soc/fsl: Introduce DPAA 1.x QMan device driver

This driver enables the Freescale DPAA 1.x Queue Manager block.
QMan is a hardware accelerator that manages frame queues.  It allows
CPUs and other accelerators connected to the SoC datapath to enqueue
and dequeue ethernet frames, thus providing the infrastructure for
data exchange among CPUs and datapath accelerators.

Signed-off-by: Roy Pledge <roy.pledge@nxp.com>
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: Scott Wood <oss@buserror.net>

5 files changed, 4418 insertions, 2 deletions

diff --git a/drivers/soc/fsl/qbman/Makefile b/drivers/soc/fsl/qbman/Makefile
index 855c3ac0f89e..6e0ee30b2834 100644
--- a/drivers/soc/fsl/qbman/Makefile
+++ b/drivers/soc/fsl/qbman/Makefile
@@ -1,2 +1,3 @@
-obj-$(CONFIG_FSL_DPAA)                          += bman_ccsr.o bman_portal.o \
-						   bman.o
+obj-$(CONFIG_FSL_DPAA)                          += bman_ccsr.o qman_ccsr.o \
+						   bman_portal.o qman_portal.o \
+						   bman.o qman.o
diff --git a/drivers/soc/fsl/qbman/qman.c b/drivers/soc/fsl/qbman/qman.c
new file mode 100644
index 000000000000..119054bc922b
--- /dev/null
+++ b/drivers/soc/fsl/qbman/qman.c
@@ -0,0 +1,2881 @@
+/* Copyright 2008 - 2016 Freescale Semiconductor, Inc.
+ *
+ * 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 Freescale Semiconductor nor the
+ *	 names of its contributors may be used to endorse or promote products
+ *	 derived from this software without specific prior written permission.
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``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 Freescale Semiconductor 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.
+ */
+
+#include "qman_priv.h"
+
+#define DQRR_MAXFILL	15
+#define EQCR_ITHRESH	4	/* if EQCR congests, interrupt threshold */
+#define IRQNAME		"QMan portal %d"
+#define MAX_IRQNAME	16	/* big enough for "QMan portal %d" */
+#define QMAN_POLL_LIMIT 32
+#define QMAN_PIRQ_DQRR_ITHRESH 12
+#define QMAN_PIRQ_MR_ITHRESH 4
+#define QMAN_PIRQ_IPERIOD 100
+
+/* Portal register assists */
+
+/* Cache-inhibited register offsets */
+#define QM_REG_EQCR_PI_CINH	0x0000
+#define QM_REG_EQCR_CI_CINH	0x0004
+#define QM_REG_EQCR_ITR		0x0008
+#define QM_REG_DQRR_PI_CINH	0x0040
+#define QM_REG_DQRR_CI_CINH	0x0044
+#define QM_REG_DQRR_ITR		0x0048
+#define QM_REG_DQRR_DCAP	0x0050
+#define QM_REG_DQRR_SDQCR	0x0054
+#define QM_REG_DQRR_VDQCR	0x0058
+#define QM_REG_DQRR_PDQCR	0x005c
+#define QM_REG_MR_PI_CINH	0x0080
+#define QM_REG_MR_CI_CINH	0x0084
+#define QM_REG_MR_ITR		0x0088
+#define QM_REG_CFG		0x0100
+#define QM_REG_ISR		0x0e00
+#define QM_REG_IER		0x0e04
+#define QM_REG_ISDR		0x0e08
+#define QM_REG_IIR		0x0e0c
+#define QM_REG_ITPR		0x0e14
+
+/* Cache-enabled register offsets */
+#define QM_CL_EQCR		0x0000
+#define QM_CL_DQRR		0x1000
+#define QM_CL_MR		0x2000
+#define QM_CL_EQCR_PI_CENA	0x3000
+#define QM_CL_EQCR_CI_CENA	0x3100
+#define QM_CL_DQRR_PI_CENA	0x3200
+#define QM_CL_DQRR_CI_CENA	0x3300
+#define QM_CL_MR_PI_CENA	0x3400
+#define QM_CL_MR_CI_CENA	0x3500
+#define QM_CL_CR		0x3800
+#define QM_CL_RR0		0x3900
+#define QM_CL_RR1		0x3940
+
+/*
+ * BTW, the drivers (and h/w programming model) already obtain the required
+ * synchronisation for portal accesses and data-dependencies. Use of barrier()s
+ * or other order-preserving primitives simply degrade performance. Hence the
+ * use of the __raw_*() interfaces, which simply ensure that the compiler treats
+ * the portal registers as volatile
+ */
+
+/* Cache-enabled ring access */
+#define qm_cl(base, idx)	((void *)base + ((idx) << 6))
+
+/*
+ * Portal modes.
+ *   Enum types;
+ *     pmode == production mode
+ *     cmode == consumption mode,
+ *     dmode == h/w dequeue mode.
+ *   Enum values use 3 letter codes. First letter matches the portal mode,
+ *   remaining two letters indicate;
+ *     ci == cache-inhibited portal register
+ *     ce == cache-enabled portal register
+ *     vb == in-band valid-bit (cache-enabled)
+ *     dc == DCA (Discrete Consumption Acknowledgment), DQRR-only
+ *   As for "enum qm_dqrr_dmode", it should be self-explanatory.
