Module Name:    src
Committed By:   cliff
Date:           Sun Mar 21 20:41:43 UTC 2010

Added Files:
        src/sys/arch/mips/rmi [matt-nb5-mips64]: rmixl_fmn.c rmixl_fmnvar.h

Log Message:
add device driver for RMI Fast Message Network


To generate a diff of this commit:
cvs rdiff -u -r0 -r1.1.2.1 src/sys/arch/mips/rmi/rmixl_fmn.c \
    src/sys/arch/mips/rmi/rmixl_fmnvar.h

Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.

Added files:

Index: src/sys/arch/mips/rmi/rmixl_fmn.c
diff -u /dev/null src/sys/arch/mips/rmi/rmixl_fmn.c:1.1.2.1
--- /dev/null	Sun Mar 21 20:41:43 2010
+++ src/sys/arch/mips/rmi/rmixl_fmn.c	Sun Mar 21 20:41:43 2010
@@ -0,0 +1,1048 @@
+/*	$NetBSD: rmixl_fmn.c,v 1.1.2.1 2010/03/21 20:41:43 cliff Exp $	*/
+
+#include "opt_ddb.h"
+
+#include <sys/cdefs.h>
+#include <sys/param.h>
+#include <sys/cpu.h>
+#include <mips/cpuregs.h>
+#include <mips/rmi/rmixlreg.h>
+#include <mips/rmi/rmixlvar.h>
+#include <mips/rmi/rmixl_intr.h>
+#include <mips/rmi/rmixl_fmnvar.h>
+
+#ifdef FMN_DEBUG
+# define DPRINTF(x)	do { printf x ; } while(0)
+#else
+# define DPRINTF(x)
+#endif
+
+#ifdef DIAGNOSTIC
+# define DIAG_PRF(x)	do { printf x ; } while(0)
+#else
+# define DIAG_PRF(x)
+#endif
+
+
+
+/*
+ * index CPU-dependent table by (global) bucket ID to obtain logical Station ID
+ * see Table 12.1 in the XLS PRM
+ */
+/* use this table for XLS6xx, XLS4xx */
+static const int station_xls_4xx[] = {
+	[0 ... 7]	= RMIXL_FMN_STID_CORE0,
+	[8 ... 15]	= RMIXL_FMN_STID_CORE1,
+	[16 ... 23]	= RMIXL_FMN_STID_CORE2,
+	[24 ... 31]	= RMIXL_FMN_STID_CORE3,
+	[32 ... 63]	= RMIXL_FMN_STID_RESERVED,
+	[64 ... 71]	= RMIXL_FMN_STID_PCIE,
+	[72 ... 79]	= RMIXL_FMN_STID_RESERVED,
+	[80 ... 87]	= RMIXL_FMN_STID_GMAC_Q1,
+	[88 ... 95]	= RMIXL_FMN_STID_RESERVED,
+	[96 ... 103]	= RMIXL_FMN_STID_GMAC_Q0,
+	[104 ... 107]	= RMIXL_FMN_STID_DMA,
+	[108 ... 109]	= RMIXL_FMN_STID_CDE,
+	[110 ... 119]	= RMIXL_FMN_STID_RESERVED,
+	[120 ... 127]	= RMIXL_FMN_STID_SAE,
+};
+
+/* use this table for XLS408Lite, XLS404Lite */
+static const int station_xls_4xx_lite[] = {
+	[0 ... 7]	= RMIXL_FMN_STID_CORE0,
+	[8 ... 15]	= RMIXL_FMN_STID_CORE1,
+	[16 ... 23]	= RMIXL_FMN_STID_CORE2,
+	[24 ... 31]	= RMIXL_FMN_STID_CORE3,
+	[32 ... 79]	= RMIXL_FMN_STID_RESERVED,
+	[80 ... 87]	= RMIXL_FMN_STID_GMAC_Q1,
+	[88 ... 95]	= RMIXL_FMN_STID_RESERVED,
+	[96 ... 103]	= RMIXL_FMN_STID_GMAC_Q0,
+	[104 ... 107]	= RMIXL_FMN_STID_DMA,
+	[108 ... 109]	= RMIXL_FMN_STID_CDE,
+	[110 ... 115]	= RMIXL_FMN_STID_RESERVED,
+	[116 ... 119]	= RMIXL_FMN_STID_PCIE,
+	[120 ... 127]	= RMIXL_FMN_STID_SAE,
+};
+
+/* use this table for XLS2xx */
+static const int station_xls_2xx[] = {
+	[0 ... 7]	= RMIXL_FMN_STID_CORE0,
+	[8 ... 15]	= RMIXL_FMN_STID_CORE1,
+	[16 ... 23]	= RMIXL_FMN_STID_CORE2,
+	[24 ... 31]	= RMIXL_FMN_STID_CORE3,
+	[32 ... 63]	= RMIXL_FMN_STID_RESERVED,
+	[64 ... 71]	= RMIXL_FMN_STID_PCIE,
+	[72 ... 95]	= RMIXL_FMN_STID_RESERVED,
+	[96 ... 103]	= RMIXL_FMN_STID_GMAC_Q0,
+	[104 ... 107]	= RMIXL_FMN_STID_DMA,
+	[108 ... 119]	= RMIXL_FMN_STID_RESERVED,
+	[120 ... 127]	= RMIXL_FMN_STID_SAE,
+};
+
+/* use this table for XLS1xx */
+static const int station_xls_1xx[] = {
+	[0 ... 7]	= RMIXL_FMN_STID_CORE0,
+	[8 ... 15]	= RMIXL_FMN_STID_CORE1,
+	[16 ... 23]	= RMIXL_FMN_STID_CORE2,
+	[24 ... 31]	= RMIXL_FMN_STID_CORE3,
+	[32 ... 63]	= RMIXL_FMN_STID_RESERVED,
+	[64 ... 71]	= RMIXL_FMN_STID_PCIE,
+	[72 ... 95]	= RMIXL_FMN_STID_RESERVED,
+	[96 ... 101]	= RMIXL_FMN_STID_GMAC_Q0,
+	[102 ... 103]	= RMIXL_FMN_STID_RESERVED,
+	[104 ... 107]	= RMIXL_FMN_STID_DMA,
+	[108 ... 119]	= RMIXL_FMN_STID_RESERVED,
+	[120 ... 127]	= RMIXL_FMN_STID_SAE,
+};
+
+typedef struct fmn_station_info {
+	const char     *si_name;
+	const u_int	si_buckets_max;
+	const u_int	si_stid_first;
+	const u_int	si_stid_last;
+	const u_int	si_bucket_size_dflt;
+	const u_int	si_credits_min;
+	const u_int	si_regbase;
+} fmn_station_info_t;
+
+/* use this table for XLS6xx, XLS4xx */
+static const fmn_station_info_t station_info_xls_4xx[RMIXL_FMN_NSTID] = {
+#if 1
+	[RMIXL_FMN_STID_CORE0]   = { "core0",	8,   0,   7,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE1]   = { "core1",	8,   8,  15,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE2]   = { "core2",	8,  16,  23,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE3]   = { "core3",	8,  24,  31,  32, 4, 0 },
+#else
+	[RMIXL_FMN_STID_CORE0]   = { "core0",	8,   0,   7,  32, 8, 0 },
+	[RMIXL_FMN_STID_CORE1]   = { "core1",	8,   8,  15,  32, 8, 0 },
+	[RMIXL_FMN_STID_CORE2]   = { "core2",	8,  16,  23,  32, 8, 0 },
+	[RMIXL_FMN_STID_CORE3]   = { "core3",	8,  24,  31,  32, 8, 0 },
+#endif
+	[RMIXL_FMN_STID_GMAC_Q0] = { "gmac_q0",	3,  80,  87,  32, 0, RMIXL_IO_DEV_GMAC_0 },
+	[RMIXL_FMN_STID_GMAC_Q1] = { "gmac_q1",	3,  96, 103,  32, 0, RMIXL_IO_DEV_GMAC_4 },
+	[RMIXL_FMN_STID_DMA]     = { "dma",	4, 104, 107,  64, 0, RMIXL_IO_DEV_DMA },
+	[RMIXL_FMN_STID_CDE]     = { "cde",	4, 108, 109, 128, 0, RMIXL_IO_DEV_CDE },
+	[RMIXL_FMN_STID_PCIE]    = { "pcie",	8,  64,  71,  32, 0, RMIXL_IO_DEV_PCIE_BE },
+	[RMIXL_FMN_STID_SAE]     = { "sae",	2, 120, 121, 128, 0, RMIXL_IO_DEV_SAE },
+};
+
+/* use this table for XLS4xxLite */
