diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2016-10-14 11:07:42 -0700 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2016-10-14 11:07:42 -0700 |
commit | d8bfb96a2e3c42f91a9944a19638c043cb86dd8f (patch) | |
tree | 58dfdccc4f15a6f35e0875d21373e303f13553a4 /include | |
parent | 11ed3e0ef354cdc9c4577a187fa60a053137b507 (diff) | |
parent | 08bf75ba852ef8304a84b6a030466b4b4850382e (diff) |
Merge tag 'powerpc-4.9-2' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux
Pull more powerpc updates from Michael Ellerman:
"Some more powerpc updates for 4.9:
Freescale updates from Scott Wood:
- qbman support (a prerequisite for datapath drivers such as ethernet)
- a PCI DMA fix+improvement
- reset handler changes
- more 8xx optimizations
- some cleanups and fixes.'
Fixes:
- selftests/powerpc: Add missing binaries to .gitignores (Michael Ellerman)
- selftests/powerpc: Fix build break caused by EXPORT_SYMBOL changes (Michael Ellerman)
- powerpc/pseries: Fix stack corruption in htpe code (Laurent Dufour)
- powerpc/64s: Fix power4_fixup_nap placement (Nicholas Piggin)
- powerpc/64: Fix incorrect return value from __copy_tofrom_user (Paul Mackerras)
- powerpc/mm/hash64: Fix might_have_hea() check (Michael Ellerman)
Other:
- MAINTAINERS: Remove myself from PA Semi entries (Olof Johansson)
- MAINTAINERS: Drop separate pseries entry (Michael Ellerman)
- MAINTAINERS: Update powerpc website & add selftests (Michael Ellerman):
* tag 'powerpc-4.9-2' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux: (35 commits)
powerpc/mm/hash64: Fix might_have_hea() check
powerpc/64: Fix incorrect return value from __copy_tofrom_user
powerpc/64s: Fix power4_fixup_nap placement
powerpc/pseries: Fix stack corruption in htpe code
selftests/powerpc: Fix build break caused by EXPORT_SYMBOL changes
MAINTAINERS: Update powerpc website & add selftests
MAINTAINERS: Drop separate pseries entry
MAINTAINERS: Remove myself from PA Semi entries
selftests/powerpc: Add missing binaries to .gitignores
arch/powerpc: Add CONFIG_FSL_DPAA to corenetXX_smp_defconfig
soc/qman: Add self-test for QMan driver
soc/bman: Add self-test for BMan driver
soc/fsl: Introduce DPAA 1.x QMan device driver
soc/fsl: Introduce DPAA 1.x BMan device driver
powerpc/8xx: make user addr DTLB miss the short path
powerpc/8xx: Move additional DTLBMiss handlers out of exception area
powerpc/8xx: use r3 to scratch CR in ITLBmiss
soc/fsl/qe: fix gpio save_regs functions
powerpc/8xx: add dedicated machine check handler
powerpc/8xx: add system_reset_exception
...
Diffstat (limited to 'include')
-rw-r--r-- | include/soc/fsl/bman.h | 129 | ||||
-rw-r--r-- | include/soc/fsl/qman.h | 1074 |
2 files changed, 1203 insertions, 0 deletions
diff --git a/include/soc/fsl/bman.h b/include/soc/fsl/bman.h new file mode 100644 index 000000000000..eaaf56df4086 --- /dev/null +++ b/include/soc/fsl/bman.h @@ -0,0 +1,129 @@ +/* Copyright 2008 - 2016 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef __FSL_BMAN_H +#define __FSL_BMAN_H + +/* wrapper for 48-bit buffers */ +struct bm_buffer { + union { + struct { + __be16 bpid; /* hi 8-bits reserved */ + __be16 hi; /* High 16-bits of 48-bit address */ + __be32 lo; /* Low 32-bits of 48-bit address */ + }; + __be64 data; + }; +} __aligned(8); +/* + * Restore the 48 bit address previously stored in BMan + * hardware pools as a dma_addr_t + */ +static inline dma_addr_t bm_buf_addr(const struct bm_buffer *buf) +{ + return be64_to_cpu(buf->data) & 0xffffffffffffLLU; +} + +static inline u64 bm_buffer_get64(const struct bm_buffer *buf) +{ + return be64_to_cpu(buf->data) & 0xffffffffffffLLU; +} + +static inline void bm_buffer_set64(struct bm_buffer *buf, u64 addr) +{ + buf->hi = cpu_to_be16(upper_32_bits(addr)); + buf->lo = cpu_to_be32(lower_32_bits(addr)); +} + +static inline u8 bm_buffer_get_bpid(const struct bm_buffer *buf) +{ + return be16_to_cpu(buf->bpid) & 0xff; +} + +static inline void bm_buffer_set_bpid(struct bm_buffer *buf, int bpid) +{ + buf->bpid = cpu_to_be16(bpid & 0xff); +} + +/* Managed portal, high-level i/face */ + +/* Portal and Buffer Pools */ +struct