Merge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus

Pull MIPS updates from Ralf Baechle:
 "This is the main pull request for 4.3 for MIPS.  Here's the summary:

  Three fixes that didn't make 4.2-stable:

   - a -Os build might compile the kernel using the MIPS16 instruction
     set but the R2 optimized inline functions in <uapi/asm/swab.h> are
     implemented using 32-bit wide instructions which is invalid.

   - a build error in pgtable-bits.h for a particular kernel
     configuration.

   - accessing registers of the CM GCR might have been compiled to use
     64 bit accesses but these registers are onl 32 bit wide.

  And also a few new bits:

   - move the ATH79 GPIO driver to drivers/gpio

   - the definition of IRQCHIP_DECLARE has moved to linux/irqchip.h,
     change ATH79 accordingly.

   - fix definition of pgprot_writecombine

   - add an implementation of dma_map_ops.mmap

   - fix alignment of quiet build output for vmlinuz link

   - BCM47xx: Use kmemdup rather than duplicating its implementation

   - Netlogic: Fix 0x0x prefixes of constants.

   - merge Bjorn Helgaas' series to remove most of the weak keywords
     from function declarations.

   - CP0 and CP1 registers are best considered treated as unsigned
     values to avoid large values from becoming negative values.

   - improve support for the MIPS GIC timer.

   - enable common clock framework for Malta and SEAD3.

   - a number of improvments and fixes to dump_tlb().

   - document the MIPS TLB dump functionality in Magic SysRq.

   - Cavium Octeon CN68XX improvments.

   - NetLogic improvments.

   - irq: Use access helper irq_data_get_affinity_mask.

   - handle MSA unaligned accesses.

   - a number of R6-related math-emu fixes.

   - support for I6400.

   - improvments to MSA support.

   - add uprobes support.

   - move from deprecated __initcall to arch_initcall.

   - remove finish_arch_switch().

   - IRQ cleanups by Thomas Gleixner.

   - migrate to new 'set-state' interface.

   - random small cleanups"

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus: (148 commits)
  MIPS: UAPI: Fix unrecognized opcode WSBH/DSBH/DSHD when using MIPS16.
  MIPS: Fix alignment of quiet build output for vmlinuz link
  MIPS: math-emu: Remove unused handle_dsemul function declaration
  MIPS: math-emu: Add support for the MIPS R6 MAX{, A} FPU instruction
  MIPS: math-emu: Add support for the MIPS R6 MIN{, A} FPU instruction
  MIPS: math-emu: Add support for the MIPS R6 CLASS FPU instruction
  MIPS: math-emu: Add support for the MIPS R6 RINT FPU instruction
  MIPS: math-emu: Add support for the MIPS R6 MSUBF FPU instruction
  MIPS: math-emu: Add support for the MIPS R6 MADDF FPU instruction
  MIPS: math-emu: Add support for the MIPS R6 SELNEZ FPU instruction
  MIPS: math-emu: Add support for the MIPS R6 SELEQZ FPU instruction
  MIPS: math-emu: Add support for the CMP.condn.fmt R6 instruction
  MIPS: inst.h: Add new MIPS R6 FPU opcodes
  MIPS: Octeon: Fix management port MII address on Kontron S1901
  MIPS: BCM47xx: Use kmemdup rather than duplicating its implementation
  STAGING: Octeon: Use common helpers for determining interface and port
  MIPS: Octeon: Support interfaces 4 and 5
  MIPS: Octeon: Set up 1:1 mapping between CN68XX PKO queues and ports
  MIPS: Octeon: Initialize CN68XX PKO
  STAGING: Octeon: Support CN68XX style WQE
  ...

4 files changed, 118 insertions, 63 deletions

diff --git a/drivers/staging/octeon/ethernet-rx.c b/drivers/staging/octeon/ethernet-rx.c
index 22853d33da05..d1a33a927f6d 100644
--- a/drivers/staging/octeon/ethernet-rx.c
+++ b/drivers/staging/octeon/ethernet-rx.c
@@ -70,7 +70,14 @@ static irqreturn_t cvm_oct_do_interrupt(int cpl, void *dev_id)
  */
 static inline int cvm_oct_check_rcv_error(cvmx_wqe_t *work)
 {
-	if ((work->word2.snoip.err_code == 10) && (work->len <= 64)) {
+	int port;
+
+	if (octeon_has_feature(OCTEON_FEATURE_PKND))
+		port = work->word0.pip.cn68xx.pknd;
+	else
+		port = work->word1.cn38xx.ipprt;
+
+	if ((work->word2.snoip.err_code == 10) && (work->word1.len <= 64)) {
 		/*
 		 * Ignore length errors on min size packets. Some
 		 * equipment incorrectly pads packets to 64+4FCS
@@ -87,8 +94,8 @@ static inline int cvm_oct_check_rcv_error(cvmx_wqe_t *work)
 		 * packet to determine if we can remove a non spec
 		 * preamble and generate a correct packet.
 		 */
-		int interface = cvmx_helper_get_interface_num(work->ipprt);
-		int index = cvmx_helper_get_interface_index_num(work->ipprt);
+		int interface = cvmx_helper_get_interface_num(port);
+		int index = cvmx_helper_get_interface_index_num(port);
 		union cvmx_gmxx_rxx_frm_ctl gmxx_rxx_frm_ctl;
 
