Merge branch '32compat'

The design of the uAPI had intended all structs to share the same layout on 32
and 64 bit compiles. Unfortunately over the years some errors have crept in.

This series fixes all the incompatabilities. It goes along with a userspace
rdma-core series that causes the providers to use these structs directly and
then does various self-checks on the command formation.

Those checks were combined with output from pahole on 32 and 64 bit compiles
to confirm that the structure layouts are the same.

This series does not make implicit padding explicit, as long as the implicit
padding is the same on 32 and 64 bit compiles.

Finally, the issue is put to rest by using __aligned_u64 in the uapi headers,
if new code copies that type, and is checked in userspace, it is unlikely we
will see problems in future.

There are two patches that break the ABI for a 32 bit kernel, one for rxe and
one for mlx4. Both patches have notes, but the overall feeling from Doug and I
is that providing compat is just too difficult and not necessary since there
is no real user of a 32 bit userspace and 32 bit kernel for various good
reasons.

The 32 bit userspace / 64 bit kernel case however does seem to have some real
users and does need to work as designed.

* 32compat:
  RDMA: Change all uapi headers to use __aligned_u64 instead of __u64
  RDMA/rxe: Fix uABI structure layouts for 32/64 compat
  RDMA/mlx4: Fix uABI structure layouts for 32/64 compat
  RDMA/qedr: Fix uABI structure layouts for 32/64 compat
  RDMA/ucma: Fix uABI structure layouts for 32/64 compat
  RDMA: Remove minor pahole differences between 32/64

19 files changed, 298 insertions, 277 deletions

diff --git a/include/uapi/rdma/bnxt_re-abi.h b/include/uapi/rdma/bnxt_re-abi.h
index 2d3c9aac661a..a7a6111e50c7 100644
--- a/include/uapi/rdma/bnxt_re-abi.h
+++ b/include/uapi/rdma/bnxt_re-abi.h
@@ -65,8 +65,8 @@ struct bnxt_re_pd_resp {
 } __attribute__((packed, aligned(4)));
 
 struct bnxt_re_cq_req {
-	__u64 cq_va;
-	__u64 cq_handle;
+	__aligned_u64 cq_va;
+	__aligned_u64 cq_handle;
 };
 
 struct bnxt_re_cq_resp {
@@ -77,9 +77,9 @@ struct bnxt_re_cq_resp {
 };
 
 struct bnxt_re_qp_req {
-	__u64 qpsva;
-	__u64 qprva;
-	__u64 qp_handle;
+	__aligned_u64 qpsva;
+	__aligned_u64 qprva;
+	__aligned_u64 qp_handle;
 };
 
 struct bnxt_re_qp_resp {
@@ -88,8 +88,8 @@ struct bnxt_re_qp_resp {
 };
 
 struct bnxt_re_srq_req {
-	__u64 srqva;
-	__u64 srq_handle;
+	__aligned_u64 srqva;
+	__aligned_u64 srq_handle;
 };
 
 struct bnxt_re_srq_resp {
diff --git a/include/uapi/rdma/cxgb3-abi.h b/include/uapi/rdma/cxgb3-abi.h
index 17116c1c7925..9acb4b7a6246 100644
--- a/include/uapi/rdma/cxgb3-abi.h
+++ b/include/uapi/rdma/cxgb3-abi.h
@@ -41,21 +41,21 @@
  * Make sure that all structs defined in this file remain laid out so
  * that they pack the same way on 32-bit and 64-bit architectures (to
  * avoid incompatibility between 32-bit userspace and 64-bit kernels).
- * In particular do not use pointer types -- pass pointers in __u64
+ * In particular do not use pointer types -- pass pointers in __aligned_u64
  * instead.
  */
 struct iwch_create_cq_req {
-	__u64 user_rptr_addr;
+	__aligned_u64 user_rptr_addr;
 };
 
 struct iwch_create_cq_resp_v0 {
-	__u64 key;
+	__aligned_u64 key;
 	__u32 cqid;
 	__u32 size_log2;
 };
 
 struct iwch_create_cq_resp {
-	__u64 key;
+	__aligned_u64 key;
 	__u32 cqid;
 	__u32 size_log2;
 	__u32 memsize;
@@ -63,8 +63,8 @@ struct iwch_create_cq_resp {
 };
 