bman_portal; +struct bman_pool; + +#define BM_POOL_MAX 64 /* max # of buffer pools */ + +/** + * bman_new_pool - Allocates a Buffer Pool object + * + * Creates a pool object, and returns a reference to it or NULL on error. + */ +struct bman_pool *bman_new_pool(void); + +/** + * bman_free_pool - Deallocates a Buffer Pool object + * @pool: the pool object to release + */ +void bman_free_pool(struct bman_pool *pool); + +/** + * bman_get_bpid - Returns a pool object's BPID. + * @pool: the pool object + * + * The returned value is the index of the encapsulated buffer pool, + * in the range of [0, @BM_POOL_MAX-1]. + */ +int bman_get_bpid(const struct bman_pool *pool); + +/** + * bman_release - Release buffer(s) to the buffer pool + * @pool: the buffer pool object to release to + * @bufs: an array of buffers to release + * @num: the number of buffers in @bufs (1-8) + * + * Adds the given buffers to RCR entries. If the RCR ring is unresponsive, + * the function will return -ETIMEDOUT. Otherwise, it returns zero. + */ +int bman_release(struct bman_pool *pool, const struct bm_buffer *bufs, u8 num); + +/** + * bman_acquire - Acquire buffer(s) from a buffer pool + * @pool: the buffer pool object to acquire from + * @bufs: array for storing the acquired buffers + * @num: the number of buffers desired (@bufs is at least this big) + * + * Issues an "Acquire" command via the portal's management command interface. + * The return value will be the number of buffers obtained from the pool, or a + * negative error code if a h/w error or pool starvation was encountered. In + * the latter case, the content of @bufs is undefined. + */ +int bman_acquire(struct bman_pool *pool, struct bm_buffer *bufs, u8 num); + +#endif /* __FSL_BMAN_H */ diff --git a/include/soc/fsl/qman.h b/include/soc/fsl/qman.h new file mode 100644 index 000000000000..37f3eb001a16 --- /dev/null +++ b/include/soc/fsl/qman.h @@ -0,0 +1,1074 @@ +/* Copyright 2008 - 2016 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef __FSL_QMAN_H +#define __FSL_QMAN_H + +#include <linux/bitops.h> + +/* Hardware constants */ +#define QM_CHANNEL_SWPORTAL0 0 +#define QMAN_CHANNEL_POOL1 0x21 +#define QMAN_CHANNEL_POOL1_REV3 0x401 +extern u16 qm_channel_pool1; + +/* Portal processing (interrupt) sources */ +#define QM_PIRQ_CSCI 0x00100000 /* Congestion State Change */ +#define QM_PIRQ_EQCI 0x00080000 /* Enqueue Command Committed */ +#define QM_PIRQ_EQRI 0x00040000 /* EQCR Ring (below threshold) */ +#define QM_PIRQ_DQRI 0x00020000 /* DQRR Ring (non-empty) */ +#define QM_PIRQ_MRI 0x00010000 /* MR Ring (non-empty) */ +/* + * This mask contains all the interrupt sources that need handling except DQRI, + * ie. that if present should trigger slow-path processing. + */ +#define QM_PIRQ_SLOW (QM_PIRQ_CSCI | QM_PIRQ_EQCI | QM_PIRQ_EQRI | \ + QM_PIRQ_MRI) + +/* For qman_static_dequeue_*** APIs */ +#define QM_SDQCR_CHANNELS_POOL_MASK 0x00007fff +/* for n in [1,15] */ +#define QM_SDQCR_CHANNELS_POOL(n) (0x00008000 >> (n)) +/* for conversion from n of qm_channel */ +static inline u32 QM_SDQCR_CHANNELS_POOL_CONV(u16 channel) +{ + return QM_SDQCR_CHANNELS_POOL(channel + 1 - qm_channel_pool1); +} + +/* --- QMan data structures (and associated constants) --- */ + +/* "Frame Descriptor (FD)" */ +struct qm_fd { + union { + struct { + u8 cfg8b_w1; + u8 bpid; /* Buffer Pool ID */ + u8 cfg8b_w3; + u8 addr_hi; /* high 8-bits of 40-bit address */ + __be32 addr_lo; /* low 32-bits of 40-bit address */ + } __packed; + __be64 data; + }; + __be32 cfg; /* format, offset, length / congestion */ + union { + __be32 cmd; + __be32 status; + }; +} __aligned(8); + +#define QM_FD_FORMAT_SG BIT(31) +#define QM_FD_FORMAT_LONG BIT(30) +#define QM_FD_FORMAT_COMPOUND BIT(29) +#define QM_FD_FORMAT_MASK GENMASK(31, 29) +#define QM_FD_OFF_SHIFT 20 +#define QM_FD_OFF_MASK GENMASK(28, 20) +#define QM_FD_LEN_MASK GENMASK(19, 0) +#define QM_FD_LEN_BIG_MASK GENMASK(28, 0) + +enum qm_fd_format { + /* + * 'contig' implies a contiguous buffer, whereas 'sg' implies a + * scatter-gather table. 'big' implies a 29-bit length with no offset + * field, otherwise length is 20-bit and offset is 9-bit. 