+static const fmn_station_info_t station_info_xls_4xx_lite[RMIXL_FMN_NSTID] = {
+	[RMIXL_FMN_STID_CORE0]   = { "core0",	8,   0,   7,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE1]   = { "core1",	8,   8,  15,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE2]   = { "core2",	8,  16,  23,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE3]   = { "core3",	8,  24,  31,  32, 4, 0 },
+	[RMIXL_FMN_STID_GMAC_Q0] = { "gmac_q0",	3,  80,  87,  32, 0, RMIXL_IO_DEV_GMAC_0 },
+	[RMIXL_FMN_STID_GMAC_Q1] = { "gmac_q1",	3,  96, 103,  32, 0, RMIXL_IO_DEV_GMAC_4 },
+	[RMIXL_FMN_STID_DMA]     = { "dma",	4, 104, 107,  64, 0, RMIXL_IO_DEV_DMA },
+	[RMIXL_FMN_STID_CDE]     = { "cde",	4, 108, 109, 128, 0, RMIXL_IO_DEV_CDE },
+	[RMIXL_FMN_STID_PCIE]    = { "pcie",	4, 116, 119,  64, 0, RMIXL_IO_DEV_PCIE_BE },
+	[RMIXL_FMN_STID_SAE]     = { "sae",	2, 120, 121, 128, 0, RMIXL_IO_DEV_SAE },
+};
+
+/* use this table for XLS2xx */
+static const fmn_station_info_t station_info_xls_2xx[RMIXL_FMN_NSTID] = {
+	[RMIXL_FMN_STID_CORE0]   = { "core0",	8,   0,   7,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE1]   = { "core1",	8,   8,  15,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE2]   = { "core2",	8,  16,  23,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE3]   = { "core3",	8,  24,  31,  32, 4, 0 },
+	[RMIXL_FMN_STID_GMAC_Q0] = { "gmac_q0",	3,  96, 103,  32, 0, RMIXL_IO_DEV_GMAC_0 },
+	[RMIXL_FMN_STID_DMA]     = { "dma",	4, 104, 107,  64, 0, RMIXL_IO_DEV_DMA },
+	[RMIXL_FMN_STID_PCIE]    = { "pcie",	8,  64,  71,  32, 0, RMIXL_IO_DEV_PCIE_BE },
+	[RMIXL_FMN_STID_SAE]     = { "sae",	2, 120, 121, 128, 0, RMIXL_IO_DEV_SAE },
+};
+
+/* use this table for XLS1xx */
+static const fmn_station_info_t station_info_xls_1xx[RMIXL_FMN_NSTID] = {
+	[RMIXL_FMN_STID_CORE0]   = { "core0",	8,   0,   7,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE1]   = { "core1",	8,   8,  15,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE2]   = { "core2",	8,  16,  23,  32, 4, 0 },
+	[RMIXL_FMN_STID_CORE3]   = { "core3",	8,  24,  31,  32, 4, 0 },
+	[RMIXL_FMN_STID_GMAC_Q0] = { "gmac_q0",	3,  96, 101,  32, 0, RMIXL_IO_DEV_GMAC_0 },
+	[RMIXL_FMN_STID_DMA]     = { "dma",	4, 104, 107,  64, 0, RMIXL_IO_DEV_PCIE_BE },
+	[RMIXL_FMN_STID_PCIE]    = { "pcie",	4,  64,  67,  32, 0, RMIXL_IO_DEV_PCIE_BE },
+	[RMIXL_FMN_STID_SAE]     = { "sae",	2, 120, 121, 128, 0, RMIXL_IO_DEV_SAE },
+};
+
+
+typedef struct fmn_intrhand {
+	int		(*ih_func)(void *, rmixl_fmn_rxmsg_t *);
+	void		*ih_arg;
+	struct evcnt	ih_count;
+} fmn_intrhand_t;
+
+/*
+ * per-core FMN structure
+ */
+typedef struct fmn {
+	kmutex_t		       *fmn_lock; 
+	u_int				fmn_core;
+	u_int				fmn_thread;
+	const int		       *fmn_stidtab;
+	const fmn_station_info_t       *fmn_stinfo;
+	void			       *fmn_ih;
+	fmn_intrhand_t			fmn_intrhand[RMIXL_FMN_NSTID];
+} fmn_t;
+
+static fmn_t fmn[1 << 10];	/* index by cpuid) *//* XXX assumes 1 node */
+#define NFMN	(sizeof(fmn) / sizeof(fmn[0]))
+
+static fmn_t *
+fmn_lookup(cpuid_t cpuid)
+{
+	KASSERT(cpuid < (cpuid_t)NFMN);
+	return &fmn[cpuid];
+}
+
+static void	rmixl_fmn_config_noncore(fmn_t *);
+static void	rmixl_fmn_config_core(fmn_t *);
+#ifdef NOTYET
+static int	rmixl_fmn_intr_dispatch(void *);
+#endif	/* NOTYET */
+static int	rmixl_fmn_msg_recv_subr(u_int, rmixl_fmn_rxmsg_t *);
+
+#ifdef FMN_DEBUG
+void	rmixl_fmn_cp2_dump(void);
+void	rmixl_fmn_cc_dump(void);
+#endif
+
+/*
+ * macros used because mtc2, mfc2, dmtc2, dmfc2 instructions
+ * must use literal values for rd and sel operands
+ * so let the compiler sort it out
+ */
+
+/*
+ * write v to all 8 SELs for given RD
+ */
+#define FMN_CP2_4SEL_READ(rd, sel, vp)					\
+	do {								\
+		uint32_t *rp = vp;					\
+		RMIXL_MFC2(rd, sel,   rp[0]);				\
+		RMIXL_MFC2(rd, sel+1, rp[1]);				\
+		RMIXL_MFC2(rd, sel+2, rp[2]);				\
+		RMIXL_MFC2(rd, sel+3, rp[3]);				\
+	} while (0)
+
+/*
+ * write v to all 8 SELs for given RD
+ */
+#define FMN_CP2_4SEL_WRITE(rd, sel, v)					\
+	do {								\
+		RMIXL_MTC2(rd, sel,   v);				\
+		RMIXL_MTC2(rd, sel+1, v);				\
+		RMIXL_MTC2(rd, sel+2, v);				\
+		RMIXL_MTC2(rd, sel+3, v);				\
+	} while (0)
+
+#define FMN_CP2_8SEL_WRITE(rd, v)					\
+	do {								\
+		RMIXL_MTC2(rd, 0, v);					\
+		RMIXL_MTC2(rd, 1, v);					\
+		RMIXL_MTC2(rd, 2, v);					\
+		RMIXL_MTC2(rd, 3, v);					\
+		RMIXL_MTC2(rd, 4, v);					\
+		RMIXL_MTC2(rd, 5, v);					\
+		RMIXL_MTC2(rd, 6, v);					\
+		RMIXL_MTC2(rd, 7, v);					\
+	} while (0)
+
+
+#define FMN_CP2_SEL_CASE_READ(rd, sel, v)				\
+	case sel:							\
+		RMIXL_MFC2(rd, sel, v);					\
+		break
+#define FMN_CP2_SEL_CASE_WRITE(rd, sel, v)				\
+	case sel:							\
+		RMIXL_MTC2(rd, sel, v);					\
+		break
+/*
+ * read/write a single arbitrary sel for the given rd
+ */
+#define FMN_CP2_SEL_SWITCH_RW(rw, rd, sel, val)				\
+	do {								\
+		switch (sel) {						\
+			FMN_CP2_SEL_CASE_ ## rw(rd, 0, val);		\
+			FMN_CP2_SEL_CASE_ ## rw(rd, 1, val);		\
+			FMN_CP2_SEL_CASE_ ## rw(rd, 2, val);		\
+			FMN_CP2_SEL_CASE_ ## rw(rd, 3, val);		\
+			FMN_CP2_SEL_CASE_ ## rw(rd, 4, val);		\
+			FMN_CP2_SEL_CASE_ ## rw(rd, 5, val);		\
+			FMN_CP2_SEL_CASE_ ## rw(rd, 6, val);		\
+			FMN_CP2_SEL_CASE_ ## rw(rd, 7, val);		\
+			default:					\
+				panic("%s:%d: bad sel %d\n",		\