 struct iwch_create_qp_resp {
-	__u64 key;
-	__u64 db_key;
+	__aligned_u64 key;
+	__aligned_u64 db_key;
 	__u32 qpid;
 	__u32 size_log2;
 	__u32 sq_size_log2;
diff --git a/include/uapi/rdma/cxgb4-abi.h b/include/uapi/rdma/cxgb4-abi.h
index c398a1ee8d00..1fefd0140c26 100644
--- a/include/uapi/rdma/cxgb4-abi.h
+++ b/include/uapi/rdma/cxgb4-abi.h
@@ -41,13 +41,13 @@
  * Make sure that all structs defined in this file remain laid out so
  * that they pack the same way on 32-bit and 64-bit architectures (to
  * avoid incompatibility between 32-bit userspace and 64-bit kernels).
- * In particular do not use pointer types -- pass pointers in __u64
+ * In particular do not use pointer types -- pass pointers in __aligned_u64
  * instead.
  */
 struct c4iw_create_cq_resp {
-	__u64 key;
-	__u64 gts_key;
-	__u64 memsize;
+	__aligned_u64 key;
+	__aligned_u64 gts_key;
+	__aligned_u64 memsize;
 	__u32 cqid;
 	__u32 size;
 	__u32 qid_mask;
@@ -59,13 +59,13 @@ enum {
 };
 
 struct c4iw_create_qp_resp {
-	__u64 ma_sync_key;
-	__u64 sq_key;
-	__u64 rq_key;
-	__u64 sq_db_gts_key;
-	__u64 rq_db_gts_key;
-	__u64 sq_memsize;
-	__u64 rq_memsize;
+	__aligned_u64 ma_sync_key;
+	__aligned_u64 sq_key;
+	__aligned_u64 rq_key;
+	__aligned_u64 sq_db_gts_key;
+	__aligned_u64 rq_db_gts_key;
+	__aligned_u64 sq_memsize;
+	__aligned_u64 rq_memsize;
 	__u32 sqid;
 	__u32 rqid;
 	__u32 sq_size;
@@ -75,7 +75,7 @@ struct c4iw_create_qp_resp {
 };
 
 struct c4iw_alloc_ucontext_resp {
-	__u64 status_page_key;
+	__aligned_u64 status_page_key;
 	__u32 status_page_size;
 	__u32 reserved; /* explicit padding (optional for i386) */
 };
diff --git a/include/uapi/rdma/hfi/hfi1_ioctl.h b/include/uapi/rdma/hfi/hfi1_ioctl.h
index 9de78c5ee913..8f3d9fe7b141 100644
--- a/include/uapi/rdma/hfi/hfi1_ioctl.h
+++ b/include/uapi/rdma/hfi/hfi1_ioctl.h
@@ -79,7 +79,7 @@ struct hfi1_user_info {
 };
 
 struct hfi1_ctxt_info {
-	__u64 runtime_flags;    /* chip/drv runtime flags (HFI1_CAP_*) */
+	__aligned_u64 runtime_flags;    /* chip/drv runtime flags (HFI1_CAP_*) */
 	__u32 rcvegr_size;      /* size of each eager buffer */
 	__u16 num_active;       /* number of active units */
 	__u16 unit;             /* unit (chip) assigned to caller */
@@ -98,9 +98,9 @@ struct hfi1_ctxt_info {
 
