From 531b17a780dc5f486e8be8d82d1ccd077ea25f14 Mon Sep 17 00:00:00 2001 From: Hemant Agrawal Date: Tue, 11 Apr 2017 19:07:09 +0530 Subject: [PATCH] bus/fslmc: add QBMAN driver to bus QBMAN, is a hardware block which interfaces with the other accelerating hardware blocks (For e.g., WRIOP) on NXP's DPAA2 SoC for queue, buffer and packet scheduling. This patch introduces a userspace driver for interfacing with the QBMAN hw block. The qbman-portal component provides APIs to do the low level hardware bit twiddling for operations such as: -initializing Qman software portals -building and sending portal commands -portal interrupt configuration and processing This same/similar code is used in kernel and compat file is used to make it working in user space. Signed-off-by: Geoff Thorpe Signed-off-by: Roy Pledge Signed-off-by: Hemant Agrawal --- drivers/bus/fslmc/Makefile | 4 + drivers/bus/fslmc/qbman/include/compat.h | 410 +++++ .../bus/fslmc/qbman/include/fsl_qbman_base.h | 160 ++ .../fslmc/qbman/include/fsl_qbman_portal.h | 1093 ++++++++++++ drivers/bus/fslmc/qbman/qbman_portal.c | 1496 +++++++++++++++++ drivers/bus/fslmc/qbman/qbman_portal.h | 277 +++ drivers/bus/fslmc/qbman/qbman_private.h | 174 ++ drivers/bus/fslmc/qbman/qbman_sys.h | 385 +++++ drivers/bus/fslmc/qbman/qbman_sys_decl.h | 73 + drivers/bus/fslmc/rte_bus_fslmc_version.map | 19 + 10 files changed, 4091 insertions(+) create mode 100644 drivers/bus/fslmc/qbman/include/compat.h create mode 100644 drivers/bus/fslmc/qbman/include/fsl_qbman_base.h create mode 100644 drivers/bus/fslmc/qbman/include/fsl_qbman_portal.h create mode 100644 drivers/bus/fslmc/qbman/qbman_portal.c create mode 100644 drivers/bus/fslmc/qbman/qbman_portal.h create mode 100644 drivers/bus/fslmc/qbman/qbman_private.h create mode 100644 drivers/bus/fslmc/qbman/qbman_sys.h create mode 100644 drivers/bus/fslmc/qbman/qbman_sys_decl.h diff --git a/drivers/bus/fslmc/Makefile b/drivers/bus/fslmc/Makefile index 5561bb81c6..259376123d 100644 --- a/drivers/bus/fslmc/Makefile +++ b/drivers/bus/fslmc/Makefile @@ -49,6 +49,7 @@ CFLAGS += $(WERROR_FLAGS) endif CFLAGS += -I$(RTE_SDK)/drivers/bus/fslmc +CFLAGS += -I$(RTE_SDK)/drivers/bus/fslmc/qbman/include CFLAGS += -I$(RTE_SDK)/lib/librte_eal/linuxapp/eal # versioning export map @@ -57,6 +58,9 @@ EXPORT_MAP := rte_bus_fslmc_version.map # library version LIBABIVER := 1 +SRCS-$(CONFIG_RTE_LIBRTE_FSLMC_BUS) += \ + qbman/qbman_portal.c + SRCS-$(CONFIG_RTE_LIBRTE_FSLMC_BUS) += fslmc_bus.c include $(RTE_SDK)/mk/rte.lib.mk diff --git a/drivers/bus/fslmc/qbman/include/compat.h b/drivers/bus/fslmc/qbman/include/compat.h new file mode 100644 index 0000000000..41effe37e2 --- /dev/null +++ b/drivers/bus/fslmc/qbman/include/compat.h @@ -0,0 +1,410 @@ +/*- + * BSD LICENSE + * + * Copyright (c) 2008-2016 Freescale Semiconductor, Inc. + * 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 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. + * + * 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. + */ + +#ifndef HEADER_COMPAT_H +#define HEADER_COMPAT_H + +#include + +#ifndef _GNU_SOURCE +#define _GNU_SOURCE +#endif +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* The following definitions are primarily to allow the single-source driver + * interfaces to be included by arbitrary program code. Ie. for interfaces that + * are also available in kernel-space, these definitions provide compatibility + * with certain attributes and types used in those interfaces. + */ + +/* Required compiler attributes */ +#define __user +#define likely(x) __builtin_expect(!!(x), 1) +#define unlikely(x) __builtin_expect(!!(x), 0) +#define ____cacheline_aligned __attribute__((aligned(L1_CACHE_BYTES))) +#undef container_of +#define container_of(ptr, type, member) ({ \ + typeof(((type *)0)->member)(*__mptr) = (ptr); \ + (type *)((char *)__mptr - offsetof(type, member)); }) +#define __stringify_1(x) #x +#define __stringify(x) __stringify_1(x) + +#ifdef ARRAY_SIZE +#undef ARRAY_SIZE +#endif +#define ARRAY_SIZE(a) (sizeof(a) / sizeof((a)[0])) + +/* Required types */ +typedef uint8_t u8; +typedef uint16_t u16; +typedef uint32_t u32; +typedef uint64_t u64; +typedef uint64_t dma_addr_t; +typedef cpu_set_t cpumask_t; +typedef u32 compat_uptr_t; + +static inline void __user *compat_ptr(compat_uptr_t uptr) +{ + return (void __user *)(unsigned long)uptr; +} + +static inline compat_uptr_t ptr_to_compat(void __user *uptr) +{ + return (u32)(unsigned long)uptr; +} + +/* I/O operations */ +static inline u32 in_be32(volatile void *__p) +{ + volatile u32 *p = __p; + return *p; +} + +static inline void out_be32(volatile void *__p, u32 val) +{ + volatile u32 *p = __p; + *p = val; +} + +/* Debugging */ +#define prflush(fmt, args...) \ + do { \ + printf(fmt, ##args); \ + fflush(stdout); \ + } while (0) +#define pr_crit(fmt, args...) prflush("CRIT:" fmt, ##args) +#define pr_err(fmt, args...) prflush("ERR:" fmt, ##args) +#define pr_warn(fmt, args...) prflush("WARN:" fmt, ##args) +#define pr_info(fmt, args...) prflush(fmt, ##args) + +#ifdef pr_debug +#undef pr_debug +#endif +#define pr_debug(fmt, args...) {} +#define might_sleep_if(c) {} +#define msleep(x) {} +#define WARN_ON(c, str) \ +do { \ + static int warned_##__LINE__; \ + if ((c) && !warned_##__LINE__) { \ + pr_warn("%s\n", str); \ + pr_warn("(%s:%d)\n", __FILE__, __LINE__); \ + warned_##__LINE__ = 1; \ + } \ +} while (0) +#ifdef CONFIG_BUGON +#define QBMAN_BUG_ON(c) WARN_ON(c, "BUG") +#else +#define QBMAN_BUG_ON(c) {} +#endif + +#define ALIGN(x, a) (((x) + ((typeof(x))(a) - 1)) & ~((typeof(x))(a) - 1)) + +/****************/ +/* Linked-lists */ +/****************/ + +struct list_head { + struct list_head *prev; + struct list_head *next; +}; + +#define LIST_HEAD(n) \ +struct list_head n = { \ + .prev = &n, \ + .next = &n \ +} + +#define INIT_LIST_HEAD(p) \ +do { \ + struct list_head *__p298 = (p); \ + __p298->next = __p298; \ + __p298->prev = __p298->next; \ +} while (0) +#define list_entry(node, type, member) \ + (type *)((void *)node - offsetof(type, member)) +#define list_empty(p) \ +({ \ + const struct list_head *__p298 = (p); \ + ((__p298->next == __p298) && (__p298->prev == __p298)); \ +}) +#define list_add(p, l) \ +do { \ + struct list_head *__p298 = (p); \ + struct list_head *__l298 = (l); \ + __p298->next = __l298->next; \ + __p298->prev = __l298; \ + __l298->next->prev = __p298; \ + __l298->next = __p298; \ +} while (0) +#define list_add_tail(p, l) \ +do { \ + struct list_head *__p298 = (p); \ + struct list_head *__l298 = (l); \ + __p298->prev = __l298->prev; \ + __p298->next = __l298; \ + __l298->prev->next = __p298; \ + __l298->prev = __p298; \ +} while (0) +#define list_for_each(i, l) \ + for (i = (l)->next; i != (l); i = i->next) +#define list_for_each_safe(i, j, l) \ + for (i = (l)->next, j = i->next; i != (l); \ + i = j, j = i->next) +#define list_for_each_entry(i, l, name) \ + for (i = list_entry((l)->next, typeof(*i), name); &i->name != (l); \ + i = list_entry(i->name.next, typeof(*i), name)) +#define list_for_each_entry_safe(i, j, l, name) \ + for (i = list_entry((l)->next, typeof(*i), name), \ + j = list_entry(i->name.next, typeof(*j), name); \ + &i->name != (l); \ + i = j, j = list_entry(j->name.next, typeof(*j), name)) +#define list_del(i) \ +do { \ + (i)->next->prev = (i)->prev; \ + (i)->prev->next = (i)->next; \ +} while (0) + +/* Other miscellaneous interfaces our APIs depend on; */ + +#define lower_32_bits(x) ((u32)(x)) +#define upper_32_bits(x) ((u32)(((x) >> 16) >> 16)) + +/* Compiler/type stuff */ +typedef unsigned int gfp_t; +typedef uint32_t phandle; + +#define __iomem +#define EINTR 4 +#define ENODEV 19 +#define GFP_KERNEL 0 +#define __raw_readb(p) (*(const volatile unsigned char *)(p)) +#define __raw_readl(p) (*(const volatile unsigned int *)(p)) +#define __raw_writel(v, p) {*(volatile unsigned int *)(p) = (v); } + +/* memcpy() stuff - when you know alignments in advance */ +#ifdef CONFIG_TRY_BETTER_MEMCPY +static inline void copy_words(void *dest, const void *src, size_t sz) +{ + u32 *__dest = dest; + const u32 *__src = src; + size_t __sz = sz >> 2; + + QBMAN_BUG_ON((unsigned long)dest & 0x3); + QBMAN_BUG_ON((unsigned long)src & 0x3); + QBMAN_BUG_ON(sz & 0x3); + while (__sz--) + *(__dest++) = *(__src++); +} + +static inline void copy_shorts(void *dest, const void *src, size_t sz) +{ + u16 *__dest = dest; + const u16 *__src = src; + size_t __sz = sz >> 1; + + QBMAN_BUG_ON((unsigned long)dest & 0x1); + QBMAN_BUG_ON((unsigned long)src & 0x1); + QBMAN_BUG_ON(sz & 0x1); + while (__sz--) + *(__dest++) = *(__src++); +} + +static inline void copy_bytes(void *dest, const void *src, size_t sz) +{ + u8 *__dest = dest; + const u8 *__src = src; + + while (sz--) + *(__dest++) = *(__src++); +} +#else +#define copy_words memcpy +#define copy_shorts memcpy +#define copy_bytes memcpy +#endif + +/* Completion stuff */ +#define DECLARE_COMPLETION(n) int n = 0 +#define complete(n) { *n = 1; } +#define wait_for_completion(n) \ +do { \ + while (!*n) { \ + bman_poll(); \ + qman_poll(); \ + } \ + *n = 0; \ +} while (0) + +/* Allocator stuff */ +#define kmalloc(sz, t) malloc(sz) +#define vmalloc(sz) malloc(sz) +#define kfree(p) { if (p) free(p); } +static inline void *kzalloc(size_t sz, gfp_t __foo __rte_unused) +{ + void *ptr = malloc(sz); + + if (ptr) + memset(ptr, 0, sz); + return ptr; +} + +static inline unsigned long get_zeroed_page(gfp_t __foo __rte_unused) +{ + void *p; + + if (posix_memalign(&p, 4096, 4096)) + return 0; + memset(p, 0, 4096); + return (unsigned long)p; +} + +static inline void free_page(unsigned long p) +{ + free((void *)p); +} + +/* Bitfield stuff. */ +#define BITS_PER_ULONG (sizeof(unsigned long) << 3) +#define SHIFT_PER_ULONG (((1 << 5) == BITS_PER_ULONG) ? 5 : 6) +#define BITS_MASK(idx) ((unsigned long)1 << ((idx) & (BITS_PER_ULONG - 1))) +#define BITS_IDX(idx) ((idx) >> SHIFT_PER_ULONG) +static inline unsigned long test_bits(unsigned long mask, + volatile unsigned long *p) +{ + return *p & mask; +} + +static inline int test_bit(int idx, volatile unsigned long *bits) +{ + return test_bits(BITS_MASK(idx), bits + BITS_IDX(idx)); +} + +static inline void set_bits(unsigned long mask, volatile unsigned long *p) +{ + *p |= mask; +} + +static inline void set_bit(int idx, volatile unsigned long *bits) +{ + set_bits(BITS_MASK(idx), bits + BITS_IDX(idx)); +} + +static inline void clear_bits(unsigned long mask, volatile unsigned long *p) +{ + *p &= ~mask; +} + +static inline void clear_bit(int idx, volatile unsigned long *bits) +{ + clear_bits(BITS_MASK(idx), bits + BITS_IDX(idx)); +} + +static inline unsigned long test_and_set_bits(unsigned long mask, + volatile unsigned long *p) +{ + unsigned long ret = test_bits(mask, p); + + set_bits(mask, p); + return ret; +} + +static inline int test_and_set_bit(int idx, volatile unsigned long *bits) +{ + int ret = test_bit(idx, bits); + + set_bit(idx, bits); + return ret; +} + +static inline int test_and_clear_bit(int idx, volatile unsigned long *bits) +{ + int ret = test_bit(idx, bits); + + clear_bit(idx, bits); + return ret; +} + +static inline int find_next_zero_bit(unsigned long *bits, int limit, int idx) +{ + while ((++idx < limit) && test_bit(idx, bits)) + ; + return idx; +} + +static inline int find_first_zero_bit(unsigned long *bits, int limit) +{ + int idx = 0; + + while (test_bit(idx, bits) && (++idx < limit)) + ; + return idx; +} + +static inline u64 div64_u64(u64 n, u64 d) +{ + return n / d; +} + +#define atomic_t rte_atomic32_t +#define atomic_read(v) rte_atomic32_read(v) +#define atomic_set(v, i) rte_atomic32_set(v, i) + +#define atomic_inc(v) rte_atomic32_add(v, 1) +#define atomic_dec(v) rte_atomic32_sub(v, 1) + +#define atomic_inc_and_test(v) rte_atomic32_inc_and_test(v) +#define atomic_dec_and_test(v) rte_atomic32_dec_and_test(v) + +#define atomic_inc_return(v) rte_atomic32_add_return(v, 1) +#define atomic_dec_return(v) rte_atomic32_sub_return(v, 1) +#define atomic_sub_and_test(i, v) (rte_atomic32_sub_return(v, i) == 0) + +#endif /* HEADER_COMPAT_H */ diff --git a/drivers/bus/fslmc/qbman/include/fsl_qbman_base.h b/drivers/bus/fslmc/qbman/include/fsl_qbman_base.h new file mode 100644 index 0000000000..ee4b772c02 --- /dev/null +++ b/drivers/bus/fslmc/qbman/include/fsl_qbman_base.h @@ -0,0 +1,160 @@ +/*- + * BSD LICENSE + * + * Copyright (C) 2014 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. + * + * 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. + */ +#ifndef _FSL_QBMAN_BASE_H +#define _FSL_QBMAN_BASE_H + +typedef uint64_t dma_addr_t; + +/** + * DOC: QBMan basic structures + * + * The QBMan block descriptor, software portal descriptor and Frame descriptor + * are defined here. + * + */ + +#define QMAN_REV_4000 0x04000000 +#define QMAN_REV_4100 0x04010000 +#define QMAN_REV_4101 0x04010001 + +/** + * struct qbman_block_desc - qbman block descriptor structure + * @ccsr_reg_bar: CCSR register map. + * @irq_rerr: Recoverable error interrupt line. + * @irq_nrerr: Non-recoverable error interrupt line + * + * Descriptor for a QBMan instance on the SoC. On partitions/targets that do not + * control this QBMan instance, these values may simply be place-holders. The + * idea is simply that we be able to distinguish between them, eg. so that SWP + * descriptors can identify which QBMan instance they belong to. + */ +struct qbman_block_desc { + void *ccsr_reg_bar; + int irq_rerr; + int irq_nrerr; +}; + +enum qbman_eqcr_mode { + qman_eqcr_vb_ring = 2, /* Valid bit, with eqcr in ring mode */ + qman_eqcr_vb_array, /* Valid bit, with eqcr in array mode */ +}; + +/** + * struct qbman_swp_desc - qbman software portal descriptor structure + * @block: The QBMan instance. + * @cena_bar: Cache-enabled portal register map. + * @cinh_bar: Cache-inhibited portal register map. + * @irq: -1 if unused (or unassigned) + * @idx: SWPs within a QBMan are indexed. -1 if opaque to the user. + * @qman_version: the qman version. + * @eqcr_mode: Select the eqcr mode, currently only valid bit ring mode and + * valid bit array mode are supported. + * + * Descriptor for a QBMan software portal, expressed in terms that make sense to + * the user context. Ie. on MC, this information is likely to be true-physical, + * and instantiated statically at compile-time. On GPP, this information is + * likely to be obtained via "discovery" over a partition's "MC bus" + * (ie. in response to a MC portal command), and would take into account any + * virtualisation of the GPP user's address space and/or interrupt numbering. + */ +struct qbman_swp_desc { + const struct qbman_block_desc *block; + uint8_t *cena_bar; + uint8_t *cinh_bar; + int irq; + int idx; + uint32_t qman_version; + enum qbman_eqcr_mode eqcr_mode; +}; + +/* Driver object for managing a QBMan portal */ +struct qbman_swp; + +/** + * struct qbman_fd - basci structure for qbman frame descriptor + * @words: for easier/faster copying the whole FD structure. + * @addr_lo: the lower 32 bits of the address in FD. + * @addr_hi: the upper 32 bits of the address in FD. + * @len: the length field in FD. + * @bpid_offset: represent the bpid and offset fields in FD. offset in + * the MS 16 bits, BPID in the LS 16 bits. + * @frc: frame context + * @ctrl: the 32bit control bits including dd, sc,... va, err. + * @flc_lo: the lower 32bit of flow context. + * @flc_hi: the upper 32bits of flow context. + * + * Place-holder for FDs, we represent it via the simplest form that we need for + * now. Different overlays may be needed to support different options, etc. (It + * is impractical to define One True Struct, because the resulting encoding + * routines (lots of read-modify-writes) would be worst-case performance whether + * or not circumstances required them.) + * + * Note, as with all data-structures exchanged between software and hardware (be + * they located in the portal register map or DMA'd to and from main-memory), + * the driver ensures that the caller of the driver API sees the data-structures + * in host-endianness. "struct qbman_fd" is no exception. The 32-bit words + * contained within this structure are represented in host-endianness, even if + * hardware always treats them as little-endian. As such, if any of these fields + * are interpreted in a binary (rather than numerical) fashion by hardware + * blocks (eg. accelerators), then the user should be careful. We illustrate + * with an example; + * + * Suppose the desired behaviour of an accelerator is controlled by the "frc" + * field of the FDs that are sent to it. Suppose also that the behaviour desired + * by the user corresponds to an "frc" value which is expressed as the literal + * sequence of bytes 0xfe, 0xed, 0xab, and 0xba. So "frc" should be the 32-bit + * value in which 0xfe is the first byte and 0xba is the last byte, and as + * hardware is little-endian, this amounts to a 32-bit "value" of 0xbaabedfe. If + * the software is little-endian also, this can simply be achieved by setting + * frc=0xbaabedfe. On the other hand, if software is big-endian, it should set + * frc=0xfeedabba! The best away of avoiding trouble with this sort of thing is + * to treat the 32-bit words as numerical values, in which the offset of a field + * from the beginning of the first byte (as required or generated by hardware) + * is numerically encoded by a left-shift (ie. by raising the field to a + * corresponding power of 2). Ie. in the current example, software could set + * "frc" in the following way, and it would work correctly on both little-endian + * and big-endian operation; + * fd.frc = (0xfe << 0) | (0xed << 8) | (0xab << 16) | (0xba << 24); + */ +struct qbman_fd { + union { + uint32_t words[8]; + struct qbman_fd_simple { + uint32_t addr_lo; + uint32_t addr_hi; + uint32_t len; + uint32_t bpid_offset; + uint32_t frc; + uint32_t ctrl; + uint32_t flc_lo; + uint32_t flc_hi; + } simple; + }; +}; + +#endif /* !