+ */
+enum qm_eqcr_pmode {		/* matches QCSP_CFG::EPM */
+	qm_eqcr_pci = 0,	/* PI index, cache-inhibited */
+	qm_eqcr_pce = 1,	/* PI index, cache-enabled */
+	qm_eqcr_pvb = 2		/* valid-bit */
+};
+enum qm_dqrr_dmode {		/* matches QCSP_CFG::DP */
+	qm_dqrr_dpush = 0,	/* SDQCR  + VDQCR */
+	qm_dqrr_dpull = 1	/* PDQCR */
+};
+enum qm_dqrr_pmode {		/* s/w-only */
+	qm_dqrr_pci,		/* reads DQRR_PI_CINH */
+	qm_dqrr_pce,		/* reads DQRR_PI_CENA */
+	qm_dqrr_pvb		/* reads valid-bit */
+};
+enum qm_dqrr_cmode {		/* matches QCSP_CFG::DCM */
+	qm_dqrr_cci = 0,	/* CI index, cache-inhibited */
+	qm_dqrr_cce = 1,	/* CI index, cache-enabled */
+	qm_dqrr_cdc = 2		/* Discrete Consumption Acknowledgment */
+};
+enum qm_mr_pmode {		/* s/w-only */
+	qm_mr_pci,		/* reads MR_PI_CINH */
+	qm_mr_pce,		/* reads MR_PI_CENA */
+	qm_mr_pvb		/* reads valid-bit */
+};
+enum qm_mr_cmode {		/* matches QCSP_CFG::MM */
+	qm_mr_cci = 0,		/* CI index, cache-inhibited */
+	qm_mr_cce = 1		/* CI index, cache-enabled */
+};
+
+/* --- Portal structures --- */
+
+#define QM_EQCR_SIZE		8
+#define QM_DQRR_SIZE		16
+#define QM_MR_SIZE		8
+
+/* "Enqueue Command" */
+struct qm_eqcr_entry {
+	u8 _ncw_verb; /* writes to this are non-coherent */
+	u8 dca;
+	u16 seqnum;
+	u32 orp;	/* 24-bit */
+	u32 fqid;	/* 24-bit */
+	u32 tag;
+	struct qm_fd fd;
+	u8 __reserved3[32];
+} __packed;
+#define QM_EQCR_VERB_VBIT		0x80
+#define QM_EQCR_VERB_CMD_MASK		0x61	/* but only one value; */
+#define QM_EQCR_VERB_CMD_ENQUEUE	0x01
+#define QM_EQCR_SEQNUM_NESN		0x8000	/* Advance NESN */
+#define QM_EQCR_SEQNUM_NLIS		0x4000	/* More fragments to come */
+#define QM_EQCR_SEQNUM_SEQMASK		0x3fff	/* sequence number goes here */
+
+struct qm_eqcr {
+	struct qm_eqcr_entry *ring, *cursor;
+	u8 ci, available, ithresh, vbit;
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	u32 busy;
+	enum qm_eqcr_pmode pmode;
+#endif
+};
+
+struct qm_dqrr {
+	const struct qm_dqrr_entry *ring, *cursor;
+	u8 pi, ci, fill, ithresh, vbit;
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	enum qm_dqrr_dmode dmode;
+	enum qm_dqrr_pmode pmode;
+	enum qm_dqrr_cmode cmode;
+#endif
+};
+
+struct qm_mr {
+	union qm_mr_entry *ring, *cursor;
+	u8 pi, ci, fill, ithresh, vbit;
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	enum qm_mr_pmode pmode;
+	enum qm_mr_cmode cmode;
+#endif
+};
+
+/* MC (Management Command) command */
+/* "Query FQ" */
+struct qm_mcc_queryfq {
+	u8 _ncw_verb;
+	u8 __reserved1[3];
+	u32 fqid;	/* 24-bit */
+	u8 __reserved2[56];
+} __packed;
+/* "Alter FQ State Commands " */
+struct qm_mcc_alterfq {
+	u8 _ncw_verb;
+	u8 __reserved1[3];
+	u32 fqid;	/* 24-bit */
+	u8 __reserved2;
+	u8 count;	/* number of consecutive FQID */
+	u8 __reserved3[10];
+	u32 context_b;	/* frame queue context b */
+	u8 __reserved4[40];
+} __packed;
+
+/* "Query CGR" */
+struct qm_mcc_querycgr {
+	u8 _ncw_verb;
+	u8 __reserved1[30];
+	u8 cgid;
+	u8 __reserved2[32];
+};
+
+struct qm_mcc_querywq {
+	u8 _ncw_verb;
+	u8 __reserved;
+	/* select channel if verb != QUERYWQ_DEDICATED */
+	u16 channel_wq; /* ignores wq (3 lsbits): _res[0-2] */
+	u8 __reserved2[60];
+} __packed;
+
+#define QM_MCC_VERB_VBIT		0x80
+#define QM_MCC_VERB_MASK		0x7f	/* where the verb contains; */
+#define QM_MCC_VERB_INITFQ_PARKED	0x40
+#define QM_MCC_VERB_INITFQ_SCHED	0x41
+#define QM_MCC_VERB_QUERYFQ		0x44
+#define QM_MCC_VERB_QUERYFQ_NP		0x45	/* "non-programmable" fields */
+#define QM_MCC_VERB_QUERYWQ		0x46
+#define QM_MCC_VERB_QUERYWQ_DEDICATED	0x47