 struct hfi1_tid_info {
 	/* virtual address of first page in transfer */
-	__u64 vaddr;
+	__aligned_u64 vaddr;
 	/* pointer to tid array. this array is big enough */
-	__u64 tidlist;
+	__aligned_u64 tidlist;
 	/* number of tids programmed by this request */
 	__u32 tidcnt;
 	/* length of transfer buffer programmed by this request */
@@ -131,23 +131,23 @@ struct hfi1_base_info {
 	 */
 	__u32 bthqp;
 	/* PIO credit return address, */
-	__u64 sc_credits_addr;
+	__aligned_u64 sc_credits_addr;
 	/*
 	 * Base address of write-only pio buffers for this process.
 	 * Each buffer has sendpio_credits*64 bytes.
 	 */
-	__u64 pio_bufbase_sop;
+	__aligned_u64 pio_bufbase_sop;
 	/*
 	 * Base address of write-only pio buffers for this process.
 	 * Each buffer has sendpio_credits*64 bytes.
 	 */
-	__u64 pio_bufbase;
+	__aligned_u64 pio_bufbase;
 	/* address where receive buffer queue is mapped into */
-	__u64 rcvhdr_bufbase;
+	__aligned_u64 rcvhdr_bufbase;
 	/* base address of Eager receive buffers. */
-	__u64 rcvegr_bufbase;
+	__aligned_u64 rcvegr_bufbase;
 	/* base address of SDMA completion ring */
-	__u64 sdma_comp_bufbase;
+	__aligned_u64 sdma_comp_bufbase;
 	/*
 	 * User register base for init code, not to be used directly by
 	 * protocol or applications.  Always maps real chip register space.
@@ -155,20 +155,20 @@ struct hfi1_base_info {
 	 * ur_rcvhdrhead, ur_rcvhdrtail, ur_rcvegrhead, ur_rcvegrtail,
 	 * ur_rcvtidflow
 	 */
-	__u64 user_regbase;
+	__aligned_u64 user_regbase;
 	/* notification events */
-	__u64 events_bufbase;
+	__aligned_u64 events_bufbase;
 	/* status page */
-	__u64 status_bufbase;
+	__aligned_u64 status_bufbase;
 	/* rcvhdrtail update */
-	__u64 rcvhdrtail_base;
+	__aligned_u64 rcvhdrtail_base;
 	/*
 	 * shared memory pages for subctxts if ctxt is shared; these cover
 	 * all the processes in the group sharing a single context.
 	 * all have enough space for the num_subcontexts value on this job.
 	 */
-	__u64 subctxt_uregbase;
-	__u64 subctxt_rcvegrbuf;
-	__u64 subctxt_rcvhdrbuf;
+	__aligned_u64 subctxt_uregbase;
+	__aligned_u64 subctxt_rcvegrbuf;
+	__aligned_u64 subctxt_rcvhdrbuf;
 };
 #endif /* _LINIUX__HFI1_IOCTL_H */
diff --git a/include/uapi/rdma/hfi/hfi1_user.h b/include/uapi/rdma/hfi/hfi1_user.h
index 43b46bf6f8bb..c6a984c0c881 100644
--- a/include/uapi/rdma/hfi/hfi1_user.h
+++ b/include/uapi/rdma/hfi/hfi1_user.h
@@ -177,8 +177,8 @@ struct hfi1_sdma_comp_entry {
  * Device status and notifications from driver to user-space.
  */
 struct hfi1_status {
-	__u64 dev;      /* device/hw status bits */
-	__u64 port;     /* port state and status bits */
+	__aligned_u64 dev;      /* device/hw status bits */
+	__aligned_u64 port;     /* port state and status bits */
 	char freezemsg[0];
 };
 
diff --git a/include/uapi/rdma/hns-abi.h b/include/uapi/rdma/hns-abi.h
index aa774985a0c7..7092c8de4bd8 100644
--- a/include/uapi/rdma/hns-abi.h
+++ b/include/uapi/rdma/hns-abi.h
@@ -37,18 +37,18 @@
 #include <linux/types.h>
 
 struct hns_roce_ib_create_cq {
-	__u64   buf_addr;
-	__u64	db_addr;
+	__aligned_u64 buf_addr;
+	__aligned_u64 db_addr;
 };
 
 struct hns_roce_ib_create_cq_resp {
-	__u64	cqn; /* Only 32 bits used, 64 for compat */
-	__u64	cap_flags;
+	__aligned_u64 cqn; /* Only 32 bits used, 64 for compat */
+	__aligned_u64 cap_flags;
 };
 
 struct hns_roce_ib_create_qp {
-	__u64	buf_addr;
-	__u64   db_addr;
+	__aligned_u64 buf_addr;
+	__aligned_u64 db_addr;
 	__u8    log_sq_bb_count;
 	__u8    log_sq_stride;
 	__u8    sq_no_prefetch;
@@ -56,7 +56,7 @@ struct hns_roce_ib_create_qp {
 };
 
 struct hns_roce_ib_create_qp_resp {
-	__u64	cap_flags;
+	__aligned_u64 cap_flags;
 };
 
 struct hns_roce_ib_alloc_ucontext_resp {
diff --git a/include/uapi/rdma/i40iw-abi.h b/include/uapi/rdma/i40iw-abi.h
index bfc3aaf2e56a..79890baa6fdb 100644
--- a/include/uapi/rdma/i40iw-abi.h
+++ b/include/uapi/rdma/i40iw-abi.h
@@ -61,17 +61,17 @@ struct i40iw_alloc_pd_resp {
 };
 
 struct i40iw_create_cq_req {
-	__u64 user_cq_buffer;
-	__u64 user_shadow_area;
+	__aligned_u64 user_cq_buffer;
+	__aligned_u64 user_shadow_area;
 };
 
 struct i40iw_create_qp_req {
-	__u64 user_wqe_buffers;
-	__u64 user_compl_ctx;
+	__aligned_u64 user_wqe_buffers;
+	__aligned_u64 user_compl_ctx;
 
