hi Doug,
This patch set refactors RDMA IOCTL declarations, unifies their location, moves all current IOCTL exporters to one place and renames them to fix the differences in coding style.
IOCTLs commands need to export the size of their data for their transfer which is usually declared as a struct defined in relevant submodule. These structures should be visible at the linkage stage.
There are number of possible solutions to overcome this limitations.
One of the options is to declare anonymous variable which will be visible during the linkage. As a downside, such option won't allow direct use of this header by other parts of code without their specific includes of other logically unrelated submodules.
Another possible option will be to move declarations of that structures to that common header file. Such move will cause to bloat this file with different submodules declarations.
Our decision was to choose direct include of other exported files with declarations of such structures to simplify future usage of this file. There is no change for legacy applications and libraries.
Changes v2->v3:
* Added Jason's Reviewed-by
* Open-ceded ioctl indexes
Changes v0->v2 (v1 was skipped):
* Update commit messages of patch number 3 and 6.
Available in the "topic/ioctl-header" topic branch of this git repo:
git://git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git
Or for browsing:
https://git.kernel.org/cgit/linux/kernel/git/leon/linux-rdma.git/log/?h=topic/ioctl-header
Thanks
Leon Romanovsky (6):
RDMA/core: Commonize RDMA IOCTL declarations location RDMA/core: Move legacy MAD IOCTL declarations to common file RDMA/hfi1: Avoid redeclaration error
RDMA/core: Move HFI1 IOCTL declarations to common file RDMA/core: Rename RDMA magic number
RDMA/core: Unify style of IOCTL commands
 include/uapi/rdma/Kbuild            |   1 +
 include/uapi/rdma/hfi/Kbuild        |   1 +
 include/uapi/rdma/hfi/hfi1_ioctl.h  | 173 +++++++++++++++++++++++++++++++++++
 include/uapi/rdma/hfi/hfi1_user.h   | 175 +-----------------------------------
 include/uapi/rdma/ib_user_mad.h     |  14 +--
 include/uapi/rdma/rdma_user_ioctl.h |  87 ++++++++++++++++++
 6 files changed, 264 insertions(+), 187 deletions(-)
 create mode 100644 include/uapi/rdma/hfi/hfi1_ioctl.h
 create mode 100644 include/uapi/rdma/rdma_user_ioctl.h

Quoted 3 lines

--
2.7.4

Series applied.

Signature · 2 lines

Doug Ledford <[email protected]>
    GPG Key ID: 0E572FDD