/* * TE410P Quad-T1/E1 PCI Driver version 0.1, 12/16/02 * * Written by Mark Spencer * Based on previous works, designs, and archetectures conceived and * written by Jim Dixon . * Further modified, optimized, and maintained by * Matthew Fredrickson and * Russ Meyerriecks * * Copyright (C) 2001 Jim Dixon / Zapata Telephony. * Copyright (C) 2001-2012, Digium, Inc. * * All rights reserved. * */ /* * See http://www.asterisk.org for more information about * the Asterisk project. Please do not directly contact * any of the maintainers of this project for assistance; * the project provides a web site, mailing lists and IRC * channels for your use. * * This program is free software, distributed under the terms of * the GNU General Public License Version 2 as published by the * Free Software Foundation. See the LICENSE file included with * this program for more details. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "wct4xxp.h" #include "vpm450m.h" /* Work queues are a way to better distribute load on SMP systems */ #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20)) /* * Work queues can significantly improve performance and scalability * on multi-processor machines, but requires bypassing some kernel * API's, so it's not guaranteed to be compatible with all kernels. */ /* #define ENABLE_WORKQUEUES */ #endif /* Support first generation cards? */ #define SUPPORT_GEN1 /* Define to get more attention-grabbing but slightly more I/O using alarm status */ #undef FANCY_ALARM /* Define to support Digium Voice Processing Module expansion card */ #define VPM_SUPPORT #define DEBUG_MAIN (1 << 0) #define DEBUG_DTMF (1 << 1) #define DEBUG_REGS (1 << 2) #define DEBUG_TSI (1 << 3) #define DEBUG_ECHOCAN (1 << 4) #define DEBUG_RBS (1 << 5) #define DEBUG_FRAMER (1 << 6) /* Maximum latency to be used with Gen 5 */ #define GEN5_MAX_LATENCY 127 #ifdef ENABLE_WORKQUEUES #include /* XXX UGLY!!!! XXX We have to access the direct structures of the workqueue which are only defined within workqueue.c because they don't give us a routine to allow us to nail a work to a particular thread of the CPU. Nailing to threads gives us substantially higher scalability in multi-CPU environments though! */ /* * The per-CPU workqueue (if single thread, we always use cpu 0's). * * The sequence counters are for flush_scheduled_work(). It wants to wait * until until all currently-scheduled works are completed, but it doesn't * want to be livelocked by new, incoming ones. So it waits until * remove_sequence is >= the insert_sequence which pertained when * flush_scheduled_work() was called. */ struct cpu_workqueue_struct { spinlock_t lock; long remove_sequence; /* Least-recently added (next to run) */ long insert_sequence; /* Next to add */ struct list_head worklist; wait_queue_head_t more_work; wait_queue_head_t work_done; struct workqueue_struct *wq; task_t *thread; int run_depth; /* Detect run_workqueue() recursion depth */ } ____cacheline_aligned; /* * The externally visible workqueue abstraction is an array of * per-CPU workqueues: */ struct workqueue_struct { /* TODO: Find out exactly where the API changed */ struct cpu_workqueue_struct *cpu_wq; const char *name; struct list_head list; /* Empty if single thread */ }; /* Preempt must be disabled. */ static void __t4_queue_work(struct cpu_workqueue_struct *cwq, struct work_struct *work) { unsigned long flags; spin_lock_irqsave(&cwq->lock, flags); work->wq_data = cwq; list_add_tail(&work->entry, &cwq->worklist); cwq->insert_sequence++; wake_up(&cwq->more_work); spin_unlock_irqrestore(&cwq->lock, flags); } /* * Queue work on a workqueue. Return non-zero if it was successfully * added. * * We queue the work to the CPU it was submitted, but there is no * guarantee that it will be processed by that CPU. */ static inline int t4_queue_work(struct workqueue_struct *wq, struct work_struct *work, int cpu) { int ret = 0; get_cpu(); if (!test_and_set_bit(0, &work->pending)) { BUG_ON(!list_empty(&work->entry)); __t4_queue_work(wq->cpu_wq + cpu, work); ret = 1; } put_cpu(); return ret; } #endif /* * Define CONFIG_FORCE_EXTENDED_RESET to allow the qfalc framer extra time * to reset itself upon hardware initialization. This exits for rare * cases for customers who are seeing the qfalc returning unexpected * information at initialization */ /* #define CONFIG_FORCE_EXTENDED_RESET */ /* #define CONFIG_NOEXTENDED_RESET */ /* * Uncomment the following definition in order to disable Active-State Power * Management on the PCIe bridge for PCIe cards. This has been known to work * around issues where the BIOS enables it on the cards even though the * platform does not support it. * */ /* #define CONFIG_WCT4XXP_DISABLE_ASPM */ #if defined(CONFIG_FORCE_EXTENDED_RESET) && defined(CONFIG_NOEXTENDED_RESET) #error "You cannot define both CONFIG_FORCE_EXTENDED_RESET and " \ "CONFIG_NOEXTENDED_RESET." #endif int debug = 0; static int timingcable = 0; static int t1e1override = -1; /* deprecated */ static char *default_linemode = "auto"; static int j1mode = 0; static int sigmode = FRMR_MODE_NO_ADDR_CMP; static int alarmdebounce = 2500; /* LOF/LFA def to 2.5s AT&T TR54016*/ static int losalarmdebounce = 2500;/* LOS def to 2.5s AT&T TR54016*/ static int aisalarmdebounce = 2500;/* AIS(blue) def to 2.5s AT&T TR54016*/ static int yelalarmdebounce = 500;/* RAI(yellow) def to 0.5s AT&T devguide */ static int max_latency = GEN5_MAX_LATENCY; /* Used to set a maximum latency (if you don't wish it to hard cap it at a certain value) in milliseconds */ #ifdef VPM_SUPPORT static int vpmsupport = 1; /* If set to auto, vpmdtmfsupport is enabled for VPM400M and disabled for VPM450M */ static int vpmdtmfsupport = -1; /* -1=auto, 0=disabled, 1=enabled*/ #endif /* VPM_SUPPORT */ /* Enabling bursting can more efficiently utilize PCI bus bandwidth, but can also cause PCI bus starvation, especially in combination with other aggressive cards. Please note that burst mode has no effect on CPU utilization / max number of calls / etc. */ static int noburst; /* For 56kbps links, set this module parameter to 0x7f */ static int hardhdlcmode = 0xff; static int latency = 1; static int ms_per_irq = 1; static int ignore_rotary; #ifdef FANCY_ALARM static int altab[] = { 0, 0, 0, 1, 2, 3, 4, 6, 8, 9, 11, 13, 16, 18, 20, 22, 24, 25, 27, 28, 29, 30, 31, 31, 32, 31, 31, 30, 29, 28, 27, 25, 23, 22, 20, 18, 16, 13, 11, 9, 8, 6, 4, 3, 2, 1, 0, 0, }; #endif #define MAX_SPANS 16 #define FLAG_STARTED (1 << 0) #define FLAG_NMF (1 << 1) #define FLAG_SENDINGYELLOW (1 << 2) #define FLAG_2NDGEN (1 << 3) #define FLAG_2PORT (1 << 4) #define FLAG_VPM2GEN (1 << 5) #define FLAG_OCTOPT (1 << 6) #define FLAG_3RDGEN (1 << 7) #define FLAG_BURST (1 << 8) #define FLAG_EXPRESS (1 << 9) #define FLAG_5THGEN (1 << 10) #define FLAG_8PORT (1 << 11) #define CANARY 0xc0de /* names of available HWEC modules */ #ifdef VPM_SUPPORT #define T4_VPM_PRESENT (1 << 28) static const char *vpmoct064_name = "VPMOCT064"; static const char *vpmoct128_name = "VPMOCT128"; static const char *vpmoct256_name = "VPMOCT256"; #endif struct devtype { char *desc; unsigned int flags; }; static struct devtype wct820p5 = { "Wildcard TE820 (5th Gen)", FLAG_8PORT | FLAG_5THGEN | FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN | FLAG_EXPRESS }; static struct devtype wct420p5 = { "Wildcard TE420 (5th Gen)", FLAG_5THGEN | FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN | FLAG_EXPRESS }; static struct devtype wct410p5 = { "Wildcard TE410P (5th Gen)", FLAG_5THGEN | FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN }; static struct devtype wct405p5 = { "Wildcard TE405P (5th Gen)", FLAG_5THGEN | FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN }; static struct devtype wct220p5 = { "Wildcard TE220 (5th Gen)", FLAG_5THGEN | FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN | FLAG_2PORT | FLAG_EXPRESS }; static struct devtype wct210p5 = { "Wildcard TE210P (5th Gen)", FLAG_5THGEN | FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN | FLAG_2PORT }; static struct devtype wct205p5 = { "Wildcard TE205P (5th Gen)", FLAG_5THGEN | FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN | FLAG_2PORT }; static struct devtype wct4xxp = { "Wildcard TE410P/TE405P (1st Gen)", 0 }; static struct devtype wct420p4 = { "Wildcard TE420 (4th Gen)", FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN | FLAG_EXPRESS }; static struct devtype wct410p4 = { "Wildcard TE410P (4th Gen)", FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN }; static struct devtype wct410p3 = { "Wildcard TE410P (3rd Gen)", FLAG_2NDGEN | FLAG_3RDGEN }; static struct devtype wct405p4 = { "Wildcard TE405P (4th Gen)", FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN }; static struct devtype wct405p3 = { "Wildcard TE405P (3rd Gen)", FLAG_2NDGEN | FLAG_3RDGEN }; static struct devtype wct410p2 = { "Wildcard TE410P (2nd Gen)", FLAG_2NDGEN }; static struct devtype wct405p2 = { "Wildcard TE405P (2nd Gen)", FLAG_2NDGEN }; static struct devtype wct220p4 = { "Wildcard TE220 (4th Gen)", FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN | FLAG_2PORT | FLAG_EXPRESS }; static struct devtype wct205p4 = { "Wildcard TE205P (4th Gen)", FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN | FLAG_2PORT }; static struct devtype wct205p3 = { "Wildcard TE205P (3rd Gen)", FLAG_2NDGEN | FLAG_3RDGEN | FLAG_2PORT }; static struct devtype wct210p4 = { "Wildcard TE210P (4th Gen)", FLAG_BURST | FLAG_2NDGEN | FLAG_3RDGEN | FLAG_2PORT }; static struct devtype wct210p3 = { "Wildcard TE210P (3rd Gen)", FLAG_2NDGEN | FLAG_3RDGEN | FLAG_2PORT }; static struct devtype wct205 = { "Wildcard TE205P ", FLAG_2NDGEN | FLAG_2PORT }; static struct devtype wct210 = { "Wildcard TE210P ", FLAG_2NDGEN | FLAG_2PORT }; struct t4; enum linemode {T1, E1, J1}; struct spi_state { int wrreg; int rdreg; }; struct t4_span { struct t4 *owner; u32 *writechunk; /* Double-word aligned write memory */ u32 *readchunk; /* Double-word aligned read memory */ enum linemode linemode; int sync; int alarmtimer; int notclear; unsigned long alarm_time; unsigned long losalarm_time; unsigned long aisalarm_time; unsigned long yelalarm_time; unsigned long alarmcheck_time; int spanflags; int syncpos; #ifdef SUPPORT_GEN1 int e1check; /* E1 check */ #endif struct dahdi_span span; unsigned char txsigs[16]; /* Transmit sigs */ int loopupcnt; int loopdowncnt; #ifdef SUPPORT_GEN1 unsigned char ec_chunk1[31][DAHDI_CHUNKSIZE]; /* first EC chunk buffer */ unsigned char ec_chunk2[31][DAHDI_CHUNKSIZE]; /* second EC chunk buffer */ #endif /* HDLC controller fields */ struct dahdi_chan *sigchan; unsigned char sigmode; int sigactive; int frames_out; int frames_in; #ifdef VPM_SUPPORT unsigned long dtmfactive; unsigned long dtmfmask; unsigned long dtmfmutemask; #endif #ifdef ENABLE_WORKQUEUES struct work_struct swork; #endif struct dahdi_chan *chans[32]; /* Individual channels */ struct dahdi_echocan_state *ec[32]; /* Echocan state for each channel */ }; struct t4 { /* This structure exists one per card */ struct pci_dev *dev; /* Pointer to PCI device */ unsigned int intcount; int num; /* Which card we are */ int syncsrc; /* active sync source */ struct dahdi_device *ddev; struct t4_span *tspans[8]; /* Individual spans */ int numspans; /* Number of spans on the card */ int blinktimer; #ifdef FANCY_ALARM int alarmpos; #endif int irq; /* IRQ used by device */ int order; /* Order */ const struct devtype *devtype; unsigned int reset_required:1; /* If reset needed in serial_setup */ unsigned int falc31:1; /* are we falc v3.1 (atomic not necessary) */ unsigned int t1e1:8; /* T1 / E1 select pins */ int ledreg; /* LED Register */ int ledreg2; /* LED Register2 */ unsigned int gpio; unsigned int gpioctl; int e1recover; /* E1 recovery timer */ spinlock_t reglock; /* lock register access */ int spansstarted; /* number of spans started */ u32 *writechunk; /* Double-word aligned write memory */ u32 *readchunk; /* Double-word aligned read memory */ #ifdef ENABLE_WORKQUEUES atomic_t worklist; struct workqueue_struct *workq; #endif int last0; /* for detecting double-missed IRQ */ /* DMA related fields */ unsigned int dmactrl; dma_addr_t readdma; dma_addr_t writedma; void __iomem *membase; /* Base address of card */ /* Flags for our bottom half */ unsigned long checkflag; struct tasklet_struct t4_tlet; /* Latency related additions */ unsigned char rxident; unsigned char lastindex; int numbufs; int needed_latency; #ifdef VPM_SUPPORT struct vpm450m *vpm; #endif struct spi_state st; }; static inline bool is_pcie(const struct t4 *wc) { return (wc->devtype->flags & FLAG_EXPRESS) > 0; } static inline bool has_e1_span(const struct t4 *wc) { return (wc->t1e1 > 0); } static inline bool is_octal(const struct t4 *wc) { return (wc->devtype->flags & FLAG_8PORT) > 0; } static inline int T4_BASE_SIZE(struct t4 *wc) { if (is_octal(wc)) return DAHDI_MAX_CHUNKSIZE * 32 * 8; else return DAHDI_MAX_CHUNKSIZE * 32 * 4; } /** * ports_on_framer - The number of ports on the framers. * @wc: Board to check. * * The framer ports could be different the the number of ports on the card * since the dual spans have four ports internally but two ports extenally. * */ static inline unsigned int ports_on_framer(const struct t4 *wc) { return (is_octal(wc)) ? 8 : 4; } #ifdef VPM_SUPPORT static void t4_vpm_init(struct t4 *wc); static void echocan_free(struct dahdi_chan *chan, struct dahdi_echocan_state *ec); static const struct dahdi_echocan_features vpm_ec_features = { .NLP_automatic = 1, .CED_tx_detect = 1, .CED_rx_detect = 1, }; static const struct dahdi_echocan_ops vpm_ec_ops = { .echocan_free = echocan_free, }; #endif static void __set_clear(struct t4 *wc, int span); static int _t4_startup(struct file *file, struct dahdi_span *span); static int t4_startup(struct file *file, struct dahdi_span *span); static int t4_shutdown(struct dahdi_span *span); static int t4_rbsbits(struct dahdi_chan *chan, int bits); static int t4_maint(struct dahdi_span *span, int cmd); static int t4_clear_maint(struct dahdi_span *span); static int t4_reset_counters(struct dahdi_span *span); #ifdef SUPPORT_GEN1 static int t4_reset_dma(struct t4 *wc); #endif static void t4_hdlc_hard_xmit(struct dahdi_chan *chan); static int t4_ioctl(struct dahdi_chan *chan, unsigned int cmd, unsigned long data); static void t4_tsi_assign(struct t4 *wc, int fromspan, int fromchan, int tospan, int tochan); static void t4_tsi_unassign(struct t4 *wc, int tospan, int tochan); static void __t4_set_rclk_src(struct t4 *wc, int span); static void __t4_set_sclk_src(struct t4 *wc, int mode, int master, int slave); static void t4_check_alarms(struct t4 *wc, int span); static void t4_check_sigbits(struct t4 *wc, int span); #define WC_RDADDR 0 #define WC_WRADDR 1 #define WC_COUNT 2 #define WC_DMACTRL 3 #define WC_INTR 4 /* #define WC_GPIO 5 */ #define WC_VERSION 6 #define WC_LEDS 7 #define WC_GPIOCTL 8 #define WC_GPIO 9 #define WC_LADDR 10 #define WC_LDATA 11 #define WC_LEDS2 12 #define WC_SET_AUTH (1 << 20) #define WC_GET_AUTH (1 << 12) #define WC_LFRMR_CS (1 << 10) /* Framer's ChipSelect signal */ #define WC_LCS (1 << 11) #define WC_LCS2 (1 << 12) #define WC_LALE (1 << 13) #define WC_LFRMR_CS2 (1 << 14) /* Framer's ChipSelect signal 2 */ #define WC_LREAD (1 << 15) #define WC_LWRITE (1 << 16) #define WC_ACTIVATE (1 << 12) #define WC_OFF (0) #define WC_RED (1) #define WC_GREEN (2) #define WC_YELLOW (3) #define WC_RECOVER 0 #define WC_SELF 1 #define LIM0_T 0x36 /* Line interface mode 0 register */ #define LIM0_LL (1 << 1) /* Local Loop */ #define LIM1_T 0x37 /* Line interface mode 1 register */ #define LIM1_RL (1 << 1) /* Remote Loop */ #define FMR0 0x1C /* Framer Mode Register 0 */ #define FMR0_SIM (1 << 0) /* Alarm Simulation */ #define FMR1_T 0x1D /* Framer Mode Register 1 */ #define FMR1_ECM (1 << 2) /* Error Counter 1sec Interrupt Enable */ #define FMR5 0x21 /* Framer Mode Register 5 */ #define FMR5_XLU (1 << 4) /* Transmit loopup code */ #define FMR5_XLD (1 << 5) /* Transmit loopdown code */ #define FMR5_EIBR (1 << 6) /* Internal Bit Robbing Access */ #define DEC_T 0x60 /* Diable Error Counter */ #define IERR_T 0x1B /* Single Bit Defect Insertion Register */ #define IBV (1 << 0) /* Bipolar violation */ #define IPE (1 << 1) /* PRBS defect */ #define ICASE (1 << 2) /* CAS defect */ #define ICRCE (1 << 3) /* CRC defect */ #define IMFE (1 << 4) /* Multiframe defect */ #define IFASE (1 << 5) /* FAS defect */ #define ISR3_SEC (1 << 6) /* Internal one-second interrupt bit mask */ #define ISR3_ES (1 << 7) /* Errored Second interrupt bit mask */ #define ESM 0x47 /* Errored Second mask register */ #define FMR2_T 0x1E /* Framer Mode Register 2 */ #define FMR2_PLB (1 << 2) /* Framer Mode Register 2 */ #define FECL_T 0x50 /* Framing Error Counter Lower Byte */ #define FECH_T 0x51 /* Framing Error Counter Higher Byte */ #define CVCL_T 0x52 /* Code Violation Counter Lower Byte */ #define CVCH_T 0x53 /* Code Violation Counter Higher Byte */ #define CEC1L_T 0x54 /* CRC Error Counter 1 Lower Byte */ #define CEC1H_T 0x55 /* CRC Error Counter 1 Higher Byte */ #define EBCL_T 0x56 /* E-Bit Error Counter Lower Byte */ #define EBCH_T 0x57 /* E-Bit Error Counter Higher Byte */ #define BECL_T 0x58 /* Bit Error Counter Lower Byte */ #define BECH_T 0x59 /* Bit Error Counter Higher Byte */ #define COEC_T 0x5A /* COFA Event Counter */ #define PRBSSTA_T 0xDA /* PRBS Status Register */ #define LCR1_T 0x3B /* Loop Code Register 1 */ #define EPRM (1 << 7) /* Enable PRBS rx */ #define XPRBS (1 << 6) /* Enable PRBS tx */ #define FLLB (1 << 1) /* Framed line loop/Invert */ #define LLBP (1 << 0) /* Line Loopback Pattern */ #define TPC0_T 0xA8 /* Test Pattern Control Register */ #define FRA (1 << 6) /* Framed/Unframed Selection */ #define PRBS23 (3 << 4) /* Pattern selection (23 poly) */ #define PRM (1 << 2) /* Non framed mode */ #define FRS1_T 0x4D /* Framer Receive Status Reg 1 */ #define LLBDD (1 << 4) #define LLBAD (1 << 3) #define MAX_T4_CARDS 64 static void t4_isr_bh(unsigned long data); static struct t4 *cards[MAX_T4_CARDS]; struct t8_firm_header { u8 header[6]; __le32 chksum; u8 pad[18]; __le32 version; } __packed; #define MAX_TDM_CHAN 32 #define MAX_DTMF_DET 16 #define HDLC_IMR0_MASK (FRMR_IMR0_RME | FRMR_IMR0_RPF) #define HDLC_IMR1_MASK (FRMR_IMR1_XDU | FRMR_IMR1_XPR) static inline unsigned int __t4_pci_in(struct t4 *wc, const unsigned int addr) { unsigned int res = readl(wc->membase + (addr * sizeof(u32))); return res; } static inline void __t4_pci_out(struct t4 *wc, const unsigned int addr, const unsigned int value) { #ifdef DEBUG unsigned int tmp; #endif writel(value, wc->membase + (addr * sizeof(u32))); #ifdef DEBUG tmp = __t4_pci_in(wc, WC_VERSION); if ((tmp & 0xffff0000) != 0xc01a0000) dev_notice(&wc->dev->dev, "Version Synchronization Error!