 	/* UDA QP PHB */
-	__u64 user_sq_phb;	/* place for VA of the sq phb buff */
-	__u64 user_rq_phb;	/* place for VA of the rq phb buff */
+	__aligned_u64 user_sq_phb;	/* place for VA of the sq phb buff */
+	__aligned_u64 user_rq_phb;	/* place for VA of the rq phb buff */
 };
 
 enum i40iw_memreg_type {
diff --git a/include/uapi/rdma/ib_user_cm.h b/include/uapi/rdma/ib_user_cm.h
index f4041bdc4d08..4a8f9562f7cd 100644
--- a/include/uapi/rdma/ib_user_cm.h
+++ b/include/uapi/rdma/ib_user_cm.h
@@ -73,8 +73,8 @@ struct ib_ucm_cmd_hdr {
 };
 
 struct ib_ucm_create_id {
-	__u64 uid;
-	__u64 response;
+	__aligned_u64 uid;
+	__aligned_u64 response;
 };
 
 struct ib_ucm_create_id_resp {
@@ -82,7 +82,7 @@ struct ib_ucm_create_id_resp {
 };
 
 struct ib_ucm_destroy_id {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 id;
 	__u32 reserved;
 };
@@ -92,7 +92,7 @@ struct ib_ucm_destroy_id_resp {
 };
 
 struct ib_ucm_attr_id {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 id;
 	__u32 reserved;
 };
@@ -105,7 +105,7 @@ struct ib_ucm_attr_id_resp {
 };
 
 struct ib_ucm_init_qp_attr {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 id;
 	__u32 qp_state;
 };
@@ -123,7 +123,7 @@ struct ib_ucm_notify {
 };
 
 struct ib_ucm_private_data {
-	__u64 data;
+	__aligned_u64 data;
 	__u32 id;
 	__u8  len;
 	__u8  reserved[3];
@@ -135,9 +135,9 @@ struct ib_ucm_req {
 	__u32 qp_type;
 	__u32 psn;
 	__be64 sid;
-	__u64 data;
-	__u64 primary_path;
-	__u64 alternate_path;
+	__aligned_u64 data;
+	__aligned_u64 primary_path;
+	__aligned_u64 alternate_path;
 	__u8  len;
 	__u8  peer_to_peer;
 	__u8  responder_resources;
@@ -153,8 +153,8 @@ struct ib_ucm_req {
 };
 
 struct ib_ucm_rep {
-	__u64 uid;
-	__u64 data;
+	__aligned_u64 uid;
+	__aligned_u64 data;
 	__u32 id;
 	__u32 qpn;
 	__u32 psn;
@@ -172,15 +172,15 @@ struct ib_ucm_rep {
 struct ib_ucm_info {
 	__u32 id;
 	__u32 status;
-	__u64 info;
-	__u64 data;
+	__aligned_u64 info;
+	__aligned_u64 data;
 	__u8  info_len;
 	__u8  data_len;
 	__u8  reserved[6];
 };
 
 struct ib_ucm_mra {
-	__u64 data;
+	__aligned_u64 data;
 	__u32 id;
 	__u8  len;
 	__u8  timeout;
@@ -188,8 +188,8 @@ struct ib_ucm_mra {
 };
 
 struct ib_ucm_lap {
-	__u64 path;
-	__u64 data;
+	__aligned_u64 path;
+	__aligned_u64 data;
 	__u32 id;
 	__u8  len;
 	__u8  reserved[3];
@@ -199,8 +199,8 @@ struct ib_ucm_sidr_req {
 	__u32 id;
 	__u32 timeout;
 	__be64 sid;
-	__u64 data;
-	__u64 path;
+	__aligned_u64 data;
+	__aligned_u64 path;
 	__u16 reserved_pkey;
 	__u8  len;
 	__u8  max_cm_retries;
@@ -212,8 +212,8 @@ struct ib_ucm_sidr_rep {
 	__u32 qpn;
 	__u32 qkey;
 	__u32 status;
-	__u64 info;
-	__u64 data;
+	__aligned_u64 info;
+	__aligned_u64 data;
 	__u8  info_len;
 	__u8  data_len;
 	__u8  reserved[6];
@@ -222,9 +222,9 @@ struct ib_ucm_sidr_rep {
  * event notification ABI structures.
  */
 struct ib_ucm_event_get {
-	__u64 response;
-	__u64 data;
-	__u64 info;
+	__aligned_u64 response;
+	__aligned_u64 data;
+	__aligned_u64 info;
 	__u8  data_len;
 	__u8  info_len;
 	__u8  reserved[6];
@@ -303,7 +303,7 @@ struct ib_ucm_sidr_rep_event_resp {
 #define IB_UCM_PRES_ALTERNATE 0x08
 
