net/octeontx/base: add octeontx IO operations
authorJerin Jacob <jerin.jacob@caviumnetworks.com>
Sun, 8 Oct 2017 12:44:06 +0000 (18:14 +0530)
committerFerruh Yigit <ferruh.yigit@intel.com>
Thu, 12 Oct 2017 00:36:57 +0000 (01:36 +0100)
Adding octeontx specific io operations. Added a stub for building
against non octeontx targets.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Signed-off-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
drivers/net/octeontx/base/octeontx_io.h [new file with mode: 0644]

diff --git a/drivers/net/octeontx/base/octeontx_io.h b/drivers/net/octeontx/base/octeontx_io.h
new file mode 100644 (file)
index 0000000..ec4ce1d
--- /dev/null
@@ -0,0 +1,156 @@
+/*
+ *   BSD LICENSE
+ *
+ *   Copyright (C) Cavium Inc. 2017. All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Cavium networks nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __OCTEONTX_IO_H__
+#define __OCTEONTX_IO_H__
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include <rte_io.h>
+
+/* In Cavium OcteonTX SoC, all accesses to the device registers are
+ * implicitly strongly ordered. So, The relaxed version of IO operation is
+ * safe to use with out any IO memory barriers.
+ */
+#define octeontx_read64 rte_read64_relaxed
+#define octeontx_write64 rte_write64_relaxed
+
+/* ARM64 specific functions */
+#if defined(RTE_ARCH_ARM64)
+#define octeontx_prefetch_store_keep(_ptr) ({\
+       asm volatile("prfm pstl1keep, %a0\n" : : "p" (_ptr)); })
+
+#define octeontx_load_pair(val0, val1, addr) ({                \
+                       asm volatile(                   \
+                       "ldp %x[x0], %x[x1], [%x[p1]]"  \
+                       :[x0]"=r"(val0), [x1]"=r"(val1) \
+                       :[p1]"r"(addr)                  \
+                       ); })
+
+#define octeontx_store_pair(val0, val1, addr) ({               \
+                       asm volatile(                   \
+                       "stp %x[x0], %x[x1], [%x[p1]]"  \
+                       ::[x0]"r"(val0), [x1]"r"(val1), [p1]"r"(addr) \
+                       ); })
+#else /* Un optimized functions for building on non arm64 arch */
+
+#define octeontx_prefetch_store_keep(_ptr) do {} while (0)
+
+#define octeontx_load_pair(val0, val1, addr)           \
+do {                                                   \
+       val0 = rte_read64(addr);                        \
+       val1 = rte_read64(((uint8_t *)addr) + 8);       \
+} while (0)
+
+#define octeontx_store_pair(val0, val1, addr)          \
+do {                                                   \
+       rte_write64(val0, addr);                        \
+       rte_write64(val1, (((uint8_t *)addr) + 8));     \
+} while (0)
+#endif
+
+#if defined(RTE_ARCH_ARM64)
+/**
+ * Perform an atomic fetch-and-add operation.
+ */
+static inline uint64_t
+octeontx_reg_ldadd_u64(void *addr, int64_t off)
+{
+       uint64_t old_val;
+
+       __asm__ volatile(
+               " .cpu          generic+lse\n"
+               " ldadd %1, %0, [%2]\n"
+               : "=r" (old_val) : "r" (off), "r" (addr) : "memory");
+
+       return old_val;
+}
+
+/**
+ * Perform a LMTST operation - an atomic write of up to 128 byte to
+ * an I/O block that supports this operation type.
+ *
+ * @param lmtline_va is the address where LMTLINE is mapped
+ * @param ioreg_va is the virtual address of the device register
+ * @param cmdbuf is the array of peripheral commands to execute
+ * @param cmdsize is the number of 64-bit words in 'cmdbuf'
+ *
+ * @return N/A
+ */
+static inline void
+octeontx_reg_lmtst(void *lmtline_va, void *ioreg_va, const uint64_t cmdbuf[],
+                  uint64_t cmdsize)
+{
+       uint64_t result;
+       uint64_t word_count;
+       uint64_t *lmtline = lmtline_va;
+
+       word_count = cmdsize;
+
+       do {
+               /* Copy commands to LMTLINE */
+               for (result = 0; result < word_count; result += 2) {
+                       lmtline[result + 0] = cmdbuf[result + 0];
+                       lmtline[result + 1] = cmdbuf[result + 1];
+               }
+
+               /* LDEOR initiates atomic transfer to I/O device */
+               __asm__ volatile(
+                       " .cpu          generic+lse\n"
+                       " ldeor xzr, %0, [%1]\n"
+                       : "=r" (result) : "r" (ioreg_va) : "memory");
+       } while (!result);
+}
+
+#else
+
+static inline uint64_t
+octeontx_reg_ldadd_u64(void *addr, int64_t off)
+{
+       RTE_SET_USED(addr);
+       RTE_SET_USED(off);
+       return 0;
+}
+
+static inline void
+octeontx_reg_lmtst(void *lmtline_va, void *ioreg_va, const uint64_t cmdbuf[],
+                  uint64_t cmdsize)
+{
+       RTE_SET_USED(lmtline_va);
+       RTE_SET_USED(ioreg_va);
+       RTE_SET_USED(cmdbuf);
+       RTE_SET_USED(cmdsize);
+}
+
+#endif
+#endif /* __OCTEONTX_IO_H__ */