+#define QM_MCC_VERB_ALTER_SCHED		0x48	/* Schedule FQ */
+#define QM_MCC_VERB_ALTER_FE		0x49	/* Force Eligible FQ */
+#define QM_MCC_VERB_ALTER_RETIRE	0x4a	/* Retire FQ */
+#define QM_MCC_VERB_ALTER_OOS		0x4b	/* Take FQ out of service */
+#define QM_MCC_VERB_ALTER_FQXON		0x4d	/* FQ XON */
+#define QM_MCC_VERB_ALTER_FQXOFF	0x4e	/* FQ XOFF */
+#define QM_MCC_VERB_INITCGR		0x50
+#define QM_MCC_VERB_MODIFYCGR		0x51
+#define QM_MCC_VERB_CGRTESTWRITE	0x52
+#define QM_MCC_VERB_QUERYCGR		0x58
+#define QM_MCC_VERB_QUERYCONGESTION	0x59
+union qm_mc_command {
+	struct {
+		u8 _ncw_verb; /* writes to this are non-coherent */
+		u8 __reserved[63];
+	};
+	struct qm_mcc_initfq initfq;
+	struct qm_mcc_queryfq queryfq;
+	struct qm_mcc_alterfq alterfq;
+	struct qm_mcc_initcgr initcgr;
+	struct qm_mcc_querycgr querycgr;
+	struct qm_mcc_querywq querywq;
+	struct qm_mcc_queryfq_np queryfq_np;
+};
+
+/* MC (Management Command) result */
+/* "Query FQ" */
+struct qm_mcr_queryfq {
+	u8 verb;
+	u8 result;
+	u8 __reserved1[8];
+	struct qm_fqd fqd;	/* the FQD fields are here */
+	u8 __reserved2[30];
+} __packed;
+
+/* "Alter FQ State Commands" */
+struct qm_mcr_alterfq {
+	u8 verb;
+	u8 result;
+	u8 fqs;		/* Frame Queue Status */
+	u8 __reserved1[61];
+};
+#define QM_MCR_VERB_RRID		0x80
+#define QM_MCR_VERB_MASK		QM_MCC_VERB_MASK
+#define QM_MCR_VERB_INITFQ_PARKED	QM_MCC_VERB_INITFQ_PARKED
+#define QM_MCR_VERB_INITFQ_SCHED	QM_MCC_VERB_INITFQ_SCHED
+#define QM_MCR_VERB_QUERYFQ		QM_MCC_VERB_QUERYFQ
+#define QM_MCR_VERB_QUERYFQ_NP		QM_MCC_VERB_QUERYFQ_NP
+#define QM_MCR_VERB_QUERYWQ		QM_MCC_VERB_QUERYWQ
+#define QM_MCR_VERB_QUERYWQ_DEDICATED	QM_MCC_VERB_QUERYWQ_DEDICATED
+#define QM_MCR_VERB_ALTER_SCHED		QM_MCC_VERB_ALTER_SCHED
+#define QM_MCR_VERB_ALTER_FE		QM_MCC_VERB_ALTER_FE
+#define QM_MCR_VERB_ALTER_RETIRE	QM_MCC_VERB_ALTER_RETIRE
+#define QM_MCR_VERB_ALTER_OOS		QM_MCC_VERB_ALTER_OOS
+#define QM_MCR_RESULT_NULL		0x00
+#define QM_MCR_RESULT_OK		0xf0
+#define QM_MCR_RESULT_ERR_FQID		0xf1
+#define QM_MCR_RESULT_ERR_FQSTATE	0xf2
+#define QM_MCR_RESULT_ERR_NOTEMPTY	0xf3	/* OOS fails if FQ is !empty */
+#define QM_MCR_RESULT_ERR_BADCHANNEL	0xf4
+#define QM_MCR_RESULT_PENDING		0xf8
+#define QM_MCR_RESULT_ERR_BADCOMMAND	0xff
+#define QM_MCR_FQS_ORLPRESENT		0x02	/* ORL fragments to come */
+#define QM_MCR_FQS_NOTEMPTY		0x01	/* FQ has enqueued frames */
+#define QM_MCR_TIMEOUT			10000	/* us */
+union qm_mc_result {
+	struct {
+		u8 verb;
+		u8 result;
+		u8 __reserved1[62];
+	};
+	struct qm_mcr_queryfq queryfq;
+	struct qm_mcr_alterfq alterfq;
+	struct qm_mcr_querycgr querycgr;
+	struct qm_mcr_querycongestion querycongestion;
+	struct qm_mcr_querywq querywq;
+	struct qm_mcr_queryfq_np queryfq_np;
+};
+
+struct qm_mc {
+	union qm_mc_command *cr;
+	union qm_mc_result *rr;
+	u8 rridx, vbit;
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	enum {
+		/* Can be _mc_start()ed */
+		qman_mc_idle,
+		/* Can be _mc_commit()ed or _mc_abort()ed */
+		qman_mc_user,
+		/* Can only be _mc_retry()ed */
+		qman_mc_hw
+	} state;
+#endif
+};
+
+struct qm_addr {
+	void __iomem *ce;	/* cache-enabled */
+	void __iomem *ci;	/* cache-inhibited */
+};
+
+struct qm_portal {
+	/*
+	 * In the non-CONFIG_FSL_DPAA_CHECKING case, the following stuff up to
+	 * and including 'mc' fits within a cacheline (yay!). The 'config' part
+	 * is setup-only, so isn't a cause for a concern. In other words, don't
+	 * rearrange this structure on a whim, there be dragons ...