'compound' + * implies a s/g-like table, where each entry itself represents a frame + * (contiguous or scatter-gather) and the 29-bit "length" is + * interpreted purely for congestion calculations, ie. a "congestion + * weight". + */ + qm_fd_contig = 0, + qm_fd_contig_big = QM_FD_FORMAT_LONG, + qm_fd_sg = QM_FD_FORMAT_SG, + qm_fd_sg_big = QM_FD_FORMAT_SG | QM_FD_FORMAT_LONG, + qm_fd_compound = QM_FD_FORMAT_COMPOUND +}; + +static inline dma_addr_t qm_fd_addr(const struct qm_fd *fd) +{ + return be64_to_cpu(fd->data) & 0xffffffffffLLU; +} + +static inline u64 qm_fd_addr_get64(const struct qm_fd *fd) +{ + return be64_to_cpu(fd->data) & 0xffffffffffLLU; +} + +static inline void qm_fd_addr_set64(struct qm_fd *fd, u64 addr) +{ + fd->addr_hi = upper_32_bits(addr); + fd->addr_lo = cpu_to_be32(lower_32_bits(addr)); +} + +/* + * The 'format' field indicates the interpretation of the remaining + * 29 bits of the 32-bit word. + * If 'format' is _contig or _sg, 20b length and 9b offset. + * If 'format' is _contig_big or _sg_big, 29b length. + * If 'format' is _compound, 29b "congestion weight". + */ +static inline enum qm_fd_format qm_fd_get_format(const struct qm_fd *fd) +{ + return be32_to_cpu(fd->cfg) & QM_FD_FORMAT_MASK; +} + +static inline int qm_fd_get_offset(const struct qm_fd *fd) +{ + return (be32_to_cpu(fd->cfg) & QM_FD_OFF_MASK) >> QM_FD_OFF_SHIFT; +} + +static inline int qm_fd_get_length(const struct qm_fd *fd) +{ + return be32_to_cpu(fd->cfg) & QM_FD_LEN_MASK; +} + +static inline int qm_fd_get_len_big(const struct qm_fd *fd) +{ + return be32_to_cpu(fd->cfg) & QM_FD_LEN_BIG_MASK; +} + +static inline void qm_fd_set_param(struct qm_fd *fd, enum qm_fd_format fmt, + int off, int len) +{ + fd->cfg = cpu_to_be32(fmt | (len & QM_FD_LEN_BIG_MASK) | + ((off << QM_FD_OFF_SHIFT) & QM_FD_OFF_MASK)); +} + +#define qm_fd_set_contig(fd, off, len) \ + qm_fd_set_param(fd, qm_fd_contig, off, len) +#define qm_fd_set_sg(fd, off, len) qm_fd_set_param(fd, qm_fd_sg, off, len) +#define qm_fd_set_contig_big(fd, len) \ + qm_fd_set_param(fd, qm_fd_contig_big, 0, len) +#define qm_fd_set_sg_big(fd, len) qm_fd_set_param(fd, qm_fd_sg_big, 0, len) + +static inline void qm_fd_clear_fd(struct qm_fd *fd) +{ + fd->data = 0; + fd->cfg = 0; + fd->cmd = 0; +} + +/* Scatter/Gather table entry */ +struct qm_sg_entry { + union { + struct { + u8 __reserved1[3]; + u8 addr_hi; /* high 8-bits of 40-bit address */ + __be32 addr_lo; /* low 32-bits of 40-bit address */ + }; + __be64 data; + }; + __be32 cfg; /* E bit, F bit, length */ + u8 __reserved2; + u8 bpid; + __be16 offset; /* 13-bit, _res[13-15]*/ +} __packed; + +#define QM_SG_LEN_MASK GENMASK(29, 0) +#define QM_SG_OFF_MASK GENMASK(12, 0) +#define QM_SG_FIN BIT(30) +#define QM_SG_EXT BIT(31) + +static inline dma_addr_t qm_sg_addr(const struct qm_sg_entry *sg) +{ + return be64_to_cpu(sg->data) & 0xffffffffffLLU; +} + +static inline u64 qm_sg_entry_get64(const struct qm_sg_entry *sg) +{ + return be64_to_cpu(sg->data) & 0xffffffffffLLU; +} + +static inline void qm_sg_entry_set64(struct qm_sg_entry *sg, u64 addr) +{ + sg->addr_hi = upper_32_bits(addr); + sg->addr_lo = cpu_to_be32(lower_32_bits(addr)); +} + +static inline bool qm_sg_entry_is_final(const struct qm_sg_entry *sg) +{ + return be32_to_cpu(sg->cfg) & QM_SG_FIN; +} + +static inline bool qm_sg_entry_is_ext(const struct qm_sg_entry *sg) +{ + return be32_to_cpu(sg->cfg) & QM_SG_EXT; +} + +static inline int qm_sg_entry_get_len(const struct qm_sg_entry *sg) +{ + return be32_to_cpu(sg->cfg) & QM_SG_LEN_MASK; +} + +static inline void qm_sg_entry_set_len(struct qm_sg_entry *sg, int len) +{ + sg->cfg = cpu_to_be32(len & QM_SG_LEN_MASK); +} + +static inline void qm_sg_entry_set_f(struct qm_sg_entry *sg, int len) +{ + sg->cfg = cpu_to_be32(QM_SG_FIN | (len & QM_SG_LEN_MASK)); +} + +static inline int qm_sg_entry_get_off(const struct qm_sg_entry *sg) +{ + return be32_to_cpu(sg->offset) & QM_SG_OFF_MASK; +} + +/* "Frame Dequeue Response" */ +struct qm_dqrr_entry { + u8 verb; + u8 stat; + u16 seqnum; /* 15-bit */ + u8 tok; + u8 __reserved2[3]; + u32 fqid; /* 24-bit */ + u32 contextB; + struct qm_fd fd; + u8 __reserved4[32]; +} __packed; +#define QM_DQRR_VERB_VBIT 0x80 +#define QM_DQRR_VERB_MASK 0x7f /* where the verb contains; */ +#define QM_DQRR_VERB_FRAME_DEQUEUE 0x60 /* "this format" */ +#define QM_DQRR_STAT_FQ_EMPTY 0x80 /* FQ empty */ +#define QM_DQRR_STAT_FQ_HELDACTIVE 0x40 /* FQ held active */ +#define QM_DQRR_STAT_FQ_FORCEELIGIBLE 0x20 /* FQ was force-eligible'd */ +#define QM_DQRR_STAT_FD_VALID 0x10 /* has a non-NULL FD */ +#define QM_DQRR_STAT_UNSCHEDULED 0x02 /* Unscheduled dequeue */ +#define QM_DQRR_STAT_DQCR_EXPIRED 0x01 /* VDQCR or PDQCR expired*/ + +/* "ERN Message Response" */ +/* "FQ State Change Notification" */ +union qm_mr_entry { + struct { + u8 verb; + u8 __reserved[63]; + }; + struct { + u8 verb; + u8 dca; + u16 seqnum; + u8 rc; /* Rej Code: 8-bit */ + u8 orp_hi; /* ORP: 24-bit */ + u16 orp_lo; + u32 fqid; /* 24-bit */ + u32 tag; + struct qm_fd fd; + u8 __reserved1[32]; + } __packed ern; + struct { + u8 verb; + u8 fqs; /* Frame Queue Status */ + u8 __reserved1[6]; + u32 fqid; /* 24-bit */ + u32 contextB; + u8 __reserved2[48]; + } __packed fq; /* FQRN/FQRNI/FQRL/FQPN */ +}; +#define QM_MR_VERB_VBIT 0x80 +/* + * ERNs originating from direct-connect portals ("dcern") use 0x20 as a verb + * which would be invalid as a s/w enqueue verb. A s/w ERN can be distinguished + * from the other MR types by noting if the 0x20 bit is unset. + */ +#define QM_MR_VERB_TYPE_MASK 0x27 +#define QM_MR_VERB_DC_ERN 0x20 +#define QM_MR_VERB_FQRN 0x21 +#define QM_MR_VERB_FQRNI 0x22 +#define QM_MR_VERB_FQRL 0x23 +#define QM_MR_VERB_FQPN 0x24 +#define QM_MR_RC_MASK 0xf0 /* contains one of; */ +#define QM_MR_RC_CGR_TAILDROP 0x00 +#define QM_MR_RC_WRED 0x10 +#define QM_MR_RC_ERROR 0x20 +#define QM_MR_RC_ORPWINDOW_EARLY 0x30 +#define QM_MR_RC_ORPWINDOW_LATE 0x40 +#define QM_MR_RC_FQ_TAILDROP 0x50 +#define QM_MR_RC_ORPWINDOW_RETIRED 0x60 +#define QM_MR_RC_ORP_ZERO 0x70 +#define QM_MR_FQS_ORLPRESENT 0x02 /* ORL fragments to come */ +#define QM_MR_FQS_NOTEMPTY 0x01 /* FQ has enqueued frames */ + +/* + * An identical structure of FQD fields is present in the "Init FQ" command and + * the "Query FQ" result, it's suctioned out into the "struct qm_fqd" type. + * Within that, the 'stashing' and 'taildrop' pieces are also factored out, the + * latter has two inlines to assist with converting to/from the mant+exp + * representation. + */ +struct qm_fqd_stashing { + /* See QM_STASHING_EXCL_<...> */ + u8 exclusive; + /* Numbers of cachelines */ + u8 cl; /* _res[6-7], as[4-5], ds[2-3], cs[0-1] */ +}; + +struct qm_fqd_oac { + /* "Overhead Accounting Control", see QM_OAC_<...> */ + u8 oac; /* oac[6-7], _res[0-5] */ + /* Two's-complement value (-128 to +127) */ + s8 oal; /* "Overhead Accounting Length" */ +}; + +struct qm_fqd { + /* _res[6-7], orprws[3-5], oa[2], olws[0-1] */ + u8 orpc; + u8 cgid; + __be16 fq_ctrl; /* See QM_FQCTRL_<...> */ + __be16 dest_wq; /* channel[3-15], wq[0-2] */ + __be16 ics_cred; /* 15-bit */ + /* + * For "Initialize Frame Queue" commands, the write-enable mask + * determines whether 'td' or 'oac_init' is observed. For query + * commands, this field is always 'td', and 'oac_query' (below) reflects + * the Overhead ACcounting values. + */ + union { + __be16 td; /* "Taildrop": _res[13-15], mant[5-12], exp[0-4] */ + struct qm_fqd_oac oac_init; + }; + __be32 context_b; + union { + /* Treat it as 64-bit opaque */ + __be64 opaque; + struct { + __be32 hi; + __be32 lo; + }; + /* Treat it as s/w portal stashing config */ + /* see "FQD Context_A field used for [...]" */ + struct { + struct qm_fqd_stashing stashing; + /* + * 48-bit address of FQ context to + * stash, must be cacheline-aligned + */ + __be16 context_hi; + __be32 context_lo; + } __packed; + } context_a; + struct qm_fqd_oac oac_query; +} __packed; + +#define QM_FQD_CHAN_OFF 3 +#define QM_FQD_WQ_MASK GENMASK(2, 0) +#define QM_FQD_TD_EXP_MASK GENMASK(4, 0) +#define QM_FQD_TD_MANT_OFF 5 +#define QM_FQD_TD_MANT_MASK GENMASK(12, 5) +#define QM_FQD_TD_MAX 0xe0000000 +#define QM_FQD_TD_MANT_MAX 0xff +#define QM_FQD_OAC_OFF 6 +#define QM_FQD_AS_OFF 4 +#define QM_FQD_DS_OFF 2 +#define QM_FQD_XS_MASK 0x3 + +/* 64-bit converters for context_hi/lo */ +static inline u64 qm_fqd_stashing_get64(const struct qm_fqd *fqd) +{ + return be64_to_cpu(fqd->context_a.opaque) & 0xffffffffffffULL; +} + +static inline dma_addr_t qm_fqd_stashing_addr(const struct qm_fqd *fqd) +{ + return be64_to_cpu(fqd->context_a.opaque) & 0xffffffffffffULL; +} + +static inline u64 qm_fqd_context_a_get64(const struct qm_fqd *fqd) +{ + return qm_fqd_stashing_get64(fqd); +} + +static inline void qm_fqd_stashing_set64(struct qm_fqd *fqd, u64 addr) +{ + fqd->context_a.context_hi = upper_32_bits(addr); + fqd->context_a.context_lo = lower_32_bits(addr); +} + +static inline void qm_fqd_context_a_set64(struct qm_fqd *fqd, u64 addr) +{ + fqd->context_a.hi = cpu_to_be16(upper_32_bits(addr)); + fqd->context_a.lo = cpu_to_be32(lower_32_bits(addr)); +} + +/* convert a threshold value into mant+exp representation */ +static inline int qm_fqd_set_taildrop(struct qm_fqd *fqd, u32 val, + int roundup) +{ + u32 e = 0; + int td, oddbit = 0; + + if (val > QM_FQD_TD_MAX) + return -ERANGE; + + while (val > QM_FQD_TD_MANT_MAX) { + oddbit = val & 1; + val >>= 1; + e++; + if (roundup && oddbit) + val++; + } + + td = (val << QM_FQD_TD_MANT_OFF) & QM_FQD_TD_MANT_MASK; + td |= (e & QM_FQD_TD_EXP_MASK); + fqd->td = cpu_to_be16(td); + return 0; +} +/* and the other direction */ +static inline int qm_fqd_get_taildrop(const struct qm_fqd *fqd) +{ + int td = be16_to_cpu(fqd->td); + + return ((td & QM_FQD_TD_MANT_MASK) >> QM_FQD_TD_MANT_OFF) + << (td & QM_FQD_TD_EXP_MASK); +} + +static inline void qm_fqd_set_stashing(struct qm_fqd *fqd, u8 as, u8 ds, u8 cs) +{ + struct qm_fqd_stashing *st = &fqd->context_a.stashing; + + st->cl = ((as & QM_FQD_XS_MASK) << QM_FQD_AS_OFF) | + ((ds & QM_FQD_XS_MASK) << QM_FQD_DS_OFF) | + (cs & QM_FQD_XS_MASK); +} + +static inline u8 qm_fqd_get_stashing(const struct qm_fqd *fqd) +{ + return fqd->context_a.stashing.cl; +} + +static inline void qm_fqd_set_oac(struct qm_fqd *fqd, u8 val) +{ + fqd->oac_init.oac = val << QM_FQD_OAC_OFF; +} + +static inline void qm_fqd_set_oal(struct qm_fqd *fqd, s8 val) +{ + fqd->oac_init.oal = val; +} + +static inline void qm_fqd_set_destwq(struct qm_fqd *fqd, int ch, int wq) +{ + fqd->dest_wq = cpu_to_be16((ch << QM_FQD_CHAN_OFF) | + (wq & QM_FQD_WQ_MASK)); +} + +static inline int qm_fqd_get_chan(const struct qm_fqd *fqd) +{ + return be16_to_cpu(fqd->dest_wq) >> QM_FQD_CHAN_OFF; +} + +static inline int qm_fqd_get_wq(const struct qm_fqd *fqd) +{ + return be16_to_cpu(fqd->dest_wq) & QM_FQD_WQ_MASK; +} + +/* See "Frame Queue Descriptor (FQD)" */ +/* Frame Queue Descriptor (FQD) field 'fq_ctrl' uses these constants */ +#define QM_FQCTRL_MASK 0x07ff /* 'fq_ctrl' flags; */ +#define QM_FQCTRL_CGE 0x0400 /* Congestion Group Enable */ +#define QM_FQCTRL_TDE 0x0200 /* Tail-Drop Enable */ +#define QM_FQCTRL_CTXASTASHING 0x0080 /* Context-A stashing */ +#define QM_FQCTRL_CPCSTASH 0x0040 /* CPC Stash Enable */ +#define QM_FQCTRL_FORCESFDR 0x0008 /* High-priority SFDRs */ +#define QM_FQCTRL_AVOIDBLOCK 0x0004 /* Don't block active */ +#define QM_FQCTRL_HOLDACTIVE 0x0002 /* Hold active in portal */ +#define QM_FQCTRL_PREFERINCACHE 0x0001 /* Aggressively cache FQD */ +#define QM_FQCTRL_LOCKINCACHE QM_FQCTRL_PREFERINCACHE /* older naming */ + +/* See "FQD Context_A field used for [...] */ +/* Frame Queue Descriptor (FQD) field 'CONTEXT_A' uses these constants */ +#define QM_STASHING_EXCL_ANNOTATION 0x04 +#define QM_STASHING_EXCL_DATA 0x02 +#define QM_STASHING_EXCL_CTX 0x01 + +/* See "Intra Class Scheduling" */ +/* FQD field 'OAC' (Overhead ACcounting) uses these constants */ +#define QM_OAC_ICS 0x2 /* Accounting for Intra-Class Scheduling */ +#define QM_OAC_CG 0x1 /* Accounting for Congestion Groups */ + +/* + * This struct represents the 32-bit "WR_PARM_[GYR]" parameters in CGR fields + * and associated commands/responses. The WRED parameters are calculated from + * these fields as follows; + * MaxTH = MA * (2 ^ Mn) + * Slope = SA / (2 ^ Sn) + * MaxP = 