+					__func__, __LINE__, sel);	\
+		}							\
+	} while (0)
+
+#define FMN_CP2_RD_CASE_RW(rw, rd, sel, val)				\
+	case rd:							\
+		FMN_CP2_SEL_SWITCH_RW(rw, rd, sel, val);		\
+		break
+/*
+ * read/write a single arbitrary Credit Counter at (rd, sel)
+ * eg:
+ *	FMN_CP2_RD_SWITCH_RW(READ,  16, 2, val)
+ *	FMN_CP2_RD_SWITCH_RW(WRITE, 18, 0, val)
+ */
+#define FMN_CP2_RD_SWITCH_RW(rw, rd, sel, val)				\
+	do {								\
+		switch(rd) {						\
+			FMN_CP2_RD_CASE_RW(rw, 0,  sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 1,  sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 2,  sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 3,  sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 4,  sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 5,  sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 6,  sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 7,  sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 8,  sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 9,  sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 10, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 11, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 12, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 13, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 14, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 15, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 16, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 17, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 18, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 19, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 20, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 21, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 22, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 23, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 24, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 25, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 26, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 27, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 28, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 29, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 30, sel, val);		\
+			FMN_CP2_RD_CASE_RW(rw, 31, sel, val);		\
+			default:					\
+				panic("%s:%d: bad regno %d\n",		\
+					__func__, __LINE__, rd);	\
+		}							\
+	} while (0)
+
+
+void
+rmixl_fmn_init(void)
+{
+	int cpu;
+	fmn_t *fmnp;
+	static bool once=false;
+
+	KASSERT(cpu_number() == 0);
+	fmnp = fmn_lookup(cpu_number());
+
+	if (once == true)
+		panic("%s: call only once!", __func__);
+	once = true;
+
+	for (cpu=0; cpu < NFMN; cpu++) {
+		fmnp[cpu].fmn_core = RMIXL_CPU_CORE(cpu);
+		fmnp[cpu].fmn_thread = RMIXL_CPU_THREAD(cpu);
+	}
+
+	rmixl_fmn_init_core();		/* for initial boot cpu (#0) */
+	rmixl_fmn_config_noncore(fmnp);	/* boot cpu initializes noncore */
+}
+
+void
+rmixl_fmn_init_core(void)
+{
+	fmn_t *fmnp;
+	const fmn_station_info_t *info = NULL;
+	const int *tab = NULL;
+	kmutex_t *lk;
+
+	fmnp = fmn_lookup(cpu_number());
+	KASSERT(fmnp->fmn_core == RMIXL_CPU_CORE(cpu_number()));
+	KASSERT(fmnp->fmn_thread == RMIXL_CPU_THREAD(cpu_number()));
+
+	lk = mutex_obj_alloc(MUTEX_DEFAULT, RMIXL_FMN_INTR_IPL);
+	if (lk == NULL)
+		panic("%s: mutex_obj_alloc failed", __func__);
+	fmnp->fmn_lock = lk;
+
+	mutex_enter(fmnp->fmn_lock);
+
+	/*
+	 * link to TX station ID table for this cpu type
+	 */
+	switch (MIPS_PRID_IMPL(mips_options.mips_cpu_id)) {
+	case MIPS_XLS104:
+	case MIPS_XLS108:
+		tab = station_xls_1xx;
+		info = station_info_xls_1xx;
+		break;
+	case MIPS_XLS204:
+	case MIPS_XLS208:
+		tab = station_xls_2xx;
+		info = station_info_xls_2xx;
+		break;
+	case MIPS_XLS404:
+	case MIPS_XLS408:
+	case MIPS_XLS416:
+	case MIPS_XLS608:
+	case MIPS_XLS616:
+		tab = station_xls_4xx;
+		info = station_info_xls_4xx;
+		break;
+	case MIPS_XLS404LITE:
+	case MIPS_XLS408LITE:
+		tab = station_xls_4xx_lite;
+		info = station_info_xls_4xx_lite;
+		break;
+	default:
+		panic("%s: unknown PRID IMPL %#x\n", __func__,
+			MIPS_PRID_IMPL(mips_options.mips_cpu_id));
+	}
+
+	fmnp->fmn_stidtab = tab;
+	fmnp->fmn_stinfo = info;
+
+	/*
+	 * thread #0 for each core owns 'global' CP2 regs
+	 */
+	if (fmnp->fmn_thread == 0)
+		rmixl_fmn_config_core(fmnp);
+
+	mutex_exit(fmnp->fmn_lock);
+}
+
+/*
+ * rmixl_fmn_config_noncore
+ *
+ *	initialize bucket sizes and (minimum) credits for non-core stations to ZERO
+ *	configured through memory write operations instead of CP2
+ */
+static void
+rmixl_fmn_config_noncore(fmn_t *fmnp)
+{
+	for (u_int sid=0; sid < RMIXL_FMN_NSTID; sid++) {
+		u_int regoff = fmnp->fmn_stinfo[sid].si_regbase;
+		if (regoff != 0) {
+			u_int buckets_max = fmnp->fmn_stinfo[sid].si_buckets_max;
+			regoff += RMIXL_FMN_BS_FIRST;
+			for (u_int bucket=0; bucket < buckets_max; bucket++) {
+				RMIXL_IOREG_WRITE(regoff, 0);
+				regoff += sizeof(uint32_t);
+			}
+		}
+	}
+}
+
+/*
+ * rmixl_fmn_config_core
+ *
+ *	- assumes fmn_mutex is owned
+ *	- configure FMN 
+ *	- initialize bucket sizes and (minimum) credits for a core
+ */
+static void
+rmixl_fmn_config_core(fmn_t *fmnp)
+{
+	const fmn_station_info_t *info = fmnp->fmn_stinfo;