 		gmxx_rxx_frm_ctl.u64 =
@@ -99,7 +106,7 @@ static inline int cvm_oct_check_rcv_error(cvmx_wqe_t *work)
 			    cvmx_phys_to_ptr(work->packet_ptr.s.addr);
 			int i = 0;
 
-			while (i < work->len - 1) {
+			while (i < work->word1.len - 1) {
 				if (*ptr != 0x55)
 					break;
 				ptr++;
@@ -109,18 +116,18 @@ static inline int cvm_oct_check_rcv_error(cvmx_wqe_t *work)
 			if (*ptr == 0xd5) {
 				/*
 				  printk_ratelimited("Port %d received 0xd5 preamble\n",
-					  work->ipprt);
+					  port);
 				 */
 				work->packet_ptr.s.addr += i + 1;
-				work->len -= i + 5;
+				work->word1.len -= i + 5;
 			} else if ((*ptr & 0xf) == 0xd) {
 				/*
 				  printk_ratelimited("Port %d received 0x?d preamble\n",
-					  work->ipprt);
+					  port);
 				 */
 				work->packet_ptr.s.addr += i;
-				work->len -= i + 4;
-				for (i = 0; i < work->len; i++) {
+				work->word1.len -= i + 4;
+				for (i = 0; i < work->word1.len; i++) {
 					*ptr =
 					    ((*ptr & 0xf0) >> 4) |
 					    ((*(ptr + 1) & 0xf) << 4);
@@ -128,7 +135,7 @@ static inline int cvm_oct_check_rcv_error(cvmx_wqe_t *work)
 				}
 			} else {
 				printk_ratelimited("Port %d unknown preamble, packet dropped\n",
-						   work->ipprt);
+						   port);
 				/*
 				   cvmx_helper_dump_packet(work);
 				 */
@@ -138,7 +145,7 @@ static inline int cvm_oct_check_rcv_error(cvmx_wqe_t *work)
 		}
 	} else {
 		printk_ratelimited("Port %d receive error code %d, packet dropped\n",
-				   work->ipprt, work->word2.snoip.err_code);
+				   port, work->word2.snoip.err_code);
 		cvm_oct_free_work(work);
 		return 1;
 	}
@@ -172,9 +179,16 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 	}
 
 	/* Only allow work for our group (and preserve priorities) */
-	old_group_mask = cvmx_read_csr(CVMX_POW_PP_GRP_MSKX(coreid));
-	cvmx_write_csr(CVMX_POW_PP_GRP_MSKX(coreid),
-		       (old_group_mask & ~0xFFFFull) | 1 << pow_receive_group);
+	if (OCTEON_IS_MODEL(OCTEON_CN68XX)) {
+		old_group_mask = cvmx_read_csr(CVMX_SSO_PPX_GRP_MSK(coreid));
+		cvmx_write_csr(CVMX_SSO_PPX_GRP_MSK(coreid),
+				1ull << pow_receive_group);
+		cvmx_read_csr(CVMX_SSO_PPX_GRP_MSK(coreid)); /* Flush */
+	} else {
+		old_group_mask = cvmx_read_csr(CVMX_POW_PP_GRP_MSKX(coreid));
+		cvmx_write_csr(CVMX_POW_PP_GRP_MSKX(coreid),
+			(old_group_mask & ~0xFFFFull) | 1 << pow_receive_group);
+	}
 
 	if (USE_ASYNC_IOBDMA) {
 		cvmx_pow_work_request_async(CVMX_SCR_SCRATCH, CVMX_POW_NO_WAIT);
@@ -186,6 +200,7 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 		struct sk_buff **pskb = NULL;
 		int skb_in_hw;
 		cvmx_wqe_t *work;
+		int port;
 