+	 */
+	struct qm_addr addr;
+	struct qm_eqcr eqcr;
+	struct qm_dqrr dqrr;
+	struct qm_mr mr;
+	struct qm_mc mc;
+} ____cacheline_aligned;
+
+/* Cache-inhibited register access. */
+static inline u32 qm_in(struct qm_portal *p, u32 offset)
+{
+	return __raw_readl(p->addr.ci + offset);
+}
+
+static inline void qm_out(struct qm_portal *p, u32 offset, u32 val)
+{
+	__raw_writel(val, p->addr.ci + offset);
+}
+
+/* Cache Enabled Portal Access */
+static inline void qm_cl_invalidate(struct qm_portal *p, u32 offset)
+{
+	dpaa_invalidate(p->addr.ce + offset);
+}
+
+static inline void qm_cl_touch_ro(struct qm_portal *p, u32 offset)
+{
+	dpaa_touch_ro(p->addr.ce + offset);
+}
+
+static inline u32 qm_ce_in(struct qm_portal *p, u32 offset)
+{
+	return __raw_readl(p->addr.ce + offset);
+}
+
+/* --- EQCR API --- */
+
+#define EQCR_SHIFT	ilog2(sizeof(struct qm_eqcr_entry))
+#define EQCR_CARRY	(uintptr_t)(QM_EQCR_SIZE << EQCR_SHIFT)
+
+/* Bit-wise logic to wrap a ring pointer by clearing the "carry bit" */
+static struct qm_eqcr_entry *eqcr_carryclear(struct qm_eqcr_entry *p)
+{
+	uintptr_t addr = (uintptr_t)p;
+
+	addr &= ~EQCR_CARRY;
+
+	return (struct qm_eqcr_entry *)addr;
+}
+
+/* Bit-wise logic to convert a ring pointer to a ring index */
+static int eqcr_ptr2idx(struct qm_eqcr_entry *e)
+{
+	return ((uintptr_t)e >> EQCR_SHIFT) & (QM_EQCR_SIZE - 1);
+}
+
+/* Increment the 'cursor' ring pointer, taking 'vbit' into account */
+static inline void eqcr_inc(struct qm_eqcr *eqcr)
+{
+	/* increment to the next EQCR pointer and handle overflow and 'vbit' */
+	struct qm_eqcr_entry *partial = eqcr->cursor + 1;
+
+	eqcr->cursor = eqcr_carryclear(partial);
+	if (partial != eqcr->cursor)
+		eqcr->vbit ^= QM_EQCR_VERB_VBIT;
+}
+
+static inline int qm_eqcr_init(struct qm_portal *portal,
+				enum qm_eqcr_pmode pmode,
+				unsigned int eq_stash_thresh,
+				int eq_stash_prio)
+{
+	struct qm_eqcr *eqcr = &portal->eqcr;
+	u32 cfg;
+	u8 pi;
+
+	eqcr->ring = portal->addr.ce + QM_CL_EQCR;
+	eqcr->ci = qm_in(portal, QM_REG_EQCR_CI_CINH) & (QM_EQCR_SIZE - 1);
+	qm_cl_invalidate(portal, QM_CL_EQCR_CI_CENA);
+	pi = qm_in(portal, QM_REG_EQCR_PI_CINH) & (QM_EQCR_SIZE - 1);
+	eqcr->cursor = eqcr->ring + pi;
+	eqcr->vbit = (qm_in(portal, QM_REG_EQCR_PI_CINH) & QM_EQCR_SIZE) ?
+		     QM_EQCR_VERB_VBIT : 0;
+	eqcr->available = QM_EQCR_SIZE - 1 -
+			  dpaa_cyc_diff(QM_EQCR_SIZE, eqcr->ci, pi);
+	eqcr->ithresh = qm_in(portal, QM_REG_EQCR_ITR);
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	eqcr->busy = 0;
+	eqcr->pmode = pmode;
+#endif
+	cfg = (qm_in(portal, QM_REG_CFG) & 0x00ffffff) |
+	      (eq_stash_thresh << 28) | /* QCSP_CFG: EST */
+	      (eq_stash_prio << 26) | /* QCSP_CFG: EP */
+	      ((pmode & 0x3) << 24); /* QCSP_CFG::EPM */
+	qm_out(portal, QM_REG_CFG, cfg);
+	return 0;
+}
+
+static inline unsigned int qm_eqcr_get_ci_stashing(struct qm_portal *portal)
+{
+	return (qm_in(portal, QM_REG_CFG) >> 28) & 0x7;
+}
+
+static inline void qm_eqcr_finish(struct qm_portal *portal)
+{
+	struct qm_eqcr *eqcr = &portal->eqcr;
+	u8 pi = qm_in(portal, QM_REG_EQCR_PI_CINH) & (QM_EQCR_SIZE - 1);
+	u8 ci = qm_in(portal, QM_REG_EQCR_CI_CINH) & (QM_EQCR_SIZE - 1);
+
+	DPAA_ASSERT(!eqcr->busy);
+	if (pi != eqcr_ptr2idx(eqcr->cursor))
+		pr_crit("losing uncommited EQCR entries\n");
+	if (ci != eqcr->ci)
+		pr_crit("missing existing EQCR completions\n");
+	if (eqcr->ci != eqcr_ptr2idx(eqcr->cursor))
+		pr_crit("EQCR destroyed unquiesced\n");
+}
+
+static inline struct qm_eqcr_entry *qm_eqcr_start_no_stash(struct qm_portal
+								 *portal)
+{
+	struct qm_eqcr *eqcr = &portal->eqcr;
+
+	DPAA_ASSERT(!eqcr->busy);
+	if (!eqcr->available)
+		return NULL;
+
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	eqcr->busy = 1;
+#endif
+	dpaa_zero(eqcr->cursor);
+	return eqcr->cursor;
+}
+
+static inline struct qm_eqcr_entry *qm_eqcr_start_stash(struct qm_portal