+	uint32_t sts1;
+	uint32_t cfg;
+	uint32_t cp0_status;
+
+	KASSERT(mutex_owned(fmnp->fmn_lock) != 0);
+	KASSERT(fmnp->fmn_thread == 0);
+	cp0_status = rmixl_cp2_enable();
+
+	/* check/clear any pre-existing status1 error(s) */
+	RMIXL_MFC2(RMIXL_COP_2_MSG_STS, 1, sts1);
+	if ((sts1 & RMIXL_MSG_STS1_ERRS) != 0)
+		RMIXL_MTC2(RMIXL_COP_2_MSG_STS, 1, sts1);
+
+	/* set up MsgConfig reg */
+	cfg = ((1 << RMIXL_MSG_CFG0_WMSHIFT)			/* watermark */
+	    |  (RMIXL_INTRVEC_FMN << RMIXL_MSG_CFG0_IV_SHIFT)	/* irq */
+	    |  (1 << RMIXL_MSG_CFG0_ITM_SHIFT)			/* thread mask */
+	    |   RMIXL_MSG_CFG0_WIE				/* watermark intr enb */
+	    |   RMIXL_MSG_CFG0_EIE);				/* rx not empty intr enb */
+	RMIXL_DMTC2(RMIXL_COP_2_MSG_CFG, 0, cfg);
+
+	/* disable trace mode, credit overrun intr, messaging errors intr */
+	RMIXL_DMTC2(RMIXL_COP_2_MSG_CFG, 0, 0);
+
+	/* XXX using 4 buckets per core */
+	KASSERT(4 <= info->si_buckets_max);
+
+	/*
+	 * initialize default sizes for core buckets
+	 * zero sizes for unused buckets
+	 */
+	KASSERT(info->si_buckets_max == 8);
+	uint32_t sz = info->si_bucket_size_dflt;
+	KASSERT((sz & ~RMIXL_MSG_BSZ_SIZE) == 0);
+	RMIXL_MTC2(RMIXL_COP_2_MSG_BSZ, 0, sz);
+	RMIXL_MTC2(RMIXL_COP_2_MSG_BSZ, 1, sz);
+	RMIXL_MTC2(RMIXL_COP_2_MSG_BSZ, 2, sz);
+	RMIXL_MTC2(RMIXL_COP_2_MSG_BSZ, 3, sz);
+	RMIXL_MTC2(RMIXL_COP_2_MSG_BSZ, 4, 0);
+	RMIXL_MTC2(RMIXL_COP_2_MSG_BSZ, 5, 0);
+	RMIXL_MTC2(RMIXL_COP_2_MSG_BSZ, 6, 0);
+	RMIXL_MTC2(RMIXL_COP_2_MSG_BSZ, 7, 0);
+
+	/*
+	 * configure minimum credits for each core, 4 buckets
+	 * zero all unused credit counters for this core
+	 */
+	uint32_t cr = info->si_credits_min;
+
+	FMN_CP2_4SEL_WRITE(RMIXL_COP_2_CREDITS,   0, cr);
+	FMN_CP2_4SEL_WRITE(RMIXL_COP_2_CREDITS,   4,  0);
+	FMN_CP2_4SEL_WRITE(RMIXL_COP_2_CREDITS+1, 0, cr);
+	FMN_CP2_4SEL_WRITE(RMIXL_COP_2_CREDITS+1, 4,  0);
+	FMN_CP2_4SEL_WRITE(RMIXL_COP_2_CREDITS+2, 0, cr);
+	FMN_CP2_4SEL_WRITE(RMIXL_COP_2_CREDITS+2, 4,  0);
+	FMN_CP2_4SEL_WRITE(RMIXL_COP_2_CREDITS+3, 0, cr);
+	FMN_CP2_4SEL_WRITE(RMIXL_COP_2_CREDITS+3, 4,  0);
+
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+4,     0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+5,     0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+6,     0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+7,     0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+8,     0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+9,     0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+10,    0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+11,    0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+12,    0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+13,    0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+14,    0);
+	FMN_CP2_8SEL_WRITE(RMIXL_COP_2_CREDITS+15,    0); 
+
+	RMIXL_MFC2(RMIXL_COP_2_MSG_STS, 1, sts1);
+	KASSERT((sts1 & RMIXL_MSG_STS1_ERRS) == 0);
+
+	rmixl_cp2_restore(cp0_status);
+}
+
+void
+rmixl_fmn_init_cpu_intr(void)
+{
+	fmn_t *fmnp;
+
+	fmnp = fmn_lookup(cpu_number());
+	mutex_enter(fmnp->fmn_lock);
+
+	for (int i=0; i < RMIXL_FMN_NSTID; i++)
+		evcnt_attach_dynamic(&fmnp->fmn_intrhand[i].ih_count,
+			EVCNT_TYPE_INTR, NULL, "rmixl_fmn", fmn->fmn_stinfo[i].si_name);
+
+#ifdef NOTYET
+	/*
+	 * establish dispatcher for FMN interrupt
+	 */
+	void *ih = rmixl_vec_establish(RMIXL_INTRVEC_FMN, -1, RMIXL_FMN_INTR_IPL,
+		rmixl_fmn_intr_dispatch, fmnp, "fmn");
+	if (ih == NULL)
+		panic("%s: rmixl_vec_establish failed", __func__);
+	fmnp->fmn_ih = ih;
+#endif
+
+	mutex_exit(fmnp->fmn_lock);
+}
+
+void *
+rmixl_fmn_intr_establish(int txstid, int (*func)(void *, rmixl_fmn_rxmsg_t *), void *arg)
+{
+	fmn_t *fmnp;
+	fmn_intrhand_t *ih;
+
+	fmnp = fmn_lookup(cpu_number());
+
+	mutex_enter(fmnp->fmn_lock);
+
+	ih = &fmnp->fmn_intrhand[txstid];
+
+	if (ih->ih_func != NULL) {
+#ifdef DEBUG
+		panic("%s: intrhand[%d] busy", __func__, txstid);
+#endif
+		ih = NULL;
+	}  else {
+		ih->ih_func = func;
+		ih->ih_arg = arg;
+	}
+
+	mutex_exit(fmnp->fmn_lock);
+
+	return ih;
+}
+
+void
+rmixl_fmn_intr_disestablish(void *cookie)
+{
+	fmn_t *fmnp;
+	fmn_intrhand_t *ih = cookie;
+
+	fmnp = fmn_lookup(cpu_number());
+	mutex_enter(fmnp->fmn_lock);
+
+	if (ih->ih_func != NULL) {
+		ih->ih_func = NULL;
+		ih->ih_arg = NULL;
+	}
+#ifdef DEBUG
+	else {
+		panic("%s: intrhand[%ld] not in use",
+			__func__, ih - &fmnp->fmn_intrhand[0]);
+	}
+#endif
+
+	mutex_exit(fmnp->fmn_lock);
+}
+
+void
+rmixl_fmn_intr_poll(u_int bucket, rmixl_fmn_rxmsg_t *rxmsg)
+{
+	uint32_t bit = 1 << bucket;
+	uint32_t cp0_status;
+
+	KASSERT(bucket < 8);
+
+	cp0_status = rmixl_cp2_enable();
+
+	for(;;) {
+		rmixl_fmn_msgwait(bit);
+		if (rmixl_fmn_msg_recv(bucket, rxmsg) == 0)
+			break;
+		DELAY(10);		/* XXX */
+	}
+
+	rmixl_cp2_restore(cp0_status);
+}
+
+#ifdef NOTYET
+static int
+rmixl_fmn_intr_dispatch(void *arg)
+{
+	fmn_t *fmnp = arg;
+	uint32_t msg_status;
+	uint32_t cp0_status;
+	uint32_t rfbne;
+	int txstid;
+	int rv = 0;
+
+	mutex_enter(fmnp->fmn_lock);
+	cp0_status = rmixl_cp2_enable();
+
+	RMIXL_MFC2(RMIXL_COP_2_MSG_STS, 0, msg_status);
+	rfbne = (~msg_status) >> RMIXL_MSG_STS0_RFBE_SHIFT;
+
+	if (rfbne != 0) {
+		DPRINTF(("%s: rfbne %#x\n", __func__, rfbne));