 struct ib_ucm_event_resp {
-	__u64 uid;
+	__aligned_u64 uid;
 	__u32 id;
 	__u32 event;
 	__u32 present;
diff --git a/include/uapi/rdma/ib_user_mad.h b/include/uapi/rdma/ib_user_mad.h
index 330a3c5f1aa8..ef92118dad97 100644
--- a/include/uapi/rdma/ib_user_mad.h
+++ b/include/uapi/rdma/ib_user_mad.h
@@ -143,7 +143,7 @@ struct ib_user_mad_hdr {
  */
 struct ib_user_mad {
 	struct ib_user_mad_hdr hdr;
-	__u64	data[0];
+	__aligned_u64	data[0];
 };
 
 /*
@@ -225,7 +225,7 @@ struct ib_user_mad_reg_req2 {
 	__u8	mgmt_class_version;
 	__u16   res;
 	__u32   flags;
-	__u64   method_mask[2];
+	__aligned_u64 method_mask[2];
 	__u32   oui;
 	__u8	rmpp_version;
 	__u8	reserved[3];
diff --git a/include/uapi/rdma/ib_user_verbs.h b/include/uapi/rdma/ib_user_verbs.h
index d56fba09dc8a..aa0615105563 100644
--- a/include/uapi/rdma/ib_user_verbs.h
+++ b/include/uapi/rdma/ib_user_verbs.h
@@ -117,13 +117,13 @@ enum {
  */
 
 struct ib_uverbs_async_event_desc {
-	__u64 element;
+	__aligned_u64 element;
 	__u32 event_type;	/* enum ib_event_type */
 	__u32 reserved;
 };
 
 struct ib_uverbs_comp_event_desc {
-	__u64 cq_handle;
+	__aligned_u64 cq_handle;
 };
 
 struct ib_uverbs_cq_moderation_caps {
@@ -150,15 +150,15 @@ struct ib_uverbs_cmd_hdr {
 };
 
 struct ib_uverbs_ex_cmd_hdr {
-	__u64 response;
+	__aligned_u64 response;
 	__u16 provider_in_words;
 	__u16 provider_out_words;
 	__u32 cmd_hdr_reserved;
 };
 
 struct ib_uverbs_get_context {
-	__u64 response;
-	__u64 driver_data[0];
+	__aligned_u64 response;
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_get_context_resp {
@@ -167,16 +167,16 @@ struct ib_uverbs_get_context_resp {
 };
 
 struct ib_uverbs_query_device {
-	__u64 response;
-	__u64 driver_data[0];
+	__aligned_u64 response;
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_query_device_resp {
-	__u64 fw_ver;
+	__aligned_u64 fw_ver;
 	__be64 node_guid;
 	__be64 sys_image_guid;
-	__u64 max_mr_size;
-	__u64 page_size_cap;
+	__aligned_u64 max_mr_size;
+	__aligned_u64 page_size_cap;
 	__u32 vendor_id;
 	__u32 vendor_part_id;
 	__u32 hw_ver;
@@ -221,7 +221,7 @@ struct ib_uverbs_ex_query_device {
 };
 
 struct ib_uverbs_odp_caps {
-	__u64 general_caps;
+	__aligned_u64 general_caps;
 	struct {
 		__u32 rc_odp_caps;
 		__u32 uc_odp_caps;
@@ -260,9 +260,9 @@ struct ib_uverbs_ex_query_device_resp {
 	__u32 comp_mask;
 	__u32 response_length;
 	struct ib_uverbs_odp_caps odp_caps;
-	__u64 timestamp_mask;
-	__u64 hca_core_clock; /* in KHZ */
-	__u64 device_cap_flags_ex;
+	__aligned_u64 timestamp_mask;
+	__aligned_u64 hca_core_clock; /* in KHZ */
+	__aligned_u64 device_cap_flags_ex;
 	struct ib_uverbs_rss_caps rss_caps;
 	__u32  max_wq_type_rq;
 	__u32 raw_packet_caps;
@@ -271,10 +271,10 @@ struct ib_uverbs_ex_query_device_resp {
 };
 
 struct ib_uverbs_query_port {
-	__u64 response;
+	__aligned_u64 response;
 	__u8  port_num;
 	__u8  reserved[7];
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_query_port_resp {
@@ -302,8 +302,8 @@ struct ib_uverbs_query_port_resp {
 };
 