+								*portal)
+{
+	struct qm_eqcr *eqcr = &portal->eqcr;
+	u8 diff, old_ci;
+
+	DPAA_ASSERT(!eqcr->busy);
+	if (!eqcr->available) {
+		old_ci = eqcr->ci;
+		eqcr->ci = qm_ce_in(portal, QM_CL_EQCR_CI_CENA) &
+			   (QM_EQCR_SIZE - 1);
+		diff = dpaa_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci);
+		eqcr->available += diff;
+		if (!diff)
+			return NULL;
+	}
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	eqcr->busy = 1;
+#endif
+	dpaa_zero(eqcr->cursor);
+	return eqcr->cursor;
+}
+
+static inline void eqcr_commit_checks(struct qm_eqcr *eqcr)
+{
+	DPAA_ASSERT(eqcr->busy);
+	DPAA_ASSERT(eqcr->cursor->orp == (eqcr->cursor->orp & 0x00ffffff));
+	DPAA_ASSERT(eqcr->cursor->fqid == (eqcr->cursor->fqid & 0x00ffffff));
+	DPAA_ASSERT(eqcr->available >= 1);
+}
+
+static inline void qm_eqcr_pvb_commit(struct qm_portal *portal, u8 myverb)
+{
+	struct qm_eqcr *eqcr = &portal->eqcr;
+	struct qm_eqcr_entry *eqcursor;
+
+	eqcr_commit_checks(eqcr);
+	DPAA_ASSERT(eqcr->pmode == qm_eqcr_pvb);
+	dma_wmb();
+	eqcursor = eqcr->cursor;
+	eqcursor->_ncw_verb = myverb | eqcr->vbit;
+	dpaa_flush(eqcursor);
+	eqcr_inc(eqcr);
+	eqcr->available--;
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	eqcr->busy = 0;
+#endif
+}
+
+static inline void qm_eqcr_cce_prefetch(struct qm_portal *portal)
+{
+	qm_cl_touch_ro(portal, QM_CL_EQCR_CI_CENA);
+}
+
+static inline u8 qm_eqcr_cce_update(struct qm_portal *portal)
+{
+	struct qm_eqcr *eqcr = &portal->eqcr;
+	u8 diff, old_ci = eqcr->ci;
+
+	eqcr->ci = qm_ce_in(portal, QM_CL_EQCR_CI_CENA) & (QM_EQCR_SIZE - 1);
+	qm_cl_invalidate(portal, QM_CL_EQCR_CI_CENA);
+	diff = dpaa_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci);
+	eqcr->available += diff;
+	return diff;
+}
+
+static inline void qm_eqcr_set_ithresh(struct qm_portal *portal, u8 ithresh)
+{
+	struct qm_eqcr *eqcr = &portal->eqcr;
+
+	eqcr->ithresh = ithresh;
+	qm_out(portal, QM_REG_EQCR_ITR, ithresh);
+}
+
+static inline u8 qm_eqcr_get_avail(struct qm_portal *portal)
+{
+	struct qm_eqcr *eqcr = &portal->eqcr;
+
+	return eqcr->available;
+}
+
+static inline u8 qm_eqcr_get_fill(struct qm_portal *portal)
+{
+	struct qm_eqcr *eqcr = &portal->eqcr;
+
+	return QM_EQCR_SIZE - 1 - eqcr->available;
+}
+
+/* --- DQRR API --- */
+
+#define DQRR_SHIFT	ilog2(sizeof(struct qm_dqrr_entry))
+#define DQRR_CARRY	(uintptr_t)(QM_DQRR_SIZE << DQRR_SHIFT)
+
+static const struct qm_dqrr_entry *dqrr_carryclear(
+					const struct qm_dqrr_entry *p)
+{
+	uintptr_t addr = (uintptr_t)p;
+
+	addr &= ~DQRR_CARRY;
+
+	return (const struct qm_dqrr_entry *)addr;
+}
+
+static inline int dqrr_ptr2idx(const struct qm_dqrr_entry *e)
+{
+	return ((uintptr_t)e >> DQRR_SHIFT) & (QM_DQRR_SIZE - 1);
+}
+
+static const struct qm_dqrr_entry *dqrr_inc(const struct qm_dqrr_entry *e)
+{
+	return dqrr_carryclear(e + 1);
+}
+
+static inline void qm_dqrr_set_maxfill(struct qm_portal *portal, u8 mf)
+{
+	qm_out(portal, QM_REG_CFG, (qm_in(portal, QM_REG_CFG) & 0xff0fffff) |
+				   ((mf & (QM_DQRR_SIZE - 1)) << 20));
+}
+
+static inline int qm_dqrr_init(struct qm_portal *portal,
+			       const struct qm_portal_config *config,
+			       enum qm_dqrr_dmode dmode,
+			       enum qm_dqrr_pmode pmode,
+			       enum qm_dqrr_cmode cmode, u8 max_fill)
+{
+	struct qm_dqrr *dqrr = &portal->dqrr;
+	u32 cfg;
+
+	/* Make sure the DQRR will be idle when we enable */
+	qm_out(portal, QM_REG_DQRR_SDQCR, 0);
+	qm_out(portal, QM_REG_DQRR_VDQCR, 0);
+	qm_out(portal, QM_REG_DQRR_PDQCR, 0);
+	dqrr->ring = portal->addr.ce + QM_CL_DQRR;
+	dqrr->pi = qm_in(portal, QM_REG_DQRR_PI_CINH) & (QM_DQRR_SIZE - 1);
+	dqrr->ci = qm_in(portal, QM_REG_DQRR_CI_CINH) & (QM_DQRR_SIZE - 1);
+	dqrr->cursor = dqrr->ring + dqrr->ci;
+	dqrr->fill = dpaa_cyc_diff(QM_DQRR_SIZE, dqrr->ci, dqrr->pi);
+	dqrr->vbit = (qm_in(portal, QM_REG_DQRR_PI_CINH) & QM_DQRR_SIZE) ?