_FSL_QBMAN_BASE_H */ diff --git a/drivers/bus/fslmc/qbman/include/fsl_qbman_portal.h b/drivers/bus/fslmc/qbman/include/fsl_qbman_portal.h new file mode 100644 index 0000000000..773177232e --- /dev/null +++ b/drivers/bus/fslmc/qbman/include/fsl_qbman_portal.h @@ -0,0 +1,1093 @@ +/*- + * BSD LICENSE + * + * Copyright (C) 2014 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. + * + * 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. + */ +#ifndef _FSL_QBMAN_PORTAL_H +#define _FSL_QBMAN_PORTAL_H + +#include + +/** + * DOC - QBMan portal APIs to implement the following functions: + * - Initialize and destroy Software portal object. + * - Read and write Software portal interrupt registers. + * - Enqueue, including setting the enqueue descriptor, and issuing enqueue + * command etc. + * - Dequeue, including setting the dequeue descriptor, issuing dequeue command, + * parsing the dequeue response in DQRR and memeory, parsing the state change + * notifications etc. + * - Release, including setting the release descriptor, and issuing the buffer + * release command. + * - Acquire, acquire the buffer from the given buffer pool. + * - FQ management. + * - Channel management, enable/disable CDAN with or without context. + */ + +/** + * qbman_swp_init() - Create a functional object representing the given + * QBMan portal descriptor. + * @d: the given qbman swp descriptor + * + * Return qbman_swp portal object for success, NULL if the object cannot + * be created. + */ +struct qbman_swp *qbman_swp_init(const struct qbman_swp_desc *d); + +/** + * qbman_swp_finish() - Create and destroy a functional object representing + * the given QBMan portal descriptor. + * @p: the qbman_swp object to be destroyed. + * + */ +void qbman_swp_finish(struct qbman_swp *p); + +/** + * qbman_swp_get_desc() - Get the descriptor of the given portal object. + * @p: the given portal object. + * + * Return the descriptor for this portal. + */ +const struct qbman_swp_desc *qbman_swp_get_desc(struct qbman_swp *p); + + /**************/ + /* Interrupts */ + /**************/ + +/* EQCR ring interrupt */ +#define QBMAN_SWP_INTERRUPT_EQRI ((uint32_t)0x00000001) +/* Enqueue command dispatched interrupt */ +#define QBMAN_SWP_INTERRUPT_EQDI ((uint32_t)0x00000002) +/* DQRR non-empty interrupt */ +#define QBMAN_SWP_INTERRUPT_DQRI ((uint32_t)0x00000004) +/* RCR ring interrupt */ +#define QBMAN_SWP_INTERRUPT_RCRI ((uint32_t)0x00000008) +/* Release command dispatched interrupt */ +#define QBMAN_SWP_INTERRUPT_RCDI ((uint32_t)0x00000010) +/* Volatile dequeue command interrupt */ +#define QBMAN_SWP_INTERRUPT_VDCI ((uint32_t)0x00000020) + +/** + * qbman_swp_interrupt_get_vanish() - Get the data in software portal + * interrupt status disable register. + * @p: the given software portal object. + * + * Return the settings in SWP_ISDR register. + */ +uint32_t qbman_swp_interrupt_get_vanish(struct qbman_swp *p); + +/** + * qbman_swp_interrupt_set_vanish() - Set the data in software portal + * interrupt status disable register. + * @p: the given software portal object. + * @mask: The value to set in SWP_IDSR register. + */ +void qbman_swp_interrupt_set_vanish(struct qbman_swp *p, uint32_t mask); + +/** + * qbman_swp_interrupt_read_status() - Get the data in software portal + * interrupt status register. + * @p: the given software portal object. + * + * Return the settings in SWP_ISR register. + */ +uint32_t qbman_swp_interrupt_read_status(struct qbman_swp *p); + +/** + * qbman_swp_interrupt_clear_status() - Set the data in software portal + * interrupt status register. + * @p: the given software portal object. + * @mask: The value to set in SWP_ISR register. + */ +void qbman_swp_interrupt_clear_status(struct qbman_swp *p, uint32_t mask); + +/** + * qbman_swp_interrupt_get_trigger() - Get the data in software portal + * interrupt enable register. + * @p: the given software portal object. + * + * Return the settings in SWP_IER register. + */ +uint32_t qbman_swp_interrupt_get_trigger(struct qbman_swp *p); + +/** + * qbman_swp_interrupt_set_trigger() - Set the data in software portal + * interrupt enable register. + * @p: the given software portal object. + * @mask: The value to set in SWP_IER register. + */ +void qbman_swp_interrupt_set_trigger(struct qbman_swp *p, uint32_t mask); + +/** + * qbman_swp_interrupt_get_inhibit() - Get the data in software portal + * interrupt inhibit register. + * @p: the given software portal object. + * + * Return the settings in SWP_IIR register. + */ +int qbman_swp_interrupt_get_inhibit(struct qbman_swp *p); + +/** + * qbman_swp_interrupt_set_inhibit() - Set the data in software portal + * interrupt inhibit register. + * @p: the given software portal object. + * @mask: The value to set in SWP_IIR register. + */ +void qbman_swp_interrupt_set_inhibit(struct qbman_swp *p, int inhibit); + + /************/ + /* Dequeues */ + /************/ + +/** + * struct qbman_result - structure for qbman dequeue response and/or + * notification. + * @dont_manipulate_directly: the 16 32bit data to represent the whole + * possible qbman dequeue result. + */ +struct qbman_result { + uint32_t dont_manipulate_directly[16]; +}; + +/* TODO: + *A DQRI interrupt can be generated when there are dequeue results on the + * portal's DQRR (this mechanism does not deal with "pull" dequeues to + * user-supplied 'storage' addresses). There are two parameters to this + * interrupt source, one is a threshold and the other is a timeout. The + * interrupt will fire if either the fill-level of the ring exceeds 'thresh', or + * if the ring has been non-empty for been longer than 'timeout' nanoseconds. + * For timeout, an approximation to the desired nanosecond-granularity value is + * made, so there are get and set APIs to allow the user to see what actual + * timeout is set (compared to the timeout that was requested). + */ +int qbman_swp_dequeue_thresh(struct qbman_swp *s, unsigned int thresh); +int qbman_swp_dequeue_set_timeout(struct qbman_swp *s, unsigned int timeout); +int qbman_swp_dequeue_get_timeout(struct qbman_swp *s, unsigned int *timeout); + +/* ------------------- */ +/* Push-mode dequeuing */ +/* ------------------- */ + +/* The user of a portal can enable and disable push-mode dequeuing of up to 16 + * channels independently. It does not specify this toggling by channel IDs, but + * rather by specifying the index (from 0 to 15) that has been mapped to the + * desired channel. + */ + +/** + * qbman_swp_push_get() - Get the push dequeue setup. + * @s: the software portal object. + * @channel_idx: the channel index to query. + * @enabled: returned boolean to show whether the push dequeue is enabled for + * the given channel. + */ +void qbman_swp_push_get(struct qbman_swp *s, uint8_t channel_idx, int *enabled); + +/** + * qbman_swp_push_set() - Enable or disable push dequeue. + * @s: the software portal object. + * @channel_idx: the channel index.. + * @enable: enable or disable push dequeue. + * + * The user of a portal can enable and disable push-mode dequeuing of up to 16 + * channels independently. It does not specify this toggling by channel IDs, but + * rather by specifying the index (from 0 to 15) that has been mapped to the + * desired channel. + */ +void qbman_swp_push_set(struct qbman_swp *s, uint8_t channel_idx, int enable); + +/* ------------------- */ +/* Pull-mode dequeuing */ +/* ------------------- */ + +/** + * struct qbman_pull_desc - the structure for pull dequeue descriptor + * @dont_manipulate_directly: the 6 32bit data to represent the whole + * possible settings for pull dequeue descriptor. + */ +struct qbman_pull_desc { + uint32_t dont_manipulate_directly[6]; +}; + +enum qbman_pull_type_e { + /* dequeue with priority precedence, respect intra-class scheduling */ + qbman_pull_type_prio = 1, + /* dequeue with active FQ precedence, respect ICS */ + qbman_pull_type_active, + /* dequeue with active FQ precedence, no ICS */ + qbman_pull_type_active_noics +}; + +/** + * qbman_pull_desc_clear() - Clear the contents of a descriptor to + * default/starting state. + * @d: the pull dequeue descriptor to be cleared. + */ +void qbman_pull_desc_clear(struct qbman_pull_desc *d); + +/** + * qbman_pull_desc_set_storage()- Set the pull dequeue storage + * @d: the pull dequeue descriptor to be set. + * @storage: the pointer of the memory to store the dequeue result. + * @storage_phys: the physical address of the storage memory. + * @stash: to indicate whether write allocate is enabled. + * + * If not called, or if called with 'storage' as NULL, the result pull dequeues + * will produce results to DQRR. If 'storage' is non-NULL, then results are + * produced to the given memory location (using the physical/DMA address which + * the caller provides in 'storage_phys'), and 'stash' controls whether or not + * those writes to main-memory express a cache-warming attribute. + */ +void qbman_pull_desc_set_storage(struct qbman_pull_desc *d, + struct qbman_result *storage, + dma_addr_t storage_phys, + int stash); +/** + * qbman_pull_desc_set_numframes() - Set the number of frames to be dequeued. + * @d: the pull dequeue descriptor to be set. + * @numframes: number of frames to be set, must be between 1 and 16, inclusive. + */ +void qbman_pull_desc_set_numframes(struct qbman_pull_desc *d, + uint8_t numframes); +/** + * qbman_pull_desc_set_token() - Set dequeue token for pull command + * @d: the dequeue descriptor + * @token: the token to be set + * + * token is the value that shows up in the dequeue response that can be used to + * detect when the results have been published. The easiest technique is to zero + * result "storage" before issuing a dequeue, and use any non-zero 'token' value + */ +void qbman_pull_desc_set_token(struct qbman_pull_desc *d, uint8_t token); + +/* Exactly one of the following descriptor "actions" should be set. (Calling any + * one of these will replace the effect of any prior call to one of these.) + * - pull dequeue from the given frame queue (FQ) + * - pull dequeue from any FQ in the given work queue (WQ) + * - pull dequeue from any FQ in any WQ in the given channel + */ +/** + * qbman_pull_desc_set_fq() - Set fqid from which the dequeue command dequeues. + * @fqid: the frame queue index of the given FQ. + */ +void qbman_pull_desc_set_fq(struct qbman_pull_desc *d, uint32_t fqid); + +/** + * qbman_pull_desc_set_wq() - Set wqid from which the dequeue command dequeues. + * @wqid: composed of channel id and wqid within the channel. + * @dct: the dequeue command type. + */ +void qbman_pull_desc_set_wq(struct qbman_pull_desc *d, uint32_t wqid, + enum qbman_pull_type_e dct); + +/* qbman_pull_desc_set_channel() - Set channelid from which the dequeue command + * dequeues. + * @chid: the channel id to be dequeued. + * @dct: the dequeue command type. + */ +void qbman_pull_desc_set_channel(struct qbman_pull_desc *d, uint32_t chid, + enum qbman_pull_type_e dct); + +/** + * qbman_swp_pull() - Issue the pull dequeue command + * @s: the software portal object. + * @d: the software portal descriptor which has been configured with + * the set of qbman_pull_desc_set_*() calls. + * + * Return 0 for success, and -EBUSY if the software portal is not ready + * to do pull dequeue. + */ +int qbman_swp_pull(struct qbman_swp *s, struct qbman_pull_desc *d); + +/* -------------------------------- */ +/* Polling DQRR for dequeue results */ +/* -------------------------------- */ + +/** + * qbman_swp_dqrr_next() - Get an valid DQRR entry. + * @s: the software portal object. + * + * Return NULL if there are no unconsumed DQRR entries. Return a DQRR entry + * only once, so repeated calls can return a sequence of DQRR entries, without + * requiring they be consumed immediately or in any particular order. + */ +const struct qbman_result *qbman_swp_dqrr_next(struct qbman_swp *p); + +/** + * qbman_swp_dqrr_consume() - Consume DQRR entries previously returned from + * qbman_swp_dqrr_next(). + * @s: the software portal object. + * @dq: the DQRR entry to be consumed. + */ +void qbman_swp_dqrr_consume(struct qbman_swp *s, const struct qbman_result *dq); + +/** + * qbman_get_dqrr_idx() - Get dqrr index from the given dqrr + * @dqrr: the given dqrr object. + * + * Return dqrr index. + */ +uint8_t qbman_get_dqrr_idx(struct qbman_result *dqrr); + +/** + * qbman_get_dqrr_from_idx() - Use index to get the dqrr entry from the + * given portal + * @s: the given portal. + * @idx: the dqrr index. + * + * Return dqrr entry object. + */ +struct qbman_result *qbman_get_dqrr_from_idx(struct qbman_swp *s, uint8_t idx); + +/* ------------------------------------------------- */ +/* Polling user-provided storage for dequeue results */ +/* ------------------------------------------------- */ + +/** + * qbman_result_has_new_result() - Check and get the dequeue response from the + * dq storage memory set in pull dequeue command + * @s: the software portal object. + * @dq: the dequeue result read from the memory. + * + * Only used for user-provided storage of dequeue results, not DQRR. For + * efficiency purposes, the driver will perform any required endianness + * conversion to ensure that the user's dequeue result storage is in host-endian + * format (whether or not that is the same as the little-endian format that + * hardware DMA'd to the user's storage). As such, once the user has called + * qbman_result_has_new_result() and been returned a valid dequeue result, + * they should not call it again on the same memory location (except of course + * if another dequeue command has been executed to produce a new result to that + * location). + * + * Return 1 for getting a valid dequeue result, or 0 for not getting a valid + * dequeue result. + */ +int qbman_result_has_new_result(struct qbman_swp *s, + const struct qbman_result *dq); + +/* -------------------------------------------------------- */ +/* Parsing dequeue entries (DQRR and user-provided storage) */ +/* -------------------------------------------------------- */ + +/** + * qbman_result_is_DQ() - check the dequeue result is a dequeue response