+		for (u_int bucket=0; bucket < 8; bucket++) {
+			rmixl_fmn_rxmsg_t rxmsg;
+			fmn_intrhand_t *ih;
+
+			if ((rfbne & (1 << bucket)) == 0)
+				continue;
+			if (rmixl_fmn_msg_recv_subr(bucket, &rxmsg) != 0)
+				continue;
+			rv = 1;
+			txstid = fmnp->fmn_stidtab[rxmsg.rxsid];
+			ih = &fmnp->fmn_intrhand[txstid];
+			if (ih->ih_func != NULL)
+				if ((ih->ih_func)(ih->ih_arg, &rxmsg) != 0)
+					ih->ih_count.ev_count++;
+		}
+
+	}
+	rmixl_cp2_restore(cp0_status);
+	mutex_exit(fmnp->fmn_lock);
+
+	return rv;
+}
+#endif	/* NOTYET */
+
+int
+rmixl_fmn_msg_send(u_int size, u_int code, u_int dest_id, rmixl_fmn_msg_t *msg)
+{
+	fmn_t *fmnp;
+        uint32_t cp0_status;
+        uint32_t msg_status;
+        uint32_t msg_status1;
+	uint32_t desc;
+	int rv = 0;
+    
+	KASSERT((size >= 1) && size <= 4);
+	KASSERT(code <= 0xff);
+	KASSERT(dest_id <= 0xff);
+
+	fmnp = fmn_lookup(cpu_number());
+	mutex_enter(fmnp->fmn_lock);
+        cp0_status = rmixl_cp2_enable();
+
+	switch(size) {
+	case 1:
+		RMIXL_DMTC2(RMIXL_COP_2_TXBUF, 0, msg->data[0]);
+		break;
+	case 2:
+		RMIXL_DMTC2(RMIXL_COP_2_TXBUF, 0, msg->data[0]);
+		RMIXL_DMTC2(RMIXL_COP_2_TXBUF, 1, msg->data[1]);
+		break;
+	case 3:
+		RMIXL_DMTC2(RMIXL_COP_2_TXBUF, 0, msg->data[0]);
+		RMIXL_DMTC2(RMIXL_COP_2_TXBUF, 1, msg->data[1]);
+		RMIXL_DMTC2(RMIXL_COP_2_TXBUF, 2, msg->data[2]);
+		break;
+	case 4:
+		RMIXL_DMTC2(RMIXL_COP_2_TXBUF, 0, msg->data[0]);
+		RMIXL_DMTC2(RMIXL_COP_2_TXBUF, 1, msg->data[1]);
+		RMIXL_DMTC2(RMIXL_COP_2_TXBUF, 2, msg->data[2]);
+		RMIXL_DMTC2(RMIXL_COP_2_TXBUF, 3, msg->data[3]);
+		break;
+	default:
+		DIAG_PRF(("%s: bad size %d", __func__, size));
+		rv = -1;
+		goto out;
+	}
+
+	for (int try=16; try--; ) {
+		RMIXL_MFC2(RMIXL_COP_2_MSG_STS, 0, msg_status);
+		if ((msg_status & (RMIXL_MSG_STS0_LPF|RMIXL_MSG_STS0_SPF|RMIXL_MSG_STS0_SMP)) == 0)
+			goto send;
+		DELAY(10);		/* XXX ??? */
+	}
+	DIAG_PRF(("%s: cpu=%ld, msg %p, dst_id=%d, sts=%#x: can't send\n",
+		__func__, cpu_number(), msg, dest_id, msg_status));
+	rv = -1;
+	goto out;
+ send:
+	desc = RMIXL_MSGSND_DESC(size, code, dest_id);
+	DPRINTF(("%s: cpu %ld, desc %#x\n", __func__, cpu_number(), desc));
+	for (int try=16; try--; ) {
+		rmixl_msgsnd(desc);
+		RMIXL_MFC2(RMIXL_COP_2_MSG_STS, 0, msg_status);
+		RMIXL_MFC2(RMIXL_COP_2_MSG_STS, 1, msg_status1);
+		if (((msg_status  & RMIXL_MSG_STS0_SCF) == 0)
+		&&  ((msg_status1 & RMIXL_MSG_STS1_ERRS) == 0))
+			goto out;
+#if 0
+#ifdef DEBUG
+		if ((msg_status & RMIXL_MSG_STS0_SCF) != 0) {
+			uint32_t r;
+			u_int regno = RMIXL_COP_2_CREDITS+fmnp->fmn_core;
+			u_int sel = fmnp->fmn_thread;
+			printf("%s: CC[%d,%d]=", __func__, regno, sel);
+			FMN_CP2_RD_SWITCH_RW(READ, regno, sel, r);
+			printf("%s: CC[%d,%d]=%d\n", __func__, regno, sel, r);
+		}
+#endif	/* DEBUG */
+#endif	/* 0 */
+		/* clear status1 error(s) */
+		if ((msg_status1 & RMIXL_MSG_STS1_ERRS) != 0) {
+			RMIXL_MFC2(RMIXL_COP_2_MSG_STS, 1, msg_status1);
+			RMIXL_MTC2(RMIXL_COP_2_MSG_STS, 1, msg_status1);
+		}
+		DIAG_PRF(("%s: src=%ld, dst=%d, sts=%#x, %#x: send error, try %d\n",
+			__func__, cpu_number(), dest_id, msg_status, msg_status1, try));
+		DELAY(10);
+	}
+	rv = -1;
+ out:
+	rmixl_cp2_restore(cp0_status);
+	mutex_exit(fmnp->fmn_lock);
+
+	return rv;
+}
+
+/*
+ * rmixl_fmn_msg_recv
+ *
+ *	- grab fmn_lock and call rmixl_fmn_msg_recv_subr to do the real work
+ *	- assume cp2 access is already enabled
+ */
+int
+rmixl_fmn_msg_recv(u_int bucket, rmixl_fmn_rxmsg_t *rxmsg)
+{
+	fmn_t *fmnp;
+	int rv;
+
+	fmnp = fmn_lookup(cpu_number());
+	mutex_enter(fmnp->fmn_lock);
+	rv = rmixl_fmn_msg_recv_subr(bucket, rxmsg);
+	mutex_exit(fmnp->fmn_lock);
+
+	return rv;
+}
+
+/*
+ * rmixl_fmn_msg_recv_subr
+ *
+ *	- assume fmn_lock is owned
+ *	- assume cp2 access is already enabled 
+ */
+static int
+rmixl_fmn_msg_recv_subr(u_int bucket, rmixl_fmn_rxmsg_t *rxmsg)
+{
+	fmn_t *fmnp;
+	uint32_t msg_status;
+	int rv;
+
+	fmnp = fmn_lookup(cpu_number());
+	KASSERT(mutex_owned(fmnp->fmn_lock) != 0);
+
+	for (int try=16; try--; ) {
+		RMIXL_MFC2(RMIXL_COP_2_MSG_STS, 0, msg_status);
+		if ((msg_status & (RMIXL_MSG_STS0_LPF)) == 0)
+			goto recv;
+	}
+	DIAG_PRF(("%s: cpu=%ld, bucket=%d, sts=%#x: Load Pending Fail\n",
+		__func__, cpu_number(), bucket, msg_status));
+	rv = -1;
+	goto out;
+ recv:
+	rmixl_msgld(bucket);
+	RMIXL_MFC2(RMIXL_COP_2_MSG_STS, 0, msg_status);
+	DPRINTF(("%s: cpu=%ld, bucket=%d, sts=%#x\n",
+		__func__, cpu_number(), bucket, msg_status));
+	rv = msg_status & (RMIXL_MSG_STS0_LEF|RMIXL_MSG_STS0_LPF);
+	if (rv == 0) {
+		rxmsg->rxsid = (msg_status & RMIXL_MSG_STS0_RMSID)
+				>> RMIXL_MSG_STS0_RMSID_SHIFT;
+		rxmsg->code = (msg_status & RMIXL_MSG_STS0_RMSC)
+				>> RMIXL_MSG_STS0_RMSC_SHIFT;
+		rxmsg->size = ((msg_status & RMIXL_MSG_STS0_RMS)
+				>> RMIXL_MSG_STS0_RMS_SHIFT) + 1;
+		switch(rxmsg->size) {
+		case 1:
+			RMIXL_DMFC2(RMIXL_COP_2_RXBUF, 0, rxmsg->msg.data[0]);
+			break;
+		case 2:
+			RMIXL_DMFC2(RMIXL_COP_2_RXBUF, 0, rxmsg->msg.data[0]);
+			RMIXL_DMFC2(RMIXL_COP_2_RXBUF, 1, rxmsg->msg.data[1]);
+			break;
+		case 3:
+			RMIXL_DMFC2(RMIXL_COP_2_RXBUF, 0, rxmsg->msg.data[0]);
+			RMIXL_DMFC2(RMIXL_COP_2_RXBUF, 1, rxmsg->msg.data[1]);