4 * (Pn + 1) + */ +struct qm_cgr_wr_parm { + /* MA[24-31], Mn[19-23], SA[12-18], Sn[6-11], Pn[0-5] */ + u32 word; +}; +/* + * This struct represents the 13-bit "CS_THRES" CGR field. In the corresponding + * management commands, this is padded to a 16-bit structure field, so that's + * how we represent it here. The congestion state threshold is calculated from + * these fields as follows; + * CS threshold = TA * (2 ^ Tn) + */ +struct qm_cgr_cs_thres { + /* _res[13-15], TA[5-12], Tn[0-4] */ + u16 word; +}; +/* + * This identical structure of CGR fields is present in the "Init/Modify CGR" + * commands and the "Query CGR" result. It's suctioned out here into its own + * struct. + */ +struct __qm_mc_cgr { + struct qm_cgr_wr_parm wr_parm_g; + struct qm_cgr_wr_parm wr_parm_y; + struct qm_cgr_wr_parm wr_parm_r; + u8 wr_en_g; /* boolean, use QM_CGR_EN */ + u8 wr_en_y; /* boolean, use QM_CGR_EN */ + u8 wr_en_r; /* boolean, use QM_CGR_EN */ + u8 cscn_en; /* boolean, use QM_CGR_EN */ + union { + struct { + u16 cscn_targ_upd_ctrl; /* use QM_CSCN_TARG_UDP_ */ + u16 cscn_targ_dcp_low; /* CSCN_TARG_DCP low-16bits */ + }; + u32 cscn_targ; /* use QM_CGR_TARG_* */ + }; + u8 cstd_en; /* boolean, use QM_CGR_EN */ + u8 cs; /* boolean, only used in query response */ + struct qm_cgr_cs_thres cs_thres; /* use qm_cgr_cs_thres_set64() */ + u8 mode; /* QMAN_CGR_MODE_FRAME not supported in rev1.0 */ +} __packed; +#define QM_CGR_EN 0x01 /* For wr_en_*, cscn_en, cstd_en */ +#define QM_CGR_TARG_UDP_CTRL_WRITE_BIT 0x8000 /* value written to portal bit*/ +#define QM_CGR_TARG_UDP_CTRL_DCP 0x4000 /* 0: SWP, 1: DCP */ +#define QM_CGR_TARG_PORTAL(n) (0x80000000 >> (n)) /* s/w portal, 0-9 */ +#define QM_CGR_TARG_FMAN0 0x00200000 /* direct-connect portal: fman0 */ +#define QM_CGR_TARG_FMAN1 0x00100000 /* : fman1 */ +/* Convert CGR thresholds to/from "cs_thres" format */ +static inline u64 qm_cgr_cs_thres_get64(const struct qm_cgr_cs_thres *th) +{ + return ((th->word >> 5) & 0xff) << (th->word & 0x1f); +} + +static inline int qm_cgr_cs_thres_set64(struct qm_cgr_cs_thres *th, u64 val, + int roundup) +{ + u32 e = 0; + int oddbit = 0; + + while (val > 0xff) { + oddbit = val & 1; + val >>= 1; + e++; + if (roundup && oddbit) + val++; + } + th->word = ((val & 0xff) << 5) | (e & 0x1f); + return 0; +} + +/* "Initialize FQ" */ +struct qm_mcc_initfq { + u8 __reserved1[2]; + u16 we_mask; /* Write Enable Mask */ + u32 fqid; /* 24-bit */ + u16 count; /* Initialises 'count+1' FQDs */ + struct qm_fqd fqd; /* the FQD fields go here */ + u8 __reserved2[30]; +} __packed; +/* "Initialize/Modify CGR" */ +struct qm_mcc_initcgr { + u8 __reserve1[2]; + u16 we_mask; /* Write Enable Mask */ + struct __qm_mc_cgr cgr; /* CGR fields */ + u8 __reserved2[2]; + u8 cgid; + u8 __reserved3[32]; +} __packed; + +/* INITFQ-specific flags */ +#define QM_INITFQ_WE_MASK 0x01ff /* 'Write Enable' flags; */ +#define QM_INITFQ_WE_OAC 0x0100 +#define QM_INITFQ_WE_ORPC 0x0080 +#define QM_INITFQ_WE_CGID 0x0040 +#define QM_INITFQ_WE_FQCTRL 0x0020 +#define QM_INITFQ_WE_DESTWQ 0x0010 +#define QM_INITFQ_WE_ICSCRED 0x0008 +#define QM_INITFQ_WE_TDTHRESH 0x0004 +#define QM_INITFQ_WE_CONTEXTB 0x0002 +#define QM_INITFQ_WE_CONTEXTA 0x0001 +/* INITCGR/MODIFYCGR-specific flags */ +#define QM_CGR_WE_MASK 0x07ff /* 'Write Enable Mask'; */ +#define QM_CGR_WE_WR_PARM_G 0x0400 +#define QM_CGR_WE_WR_PARM_Y 0x0200 +#define QM_CGR_WE_WR_PARM_R 0x0100 +#define QM_CGR_WE_WR_EN_G 0x0080 +#define QM_CGR_WE_WR_EN_Y 0x0040 +#define QM_CGR_WE_WR_EN_R 0x0020 +#define QM_CGR_WE_CSCN_EN 0x0010 +#define QM_CGR_WE_CSCN_TARG 0x0008 +#define QM_CGR_WE_CSTD_EN 0x0004 +#define QM_CGR_WE_CS_THRES 0x0002 +#define QM_CGR_WE_MODE 0x0001 + +#define QMAN_CGR_FLAG_USE_INIT 0x00000001 + + /* Portal and Frame Queues */ +/* Represents a managed portal */ +struct qman_portal; + +/* + * This object type represents QMan frame queue descriptors (FQD), it is + * cacheline-aligned, and initialised by qman_create_fq(). The structure is + * defined further down. + */ +struct qman_fq; + +/* + * This object type represents a QMan congestion group, it is defined further + * down. + */ +struct qman_cgr; + +/* + * This enum, and the callback type that returns it, are used when handling + * dequeued frames via DQRR. Note that for "null" callbacks registered with the + * portal object (for handling dequeues that do not demux because contextB is + * NULL), the return value *MUST* be qman_cb_dqrr_consume. + */ +enum qman_cb_dqrr_result { + /* DQRR entry can be consumed */ + qman_cb_dqrr_consume, + /* Like _consume, but requests parking - FQ must be held-active */ + qman_cb_dqrr_park, + /* Does not consume, for DCA mode only. */ + qman_cb_dqrr_defer, + /* + * Stop processing without consuming this ring entry. Exits the current + * qman_p_poll_dqrr() or interrupt-handling, as appropriate. If within + * an interrupt handler, the callback would typically call + * qman_irqsource_remove(QM_PIRQ_DQRI) before returning this value, + * otherwise the interrupt will reassert immediately. + */ + qman_cb_dqrr_stop, + /* Like qman_cb_dqrr_stop, but consumes the current entry. */ + qman_cb_dqrr_consume_stop +}; +typedef enum qman_cb_dqrr_result (*qman_cb_dqrr)(struct qman_portal *qm, + struct qman_fq *fq, + const struct qm_dqrr_entry *dqrr); + +/* + * This callback type is used when handling ERNs, FQRNs and FQRLs via MR. They + * are always consumed after the callback returns. + */ +typedef void (*qman_cb_mr)(struct qman_portal *qm, struct qman_fq *fq, + const union qm_mr_entry *msg); + +/* + * s/w-visible states. Ie. tentatively scheduled + truly scheduled + active + + * held-active + held-suspended are just "sched". Things like "retired" will not + * be assumed until it is complete (ie. QMAN_FQ_STATE_CHANGING is set until + * then, to indicate it's completing and to gate attempts to retry the retire + * command). Note, park commands do not set QMAN_FQ_STATE_CHANGING because it's + * technically impossible in the case of enqueue DCAs (which refer to DQRR ring + * index rather than the FQ that ring entry corresponds to), so repeated park + * commands are allowed (if you're silly enough to try) but won't change FQ + * state, and the resulting park notifications move FQs from "sched" to + * "parked". + */ +enum qman_fq_state { + qman_fq_state_oos, + qman_fq_state_parked, + qman_fq_state_sched, + qman_fq_state_retired +}; + +#define QMAN_FQ_STATE_CHANGING 0x80000000 /* 'state' is changing */ +#define QMAN_FQ_STATE_NE 0x40000000 /* retired FQ isn't empty */ +#define QMAN_FQ_STATE_ORL 0x20000000 /* retired FQ has ORL */ +#define QMAN_FQ_STATE_BLOCKOOS 0xe0000000 /* if any are set, no OOS */ +#define QMAN_FQ_STATE_CGR_EN 0x10000000 /* CGR enabled */ +#define QMAN_FQ_STATE_VDQCR 0x08000000 /* being volatile dequeued */ + +/* + * Frame queue objects (struct qman_fq) are stored within memory passed to + * qman_create_fq(), as this allows stashing of caller-provided demux callback + * pointers at no extra cost to stashing of (driver-internal) FQ state. If the + * caller wishes to add per-FQ state and have it benefit from dequeue-stashing, + * they should; + * + * (a) extend the qman_fq structure with their state; eg. + * + * // myfq is allocated and driver_fq callbacks filled in; + * struct my_fq { + * struct qman_fq base; + * int an_extra_field; + * [ ... add other fields to be associated with each FQ ...] + * } *myfq = some_my_fq_allocator(); + * struct qman_fq *fq = qman_create_fq(fqid, flags, &myfq->base); + * + * // in a dequeue callback, access extra fields from 'fq' via a cast; + * struct my_fq *myfq = (struct my_fq *)fq; + * do_something_with(myfq->an_extra_field); + * [...] + * + * (b) when and if configuring the FQ for context stashing, specify how ever + * many cachelines are required to stash 'struct my_fq', to accelerate not + * only the QMan driver but the callback as well. + */ + +struct qman_fq_cb { + qman_cb_dqrr dqrr; /* for dequeued frames */ + qman_cb_mr ern; /* for s/w ERNs */ + qman_cb_mr fqs; /* frame-queue state changes*/ +}; + +struct qman_fq { + /* Caller of qman_create_fq() provides these demux callbacks */ + struct qman_fq_cb cb; + /* + * These are internal to the driver, don't touch. In particular, they + * may change, be removed, or extended (so you shouldn't rely on + * sizeof(qman_fq) being a constant). + */ + u32 fqid, idx; + unsigned long flags; + enum qman_fq_state state; + int cgr_groupid; +}; + +/* + * This callback type is used when handling congestion group entry/exit. + * 'congested' is non-zero on congestion-entry, and zero on congestion-exit. + */ +typedef void (*qman_cb_cgr)(struct qman_portal *qm, + struct qman_cgr *cgr, int congested); + +struct qman_cgr { + /* Set these prior to qman_create_cgr() */ + u32 cgrid; /* 0..255, but u32 to allow specials like -1, 256, etc.