or not + * @dq: the dequeue result to be checked. + * + * DQRR entries may contain non-dequeue results, ie. notifications + */ +int qbman_result_is_DQ(const struct qbman_result *dq); + +/** + * qbman_result_is_SCN() - Check the dequeue result is notification or not + * @dq: the dequeue result to be checked. + * + * All the non-dequeue results (FQDAN/CDAN/CSCN/...) are "state change + * notifications" of one type or another. Some APIs apply to all of them, of the + * form qbman_result_SCN_***(). + */ +static inline int qbman_result_is_SCN(const struct qbman_result *dq) +{ + return !qbman_result_is_DQ(dq); +} + +/* Recognise different notification types, only required if the user allows for + * these to occur, and cares about them when they do. + */ + +/** + * qbman_result_is_FQDAN() - Check for FQ Data Availability + * @dq: the qbman_result object. + * + * Return 1 if this is FQDAN. + */ +int qbman_result_is_FQDAN(const struct qbman_result *dq); + +/** + * qbman_result_is_CDAN() - Check for Channel Data Availability + * @dq: the qbman_result object to check. + * + * Return 1 if this is CDAN. + */ +int qbman_result_is_CDAN(const struct qbman_result *dq); + +/** + * qbman_result_is_CSCN() - Check for Congestion State Change + * @dq: the qbman_result object to check. + * + * Return 1 if this is CSCN. + */ +int qbman_result_is_CSCN(const struct qbman_result *dq); + +/** + * qbman_result_is_BPSCN() - Check for Buffer Pool State Change. + * @dq: the qbman_result object to check. + * + * Return 1 if this is BPSCN. + */ +int qbman_result_is_BPSCN(const struct qbman_result *dq); + +/** + * qbman_result_is_CGCU() - Check for Congestion Group Count Update. + * @dq: the qbman_result object to check. + * + * Return 1 if this is CGCU. + */ +int qbman_result_is_CGCU(const struct qbman_result *dq); + +/* Frame queue state change notifications; (FQDAN in theory counts too as it + * leaves a FQ parked, but it is primarily a data availability notification) + */ + +/** + * qbman_result_is_FQRN() - Check for FQ Retirement Notification. + * @dq: the qbman_result object to check. + * + * Return 1 if this is FQRN. + */ +int qbman_result_is_FQRN(const struct qbman_result *dq); + +/** + * qbman_result_is_FQRNI() - Check for FQ Retirement Immediate + * @dq: the qbman_result object to check. + * + * Return 1 if this is FQRNI. + */ +int qbman_result_is_FQRNI(const struct qbman_result *dq); + +/** + * qbman_result_is_FQPN() - Check for FQ Park Notification + * @dq: the qbman_result object to check. + * + * Return 1 if this is FQPN. + */ +int qbman_result_is_FQPN(const struct qbman_result *dq); + +/* Parsing frame dequeue results (qbman_result_is_DQ() must be TRUE) + */ +/* FQ empty */ +#define QBMAN_DQ_STAT_FQEMPTY 0x80 +/* FQ held active */ +#define QBMAN_DQ_STAT_HELDACTIVE 0x40 +/* FQ force eligible */ +#define QBMAN_DQ_STAT_FORCEELIGIBLE 0x20 +/* Valid frame */ +#define QBMAN_DQ_STAT_VALIDFRAME 0x10 +/* FQ ODP enable */ +#define QBMAN_DQ_STAT_ODPVALID 0x04 +/* Volatile dequeue */ +#define QBMAN_DQ_STAT_VOLATILE 0x02 +/* volatile dequeue command is expired */ +#define QBMAN_DQ_STAT_EXPIRED 0x01 + +/** + * qbman_result_DQ_flags() - Get the STAT field of dequeue response + * @dq: the dequeue result. + * + * Return the state field. + */ +uint32_t qbman_result_DQ_flags(const struct qbman_result *dq); + +/** + * qbman_result_DQ_is_pull() - Check whether the dq response is from a pull + * command. + * @dq: the dequeue result. + * + * Return 1 for volatile(pull) dequeue, 0 for static dequeue. + */ +static inline int qbman_result_DQ_is_pull(const struct qbman_result *dq) +{ + return (int)(qbman_result_DQ_flags(dq) & QBMAN_DQ_STAT_VOLATILE); +} + +/** + * qbman_result_DQ_is_pull_complete() - Check whether the pull command is + * completed. + * @dq: the dequeue result. + * + * Return boolean. + */ +static inline int qbman_result_DQ_is_pull_complete( + const struct qbman_result *dq) +{ + return (int)(qbman_result_DQ_flags(dq) & QBMAN_DQ_STAT_EXPIRED); +} + +/** + * qbman_result_DQ_seqnum() - Get the seqnum field in dequeue response + * seqnum is valid only if VALIDFRAME flag is TRUE + * @dq: the dequeue result. + * + * Return seqnum. + */ +uint16_t qbman_result_DQ_seqnum(const struct qbman_result *dq); + +/** + * qbman_result_DQ_odpid() - Get the seqnum field in dequeue response + * odpid is valid only if ODPVAILD flag is TRUE. + * @dq: the dequeue result. + * + * Return odpid. + */ +uint16_t qbman_result_DQ_odpid(const struct qbman_result *dq); + +/** + * qbman_result_DQ_fqid() - Get the fqid in dequeue response + * @dq: the dequeue result. + * + * Return fqid. + */ +uint32_t qbman_result_DQ_fqid(const struct qbman_result *dq); + +/** + * qbman_result_DQ_byte_count() - Get the byte count in dequeue response + * @dq: the dequeue result. + * + * Return the byte count remaining in the FQ. + */ +uint32_t qbman_result_DQ_byte_count(const struct qbman_result *dq); + +/** + * qbman_result_DQ_frame_count - Get the frame count in dequeue response + * @dq: the dequeue result. + * + * Return the frame count remaining in the FQ. + */ +uint32_t qbman_result_DQ_frame_count(const struct qbman_result *dq); + +/** + * qbman_result_DQ_fqd_ctx() - Get the frame queue context in dequeue response + * @dq: the dequeue result. + * + * Return the frame queue context. + */ +uint64_t qbman_result_DQ_fqd_ctx(const struct qbman_result *dq); + +/** + * qbman_result_DQ_fd() - Get the frame descriptor in dequeue response + * @dq: the dequeue result. + * + * Return the frame descriptor. + */ +const struct qbman_fd *qbman_result_DQ_fd(const struct qbman_result *dq); + +/* State-change notifications (FQDAN/CDAN/CSCN/...). */ + +/** + * qbman_result_SCN_state() - Get the state field in State-change notification + * @scn: the state change notification. + * + * Return the state in the notifiation. + */ +uint8_t qbman_result_SCN_state(const struct qbman_result *scn); + +/** + * qbman_result_SCN_rid() - Get the resource id from the notification + * @scn: the state change notification. + * + * Return the resource id. + */ +uint32_t qbman_result_SCN_rid(const struct qbman_result *scn); + +/** + * qbman_result_SCN_ctx() - get the context from the notification + * @scn: the state change notification. + * + * Return the context. + */ +uint64_t qbman_result_SCN_ctx(const struct qbman_result *scn); + +/** + * qbman_result_SCN_state_in_mem() - Get the state in notification written + * in memory + * @scn: the state change notification. + * + * Return the state. + */ +uint8_t qbman_result_SCN_state_in_mem(const struct qbman_result *scn); + +/** + * qbman_result_SCN_rid_in_mem() - Get the resource id in notification written + * in memory. + * @scn: the state change notification. + * + * Return the resource id. + */ +uint32_t qbman_result_SCN_rid_in_mem(const struct qbman_result *scn); + +/* Type-specific "resource IDs". Mainly for illustration purposes, though it + * also gives the appropriate type widths. + */ +/* Get the FQID from the FQDAN */ +#define qbman_result_FQDAN_fqid(dq) qbman_result_SCN_rid(dq) +/* Get the FQID from the FQRN */ +#define qbman_result_FQRN_fqid(dq) qbman_result_SCN_rid(dq) +/* Get the FQID from the FQRNI */ +#define qbman_result_FQRNI_fqid(dq) qbman_result_SCN_rid(dq) +/* Get the FQID from the FQPN */ +#define qbman_result_FQPN_fqid(dq) qbman_result_SCN_rid(dq) +/* Get the channel ID from the CDAN */ +#define qbman_result_CDAN_cid(dq) ((uint16_t)qbman_result_SCN_rid(dq)) +/* Get the CGID from the CSCN */ +#define qbman_result_CSCN_cgid(dq) ((uint16_t)qbman_result_SCN_rid(dq)) + +/** + * qbman_result_bpscn_bpid() - Get the bpid from BPSCN + * @scn: the state change notification. + * + * Return the buffer pool id. + */ +uint16_t qbman_result_bpscn_bpid(const struct qbman_result *scn); + +/** + * qbman_result_bpscn_has_free_bufs() - Check whether there are free + * buffers in the pool from BPSCN. + * @scn: the state change notification. + * + * Return the number of free buffers. + */ +int qbman_result_bpscn_has_free_bufs(const struct qbman_result *scn); + +/** + * qbman_result_bpscn_is_depleted() - Check BPSCN to see whether the + * buffer pool is depleted. + * @scn: the state change notification. + * + * Return the status of buffer pool depletion. + */ +int qbman_result_bpscn_is_depleted(const struct qbman_result *scn); + +/** + * qbman_result_bpscn_is_surplus() - Check BPSCN to see whether the buffer + * pool is surplus or not. + * @scn: the state change notification. + * + * Return the status of buffer pool surplus. + */ +int qbman_result_bpscn_is_surplus(const struct qbman_result *scn); + +/** + * qbman_result_bpscn_ctx() - Get the BPSCN CTX from BPSCN message + * @scn: the state change notification. + * + * Return the BPSCN context. + */ +uint64_t qbman_result_bpscn_ctx(const struct qbman_result *scn); + +/* Parsing CGCU */ +/** + * qbman_result_cgcu_cgid() - Check CGCU resouce id, i.e. cgid + * @scn: the state change notification. + * + * Return the CGCU resource id. + */ +uint16_t qbman_result_cgcu_cgid(const struct qbman_result *scn); + +/** + * qbman_result_cgcu_icnt() - Get the I_CNT from CGCU + * @scn: the state change notification. + * + * Return instantaneous count in the CGCU notification. + */ +uint64_t qbman_result_cgcu_icnt(const struct qbman_result *scn); + + /************/ + /* Enqueues */ + /************/ + +/** + * struct qbman_eq_desc - structure of enqueue descriptor + * @dont_manipulate_directly: the 8 32bit data to represent the whole + * possible qbman enqueue setting in enqueue descriptor. + */ +struct qbman_eq_desc { + uint32_t dont_manipulate_directly[8]; +}; + +/** + * struct qbman_eq_response - structure of enqueue response + * @dont_manipulate_directly: the 16 32bit data to represent the whole + * enqueue response. + */ +struct qbman_eq_response { + uint32_t dont_manipulate_directly[16]; +}; + +/** + * qbman_eq_desc_clear() - Clear the contents of a descriptor to + * default/starting state. + * @d: the given enqueue descriptor. + */ +void qbman_eq_desc_clear(struct qbman_eq_desc *d); + +/* Exactly one of the following descriptor "actions" should be set. (Calling + * any one of these will replace the effect of any prior call to one of these.) + * - enqueue without order-restoration + * - enqueue with order-restoration + * - fill a hole in the order-restoration sequence, without any enqueue + * - advance NESN (Next Expected Sequence Number), without any enqueue + * 'respond_success' indicates whether an enqueue response should be DMA'd + * after success (otherwise a response is DMA'd only after failure). + * 'incomplete' indicates that other fragments of the same 'seqnum' are yet to + * be enqueued. + */ + +/** + * qbman_eq_desc_set_no_orp() - Set enqueue descriptor without orp + * @d: the enqueue descriptor. + * @response_success: 1 = enqueue with response always; 0 = enqueue with + * rejections returned on a FQ. + */ +void qbman_eq_desc_set_no_orp(struct qbman_eq_desc *d, int respond_success); +/** + * qbman_eq_desc_set_orp() - Set order-resotration in the enqueue descriptor + * @d: the enqueue descriptor. + * @response_success: 1 = enqueue with response always; 0 = enqueue with + * rejections returned on a FQ. + * @opr_id: the order point record id. + * @seqnum: the order restoration sequence number. + * @incomplete: indiates whether this is the last fragments using the same + * sequeue number. + */ +void qbman_eq_desc_set_orp(struct qbman_eq_desc *d, int respond_success, + uint32_t opr_id, uint32_t seqnum, int incomplete); + +/** + * qbman_eq_desc_set_orp_hole() - fill a hole in the order-restoration sequence + * without any enqueue + * @d: the enqueue descriptor. + * @opr_id: the order point record id. + * @seqnum: the order restoration sequence number. + */ +void qbman_eq_desc_set_orp_hole(struct qbman_eq_desc *d, uint32_t opr_id, + uint32_t seqnum); + +/** + * qbman_eq_desc_set_orp_nesn() - advance NESN (Next Expected Sequence Number) + * without any enqueue + * @d: the enqueue descriptor. + * @opr_id: the order point record id. + * @seqnum: the order restoration sequence number. + */ +void qbman_eq_desc_set_orp_nesn(struct qbman_eq_desc *d, uint32_t opr_id, + uint32_t seqnum); +/** + * qbman_eq_desc_set_response() - Set the enqueue response info. + * @d: the enqueue descriptor + * @storage_phys: the physical address of the enqueue response in memory. + * @stash: indicate that the write allocation enabled or not. + * + * In the case where an enqueue response is DMA'd, this determines where that + * response should go. (The physical/DMA address is given for hardware's + * benefit, but software should interpret it as a "struct qbman_eq_response" + * data structure.) 'stash' controls whether or not the write to main-memory + * expresses a cache-warming attribute. + */ +void qbman_eq_desc_set_response(struct qbman_eq_desc *d, + dma_addr_t storage_phys, + int stash); + +/** + * qbman_eq_desc_set_token() - Set token for the enqueue command + * @d: the enqueue descriptor + * @token: the token to be set. + * + * token is the value that shows up in an enqueue response that can be used to + * detect when the results have been published. The easiest technique is to zero + * result "storage" before issuing an enqueue, and use any non-zero 'token' + * value. + */ +void qbman_eq_desc_set_token(struct qbman_eq_desc *d, uint8_t token); + +/** + * Exactly one of the following descriptor "targets" should be set. (Calling any + * one of these will replace the effect of any prior call to one of these.) + * - enqueue to a frame queue + * - enqueue to a queuing destination + * Note, that none of these will have any affect if the "action" type has been + * set to "orp_hole" or "orp_nesn". + */ +/** + * qbman_eq_desc_set_fq() - Set Frame Queue id for the enqueue command + * @d: the enqueue descriptor + * @fqid: the id of the frame queue to be enqueued. + */ +void qbman_eq_desc_set_fq(struct qbman_eq_desc *d, uint32_t fqid); + +/** + * qbman_eq_desc_set_qd() - Set Queuing Destination for the enqueue command. + * @d: the enqueue descriptor + * @qdid: the id of the queuing destination to be enqueued. + * @qd_bin: the queuing destination bin + * @qd_prio: the queuing destination priority. + */ +void qbman_eq_desc_set_qd(struct qbman_eq_desc *d, uint32_t qdid, + uint32_t qd_bin, uint32_t qd_prio); + +/** + * qbman_eq_desc_set_eqdi() - enable/disable EQDI interrupt + * @d: the enqueue descriptor + * @enable: boolean to enable/disable EQDI + * + * Determines whether or not the portal's EQDI interrupt source should be + * asserted after the enqueue command is completed. + */ +void qbman_eq_desc_set_eqdi(struct qbman_eq_desc *d, int enable); + +/** + * qbman_eq_desc_set_dca() - Set DCA mode in the enqueue command. + * @d: the enqueue descriptor. + * @enable: enabled/disable DCA mode. + * @dqrr_idx: DCAP_CI, the DCAP consumer index. + * @park: determine the whether park the FQ or not + * + * Determines whether or not a portal DQRR entry should be consumed once the + * enqueue command is completed. (And if so, and the DQRR entry corresponds to a + * held-active (order-preserving) FQ, whether the FQ should be parked instead of + * being rescheduled.) + */ +void qbman_eq_desc_set_dca(struct qbman_eq_desc *d, int enable, + uint32_t dqrr_idx, int park); + +/** + * qbman_swp_enqueue() - Issue an enqueue command. + * @s: the software portal used for enqueue. + * @d: the enqueue descriptor. + * @fd: the frame descriptor to be enqueued. + * + * Please note that 'fd' should only be NULL if the "action" of the + * descriptor is "orp_hole" or "orp_nesn". + * + * Return 0 for a successful enqueue, -EBUSY if the EQCR is not ready. + */ +int qbman_swp_enqueue(struct qbman_swp *s, const struct qbman_eq_desc *d, + const struct qbman_fd *fd); + +/* TODO: + * qbman_swp_enqueue_thresh() - Set threshold for EQRI interrupt. + * @s: the software portal. + * @thresh: the threshold to trigger the EQRI interrupt. + * + * An EQRI interrupt can be