+			RMIXL_DMFC2(RMIXL_COP_2_RXBUF, 2, rxmsg->msg.data[2]);
+			break;
+		case 4:
+			RMIXL_DMFC2(RMIXL_COP_2_RXBUF, 0, rxmsg->msg.data[0]);
+			RMIXL_DMFC2(RMIXL_COP_2_RXBUF, 1, rxmsg->msg.data[1]);
+			RMIXL_DMFC2(RMIXL_COP_2_RXBUF, 2, rxmsg->msg.data[2]);
+			RMIXL_DMFC2(RMIXL_COP_2_RXBUF, 3, rxmsg->msg.data[3]);
+			break;
+		default:
+			/* "impossible" due to bitfield width */
+			panic("%s: bad size %d", __func__, rxmsg->size);
+		}
+	}
+ out:
+
+	return rv;
+}
+
+#ifdef FMN_DEBUG
+void
+rmixl_fmn_cp2_dump(void)
+{
+	uint32_t cp0_status;
+
+	cp0_status = rmixl_cp2_enable();
+
+	CPU2_PRINT_8(RMIXL_COP_2_TXBUF, 0);
+	CPU2_PRINT_8(RMIXL_COP_2_TXBUF, 1);
+	CPU2_PRINT_8(RMIXL_COP_2_TXBUF, 2);
+	CPU2_PRINT_8(RMIXL_COP_2_TXBUF, 3);
+
+	CPU2_PRINT_8(RMIXL_COP_2_RXBUF, 0);
+	CPU2_PRINT_8(RMIXL_COP_2_RXBUF, 1);
+	CPU2_PRINT_8(RMIXL_COP_2_RXBUF, 2);
+	CPU2_PRINT_8(RMIXL_COP_2_RXBUF, 3);
+
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_STS, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_STS, 1);
+
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_CFG, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_CFG, 1);
+
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_BSZ, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_BSZ, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_BSZ, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_BSZ, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_BSZ, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_BSZ, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_BSZ, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_MSG_BSZ, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 0, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 0, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 0, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 0, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 0, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 0, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 0, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 0, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 1, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 1, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 1, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 1, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 1, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 1, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 1, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 1, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 2, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 2, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 2, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 2, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 2, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 2, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 2, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 2, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 3, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 3, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 3, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 3, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 3, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 3, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 3, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 3, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 4, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 4, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 4, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 4, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 4, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 4, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 4, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 4, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 5, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 5, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 5, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 5, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 5, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 5, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 5, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 5, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 6, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 6, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 6, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 6, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 6, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 6, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 6, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 6, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 7, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 7, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 7, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 7, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 7, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 7, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 7, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 7, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 8, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 8, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 8, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 8, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 8, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 8, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 8, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 8, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 9, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 9, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 9, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 9, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 9, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 9, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 9, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 9, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 10, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 10, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 10, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 10, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 10, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 10, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 10, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 10, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 11, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 11, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 11, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 11, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 11, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 11, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 11, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 11, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 12, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 12, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 12, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 12, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 12, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 12, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 12, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 12, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 13, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 13, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 13, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 13, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 13, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 13, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 13, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 13, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 14, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 14, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 14, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 14, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 14, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 14, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 14, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 14, 7);
+
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 15, 0);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 15, 1);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 15, 2);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 15, 3);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 15, 4);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 15, 5);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 15, 6);
+	CPU2_PRINT_4(RMIXL_COP_2_CREDITS + 15, 7);
+
+	rmixl_cp2_restore(cp0_status);
+}
+
+
+void
+rmixl_fmn_cc_dump(void)
+{
+	uint32_t cc[4][8];
+
+	FMN_CP2_4SEL_READ(RMIXL_COP_2_CREDITS,   0, &cc[0][0]);
+	FMN_CP2_4SEL_READ(RMIXL_COP_2_CREDITS,   4, &cc[0][4]);
+	FMN_CP2_4SEL_READ(RMIXL_COP_2_CREDITS+1, 0, &cc[1][0]);
+	FMN_CP2_4SEL_READ(RMIXL_COP_2_CREDITS+1, 4, &cc[1][4]);
+	FMN_CP2_4SEL_READ(RMIXL_COP_2_CREDITS+2, 0, &cc[2][0]);
+	FMN_CP2_4SEL_READ(RMIXL_COP_2_CREDITS+2, 4, &cc[2][4]);
+	FMN_CP2_4SEL_READ(RMIXL_COP_2_CREDITS+3, 0, &cc[3][0]);
+	FMN_CP2_4SEL_READ(RMIXL_COP_2_CREDITS+3, 4, &cc[3][4]);
+
+	printf("%s: cpu %ld\n", __func__, cpu_number());
+	for (int i=0; i < 4; i++) {
+		for (int j=0; j < 8; j++)
+			printf(" %#x,", cc[i][j]);
+		printf("\n");
+	}
+}
+
+#endif	/* FMN_DEBUG */
Index: src/sys/arch/mips/rmi/rmixl_fmnvar.h
diff -u /dev/null src/sys/arch/mips/rmi/rmixl_fmnvar.h:1.1.2.1
--- /dev/null	Sun Mar 21 20:41:43 2010
+++ src/sys/arch/mips/rmi/rmixl_fmnvar.h	Sun Mar 21 20:41:43 2010
@@ -0,0 +1,231 @@
+/*	$Id: rmixl_fmnvar.h,v 1.1.2.1 2010/03/21 20:41:43 cliff Exp $	*/
+
+#ifndef _ARCH_MIPS_RMIXL_RMIXL_FMNVAR_H_
+#define _ARCH_MIPS_RMIXL_RMIXL_FMNVAR_H_
+
+#include <mips/cpuregs.h>
+
+#define RMIXL_FMN_CODE_PSB_WAKEUP	200	/* firmware MSGRNG_CODE_BOOT_WAKEUP */
+#define RMIXL_FMN_CODE_HELLO_REQ	201
+#define RMIXL_FMN_CODE_HELLO_ACK	202
+
+#define RMIXL_FMN_HELLO_REQ_SZ		4
+#define RMIXL_FMN_HELLO_ACK_SZ		4
+
+typedef struct rmixl_fmn_msg {
+	uint64_t data[4];
+} rmixl_fmn_msg_t;
+
+typedef struct rmixl_fmn_rxmsg {