+			QM_DQRR_VERB_VBIT : 0;
+	dqrr->ithresh = qm_in(portal, QM_REG_DQRR_ITR);
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	dqrr->dmode = dmode;
+	dqrr->pmode = pmode;
+	dqrr->cmode = cmode;
+#endif
+	/* Invalidate every ring entry before beginning */
+	for (cfg = 0; cfg < QM_DQRR_SIZE; cfg++)
+		dpaa_invalidate(qm_cl(dqrr->ring, cfg));
+	cfg = (qm_in(portal, QM_REG_CFG) & 0xff000f00) |
+		((max_fill & (QM_DQRR_SIZE - 1)) << 20) | /* DQRR_MF */
+		((dmode & 1) << 18) |			/* DP */
+		((cmode & 3) << 16) |			/* DCM */
+		0xa0 |					/* RE+SE */
+		(0 ? 0x40 : 0) |			/* Ignore RP */
+		(0 ? 0x10 : 0);				/* Ignore SP */
+	qm_out(portal, QM_REG_CFG, cfg);
+	qm_dqrr_set_maxfill(portal, max_fill);
+	return 0;
+}
+
+static inline void qm_dqrr_finish(struct qm_portal *portal)
+{
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	struct qm_dqrr *dqrr = &portal->dqrr;
+
+	if (dqrr->cmode != qm_dqrr_cdc &&
+	    dqrr->ci != dqrr_ptr2idx(dqrr->cursor))
+		pr_crit("Ignoring completed DQRR entries\n");
+#endif
+}
+
+static inline const struct qm_dqrr_entry *qm_dqrr_current(
+						struct qm_portal *portal)
+{
+	struct qm_dqrr *dqrr = &portal->dqrr;
+
+	if (!dqrr->fill)
+		return NULL;
+	return dqrr->cursor;
+}
+
+static inline u8 qm_dqrr_next(struct qm_portal *portal)
+{
+	struct qm_dqrr *dqrr = &portal->dqrr;
+
+	DPAA_ASSERT(dqrr->fill);
+	dqrr->cursor = dqrr_inc(dqrr->cursor);
+	return --dqrr->fill;
+}
+
+static inline void qm_dqrr_pvb_update(struct qm_portal *portal)
+{
+	struct qm_dqrr *dqrr = &portal->dqrr;
+	struct qm_dqrr_entry *res = qm_cl(dqrr->ring, dqrr->pi);
+
+	DPAA_ASSERT(dqrr->pmode == qm_dqrr_pvb);
+#ifndef CONFIG_FSL_PAMU
+	/*
+	 * If PAMU is not available we need to invalidate the cache.
+	 * When PAMU is available the cache is updated by stash
+	 */
+	dpaa_invalidate_touch_ro(res);
+#endif
+	/*
+	 *  when accessing 'verb', use __raw_readb() to ensure that compiler
+	 * inlining doesn't try to optimise out "excess reads".
+	 */
+	if ((__raw_readb(&res->verb) & QM_DQRR_VERB_VBIT) == dqrr->vbit) {
+		dqrr->pi = (dqrr->pi + 1) & (QM_DQRR_SIZE - 1);
+		if (!dqrr->pi)
+			dqrr->vbit ^= QM_DQRR_VERB_VBIT;
+		dqrr->fill++;
+	}
+}
+
+static inline void qm_dqrr_cdc_consume_1ptr(struct qm_portal *portal,
+					const struct qm_dqrr_entry *dq,
+					int park)
+{
+	__maybe_unused struct qm_dqrr *dqrr = &portal->dqrr;
+	int idx = dqrr_ptr2idx(dq);
+
+	DPAA_ASSERT(dqrr->cmode == qm_dqrr_cdc);
+	DPAA_ASSERT((dqrr->ring + idx) == dq);
+	DPAA_ASSERT(idx < QM_DQRR_SIZE);
+	qm_out(portal, QM_REG_DQRR_DCAP, (0 << 8) | /* DQRR_DCAP::S */
+	       ((park ? 1 : 0) << 6) |		    /* DQRR_DCAP::PK */
+	       idx);				    /* DQRR_DCAP::DCAP_CI */
+}
+
+static inline void qm_dqrr_cdc_consume_n(struct qm_portal *portal, u32 bitmask)
+{
+	__maybe_unused struct qm_dqrr *dqrr = &portal->dqrr;
+
+	DPAA_ASSERT(dqrr->cmode == qm_dqrr_cdc);
+	qm_out(portal, QM_REG_DQRR_DCAP, (1 << 8) | /* DQRR_DCAP::S */
+	       (bitmask << 16));		    /* DQRR_DCAP::DCAP_CI */
+}
+
+static inline void qm_dqrr_sdqcr_set(struct qm_portal *portal, u32 sdqcr)
+{
+	qm_out(portal, QM_REG_DQRR_SDQCR, sdqcr);
+}
+
+static inline void qm_dqrr_vdqcr_set(struct qm_portal *portal, u32 vdqcr)
+{
+	qm_out(portal, QM_REG_DQRR_VDQCR, vdqcr);
+}
+
+static inline void qm_dqrr_set_ithresh(struct qm_portal *portal, u8 ithresh)
+{
+	qm_out(portal, QM_REG_DQRR_ITR, ithresh);
+}
+
+/* --- MR API --- */
+
+#define MR_SHIFT	ilog2(sizeof(union qm_mr_entry))
+#define MR_CARRY	(uintptr_t)(QM_MR_SIZE << MR_SHIFT)
+
+static union qm_mr_entry *mr_carryclear(union qm_mr_entry *p)
+{
+	uintptr_t addr = (uintptr_t)p;
+
+	addr &= ~MR_CARRY;
+
+	return (union qm_mr_entry *)addr;
+}
+
+static inline int mr_ptr2idx(const union qm_mr_entry *e)
+{
+	return ((uintptr_t)e >> MR_SHIFT) & (QM_MR_SIZE - 1);
+}
+
+static inline union qm_mr_entry *mr_inc(union qm_mr_entry *e)
+{
+	return mr_carryclear(e + 1);
+}
+
+static inline int qm_mr_init(struct qm_portal *portal, enum qm_mr_pmode pmode,
+			     enum qm_mr_cmode cmode)
+{
+	struct qm_mr *mr = &portal->mr;
+	u32 cfg;
+
+	mr->ring = portal->addr.ce + QM_CL_MR;