generated when the fill-level of EQCR falls below + * the 'thresh' value set here. Setting thresh==0 (the default) disables. + */ +int qbman_swp_enqueue_thresh(struct qbman_swp *s, unsigned int thresh); + + /*******************/ + /* Buffer releases */ + /*******************/ +/** + * struct qbman_release_desc - The structure for buffer release descriptor + * @dont_manipulate_directly: the 32bit data to represent the whole + * possible settings of qbman release descriptor. + */ +struct qbman_release_desc { + uint32_t dont_manipulate_directly[1]; +}; + +/** + * qbman_release_desc_clear() - Clear the contents of a descriptor to + * default/starting state. + * @d: the qbman release descriptor. + */ +void qbman_release_desc_clear(struct qbman_release_desc *d); + +/** + * qbman_release_desc_set_bpid() - Set the ID of the buffer pool to release to + * @d: the qbman release descriptor. + */ +void qbman_release_desc_set_bpid(struct qbman_release_desc *d, uint32_t bpid); + +/** + * qbman_release_desc_set_rcdi() - Determines whether or not the portal's RCDI + * interrupt source should be asserted after the release command is completed. + * @d: the qbman release descriptor. + */ +void qbman_release_desc_set_rcdi(struct qbman_release_desc *d, int enable); + +/** + * qbman_swp_release() - Issue a buffer release command. + * @s: the software portal object. + * @d: the release descriptor. + * @buffers: a pointer pointing to the buffer address to be released. + * @num_buffers: number of buffers to be released, must be less than 8. + * + * Return 0 for success, -EBUSY if the release command ring is not ready. + */ +int qbman_swp_release(struct qbman_swp *s, const struct qbman_release_desc *d, + const uint64_t *buffers, unsigned int num_buffers); + +/* TODO: + * qbman_swp_release_thresh() - Set threshold for RCRI interrupt + * @s: the software portal. + * @thresh: the threshold. + * An RCRI interrupt can be generated when the fill-level of RCR falls below + * the 'thresh' value set here. Setting thresh==0 (the default) disables. + */ +int qbman_swp_release_thresh(struct qbman_swp *s, unsigned int thresh); + + /*******************/ + /* Buffer acquires */ + /*******************/ +/** + * qbman_swp_acquire() - Issue a buffer acquire command. + * @s: the software portal object. + * @bpid: the buffer pool index. + * @buffers: a pointer pointing to the acquired buffer address|es. + * @num_buffers: number of buffers to be acquired, must be less than 8. + * + * Return 0 for success, or negative error code if the acquire command + * fails. + */ +int qbman_swp_acquire(struct qbman_swp *s, uint32_t bpid, uint64_t *buffers, + unsigned int num_buffers); + + /*****************/ + /* FQ management */ + /*****************/ +/** + * qbman_swp_fq_schedule() - Move the fq to the scheduled state. + * @s: the software portal object. + * @fqid: the index of frame queue to be scheduled. + * + * There are a couple of different ways that a FQ can end up parked state, + * This schedules it. + * + * Return 0 for success, or negative error code for failure. + */ +int qbman_swp_fq_schedule(struct qbman_swp *s, uint32_t fqid); + +/** + * qbman_swp_fq_force() - Force the FQ to fully scheduled state. + * @s: the software portal object. + * @fqid: the index of frame queue to be forced. + * + * Force eligible will force a tentatively-scheduled FQ to be fully-scheduled + * and thus be available for selection by any channel-dequeuing behaviour (push + * or pull). If the FQ is subsequently "dequeued" from the channel and is still + * empty at the time this happens, the resulting dq_entry will have no FD. + * (qbman_result_DQ_fd() will return NULL.) + * + * Return 0 for success, or negative error code for failure. + */ +int qbman_swp_fq_force(struct qbman_swp *s, uint32_t fqid); + +/** + * These functions change the FQ flow-control stuff between XON/XOFF. (The + * default is XON.) This setting doesn't affect enqueues to the FQ, just + * dequeues. XOFF FQs will remain in the tenatively-scheduled state, even when + * non-empty, meaning they won't be selected for scheduled dequeuing. If a FQ is + * changed to XOFF after it had already become truly-scheduled to a channel, and + * a pull dequeue of that channel occurs that selects that FQ for dequeuing, + * then the resulting dq_entry will have no FD. (qbman_result_DQ_fd() will + * return NULL.) + */ +/** + * qbman_swp_fq_xon() - XON the frame queue. + * @s: the software portal object. + * @fqid: the index of frame queue. + * + * Return 0 for success, or negative error code for failure. + */ +int qbman_swp_fq_xon(struct qbman_swp *s, uint32_t fqid); +/** + * qbman_swp_fq_xoff() - XOFF the frame queue. + * @s: the software portal object. + * @fqid: the index of frame queue. + * + * Return 0 for success, or negative error code for failure. + */ +int qbman_swp_fq_xoff(struct qbman_swp *s, uint32_t fqid); + + /**********************/ + /* Channel management */ + /**********************/ + +/** + * If the user has been allocated a channel object that is going to generate + * CDANs to another channel, then these functions will be necessary. + * CDAN-enabled channels only generate a single CDAN notification, after which + * it they need to be reenabled before they'll generate another. (The idea is + * that pull dequeuing will occur in reaction to the CDAN, followed by a + * reenable step.) Each function generates a distinct command to hardware, so a + * combination function is provided if the user wishes to modify the "context" + * (which shows up in each CDAN message) each time they reenable, as a single + * command to hardware. + */ + +/** + * qbman_swp_CDAN_set_context() - Set CDAN context + * @s: the software portal object. + * @channelid: the channel index. + * @ctx: the context to be set in CDAN. + * + * Return 0 for success, or negative error code for failure. + */ +int qbman_swp_CDAN_set_context(struct qbman_swp *s, uint16_t channelid, + uint64_t ctx); + +/** + * qbman_swp_CDAN_enable() - Enable CDAN for the channel. + * @s: the software portal object. + * @channelid: the index of the channel to generate CDAN. + * + * Return 0 for success, or negative error code for failure. + */ +int qbman_swp_CDAN_enable(struct qbman_swp *s, uint16_t channelid); + +/** + * qbman_swp_CDAN_disable() - disable CDAN for the channel. + * @s: the software portal object. + * @channelid: the index of the channel to generate CDAN. + * + * Return 0 for success, or negative error code for failure. + */ +int qbman_swp_CDAN_disable(struct qbman_swp *s, uint16_t channelid); + +/** + * qbman_swp_CDAN_set_context_enable() - Set CDAN contest and enable CDAN + * @s: the software portal object. + * @channelid: the index of the channel to generate CDAN. + * @ctx: the context set in CDAN. + * + * Return 0 for success, or negative error code for failure. + */ +int qbman_swp_CDAN_set_context_enable(struct qbman_swp *s, uint16_t channelid, + uint64_t ctx); +int qbman_swp_fill_ring(struct qbman_swp *s, + const struct qbman_eq_desc *d, + const struct qbman_fd *fd, + uint8_t burst_index); +int qbman_swp_flush_ring(struct qbman_swp *s); +void qbman_sync(void); +int qbman_swp_send_multiple(struct qbman_swp *s, + const struct qbman_eq_desc *d, + const struct qbman_fd *fd, + int frames_to_send); + +int qbman_check_command_complete(struct qbman_swp *s, + const struct qbman_result *dq); + +int qbman_get_version(void); +#endif /* !_FSL_QBMAN_PORTAL_H */ diff --git a/drivers/bus/fslmc/qbman/qbman_portal.c b/drivers/bus/fslmc/qbman/qbman_portal.c new file mode 100644 index 0000000000..5d407cc01d --- /dev/null +++ b/drivers/bus/fslmc/qbman/qbman_portal.c @@ -0,0 +1,1496 @@ +/*- + * BSD LICENSE + * + * Copyright (C) 2014-2016 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. + * + * 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 "qbman_portal.h" + +/* QBMan portal management command codes */ +#define QBMAN_MC_ACQUIRE 0x30 +#define QBMAN_WQCHAN_CONFIGURE 0x46 + +/* CINH register offsets */ +#define QBMAN_CINH_SWP_EQCR_PI 0x800 +#define QBMAN_CINH_SWP_EQCR_CI 0x840 +#define QBMAN_CINH_SWP_EQAR 0x8c0 +#define QBMAN_CINH_SWP_DQPI 0xa00 +#define QBMAN_CINH_SWP_DCAP 0xac0 +#define QBMAN_CINH_SWP_SDQCR 0xb00 +#define QBMAN_CINH_SWP_RAR 0xcc0 +#define QBMAN_CINH_SWP_ISR 0xe00 +#define QBMAN_CINH_SWP_IER 0xe40 +#define QBMAN_CINH_SWP_ISDR 0xe80 +#define QBMAN_CINH_SWP_IIR 0xec0 + +/* CENA register offsets */ +#define QBMAN_CENA_SWP_EQCR(n) (0x000 + ((uint32_t)(n) << 6)) +#define QBMAN_CENA_SWP_DQRR(n) (0x200 + ((uint32_t)(n) << 6)) +#define QBMAN_CENA_SWP_RCR(n) (0x400 + ((uint32_t)(n) << 6)) +#define QBMAN_CENA_SWP_CR 0x600 +#define QBMAN_CENA_SWP_RR(vb) (0x700 + ((uint32_t)(vb) >> 1)) +#define QBMAN_CENA_SWP_VDQCR 0x780 +#define QBMAN_CENA_SWP_EQCR_CI 0x840 + +/* Reverse mapping of QBMAN_CENA_SWP_DQRR() */ +#define QBMAN_IDX_FROM_DQRR(p) (((unsigned long)p & 0x1ff) >> 6) + +/* QBMan FQ management command codes */ +#define QBMAN_FQ_SCHEDULE 0x48 +#define QBMAN_FQ_FORCE 0x49 +#define QBMAN_FQ_XON 0x4d +#define QBMAN_FQ_XOFF 0x4e + +/*******************************/ +/* Pre-defined attribute codes */ +/*******************************/ + +struct qb_attr_code code_generic_verb = QB_CODE(0, 0, 7); +struct qb_attr_code code_generic_rslt = QB_CODE(0, 8, 8); + +/*************************/ +/* SDQCR attribute codes */ +/*************************/ + +/* we put these here because at least some of them are required by + * qbman_swp_init() + */ +struct qb_attr_code code_sdqcr_dct = QB_CODE(0, 24, 2); +struct qb_attr_code code_sdqcr_fc = QB_CODE(0, 29, 1); +struct qb_attr_code code_sdqcr_tok = QB_CODE(0, 16, 8); +static struct qb_attr_code code_eq_dca_idx; +#define CODE_SDQCR_DQSRC(n) QB_CODE(0, n, 1) +enum qbman_sdqcr_dct { + qbman_sdqcr_dct_null = 0, + qbman_sdqcr_dct_prio_ics, + qbman_sdqcr_dct_active_ics, + qbman_sdqcr_dct_active +}; + +enum qbman_sdqcr_fc { + qbman_sdqcr_fc_one = 0, + qbman_sdqcr_fc_up_to_3 = 1 +}; + +struct qb_attr_code code_sdqcr_dqsrc = QB_CODE(0, 0, 16); + +/* We need to keep track of which SWP triggered a pull command + * so keep an array of portal IDs and use the token field to + * be able to find the proper portal + */ +#define MAX_QBMAN_PORTALS 35 +static struct qbman_swp *portal_idx_map[MAX_QBMAN_PORTALS]; + +uint32_t qman_version; + +/*********************************/ +/* Portal constructor/destructor */ +/*********************************/ + +/* Software portals should always be in the power-on state when we initialise, + * due to the CCSR-based portal reset functionality that MC has. + * + * Erk! Turns out that QMan versions prior to 4.1 do not correctly reset DQRR + * valid-bits, so we need to support a workaround where we don't trust + * valid-bits when detecting new entries until any stale ring entries have been + * overwritten at least once. The idea is that we read PI for the first few + * entries, then switch to valid-bit after that. The trick is to clear the + * bug-work-around boolean once the PI wraps around the ring for the first time. + * + * Note: this still carries a slight additional cost once the decrementer hits + * zero. + */ +struct qbman_swp *qbman_swp_init(const struct qbman_swp_desc *d) +{ + int ret; + uint32_t eqcr_pi; + struct qbman_swp *p = kmalloc(sizeof(*p), GFP_KERNEL); + + if (!p) + return NULL; + p->desc = *d; +#ifdef QBMAN_CHECKING + p->mc.check = swp_mc_can_start; +#endif + p->mc.valid_bit = QB_VALID_BIT; + p->sdq = 0; + qb_attr_code_encode(&code_sdqcr_dct, &p->sdq, qbman_sdqcr_dct_prio_ics); + qb_attr_code_encode(&code_sdqcr_fc, &p->sdq, qbman_sdqcr_fc_up_to_3); + qb_attr_code_encode(&code_sdqcr_tok, &p->sdq, 0xbb); + atomic_set(&p->vdq.busy, 1); + p->vdq.valid_bit = QB_VALID_BIT; + p->dqrr.next_idx = 0; + p->dqrr.valid_bit = QB_VALID_BIT; + qman_version = p->desc.qman_version; + if ((qman_version & 0xFFFF0000) < QMAN_REV_4100) { + p->dqrr.dqrr_size = 4; + p->dqrr.reset_bug = 1; + /* Set size of DQRR to 4, encoded in 2 bits */ + code_eq_dca_idx = (struct qb_attr_code)QB_CODE(0, 8, 2); + } else { + p->dqrr.dqrr_size = 8; + p->dqrr.reset_bug = 0; + /* Set size of DQRR to 8, encoded in 3 bits */ + code_eq_dca_idx = (struct qb_attr_code)QB_CODE(0, 8, 3); + } + + ret = qbman_swp_sys_init(&p->sys, d, p->dqrr.dqrr_size); + if (ret) { + kfree(p); + pr_err("qbman_swp_sys_init() failed %d\n", ret); + return NULL; + } + /* SDQCR needs to be initialized to 0 when no channels are + * being dequeued from or else the QMan HW will indicate an + * error. The values that were calculated above will be + * applied when dequeues from a specific channel are enabled + */ + qbman_cinh_write(&p->sys, QBMAN_CINH_SWP_SDQCR, 0); + eqcr_pi = qbman_cinh_read(&p->sys, QBMAN_CINH_SWP_EQCR_PI); + p->eqcr.pi = eqcr_pi & 0xF; + p->eqcr.pi_vb = eqcr_pi & QB_VALID_BIT; + p->eqcr.ci = qbman_cinh_read(&p->sys, QBMAN_CINH_SWP_EQCR_CI) & 0xF; + p->eqcr.available = QBMAN_EQCR_SIZE - qm_cyc_diff(QBMAN_EQCR_SIZE, + p->eqcr.ci, p->eqcr.pi); + + portal_idx_map[p->desc.idx] = p; + return p; +} + +void qbman_swp_finish(struct qbman_swp *p) +{ +#ifdef QBMAN_CHECKING + QBMAN_BUG_ON(p->mc.check != swp_mc_can_start); +#endif + qbman_swp_sys_finish(&p->sys); + portal_idx_map[p->desc.idx] = NULL; + kfree(p); +} + +const struct qbman_swp_desc *qbman_swp_get_desc(struct qbman_swp *p) +{ + return &p->desc; +} + +/**************/ +/* Interrupts */ +/**************/ + +uint32_t qbman_swp_interrupt_get_vanish(struct qbman_swp *p) +{ + return qbman_cinh_read(&p->sys, QBMAN_CINH_SWP_ISDR); +} + +void qbman_swp_interrupt_set_vanish(struct qbman_swp *p, uint32_t mask) +{ + qbman_cinh_write(&p->sys, QBMAN_CINH_SWP_ISDR, mask); +} + +uint32_t qbman_swp_interrupt_read_status(struct qbman_swp *p) +{ + return qbman_cinh_read(&p->sys, QBMAN_CINH_SWP_ISR); +} + +void qbman_swp_interrupt_clear_status(struct qbman_swp *p, uint32_t mask) +{ + qbman_cinh_write(&p->sys, QBMAN_CINH_SWP_ISR, mask); +} + +uint32_t qbman_swp_interrupt_get_trigger(struct qbman_swp *p) +{ + return qbman_cinh_read(&p->sys, QBMAN_CINH_SWP_IER); +} + +void qbman_swp_interrupt_set_trigger(struct qbman_swp *p, uint32_t mask) +{ + qbman_cinh_write(&p->sys, QBMAN_CINH_SWP_IER, mask); +} + +int qbman_swp_interrupt_get_inhibit(struct qbman_swp *p) +{ + return qbman_cinh_read(&p->sys, QBMAN_CINH_SWP_IIR); +} + +void qbman_swp_interrupt_set_inhibit(struct qbman_swp *p, int inhibit) +{ + qbman_cinh_write(&p->sys, QBMAN_CINH_SWP_IIR, inhibit ? 0xffffffff : 0); +} + +/***********************/ +/* Management commands */ +/***********************/ + +/* + * Internal code common to all types of management commands. + */ + +void *qbman_swp_mc_start(struct qbman_swp *p) +{ + void *ret; +#ifdef QBMAN_CHECKING + QBMAN_BUG_ON(p->mc.check != swp_mc_can_start); +#endif + ret = qbman_cena_write_start(&p->sys, QBMAN_CENA_SWP_CR); +#ifdef QBMAN_CHECKING + if (!ret) + p->mc.check = swp_mc_can_submit; +#endif + return ret; +} + +void qbman_swp_mc_submit(struct qbman_swp *p, void *cmd, uint32_t cmd_verb) +{ + uint32_t *v = cmd; +#ifdef QBMAN_CHECKING + QBMAN_BUG_ON(!