 		if (USE_ASYNC_IOBDMA && did_work_request)
 			work = cvmx_pow_work_response_async(CVMX_SCR_SCRATCH);
@@ -195,12 +210,19 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 		prefetch(work);
 		did_work_request = 0;
 		if (work == NULL) {
-			union cvmx_pow_wq_int wq_int;
+			if (OCTEON_IS_MODEL(OCTEON_CN68XX)) {
+				cvmx_write_csr(CVMX_SSO_WQ_IQ_DIS,
+					       1ull << pow_receive_group);
+				cvmx_write_csr(CVMX_SSO_WQ_INT,
+					       1ull << pow_receive_group);
+			} else {
+				union cvmx_pow_wq_int wq_int;
 
-			wq_int.u64 = 0;
-			wq_int.s.iq_dis = 1 << pow_receive_group;
-			wq_int.s.wq_int = 1 << pow_receive_group;
-			cvmx_write_csr(CVMX_POW_WQ_INT, wq_int.u64);
+				wq_int.u64 = 0;
+				wq_int.s.iq_dis = 1 << pow_receive_group;
+				wq_int.s.wq_int = 1 << pow_receive_group;
+				cvmx_write_csr(CVMX_POW_WQ_INT, wq_int.u64);
+			}
 			break;
 		}
 		pskb = (struct sk_buff **)(cvm_oct_get_buffer_ptr(work->packet_ptr) -
@@ -220,7 +242,13 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 			prefetch(&skb->head);
 			prefetch(&skb->len);
 		}
-		prefetch(cvm_oct_device[work->ipprt]);
+
+		if (octeon_has_feature(OCTEON_FEATURE_PKND))
+			port = work->word0.pip.cn68xx.pknd;
+		else
+			port = work->word1.cn38xx.ipprt;
+
+		prefetch(cvm_oct_device[port]);
 
 		/* Immediately throw away all packets with receive errors */
 		if (unlikely(work->word2.snoip.rcv_error)) {
@@ -237,7 +265,7 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 			skb->data = skb->head + work->packet_ptr.s.addr -
 				cvmx_ptr_to_phys(skb->head);
 			prefetch(skb->data);
-			skb->len = work->len;
+			skb->len = work->word1.len;
 			skb_set_tail_pointer(skb, skb->len);
 			packet_not_copied = 1;
 		} else {
@@ -245,7 +273,7 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 			 * We have to copy the packet. First allocate
 			 * an skbuff for it.
 			 */
-			skb = dev_alloc_skb(work->len);
+			skb = dev_alloc_skb(work->word1.len);
 			if (!skb) {
 				cvm_oct_free_work(work);
 				continue;
@@ -268,13 +296,14 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 					else
 						ptr += 6;
 				}
-				memcpy(skb_put(skb, work->len), ptr, work->len);
+				memcpy(skb_put(skb, work->word1.len), ptr,
+				       work->word1.len);
 				/* No packet buffers to free */
 			} else {
 				int segments = work->word2.s.bufs;
 				union cvmx_buf_ptr segment_ptr =
 				    work->packet_ptr;
-				int len = work->len;
+				int len = work->word1.len;
 
 				while (segments--) {
 					union cvmx_buf_ptr next_ptr =
@@ -310,10 +339,9 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 			}
 			packet_not_copied = 0;
 		}
-
-		if (likely((work->ipprt < TOTAL_NUMBER_OF_PORTS) &&
-			   cvm_oct_device[work->ipprt])) {
-			struct net_device *dev = cvm_oct_device[work->ipprt];
+		if (likely((port < TOTAL_NUMBER_OF_PORTS) &&
+			   cvm_oct_device[port])) {
+			struct net_device *dev = cvm_oct_device[port];
 			struct octeon_ethernet *priv = netdev_priv(dev);
 
 			/*
@@ -333,7 +361,7 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 					skb->ip_summed = CHECKSUM_UNNECESSARY;
 