+	mr->pi = qm_in(portal, QM_REG_MR_PI_CINH) & (QM_MR_SIZE - 1);
+	mr->ci = qm_in(portal, QM_REG_MR_CI_CINH) & (QM_MR_SIZE - 1);
+	mr->cursor = mr->ring + mr->ci;
+	mr->fill = dpaa_cyc_diff(QM_MR_SIZE, mr->ci, mr->pi);
+	mr->vbit = (qm_in(portal, QM_REG_MR_PI_CINH) & QM_MR_SIZE)
+		? QM_MR_VERB_VBIT : 0;
+	mr->ithresh = qm_in(portal, QM_REG_MR_ITR);
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	mr->pmode = pmode;
+	mr->cmode = cmode;
+#endif
+	cfg = (qm_in(portal, QM_REG_CFG) & 0xfffff0ff) |
+	      ((cmode & 1) << 8);	/* QCSP_CFG:MM */
+	qm_out(portal, QM_REG_CFG, cfg);
+	return 0;
+}
+
+static inline void qm_mr_finish(struct qm_portal *portal)
+{
+	struct qm_mr *mr = &portal->mr;
+
+	if (mr->ci != mr_ptr2idx(mr->cursor))
+		pr_crit("Ignoring completed MR entries\n");
+}
+
+static inline const union qm_mr_entry *qm_mr_current(struct qm_portal *portal)
+{
+	struct qm_mr *mr = &portal->mr;
+
+	if (!mr->fill)
+		return NULL;
+	return mr->cursor;
+}
+
+static inline int qm_mr_next(struct qm_portal *portal)
+{
+	struct qm_mr *mr = &portal->mr;
+
+	DPAA_ASSERT(mr->fill);
+	mr->cursor = mr_inc(mr->cursor);
+	return --mr->fill;
+}
+
+static inline void qm_mr_pvb_update(struct qm_portal *portal)
+{
+	struct qm_mr *mr = &portal->mr;
+	union qm_mr_entry *res = qm_cl(mr->ring, mr->pi);
+
+	DPAA_ASSERT(mr->pmode == qm_mr_pvb);
+	/*
+	 *  when accessing 'verb', use __raw_readb() to ensure that compiler
+	 * inlining doesn't try to optimise out "excess reads".
+	 */
+	if ((__raw_readb(&res->verb) & QM_MR_VERB_VBIT) == mr->vbit) {
+		mr->pi = (mr->pi + 1) & (QM_MR_SIZE - 1);
+		if (!mr->pi)
+			mr->vbit ^= QM_MR_VERB_VBIT;
+		mr->fill++;
+		res = mr_inc(res);
+	}
+	dpaa_invalidate_touch_ro(res);
+}
+
+static inline void qm_mr_cci_consume(struct qm_portal *portal, u8 num)
+{
+	struct qm_mr *mr = &portal->mr;
+
+	DPAA_ASSERT(mr->cmode == qm_mr_cci);
+	mr->ci = (mr->ci + num) & (QM_MR_SIZE - 1);
+	qm_out(portal, QM_REG_MR_CI_CINH, mr->ci);
+}
+
+static inline void qm_mr_cci_consume_to_current(struct qm_portal *portal)
+{
+	struct qm_mr *mr = &portal->mr;
+
+	DPAA_ASSERT(mr->cmode == qm_mr_cci);
+	mr->ci = mr_ptr2idx(mr->cursor);
+	qm_out(portal, QM_REG_MR_CI_CINH, mr->ci);
+}
+
+static inline void qm_mr_set_ithresh(struct qm_portal *portal, u8 ithresh)
+{
+	qm_out(portal, QM_REG_MR_ITR, ithresh);
+}
+
+/* --- Management command API --- */
+
+static inline int qm_mc_init(struct qm_portal *portal)
+{
+	struct qm_mc *mc = &portal->mc;
+
+	mc->cr = portal->addr.ce + QM_CL_CR;
+	mc->rr = portal->addr.ce + QM_CL_RR0;
+	mc->rridx = (__raw_readb(&mc->cr->_ncw_verb) & QM_MCC_VERB_VBIT)
+		    ? 0 : 1;
+	mc->vbit = mc->rridx ? QM_MCC_VERB_VBIT : 0;
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	mc->state = qman_mc_idle;
+#endif
+	return 0;
+}
+
+static inline void qm_mc_finish(struct qm_portal *portal)
+{
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	struct qm_mc *mc = &portal->mc;
+
+	DPAA_ASSERT(mc->state == qman_mc_idle);
+	if (mc->state != qman_mc_idle)
+		pr_crit("Losing incomplete MC command\n");
+#endif
+}
+
+static inline union qm_mc_command *qm_mc_start(struct qm_portal *portal)
+{
+	struct qm_mc *mc = &portal->mc;
+
+	DPAA_ASSERT(mc->state == qman_mc_idle);
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	mc->state = qman_mc_user;
+#endif
+	dpaa_zero(mc->cr);
+	return mc->cr;
+}
+
+static inline void qm_mc_commit(struct qm_portal *portal, u8 myverb)
+{
+	struct qm_mc *mc = &portal->mc;
+	union qm_mc_result *rr = mc->rr + mc->rridx;
+
+	DPAA_ASSERT(mc->state == qman_mc_user);
+	dma_wmb();
+	mc->cr->_ncw_verb = myverb | mc->vbit;
+	dpaa_flush(mc->cr);
+	dpaa_invalidate_touch_ro(rr);
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	mc->state = qman_mc_hw;
+#endif
+}
+
+static inline union qm_mc_result *qm_mc_result(struct qm_portal *portal)
+{
+	struct qm_mc *mc = &portal->mc;
+	union qm_mc_result *rr = mc->rr + mc->rridx;
+
+	DPAA_ASSERT(mc->state == qman_mc_hw);
+	/*
+	 *  The inactive response register's verb byte always returns zero until
+	 * its command is submitted and completed. This includes the valid-bit,
+	 * in case you were wondering...