(p->mc.check != swp_mc_can_submit)); +#endif + /* TBD: "|=" is going to hurt performance. Need to move as many fields + * out of word zero, and for those that remain, the "OR" needs to occur + * at the caller side. This debug check helps to catch cases where the + * caller wants to OR but has forgotten to do so. + */ + QBMAN_BUG_ON((*v & cmd_verb) != *v); + *v = cmd_verb | p->mc.valid_bit; + qbman_cena_write_complete(&p->sys, QBMAN_CENA_SWP_CR, cmd); +#ifdef QBMAN_CHECKING + p->mc.check = swp_mc_can_poll; +#endif +} + +void *qbman_swp_mc_result(struct qbman_swp *p) +{ + uint32_t *ret, verb; +#ifdef QBMAN_CHECKING + QBMAN_BUG_ON(p->mc.check != swp_mc_can_poll); +#endif + qbman_cena_invalidate_prefetch(&p->sys, + QBMAN_CENA_SWP_RR(p->mc.valid_bit)); + ret = qbman_cena_read(&p->sys, QBMAN_CENA_SWP_RR(p->mc.valid_bit)); + /* Remove the valid-bit - command completed iff the rest is non-zero */ + verb = ret[0] & ~QB_VALID_BIT; + if (!verb) + return NULL; +#ifdef QBMAN_CHECKING + p->mc.check = swp_mc_can_start; +#endif + p->mc.valid_bit ^= QB_VALID_BIT; + return ret; +} + +/***********/ +/* Enqueue */ +/***********/ + +/* These should be const, eventually */ +static struct qb_attr_code code_eq_cmd = QB_CODE(0, 0, 2); +static struct qb_attr_code code_eq_eqdi = QB_CODE(0, 3, 1); +static struct qb_attr_code code_eq_dca_en = QB_CODE(0, 15, 1); +static struct qb_attr_code code_eq_dca_pk = QB_CODE(0, 14, 1); +/* Can't set code_eq_dca_idx width. Need qman version. Read at runtime */ +static struct qb_attr_code code_eq_orp_en = QB_CODE(0, 2, 1); +static struct qb_attr_code code_eq_orp_is_nesn = QB_CODE(0, 31, 1); +static struct qb_attr_code code_eq_orp_nlis = QB_CODE(0, 30, 1); +static struct qb_attr_code code_eq_orp_seqnum = QB_CODE(0, 16, 14); +static struct qb_attr_code code_eq_opr_id = QB_CODE(1, 0, 16); +static struct qb_attr_code code_eq_tgt_id = QB_CODE(2, 0, 24); +/* static struct qb_attr_code code_eq_tag = QB_CODE(3, 0, 32); */ +static struct qb_attr_code code_eq_qd_en = QB_CODE(0, 4, 1); +static struct qb_attr_code code_eq_qd_bin = QB_CODE(4, 0, 16); +static struct qb_attr_code code_eq_qd_pri = QB_CODE(4, 16, 4); +static struct qb_attr_code code_eq_rsp_stash = QB_CODE(5, 16, 1); +static struct qb_attr_code code_eq_rsp_id = QB_CODE(5, 24, 8); +static struct qb_attr_code code_eq_rsp_lo = QB_CODE(6, 0, 32); + +enum qbman_eq_cmd_e { + /* No enqueue, primarily for plugging ORP gaps for dropped frames */ + qbman_eq_cmd_empty, + /* DMA an enqueue response once complete */ + qbman_eq_cmd_respond, + /* DMA an enqueue response only if the enqueue fails */ + qbman_eq_cmd_respond_reject +}; + +void qbman_eq_desc_clear(struct qbman_eq_desc *d) +{ + memset(d, 0, sizeof(*d)); +} + +void qbman_eq_desc_set_no_orp(struct qbman_eq_desc *d, int respond_success) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_eq_orp_en, cl, 0); + qb_attr_code_encode(&code_eq_cmd, cl, + respond_success ? qbman_eq_cmd_respond : + qbman_eq_cmd_respond_reject); +} + +void qbman_eq_desc_set_orp(struct qbman_eq_desc *d, int respond_success, + uint32_t opr_id, uint32_t seqnum, int incomplete) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_eq_orp_en, cl, 1); + qb_attr_code_encode(&code_eq_cmd, cl, + respond_success ? qbman_eq_cmd_respond : + qbman_eq_cmd_respond_reject); + qb_attr_code_encode(&code_eq_opr_id, cl, opr_id); + qb_attr_code_encode(&code_eq_orp_seqnum, cl, seqnum); + qb_attr_code_encode(&code_eq_orp_nlis, cl, !!incomplete); +} + +void qbman_eq_desc_set_orp_hole(struct qbman_eq_desc *d, uint32_t opr_id, + uint32_t seqnum) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_eq_orp_en, cl, 1); + qb_attr_code_encode(&code_eq_cmd, cl, qbman_eq_cmd_empty); + qb_attr_code_encode(&code_eq_opr_id, cl, opr_id); + qb_attr_code_encode(&code_eq_orp_seqnum, cl, seqnum); + qb_attr_code_encode(&code_eq_orp_nlis, cl, 0); + qb_attr_code_encode(&code_eq_orp_is_nesn, cl, 0); +} + +void qbman_eq_desc_set_orp_nesn(struct qbman_eq_desc *d, uint32_t opr_id, + uint32_t seqnum) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_eq_orp_en, cl, 1); + qb_attr_code_encode(&code_eq_cmd, cl, qbman_eq_cmd_empty); + qb_attr_code_encode(&code_eq_opr_id, cl, opr_id); + qb_attr_code_encode(&code_eq_orp_seqnum, cl, seqnum); + qb_attr_code_encode(&code_eq_orp_nlis, cl, 0); + qb_attr_code_encode(&code_eq_orp_is_nesn, cl, 1); +} + +void qbman_eq_desc_set_response(struct qbman_eq_desc *d, + dma_addr_t storage_phys, + int stash) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode_64(&code_eq_rsp_lo, (uint64_t *)cl, storage_phys); + qb_attr_code_encode(&code_eq_rsp_stash, cl, !!stash); +} + +void qbman_eq_desc_set_token(struct qbman_eq_desc *d, uint8_t token) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_eq_rsp_id, cl, (uint32_t)token); +} + +void qbman_eq_desc_set_fq(struct qbman_eq_desc *d, uint32_t fqid) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_eq_qd_en, cl, 0); + qb_attr_code_encode(&code_eq_tgt_id, cl, fqid); +} + +void qbman_eq_desc_set_qd(struct qbman_eq_desc *d, uint32_t qdid, + uint32_t qd_bin, uint32_t qd_prio) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_eq_qd_en, cl, 1); + qb_attr_code_encode(&code_eq_tgt_id, cl, qdid); + qb_attr_code_encode(&code_eq_qd_bin, cl, qd_bin); + qb_attr_code_encode(&code_eq_qd_pri, cl, qd_prio); +} + +void qbman_eq_desc_set_eqdi(struct qbman_eq_desc *d, int enable) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_eq_eqdi, cl, !!enable); +} + +void qbman_eq_desc_set_dca(struct qbman_eq_desc *d, int enable, + uint32_t dqrr_idx, int park) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_eq_dca_en, cl, !!enable); + if (enable) { + qb_attr_code_encode(&code_eq_dca_pk, cl, !!park); + qb_attr_code_encode(&code_eq_dca_idx, cl, dqrr_idx); + } +} + +#define EQAR_IDX(eqar) ((eqar) & 0x7) +#define EQAR_VB(eqar) ((eqar) & 0x80) +#define EQAR_SUCCESS(eqar) ((eqar) & 0x100) +static int qbman_swp_enqueue_array_mode(struct qbman_swp *s, + const struct qbman_eq_desc *d, + const struct qbman_fd *fd) +{ + uint32_t *p; + const uint32_t *cl = qb_cl(d); + uint32_t eqar = qbman_cinh_read(&s->sys, QBMAN_CINH_SWP_EQAR); + + pr_debug("EQAR=%08x\n", eqar); + if (!EQAR_SUCCESS(eqar)) + return -EBUSY; + p = qbman_cena_write_start_wo_shadow(&s->sys, + QBMAN_CENA_SWP_EQCR(EQAR_IDX(eqar))); + word_copy(&p[1], &cl[1], 7); + word_copy(&p[8], fd, sizeof(*fd) >> 2); + /* Set the verb byte, have to substitute in the valid-bit */ + lwsync(); + p[0] = cl[0] | EQAR_VB(eqar); + qbman_cena_write_complete_wo_shadow(&s->sys, + QBMAN_CENA_SWP_EQCR(EQAR_IDX(eqar))); + return 0; +} + +static int qbman_swp_enqueue_ring_mode(struct qbman_swp *s, + const struct qbman_eq_desc *d, + const struct qbman_fd *fd) +{ + uint32_t *p; + const uint32_t *cl = qb_cl(d); + uint32_t eqcr_ci; + uint8_t diff; + + if (!s->eqcr.available) { + eqcr_ci = s->eqcr.ci; + s->eqcr.ci = qbman_cena_read_reg(&s->sys, + QBMAN_CENA_SWP_EQCR_CI) & 0xF; + diff = qm_cyc_diff(QBMAN_EQCR_SIZE, + eqcr_ci, s->eqcr.ci); + s->eqcr.available += diff; + if (!diff) + return -EBUSY; + } + + p = qbman_cena_write_start_wo_shadow(&s->sys, + QBMAN_CENA_SWP_EQCR(s->eqcr.pi & 7)); + word_copy(&p[1], &cl[1], 7); + word_copy(&p[8], fd, sizeof(*fd) >> 2); + lwsync(); + /* Set the verb byte, have to substitute in the valid-bit */ + p[0] = cl[0] | s->eqcr.pi_vb; + qbman_cena_write_complete_wo_shadow(&s->sys, + QBMAN_CENA_SWP_EQCR(s->eqcr.pi & 7)); + s->eqcr.pi++; + s->eqcr.pi &= 0xF; + s->eqcr.available--; + if (!(s->eqcr.pi & 7)) + s->eqcr.pi_vb ^= QB_VALID_BIT; + return 0; +} + +int qbman_swp_fill_ring(struct qbman_swp *s, + const struct qbman_eq_desc *d, + const struct qbman_fd *fd, + __attribute__((unused)) uint8_t burst_index) +{ + uint32_t *p; + const uint32_t *cl = qb_cl(d); + uint32_t eqcr_ci; + uint8_t diff; + + if (!s->eqcr.available) { + eqcr_ci = s->eqcr.ci; + s->eqcr.ci = qbman_cena_read_reg(&s->sys, + QBMAN_CENA_SWP_EQCR_CI) & 0xF; + diff = qm_cyc_diff(QBMAN_EQCR_SIZE, + eqcr_ci, s->eqcr.ci); + s->eqcr.available += diff; + if (!diff) + return -EBUSY; + } + p = qbman_cena_write_start_wo_shadow(&s->sys, + QBMAN_CENA_SWP_EQCR((s->eqcr.pi/* +burst_index */) & 7)); + /* word_copy(&p[1], &cl[1], 7); */ + memcpy(&p[1], &cl[1], 7 * 4); + /* word_copy(&p[8], fd, sizeof(*fd) >> 2); */ + memcpy(&p[8], fd, sizeof(struct qbman_fd)); + + /* lwsync(); */ + p[0] = cl[0] | s->eqcr.pi_vb; + + s->eqcr.pi++; + s->eqcr.pi &= 0xF; + s->eqcr.available--; + if (!(s->eqcr.pi & 7)) + s->eqcr.pi_vb ^= QB_VALID_BIT; + + return 0; +} + +int qbman_swp_flush_ring(struct qbman_swp *s) +{ + void *ptr = s->sys.addr_cena; + + dcbf((uint64_t)ptr); + dcbf((uint64_t)ptr + 0x40); + dcbf((uint64_t)ptr + 0x80); + dcbf((uint64_t)ptr + 0xc0); + dcbf((uint64_t)ptr + 0x100); + dcbf((uint64_t)ptr + 0x140); + dcbf((uint64_t)ptr + 0x180); + dcbf((uint64_t)ptr + 0x1c0); + + return 0; +} + +void qbman_sync(void) +{ + lwsync(); +} + +int qbman_swp_enqueue(struct qbman_swp *s, const struct qbman_eq_desc *d, + const struct qbman_fd *fd) +{ + if (s->sys.eqcr_mode == qman_eqcr_vb_array) + return qbman_swp_enqueue_array_mode(s, d, fd); + else /* Use ring mode by default */ + return qbman_swp_enqueue_ring_mode(s, d, fd); +} + +/*************************/ +/* Static (push) dequeue */ +/*************************/ + +void qbman_swp_push_get(struct qbman_swp *s, uint8_t channel_idx, int *enabled) +{ + struct qb_attr_code code = CODE_SDQCR_DQSRC(channel_idx); + + QBMAN_BUG_ON(channel_idx > 15); + *enabled = (int)qb_attr_code_decode(&code, &s->sdq); +} + +void qbman_swp_push_set(struct qbman_swp *s, uint8_t channel_idx, int enable) +{ + uint16_t dqsrc; + struct qb_attr_code code = CODE_SDQCR_DQSRC(channel_idx); + + QBMAN_BUG_ON(channel_idx > 15); + qb_attr_code_encode(&code, &s->sdq, !!enable); + /* Read make the complete src map. If no channels are enabled + * the SDQCR must be 0 or else QMan will assert errors + */ + dqsrc = (uint16_t)qb_attr_code_decode(&code_sdqcr_dqsrc, &s->sdq); + if (dqsrc != 0) + qbman_cinh_write(&s->sys, QBMAN_CINH_SWP_SDQCR, s->sdq); + else + qbman_cinh_write(&s->sys, QBMAN_CINH_SWP_SDQCR, 0); +} + +/***************************/ +/* Volatile (pull) dequeue */ +/***************************/ + +/* These should be const, eventually */ +static struct qb_attr_code code_pull_dct = QB_CODE(0, 0, 2); +static struct qb_attr_code code_pull_dt = QB_CODE(0, 2, 2); +static struct qb_attr_code code_pull_rls = QB_CODE(0, 4, 1); +static struct qb_attr_code code_pull_stash = QB_CODE(0, 5, 1); +static struct qb_attr_code code_pull_numframes = QB_CODE(0, 8, 4); +static struct qb_attr_code code_pull_token = QB_CODE(0, 16, 8); +static struct qb_attr_code code_pull_dqsource = QB_CODE(1, 0, 24); +static struct qb_attr_code code_pull_rsp_lo = QB_CODE(2, 0, 32); + +enum qb_pull_dt_e { + qb_pull_dt_channel, + qb_pull_dt_workqueue, + qb_pull_dt_framequeue +}; + +void qbman_pull_desc_clear(struct qbman_pull_desc *d) +{ + memset(d, 0, sizeof(*d)); +} + +void qbman_pull_desc_set_storage(struct qbman_pull_desc *d, + struct qbman_result *storage, + dma_addr_t storage_phys, + int stash) +{ + uint32_t *cl = qb_cl(d); + /* Squiggle the pointer 'storage' into the extra 2 words of the + * descriptor (which aren't copied to the hw command) + */ + *(void **)&cl[4] = storage; + if (!storage) { + qb_attr_code_encode(&code_pull_rls, cl, 0); + return; + } + qb_attr_code_encode(&code_pull_rls, cl, 1); + qb_attr_code_encode(&code_pull_stash, cl, !!stash); + qb_attr_code_encode_64(&code_pull_rsp_lo, (uint64_t *)cl, storage_phys); +} + +void qbman_pull_desc_set_numframes(struct qbman_pull_desc *d, uint8_t numframes) +{ + uint32_t *cl = qb_cl(d); + + QBMAN_BUG_ON(!numframes || (numframes > 16)); + qb_attr_code_encode(&code_pull_numframes, cl, + (uint32_t)(numframes - 1)); +} + +void qbman_pull_desc_set_token(struct qbman_pull_desc *d, uint8_t token) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_pull_token, cl, token); +} + +void qbman_pull_desc_set_fq(struct qbman_pull_desc *d, uint32_t fqid) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_pull_dct, cl, 1); + qb_attr_code_encode(&code_pull_dt, cl, qb_pull_dt_framequeue); + qb_attr_code_encode(&code_pull_dqsource, cl, fqid); +} + +void qbman_pull_desc_set_wq(struct qbman_pull_desc *d, uint32_t wqid, + enum qbman_pull_type_e dct) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_pull_dct, cl, dct); + qb_attr_code_encode(&code_pull_dt, cl, qb_pull_dt_workqueue); + qb_attr_code_encode(&code_pull_dqsource, cl, wqid); +} + +void qbman_pull_desc_set_channel(struct qbman_pull_desc *d, uint32_t chid, + enum qbman_pull_type_e dct) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_pull_dct, cl, dct); + qb_attr_code_encode(&code_pull_dt, cl, qb_pull_dt_channel); + qb_attr_code_encode(&code_pull_dqsource, cl, chid); +} + +int qbman_swp_pull(struct qbman_swp *s, struct qbman_pull_desc *d) +{ + uint32_t *p; + uint32_t *cl = qb_cl(d); + + if (!atomic_dec_and_test(&s->vdq.busy)) { + atomic_inc(&s->vdq.busy); + return -EBUSY; + } + s->vdq.storage = *(void **)&cl[4]; + /* We use portal index +1 as token so that 0 still indicates + * that the result isn't valid yet. + */ + qb_attr_code_encode(&code_pull_token, cl, s->desc.idx + 1); + p = qbman_cena_write_start_wo_shadow(&s->sys, QBMAN_CENA_SWP_VDQCR); + word_copy(&p[1], &cl[1], 3); + /* Set the verb byte, have to substitute in the valid-bit */ + lwsync(); + p[0] = cl[0] | s->vdq.valid_bit; + s->vdq.valid_bit ^= QB_VALID_BIT; + qbman_cena_write_complete_wo_shadow(&s->sys, QBMAN_CENA_SWP_VDQCR); + return 0; +} + +/****************/ +/* Polling DQRR */ +/****************/ + +static struct qb_attr_code code_dqrr_verb = QB_CODE(0, 0, 8); +static struct qb_attr_code code_dqrr_response = QB_CODE(0, 0, 7); +static struct qb_attr_code code_dqrr_stat = QB_CODE(0, 8, 8); +static struct qb_attr_code code_dqrr_seqnum = QB_CODE(0, 16, 14); +static struct qb_attr_code code_dqrr_odpid = QB_CODE(1, 0, 16); +/* static struct qb_attr_code code_dqrr_tok = QB_CODE(1, 24, 8); */ +static struct qb_attr_code code_dqrr_fqid = QB_CODE(2, 0, 24); +static struct qb_attr_code code_dqrr_byte_count = QB_CODE(4, 0, 32); +static struct qb_attr_code code_dqrr_frame_count = QB_CODE(5, 0, 24); +static struct qb_attr_code code_dqrr_ctx_lo = QB_CODE(6, 0, 32); + +#define QBMAN_RESULT_DQ 0x60 +#define QBMAN_RESULT_FQRN 0x21 +#define QBMAN_RESULT_FQRNI 0x22 +#define QBMAN_RESULT_FQPN 0x24 +#define QBMAN_RESULT_FQDAN 0x25 +#define QBMAN_RESULT_CDAN 0x26 +#define QBMAN_RESULT_CSCN_MEM 0x27 +#define QBMAN_RESULT_CGCU 0x28 +#define QBMAN_RESULT_BPSCN 0x29 +#define QBMAN_RESULT_CSCN_WQ 0x2a + +static struct qb_attr_code code_dqpi_pi = QB_CODE(0, 0, 4); + +/* NULL return if there are no unconsumed DQRR entries. Returns a DQRR entry + * only once, so repeated calls can return a sequence of DQRR entries, without + * requiring they be consumed immediately or in any particular order. + */ +const struct qbman_result *qbman_swp_dqrr_next(struct qbman_swp *s) +{ + uint32_t verb; + uint32_t response_verb; + uint32_t flags; + const struct qbman_result *dq; + const uint32_t *p; + + /* Before using valid-bit to detect if something is there, we have to + * handle the case of the DQRR reset bug... + */ + if (unlikely(s->dqrr.reset_bug)) { + /* We pick up new entries by cache-inhibited producer index, + * which means that a non-coherent mapping would require us to + * invalidate and read *only* once that PI has indicated that + * there's an entry here. The first trip around the DQRR ring + * will be much less efficient than all subsequent trips around + * it... + */ + uint32_t dqpi = qbman_cinh_read(&s->sys, QBMAN_CINH_SWP_DQPI); + uint32_t pi = qb_attr_code_decode(&code_dqpi_pi, &dqpi); + /* there are new entries iff pi != next_idx */ + if (pi == s->dqrr.next_idx) + return NULL; + /* if next_idx is/was the last ring index, and 'pi' is + * different, we can disable the workaround as all the ring + * entries have now been DMA'd to so valid-bit checking is + * repaired. Note: this logic needs to be based on next_idx + * (which increments one at a time), rather than on pi (which + * can burst and wrap-around between our snapshots of it). + */ + QBMAN_BUG_ON((s->dqrr.dqrr_size - 1) < 0); + if (s->dqrr.next_idx == (s->dqrr.dqrr_size - 1u)) { + pr_debug("DEBUG: next_idx=%d, pi=%d, clear reset bug\n", + s->dqrr.next_idx, pi); + s->dqrr.reset_bug = 0; + } + qbman_cena_invalidate_prefetch(&s->sys, + QBMAN_CENA_SWP_DQRR(s->dqrr.next_idx)); + } + dq = qbman_cena_read_wo_shadow(&s->sys, + QBMAN_CENA_SWP_DQRR(s->dqrr.next_idx)); + p = qb_cl(dq); + verb = qb_attr_code_decode(&code_dqrr_verb, p); + /* If the valid-bit isn't of the expected polarity, nothing there. Note, + * in the DQRR reset bug workaround, we shouldn't need to skip these + * check, because we've already determined that a new entry is available + * and we've invalidated the cacheline before reading it, so the + * valid-bit behaviour is repaired and should tell