 struct ib_uverbs_alloc_pd {
-	__u64 response;
-	__u64 driver_data[0];
+	__aligned_u64 response;
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_alloc_pd_resp {
@@ -315,10 +315,10 @@ struct ib_uverbs_dealloc_pd {
 };
 
 struct ib_uverbs_open_xrcd {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 fd;
 	__u32 oflags;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_open_xrcd_resp {
@@ -330,13 +330,13 @@ struct ib_uverbs_close_xrcd {
 };
 
 struct ib_uverbs_reg_mr {
-	__u64 response;
-	__u64 start;
-	__u64 length;
-	__u64 hca_va;
+	__aligned_u64 response;
+	__aligned_u64 start;
+	__aligned_u64 length;
+	__aligned_u64 hca_va;
 	__u32 pd_handle;
 	__u32 access_flags;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_reg_mr_resp {
@@ -346,12 +346,12 @@ struct ib_uverbs_reg_mr_resp {
 };
 
 struct ib_uverbs_rereg_mr {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 mr_handle;
 	__u32 flags;
-	__u64 start;
-	__u64 length;
-	__u64 hca_va;
+	__aligned_u64 start;
+	__aligned_u64 length;
+	__aligned_u64 hca_va;
 	__u32 pd_handle;
 	__u32 access_flags;
 };
@@ -366,7 +366,7 @@ struct ib_uverbs_dereg_mr {
 };
 
 struct ib_uverbs_alloc_mw {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 pd_handle;
 	__u8  mw_type;
 	__u8  reserved[3];
@@ -382,7 +382,7 @@ struct ib_uverbs_dealloc_mw {
 };
 
 struct ib_uverbs_create_comp_channel {
-	__u64 response;
+	__aligned_u64 response;
 };
 
 struct ib_uverbs_create_comp_channel_resp {
@@ -390,13 +390,13 @@ struct ib_uverbs_create_comp_channel_resp {
 };
 
 struct ib_uverbs_create_cq {
-	__u64 response;
-	__u64 user_handle;
+	__aligned_u64 response;
+	__aligned_u64 user_handle;
 	__u32 cqe;
 	__u32 comp_vector;
 	__s32 comp_channel;
 	__u32 reserved;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 enum ib_uverbs_ex_create_cq_flags {
@@ -405,7 +405,7 @@ enum ib_uverbs_ex_create_cq_flags {
 };
 
 struct ib_uverbs_ex_create_cq {
-	__u64 user_handle;
+	__aligned_u64 user_handle;
 	__u32 cqe;
 	__u32 comp_vector;
 	__s32 comp_channel;
@@ -426,26 +426,26 @@ struct ib_uverbs_ex_create_cq_resp {
 };
 
 struct ib_uverbs_resize_cq {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 cq_handle;
 	__u32 cqe;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_resize_cq_resp {
 	__u32 cqe;
 	__u32 reserved;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_poll_cq {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 cq_handle;
 	__u32 ne;
 };
 
 struct ib_uverbs_wc {
-	__u64 wr_id;
+	__aligned_u64 wr_id;
 	__u32 status;
 	__u32 opcode;
 	__u32 vendor_err;
@@ -477,7 +477,7 @@ struct ib_uverbs_req_notify_cq {
 };
 
 struct ib_uverbs_destroy_cq {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 cq_handle;
 	__u32 reserved;
 };
@@ -546,8 +546,8 @@ struct ib_uverbs_qp_attr {
 };
 
 struct ib_uverbs_create_qp {
-	__u64 response;
-	__u64 user_handle;
+	__aligned_u64 response;
+	__aligned_u64 user_handle;
 	__u32 pd_handle;
 	__u32 send_cq_handle;
 	__u32 recv_cq_handle;
@@ -561,7 +561,7 @@ struct ib_uverbs_create_qp {
 	__u8  qp_type;
 	__u8  is_srq;
 	__u8  reserved;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 enum ib_uverbs_create_qp_mask {
@@ -587,7 +587,7 @@ enum {
 };
 
 struct ib_uverbs_ex_create_qp {
-	__u64 user_handle;
+	__aligned_u64 user_handle;
 	__u32 pd_handle;
 	__u32 send_cq_handle;
 	__u32 recv_cq_handle;
@@ -608,13 +608,13 @@ struct ib_uverbs_ex_create_qp {
 };
 
 struct ib_uverbs_open_qp {
-	__u64 response;
-	__u64 user_handle;
+	__aligned_u64 response;
+	__aligned_u64 user_handle;
 	__u32 pd_handle;
 	__u32 qpn;
 	__u8  qp_type;
 	__u8  reserved[7];
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 /* also used for open response */
@@ -655,10 +655,10 @@ struct ib_uverbs_qp_dest {
 };
 