+	u_int rxsid;
+	u_int code;
+	u_int size;
+	rmixl_fmn_msg_t msg;
+} rmixl_fmn_rxmsg_t;
+
+
+/*
+ * compute FMN dest_id from MIPS cpuid
+ * - each Core FMN sation has 8 buckets
+ * - each Core has 4 threads
+ * - here we use 1 bucket per thread
+ *   (the first four buckets)
+ * - if we need { hi, lo } priority buckets per thread
+ *   need to adjust the RMIXL_FMN_DESTID macro
+ *   and use the 'pri' parameter
+ * - i.e. for now there is only one priority
+ */
+#define RMIXL_CPU_CORE(cpuid)	((uint32_t)((cpuid) & __BITS(9,0)) >> 2)
+#define RMIXL_CPU_THREAD(cpuid)	((uint32_t)((cpuid) & __BITS(1,0)))
+#define RMIXL_FMN_CORE_DESTID(core, bucket)	\
+		 (((core) << 3) | (bucket))
+
+
+#define RMIXL_DMFC2(regnum, sel, rv)					\
+do {									\
+	uint64_t __val;							\
+									\
+	__asm volatile(							\
+		".set push" 			"\n\t"			\
+		".set mips64"			"\n\t"			\
+		".set noat"			"\n\t"			\
+		"dmfc2 %0,$%1,%2"		"\n\t"			\
+		".set pop"			"\n\t"			\
+	    : "=r"(__val) : "n"(regnum), "n"(sel));			\
+	rv = __val;							\
+} while (0)
+
+#define RMIXL_DMTC2(regnum, sel, val)					\
+do {									\
+	uint64_t __val = val;						\
+									\
+	__asm volatile(							\
+		".set push" 			"\n\t"			\
+		".set mips64"			"\n\t"			\
+		".set noat"			"\n\t"			\
+		"dmtc2 %0,$%1,%2"		"\n\t"			\
+		".set pop"			"\n\t"			\
+	    :: "r"(__val), "n"(regnum), "n"(sel));			\
+} while (0)
+
+#define RMIXL_MFC2(regnum, sel, rv)					\
+do {									\
+	uint32_t __val;							\
+									\
+	__asm volatile(							\
+		".set push"			"\n\t"			\
+		".set mips64"			"\n\t"			\
+		"mfc2 %0,$%1,%2"		"\n\t"			\
+		".set pop"			"\n\t"			\
+	    : "=r"(__val) : "n"(regnum), "n"(sel));			\
+	rv = __val;							\
+} while (0)
+
+#define RMIXL_MTC2(regnum, sel, val)					\
+do {									\
+	uint32_t __val = val;						\
+									\
+	__asm volatile(							\
+		".set push"			"\n\t"			\
+		".set mips64"			"\n\t"			\
+		"mtc2 %0,$%1,%2"		"\n\t"			\
+		".set pop"			"\n\t"			\
+	    :: "r"(__val), "n"(regnum), "n"(sel));			\
+} while (0)
+
+#define CPU2_PRINT_8(regno, sel)					\
+do {									\
+	uint64_t r;							\
+	RMIXL_DMFC2(regno, sel, r);					\
+	printf("%s: CP2(%d,%d) = %#"PRIx64"\n",				\
+		__func__, regno, sel, r);				\
+} while (0)
+
+#define CPU2_PRINT_4(regno, sel)					\
+do {									\
+	uint32_t r;							\
+	RMIXL_MFC2(regno, sel, r);					\
+	printf("%s: CP2(%d,%d) = %#x\n",				\
+		__func__, regno, sel, r);				\
+} while (0)
+
+
+/*
+ * encode 'dest' for msgsnd op 'rt'
+ */
+#define RMIXL_MSGSND_DESC(size, code, dest_id)	\
+		((((size) - 1) << 16) | ((code) << 8) | (dest_id))
+
+static inline void
+rmixl_msgsnd(uint32_t desc)
+{
+	__asm__ volatile (
+		".set push"		"\n\t"
+		".set noreorder"	"\n\t"
+		".set mips64"		"\n\t"
+		"sync"			"\n\t"
+		"msgsnd %0"		"\n\t"
+		".set pop"		"\n\t"
+			:: "r"(desc));
+}
+
+static inline void
+rmixl_msgld(uint32_t bucket)
+{
+	__asm__ volatile (
+		".set push"		"\n\t"
+		".set noreorder"	"\n\t"
+		".set mips64"		"\n\t"
+		"msgld %0"		"\n\t"
+		".set pop"		"\n\t"
+			:: "r"(bucket));
+}
+
+/*
+ * the seemingly-spurious add is recommended by RMI
+ * see XLS PRM (rev. 3.21) 5.3.9
+ */
+static inline void
+rmixl_fmn_msgwait(u_int mask)
+{
+	__asm__ volatile (
+		".set push"		"\n\t"
+		".set noreorder"	"\n\t"
+		".set mips64"		"\n\t"
+		"addu %0,%0,0"		"\n\t"
+		"msgwait %0"		"\n\t"
+		".set pop"		"\n\t"
+			:: "r"(mask));
+}
+
+static inline uint32_t
+rmixl_cp2_enable(void)
+{
+	uint32_t rv;
+	uint32_t cu2;
+
+	KASSERT(curcpu()->ci_cpl == IPL_HIGH);
+	__asm volatile(
+		".set push"		"\n\t"
+		".set noreorder"	"\n\t"
+		"li	%1,%3"		"\n\t"
+		"mfc0	%0,$%2"		"\n\t"
+		"or	%1,%1,%0"	"\n\t"
+		"mtc0	%1,$%2"		"\n\t"
+		".set pop"		"\n\t"
+			: "=r"(rv), "=r"(cu2)
+			: "n"(MIPS_COP_0_STATUS), "n"(1 << 30));
+
+	return (rv & (1 << 30));
+}
+
+static inline void
+rmixl_cp2_restore(uint32_t ocu)
+{
+	uint32_t cu2;
+	uint32_t mask = ~(1 << 30);
+
+	KASSERT(curcpu()->ci_cpl == IPL_HIGH);
+	__asm volatile(
+		".set push"		"\n\t"
+		".set noreorder"	"\n\t"
+		"mfc0	%0,$%1"		"\n\t"
+		"and	%0,%2,%0"	"\n\t"
+		"or	%0,%3,%0"	"\n\t"
+		"mtc0	%0,$%1"		"\n\t"
+		".set pop"		"\n\t"
+			: "=r"(cu2)
+			: "n"(MIPS_COP_0_STATUS), "r"(mask), "r"(ocu));
+}
+
+/*
+ * logical station IDs
+ * see Table 12.1 in the XLS PRM
+ */
+#define RMIXL_FMN_STID_CORE0		0
+#define RMIXL_FMN_STID_CORE1		1
+#define RMIXL_FMN_STID_CORE2		2
+#define RMIXL_FMN_STID_CORE3		3
+#define RMIXL_FMN_STID_GMAC_Q0		4
+#define RMIXL_FMN_STID_GMAC_Q1		5
+#define RMIXL_FMN_STID_DMA		6
+#define RMIXL_FMN_STID_CDE		7
+#define RMIXL_FMN_STID_PCIE		8
+#define RMIXL_FMN_STID_SAE		9
+#define RMIXL_FMN_NSTID		(RMIXL_FMN_STID_SAE+1)
+#define RMIXL_FMN_STID_RESERVED	-1
+
+#define RMIXL_FMN_INTR_IPL	IPL_HIGH
+
+void	rmixl_fmn_init(void);
+void	rmixl_fmn_init_core(void);
+void	rmixl_fmn_init_cpu_intr(void);
+void   *rmixl_fmn_intr_establish(int, int (*)(void *, rmixl_fmn_rxmsg_t *), void *);
+void	rmixl_fmn_intr_disestablish(void *);
+void	rmixl_fmn_intr_poll(u_int, rmixl_fmn_rxmsg_t *);
+int	rmixl_fmn_msg_send(u_int, u_int, u_int, rmixl_fmn_msg_t *);
+int	rmixl_fmn_msg_recv(u_int, rmixl_fmn_rxmsg_t *);
+
+
+
+#endif	/* _ARCH_MIPS_RMIXL_RMIXL_FMNVAR_H_ */

Reply via email to