\n"); #else __t4_pci_in(wc, WC_VERSION); #endif } static inline void __t4_gpio_set(struct t4 *wc, unsigned bits, unsigned int val) { unsigned int newgpio; newgpio = wc->gpio & (~bits); newgpio |= val; if (newgpio != wc->gpio) { wc->gpio = newgpio; __t4_pci_out(wc, WC_GPIO, wc->gpio); } } static inline void __t4_gpio_setdir(struct t4 *wc, unsigned int bits, unsigned int val) { unsigned int newgpioctl; newgpioctl = wc->gpioctl & (~bits); newgpioctl |= val; if (newgpioctl != wc->gpioctl) { wc->gpioctl = newgpioctl; __t4_pci_out(wc, WC_GPIOCTL, wc->gpioctl); } } static inline void t4_gpio_setdir(struct t4 *wc, unsigned int bits, unsigned int val) { unsigned long flags; spin_lock_irqsave(&wc->reglock, flags); __t4_gpio_setdir(wc, bits, val); spin_unlock_irqrestore(&wc->reglock, flags); } static inline void t4_gpio_set(struct t4 *wc, unsigned int bits, unsigned int val) { unsigned long flags; spin_lock_irqsave(&wc->reglock, flags); __t4_gpio_set(wc, bits, val); spin_unlock_irqrestore(&wc->reglock, flags); } static inline void t4_pci_out(struct t4 *wc, const unsigned int addr, const unsigned int value) { unsigned long flags; spin_lock_irqsave(&wc->reglock, flags); __t4_pci_out(wc, addr, value); spin_unlock_irqrestore(&wc->reglock, flags); } static inline void __t4_set_led(struct t4 *wc, int span, int color) { if (span <= 3) { int oldreg = wc->ledreg; wc->ledreg &= ~(0x3 << (span << 1)); wc->ledreg |= (color << (span << 1)); if (oldreg != wc->ledreg) __t4_pci_out(wc, WC_LEDS, wc->ledreg); } else { int oldreg = wc->ledreg2; span &= 3; wc->ledreg2 &= ~(0x3 << (span << 1)); wc->ledreg2 |= (color << (span << 1)); if (oldreg != wc->ledreg2) __t4_pci_out(wc, WC_LEDS2, wc->ledreg2); } } static inline void t4_activate(struct t4 *wc) { wc->ledreg |= WC_ACTIVATE; t4_pci_out(wc, WC_LEDS, wc->ledreg); } static inline unsigned int t4_pci_in(struct t4 *wc, const unsigned int addr) { unsigned int ret; unsigned long flags; spin_lock_irqsave(&wc->reglock, flags); ret = __t4_pci_in(wc, addr); spin_unlock_irqrestore(&wc->reglock, flags); return ret; } static unsigned int __t4_framer_in(const struct t4 *wc, int unit, const unsigned int addr) { unsigned int ret; register u32 val; void __iomem *const wc_laddr = wc->membase + (WC_LADDR*sizeof(u32)); void __iomem *const wc_version = wc->membase + (WC_VERSION*sizeof(u32)); void __iomem *const wc_ldata = wc->membase + (WC_LDATA*sizeof(u32)); int haddr = (((unit & 4) ? 0 : WC_LFRMR_CS2)); unit &= 0x3; val = ((unit & 0x3) << 8) | (addr & 0xff) | haddr; writel(val, wc_laddr); readl(wc_version); writel(val | WC_LFRMR_CS | WC_LREAD, wc_laddr); readl(wc_version); ret = readb(wc_ldata); writel(val, wc_laddr); readl(wc_version); return ret; } static unsigned int t4_framer_in(struct t4 *wc, int unit, const unsigned int addr) { unsigned long flags; unsigned int ret; spin_lock_irqsave(&wc->reglock, flags); ret = __t4_framer_in(wc, unit, addr); spin_unlock_irqrestore(&wc->reglock, flags); return ret; } static void __t4_framer_out(const struct t4 *wc, int unit, const u8 addr, const unsigned int value) { register u32 val; void __iomem *const wc_laddr = wc->membase + (WC_LADDR*sizeof(u32)); void __iomem *const wc_version = wc->membase + (WC_VERSION*sizeof(u32)); void __iomem *const wc_ldata = wc->membase + (WC_LDATA*sizeof(u32)); int haddr = (((unit & 4) ? 0 : WC_LFRMR_CS2)); val = ((unit & 0x3) << 8) | (addr & 0xff) | haddr; writel(val, wc_laddr); readl(wc_version); writel(value, wc_ldata); readl(wc_version); writel(val | WC_LFRMR_CS | WC_LWRITE, wc_laddr); readl(wc_version); writel(val, wc_laddr); readl(wc_version); } static void t4_framer_out(struct t4 *wc, int unit, const unsigned int addr, const unsigned int value) { unsigned long flags; spin_lock_irqsave(&wc->reglock, flags); __t4_framer_out(wc, unit, addr, value); spin_unlock_irqrestore(&wc->reglock, flags); } #ifdef VPM_SUPPORT static inline void __t4_raw_oct_out(struct t4 *wc, const unsigned int addr, const unsigned int value) { int octopt = wc->tspans[0]->spanflags & FLAG_OCTOPT; if (!octopt) __t4_gpio_set(wc, 0xff, (addr >> 8)); __t4_pci_out(wc, WC_LDATA, 0x10000 | (addr & 0xffff)); if (!octopt) __t4_pci_out(wc, WC_LADDR, (WC_LWRITE)); __t4_pci_out(wc, WC_LADDR, (WC_LWRITE | WC_LALE)); if (!octopt) __t4_gpio_set(wc, 0xff, (value >> 8)); __t4_pci_out(wc, WC_LDATA, (value & 0xffff)); __t4_pci_out(wc, WC_LADDR, (WC_LWRITE | WC_LALE | WC_LCS)); __t4_pci_out(wc, WC_LADDR, (0)); } static inline unsigned int __t4_raw_oct_in(struct t4 *wc, const unsigned int addr) { unsigned int ret; int octopt = wc->tspans[0]->spanflags & FLAG_OCTOPT; if (!octopt) __t4_gpio_set(wc, 0xff, (addr >> 8)); __t4_pci_out(wc, WC_LDATA, 0x10000 | (addr & 0xffff)); if (!octopt) __t4_pci_out(wc, WC_LADDR, (WC_LWRITE)); __t4_pci_out(wc, WC_LADDR, (WC_LWRITE | WC_LALE)); __t4_pci_out(wc, WC_LADDR, (WC_LALE)); if (!octopt) { __t4_gpio_setdir(wc, 0xff, 0x00); __t4_gpio_set(wc, 0xff, 0x00); } __t4_pci_out(wc, WC_LADDR, (WC_LREAD | WC_LALE | WC_LCS)); if (octopt) { ret = __t4_pci_in(wc, WC_LDATA) & 0xffff; } else { ret = __t4_pci_in(wc, WC_LDATA) & 0xff; ret |= (__t4_pci_in(wc, WC_GPIO) & 0xff) << 8; } __t4_pci_out(wc, WC_LADDR, (0)); if (!octopt) __t4_gpio_setdir(wc, 0xff, 0xff); return ret & 0xffff; } static inline unsigned int __t4_oct_in(struct t4 *wc, unsigned int addr) { #ifdef PEDANTIC_OCTASIC_CHECKING int count = 1000; #endif __t4_raw_oct_out(wc, 0x0008, (addr >> 20)); __t4_raw_oct_out(wc, 0x000a, (addr >> 4) & ((1 << 16) - 1)); __t4_raw_oct_out(wc, 0x0000, (((addr >> 1) & 0x7) << 9) | (1 << 8) | (1)); #ifdef PEDANTIC_OCTASIC_CHECKING while((__t4_raw_oct_in(wc, 0x0000) & (1 << 8)) && --count); if (count != 1000) dev_notice(&wc->dev->dev, "Yah, read can be slow...\n"); if (!count) dev_notice(&wc->dev->dev, "Read timed out!\n"); #endif return __t4_raw_oct_in(wc, 0x0004); } static inline unsigned int t4_oct_in(struct t4 *wc, const unsigned int addr) { unsigned long flags; unsigned int ret; spin_lock_irqsave(&wc->reglock, flags); ret = __t4_oct_in(wc, addr); spin_unlock_irqrestore(&wc->reglock, flags); return ret; } static inline void __t4_oct_out(struct t4 *wc, unsigned int addr, unsigned int value) { #ifdef PEDANTIC_OCTASIC_CHECKING int count = 1000; #endif __t4_raw_oct_out(wc, 0x0008, (addr >> 20)); __t4_raw_oct_out(wc, 0x000a, (addr >> 4) & ((1 << 16) - 1)); __t4_raw_oct_out(wc, 0x0004, value); __t4_raw_oct_out(wc, 0x0000, (((addr >> 1) & 0x7) << 9) | (1 << 8) | (3 << 12) | 1); #ifdef PEDANTIC_OCTASIC_CHECKING while((__t4_raw_oct_in(wc, 0x0000) & (1 << 8)) && --count); if (count != 1000) dev_notice(&wc->dev->dev, "Yah, write can be slow\n"); if (!count) dev_notice(&wc->dev->dev, "Write timed out!