/* $OpenBSD: awacs.c,v 1.29 2014/07/12 18:44:42 tedu Exp $ */ /* $NetBSD: awacs.c,v 1.4 2001/02/26 21:07:51 wiz Exp $ */ /*- * Copyright (c) 2000 Tsubai Masanari. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. 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. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 THE AUTHOR 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 #include #include #include #include #include #include #include #include #include #ifdef AWACS_DEBUG # define DPRINTF printf #else # define DPRINTF while (0) printf #endif #define AWACS_DMALIST_MAX 32 #define AWACS_DMASEG_MAX NBPG struct awacs_dma { bus_dmamap_t map; caddr_t addr; bus_dma_segment_t segs[AWACS_DMALIST_MAX]; int nsegs; size_t size; struct awacs_dma *next; }; struct awacs_softc { struct device sc_dev; void (*sc_ointr)(void *); /* dma completion intr handler */ void *sc_oarg; /* arg for sc_ointr() */ void (*sc_iintr)(void *); /* dma completion intr handler */ void *sc_iarg; /* arg for sc_iintr() */ u_int sc_record_source; /* recording source mask */ u_int sc_output_mask; /* output source mask */ char *sc_reg; u_int sc_codecctl0; u_int sc_codecctl1; u_int sc_codecctl2; u_int sc_codecctl4; u_int sc_soundctl; bus_dma_tag_t sc_dmat; struct dbdma_regmap *sc_odma; struct dbdma_regmap *sc_idma; struct dbdma_command *sc_odmacmd, *sc_odmap; struct dbdma_command *sc_idmacmd, *sc_idmap; dbdma_t sc_odbdma, sc_idbdma; struct awacs_dma *sc_dmas; }; int awacs_match(struct device *, void *, void *); void awacs_attach(struct device *, struct device *, void *); int awacs_intr(void *); int awacs_tx_intr(void *); int awacs_rx_intr(void *); int awacs_open(void *, int); void awacs_close(void *); int awacs_query_encoding(void *, struct audio_encoding *); int awacs_set_params(void *, int, int, struct audio_params *, struct audio_params *); int awacs_round_blocksize(void *, int); int awacs_trigger_output(void *, void *, void *, int, void (*)(void *), void *, struct audio_params *); int awacs_trigger_input(void *, void *, void *, int, void (*)(void *), void *, struct audio_params *); int awacs_halt_output(void *); int awacs_halt_input(void *); int awacs_getdev(void *, struct audio_device *); int awacs_set_port(void *, mixer_ctrl_t *); int awacs_get_port(void *, mixer_ctrl_t *); int awacs_query_devinfo(void *, mixer_devinfo_t *); size_t awacs_round_buffersize(void *, int, size_t); paddr_t awacs_mappage(void *, void *, off_t, int); int awacs_get_props(void *); void *awacs_allocm(void *, int, size_t, int, int); static inline u_int awacs_read_reg(struct awacs_softc *, int); static inline void awacs_write_reg(struct awacs_softc *, int, int); void awacs_write_codec(struct awacs_softc *, int); void awacs_set_speaker_volume(struct awacs_softc *, int, int); void awacs_set_ext_volume(struct awacs_softc *, int, int); void awacs_set_rate(struct awacs_softc *, struct audio_params *); struct cfattach awacs_ca = { sizeof(struct awacs_softc), awacs_match, awacs_attach }; struct cfdriver awacs_cd = { NULL, "awacs", DV_DULL }; struct audio_hw_if awacs_hw_if = { awacs_open, awacs_close, NULL, /* drain */ awacs_query_encoding, awacs_set_params, awacs_round_blocksize, NULL, /* commit_setting */ NULL, /* init_output */ NULL, /* init_input */ NULL, /* start_output */ NULL, /* start_input */ awacs_halt_output, awacs_halt_input, NULL, /* speaker_ctl */ awacs_getdev, NULL, /* getfd */ awacs_set_port, awacs_get_port, awacs_query_devinfo, awacs_allocm, /* allocm */ NULL, /* freem */ awacs_round_buffersize, /* round_buffersize */ awacs_mappage, awacs_get_props, awacs_trigger_output, awacs_trigger_input, NULL }; struct audio_device awacs_device = { "AWACS", "", "awacs" }; /* register offset */ #define AWACS_SOUND_CTRL 0x00 #define AWACS_CODEC_CTRL 0x10 #define AWACS_CODEC_STATUS 0x20 #define AWACS_CLIP_COUNT 0x30 #define AWACS_BYTE_SWAP 0x40 /* sound control */ #define AWACS_INPUT_SUBFRAME0 0x00000001 #define AWACS_INPUT_SUBFRAME1 0x00000002 #define AWACS_INPUT_SUBFRAME2 0x00000004 #define AWACS_INPUT_SUBFRAME3 0x00000008 #define AWACS_OUTPUT_SUBFRAME0 0x00000010 #define AWACS_OUTPUT_SUBFRAME1 0x00000020 #define AWACS_OUTPUT_SUBFRAME2 0x00000040 #define AWACS_OUTPUT_SUBFRAME3 0x00000080 #define AWACS_RATE_44100 0x00000000 #define AWACS_RATE_29400 0x00000100 #define AWACS_RATE_22050 0x00000200 #define AWACS_RATE_17640 0x00000300 #define AWACS_RATE_14700 0x00000400 #define AWACS_RATE_11025 0x00000500 #define AWACS_RATE_8820 0x00000600 #define AWACS_RATE_7350 0x00000700 #define AWACS_RATE_MASK 0x00000700 #define AWACS_CTL_CNTRLERR (1 << 11) #define AWACS_CTL_PORTCHG (1 << 12) #define AWACS_INT_CNTRLERR (1 << 13) #define AWACS_INT_PORTCHG (1 << 14) /* codec control */ #define AWACS_CODEC_ADDR0 0x00000000 #define AWACS_CODEC_ADDR1 0x00001000 #define AWACS_CODEC_ADDR2 0x00002000 #define AWACS_CODEC_ADDR4 0x00004000 #define AWACS_CODEC_EMSEL0 0x00000000 #define AWACS_CODEC_EMSEL1 0x00400000 #define AWACS_CODEC_EMSEL2 0x00800000 #define AWACS_CODEC_EMSEL4 0x00c00000 #define AWACS_CODEC_BUSY 0x01000000 /* cc0 */ #define AWACS_DEFAULT_CD_GAIN 0x000000bb #define AWACS_INPUT_CD 0x00000200 #define AWACS_INPUT_LINE 0x00000400 #define AWACS_INPUT_MICROPHONE 0x00000800 #define AWACS_INPUT_MASK 0x00000e00 /* cc1 */ #define AWACS_MUTE_SPEAKER 0x00000080 #define AWACS_MUTE_HEADPHONE 0x00000200 const struct awacs_speed_tab { int rate; u_int32_t bits; } awacs_speeds[] = { { 7350, AWACS_RATE_7350 }, { 8820, AWACS_RATE_8820 }, { 11025, AWACS_RATE_11025 }, { 14700, AWACS_RATE_14700 }, { 17640, AWACS_RATE_17640 }, { 22050, AWACS_RATE_22050 }, { 29400, AWACS_RATE_29400 }, { 44100, AWACS_RATE_44100 }, }; int awacs_match(struct device *parent, void *match, void *aux) { struct confargs *ca = aux; if (strcmp(ca->ca_name, "awacs") != 0 && strcmp(ca->ca_name, "davbus") != 0) return 0; #ifdef DEBUG printf("awacs: matched %s nreg %d nintr %d\n", ca->ca_name, ca->ca_nreg, ca->ca_nintr); #endif if (ca->ca_nreg < 24 || ca->ca_nintr < 12) return 0; /* XXX for now if (ca->ca_nintr > 12) return 0; */ return 1; } void awacs_attach(struct device *parent, struct device *self, void *aux) { struct awacs_softc *sc = (struct awacs_softc *)self; struct confargs *ca = aux; int cirq, oirq, iirq; int cirq_type, oirq_type, iirq_type; ca->ca_reg[0] += ca->ca_baseaddr; ca->ca_reg[2] += ca->ca_baseaddr; ca->ca_reg[4] += ca->ca_baseaddr; sc->sc_reg = mapiodev(ca->ca_reg[0], ca->ca_reg[1]); sc->sc_dmat = ca->ca_dmat; sc->sc_odma = mapiodev(ca->ca_reg[2], ca->ca_reg[3]); /* out */ sc->sc_idma = mapiodev(ca->ca_reg[4], ca->ca_reg[5]); /* in */ sc->sc_odbdma = dbdma_alloc(sc->sc_dmat, AWACS_DMALIST_MAX); sc->sc_odmacmd = sc->sc_odbdma->d_addr; sc->sc_idbdma = dbdma_alloc(sc->sc_dmat, AWACS_DMALIST_MAX); sc->sc_idmacmd = sc->sc_idbdma->d_addr; if (ca->ca_nintr == 24) { cirq = ca->ca_intr[0]; oirq = ca->ca_intr[2]; iirq = ca->ca_intr[4]; cirq_type = ca->ca_intr[1] ? IST_LEVEL : IST_EDGE; oirq_type = ca->ca_intr[3] ? IST_LEVEL : IST_EDGE; iirq_type = ca->ca_intr[5] ? IST_LEVEL : IST_EDGE; } else { cirq = ca->ca_intr[0]; oirq = ca->ca_intr[1]; iirq = ca->ca_intr[2]; cirq_type = oirq_type = iirq_type = IST_LEVEL; } mac_intr_establish(parent, cirq, cirq_type, IPL_AUDIO, awacs_intr, sc, sc->sc_dev.dv_xname); mac_intr_establish(parent, oirq, oirq_type, IPL_AUDIO, awacs_tx_intr, sc, sc->sc_dev.dv_xname); mac_intr_establish(parent, iirq, iirq_type, IPL_AUDIO, awacs_rx_intr, sc, sc->sc_dev.dv_xname); printf(": irq %d,%d,%d", cirq, oirq, iirq); sc->sc_soundctl = AWACS_INPUT_SUBFRAME0 | AWACS_OUTPUT_SUBFRAME0 | AWACS_RATE_44100 | AWACS_INT_PORTCHG; awacs_write_reg(sc, AWACS_SOUND_CTRL, sc->sc_soundctl); sc->sc_codecctl0 = AWACS_CODEC_ADDR0 | AWACS_CODEC_EMSEL0; sc->sc_codecctl1 = AWACS_CODEC_ADDR1 | AWACS_CODEC_EMSEL0; sc->sc_codecctl2 = AWACS_CODEC_ADDR2 | AWACS_CODEC_EMSEL0; sc->sc_codecctl4 = AWACS_CODEC_ADDR4 | AWACS_CODEC_EMSEL0; sc->sc_codecctl0 |= AWACS_INPUT_CD | AWACS_DEFAULT_CD_GAIN; awacs_write_codec(sc, sc->sc_codecctl0); /* Set initial volume[s] */ awacs_set_speaker_volume(sc, 80, 80); awacs_set_ext_volume(sc, 80, 80); /* Set loopback (for CD?) */ /* sc->sc_codecctl1 |= 0x440; */ sc->sc_codecctl1 |= 0x40; awacs_write_codec(sc, sc->sc_codecctl1); /* check for headphone present */ if (awacs_read_reg(sc, AWACS_CODEC_STATUS) & 0x8) { /* default output to speakers */ printf(" headphones"); sc->sc_output_mask = 1 << 1; sc->sc_codecctl1 &= ~AWACS_MUTE_HEADPHONE; sc->sc_codecctl1 |= AWACS_MUTE_SPEAKER; awacs_write_codec(sc, sc->sc_codecctl1); } else { /* default output to speakers */ printf(" speaker"); sc->sc_output_mask = 1 << 0; sc->sc_codecctl1 &= ~AWACS_MUTE_SPEAKER; sc->sc_codecctl1 |= AWACS_MUTE_HEADPHONE; awacs_write_codec(sc, sc->sc_codecctl1); } /* default input from CD */ sc->sc_record_source = 1 << 0; sc->sc_codecctl0 &= ~AWACS_INPUT_MASK; sc->sc_codecctl0 |= AWACS_INPUT_CD; awacs_write_codec(sc, sc->sc_codecctl0); /* Enable interrupts and looping mode. */ /* XXX ... */ awacs_halt_output(sc); awacs_halt_input(sc); printf("\n"); audio_attach_mi(&awacs_hw_if, sc, &sc->sc_dev); } u_int awacs_read_reg(struct awacs_softc *sc, int reg) { char *addr = sc->sc_reg; return in32rb(addr + reg); } void awacs_write_reg(struct awacs_softc *sc, int reg, int val) { char *addr = sc->sc_reg; out32rb(addr + reg, val); } void awacs_write_codec(struct awacs_softc *sc, int value) { awacs_write_reg(sc, AWACS_CODEC_CTRL, value); while (awacs_read_reg(sc, AWACS_CODEC_CTRL) & AWACS_CODEC_BUSY); } int awacs_intr(void *v) { int reason; struct awacs_softc *sc = v; mtx_enter(&audio_lock); reason = awacs_read_reg(sc, AWACS_SOUND_CTRL); if (reason & AWACS_CTL_CNTRLERR) { /* change outputs ?? */ } if (reason & AWACS_CTL_PORTCHG) { #ifdef DEBUG printf("status = %x\n", awacs_read_reg(sc, AWACS_CODEC_STATUS)); #endif if (awacs_read_reg(sc, AWACS_CODEC_STATUS) & 0x8) { /* default output to speakers */ sc->sc_output_mask = 1 << 1; sc->sc_codecctl1 &= ~AWACS_MUTE_HEADPHONE; sc->sc_codecctl1 |= AWACS_MUTE_SPEAKER; awacs_write_codec(sc, sc->sc_codecctl1); } else { /* default output to speakers */ sc->sc_output_mask = 1 << 0; sc->sc_codecctl1 &= ~AWACS_MUTE_SPEAKER; sc->sc_codecctl1 |= AWACS_MUTE_HEADPHONE; awacs_write_codec(sc, sc->sc_codecctl1); } } awacs_write_reg(sc, AWACS_SOUND_CTRL, reason); /* clear interrupt */ mtx_leave(&audio_lock); return 1; } int awacs_tx_intr(void *v) { struct awacs_softc *sc = v; struct dbdma_command *cmd = sc->sc_odmap; u_int16_t c, status; /* if not set we are not running */ if (!cmd) return (0); mtx_enter(&audio_lock); c = in16rb(&cmd->d_command); status = in16rb(&cmd->d_status); if (c >> 12 == DBDMA_CMD_OUT_LAST) sc->sc_odmap = sc->sc_odmacmd; else sc->sc_odmap++; if (c & (DBDMA_INT_ALWAYS << 4)) { cmd->d_status = 0; if (status) /* status == 0x8400 */ if (sc->sc_ointr) (*sc->sc_ointr)(sc->sc_oarg); } mtx_leave(&audio_lock); return (1); } int awacs_rx_intr(void *v) { struct awacs_softc *sc = v; struct dbdma_command *cmd = sc->sc_idmap; u_int16_t c, status; /* if not set we are not running */ if (!cmd) return (0); mtx_enter(&audio_lock); c = in16rb(&cmd->d_command); status = in16rb(&cmd->d_status); if (c >> 12 == DBDMA_CMD_IN_LAST) sc->sc_idmap = sc->sc_idmacmd; else sc->sc_idmap++; if (c & (DBDMA_INT_ALWAYS << 4)) { cmd->d_status = 0; if (status) /* status == 0x8400 */ if (sc->sc_iintr) (*sc->sc_iintr)(sc->sc_iarg); } mtx_leave(&audio_lock); return (1); } int awacs_open(void *h, int flags) { return 0; } /* * Close function is called at splaudio(). */ void awacs_close(void *h) { struct awacs_softc *sc = h; /* XXX: halt_xxx() already called by upper layer */ awacs_halt_output(sc); awacs_halt_input(sc); sc->sc_ointr = 0; sc->sc_iintr = 0; } int awacs_query_encoding(void *h, struct audio_encoding *ae) { switch (ae->index) { case 0: strlcpy(ae->name, AudioEslinear, sizeof ae->name); ae->encoding = AUDIO_ENCODING_SLINEAR; ae->precision = 16; ae->flags = 0; break; case 1: strlcpy(ae->name, AudioEslinear_be, sizeof ae->name); ae->encoding = AUDIO_ENCODING_SLINEAR_BE; ae->precision = 16; ae->flags = 0; break; case 2: strlcpy(ae->name, AudioEslinear_le, sizeof ae->name); ae->encoding = AUDIO_ENCODING_SLINEAR_LE; ae->precision = 16; ae->flags = 0; break; case 3: strlcpy(ae->name, AudioEmulaw, sizeof ae->name); ae->encoding = AUDIO_ENCODING_ULAW; ae->precision = 8; ae->flags = AUDIO_ENCODINGFLAG_EMULATED; break; case 4: strlcpy(ae->name, AudioEalaw, sizeof ae->name); ae->encoding = AUDIO_ENCODING_ALAW; ae->precision = 8; ae->flags = AUDIO_ENCODINGFLAG_EMULATED; break; case 5: strlcpy(ae->name, AudioEulinear, sizeof ae->name); ae->encoding = AUDIO_ENCODING_ULINEAR; ae->precision = 16; ae->flags = AUDIO_ENCODINGFLAG_EMULATED; break; case 6: strlcpy(ae->name, AudioEulinear_le, sizeof ae->name); ae->encoding = AUDIO_ENCODING_ULINEAR_LE; ae->precision = 16; ae->flags = AUDIO_ENCODINGFLAG_EMULATED; break; case 7: strlcpy(ae->name, AudioEulinear_be, sizeof ae->name); ae->encoding = AUDIO_ENCODING_ULINEAR_BE; ae->precision = 16; ae->flags = AUDIO_ENCODINGFLAG_EMULATED; break; default: return (EINVAL); } ae->bps = AUDIO_BPS(ae->precision); ae->msb = 1; return (0); } int awacs_set_params(void *h, int setmode, int usemode, struct audio_params *play, struct audio_params *rec) { struct awacs_softc *sc = h; struct audio_params *p; int mode; /* * This device only has one clock, so make the sample rates match. */ if (play->sample_rate != rec->sample_rate && usemode == (AUMODE_PLAY | AUMODE_RECORD)) { if (setmode == AUMODE_PLAY) { rec->sample_rate = play->sample_rate; setmode |= AUMODE_RECORD; } else if (setmode == AUMODE_RECORD) { play->sample_rate = rec->sample_rate; setmode |= AUMODE_PLAY; } else return EINVAL; } for (mode = AUMODE_RECORD; mode != -1; mode = mode == AUMODE_RECORD ? AUMODE_PLAY : -1) { if ((setmode & mode) == 0) continue; p = mode == AUMODE_PLAY ? play : rec; if (p->sample_rate < 4000) p->sample_rate = 4000; if (p->sample_rate > 50000) p->sample_rate = 50000; if (p->precision > 16) p->precision = 16; if (p->channels > 2) p->channels = 2; p->factor = 1; p->sw_code = NULL; awacs_write_reg(sc, AWACS_BYTE_SWAP, 0); switch (p->encoding) { case AUDIO_ENCODING_SLINEAR_LE: if (p->precision != 16) p->precision = 16; if (p->channels == 2) p->sw_code = swap_bytes; else { p->factor = 2; p->sw_code = swap_bytes_mts; } break; case AUDIO_ENCODING_SLINEAR_BE: if (p->precision != 16) p->precision = 16; if (p->channels == 1) { p->factor = 2; p->sw_code = noswap_bytes_mts; } break; case AUDIO_ENCODING_ULINEAR_LE: if (p->precision != 16) p->precision = 16; if (p->channels == 2) p->sw_code = swap_bytes_change_sign16_be; else { p->factor = 2; p->sw_code = swap_bytes_change_sign16_be_mts; } break; case AUDIO_ENCODING_ULINEAR_BE: if (p->precision != 16) p->precision = 16; if (p->channels == 2) p->sw_code = change_sign16_be; else { p->factor = 2; p->sw_code = change_sign16_be_mts; } break; case AUDIO_ENCODING_ULAW: if (mode == AUMODE_PLAY) { p->factor = 2; p->sw_code = mulaw_to_slinear16_be; break; } else break; /* XXX */ return (EINVAL); case AUDIO_ENCODING_ALAW: if (mode == AUMODE_PLAY) { p->factor = 2; p->sw_code = alaw_to_slinear16_be; break; } return (EINVAL); default: return (EINVAL); } p->bps = AUDIO_BPS(p->precision); p->msb = 1; } /* Set the speed */ awacs_set_rate(sc, p); return (0); } int awacs_round_blocksize(void *h, int size) { if (size < PAGE_SIZE) size = PAGE_SIZE; return (size + PAGE_SIZE / 2) & ~(PGOFSET); } int awacs_halt_output(void *h) { struct awacs_softc *sc = h; mtx_enter(&audio_lock); dbdma_stop(sc->sc_odma); dbdma_reset(sc->sc_odma); dbdma_stop(sc->sc_odma); sc->sc_odmap = NULL; mtx_leave(&audio_lock); return 0; } int awacs_halt_input(void *h) { struct awacs_softc *sc = h; mtx_enter(&audio_lock); dbdma_stop(sc->sc_idma); dbdma_reset(sc->sc_idma); mtx_leave(&audio_lock); return 0; } int awacs_getdev(void *h, struct audio_device *retp) { *retp = awacs_device; return 0; } enum { AWACS_OUTPUT_SELECT, AWACS_VOL_SPEAKER, AWACS_VOL_HEADPHONE, AWACS_OUTPUT_CLASS, AWACS_MONITOR_CLASS, AWACS_INPUT_SELECT, AWACS_VOL_INPUT, AWACS_INPUT_CLASS, AWACS_RECORD_CLASS, AWACS_ENUM_LAST }; int awacs_set_port(void *h, mixer_ctrl_t *mc) { struct awacs_softc *sc = h; int l, r; DPRINTF("awacs_set_port dev = %d, type = %d\n", mc->dev, mc->type); l = mc->un.value.level[AUDIO_MIXER_LEVEL_LEFT]; r = mc->un.value.level[AUDIO_MIXER_LEVEL_RIGHT]; switch (mc->dev) { case AWACS_OUTPUT_SELECT: /* no change necessary? */ if (mc->un.mask == sc->sc_output_mask) return 0; sc->sc_codecctl1 |= AWACS_MUTE_SPEAKER | AWACS_MUTE_HEADPHONE; if (mc->un.mask & 1 << 0) sc->sc_codecctl1 &= ~AWACS_MUTE_SPEAKER; if (mc->un.mask & 1 << 1) sc->sc_codecctl1 &= ~AWACS_MUTE_HEADPHONE; awacs_write_codec(sc, sc->sc_codecctl1); sc->sc_output_mask = mc->un.mask; return 0; case AWACS_VOL_SPEAKER: awacs_set_speaker_volume(sc, l, r); return 0; case AWACS_VOL_HEADPHONE: awacs_set_ext_volume(sc, l, r); return 0; case AWACS_VOL_INPUT: sc->sc_codecctl0 &= ~0xff; sc->sc_codecctl0 |= (l & 0xf0) | (r >> 4); awacs_write_codec(sc, sc->sc_codecctl0); return 0; case AWACS_INPUT_SELECT: /* no change necessary? */ if (mc->un.mask == sc->sc_record_source) return 0; switch(mc->un.mask) { case 1<<0: /* CD */ sc->sc_codecctl0 &= ~AWACS_INPUT_MASK; sc->sc_codecctl0 |= AWACS_INPUT_CD; awacs_write_codec(sc, sc->sc_codecctl0); break; case 1<<1: /* microphone */ sc->sc_codecctl0 &= ~AWACS_INPUT_MASK; sc->sc_codecctl0 |= AWACS_INPUT_MICROPHONE; awacs_write_codec(sc, sc->sc_codecctl0); break; case 1<<2: /* line in */ sc->sc_codecctl0 &= ~AWACS_INPUT_MASK; sc->sc_codecctl0 |= AWACS_INPUT_LINE; awacs_write_codec(sc, sc->sc_codecctl0); break; default: /* invalid argument */ return -1; } sc->sc_record_source = mc->un.mask; return 0; } return ENXIO; } int awacs_get_port(void *h, mixer_ctrl_t *mc) { struct awacs_softc *sc = h; int vol, l, r; DPRINTF("awacs_get_port dev = %d, type = %d\n", mc->dev, mc->type); switch (mc->dev) { case AWACS_OUTPUT_SELECT: mc->un.mask = sc->sc_output_mask; return 0; case AWACS_VOL_SPEAKER: vol = sc->sc_codecctl4; l = (15 - ((vol & 0x3c0) >> 6)) * 16; r = (15 - (vol & 0x0f)) * 16; mc->un.mask = 1 << 0; mc->un.value.num_channels = 2; mc->un.value.level[AUDIO_MIXER_LEVEL_LEFT] = l; mc->un.value.level[AUDIO_MIXER_LEVEL_RIGHT] = r; return 0; case AWACS_VOL_HEADPHONE: vol = sc->sc_codecctl2; l = (15 - ((vol & 0x3c0) >> 6)) * 16; r = (15 - (vol & 0x0f)) * 16; mc->un.mask = 1 << 1; mc->un.value.num_channels = 2; mc->un.value.level[AUDIO_MIXER_LEVEL_LEFT] = l; mc->un.value.level[AUDIO_MIXER_LEVEL_RIGHT] = r; return 0; case AWACS_INPUT_SELECT: mc->un.mask = sc->sc_record_source; return 0; case AWACS_VOL_INPUT: vol = sc->sc_codecctl0 & 0xff; l = (vol & 0xf0); r = (vol & 0x0f) << 4; mc->un.mask = sc->sc_record_source; mc->un.value.num_channels = 2; mc->un.value.level[AUDIO_MIXER_LEVEL_LEFT] = l; mc->un.value.level[AUDIO_MIXER_LEVEL_RIGHT] = r; return 0; default: return ENXIO; } return 0; } int awacs_query_devinfo(void *h, mixer_devinfo_t *dip) { DPRINTF("query_devinfo %d\n", dip->index); switch (dip->index) { case AWACS_OUTPUT_SELECT: dip->mixer_class = AWACS_OUTPUT_CLASS; strlcpy(dip->label.name, AudioNselect, sizeof dip->label.name); dip->type = AUDIO_MIXER_SET; dip->prev = dip->next = AUDIO_MIXER_LAST; dip->un.s.num_mem = 2; strlcpy(dip->un.s.member[0].label.name, AudioNspeaker, sizeof dip->un.s.member[0].label.name); dip->un.s.member[0].mask = 1 << 0; strlcpy(dip->un.s.member[1].label.name, AudioNheadphone, sizeof dip->un.s.member[0].label.name); dip->un.s.member[1].mask = 1 << 1; return 0; case AWACS_VOL_SPEAKER: dip->mixer_class = AWACS_OUTPUT_CLASS; strlcpy(dip->label.name, AudioNspeaker, sizeof dip->label.name); dip->type = AUDIO_MIXER_VALUE; dip->prev = dip->next = AUDIO_MIXER_LAST; dip->un.v.num_channels = 2; strlcpy(dip->un.v.units.name, AudioNvolume, sizeof dip->un.v.units.name); return 0; case AWACS_VOL_HEADPHONE: dip->mixer_class = AWACS_OUTPUT_CLASS; strlcpy(dip->label.name, AudioNheadphone, sizeof dip->label.name); dip->type = AUDIO_MIXER_VALUE; dip->prev = dip->next = AUDIO_MIXER_LAST; dip->un.v.num_channels = 2; strlcpy(dip->un.v.units.name, AudioNvolume, sizeof dip->un.v.units.name); return 0; case AWACS_INPUT_SELECT: dip->mixer_class = AWACS_RECORD_CLASS; strlcpy(dip->label.name, AudioNsource, sizeof dip->label.name); dip->type = AUDIO_MIXER_SET; dip->prev = dip->next = AUDIO_MIXER_LAST; dip->un.s.num_mem = 3; strlcpy(dip->un.s.member[0].label.name, AudioNcd, sizeof dip->un.s.member[0].label.name); dip->un.s.member[0].mask = 1 << 0; strlcpy(dip->un.s.member[1].label.name, AudioNmicrophone, sizeof dip->un.s.member[1].label.name); dip->un.s.member[1].mask = 1 << 1; strlcpy(dip->un.s.member[2].label.name, AudioNline, sizeof dip->un.s.member[2].label.name); dip->un.s.member[2].mask = 1 << 2; return 0; case AWACS_VOL_INPUT: dip->mixer_class = AWACS_RECORD_CLASS; strlcpy(dip->label.name, AudioNmaster, sizeof dip->label.name); dip->type = AUDIO_MIXER_VALUE; dip->prev = dip->next = AUDIO_MIXER_LAST; dip->un.v.num_channels = 2; strlcpy(dip->un.v.units.name, AudioNvolume, sizeof dip->un.v.units.name); return 0; case AWACS_MONITOR_CLASS: dip->mixer_class = AWACS_MONITOR_CLASS; strlcpy(dip->label.name, AudioCmonitor, sizeof dip->label.name); dip->type = AUDIO_MIXER_CLASS; dip->next = dip->prev = AUDIO_MIXER_LAST; return 0; case AWACS_OUTPUT_CLASS: dip->mixer_class = AWACS_OUTPUT_CLASS; strlcpy(dip->label.name, AudioCoutputs, sizeof dip->label.name); dip->type = AUDIO_MIXER_CLASS; dip->next = dip->prev = AUDIO_MIXER_LAST; return 0; case AWACS_RECORD_CLASS: dip->mixer_class = AWACS_MONITOR_CLASS; strlcpy(dip->label.name, AudioCrecord, sizeof dip->label.name); dip->type = AUDIO_MIXER_CLASS; dip->next = dip->prev = AUDIO_MIXER_LAST; return 0; } return ENXIO; } size_t awacs_round_buffersize(void *h, int dir, size_t size) { size = (size + PGOFSET) & ~(PGOFSET); if (size > AWACS_DMALIST_MAX * AWACS_DMASEG_MAX) size = AWACS_DMALIST_MAX * AWACS_DMASEG_MAX; return (size); } void * awacs_allocm(void *h, int dir, size_t size, int type, int flags) { struct awacs_softc *sc = h; struct awacs_dma *p; int error; if (size > AWACS_DMALIST_MAX * AWACS_DMASEG_MAX) return (NULL); p = malloc(sizeof(*p), type, flags | M_ZERO); if (!p) return (NULL); /* convert to the bus.h style, not used otherwise */ if (flags & M_NOWAIT) flags = BUS_DMA_NOWAIT; p->size = size; if ((error = bus_dmamem_alloc(sc->sc_dmat, p->size, NBPG, 0, p->segs, 1, &p->nsegs, flags)) != 0) { printf("%s: unable to allocate dma, error = %d\n", sc->sc_dev.dv_xname, error); free(p, type, 0); return NULL; } if ((error = bus_dmamem_map(sc->sc_dmat, p->segs, p->nsegs, p->size, &p->addr, flags | BUS_DMA_COHERENT)) != 0) { printf("%s: unable to map dma, error = %d\n", sc->sc_dev.dv_xname, error); bus_dmamem_free(sc->sc_dmat, p->segs, p->nsegs); free(p, type, 0); return NULL; } if ((error = bus_dmamap_create(sc->sc_dmat, p->size, 1, p->size, 0, flags, &p->map)) != 0) { printf("%s: unable to create dma map, error = %d\n", sc->sc_dev.dv_xname, error); bus_dmamem_unmap(sc->sc_dmat, p->addr, size); bus_dmamem_free(sc->sc_dmat, p->segs, p->nsegs); free(p, type, 0); return NULL; } if ((error = bus_dmamap_load(sc->sc_dmat, p->map, p->addr, p->size, NULL, flags)) != 0) { printf("%s: unable to load dma map, error = %d\n", sc->sc_dev.dv_xname, error); bus_dmamap_destroy(sc->sc_dmat, p->map); bus_dmamem_unmap(sc->sc_dmat, p->addr, size); bus_dmamem_free(sc->sc_dmat, p->segs, p->nsegs); free(p, type, 0); return NULL; } p->next = sc->sc_dmas; sc->sc_dmas = p; return p->addr; } paddr_t awacs_mappage(void *h, void *mem, off_t off, int prot) { if (off < 0) return -1; return -1; /* XXX */ } int awacs_get_props(void *h) { return AUDIO_PROP_FULLDUPLEX /* | AUDIO_PROP_MMAP */; } int awacs_trigger_output(void *h, void *start, void *end, int bsize, void (*intr)(void *), void *arg, struct audio_params *param) { struct awacs_softc *sc = h; struct awacs_dma *p; struct dbdma_command *cmd = sc->sc_odmacmd; vaddr_t spa, pa, epa; int c; DPRINTF("trigger_output %p %p 0x%x\n", start, end, bsize); for (p = sc->sc_dmas; p && p->addr != start; p = p->next); if (!p) return -1; sc->sc_ointr = intr; sc->sc_oarg = arg; sc->sc_odmap = sc->sc_odmacmd; mtx_enter(&audio_lock); spa = p->segs[0].ds_addr; c = DBDMA_CMD_OUT_MORE; for (pa = spa, epa = spa + (end - start); pa < epa; pa += bsize, cmd++) { if (pa + bsize == epa) c = DBDMA_CMD_OUT_LAST; DBDMA_BUILD(cmd, c, 0, bsize, pa, DBDMA_INT_ALWAYS, DBDMA_WAIT_NEVER, DBDMA_BRANCH_NEVER); } DBDMA_BUILD(cmd, DBDMA_CMD_NOP, 0, 0, 0, DBDMA_INT_NEVER, DBDMA_WAIT_NEVER, DBDMA_BRANCH_ALWAYS); dbdma_st32(&cmd->d_cmddep, sc->sc_odbdma->d_paddr); dbdma_start(sc->sc_odma, sc->sc_odbdma); mtx_leave(&audio_lock); return 0; } int awacs_trigger_input(void *h, void *start, void *end, int bsize, void (*intr)(void *), void *arg, struct audio_params *param) { struct awacs_softc *sc = h; struct awacs_dma *p; struct dbdma_command *cmd = sc->sc_idmacmd; vaddr_t spa, pa, epa; int c; DPRINTF("trigger_input %p %p 0x%x\n", start, end, bsize); for (p = sc->sc_dmas; p && p->addr != start; p = p->next); if (!p) return -1; sc->sc_iintr = intr; sc->sc_iarg = arg; sc->sc_idmap = sc->sc_idmacmd; mtx_enter(&audio_lock); spa = p->segs[0].ds_addr; c = DBDMA_CMD_IN_MORE; for (pa = spa, epa = spa + (end - start); pa < epa; pa += bsize, cmd++) { if (pa + bsize == epa) c = DBDMA_CMD_IN_LAST; DBDMA_BUILD(cmd, c, 0, bsize, pa, DBDMA_INT_ALWAYS, DBDMA_WAIT_NEVER, DBDMA_BRANCH_NEVER); } DBDMA_BUILD(cmd, DBDMA_CMD_NOP, 0, 0, 0, DBDMA_INT_NEVER, DBDMA_WAIT_NEVER, DBDMA_BRANCH_ALWAYS); dbdma_st32(&cmd->d_cmddep, sc->sc_idbdma->d_paddr); dbdma_start(sc->sc_idma, sc->sc_idbdma); mtx_leave(&audio_lock); return 0; } void awacs_set_speaker_volume(struct awacs_softc *sc, int left, int right) { int lval = 15 - (left & 0xff) / 16; int rval = 15 - (right & 0xff) / 16; DPRINTF("speaker_volume %d %d\n", lval, rval); sc->sc_codecctl4 &= ~0x3cf; sc->sc_codecctl4 |= (lval << 6) | rval; awacs_write_codec(sc, sc->sc_codecctl4); } void awacs_set_ext_volume(struct awacs_softc *sc, int left, int right) { int lval = 15 - (left & 0xff) / 16; int rval = 15 - (right & 0xff) / 16; DPRINTF("ext_volume %d %d\n", lval, rval); sc->sc_codecctl2 &= ~0x3cf; sc->sc_codecctl2 |= (lval << 6) | rval; awacs_write_codec(sc, sc->sc_codecctl2); } void awacs_set_rate(struct awacs_softc *sc, struct audio_params *p) { int selected = -1; size_t n, i; n = sizeof(awacs_speeds)/sizeof(awacs_speeds[0]); if (p->sample_rate < awacs_speeds[0].rate) selected = 0; if (p->sample_rate > awacs_speeds[n - 1].rate) selected = n - 1; for (i = 1; selected == -1 && i < n; i++) { if (p->sample_rate == awacs_speeds[i].rate) selected = i; else if (p->sample_rate < awacs_speeds[i].rate) { u_int diff1, diff2; diff1 = p->sample_rate - awacs_speeds[i - 1].rate; diff2 = awacs_speeds[i].rate - p->sample_rate; selected = (diff1 < diff2) ? i - 1 : i; } } if (selected == -1) selected = 0; sc->sc_soundctl &= ~AWACS_RATE_MASK; sc->sc_soundctl |= awacs_speeds[selected].bits; p->sample_rate = awacs_speeds[selected].rate; awacs_write_reg(sc, AWACS_SOUND_CTRL, sc->sc_soundctl); }