 				/* Increment RX stats for virtual ports */
-				if (work->ipprt >= CVMX_PIP_NUM_INPUT_PORTS) {
+				if (port >= CVMX_PIP_NUM_INPUT_PORTS) {
 #ifdef CONFIG_64BIT
 					atomic64_add(1,
 						     (atomic64_t *)&priv->stats.rx_packets);
@@ -368,7 +396,7 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 			 * doesn't exist.
 			 */
 			printk_ratelimited("Port %d not controlled by Linux, packet dropped\n",
-				   work->ipprt);
+				   port);
 			dev_kfree_skb_irq(skb);
 		}
 		/*
@@ -390,7 +418,13 @@ static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
 		}
 	}
 	/* Restore the original POW group mask */
-	cvmx_write_csr(CVMX_POW_PP_GRP_MSKX(coreid), old_group_mask);
+	if (OCTEON_IS_MODEL(OCTEON_CN68XX)) {
+		cvmx_write_csr(CVMX_SSO_PPX_GRP_MSK(coreid), old_group_mask);
+		cvmx_read_csr(CVMX_SSO_PPX_GRP_MSK(coreid)); /* Flush */
+	} else {
+		cvmx_write_csr(CVMX_POW_PP_GRP_MSKX(coreid), old_group_mask);
+	}
+
 	if (USE_ASYNC_IOBDMA) {
 		/* Restore the scratch area */
 		cvmx_scratch_write64(CVMX_SCR_SCRATCH, old_scratch);
@@ -422,8 +456,6 @@ void cvm_oct_rx_initialize(void)
 {
 	int i;
 	struct net_device *dev_for_napi = NULL;
-	union cvmx_pow_wq_int_thrx int_thr;
-	union cvmx_pow_wq_int_pc int_pc;
 
 	for (i = 0; i < TOTAL_NUMBER_OF_PORTS; i++) {
 		if (cvm_oct_device[i]) {
@@ -449,15 +481,34 @@ void cvm_oct_rx_initialize(void)
 
 	disable_irq_nosync(OCTEON_IRQ_WORKQ0 + pow_receive_group);
 
-	int_thr.u64 = 0;
-	int_thr.s.tc_en = 1;
-	int_thr.s.tc_thr = 1;
 	/* Enable POW interrupt when our port has at least one packet */
-	cvmx_write_csr(CVMX_POW_WQ_INT_THRX(pow_receive_group), int_thr.u64);
-
-	int_pc.u64 = 0;
-	int_pc.s.pc_thr = 5;
-	cvmx_write_csr(CVMX_POW_WQ_INT_PC, int_pc.u64);
+	if (OCTEON_IS_MODEL(OCTEON_CN68XX)) {
+		union cvmx_sso_wq_int_thrx int_thr;
+		union cvmx_pow_wq_int_pc int_pc;
+
+		int_thr.u64 = 0;
+		int_thr.s.tc_en = 1;
+		int_thr.s.tc_thr = 1;
+		cvmx_write_csr(CVMX_SSO_WQ_INT_THRX(pow_receive_group),
+			       int_thr.u64);
+
+		int_pc.u64 = 0;
+		int_pc.s.pc_thr = 5;
+		cvmx_write_csr(CVMX_SSO_WQ_INT_PC, int_pc.u64);
+	} else {
+		union cvmx_pow_wq_int_thrx int_thr;
+		union cvmx_pow_wq_int_pc int_pc;
+
+		int_thr.u64 = 0;
+		int_thr.s.tc_en = 1;
+		int_thr.s.tc_thr = 1;
+		cvmx_write_csr(CVMX_POW_WQ_INT_THRX(pow_receive_group),
+			       int_thr.u64);
+
+		int_pc.u64 = 0;
+		int_pc.s.pc_thr = 5;
+		cvmx_write_csr(CVMX_POW_WQ_INT_PC, int_pc.u64);
+	}
 
 	/* Schedule NAPI now. This will indirectly enable the interrupt. */
 	napi_schedule(&cvm_oct_napi);
diff --git a/drivers/staging/octeon/ethernet-tx.c b/drivers/staging/octeon/ethernet-tx.c
index e2df041ca82d..9e2116f4c915 100644
--- a/drivers/staging/octeon/ethernet-tx.c
+++ b/drivers/staging/octeon/ethernet-tx.c
@@ -591,13 +591,14 @@ int cvm_oct_xmit_pow(struct sk_buff *skb, struct net_device *dev)
 	 * Fill in some of the work queue fields. We may need to add
 	 * more if the software at the other end needs them.
 	 */
-	work->hw_chksum = skb->csum;
-	work->len = skb->len;
-	work->ipprt = priv->port;
-	work->qos = priv->port & 0x7;
-	work->grp = pow_send_group;
-	work->tag_type = CVMX_HELPER_INPUT_TAG_TYPE;
-	work->tag = pow_send_group;	/* FIXME */
+	if (!OCTEON_IS_MODEL(OCTEON_CN68XX))
+		work->word0.pip.cn38xx.hw_chksum = skb->csum;
+	work->word1.len = skb->len;
+	cvmx_wqe_set_port(work, priv->port);
+	cvmx_wqe_set_qos(work, priv->port & 0x7);
+	cvmx_wqe_set_grp(work, pow_send_group);
+	work->word1.tag_type = CVMX_HELPER_INPUT_TAG_TYPE;
+	work->word1.tag = pow_send_group;	/* FIXME */
 	/* Default to zero. Sets of zero later are commented out */
 	work->word2.u64 = 0;
 	work->word2.s.bufs = 1;
@@ -677,8 +678,8 @@ int cvm_oct_xmit_pow(struct sk_buff *skb, struct net_device *dev)
 	}
 