us what we already + * knew from reading PI. + */ + if ((verb & QB_VALID_BIT) != s->dqrr.valid_bit) + return NULL; + + /* There's something there. Move "next_idx" attention to the next ring + * entry (and prefetch it) before returning what we found. + */ + s->dqrr.next_idx++; + if (s->dqrr.next_idx == s->dqrr.dqrr_size) { + s->dqrr.next_idx = 0; + s->dqrr.valid_bit ^= QB_VALID_BIT; + } + /* If this is the final response to a volatile dequeue command + * indicate that the vdq is no longer busy. + */ + flags = qbman_result_DQ_flags(dq); + response_verb = qb_attr_code_decode(&code_dqrr_response, &verb); + if ((response_verb == QBMAN_RESULT_DQ) && + (flags & QBMAN_DQ_STAT_VOLATILE) && + (flags & QBMAN_DQ_STAT_EXPIRED)) + atomic_inc(&s->vdq.busy); + + return dq; +} + +/* Consume DQRR entries previously returned from qbman_swp_dqrr_next(). */ +void qbman_swp_dqrr_consume(struct qbman_swp *s, + const struct qbman_result *dq) +{ + qbman_cinh_write(&s->sys, QBMAN_CINH_SWP_DCAP, QBMAN_IDX_FROM_DQRR(dq)); +} + +/*********************************/ +/* Polling user-provided storage */ +/*********************************/ + +int qbman_result_has_new_result(__attribute__((unused)) struct qbman_swp *s, + const struct qbman_result *dq) +{ + /* To avoid converting the little-endian DQ entry to host-endian prior + * to us knowing whether there is a valid entry or not (and run the + * risk of corrupting the incoming hardware LE write), we detect in + * hardware endianness rather than host. This means we need a different + * "code" depending on whether we are BE or LE in software, which is + * where DQRR_TOK_OFFSET comes in... + */ + static struct qb_attr_code code_dqrr_tok_detect = + QB_CODE(0, DQRR_TOK_OFFSET, 8); + /* The user trying to poll for a result treats "dq" as const. It is + * however the same address that was provided to us non-const in the + * first place, for directing hardware DMA to. So we can cast away the + * const because it is mutable from our perspective. + */ + uint32_t *p = (uint32_t *)(unsigned long)qb_cl(dq); + uint32_t token; + + token = qb_attr_code_decode(&code_dqrr_tok_detect, &p[1]); + if (token == 0) + return 0; + /* Entry is valid - overwrite token back to 0 so + * a) If this memory is reused tokesn will be 0 + * b) If someone calls "has_new_result()" again on this entry it + * will not appear to be new + */ + qb_attr_code_encode(&code_dqrr_tok_detect, &p[1], 0); + + /* Only now do we convert from hardware to host endianness. Also, as we + * are returning success, the user has promised not to call us again, so + * there's no risk of us converting the endianness twice... + */ + make_le32_n(p, 16); + return 1; +} + +int qbman_check_command_complete(struct qbman_swp *s, + const struct qbman_result *dq) +{ + /* To avoid converting the little-endian DQ entry to host-endian prior + * to us knowing whether there is a valid entry or not (and run the + * risk of corrupting the incoming hardware LE write), we detect in + * hardware endianness rather than host. This means we need a different + * "code" depending on whether we are BE or LE in software, which is + * where DQRR_TOK_OFFSET comes in... + */ + static struct qb_attr_code code_dqrr_tok_detect = + QB_CODE(0, DQRR_TOK_OFFSET, 8); + /* The user trying to poll for a result treats "dq" as const. It is + * however the same address that was provided to us non-const in the + * first place, for directing hardware DMA to. So we can cast away the + * const because it is mutable from our perspective. + */ + uint32_t *p = (uint32_t *)(unsigned long)qb_cl(dq); + uint32_t token; + + token = qb_attr_code_decode(&code_dqrr_tok_detect, &p[1]); + if (token == 0) + return 0; + /* TODO: Remove qbman_swp from parameters and make it a local + * once we've tested the reserve portal map change + */ + s = portal_idx_map[token - 1]; + /* When token is set it indicates that VDQ command has been fetched + * by qbman and is working on it. It is safe for software to issue + * another VDQ command, so incrementing the busy variable. + */ + if (s->vdq.storage == dq) { + s->vdq.storage = NULL; + atomic_inc(&s->vdq.busy); + } + return 1; +} + +/********************************/ +/* Categorising qbman results */ +/********************************/ + +static struct qb_attr_code code_result_in_mem = + QB_CODE(0, QBMAN_RESULT_VERB_OFFSET_IN_MEM, 7); + +static inline int __qbman_result_is_x(const struct qbman_result *dq, + uint32_t x) +{ + const uint32_t *p = qb_cl(dq); + uint32_t response_verb = qb_attr_code_decode(&code_dqrr_response, p); + + return (response_verb == x); +} + +static inline int __qbman_result_is_x_in_mem(const struct qbman_result *dq, + uint32_t x) +{ + const uint32_t *p = qb_cl(dq); + uint32_t response_verb = qb_attr_code_decode(&code_result_in_mem, p); + + return (response_verb == x); +} + +int qbman_result_is_DQ(const struct qbman_result *dq) +{ + return __qbman_result_is_x(dq, QBMAN_RESULT_DQ); +} + +int qbman_result_is_FQDAN(const struct qbman_result *dq) +{ + return __qbman_result_is_x(dq, QBMAN_RESULT_FQDAN); +} + +int qbman_result_is_CDAN(const struct qbman_result *dq) +{ + return __qbman_result_is_x(dq, QBMAN_RESULT_CDAN); +} + +int qbman_result_is_CSCN(const struct qbman_result *dq) +{ + return __qbman_result_is_x_in_mem(dq, QBMAN_RESULT_CSCN_MEM) || + __qbman_result_is_x(dq, QBMAN_RESULT_CSCN_WQ); +} + +int qbman_result_is_BPSCN(const struct qbman_result *dq) +{ + return __qbman_result_is_x_in_mem(dq, QBMAN_RESULT_BPSCN); +} + +int qbman_result_is_CGCU(const struct qbman_result *dq) +{ + return __qbman_result_is_x_in_mem(dq, QBMAN_RESULT_CGCU); +} + +int qbman_result_is_FQRN(const struct qbman_result *dq) +{ + return __qbman_result_is_x_in_mem(dq, QBMAN_RESULT_FQRN); +} + +int qbman_result_is_FQRNI(const struct qbman_result *dq) +{ + return __qbman_result_is_x_in_mem(dq, QBMAN_RESULT_FQRNI); +} + +int qbman_result_is_FQPN(const struct qbman_result *dq) +{ + return __qbman_result_is_x(dq, QBMAN_RESULT_FQPN); +} + +/*********************************/ +/* Parsing frame dequeue results */ +/*********************************/ + +/* These APIs assume qbman_result_is_DQ() is TRUE */ + +uint32_t qbman_result_DQ_flags(const struct qbman_result *dq) +{ + const uint32_t *p = qb_cl(dq); + + return qb_attr_code_decode(&code_dqrr_stat, p); +} + +uint16_t qbman_result_DQ_seqnum(const struct qbman_result *dq) +{ + const uint32_t *p = qb_cl(dq); + + return (uint16_t)qb_attr_code_decode(&code_dqrr_seqnum, p); +} + +uint16_t qbman_result_DQ_odpid(const struct qbman_result *dq) +{ + const uint32_t *p = qb_cl(dq); + + return (uint16_t)qb_attr_code_decode(&code_dqrr_odpid, p); +} + +uint32_t qbman_result_DQ_fqid(const struct qbman_result *dq) +{ + const uint32_t *p = qb_cl(dq); + + return qb_attr_code_decode(&code_dqrr_fqid, p); +} + +uint32_t qbman_result_DQ_byte_count(const struct qbman_result *dq) +{ + const uint32_t *p = qb_cl(dq); + + return qb_attr_code_decode(&code_dqrr_byte_count, p); +} + +uint32_t qbman_result_DQ_frame_count(const struct qbman_result *dq) +{ + const uint32_t *p = qb_cl(dq); + + return qb_attr_code_decode(&code_dqrr_frame_count, p); +} + +uint64_t qbman_result_DQ_fqd_ctx(const struct qbman_result *dq) +{ + const uint64_t *p = (const uint64_t *)qb_cl(dq); + + return qb_attr_code_decode_64(&code_dqrr_ctx_lo, p); +} + +const struct qbman_fd *qbman_result_DQ_fd(const struct qbman_result *dq) +{ + const uint32_t *p = qb_cl(dq); + + return (const struct qbman_fd *)&p[8]; +} + +/**************************************/ +/* Parsing state-change notifications */ +/**************************************/ + +static struct qb_attr_code code_scn_state = QB_CODE(0, 16, 8); +static struct qb_attr_code code_scn_rid = QB_CODE(1, 0, 24); +static struct qb_attr_code code_scn_state_in_mem = + QB_CODE(0, SCN_STATE_OFFSET_IN_MEM, 8); +static struct qb_attr_code code_scn_rid_in_mem = + QB_CODE(1, SCN_RID_OFFSET_IN_MEM, 24); +static struct qb_attr_code code_scn_ctx_lo = QB_CODE(2, 0, 32); + +uint8_t qbman_result_SCN_state(const struct qbman_result *scn) +{ + const uint32_t *p = qb_cl(scn); + + return (uint8_t)qb_attr_code_decode(&code_scn_state, p); +} + +uint32_t qbman_result_SCN_rid(const struct qbman_result *scn) +{ + const uint32_t *p = qb_cl(scn); + + return qb_attr_code_decode(&code_scn_rid, p); +} + +uint64_t qbman_result_SCN_ctx(const struct qbman_result *scn) +{ + const uint64_t *p = (const uint64_t *)qb_cl(scn); + + return qb_attr_code_decode_64(&code_scn_ctx_lo, p); +} + +uint8_t qbman_result_SCN_state_in_mem(const struct qbman_result *scn) +{ + const uint32_t *p = qb_cl(scn); + + return (uint8_t)qb_attr_code_decode(&code_scn_state_in_mem, p); +} + +uint32_t qbman_result_SCN_rid_in_mem(const struct qbman_result *scn) +{ + const uint32_t *p = qb_cl(scn); + uint32_t result_rid; + + result_rid = qb_attr_code_decode(&code_scn_rid_in_mem, p); + return make_le24(result_rid); +} + +/*****************/ +/* Parsing BPSCN */ +/*****************/ +uint16_t qbman_result_bpscn_bpid(const struct qbman_result *scn) +{ + return (uint16_t)qbman_result_SCN_rid_in_mem(scn) & 0x3FFF; +} + +int qbman_result_bpscn_has_free_bufs(const struct qbman_result *scn) +{ + return !(int)(qbman_result_SCN_state_in_mem(scn) & 0x1); +} + +int qbman_result_bpscn_is_depleted(const struct qbman_result *scn) +{ + return (int)(qbman_result_SCN_state_in_mem(scn) & 0x2); +} + +int qbman_result_bpscn_is_surplus(const struct qbman_result *scn) +{ + return (int)(qbman_result_SCN_state_in_mem(scn) & 0x4); +} + +uint64_t qbman_result_bpscn_ctx(const struct qbman_result *scn) +{ + uint64_t ctx; + uint32_t ctx_hi, ctx_lo; + + ctx = qbman_result_SCN_ctx(scn); + ctx_hi = upper32(ctx); + ctx_lo = lower32(ctx); + return ((uint64_t)make_le32(ctx_hi) << 32 | + (uint64_t)make_le32(ctx_lo)); +} + +/*****************/ +/* Parsing CGCU */ +/*****************/ +uint16_t qbman_result_cgcu_cgid(const struct qbman_result *scn) +{ + return (uint16_t)qbman_result_SCN_rid_in_mem(scn) & 0xFFFF; +} + +uint64_t qbman_result_cgcu_icnt(const struct qbman_result *scn) +{ + uint64_t ctx; + uint32_t ctx_hi, ctx_lo; + + ctx = qbman_result_SCN_ctx(scn); + ctx_hi = upper32(ctx); + ctx_lo = lower32(ctx); + return ((uint64_t)(make_le32(ctx_hi) & 0xFF) << 32) | + (uint64_t)make_le32(ctx_lo); +} + +/******************/ +/* Buffer release */ +/******************/ + +/* These should be const, eventually */ +/* static struct qb_attr_code code_release_num = QB_CODE(0, 0, 3); */ +static struct qb_attr_code code_release_set_me = QB_CODE(0, 5, 1); +static struct qb_attr_code code_release_rcdi = QB_CODE(0, 6, 1); +static struct qb_attr_code code_release_bpid = QB_CODE(0, 16, 16); + +void qbman_release_desc_clear(struct qbman_release_desc *d) +{ + uint32_t *cl; + + memset(d, 0, sizeof(*d)); + cl = qb_cl(d); + qb_attr_code_encode(&code_release_set_me, cl, 1); +} + +void qbman_release_desc_set_bpid(struct qbman_release_desc *d, uint32_t bpid) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_release_bpid, cl, bpid); +} + +void qbman_release_desc_set_rcdi(struct qbman_release_desc *d, int enable) +{ + uint32_t *cl = qb_cl(d); + + qb_attr_code_encode(&code_release_rcdi, cl, !!enable); +} + +#define RAR_IDX(rar) ((rar) & 0x7) +#define RAR_VB(rar) ((rar) & 0x80) +#define RAR_SUCCESS(rar) ((rar) & 0x100) + +int qbman_swp_release(struct qbman_swp *s, const struct qbman_release_desc *d, + const uint64_t *buffers, unsigned int num_buffers) +{ + uint32_t *p; + const uint32_t *cl = qb_cl(d); + uint32_t rar = qbman_cinh_read(&s->sys, QBMAN_CINH_SWP_RAR); + + pr_debug("RAR=%08x\n", rar); + if (!RAR_SUCCESS(rar)) + return -EBUSY; + QBMAN_BUG_ON(!num_buffers || (num_buffers > 7)); + /* Start the release command */ + p = qbman_cena_write_start_wo_shadow(&s->sys, + QBMAN_CENA_SWP_RCR(RAR_IDX(rar))); + /* Copy the caller's buffer pointers to the command */ + u64_to_le32_copy(&p[2], buffers, num_buffers); + /* Set the verb byte, have to substitute in the valid-bit and the number + * of buffers. + */ + lwsync(); + p[0] = cl[0] | RAR_VB(rar) | num_buffers; + qbman_cena_write_complete_wo_shadow(&s->sys, + QBMAN_CENA_SWP_RCR(RAR_IDX(rar))); + return 0; +} + +/*******************/ +/* Buffer acquires */ +/*******************/ + +/* These should be const, eventually */ +static struct qb_attr_code code_acquire_bpid = QB_CODE(0, 16, 16); +static struct qb_attr_code code_acquire_num = QB_CODE(1, 0, 3); +static struct qb_attr_code code_acquire_r_num = QB_CODE(1, 0, 3); + +int qbman_swp_acquire(struct qbman_swp *s, uint32_t bpid, uint64_t *buffers, + unsigned int num_buffers) +{ + uint32_t *p; + uint32_t rslt, num; + + QBMAN_BUG_ON(!num_buffers || (num_buffers > 7)); + + /* Start the management command */ + p = qbman_swp_mc_start(s); + + if (!p) + return -EBUSY; + + /* Encode the caller-provided attributes */ + qb_attr_code_encode(&code_acquire_bpid, p, bpid); + qb_attr_code_encode(&code_acquire_num, p, num_buffers); + + /* Complete the management command */ + p = qbman_swp_mc_complete(s, p, p[0] | QBMAN_MC_ACQUIRE); + + /* Decode the outcome */ + rslt = qb_attr_code_decode(&code_generic_rslt, p); + num = qb_attr_code_decode(&code_acquire_r_num, p); + QBMAN_BUG_ON(qb_attr_code_decode(&code_generic_verb, p) != + QBMAN_MC_ACQUIRE); + + /* Determine success or failure */ + if (unlikely(rslt != QBMAN_MC_RSLT_OK)) { + pr_err("Acquire buffers from BPID 0x%x failed, code=0x%02x\n", + bpid, rslt); + return -EIO; + } + QBMAN_BUG_ON(num > num_buffers); + /* Copy the acquired buffers to the caller's array */ + u64_from_le32_copy(buffers, &p[2], num); + return (int)num; +} + +/*****************/ +/* FQ management */ +/*****************/ + +static struct qb_attr_code code_fqalt_fqid = QB_CODE(1, 0, 32); + +static int qbman_swp_alt_fq_state(struct qbman_swp *s, uint32_t fqid, + uint8_t alt_fq_verb) +{ + uint32_t *p; + uint32_t rslt; + + /* Start the management command */ + p = qbman_swp_mc_start(s); + if (!p) + return -EBUSY; + + qb_attr_code_encode(&code_fqalt_fqid, p, fqid); + /* Complete the management command */ + p = qbman_swp_mc_complete(s, p, p[0] | alt_fq_verb); + + /* Decode the outcome */ + rslt = qb_attr_code_decode(&code_generic_rslt, p); + QBMAN_BUG_ON(qb_attr_code_decode(&code_generic_verb, p) != alt_fq_verb); + + /* Determine success or failure */ + if (unlikely(rslt != QBMAN_MC_RSLT_OK)) { + pr_err("ALT FQID %d failed: verb = 0x%08x, code = 0x%02x\n", + fqid, alt_fq_verb, rslt); + return -EIO; + } + + return 0; +} + +int qbman_swp_fq_schedule(struct qbman_swp *s, uint32_t fqid) +{ + return qbman_swp_alt_fq_state(s, fqid, QBMAN_FQ_SCHEDULE); +} + +int qbman_swp_fq_force(struct qbman_swp *s, uint32_t fqid) +{ + return qbman_swp_alt_fq_state(s, fqid, QBMAN_FQ_FORCE); +} + +int qbman_swp_fq_xon(struct qbman_swp *s, uint32_t fqid) +{ + return qbman_swp_alt_fq_state(s, fqid, QBMAN_FQ_XON); +} + +int qbman_swp_fq_xoff(struct qbman_swp *s, uint32_t fqid) +{ + return qbman_swp_alt_fq_state(s, fqid, QBMAN_FQ_XOFF); +} + +/**********************/ +/* Channel management */ +/**********************/ + +static struct qb_attr_code code_cdan_cid = QB_CODE(0, 16, 12); +static struct qb_attr_code code_cdan_we = QB_CODE(1, 0, 8); +static struct qb_attr_code code_cdan_en = QB_CODE(1, 8, 1); +static struct qb_attr_code code_cdan_ctx_lo = QB_CODE(2, 0, 32); + +/* Hide "ICD" for now as we don't use it, don't set it, and don't test it, so it + * would be irresponsible to expose it. + */ +#define CODE_CDAN_WE_EN 0x1 +#define CODE_CDAN_WE_CTX 0x4 + +static int qbman_swp_CDAN_set(struct qbman_swp *s, uint16_t channelid, + uint8_t we_mask, uint8_t cdan_en, + uint64_t ctx) +{ + uint32_t *p; + uint32_t rslt; + + /* Start the management command */ + p = qbman_swp_mc_start(s); + if (!p) + return -EBUSY; + + /* Encode the caller-provided attributes */ + qb_attr_code_encode(&code_cdan_cid, p, channelid); + qb_attr_code_encode(&code_cdan_we, p, we_mask); + qb_attr_code_encode(&code_cdan_en, p, cdan_en); + qb_attr_code_encode_64(&code_cdan_ctx_lo, (uint64_t *)p, ctx); + /* Complete