Add support for SVE Direct WQE

Some Kunpeng SoCs do not support the DWQE through NEON
instructions. In this case, the IO path works normally,
but the performance will deteriorate.

For these SoCs that do not support NEON DWQE, they support
DWQE through SVE instructions. This patch supports SVE DWQE
to guarantee the performance of these SoCs. In addition, in
this scenario, DWQE only supports acceleration through SVE's
ldr and str instructions. Other load and store instructions
also cause performance degradation.

Signed-off-by: Juan Zhou <zhoujuan51@h-partners.com>
(cherry picked from commit 268e25f9374021fc4c0d6dabd62e0f360193081f)

0042-Update-kernel-headers.patch

@@ -0,0 +1,28 @@
+From 0c132b9216282269974bf5d21f877413cc222950 Mon Sep 17 00:00:00 2001
+From: Yixing Liu <liuyixing1@huawei.com>
+Date: Mon, 17 Apr 2023 09:48:09 +0800
+Subject: [PATCH 1/2] Update kernel headers
+
+    To commit ?? ("RDMA/hns: Add SVE DIRECT WQE flag to support libhns").
+
+Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
+Reviewed-by: Yangyang Li <liyangyang20@huawei.com>
+---
+ kernel-headers/rdma/hns-abi.h | 1 +
+ 1 file changed, 1 insertion(+)
+
+diff --git a/kernel-headers/rdma/hns-abi.h b/kernel-headers/rdma/hns-abi.h
+index bd19927..cab941f 100644
+--- a/kernel-headers/rdma/hns-abi.h
++++ b/kernel-headers/rdma/hns-abi.h
+@@ -92,6 +92,7 @@ enum hns_roce_qp_cap_flags {
+ 	HNS_ROCE_QP_CAP_RQ_RECORD_DB = 1 << 0,
+ 	HNS_ROCE_QP_CAP_SQ_RECORD_DB = 1 << 1,
+ 	HNS_ROCE_QP_CAP_OWNER_DB = 1 << 2,
++	HNS_ROCE_QP_CAP_SVE_DIRECT_WQE = 1 << 3,
+ 	HNS_ROCE_QP_CAP_DYNAMIC_CTX_ATTACH = 1 << 4,
+ 	HNS_ROCE_QP_CAP_DIRECT_WQE = 1 << 5,
+ 	HNS_ROCE_QP_CAP_DYNAMIC_CTX_DETACH = 1 << 6,
+-- 
+2.25.1
+