 struct ib_uverbs_query_qp {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 qp_handle;
 	__u32 attr_mask;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_query_qp_resp {
@@ -692,7 +692,7 @@ struct ib_uverbs_query_qp_resp {
 	__u8  alt_timeout;
 	__u8  sq_sig_all;
 	__u8  reserved[5];
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_modify_qp {
@@ -722,7 +722,7 @@ struct ib_uverbs_modify_qp {
 	__u8  alt_port_num;
 	__u8  alt_timeout;
 	__u8  reserved[2];
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_ex_modify_qp {
@@ -740,7 +740,7 @@ struct ib_uverbs_ex_modify_qp_resp {
 };
 
 struct ib_uverbs_destroy_qp {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 qp_handle;
 	__u32 reserved;
 };
@@ -756,13 +756,13 @@ struct ib_uverbs_destroy_qp_resp {
  * document the ABI.
  */
 struct ib_uverbs_sge {
-	__u64 addr;
+	__aligned_u64 addr;
 	__u32 length;
 	__u32 lkey;
 };
 
 struct ib_uverbs_send_wr {
-	__u64 wr_id;
+	__aligned_u64 wr_id;
 	__u32 num_sge;
 	__u32 opcode;
 	__u32 send_flags;
@@ -772,14 +772,14 @@ struct ib_uverbs_send_wr {
 	} ex;
 	union {
 		struct {
-			__u64 remote_addr;
+			__aligned_u64 remote_addr;
 			__u32 rkey;
 			__u32 reserved;
 		} rdma;
 		struct {
-			__u64 remote_addr;
-			__u64 compare_add;
-			__u64 swap;
+			__aligned_u64 remote_addr;
+			__aligned_u64 compare_add;
+			__aligned_u64 swap;
 			__u32 rkey;
 			__u32 reserved;
 		} atomic;
@@ -793,7 +793,7 @@ struct ib_uverbs_send_wr {
 };
 
 struct ib_uverbs_post_send {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 qp_handle;
 	__u32 wr_count;
 	__u32 sge_count;
@@ -806,13 +806,13 @@ struct ib_uverbs_post_send_resp {
 };
 
 struct ib_uverbs_recv_wr {
-	__u64 wr_id;
+	__aligned_u64 wr_id;
 	__u32 num_sge;
 	__u32 reserved;
 };
 
 struct ib_uverbs_post_recv {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 qp_handle;
 	__u32 wr_count;
 	__u32 sge_count;
@@ -825,7 +825,7 @@ struct ib_uverbs_post_recv_resp {
 };
 
 struct ib_uverbs_post_srq_recv {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 srq_handle;
 	__u32 wr_count;
 	__u32 sge_count;
@@ -838,8 +838,8 @@ struct ib_uverbs_post_srq_recv_resp {
 };
 
 struct ib_uverbs_create_ah {
-	__u64 response;
-	__u64 user_handle;
+	__aligned_u64 response;
+	__aligned_u64 user_handle;
 	__u32 pd_handle;
 	__u32 reserved;
 	struct ib_uverbs_ah_attr attr;
@@ -858,7 +858,7 @@ struct ib_uverbs_attach_mcast {
 	__u32 qp_handle;
 	__u16 mlid;
 	__u16 reserved;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_detach_mcast {
@@ -866,7 +866,7 @@ struct ib_uverbs_detach_mcast {
 	__u32 qp_handle;
 	__u16 mlid;
 	__u16 reserved;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_flow_spec_hdr {
@@ -874,7 +874,7 @@ struct ib_uverbs_flow_spec_hdr {
 	__u16 size;
 	__u16 reserved;
 	/* followed by flow_spec */
-	__u64 flow_spec_data[0];
+	__aligned_u64 flow_spec_data[0];
 };
 
 struct ib_uverbs_flow_eth_filter {
@@ -1033,18 +1033,18 @@ struct ib_uverbs_destroy_flow  {
 };
 