\n"); #endif } static inline void t4_oct_out(struct t4 *wc, const unsigned int addr, const unsigned int value) { unsigned long flags; spin_lock_irqsave(&wc->reglock, flags); __t4_oct_out(wc, addr, value); spin_unlock_irqrestore(&wc->reglock, flags); } static void t4_check_vpm(struct t4 *wc) { int channel, tone, start, span; if (vpm450m_checkirq(wc->vpm)) { while(vpm450m_getdtmf(wc->vpm, &channel, &tone, &start)) { span = channel & 0x3; channel >>= 2; if (!has_e1_span(wc)) channel -= 5; else channel -= 1; if (unlikely(debug)) dev_info(&wc->dev->dev, "Got tone %s of '%c' " "on channel %d of span %d\n", (start ? "START" : "STOP"), tone, channel, span + 1); if (test_bit(channel, &wc->tspans[span]->dtmfmask) && (tone != 'u')) { if (start) { /* The octasic is supposed to mute us, but... Yah, you guessed it. */ if (test_bit(channel, &wc->tspans[span]->dtmfmutemask)) { unsigned long flags; struct dahdi_chan *chan = wc->tspans[span]->span.chans[channel]; int y; spin_lock_irqsave(&chan->lock, flags); for (y=0;ynumbufs;y++) { if ((chan->inreadbuf > -1) && (chan->readidx[y])) memset(chan->readbuf[chan->inreadbuf], DAHDI_XLAW(0, chan), chan->readidx[y]); } spin_unlock_irqrestore(&chan->lock, flags); } set_bit(channel, &wc->tspans[span]->dtmfactive); dahdi_qevent_lock(wc->tspans[span]->span.chans[channel], (DAHDI_EVENT_DTMFDOWN | tone)); } else { clear_bit(channel, &wc->tspans[span]->dtmfactive); dahdi_qevent_lock(wc->tspans[span]->span.chans[channel], (DAHDI_EVENT_DTMFUP | tone)); } } } } } #endif /* VPM_SUPPORT */ static void hdlc_stop(struct t4 *wc, unsigned int span) { struct t4_span *t = wc->tspans[span]; unsigned char imr0, imr1, mode; unsigned long flags; int i = 0; if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "Stopping HDLC controller on span " "%d\n", span+1); /* Clear receive and transmit timeslots */ for (i = 0; i < 4; i++) { t4_framer_out(wc, span, FRMR_RTR_BASE + i, 0x00); t4_framer_out(wc, span, FRMR_TTR_BASE + i, 0x00); } spin_lock_irqsave(&wc->reglock, flags); imr0 = __t4_framer_in(wc, span, FRMR_IMR0); imr1 = __t4_framer_in(wc, span, FRMR_IMR1); /* Disable HDLC interrupts */ imr0 |= HDLC_IMR0_MASK; __t4_framer_out(wc, span, FRMR_IMR0, imr0); imr1 |= HDLC_IMR1_MASK; __t4_framer_out(wc, span, FRMR_IMR1, imr1); mode = __t4_framer_in(wc, span, FRMR_MODE); mode &= ~FRMR_MODE_HRAC; __t4_framer_out(wc, span, FRMR_MODE, mode); spin_unlock_irqrestore(&wc->reglock, flags); t->sigactive = 0; } static inline void __t4_framer_cmd(struct t4 *wc, unsigned int span, int cmd) { __t4_framer_out(wc, span, FRMR_CMDR, cmd); } static inline void t4_framer_cmd_wait(struct t4 *wc, unsigned int span, int cmd) { int sis; int loops = 0; /* XXX could be time consuming XXX */ for (;;) { sis = t4_framer_in(wc, span, FRMR_SIS); if (!(sis & 0x04)) break; if (!loops++ && (debug & DEBUG_FRAMER)) { dev_notice(&wc->dev->dev, "!!!SIS Waiting before cmd " "%02x\n", cmd); } } if (loops && (debug & DEBUG_FRAMER)) dev_notice(&wc->dev->dev, "!!!SIS waited %d loops\n", loops); t4_framer_out(wc, span, FRMR_CMDR, cmd); } static int hdlc_start(struct t4 *wc, unsigned int span, struct dahdi_chan *chan, unsigned char mode) { struct t4_span *t = wc->tspans[span]; unsigned char imr0, imr1; int offset = chan->chanpos; unsigned long flags; if (debug & DEBUG_FRAMER) dev_info(&wc->dev->dev, "Starting HDLC controller for channel " "%d span %d\n", offset, span+1); if (mode != FRMR_MODE_NO_ADDR_CMP) return -1; mode |= FRMR_MODE_HRAC; spin_lock_irqsave(&wc->reglock, flags); /* Make sure we're in the right mode */ __t4_framer_out(wc, span, FRMR_MODE, mode); __t4_framer_out(wc, span, FRMR_TSEO, 0x00); __t4_framer_out(wc, span, FRMR_TSBS1, hardhdlcmode); /* Set the interframe gaps, etc */ __t4_framer_out(wc, span, FRMR_CCR1, FRMR_CCR1_ITF|FRMR_CCR1_EITS); __t4_framer_out(wc, span, FRMR_CCR2, FRMR_CCR2_RCRC); /* Set up the time slot that we want to tx/rx on */ __t4_framer_out(wc, span, FRMR_TTR_BASE + (offset / 8), (0x80 >> (offset % 8))); __t4_framer_out(wc, span, FRMR_RTR_BASE + (offset / 8), (0x80 >> (offset % 8))); imr0 = __t4_framer_in(wc, span, FRMR_IMR0); imr1 = __t4_framer_in(wc, span, FRMR_IMR1); /* Enable our interrupts again */ imr0 &= ~HDLC_IMR0_MASK; __t4_framer_out(wc, span, FRMR_IMR0, imr0); imr1 &= ~HDLC_IMR1_MASK; __t4_framer_out(wc, span, FRMR_IMR1, imr1); spin_unlock_irqrestore(&wc->reglock, flags); /* Reset the signaling controller */ t4_framer_cmd_wait(wc, span, FRMR_CMDR_SRES); spin_lock_irqsave(&wc->reglock, flags); t->sigchan = chan; spin_unlock_irqrestore(&wc->reglock, flags); t->sigactive = 0; return 0; } static void __set_clear(struct t4 *wc, int span) { int i,j; int oldnotclear; unsigned short val=0; struct t4_span *ts = wc->tspans[span]; oldnotclear = ts->notclear; if (E1 != ts->linemode) { for (i=0;i<24;i++) { j = (i/8); if (ts->span.chans[i]->flags & DAHDI_FLAG_CLEAR) { val |= 1 << (7 - (i % 8)); ts->notclear &= ~(1 << i); } else ts->notclear |= (1 << i); if ((i % 8)==7) { if (debug) dev_notice(&wc->dev->dev, "Putting %d " "in register %02x on span %d" "\n", val, 0x2f + j, span + 1); __t4_framer_out(wc, span, 0x2f + j, val); val = 0; } } } else { for (i=0;i<31;i++) { if (ts->span.chans[i]->flags & DAHDI_FLAG_CLEAR) ts->notclear &= ~(1 << i); else ts->notclear |= (1 << i); } } if (ts->notclear != oldnotclear) { unsigned char reg; reg = __t4_framer_in(wc, span, FRMR_IMR0); if (ts->notclear) reg &= ~0x08; else reg |= 0x08; __t4_framer_out(wc, span, FRMR_IMR0, reg); } } static int t4_dacs(struct dahdi_chan *dst, struct dahdi_chan *src) { struct t4 *wc; struct t4_span *ts; wc = dst->pvt; ts = wc->tspans[dst->span->offset]; if (src && (src->pvt != dst->pvt)) { if (ts->spanflags & FLAG_2NDGEN) t4_tsi_unassign(wc, dst->span->offset, dst->chanpos); wc = src->pvt; if (ts->spanflags & FLAG_2NDGEN) t4_tsi_unassign(wc, src->span->offset, src->chanpos); if (debug) dev_notice(&wc->dev->dev, "Unassigning %d/%d by " "default and...\n", src->span->offset, src->chanpos); if (debug) dev_notice(&wc->dev->dev, "Unassigning %d/%d by " "default\n", dst->span->offset, dst->chanpos); return -1; } if (src) { t4_tsi_assign(wc, src->span->offset, src->chanpos, dst->span->offset, dst->chanpos); if (debug) dev_notice(&wc->dev->dev, "Assigning channel %d/%d -> " "%d/%d!\n", src->span->offset, src->chanpos, dst->span->offset, dst->chanpos); } else { t4_tsi_unassign(wc, dst->span->offset, dst->chanpos); if (debug) dev_notice(&wc->dev->dev, "Unassigning channel %d/%d!" "\n", dst->span->offset, dst->chanpos); } return 0; } #ifdef VPM_SUPPORT void oct_set_reg(void *data, unsigned int reg, unsigned int val) { struct t4 *wc = data; t4_oct_out(wc, reg, val); } unsigned int oct_get_reg(void *data, unsigned int reg) { struct t4 *wc = data; unsigned int ret; ret = t4_oct_in(wc, reg); return ret; } static const char *__t4_echocan_name(struct t4 *wc) { if (wc->vpm) { if (wc->numspans == 2) return vpmoct064_name; else if (wc->numspans == 4) return vpmoct128_name; else if (wc->numspans == 8) return vpmoct256_name; } return NULL; } static const char *t4_echocan_name(const struct dahdi_chan *chan) { struct t4 *wc = chan->pvt; return __t4_echocan_name(wc); } static int t4_echocan_create(struct dahdi_chan *chan, struct dahdi_echocanparams *ecp, struct dahdi_echocanparam *p, struct dahdi_echocan_state **ec) { struct t4 *wc = chan->pvt; struct t4_span *tspan = container_of(chan->span, struct t4_span, span); int channel; const bool alaw = (chan->span->deflaw == 2); if (!vpmsupport || !wc->vpm) return -ENODEV; if (ecp->param_count > 0) { dev_warn(&wc->dev->dev, "%s echo canceller does not support " "parameters; failing request\n", chan->ec_factory->get_name(chan)); return -EINVAL; } *ec = tspan->ec[chan->chanpos - 1]; (*ec)->ops = &vpm_ec_ops; (*ec)->features = vpm_ec_features; channel = has_e1_span(wc) ? chan->chanpos : chan->chanpos + 4; if (is_octal(wc)) channel = channel << 3; else channel = channel << 2; channel |= chan->span->offset; if (debug & DEBUG_ECHOCAN) { dev_notice(&wc->dev->dev, "echocan: Card is %d, Channel is %d, Span is %d, offset is %d length %d\n", wc->num, chan->chanpos, chan->span->offset, channel, ecp->tap_length); } vpm450m_set_alaw_companding(wc->vpm, channel, alaw); vpm450m_setec(wc->vpm, channel, ecp->tap_length); return 0; } static void echocan_free(struct dahdi_chan *chan, struct dahdi_echocan_state *ec) { struct t4 *wc = chan->pvt; int channel; if (!wc->vpm) return; memset(ec, 0, sizeof(*ec)); channel = has_e1_span(wc) ? chan->chanpos : chan->chanpos + 4; if (is_octal(wc)) channel = channel << 3; else channel = channel << 2; channel |= chan->span->offset; if (debug & DEBUG_ECHOCAN) { dev_notice(&wc->dev->dev, "echocan: Card is %d, Channel is %d, Span is %d, offset is %d length 0\n", wc->num, chan->chanpos, chan->span->offset, channel); } vpm450m_setec(wc->vpm, channel, 0); } #endif static int t4_ioctl(struct dahdi_chan *chan, unsigned int cmd, unsigned long data) { struct t4_regs regs; struct t4_reg reg; int x; struct t4 *wc = chan->pvt; #ifdef VPM_SUPPORT int j; int channel; struct t4_span *ts = wc->tspans[chan->span->offset]; #endif switch(cmd) { case WCT4_SET_REG: if (copy_from_user(®, (struct t4_reg __user *)data, sizeof(reg))) return -EFAULT; t4_pci_out(wc, reg.reg, reg.val); break; case WCT4_GET_REG: if (copy_from_user(®, (struct t4_reg __user *)data, sizeof(reg))) return -EFAULT; reg.val = t4_pci_in(wc, reg.reg); if (copy_to_user((struct t4_reg __user *)data, ®, sizeof(reg))) return -EFAULT; break; case WCT4_GET_REGS: for (x=0;xspan->offset, x); if (copy_to_user((void __user *) data, ®s, sizeof(regs))) return -EFAULT; break; #ifdef VPM_SUPPORT case DAHDI_TONEDETECT: if (get_user(j, (__user int *) data)) return -EFAULT; if (!wc->vpm) return -ENOSYS; if (j && (vpmdtmfsupport == 0)) return -ENOSYS; if (j & DAHDI_TONEDETECT_ON) set_bit(chan->chanpos - 1, &ts->dtmfmask); else clear_bit(chan->chanpos - 1, &ts->dtmfmask); if (j & DAHDI_TONEDETECT_MUTE) set_bit(chan->chanpos - 1, &ts->dtmfmutemask); else clear_bit(chan->chanpos - 1, &ts->dtmfmutemask); channel = has_e1_span(wc) ? chan->chanpos : chan->chanpos + 4; if (is_octal(wc)) channel = channel << 3; else channel = channel << 2; channel |= chan->span->offset; vpm450m_setdtmf(wc->vpm, channel, j & DAHDI_TONEDETECT_ON, j & DAHDI_TONEDETECT_MUTE); return 0; #endif default: return -ENOTTY; } return 0; } static void inline t4_hdlc_xmit_fifo(struct t4 *wc, unsigned int span, struct t4_span *ts) { int res, i; unsigned int size = 32; unsigned char buf[32]; res = dahdi_hdlc_getbuf(ts->sigchan, buf, &size); if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "Got buffer sized %d and res %d " "for %d\n", size, res, span); if (size > 0) { ts->sigactive = 1; if (debug & DEBUG_FRAMER) { dev_notice(&wc->dev->dev, "TX("); for (i = 0; i < size; i++) dev_notice(&wc->dev->dev, "%s%02x", (i ? " " : ""), buf[i]); dev_notice(&wc->dev->dev, ")\n"); } for (i = 0; i < size; i++) t4_framer_out(wc, span, FRMR_TXFIFO, buf[i]); if (res) /* End of message */ { if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "transmiting XHF|XME\n"); t4_framer_cmd_wait(wc, span, FRMR_CMDR_XHF | FRMR_CMDR_XME); ++ts->frames_out; if ((debug & DEBUG_FRAMER) && !(ts->frames_out & 0x0f)) dev_notice(&wc->dev->dev, "Transmitted %d " "frames on span %d\n", ts->frames_out, span); } else { /* Still more to transmit */ if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "transmiting XHF\n"); t4_framer_cmd_wait(wc, span, FRMR_CMDR_XHF); } } else if (res < 0) ts->sigactive = 0; } static void t4_hdlc_hard_xmit(struct dahdi_chan *chan) { struct t4 *wc = chan->pvt; int span = chan->span->offset; struct t4_span *ts = wc->tspans[span]; unsigned long flags; spin_lock_irqsave(&wc->reglock, flags); if (!ts->sigchan) { dev_notice(&wc->dev->dev, "t4_hdlc_hard_xmit: Invalid (NULL) " "signalling channel\n"); spin_unlock_irqrestore(&wc->reglock, flags); return; } spin_unlock_irqrestore(&wc->reglock, flags); if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "t4_hdlc_hard_xmit on channel %s " "(sigchan %s), sigactive=%d\n", chan->name, ts->sigchan->name, ts->sigactive); if ((ts->sigchan == chan) && !ts->sigactive) t4_hdlc_xmit_fifo(wc, span, ts); } /** * t4_set_framer_bits - Atomically set bits in a framer register. */ static void t4_set_framer_bits(struct t4 *wc, unsigned int spanno, unsigned int const addr, u16 bits) { unsigned long flags; unsigned int reg; spin_lock_irqsave(&wc->reglock, flags); reg = __t4_framer_in(wc, spanno, addr); __t4_framer_out(wc, spanno, addr, (reg | bits)); spin_unlock_irqrestore(&wc->reglock, flags); } static int t4_maint(struct dahdi_span *span, int cmd) { struct t4_span *ts = container_of(span, struct t4_span, span); struct t4 *wc = ts->owner; unsigned int reg; unsigned long flags; if (E1 == ts->linemode) { switch(cmd) { case DAHDI_MAINT_NONE: dev_info(&wc->dev->dev, "Clearing all maint modes\n"); t4_clear_maint(span); break; case DAHDI_MAINT_LOCALLOOP: dev_info(&wc->dev->dev, "Turning on local loopback\n"); t4_clear_maint(span); t4_set_framer_bits(wc, span->offset, LIM0_T, LIM0_LL); break; case DAHDI_MAINT_NETWORKLINELOOP: dev_info(&wc->dev->dev, "Turning on network line loopback\n"); t4_clear_maint(span); t4_set_framer_bits(wc, span->offset, LIM1_T, LIM1_RL); break; case DAHDI_MAINT_NETWORKPAYLOADLOOP: dev_info(&wc->dev->dev, "Turning on network payload loopback\n"); t4_clear_maint(span); t4_set_framer_bits(wc, span->offset, FMR2_T, FMR2_PLB); break; case DAHDI_MAINT_LOOPUP: case DAHDI_MAINT_LOOPDOWN: dev_info(&wc->dev->dev, "Loopup & loopdown not supported in E1 mode\n"); return -ENOSYS; case DAHDI_MAINT_FAS_DEFECT: t4_framer_out(wc, span->offset, IERR_T, IFASE); break; case DAHDI_MAINT_MULTI_DEFECT: t4_framer_out(wc, span->offset, IERR_T, IMFE); break; case DAHDI_MAINT_CRC_DEFECT: t4_framer_out(wc, span->offset, IERR_T, ICRCE); break; case DAHDI_MAINT_CAS_DEFECT: t4_framer_out(wc, span->offset, IERR_T, ICASE); break; case DAHDI_MAINT_PRBS_DEFECT: t4_framer_out(wc, span->offset, IERR_T, IPE); break; case DAHDI_MAINT_BIPOLAR_DEFECT: t4_framer_out(wc, span->offset, IERR_T, IBV); break; case DAHDI_RESET_COUNTERS: t4_reset_counters(span); break; case DAHDI_MAINT_ALARM_SIM: dev_info(&wc->dev->dev, "Invoking alarm state"); t4_set_framer_bits(wc, span->offset, FMR0, FMR0_SIM); break; default: dev_info(&wc->dev->dev, "Unknown E1 maint command: %d\n", cmd); return -ENOSYS; } } else { switch(cmd) { case DAHDI_MAINT_NONE: dev_info(&wc->dev->dev, "Clearing all maint modes\n"); t4_clear_maint(span); break; case DAHDI_MAINT_LOCALLOOP: dev_info(&wc->dev->dev, "Turning on local loopback\n"); t4_clear_maint(span); t4_set_framer_bits(wc, span->offset, LIM0_T, LIM0_LL); break; case DAHDI_MAINT_NETWORKLINELOOP: dev_info(&wc->dev->dev, "Turning on network line loopback\n"); t4_clear_maint(span); t4_set_framer_bits(wc, span->offset, LIM1_T, LIM1_RL); break; case DAHDI_MAINT_NETWORKPAYLOADLOOP: dev_info(&wc->dev->dev, "Turning on network payload loopback\n"); t4_clear_maint(span); t4_set_framer_bits(wc, span->offset, FMR2_T, FMR2_PLB); break; case DAHDI_MAINT_LOOPUP: dev_info(&wc->dev->dev, "Transmitting loopup code\n"); t4_clear_maint(span); t4_set_framer_bits(wc, span->offset, FMR5, FMR5_XLU); ts->span.maintstat = DAHDI_MAINT_REMOTELOOP; break; case DAHDI_MAINT_LOOPDOWN: dev_info(&wc->dev->dev, "Transmitting loopdown code\n"); t4_clear_maint(span); t4_set_framer_bits(wc, span->offset, FMR5, FMR5_XLD); break; case DAHDI_MAINT_FAS_DEFECT: t4_framer_out(wc, span->offset, IERR_T, IFASE); break; case DAHDI_MAINT_MULTI_DEFECT: t4_framer_out(wc, span->offset, IERR_T, IMFE); break; case DAHDI_MAINT_CRC_DEFECT: t4_framer_out(wc, span->offset, IERR_T, ICRCE); break; case DAHDI_MAINT_CAS_DEFECT: t4_framer_out(wc, span->offset, IERR_T, ICASE); break; case DAHDI_MAINT_PRBS_DEFECT: t4_framer_out(wc, span->offset, IERR_T, IPE); break; case DAHDI_MAINT_BIPOLAR_DEFECT: t4_framer_out(wc, span->offset, IERR_T, IBV); break; case DAHDI_MAINT_PRBS: dev_info(&wc->dev->dev, "PRBS not supported\n"); return -ENOSYS; case DAHDI_RESET_COUNTERS: t4_reset_counters(span); break; case DAHDI_MAINT_ALARM_SIM: spin_lock_irqsave(&wc->reglock, flags); reg = __t4_framer_in(wc, span->offset, FMR0); /* * The alarm simulation state machine requires us to * bring this bit up and down for at least 1 clock cycle */ __t4_framer_out(wc, span->offset, FMR0, (reg | FMR0_SIM)); udelay(1); __t4_framer_out(wc, span->offset, FMR0, (reg & ~FMR0_SIM)); udelay(1); spin_unlock_irqrestore(&wc->reglock, flags); reg = t4_framer_in(wc, span->offset, 0x4e); if (debug & DEBUG_MAIN) { dev_info(&wc->dev->dev, "FRS2(alarm state): %d\n", ((reg & 0xe0) >> 5)); } break; default: dev_info(&wc->dev->dev, "Unknown T1 maint command:%d\n", cmd); break; } } return 0; } static int t4_clear_maint(struct dahdi_span *span) { struct t4_span *ts = container_of(span, struct t4_span, span); struct t4 *wc = ts->owner; unsigned int reg; unsigned long flags; spin_lock_irqsave(&wc->reglock, flags); /* Clear local loop */ reg = __t4_framer_in(wc, span->offset, LIM0_T); __t4_framer_out(wc, span->offset, LIM0_T, (reg & ~LIM0_LL)); /* Clear Remote Loop */ reg = __t4_framer_in(wc, span->offset, LIM1_T); __t4_framer_out(wc, span->offset, LIM1_T, (reg & ~LIM1_RL)); /* Clear Remote Payload Loop */ reg = __t4_framer_in(wc, span->offset, FMR2_T); __t4_framer_out(wc, span->offset, FMR2_T, (reg & ~FMR2_PLB)); /* Clear PRBS */ reg = __t4_framer_in(wc, span->offset, LCR1_T); __t4_framer_out(wc, span->offset, LCR1_T, (reg & ~(XPRBS | EPRM))); /* Clear loopup/loopdown signals on the line */ reg = __t4_framer_in(wc, span->offset, FMR5); __t4_framer_out(wc, span->offset, FMR5, (reg & ~(FMR5_XLU | FMR5_XLD))); spin_unlock_irqrestore(&wc->reglock, flags); span->mainttimer = 0; return 0; } static int t4_reset_counters(struct dahdi_span *span) { struct t4_span *ts = container_of(span, struct t4_span, span); memset(&ts->span.count, 0, sizeof(ts->span.count)); return 0; } static int t4_rbsbits(struct dahdi_chan *chan, int bits) { u_char m,c; int k,n,b; struct t4 *wc = chan->pvt; struct t4_span *ts = wc->tspans[chan->span->offset]; unsigned long flags; if (debug & DEBUG_RBS) dev_notice(&wc->dev->dev, "Setting bits to %d on channel %s\n", bits, chan->name); spin_lock_irqsave(&wc->reglock, flags); k = chan->span->offset; if (E1 == ts->linemode) { if (chan->chanpos == 16) { spin_unlock_irqrestore(&wc->reglock, flags); return 0; } n = chan->chanpos - 1; if (chan->chanpos > 15) n--; b = (n % 15); c = ts->txsigs[b]; m = (n / 15) << 2; /* nibble selector */ c &= (0xf << m); /* keep the other nibble */ c |= (bits & 0xf) << (4 - m); /* put our new nibble here */ ts->txsigs[b] = c; /* output them to the chip */ __t4_framer_out(wc,k,0x71 + b,c); } else if (ts->span.lineconfig & DAHDI_CONFIG_D4) { n = chan->chanpos - 1; b = (n/4); c = ts->txsigs[b]; m = ((3 - (n % 4)) << 1); /* nibble selector */ c &= ~(0x3 << m); /* keep the other nibble */ c |= ((bits >> 2) & 0x3) << m; /* put our new nibble here */ ts->txsigs[b] = c; /* output them to the chip */ __t4_framer_out(wc,k,0x70 + b,c); __t4_framer_out(wc,k,0x70 + b + 6,c); } else if (ts->span.lineconfig & DAHDI_CONFIG_ESF) { n = chan->chanpos - 1; b = (n/2); c = ts->txsigs[b]; m = ((n % 2) << 2); /* nibble selector */ c &= (0xf << m); /* keep the other nibble */ c |= (bits & 0xf) << (4 - m); /* put our new nibble here */ ts->txsigs[b] = c; /* output them to the chip */ __t4_framer_out(wc,k,0x70 + b,c); } spin_unlock_irqrestore(&wc->reglock, flags); if (debug & DEBUG_RBS) dev_notice(&wc->dev->dev, "Finished setting RBS bits\n"); return 0; } static int t4_shutdown(struct dahdi_span *span) { int tspan; int wasrunning; unsigned long flags; struct t4_span *ts = container_of(span, struct t4_span, span); struct t4 *wc = ts->owner; tspan = span->offset + 1; if (tspan < 0) { dev_notice(&wc->dev->dev, "T%dXXP: Span '%d' isn't us?\n", wc->numspans, span->spanno); return -1; } if (debug & DEBUG_MAIN) dev_notice(&wc->dev->dev, "Shutting down span %d (%s)\n", span->spanno, span->name); /* Stop HDLC controller if runned */ if (ts->sigchan) hdlc_stop(wc, span->offset); spin_lock_irqsave(&wc->reglock, flags); wasrunning = span->flags & DAHDI_FLAG_RUNNING; span->flags &= ~DAHDI_FLAG_RUNNING; __t4_set_led(wc, span->offset, WC_OFF); if (((wc->numspans == 8) && (!(wc->tspans[0]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[1]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[2]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[3]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[4]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[5]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[6]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[7]->span.flags & DAHDI_FLAG_RUNNING))) || ((wc->numspans == 4) && (!(wc->tspans[0]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[1]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[2]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[3]->span.flags & DAHDI_FLAG_RUNNING))) || ((wc->numspans == 2) && (!(wc->tspans[0]->span.flags & DAHDI_FLAG_RUNNING)) && (!(wc->tspans[1]->span.flags & DAHDI_FLAG_RUNNING)))) { /* No longer in use, disable interrupts */ dev_info(&wc->dev->dev, "TE%dXXP: Disabling interrupts since " "there are no active spans\n", wc->numspans); set_bit(T4_STOP_DMA, &wc->checkflag); } else set_bit(T4_CHECK_TIMING, &wc->checkflag); spin_unlock_irqrestore(&wc->reglock, flags); /* Wait for interrupt routine to shut itself down */ msleep(10); if (wasrunning) wc->spansstarted--; if (debug & DEBUG_MAIN) dev_notice(&wc->dev->dev, "Span %d (%s) shutdown\n", span->spanno, span->name); return 0; } static void t4_chan_set_sigcap(struct dahdi_span *span, int x) { struct t4_span *wc = container_of(span, struct t4_span, span); struct dahdi_chan *chan = wc->chans[x]; chan->sigcap = DAHDI_SIG_CLEAR; /* E&M variant supported depends on span type */ if (E1 == wc->linemode) { /* E1 sigcap setup */ if (span->lineconfig & DAHDI_CONFIG_CCS) { /* CCS setup */ chan->sigcap |= DAHDI_SIG_MTP2 | DAHDI_SIG_SF | DAHDI_SIG_HARDHDLC; return; } /* clear out sig and sigcap for channel 16 on E1 CAS * lines, otherwise, set it correctly */ if (x == 15) { /* CAS signaling channel setup */ wc->chans[15]->sigcap = 0; wc->chans[15]->sig = 0; return; } /* normal CAS setup */ chan->sigcap |= DAHDI_SIG_EM_E1 | DAHDI_SIG_FXSLS | DAHDI_SIG_FXSGS | DAHDI_SIG_FXSKS | DAHDI_SIG_SF | DAHDI_SIG_FXOLS | DAHDI_SIG_FXOGS | DAHDI_SIG_FXOKS | DAHDI_SIG_CAS | DAHDI_SIG_DACS_RBS; } else { /* T1 sigcap setup */ chan->sigcap |= DAHDI_SIG_EM | DAHDI_SIG_FXSLS | DAHDI_SIG_FXSGS | DAHDI_SIG_FXSKS | DAHDI_SIG_MTP2 | DAHDI_SIG_SF | DAHDI_SIG_FXOLS | DAHDI_SIG_FXOGS | DAHDI_SIG_FXOKS | DAHDI_SIG_CAS | DAHDI_SIG_DACS_RBS | DAHDI_SIG_HARDHDLC; } } static int _t4_spanconfig(struct file *file, struct dahdi_span *span, struct dahdi_lineconfig *lc) { int i; struct t4_span *ts = container_of(span, struct t4_span, span); struct t4 *wc = ts->owner; if (debug) dev_info(&wc->dev->dev, "About to enter spanconfig!\n"); if (debug & DEBUG_MAIN) dev_notice(&wc->dev->dev, "TE%dXXP: Configuring span %d\n", wc->numspans, span->spanno); if (lc->sync < 0) lc->sync = 0; if (lc->sync > wc->numspans) { dev_warn(&wc->dev->dev, "WARNING: Cannot set priority on span %d to %d. Please set to a number between 1 and %d\n", span->spanno, lc->sync, wc->numspans); lc->sync = 0; } /* remove this span number from the current sync sources, if there */ for(i = 0; i < wc->numspans; i++) { if (wc->tspans[i]->sync == span->spanno) wc->tspans[i]->sync = 0; } wc->tspans[span->offset]->syncpos = lc->sync; /* if a sync src, put it in proper place */ if (lc->sync) wc->tspans[lc->sync - 1]->sync = span->spanno; set_bit(T4_CHECK_TIMING, &wc->checkflag); /* Make sure this is clear in case of multiple startup and shutdown * iterations */ clear_bit(T4_STOP_DMA, &wc->checkflag); /* make sure that sigcaps gets updated if necessary */ for (i = 0; i < span->channels; i++) t4_chan_set_sigcap(span, i); /* If we're already running, then go ahead and apply the changes */ if (span->flags & DAHDI_FLAG_RUNNING) return _t4_startup(file, span); if (debug) dev_info(&wc->dev->dev, "Done with spanconfig!\n"); return 0; } static int t4_spanconfig(struct file *file, struct dahdi_span *span, struct dahdi_lineconfig *lc) { int ret; struct dahdi_device *const ddev = span->parent; struct dahdi_span *s; ret = _t4_spanconfig(file, span, lc); /* Make sure all the spans have a basic configuration in case they are * not all specified in the configuration files. */ lc->sync = 0; list_for_each_entry(s, &ddev->spans, device_node) { WARN_ON(!s->channels); if (!s->channels) continue; if (!s->chans[0]->sigcap) _t4_spanconfig(file, s, lc); } return ret; } static int t4_chanconfig(struct file *file, struct dahdi_chan *chan, int sigtype) { int alreadyrunning; unsigned long flags; struct t4 *wc = chan->pvt; struct t4_span *ts = wc->tspans[chan->span->offset]; alreadyrunning = ts->span.flags & DAHDI_FLAG_RUNNING; if (debug & DEBUG_MAIN) { if (alreadyrunning) dev_notice(&wc->dev->dev, "TE%dXXP: Reconfigured " "channel %d (%s) sigtype %d\n", wc->numspans, chan->channo, chan->name, sigtype); else dev_notice(&wc->dev->dev, "TE%dXXP: Configured channel" " %d (%s) sigtype %d\n", wc->numspans, chan->channo, chan->name, sigtype); } spin_lock_irqsave(&wc->reglock, flags); if (alreadyrunning) __set_clear(wc, chan->span->offset); spin_unlock_irqrestore(&wc->reglock, flags); /* (re)configure signalling channel */ if ((sigtype == DAHDI_SIG_HARDHDLC) || (ts->sigchan == chan)) { if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "%sonfiguring hardware HDLC " "on %s\n", ((sigtype == DAHDI_SIG_HARDHDLC) ? "C" : "Unc"), chan->name); if (alreadyrunning) { if (ts->sigchan) hdlc_stop(wc, ts->sigchan->span->offset); if (sigtype == DAHDI_SIG_HARDHDLC) { if (hdlc_start(wc, chan->span->offset, chan, ts->sigmode)) { dev_notice(&wc->dev->dev, "Error " "initializing signalling " "controller\n"); return -1; } } else { spin_lock_irqsave(&wc->reglock, flags); ts->sigchan = NULL; spin_unlock_irqrestore(&wc->reglock, flags); } } else { spin_lock_irqsave(&wc->reglock, flags); ts->sigchan = (sigtype == DAHDI_SIG_HARDHDLC) ? chan : NULL; spin_unlock_irqrestore(&wc->reglock, flags); ts->sigactive = 0; } } return 0; } static int t4_open(struct dahdi_chan *chan) { return 0; } static int t4_close(struct dahdi_chan *chan) { return 0; } static int set_span_devicetype(struct t4 *wc) { #ifdef VPM_SUPPORT const char *vpmstring = __t4_echocan_name(wc); if (vpmstring) { wc->ddev->devicetype = kasprintf(GFP_KERNEL, "%s (%s)", wc->devtype->desc, vpmstring); } else { wc->ddev->devicetype = kasprintf(GFP_KERNEL, wc->devtype->desc); } #else wc->ddev->devicetype = kasprintf(GFP_KERNEL, wc->devtype->desc); #endif if (!wc->ddev->devicetype) return -ENOMEM; return 0; } /* The number of cards we have seen with each possible 'order' switch setting. */ static unsigned int order_index[16]; static void setup_chunks(struct t4 *wc, int which) { struct t4_span *ts; int offset = 1; int x, y; int gen2; int basesize = T4_BASE_SIZE(wc) >> 2; if (!has_e1_span(wc)) offset += 4; gen2 = (wc->tspans[0]->spanflags & FLAG_2NDGEN); for (x = 0; x < wc->numspans; x++) { ts = wc->tspans[x]; ts->writechunk = (void *)(wc->writechunk + (x * 32 * 2) + (which * (basesize))); ts->readchunk = (void *)(wc->readchunk + (x * 32 * 2) + (which * (basesize))); for (y=0;ytspans[x]->span.channels;y++) { struct dahdi_chan *mychans = ts->chans[y]; if (gen2) { mychans->writechunk = (void *)(wc->writechunk + ((x * 32 + y + offset) * 2) + (which * (basesize))); mychans->readchunk = (void *)(wc->readchunk + ((x * 32 + y + offset) * 2) + (which * (basesize))); } } } } static int __t4_hardware_init_1(struct t4 *wc, unsigned int cardflags, bool first_time); static int __t4_hardware_init_2(struct t4 *wc, bool first_time); static int t4_hardware_stop(struct t4 *wc); static void t4_framer_reset(struct t4 *wc) { const bool first_time = false; bool have_vpm = wc->vpm != NULL; if (have_vpm) { release_vpm450m(wc->vpm); wc->vpm = NULL; } t4_hardware_stop(wc); __t4_set_sclk_src(wc, WC_SELF, 0, 0); __t4_hardware_init_1(wc, wc->devtype->flags, first_time); __t4_hardware_init_2(wc, first_time); if (have_vpm) { t4_vpm_init(wc); wc->dmactrl |= (wc->vpm) ? T4_VPM_PRESENT : 0; t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); } setup_chunks(wc, 0); wc->lastindex = 0; } /** * t4_serial_setup - Setup serial parameters and system interface. * @wc: The card to configure. * */ static void t4_serial_setup(struct t4 *wc) { unsigned long flags; unsigned int unit; bool reset_required = false; if (debug) { dev_info(&wc->dev->dev, "TE%dXXP: Setting up global serial parameters\n", wc->numspans); } spin_lock_irqsave(&wc->reglock, flags); reset_required = wc->reset_required > 0; wc->reset_required = 0; spin_unlock_irqrestore(&wc->reglock, flags); if (reset_required) t4_framer_reset(wc); spin_lock_irqsave(&wc->reglock, flags); /* GPC1: Multiplex mode enabled, FSC is output, active low, RCLK from * channel 0 */ __t4_framer_out(wc, 0, 0x85, 0xe0); if (is_octal(wc)) __t4_framer_out(wc, 0, FRMR_GPC2, 0x00); /* IPC: Interrupt push/pull active low */ __t4_framer_out(wc, 0, 0x08, 0x01); if (is_octal(wc)) { /* Global clocks (16.384 Mhz CLK) */ __t4_framer_out(wc, 0, 0x92, 0x00); /* GCM1 */ __t4_framer_out(wc, 0, 0x93, 0x18); __t4_framer_out(wc, 0, 0x94, 0xfb); __t4_framer_out(wc, 0, 0x95, 0x0b); __t4_framer_out(wc, 0, 0x96, 0x01); __t4_framer_out(wc, 0, 0x97, 0x0b); __t4_framer_out(wc, 0, 0x98, 0xdb); __t4_framer_out(wc, 0, 0x99, 0xdf); } else { /* Global clocks (8.192 Mhz CLK) */ __t4_framer_out(wc, 0, 0x92, 0x00); __t4_framer_out(wc, 0, 0x93, 0x18); __t4_framer_out(wc, 0, 0x94, 0xfb); __t4_framer_out(wc, 0, 0x95, 0x0b); __t4_framer_out(wc, 0, 0x96, 0x00); __t4_framer_out(wc, 0, 0x97, 0x0b); __t4_framer_out(wc, 0, 0x98, 0xdb); __t4_framer_out(wc, 0, 0x99, 0xdf); } spin_unlock_irqrestore(&wc->reglock, flags); for (unit = 0; unit < ports_on_framer(wc); ++unit) { spin_lock_irqsave(&wc->reglock, flags); /* Configure interrupts */ /* GCR: Interrupt on Activation/Deactivation of each */ __t4_framer_out(wc, unit, FRMR_GCR, 0x00); /* Configure system interface */ if (is_octal(wc)) { /* SIC1: 16.384 Mhz clock/bus, double buffer receive / * transmit, byte interleaved */ __t4_framer_out(wc, unit, FRMR_SIC1, 0xc2 | 0x08); } else { /* SIC1: 8.192 Mhz clock/bus, double buffer receive / * transmit, byte interleaved */ __t4_framer_out(wc, unit, FRMR_SIC1, 0xc2); } /* SIC2: No FFS, no center receive eliastic buffer, phase */ __t4_framer_out(wc, unit, FRMR_SIC2, 0x20 | (unit << 1)); /* SIC3: Edges for capture */ if (is_octal(wc)) { __t4_framer_out(wc, unit, FRMR_SIC3, 0x04 | (1 << 4)); } else { __t4_framer_out(wc, unit, FRMR_SIC3, 0x04); } /* CMR2: We provide sync and clock for tx and rx. */ __t4_framer_out(wc, unit, FRMR_CMR2, 0x00); if (is_octal(wc)) { /* Set RCLK to 16 MHz */ __t4_framer_out(wc, unit, FRMR_CMR4, 0x5); if (!has_e1_span(wc)) { /* T1/J1 mode */ __t4_framer_out(wc, unit, FRMR_XC0, 0x07); __t4_framer_out(wc, unit, FRMR_XC1, 0x04); if (wc->tspans[unit]->linemode == J1) __t4_framer_out(wc, unit, FRMR_RC0, 0x87); else __t4_framer_out(wc, unit, FRMR_RC0, 0x07); __t4_framer_out(wc, unit, FRMR_RC1, 0x04); } else { /* E1 mode */ __t4_framer_out(wc, unit, FRMR_XC0, 0x00); __t4_framer_out(wc, unit, FRMR_XC1, 0x04); __t4_framer_out(wc, unit, FRMR_RC0, 0x00); __t4_framer_out(wc, unit, FRMR_RC1, 0x04); } } else { if (!has_e1_span(wc)) { /* T1/J1 mode */ __t4_framer_out(wc, unit, FRMR_XC0, 0x03); __t4_framer_out(wc, unit, FRMR_XC1, 0x84); if (J1 == wc->tspans[unit]->linemode) __t4_framer_out(wc, unit, FRMR_RC0, 0x83); else __t4_framer_out(wc, unit, FRMR_RC0, 0x03); __t4_framer_out(wc, unit, FRMR_RC1, 0x84); } else { /* E1 mode */ __t4_framer_out(wc, unit, FRMR_XC0, 0x00); __t4_framer_out(wc, unit, FRMR_XC1, 0x04); __t4_framer_out(wc, unit, FRMR_RC0, 0x04); __t4_framer_out(wc, unit, FRMR_RC1, 0x04); } } /* Configure ports */ /* PC1: SPYR/SPYX input on RPA/XPA */ __t4_framer_out(wc, unit, 0x80, 0x00); /* PC2: RMFB/XSIG output/input on RPB/XPB */ /* PC3: Some unused stuff */ /* PC4: Some more unused stuff */ if (is_octal(wc)) { __t4_framer_out(wc, unit, 0x81, 0xBB); __t4_framer_out(wc, unit, 0x82, 0xf5); __t4_framer_out(wc, unit, 0x83, 0x35); } else if (wc->falc31) { __t4_framer_out(wc, unit, 0x81, 0xBB); __t4_framer_out(wc, unit, 0x82, 0xBB); __t4_framer_out(wc, unit, 0x83, 0xBB); } else { __t4_framer_out(wc, unit, 0x81, 0x22); __t4_framer_out(wc, unit, 0x82, 0x65); __t4_framer_out(wc, unit, 0x83, 0x35); } /* PC5: XMFS active low, SCLKR is input, RCLK is output */ __t4_framer_out(wc, unit, 0x84, 0x01); if (debug & DEBUG_MAIN) { dev_notice(&wc->dev->dev, "Successfully initialized serial bus " "for unit %d\n", unit); } spin_unlock_irqrestore(&wc->reglock, flags); } } /** * t4_span_assigned - Called when the span is assigned by DAHDI. * @span: Span that has been assigned. * * When this function is called, the span has a valid spanno and all the * channels on the span have valid channel numbers assigned. * * This function is necessary because a device may be registered, and * then user space may then later decide to assign span numbers and the * channel numbers. * */ static void t4_span_assigned(struct dahdi_span *span) { struct t4_span *tspan = container_of(span, struct t4_span, span); struct t4 *wc = tspan->owner; struct dahdi_span *pos; unsigned int unassigned_spans = 0; unsigned long flags; /* We use this to make sure all the spans are assigned before * running the serial setup. */ list_for_each_entry(pos, &wc->ddev->spans, device_node) { if (!test_bit(DAHDI_FLAGBIT_REGISTERED, &pos->flags)) ++unassigned_spans; } if (0 == unassigned_spans) { t4_serial_setup(wc); set_bit(T4_CHECK_TIMING, &wc->checkflag); spin_lock_irqsave(&wc->reglock, flags); __t4_set_sclk_src(wc, WC_SELF, 0, 0); spin_unlock_irqrestore(&wc->reglock, flags); } } static void free_wc(struct t4 *wc) { unsigned int x, y; for (x = 0; x < ARRAY_SIZE(wc->tspans); x++) { if (!wc->tspans[x]) continue; for (y = 0; y < ARRAY_SIZE(wc->tspans[x]->chans); y++) { kfree(wc->tspans[x]->chans[y]); kfree(wc->tspans[x]->ec[y]); } kfree(wc->tspans[x]); } kfree(wc->ddev->devicetype); kfree(wc->ddev->location); kfree(wc->ddev->hardware_id); dahdi_free_device(wc->ddev); kfree(wc); } /** * t4_alloc_channels - Allocate the channels on a span. * @wc: The board we're allocating for. * @ts: The span we're allocating for. * @linemode: Which mode (T1/E1/J1) to use for this span. * * This function must only be called before the span is assigned it's * possible for user processes to have an open reference to the * channels. * */ static int t4_alloc_channels(struct t4 *wc, struct t4_span *ts, enum linemode linemode) { int i; if (test_bit(DAHDI_FLAGBIT_REGISTERED, &ts->span.flags)) { dev_dbg(&wc->dev->dev, "Cannot allocate channels on a span that is already " "assigned.\n"); return -EINVAL; } /* Cleanup any previously allocated channels. */ for (i = 0; i < ARRAY_SIZE(ts->chans); ++i) { kfree(ts->chans[i]); kfree(ts->ec[i]); ts->chans[i] = NULL; ts->ec[i] = NULL; } ts->linemode = linemode; for (i = 0; i < ((E1 == ts->linemode) ? 