the management command */ + p = qbman_swp_mc_complete(s, p, p[0] | QBMAN_WQCHAN_CONFIGURE); + + /* Decode the outcome */ + rslt = qb_attr_code_decode(&code_generic_rslt, p); + QBMAN_BUG_ON(qb_attr_code_decode(&code_generic_verb, p) + != QBMAN_WQCHAN_CONFIGURE); + + /* Determine success or failure */ + if (unlikely(rslt != QBMAN_MC_RSLT_OK)) { + pr_err("CDAN cQID %d failed: code = 0x%02x\n", + channelid, rslt); + return -EIO; + } + + return 0; +} + +int qbman_swp_CDAN_set_context(struct qbman_swp *s, uint16_t channelid, + uint64_t ctx) +{ + return qbman_swp_CDAN_set(s, channelid, + CODE_CDAN_WE_CTX, + 0, ctx); +} + +int qbman_swp_CDAN_enable(struct qbman_swp *s, uint16_t channelid) +{ + return qbman_swp_CDAN_set(s, channelid, + CODE_CDAN_WE_EN, + 1, 0); +} + +int qbman_swp_CDAN_disable(struct qbman_swp *s, uint16_t channelid) +{ + return qbman_swp_CDAN_set(s, channelid, + CODE_CDAN_WE_EN, + 0, 0); +} + +int qbman_swp_CDAN_set_context_enable(struct qbman_swp *s, uint16_t channelid, + uint64_t ctx) +{ + return qbman_swp_CDAN_set(s, channelid, + CODE_CDAN_WE_EN | CODE_CDAN_WE_CTX, + 1, ctx); +} + +uint8_t qbman_get_dqrr_idx(struct qbman_result *dqrr) +{ + return QBMAN_IDX_FROM_DQRR(dqrr); +} + +struct qbman_result *qbman_get_dqrr_from_idx(struct qbman_swp *s, uint8_t idx) +{ + struct qbman_result *dq; + + dq = qbman_cena_read(&s->sys, QBMAN_CENA_SWP_DQRR(idx)); + return dq; +} + +int qbman_swp_send_multiple(struct qbman_swp *s, + const struct qbman_eq_desc *d, + const struct qbman_fd *fd, + int frames_to_send) +{ + uint32_t *p; + const uint32_t *cl = qb_cl(d); + uint32_t eqcr_ci; + uint8_t diff; + int sent = 0; + int i; + int initial_pi = s->eqcr.pi; + uint64_t start_pointer; + + if (!s->eqcr.available) { + eqcr_ci = s->eqcr.ci; + s->eqcr.ci = qbman_cena_read_reg(&s->sys, + QBMAN_CENA_SWP_EQCR_CI) & 0xF; + diff = qm_cyc_diff(QBMAN_EQCR_SIZE, + eqcr_ci, s->eqcr.ci); + if (!diff) + goto done; + s->eqcr.available += diff; + } + + /* we are trying to send frames_to_send, + * if we have enough space in the ring + */ + while (s->eqcr.available && frames_to_send--) { + p = qbman_cena_write_start_wo_shadow_fast(&s->sys, + QBMAN_CENA_SWP_EQCR((initial_pi) & 7)); + /* Write command (except of first byte) and FD */ + memcpy(&p[1], &cl[1], 7 * 4); + memcpy(&p[8], &fd[sent], sizeof(struct qbman_fd)); + + initial_pi++; + initial_pi &= 0xF; + s->eqcr.available--; + sent++; + } + +done: + initial_pi = s->eqcr.pi; + lwsync(); + + /* in order for flushes to complete faster: + * we use a following trick: we record all lines in 32 bit word + */ + + initial_pi = s->eqcr.pi; + for (i = 0; i < sent; i++) { + p = qbman_cena_write_start_wo_shadow_fast(&s->sys, + QBMAN_CENA_SWP_EQCR((initial_pi) & 7)); + + p[0] = cl[0] | s->eqcr.pi_vb; + initial_pi++; + initial_pi &= 0xF; + + if (!(initial_pi & 7)) + s->eqcr.pi_vb ^= QB_VALID_BIT; + } + + initial_pi = s->eqcr.pi; + + /* We need to flush all the lines but without + * load/store operations between them. + * We assign start_pointer before we start loop so that + * in loop we do not read it from memory + */ + start_pointer = (uint64_t)s->sys.addr_cena; + for (i = 0; i < sent; i++) { + p = (uint32_t *)(start_pointer + + QBMAN_CENA_SWP_EQCR(initial_pi & 7)); + dcbf((uint64_t)p); + initial_pi++; + initial_pi &= 0xF; + } + + /* Update producer index for the next call */ + s->eqcr.pi = initial_pi; + + return sent; +} + +int qbman_get_version(void) +{ + return qman_version; +} diff --git a/drivers/bus/fslmc/qbman/qbman_portal.h b/drivers/bus/fslmc/qbman/qbman_portal.h new file mode 100644 index 0000000000..7aa1d4f66b --- /dev/null +++ b/drivers/bus/fslmc/qbman/qbman_portal.h @@ -0,0 +1,277 @@ +/*- + * BSD LICENSE + * + * Copyright (C) 2014-2016 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. + * + * 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 "qbman_private.h" +#include + +/* All QBMan command and result structures use this "valid bit" encoding */ +#define QB_VALID_BIT ((uint32_t)0x80) + +/* Management command result codes */ +#define QBMAN_MC_RSLT_OK 0xf0 + +/* QBMan DQRR size is set at runtime in qbman_portal.c */ + +#define QBMAN_EQCR_SIZE 8 + +static inline u8 qm_cyc_diff(u8 ringsize, u8 first, u8 last) +{ + /* 'first' is included, 'last' is excluded */ + if (first <= last) + return last - first; + return (2 * ringsize) + last - first; +} + +/* --------------------- */ +/* portal data structure */ +/* --------------------- */ + +struct qbman_swp { + struct qbman_swp_desc desc; + /* The qbman_sys (ie. arch/OS-specific) support code can put anything it + * needs in here. + */ + struct qbman_swp_sys sys; + /* Management commands */ + struct { +#ifdef QBMAN_CHECKING + enum swp_mc_check { + swp_mc_can_start, /* call __qbman_swp_mc_start() */ + swp_mc_can_submit, /* call __qbman_swp_mc_submit() */ + swp_mc_can_poll, /* call __qbman_swp_mc_result() */ + } check; +#endif + uint32_t valid_bit; /* 0x00 or 0x80 */ + } mc; + /* Push dequeues */ + uint32_t sdq; + /* Volatile dequeues */ + struct { + /* VDQCR supports a "1 deep pipeline", meaning that if you know + * the last-submitted command is already executing in the + * hardware (as evidenced by at least 1 valid dequeue result), + * you can write another dequeue command to the register, the + * hardware will start executing it as soon as the + * already-executing command terminates. (This minimises latency + * and stalls.) With that in mind, this "busy" variable refers + * to whether or not a command can be submitted, not whether or + * not a previously-submitted command is still executing. In + * other words, once proof is seen that the previously-submitted + * command is executing, "vdq" is no longer "busy". + */ + atomic_t busy; + uint32_t valid_bit; /* 0x00 or 0x80 */ + /* We need to determine when vdq is no longer busy. This depends + * on whether the "busy" (last-submitted) dequeue command is + * targeting DQRR or main-memory, and detected is based on the + * presence of the dequeue command's "token" showing up in + * dequeue entries in DQRR or main-memory (respectively). + */ + struct qbman_result *storage; /* NULL if DQRR */ + } vdq; + /* DQRR */ + struct { + uint32_t next_idx; + uint32_t valid_bit; + uint8_t dqrr_size; + int reset_bug; + } dqrr; + struct { + uint32_t pi; + uint32_t pi_vb; + uint32_t ci; + int available; + } eqcr; +}; + +/* -------------------------- */ +/* portal management commands */ +/* -------------------------- */ + +/* Different management commands all use this common base layer of code to issue + * commands and poll for results. The first function returns a pointer to where + * the caller should fill in their MC command (though they should ignore the + * verb byte), the second function commits merges in the caller-supplied command + * verb (which should not include the valid-bit) and submits the command to + * hardware, and the third function checks for a completed response (returns + * non-NULL if only if the response is complete). + */ +void *qbman_swp_mc_start(struct qbman_swp *p); +void qbman_swp_mc_submit(struct qbman_swp *p, void *cmd, uint32_t cmd_verb); +void *qbman_swp_mc_result(struct qbman_swp *p); + +/* Wraps up submit + poll-for-result */ +static inline void *qbman_swp_mc_complete(struct qbman_swp *swp, void *cmd, + uint32_t cmd_verb) +{ + int loopvar; + + qbman_swp_mc_submit(swp, cmd, cmd_verb); + DBG_POLL_START(loopvar); + do { + DBG_POLL_CHECK(loopvar); + cmd = qbman_swp_mc_result(swp); + } while (!cmd); + return cmd; +} + +/* ------------ */ +/* qb_attr_code */ +/* ------------ */ + +/* This struct locates a sub-field within a QBMan portal (CENA) cacheline which + * is either serving as a configuration command or a query result. The + * representation is inherently little-endian, as the indexing of the words is + * itself little-endian in nature and DPAA2 QBMan is little endian for anything + * that crosses a word boundary too (64-bit fields are the obvious examples). + */ +struct qb_attr_code { + unsigned int word; /* which uint32_t[] array member encodes the field */ + unsigned int lsoffset; /* encoding offset from ls-bit */ + unsigned int width; /* encoding width. (bool must be 1.) */ +}; + +/* Some pre-defined codes */ +extern struct qb_attr_code code_generic_verb; +extern struct qb_attr_code code_generic_rslt; + +/* Macros to define codes */ +#define QB_CODE(a, b, c) { a, b, c} +#define QB_CODE_NULL \ + QB_CODE((unsigned int)-1, (unsigned int)-1, (unsigned int)-1) + +/* Rotate a code "ms", meaning that it moves from less-significant bytes to + * more-significant, from less-significant words to more-significant, etc. The + * "ls" version does the inverse, from more-significant towards + * less-significant. + */ +static inline void qb_attr_code_rotate_ms(struct qb_attr_code *code, + unsigned int bits) +{ + code->lsoffset += bits; + while (code->lsoffset > 31) { + code->word++; + code->lsoffset -= 32; + } +} + +static inline void qb_attr_code_rotate_ls(struct qb_attr_code *code, + unsigned int bits) +{ + /* Don't be fooled, this trick should work because the types are + * unsigned. So the case that interests the while loop (the rotate has + * gone too far and the word count needs to compensate for it), is + * manifested when lsoffset is negative. But that equates to a really + * large unsigned value, starting with lots of "F"s. As such, we can + * continue adding 32 back to it until it wraps back round above zero, + * to a value of 31 or less... + */ + code->lsoffset -= bits; + while (code->lsoffset > 31) { + code->word--; + code->lsoffset += 32; + } +} + +/* Implement a loop of code rotations until 'expr' evaluates to FALSE (0). */ +#define qb_attr_code_for_ms(code, bits, expr) \ + for (; expr; qb_attr_code_rotate_ms(code, bits)) +#define qb_attr_code_for_ls(code, bits, expr) \ + for (; expr; qb_attr_code_rotate_ls(code, bits)) + +/* decode a field from a cacheline */ +static inline uint32_t qb_attr_code_decode(const struct qb_attr_code *code, + const uint32_t *cacheline) +{ + return d32_uint32_t(code->lsoffset, code->width, cacheline[code->word]); +} + +static inline uint64_t qb_attr_code_decode_64(const struct qb_attr_code *code, + const uint64_t *cacheline) +{ + return cacheline[code->word / 2]; +} + +/* encode a field to a cacheline */ +static inline void qb_attr_code_encode(const struct qb_attr_code *code, + uint32_t *cacheline, uint32_t val) +{ + cacheline[code->word] = + r32_uint32_t(code->lsoffset, code->width, cacheline[code->word]) + | e32_uint32_t(code->lsoffset, code->width, val); +} + +static inline void qb_attr_code_encode_64(const struct qb_attr_code *code, + uint64_t *cacheline, uint64_t val) +{ + cacheline[code->word / 2] = val; +} + +/* Small-width signed values (two's-complement) will decode into medium-width + * positives. (Eg. for an 8-bit signed field, which stores values from -128 to + * +127, a setting of -7 would appear to decode to the 32-bit unsigned value + * 249. Likewise -120 would decode as 136.) This function allows the caller to + * "re-sign" such fields to 32-bit signed. (Eg. -7, which was 249 with an 8-bit + * encoding, will become 0xfffffff9 if you cast the return value to uint32_t). + */ +static inline int32_t qb_attr_code_makesigned(const struct qb_attr_code *code, + uint32_t val) +{ + QBMAN_BUG_ON(val >= (1u << code->width)); + /* code->width should never exceed the width of val. If it does then a + * different function with larger val size must be used to translate + * from unsigned to signed + */ + QBMAN_BUG_ON(code->width > sizeof(val) * CHAR_BIT); + /* If the high bit was set, it was encoding a negative */ + if (val >= 1u << (code->width - 1)) + return (int32_t)0 - (int32_t)(((uint32_t)1 << code->width) - + val); + /* Otherwise, it was encoding a positive */ + return (int32_t)val; +} + +/* ---------------------- */ +/* Descriptors/cachelines */ +/* ---------------------- */ + +/* To avoid needless dynamic allocation, the driver API often gives the caller + * a "descriptor" type that the caller can instantiate however they like. + * Ultimately though, it is just a cacheline of binary storage (or something + * smaller when it is known that the descriptor doesn't need all 64 bytes) for + * holding pre-formatted pieces of hardware commands. The performance-critical + * code can then copy these descriptors directly into hardware command + * registers more efficiently than trying to construct/format commands + * on-the-fly. The API user sees the descriptor as an array of 32-bit words in + * order for the compiler to know its size, but the internal details are not + * exposed. The following macro is used within the driver for converting *any* + * descriptor pointer to a usable array pointer. The use of a macro (instead of + * an inline) is necessary to work with different descriptor types and to work + * correctly with const and non-const inputs (and similarly-qualified outputs). + */ +#define qb_cl(d) (&(d)->dont_manipulate_directly[0]) diff --git a/drivers/bus/fslmc/qbman/qbman_private.h b/drivers/bus/fslmc/qbman/qbman_private.h new file mode 100644 index 0000000000..292ec6a980 --- /dev/null +++ b/drivers/bus/fslmc/qbman/qbman_private.h @@ -0,0 +1,174 @@ +/*- + * BSD LICENSE + * + * Copyright (C) 2014-2016 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. + * + * 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. + */ + +/* Perform extra checking */ +#define QBMAN_CHECKING + +/* To maximise the amount of logic that is common between the Linux driver and + * other targets (such as the embedded MC firmware), we pivot here between the + * inclusion of two platform-specific headers. + * + * The first, qbman_sys_decl.h, includes any and all required system headers as + * well as providing any definitions for the purposes of compatibility. The + * second, qbman_sys.h, is where platform-specific routines go. + * + * The point of the split is that the platform-independent code (including this + * header) may depend on platform-specific declarations, yet other + * platform-specific routines may depend on platform-independent definitions. + */ + +#include "qbman_sys_decl.h" + +/* When things go wrong, it is a convenient trick to insert a few FOO() + * statements in the code to trace progress. TODO: remove this once we are + * hacking the code less actively. + */ +#define FOO() fsl_os_print("FOO: %s:%d\n", __FILE__, __LINE__) + +/* Any time there is a register interface which we poll on, this provides a + * "break after x iterations" scheme for it. It's handy for debugging, eg. + * where you don't want millions of lines of log output from a polling loop + * that won't, because such things tend to drown out the earlier log output + * that might explain what caused the problem. (NB: put ";" after each macro!) + * TODO: we should probably remove this once we're done sanitising the + * simulator... + */ +#define DBG_POLL_START(loopvar) (loopvar = 10) +#define DBG_POLL_CHECK(loopvar) \ +do { \ + if (!(loopvar--)) \ + QBMAN_BUG_ON(NULL == "DBG_POLL_CHECK"); \ +} while (0) + +/* For CCSR or portal-CINH registers that contain fields at arbitrary offsets + * and widths, these macro-generated encode/decode/isolate/remove inlines can + * be used. + * + * Eg. to "d"ecode a 14-bit field out of a register (into a "uint16_t" type), + * where the field is located 3 bits "up" from the least-significant bit of the + * register (ie. the field location within the 32-bit register corresponds to a + * mask of 0x0001fff8), you would do; + * uint16_t field = d32_uint16_t(3, 14, reg_value); + * + * Or to "e"ncode a 1-bit boolean value (input type is "int", zero is FALSE, + * non-zero is TRUE, so must convert all non-zero inputs to 1, hence the "!!" + * operator) into a register at bit location 0x00080000 (19 bits "in" from the + * LS bit), do; + * reg_value |= e32_int(19, 1, !!field); + * + * If you wish to read-modify-write a register, such that you leave the 14-bit + * field as-is but have all other fields set to zero, then "i"solate the 14-bit + * value using; + * reg_value = i32_uint16_t(3, 14, reg_value); + * + * Alternatively, you could "r"emove the 1-bit boolean field (setting it to + * zero) but leaving all other fields as-is; + * reg_val = r32_int(19, 1, reg_value); + * + */ +#ifdef __LP64__ +#define MAKE_MASK32(width) ((uint32_t)(( 1ULL << width) - 1)) +#else +#define MAKE_MASK32(width) (width == 32 ? 