0043-libhns-Add-support-for-SVE-Direct-WQE.patch

@@ -0,0 +1,125 @@
+From 6f08530cae5de66fabfae4cb29729a18b0e86365 Mon Sep 17 00:00:00 2001
+From: Yixing Liu <liuyixing1@huawei.com>
+Date: Mon, 17 Apr 2023 09:48:10 +0800
+Subject: [PATCH 2/2] libhns: Add support for SVE Direct WQE
+
+driver inclusion
+category: bugfix
+bugzilla: https://gitee.com/src-openeuler/rdma-core/issues/I6VLLM
+
+---------------------------------------------------------------
+
+Some Kunpeng SoCs do not support the DWQE through NEON
+instructions. In this case, the IO path works normally,
+but the performance will deteriorate.
+
+For these SoCs that do not support NEON DWQE, they support
+DWQE through SVE instructions. This patch supports SVE DWQE
+to guarantee the performance of these SoCs. In addition, in
+this scenario, DWQE only supports acceleration through SVE's
+ldr and str instructions. Other load and store instructions
+also cause performance degradation.
+
+Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
+Reviewed-by: Yangyang Li <liyangyang20@huawei.com>
+---
+ CMakeLists.txt                   |  1 +
+ buildlib/RDMA_EnableCStd.cmake   | 17 +++++++++++++++++
+ providers/hns/CMakeLists.txt     |  5 +++++
+ providers/hns/hns_roce_u_hw_v2.c | 21 ++++++++++++++++++++-
+ 4 files changed, 43 insertions(+), 1 deletion(-)
+
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index 787c8be..bc4437b 100644
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -399,6 +399,7 @@ if (NOT HAVE_SPARSE)
+ endif()
+ 
+ RDMA_Check_SSE(HAVE_TARGET_SSE)
++RDMA_Check_SVE(HAVE_TARGET_SVE)
+ 
+ # Enable development support features
+ # Prune unneeded shared libraries during linking
+diff --git a/buildlib/RDMA_EnableCStd.cmake b/buildlib/RDMA_EnableCStd.cmake
+index 3c42824..2b56f42 100644
+--- a/buildlib/RDMA_EnableCStd.cmake
++++ b/buildlib/RDMA_EnableCStd.cmake
+@@ -127,3 +127,20 @@ int main(int argc, char *argv[])
+   endif()
+   set(${TO_VAR} "${HAVE_TARGET_SSE}" PARENT_SCOPE)
+ endFunction()
++
++function(RDMA_Check_SVE TO_VAR)
++  set(SVE_CHECK_PROGRAM "
++int main(int argc, char *argv[])
++{
++	return 0;
++}
++")
++
++  RDMA_Check_C_Compiles(HAVE_TARGET_SVE "${SVE_CHECK_PROGRAM}" "-march=armv8.2-a+sve")
++  if(NOT HAVE_TARGET_SVE)
++    message("SVE is not supported")
++  else()
++    set(SVE_FLAGS "-march=armv8.2-a+sve" PARENT_SCOPE)
++  endif()
++  set(${TO_VAR} "${HAVE_TARGET_SVE}" PARENT_SCOPE)
++endFunction()
+\ No newline at end of file
+diff --git a/providers/hns/CMakeLists.txt b/providers/hns/CMakeLists.txt
+index 160e1ff..ef031a8 100644
+--- a/providers/hns/CMakeLists.txt
++++ b/providers/hns/CMakeLists.txt
+@@ -11,4 +11,9 @@ publish_headers(infiniband
+ 	hnsdv.h
+ )
+ 
++if (HAVE_TARGET_SVE)
++  add_definitions("-DHNS_SVE")
++  set_source_files_properties(hns_roce_u_hw_v2.c PROPERTIES COMPILE_FLAGS "${SVE_FLAGS}")
++endif()
++
+ rdma_pkg_config("hns" "libibverbs" "${CMAKE_THREAD_LIBS_INIT}")
+diff --git a/providers/hns/hns_roce_u_hw_v2.c b/providers/hns/hns_roce_u_hw_v2.c
+index d0067d3..a49b50d 100644
+--- a/providers/hns/hns_roce_u_hw_v2.c
++++ b/providers/hns/hns_roce_u_hw_v2.c
+@@ -321,6 +321,22 @@ static void hns_roce_write512(uint64_t *dest, uint64_t *val)
+ 	mmio_memcpy_x64(dest, val, sizeof(struct hns_roce_rc_sq_wqe));
+ }
+ 
++#if defined(HNS_SVE)
++static void hns_roce_sve_write512(uint64_t *dest, uint64_t *val)
++{
++	asm volatile(
++		"ldr z0, [%0]\n"
++		"str z0, [%1]\n"
++		::"r" (val), "r"(dest):"cc", "memory"
++	);
++}
++#else
++static void hns_roce_sve_write512(uint64_t *dest, uint64_t *val)
++{
++	return;
++}
++#endif
++
+ static void hns_roce_write_dwqe(struct hns_roce_qp *qp, void *wqe)
+ {
+ 	struct hns_roce_rc_sq_wqe *rc_sq_wqe = wqe;
+@@ -337,7 +353,10 @@ static void hns_roce_write_dwqe(struct hns_roce_qp *qp, void *wqe)
+ 	hr_reg_write(rc_sq_wqe, RCWQE_DB_SL_H, qp->sl >> HNS_ROCE_SL_SHIFT);
+ 	hr_reg_write(rc_sq_wqe, RCWQE_WQE_IDX, qp->sq.head);
+ 
+-	hns_roce_write512(qp->sq.db_reg, wqe);
++	if (qp->flags & HNS_ROCE_QP_CAP_SVE_DIRECT_WQE)
++		hns_roce_sve_write512(qp->sq.db_reg, wqe);
++	else
++		hns_roce_write512(qp->sq.db_reg, wqe);
+ }
+ 
+ static void update_cq_db(struct hns_roce_context *ctx, struct hns_roce_cq *cq)
+-- 
+2.25.1
+

rdma-core.spec

@@ -1,6 +1,6 @@
 Name:           rdma-core
 Version:        41.0
-Release:        9
+Release:        10
 Summary:        RDMA core userspace libraries and daemons
 License:        GPLv2 or BSD
 Url:            https://github.com/linux-rdma/rdma-core
@@ -47,6 +47,8 @@ Patch37: 0038-Update-kernel-headers.patch
 Patch38: 0039-libhns-Add-reset-stop-flow-mechanism.patch
 Patch39: 0040-Update-kernel-headers.patch
 Patch40: 0041-libhns-Support-congestion-control-algorithm-configur.patch
+Patch41: 0042-Update-kernel-headers.patch
+Patch42: 0043-libhns-Add-support-for-SVE-Direct-WQE.patch
 
 BuildRequires:  binutils cmake >= 2.8.11 gcc libudev-devel pkgconfig pkgconfig(libnl-3.0)
 BuildRequires:  pkgconfig(libnl-route-3.0) valgrind-devel systemd systemd-devel
@@ -294,6 +296,12 @@ fi
 %{_mandir}/*
 
 %changelog
+* Mon Apr 17 2023 Juan Zhou <zhoujuan51@h-partners.com> - 41.0-10
+- Type: bugfix
+- ID: NA
+- SUG: NA
+- DESC: Add support for SVE Direct WQE for libhns
+
 * Thu Apr 13 2023 Juan Zhou <zhoujuan51@h-partners.com> - 41.0-9
 - Type: requirement
 - ID: NA