*/ + qman_cb_cgr cb; + /* These are private to the driver */ + u16 chan; /* portal channel this object is created on */ + struct list_head node; +}; + +/* Flags to qman_create_fq() */ +#define QMAN_FQ_FLAG_NO_ENQUEUE 0x00000001 /* can't enqueue */ +#define QMAN_FQ_FLAG_NO_MODIFY 0x00000002 /* can only enqueue */ +#define QMAN_FQ_FLAG_TO_DCPORTAL 0x00000004 /* consumed by CAAM/PME/Fman */ +#define QMAN_FQ_FLAG_DYNAMIC_FQID 0x00000020 /* (de)allocate fqid */ + +/* Flags to qman_init_fq() */ +#define QMAN_INITFQ_FLAG_SCHED 0x00000001 /* schedule rather than park */ +#define QMAN_INITFQ_FLAG_LOCAL 0x00000004 /* set dest portal */ + + /* Portal Management */ +/** + * qman_p_irqsource_add - add processing sources to be interrupt-driven + * @bits: bitmask of QM_PIRQ_**I processing sources + * + * Adds processing sources that should be interrupt-driven (rather than + * processed via qman_poll_***() functions). + */ +void qman_p_irqsource_add(struct qman_portal *p, u32 bits); + +/** + * qman_p_irqsource_remove - remove processing sources from being int-driven + * @bits: bitmask of QM_PIRQ_**I processing sources + * + * Removes processing sources from being interrupt-driven, so that they will + * instead be processed via qman_poll_***() functions. + */ +void qman_p_irqsource_remove(struct qman_portal *p, u32 bits); + +/** + * qman_affine_cpus - return a mask of cpus that have affine portals + */ +const cpumask_t *qman_affine_cpus(void); + +/** + * qman_affine_channel - return the channel ID of an portal + * @cpu: the cpu whose affine portal is the subject of the query + * + * If @cpu is -1, the affine portal for the current CPU will be used. It is a + * bug to call this function for any value of @cpu (other than -1) that is not a + * member of the mask returned from qman_affine_cpus(). + */ +u16 qman_affine_channel(int cpu); + +/** + * qman_get_affine_portal - return the portal pointer affine to cpu + * @cpu: the cpu whose affine portal is the subject of the query + */ +struct qman_portal *qman_get_affine_portal(int cpu); + +/** + * qman_p_poll_dqrr - process DQRR (fast-path) entries + * @limit: the maximum number of DQRR entries to process + * + * Use of this function requires that DQRR processing not be interrupt-driven. + * The return value represents the number of DQRR entries processed. + */ +int qman_p_poll_dqrr(struct qman_portal *p, unsigned int limit); + +/** + * qman_p_static_dequeue_add - Add pool channels to the portal SDQCR + * @pools: bit-mask of pool channels, using QM_SDQCR_CHANNELS_POOL(n) + * + * Adds a set of pool channels to the portal's static dequeue command register + * (SDQCR). The requested pools are limited to those the portal has dequeue + * access to. + */ +void qman_p_static_dequeue_add(struct qman_portal *p, u32 pools); + + /* FQ management */ +/** + * qman_create_fq - Allocates a FQ + * @fqid: the index of the FQD to encapsulate, must be "Out of Service" + * @flags: bit-mask of QMAN_FQ_FLAG_*** options + * @fq: memory for storing the 'fq', with callbacks filled in + * + * Creates a frame queue object for the given @fqid, unless the + * QMAN_FQ_FLAG_DYNAMIC_FQID flag is set in @flags, in which case a FQID is + * dynamically allocated (or the function fails if none are available). Once + * created, the caller should not touch the memory at 'fq' except as extended to + * adjacent memory for user-defined fields (see the definition of "struct + * qman_fq" for more info). NO_MODIFY is only intended for enqueuing to |