0xffffffff : \ + (uint32_t)((1 << width) - 1)) +#endif +#define DECLARE_CODEC32(t) \ +static inline uint32_t e32_##t(uint32_t lsoffset, uint32_t width, t val) \ +{ \ + QBMAN_BUG_ON(width > (sizeof(t) * 8)); \ + return ((uint32_t)val & MAKE_MASK32(width)) << lsoffset; \ +} \ +static inline t d32_##t(uint32_t lsoffset, uint32_t width, uint32_t val) \ +{ \ + QBMAN_BUG_ON(width > (sizeof(t) * 8)); \ + return (t)((val >> lsoffset) & MAKE_MASK32(width)); \ +} \ +static inline uint32_t i32_##t(uint32_t lsoffset, uint32_t width, \ + uint32_t val) \ +{ \ + QBMAN_BUG_ON(width > (sizeof(t) * 8)); \ + return e32_##t(lsoffset, width, d32_##t(lsoffset, width, val)); \ +} \ +static inline uint32_t r32_##t(uint32_t lsoffset, uint32_t width, \ + uint32_t val) \ +{ \ + QBMAN_BUG_ON(width > (sizeof(t) * 8)); \ + return ~(MAKE_MASK32(width) << lsoffset) & val; \ +} +DECLARE_CODEC32(uint32_t) +DECLARE_CODEC32(uint16_t) +DECLARE_CODEC32(uint8_t) +DECLARE_CODEC32(int) + + /*********************/ + /* Debugging assists */ + /*********************/ + +static inline void __hexdump(unsigned long start, unsigned long end, + unsigned long p, size_t sz, const unsigned char *c) +{ + while (start < end) { + unsigned int pos = 0; + char buf[64]; + int nl = 0; + + pos += sprintf(buf + pos, "%08lx: ", start); + do { + if ((start < p) || (start >= (p + sz))) + pos += sprintf(buf + pos, ".."); + else + pos += sprintf(buf + pos, "%02x", *(c++)); + if (!(++start & 15)) { + buf[pos++] = '\n'; + nl = 1; + } else { + nl = 0; + if (!(start & 1)) + buf[pos++] = ' '; + if (!(start & 3)) + buf[pos++] = ' '; + } + } while (start & 15); + if (!nl) + buf[pos++] = '\n'; + buf[pos] = '\0'; + pr_info("%s", buf); + } +} + +static inline void hexdump(const void *ptr, size_t sz) +{ + unsigned long p = (unsigned long)ptr; + unsigned long start = p & ~(unsigned long)15; + unsigned long end = (p + sz + 15) & ~(unsigned long)15; + const unsigned char *c = ptr; + + __hexdump(start, end, p, sz, c); +} + +#include "qbman_sys.h" diff --git a/drivers/bus/fslmc/qbman/qbman_sys.h b/drivers/bus/fslmc/qbman/qbman_sys.h new file mode 100644 index 0000000000..5dbcaa5744 --- /dev/null +++ b/drivers/bus/fslmc/qbman/qbman_sys.h @@ -0,0 +1,385 @@ +/*- + * BSD LICENSE + * + * Copyright (C) 2014-2016 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. + * + * 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. + */ +/* qbman_sys_decl.h and qbman_sys.h are the two platform-specific files in the + * driver. They are only included via qbman_private.h, which is itself a + * platform-independent file and is included by all the other driver source. + * + * qbman_sys_decl.h is included prior to all other declarations and logic, and + * it exists to provide compatibility with any linux interfaces our + * single-source driver code is dependent on (eg. kmalloc). Ie. this file + * provides linux compatibility. + * + * This qbman_sys.h header, on the other hand, is included *after* any common + * and platform-neutral declarations and logic in qbman_private.h, and exists to + * implement any platform-specific logic of the qbman driver itself. Ie. it is + * *not* to provide linux compatibility. + */ + +/* Trace the 3 different classes of read/write access to QBMan. #undef as + * required. + */ +#undef QBMAN_CCSR_TRACE +#undef QBMAN_CINH_TRACE +#undef QBMAN_CENA_TRACE + +static inline void word_copy(void *d, const void *s, unsigned int cnt) +{ + uint32_t *dd = d; + const uint32_t *ss = s; + + while (cnt--) + *(dd++) = *(ss++); +} + +/* Currently, the CENA support code expects each 32-bit word to be written in + * host order, and these are converted to hardware (little-endian) order on + * command submission. However, 64-bit quantities are must be written (and read) + * as two 32-bit words with the least-significant word first, irrespective of + * host endianness. + */ +static inline void u64_to_le32_copy(void *d, const uint64_t *s, + unsigned int cnt) +{ + uint32_t *dd = d; + const uint32_t *ss = (const uint32_t *)s; + + while (cnt--) { + /* TBD: the toolchain was choking on the use of 64-bit types up + * until recently so this works entirely with 32-bit variables. + * When 64-bit types become usable again, investigate better + * ways of doing this. + */ +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + *(dd++) = ss[1]; + *(dd++) = ss[0]; + ss += 2; +#else + *(dd++) = *(ss++); + *(dd++) = *(ss++); +#endif + } +} + +static inline void u64_from_le32_copy(uint64_t *d, const void *s, + unsigned int cnt) +{ + const uint32_t *ss = s; + uint32_t *dd = (uint32_t *)d; + + while (cnt--) { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + dd[1] = *(ss++); + dd[0] = *(ss++); + dd += 2; +#else + *(dd++) = *(ss++); + *(dd++) = *(ss++); +#endif + } +} + +/* Convert a host-native 32bit value into little endian */ +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ +static inline uint32_t make_le32(uint32_t val) +{ + return ((val & 0xff) << 24) | ((val & 0xff00) << 8) | + ((val & 0xff0000) >> 8) | ((val & 0xff000000) >> 24); +} + +static inline uint32_t make_le24(uint32_t val) +{ + return (((val & 0xff) << 16) | (val & 0xff00) | + ((val & 0xff0000) >> 16)); +} + +static inline void make_le32_n(uint32_t *val, unsigned int num) +{ + while (num--) { + *val = make_le32(*val); + val++; + } +} + +#else +#define make_le32(val) (val) +#define make_le24(val) (val) +#define make_le32_n(val, len) do {} while (0) +#endif + + /******************/ + /* Portal access */ + /******************/ +struct qbman_swp_sys { + /* On GPP, the sys support for qbman_swp is here. The CENA region isi + * not an mmap() of the real portal registers, but an allocated + * place-holder, because the actual writes/reads to/from the portal are + * marshalled from these allocated areas using QBMan's "MC access + * registers". CINH accesses are atomic so there's no need for a + * place-holder. + */ + uint8_t *cena; + uint8_t __iomem *addr_cena; + uint8_t __iomem *addr_cinh; + uint32_t idx; + enum qbman_eqcr_mode eqcr_mode; +}; + +/* P_OFFSET is (ACCESS_CMD,0,12) - offset within the portal + * C is (ACCESS_CMD,12,1) - is inhibited? (0==CENA, 1==CINH) + * SWP_IDX is (ACCESS_CMD,16,10) - Software portal index + * P is (ACCESS_CMD,28,1) - (0==special portal, 1==any portal) + * T is (ACCESS_CMD,29,1) - Command type (0==READ, 1==WRITE) + * E is (ACCESS_CMD,31,1) - Command execute (1 to issue, poll for 0==complete) + */ + +static inline void qbman_cinh_write(struct qbman_swp_sys *s, uint32_t offset, + uint32_t val) +{ + __raw_writel(val, s->addr_cinh + offset); +#ifdef QBMAN_CINH_TRACE + pr_info("qbman_cinh_write(%p:%d:0x%03x) 0x%08x\n", + s->addr_cinh, s->idx, offset, val); +#endif +} + +static inline uint32_t qbman_cinh_read(struct qbman_swp_sys *s, uint32_t offset) +{ + uint32_t reg = __raw_readl(s->addr_cinh + offset); +#ifdef QBMAN_CINH_TRACE + pr_info("qbman_cinh_read(%p:%d:0x%03x) 0x%08x\n", + s->addr_cinh, s->idx, offset, reg); +#endif + return reg; +} + +static inline void *qbman_cena_write_start(struct qbman_swp_sys *s, + uint32_t offset) +{ + void *shadow = s->cena + offset; + +#ifdef QBMAN_CENA_TRACE + pr_info("qbman_cena_write_start(%p:%d:0x%03x) %p\n", + s->addr_cena, s->idx, offset, shadow); +#endif + QBMAN_BUG_ON(offset & 63); + dcbz(shadow); + return shadow; +} + +static inline void *qbman_cena_write_start_wo_shadow(struct qbman_swp_sys *s, + uint32_t offset) +{ +#ifdef QBMAN_CENA_TRACE + pr_info("qbman_cena_write_start(%p:%d:0x%03x)\n", + s->addr_cena, s->idx, offset); +#endif + QBMAN_BUG_ON(offset & 63); + return (s->addr_cena + offset); +} + +static inline void qbman_cena_write_complete(struct qbman_swp_sys *s, + uint32_t offset, void *cmd) +{ + const uint32_t *shadow = cmd; + int loop; +#ifdef QBMAN_CENA_TRACE + pr_info("qbman_cena_write_complete(%p:%d:0x%03x) %p\n", + s->addr_cena, s->idx, offset, shadow); + hexdump(cmd, 64); +#endif + for (loop = 15; loop >= 1; loop--) + __raw_writel(shadow[loop], s->addr_cena + + offset + loop * 4); + lwsync(); + __raw_writel(shadow[0], s->addr_cena + offset); + dcbf(s->addr_cena + offset); +} + +static inline void qbman_cena_write_complete_wo_shadow(struct qbman_swp_sys *s, + uint32_t offset) +{ +#ifdef QBMAN_CENA_TRACE + pr_info("qbman_cena_write_complete(%p:%d:0x%03x)\n", + s->addr_cena, s->idx, offset); + hexdump(cmd, 64); +#endif + dcbf(s->addr_cena + offset); +} + +static inline uint32_t qbman_cena_read_reg(struct qbman_swp_sys *s, + uint32_t offset) +{ + return __raw_readl(s->addr_cena + offset); +} + +static inline void *qbman_cena_read(struct qbman_swp_sys *s, uint32_t offset) +{ + uint32_t *shadow = (uint32_t *)(s->cena + offset); + unsigned int loop; +#ifdef QBMAN_CENA_TRACE + pr_info("qbman_cena_read(%p:%d:0x%03x) %p\n", + s->addr_cena, s->idx, offset, shadow); +#endif + + for (loop = 0; loop < 16; loop++) + shadow[loop] = __raw_readl(s->addr_cena + offset + + loop * 4); +#ifdef QBMAN_CENA_TRACE + hexdump(shadow, 64); +#endif + return shadow; +} + +static inline void *qbman_cena_read_wo_shadow(struct qbman_swp_sys *s, + uint32_t offset) +{ +#ifdef QBMAN_CENA_TRACE + pr_info("qbman_cena_read(%p:%d:0x%03x) %p\n", + s->addr_cena, s->idx, offset, shadow); +#endif + +#ifdef QBMAN_CENA_TRACE + hexdump(shadow, 64); +#endif + return s->addr_cena + offset; +} + +static inline void qbman_cena_invalidate(struct qbman_swp_sys *s, + uint32_t offset) +{ + dccivac(s->addr_cena + offset); +} + +static inline void qbman_cena_invalidate_prefetch(struct qbman_swp_sys *s, + uint32_t offset) +{ + dccivac(s->addr_cena + offset); + prefetch_for_load(s->addr_cena + offset); +} + +static inline void qbman_cena_prefetch(struct qbman_swp_sys *s, + uint32_t offset) +{ + prefetch_for_load(s->addr_cena + offset); +} + + /******************/ + /* Portal support */ + /******************/ + +/* The SWP_CFG portal register is special, in that it is used by the + * platform-specific code rather than the platform-independent code in + * qbman_portal.c. So use of it is declared locally here. + */ +#define QBMAN_CINH_SWP_CFG 0xd00 + +/* For MC portal use, we always configure with + * DQRR_MF is (SWP_CFG,20,3) - DQRR max fill (<- 0x4) + * EST is (SWP_CFG,16,3) - EQCR_CI stashing threshold (<- 0x2) + * RPM is (SWP_CFG,12,2) - RCR production notification mode (<- 0x3) + * DCM is (SWP_CFG,10,2) - DQRR consumption notification mode (<- 0x2) + * EPM is (SWP_CFG,8,2) - EQCR production notification mode (<- 0x2) + * SD is (SWP_CFG,5,1) - memory stashing drop enable (<- TRUE) + * SP is (SWP_CFG,4,1) - memory stashing priority (<- TRUE) + * SE is (SWP_CFG,3,1) - memory stashing enable (<- TRUE) + * DP is (SWP_CFG,2,1) - dequeue stashing priority (<- TRUE) + * DE is (SWP_CFG,1,1) - dequeue stashing enable (<- TRUE) + * EP is (SWP_CFG,0,1) - EQCR_CI stashing priority (<- TRUE) + */ +static inline uint32_t qbman_set_swp_cfg(uint8_t max_fill, uint8_t wn, + uint8_t est, uint8_t rpm, uint8_t dcm, + uint8_t epm, int sd, int sp, int se, + int dp, int de, int ep) +{ + uint32_t reg; + + reg = e32_uint8_t(20, (uint32_t)(3 + (max_fill >> 3)), max_fill) | + e32_uint8_t(16, 3, est) | + e32_uint8_t(12, 2, rpm) | e32_uint8_t(10, 2, dcm) | + e32_uint8_t(8, 2, epm) | e32_int(5, 1, sd) | + e32_int(4, 1, sp) | e32_int(3, 1, se) | e32_int(2, 1, dp) | + e32_int(1, 1, de) | e32_int(0, 1, ep) | e32_uint8_t(14, 1, wn); + return reg; +} + +static inline int qbman_swp_sys_init(struct qbman_swp_sys *s, + const struct qbman_swp_desc *d, + uint8_t dqrr_size) +{ + uint32_t reg; + + s->addr_cena = d->cena_bar; + s->addr_cinh = d->cinh_bar; + s->idx = (uint32_t)d->idx; + s->cena = (void *)get_zeroed_page(GFP_KERNEL); + if (!s->cena) { + pr_err("Could not allocate page for cena shadow\n"); + return -1; + } + s->eqcr_mode = d->eqcr_mode; + QBMAN_BUG_ON(d->idx < 0); +#ifdef QBMAN_CHECKING + /* We should never be asked to initialise for a portal that isn't in + * the power-on state. (Ie. don't forget to reset portals when they are + * decommissioned!) + */ + reg = qbman_cinh_read(s, QBMAN_CINH_SWP_CFG); + QBMAN_BUG_ON(reg); +#endif + if (s->eqcr_mode == qman_eqcr_vb_array) + reg = qbman_set_swp_cfg(dqrr_size, 0, 0, 3, 2, 3, 1, 1, 1, 1, + 1, 1); + else + reg = qbman_set_swp_cfg(dqrr_size, 0, 2, 3, 2, 2, 1, 1, 1, 1, + 1, 1); + qbman_cinh_write(s, QBMAN_CINH_SWP_CFG, reg); + reg = qbman_cinh_read(s, QBMAN_CINH_SWP_CFG); + if (!reg) { + pr_err("The portal %d is not enabled!\n", s->idx); + kfree(s->cena); + return -1; + } + return 0; +} + +static inline void qbman_swp_sys_finish(struct qbman_swp_sys *s) +{ + free_page((unsigned long)s->cena); +} + +static inline void * +qbman_cena_write_start_wo_shadow_fast(struct qbman_swp_sys *s, + uint32_t offset) +{ +#ifdef QBMAN_CENA_TRACE + pr_info("qbman_cena_write_start(%p:%d:0x%03x)\n", + s->addr_cena, s->idx, offset); +#endif + QBMAN_BUG_ON(offset & 63); + return (s->addr_cena + offset); +} diff --git a/drivers/bus/fslmc/qbman/qbman_sys_decl.h b/drivers/bus/fslmc/qbman/qbman_sys_decl.h new file mode 100644 index 0000000000..e52f5ed2e3 --- /dev/null +++ b/drivers/bus/fslmc/qbman/qbman_sys_decl.h @@ -0,0 +1,73 @@ +/*- + * BSD LICENSE + * + * Copyright (C) 2014-2016 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. + * + * 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 +#include + +/* Sanity check */ +#if (__BYTE_ORDER__ != __ORDER_BIG_ENDIAN__) && \ + (__BYTE_ORDER__ != __ORDER_LITTLE_ENDIAN__) +#error "Unknown endianness!" +#endif + +/* The platform-independent code shouldn't need endianness, except for + * weird/fast-path cases like qbman_result_has_token(), which needs to + * perform a passive and endianness-specific test on a read-only data structure + * very quickly. It's an exception, and this symbol is used for that case. + */ +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ +#define DQRR_TOK_OFFSET 0 +#define QBMAN_RESULT_VERB_OFFSET_IN_MEM 24 +#define SCN_STATE_OFFSET_IN_MEM 8 +#define SCN_RID_OFFSET_IN_MEM 8 +#else +#define DQRR_TOK_OFFSET 24 +#define QBMAN_RESULT_VERB_OFFSET_IN_MEM 0 +#define SCN_STATE_OFFSET_IN_MEM 16 +#define SCN_RID_OFFSET_IN_MEM 0 +#endif + +/* Similarly-named functions */ +#define upper32(a) upper_32_bits(a) +#define lower32(a) lower_32_bits(a) + + /****************/ + /* arch assists */ + /****************/ +#define dcbz(p) { asm volatile("dc zva, %0" : : "r" (p) : "memory"); } +#define lwsync() { asm volatile("dmb st" : : : "memory"); } +#define dcbf(p) { asm volatile("dc cvac, %0" : : "r"(p) : "memory"); } +#define dccivac(p) { asm volatile("dc civac, %0" : : "r"(p) : "memory"); } +static inline void prefetch_for_load(void *p) +{ + asm volatile("prfm pldl1keep, [%0, #64]" : : "r" (p)); +} + +static inline void prefetch_for_store(void *p) +{ + asm volatile("prfm pstl1keep, [%0, #64]" : : "r" (p)); +} diff --git a/drivers/bus/fslmc/rte_bus_fslmc_version.map b/drivers/bus/fslmc/rte_bus_fslmc_version.map index 41c80d9f2d..95c1804b57 100644 --- a/drivers/bus/fslmc/rte_bus_fslmc_version.map +++ b/drivers/bus/fslmc/rte_bus_fslmc_version.map @@ -1,6 +1,25 @@ DPDK_17.05 { global: + qbman_check_command_complete; + qbman_eq_desc_clear; + qbman_eq_desc_set_no_orp; + qbman_eq_desc_set_qd; + qbman_eq_desc_set_response; + qbman_get_version; + qbman_pull_desc_clear; + qbman_pull_desc_set_fq; + qbman_pull_desc_set_numframes; + qbman_pull_desc_set_storage; + qbman_release_desc_clear; + qbman_release_desc_set_bpid; + qbman_result_DQ_fd; + qbman_result_DQ_flags; + qbman_result_has_new_result; + qbman_swp_acquire; + qbman_swp_pull; + qbman_swp_release; + qbman_swp_send_multiple; rte_fslmc_driver_register; rte_fslmc_driver_unregister; -- 2.20.1