 	/* Submit the packet to the POW */
-	cvmx_pow_work_submit(work, work->tag, work->tag_type, work->qos,
-			     work->grp);
+	cvmx_pow_work_submit(work, work->word1.tag, work->word1.tag_type,
+			     cvmx_wqe_get_qos(work), cvmx_wqe_get_grp(work));
 	priv->stats.tx_packets++;
 	priv->stats.tx_bytes += skb->len;
 	dev_consume_skb_any(skb);
diff --git a/drivers/staging/octeon/ethernet-util.h b/drivers/staging/octeon/ethernet-util.h
index 1ba789a7741b..45f024bc5e33 100644
--- a/drivers/staging/octeon/ethernet-util.h
+++ b/drivers/staging/octeon/ethernet-util.h
@@ -8,6 +8,10 @@
  * published by the Free Software Foundation.
  */
 
+#include <asm/octeon/cvmx-pip.h>
+#include <asm/octeon/cvmx-helper.h>
+#include <asm/octeon/cvmx-helper-util.h>
+
 /**
  * cvm_oct_get_buffer_ptr - convert packet data address to pointer
  * @packet_ptr: Packet data hardware address
@@ -28,14 +32,12 @@ static inline void *cvm_oct_get_buffer_ptr(union cvmx_buf_ptr packet_ptr)
  */
 static inline int INTERFACE(int ipd_port)
 {
-	if (ipd_port < 32)	/* Interface 0 or 1 for RGMII,GMII,SPI, etc */
-		return ipd_port >> 4;
-	else if (ipd_port < 36)	/* Interface 2 for NPI */
-		return 2;
-	else if (ipd_port < 40)	/* Interface 3 for loopback */
-		return 3;
-	else if (ipd_port == 40)	/* Non existent interface for POW0 */
-		return 4;
+	int interface = cvmx_helper_get_interface_num(ipd_port);
+
+	if (interface >= 0)
+		return interface;
+	else if (ipd_port == CVMX_PIP_NUM_INPUT_PORTS)
+		return 10;
 	panic("Illegal ipd_port %d passed to INTERFACE\n", ipd_port);
 }
 
@@ -47,7 +49,5 @@ static inline int INTERFACE(int ipd_port)
  */
 static inline int INDEX(int ipd_port)
 {
-	if (ipd_port < 32)
-		return ipd_port & 15;
-	return ipd_port & 3;
+	return cvmx_helper_get_interface_index_num(ipd_port);
 }
diff --git a/drivers/staging/octeon/ethernet.c b/drivers/staging/octeon/ethernet.c
index 0718b35f7335..7274fda0b77b 100644
--- a/drivers/staging/octeon/ethernet.c
+++ b/drivers/staging/octeon/ethernet.c
@@ -152,7 +152,7 @@ static void cvm_oct_configure_common_hw(void)
 			     num_packet_buffers);
 	if (CVMX_FPA_OUTPUT_BUFFER_POOL != CVMX_FPA_PACKET_POOL)
 		cvm_oct_mem_fill_fpa(CVMX_FPA_OUTPUT_BUFFER_POOL,
-				     CVMX_FPA_OUTPUT_BUFFER_POOL_SIZE, 128);
+				     CVMX_FPA_OUTPUT_BUFFER_POOL_SIZE, 1024);
 
 #ifdef __LITTLE_ENDIAN
 	{
@@ -860,7 +860,10 @@ static int cvm_oct_remove(struct platform_device *pdev)
 	int port;
 
 	/* Disable POW interrupt */
-	cvmx_write_csr(CVMX_POW_WQ_INT_THRX(pow_receive_group), 0);
+	if (OCTEON_IS_MODEL(OCTEON_CN68XX))
+		cvmx_write_csr(CVMX_SSO_WQ_INT_THRX(pow_receive_group), 0);
+	else
+		cvmx_write_csr(CVMX_POW_WQ_INT_THRX(pow_receive_group), 0);
 
 	cvmx_ipd_disable();