 struct ib_uverbs_create_srq {
-	__u64 response;
-	__u64 user_handle;
+	__aligned_u64 response;
+	__aligned_u64 user_handle;
 	__u32 pd_handle;
 	__u32 max_wr;
 	__u32 max_sge;
 	__u32 srq_limit;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_create_xsrq {
-	__u64 response;
-	__u64 user_handle;
+	__aligned_u64 response;
+	__aligned_u64 user_handle;
 	__u32 srq_type;
 	__u32 pd_handle;
 	__u32 max_wr;
@@ -1053,7 +1053,7 @@ struct ib_uverbs_create_xsrq {
 	__u32 max_num_tags;
 	__u32 xrcd_handle;
 	__u32 cq_handle;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_create_srq_resp {
@@ -1068,14 +1068,14 @@ struct ib_uverbs_modify_srq {
 	__u32 attr_mask;
 	__u32 max_wr;
 	__u32 srq_limit;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_query_srq {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 srq_handle;
 	__u32 reserved;
-	__u64 driver_data[0];
+	__aligned_u64 driver_data[0];
 };
 
 struct ib_uverbs_query_srq_resp {
@@ -1086,7 +1086,7 @@ struct ib_uverbs_query_srq_resp {
 };
 
 struct ib_uverbs_destroy_srq {
-	__u64 response;
+	__aligned_u64 response;
 	__u32 srq_handle;
 	__u32 reserved;
 };
@@ -1098,7 +1098,7 @@ struct ib_uverbs_destroy_srq_resp {
 struct ib_uverbs_ex_create_wq  {
 	__u32 comp_mask;
 	__u32 wq_type;
-	__u64 user_handle;
+	__aligned_u64 user_handle;
 	__u32 pd_handle;
 	__u32 cq_handle;
 	__u32 max_wr;
diff --git a/include/uapi/rdma/mlx4-abi.h b/include/uapi/rdma/mlx4-abi.h
index a448abd07052..04f64bc4045f 100644
--- a/include/uapi/rdma/mlx4-abi.h
+++ b/include/uapi/rdma/mlx4-abi.h
@@ -77,8 +77,8 @@ struct mlx4_ib_alloc_pd_resp {
 };
 
 struct mlx4_ib_create_cq {
-	__u64	buf_addr;
-	__u64	db_addr;
+	__aligned_u64 buf_addr;
+	__aligned_u64 db_addr;
 };
 
 struct mlx4_ib_create_cq_resp {
@@ -87,12 +87,12 @@ struct mlx4_ib_create_cq_resp {
 };
 
 struct mlx4_ib_resize_cq {
-	__u64	buf_addr;
+	__aligned_u64 buf_addr;
 };
 
 struct mlx4_ib_create_srq {
-	__u64	buf_addr;
-	__u64	db_addr;
+	__aligned_u64 buf_addr;
+	__aligned_u64 db_addr;
 };
 
 struct mlx4_ib_create_srq_resp {
@@ -101,7 +101,7 @@ struct mlx4_ib_create_srq_resp {
 };
 
 struct mlx4_ib_create_qp_rss {
-	__u64   rx_hash_fields_mask; /* Use  enum mlx4_ib_rx_hash_fields */
+	__aligned_u64 rx_hash_fields_mask; /* Use  enum mlx4_ib_rx_hash_fields */
 	__u8    rx_hash_function; /* Use enum mlx4_ib_rx_hash_function_flags */
 	__u8    reserved[7];
 	__u8    rx_hash_key[40];
@@ -110,8 +110,8 @@ struct mlx4_ib_create_qp_rss {
 };
 
 struct mlx4_ib_create_qp {
-	__u64	buf_addr;
-	__u64	db_addr;
+	__aligned_u64 buf_addr;
+	__aligned_u64 db_addr;
 	__u8	log_sq_bb_count;
 	__u8	log_sq_stride;
 	__u8	sq_no_prefetch;
@@ -120,8 +120,8 @@ struct mlx4_ib_create_qp {
 };
 
 struct mlx4_ib_create_wq {
-	__u64	buf_addr;
-	__u64	db_addr;
+	__aligned_u64 buf_addr;
+	__aligned_u64 db_addr;
 	__u8	log_range_size;
 	__u8	reserved[3];
 	__u32   comp_mask;
@@ -161,7 +161,7 @@ enum mlx4_ib_rx_hash_fields {
 };
 
 struct mlx4_ib_rss_caps {
-	__u64 rx_hash_fields_mask; /* enum mlx4_ib_rx_hash_fields */
+	__aligned_u64 rx_hash_fields_mask; /* enum mlx4_ib_rx_hash_fields */
 	__u8 rx_hash_function; /* enum mlx4_ib_rx_hash_function_flags */
 	__u8 reserved[7];
 };
@@ -181,8 +181,9 @@ struct mlx4_ib_tso_caps {
 struct mlx4_uverbs_ex_query_device_resp {
 	__u32			comp_mask;
 	__u32			response_length;
-	__u64			hca_cor