+	 */
+	if (!__raw_readb(&rr->verb)) {
+		dpaa_invalidate_touch_ro(rr);
+		return NULL;
+	}
+	mc->rridx ^= 1;
+	mc->vbit ^= QM_MCC_VERB_VBIT;
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	mc->state = qman_mc_idle;
+#endif
+	return rr;
+}
+
+static inline int qm_mc_result_timeout(struct qm_portal *portal,
+				       union qm_mc_result **mcr)
+{
+	int timeout = QM_MCR_TIMEOUT;
+
+	do {
+		*mcr = qm_mc_result(portal);
+		if (*mcr)
+			break;
+		udelay(1);
+	} while (--timeout);
+
+	return timeout;
+}
+
+static inline void fq_set(struct qman_fq *fq, u32 mask)
+{
+	set_bits(mask, &fq->flags);
+}
+
+static inline void fq_clear(struct qman_fq *fq, u32 mask)
+{
+	clear_bits(mask, &fq->flags);
+}
+
+static inline int fq_isset(struct qman_fq *fq, u32 mask)
+{
+	return fq->flags & mask;
+}
+
+static inline int fq_isclear(struct qman_fq *fq, u32 mask)
+{
+	return !(fq->flags & mask);
+}
+
+struct qman_portal {
+	struct qm_portal p;
+	/* PORTAL_BITS_*** - dynamic, strictly internal */
+	unsigned long bits;
+	/* interrupt sources processed by portal_isr(), configurable */
+	unsigned long irq_sources;
+	u32 use_eqcr_ci_stashing;
+	/* only 1 volatile dequeue at a time */
+	struct qman_fq *vdqcr_owned;
+	u32 sdqcr;
+	/* probing time config params for cpu-affine portals */
+	const struct qm_portal_config *config;
+	/* needed for providing a non-NULL device to dma_map_***() */
+	struct platform_device *pdev;
+	/* 2-element array. cgrs[0] is mask, cgrs[1] is snapshot. */
+	struct qman_cgrs *cgrs;
+	/* linked-list of CSCN handlers. */
+	struct list_head cgr_cbs;
+	/* list lock */
+	spinlock_t cgr_lock;
+	struct work_struct congestion_work;
+	struct work_struct mr_work;
+	char irqname[MAX_IRQNAME];
+};
+
+static cpumask_t affine_mask;
+static DEFINE_SPINLOCK(affine_mask_lock);
+static u16 affine_channels[NR_CPUS];
+static DEFINE_PER_CPU(struct qman_portal, qman_affine_portal);
+struct qman_portal *affine_portals[NR_CPUS];
+
+static inline struct qman_portal *get_affine_portal(void)
+{
+	return &get_cpu_var(qman_affine_portal);
+}
+
+static inline void put_affine_portal(void)
+{
+	put_cpu_var(qman_affine_portal);
+}
+
+static struct workqueue_struct *qm_portal_wq;
+
+int qman_wq_alloc(void)
+{
+	qm_portal_wq = alloc_workqueue("qman_portal_wq", 0, 1);
+	if (!qm_portal_wq)
+		return -ENOMEM;
+	return 0;
+}
+
+/*
+ * This is what everything can wait on, even if it migrates to a different cpu
+ * to the one whose affine portal it is waiting on.
+ */
+static DECLARE_WAIT_QUEUE_HEAD(affine_queue);
+
+static struct qman_fq **fq_table;
+static u32 num_fqids;
+
+int qman_alloc_fq_table(u32 _num_fqids)
+{
+	num_fqids = _num_fqids;
+
+	fq_table = vzalloc(num_fqids * 2 * sizeof(struct qman_fq *));
+	if (!fq_table)
+		return -ENOMEM;
+
+	pr_debug("Allocated fq lookup table at %p, entry count %u\n",
+		 fq_table, num_fqids * 2);
+	return 0;
+}
+
+static struct qman_fq *idx_to_fq(u32 idx)
+{
+	struct qman_fq *fq;
+
+#ifdef CONFIG_FSL_DPAA_CHECKING
+	if (WARN_ON(idx >= num_fqids * 2))
+		return NULL;
+#endif
+	fq = fq_table[idx];
+	DPAA_ASSERT(!fq || idx == fq->idx);
+
+	return fq;
+}
+
+/*
+ * Only returns full-service fq objects, not enqueue-only
+ * references (QMAN_FQ_FLAG_NO_MODIFY).
+ */
+static struct qman_fq *fqid_to_fq(u32 fqid)
+{
+	return idx_to_fq(fqid * 2);
+}
+
+static struct qman_fq *tag_to_fq(u32 tag)
+{
+#if BITS_PER_LONG == 64
+	return idx_to_fq(tag);
+#else
+	return (struct qman_fq *)tag;
+#endif
+}
+
+static u32 fq_to_tag(struct qman_fq *fq)
+{
+#if BITS_PER_LONG == 64
+	return fq->idx;
+#else
+	return (u32)fq;
+#endif
+}
+
+static u32 __poll_portal_slow(struct qman_portal *p, u32 is);
+static inline unsigned int __poll_portal_fast(struct qman_portal *p,
+					unsigned int poll_limit);
+static void qm_congestion_task(struct work_struct *work);
+static void qm_mr_process_task(struct work_struct *work);
+
+static irqreturn_t portal_isr(int irq, void *ptr)
+{
+	struct qman_portal *p = ptr;
+
+	u32 clear = QM_DQAVAIL_MASK | p->irq_sources;
+	u32 is = qm_in(&p->p, QM_REG_ISR) & p->irq_sources;
+
+	if (unlikely(!is))
+		return IRQ_NONE;
+
+	/* DQRR-handling if it's interrupt-driven */
+	if (is &