31 : 24); i++) { struct dahdi_chan *chan; struct dahdi_echocan_state *ec; chan = kzalloc(sizeof(*chan), GFP_KERNEL); if (!chan) { free_wc(wc); return -ENOMEM; } ts->chans[i] = chan; ec = kzalloc(sizeof(*ec), GFP_KERNEL); if (!ec) { free_wc(wc); return -ENOMEM; } ts->ec[i] = ec; } return 0; } static void t4_init_one_span(struct t4 *wc, struct t4_span *ts) { unsigned long flags; unsigned int reg; int i; snprintf(ts->span.name, sizeof(ts->span.name) - 1, "TE%d/%d/%d", wc->numspans, wc->num, ts->span.offset + 1); snprintf(ts->span.desc, sizeof(ts->span.desc) - 1, "T%dXXP (PCI) Card %d Span %d", wc->numspans, wc->num, ts->span.offset + 1); switch (ts->linemode) { case T1: ts->span.spantype = SPANTYPE_DIGITAL_T1; break; case E1: ts->span.spantype = SPANTYPE_DIGITAL_E1; break; case J1: ts->span.spantype = SPANTYPE_DIGITAL_J1; break; } /* HDLC Specific init */ ts->sigchan = NULL; ts->sigmode = sigmode; ts->sigactive = 0; if (E1 != ts->linemode) { ts->span.channels = 24; ts->span.deflaw = DAHDI_LAW_MULAW; ts->span.linecompat = DAHDI_CONFIG_AMI | DAHDI_CONFIG_B8ZS | DAHDI_CONFIG_D4 | DAHDI_CONFIG_ESF; } else { ts->span.channels = 31; ts->span.deflaw = DAHDI_LAW_ALAW; ts->span.linecompat = DAHDI_CONFIG_AMI | DAHDI_CONFIG_HDB3 | DAHDI_CONFIG_CCS | DAHDI_CONFIG_CRC4; } ts->span.chans = ts->chans; ts->span.flags = DAHDI_FLAG_RBS; for (i = 0; i < ts->span.channels; i++) { struct dahdi_chan *const chan = ts->chans[i]; chan->pvt = wc; snprintf(chan->name, sizeof(chan->name) - 1, "%s/%d", ts->span.name, i + 1); t4_chan_set_sigcap(&ts->span, i); chan->chanpos = i + 1; } /* Enable 1sec timer interrupt */ spin_lock_irqsave(&wc->reglock, flags); reg = __t4_framer_in(wc, ts->span.offset, FMR1_T); __t4_framer_out(wc, ts->span.offset, FMR1_T, (reg | FMR1_ECM)); /* Enable Errored Second interrupt */ __t4_framer_out(wc, ts->span.offset, ESM, 0); spin_unlock_irqrestore(&wc->reglock, flags); t4_reset_counters(&ts->span); } /** * t4_set_linemode - Allows user space to change the linemode before spans are assigned. * @span: span on which to change the linemode. * @linemode: A value from enumerated spantypes * * This callback is used to override the E1/T1 mode jumper settings and set * the linemode on for each span. Called when the "spantype" attribute * is written in sysfs under the dahdi_device. * */ static int t4_set_linemode(struct dahdi_span *span, enum spantypes linemode) { struct t4_span *ts = container_of(span, struct t4_span, span); struct t4 *wc = ts->owner; int res = 0; enum linemode mode; const char *old_name; static DEFINE_MUTEX(linemode_lock); unsigned long flags; dev_dbg(&wc->dev->dev, "Setting '%s' to '%s'\n", span->name, dahdi_spantype2str(linemode)); if (span->spantype == linemode) return 0; spin_lock_irqsave(&wc->reglock, flags); wc->reset_required = 1; spin_unlock_irqrestore(&wc->reglock, flags); /* Do not allow the t1e1 member to be changed by multiple threads. */ mutex_lock(&linemode_lock); old_name = dahdi_spantype2str(span->spantype); switch (linemode) { case SPANTYPE_DIGITAL_T1: dev_info(&wc->dev->dev, "Changing from %s to T1 line mode.\n", old_name); mode = T1; wc->t1e1 &= ~(1 << span->offset); break; case SPANTYPE_DIGITAL_E1: dev_info(&wc->dev->dev, "Changing from %s to E1 line mode.\n", old_name); mode = E1; wc->t1e1 |= (1 << span->offset); break; case SPANTYPE_DIGITAL_J1: dev_info(&wc->dev->dev, "Changing from %s to J1 line mode.\n", old_name); mode = J1; wc->t1e1 &= ~(1 << span->offset); break; default: dev_err(&wc->dev->dev, "Got invalid linemode %d from dahdi\n", linemode); res = -EINVAL; } if (!res) { t4_alloc_channels(wc, ts, mode); t4_init_one_span(wc, ts); dahdi_init_span(span); } mutex_unlock(&linemode_lock); return res; } static const struct dahdi_span_ops t4_gen1_span_ops = { .owner = THIS_MODULE, .spanconfig = t4_spanconfig, .chanconfig = t4_chanconfig, .startup = t4_startup, .shutdown = t4_shutdown, .rbsbits = t4_rbsbits, .maint = t4_maint, .open = t4_open, .close = t4_close, .ioctl = t4_ioctl, .hdlc_hard_xmit = t4_hdlc_hard_xmit, .assigned = t4_span_assigned, .set_spantype = t4_set_linemode, }; static const struct dahdi_span_ops t4_gen2_span_ops = { .owner = THIS_MODULE, .spanconfig = t4_spanconfig, .chanconfig = t4_chanconfig, .startup = t4_startup, .shutdown = t4_shutdown, .rbsbits = t4_rbsbits, .maint = t4_maint, .open = t4_open, .close = t4_close, .ioctl = t4_ioctl, .hdlc_hard_xmit = t4_hdlc_hard_xmit, .dacs = t4_dacs, .assigned = t4_span_assigned, .set_spantype = t4_set_linemode, #ifdef VPM_SUPPORT .echocan_create = t4_echocan_create, .echocan_name = t4_echocan_name, #endif }; /** * init_spans - Do first initialization on all the spans * @wc: Card to initialize the spans on. * * This function is called *before* the dahdi_device is first registered * with the system. What happens in t4_init_one_span can happen between * when the device is registered and when the spans are assigned via * sysfs (or automatically). * */ static void init_spans(struct t4 *wc) { int x, y; int gen2; struct t4_span *ts; unsigned int reg; unsigned long flags; gen2 = (wc->tspans[0]->spanflags & FLAG_2NDGEN); for (x = 0; x < wc->numspans; x++) { ts = wc->tspans[x]; sprintf(ts->span.name, "TE%d/%d/%d", wc->numspans, wc->num, x + 1); snprintf(ts->span.desc, sizeof(ts->span.desc) - 1, "T%dXXP (PCI) Card %d Span %d", wc->numspans, wc->num, x+1); switch (ts->linemode) { case T1: ts->span.spantype = SPANTYPE_DIGITAL_T1; break; case E1: ts->span.spantype = SPANTYPE_DIGITAL_E1; break; case J1: ts->span.spantype = SPANTYPE_DIGITAL_J1; break; } /* HDLC Specific init */ ts->sigchan = NULL; ts->sigmode = sigmode; ts->sigactive = 0; if (E1 != ts->linemode) { ts->span.channels = 24; ts->span.deflaw = DAHDI_LAW_MULAW; ts->span.linecompat = DAHDI_CONFIG_AMI | DAHDI_CONFIG_B8ZS | DAHDI_CONFIG_D4 | DAHDI_CONFIG_ESF; } else { ts->span.channels = 31; ts->span.deflaw = DAHDI_LAW_ALAW; ts->span.linecompat = DAHDI_CONFIG_AMI | DAHDI_CONFIG_HDB3 | DAHDI_CONFIG_CCS | DAHDI_CONFIG_CRC4; } ts->span.chans = ts->chans; ts->span.flags = DAHDI_FLAG_RBS; ts->owner = wc; ts->span.offset = x; ts->writechunk = (void *)(wc->writechunk + x * 32 * 2); ts->readchunk = (void *)(wc->readchunk + x * 32 * 2); if (gen2) { ts->span.ops = &t4_gen2_span_ops; } else { ts->span.ops = &t4_gen1_span_ops; } for (y=0;ytspans[x]->span.channels;y++) { struct dahdi_chan *mychans = ts->chans[y]; sprintf(mychans->name, "TE%d/%d/%d/%d", wc->numspans, wc->num, x + 1, y + 1); t4_chan_set_sigcap(&ts->span, x); mychans->pvt = wc; mychans->chanpos = y + 1; } /* Start checking for alarms in 250 ms */ ts->alarmcheck_time = jiffies + msecs_to_jiffies(250); /* Enable 1sec timer interrupt */ spin_lock_irqsave(&wc->reglock, flags); reg = __t4_framer_in(wc, x, FMR1_T); __t4_framer_out(wc, x, FMR1_T, (reg | FMR1_ECM)); /* Enable Errored Second interrupt */ __t4_framer_out(wc, x, ESM, 0); spin_unlock_irqrestore(&wc->reglock, flags); t4_reset_counters(&ts->span); } set_span_devicetype(wc); setup_chunks(wc, 0); wc->lastindex = 0; } static int syncsrc = 0; static int syncnum = 0 /* -1 */; static int syncspan = 0; static DEFINE_SPINLOCK(synclock); static void __t4_set_rclk_src(struct t4 *wc, int span) { if (is_octal(wc)) { int cmr5 = 0x00 | (span << 5); int cmr1 = 0x38; /* Clock Mode: RCLK sourced by DCO-R1 by default, Disable Clock-Switching */ __t4_framer_out(wc, 0, 0x44, cmr1); __t4_framer_out(wc, 0, FRMR_CMR5, cmr5); } else { int cmr1 = 0x38; /* Clock Mode: RCLK sourced by DCO-R1 by default, Disable Clock-Switching */ cmr1 |= (span << 6); __t4_framer_out(wc, 0, 0x44, cmr1); } dev_info(&wc->dev->dev, "RCLK source set to span %d\n", span+1); } static void __t4_set_sclk_src(struct t4 *wc, int mode, int master, int slave) { if (slave) { wc->dmactrl |= (1 << 25); dev_info(&wc->dev->dev, "SCLK is slaved to timing cable\n"); } else { wc->dmactrl &= ~(1 << 25); } if (master) { wc->dmactrl |= (1 << 24); dev_info(&wc->dev->dev, "SCLK is master to timing cable\n"); } else { wc->dmactrl &= ~(1 << 24); } if (mode == WC_RECOVER) wc->dmactrl |= (1 << 29); /* Recover timing from RCLK */ if (mode == WC_SELF) wc->dmactrl &= ~(1 << 29);/* Provide timing from MCLK */ __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); } static ssize_t t4_timing_master_show(struct device *dev, struct device_attribute *attr, char *buf) { struct t4 *wc = dev_get_drvdata(dev); if (wc->dmactrl & (1 << 29)) return sprintf(buf, "%d\n", wc->syncsrc); else return sprintf(buf, "%d\n", -1); } static DEVICE_ATTR(timing_master, 0400, t4_timing_master_show, NULL); static void create_sysfs_files(struct t4 *wc) { int ret; ret = device_create_file(&wc->dev->dev, &dev_attr_timing_master); if (ret) { dev_info(&wc->dev->dev, "Failed to create device attributes.\n"); } } static void remove_sysfs_files(struct t4 *wc) { device_remove_file(&wc->dev->dev, &dev_attr_timing_master); } static inline void __t4_update_timing(struct t4 *wc) { int i; /* update sync src info */ if (wc->syncsrc != syncsrc) { dev_info(&wc->dev->dev, "Swapping card %d from %d to %d\n", wc->num, wc->syncsrc, syncsrc); wc->syncsrc = syncsrc; /* Update sync sources */ for (i = 0; i < wc->numspans; i++) { wc->tspans[i]->span.syncsrc = wc->syncsrc; } if (syncnum == wc->num) { __t4_set_rclk_src(wc, syncspan-1); __t4_set_sclk_src(wc, WC_RECOVER, 1, 0); if (debug) dev_notice(&wc->dev->dev, "Card %d, using sync " "span %d, master\n", wc->num, syncspan); } else { __t4_set_sclk_src(wc, WC_RECOVER, 0, 1); if (debug) dev_notice(&wc->dev->dev, "Card %d, using " "Timing Bus, NOT master\n", wc->num); } } } static int __t4_findsync(struct t4 *wc) { int i; int x; unsigned long flags; int p; int nonzero; int newsyncsrc = 0; /* DAHDI span number */ int newsyncnum = 0; /* wct4xxp card number */ int newsyncspan = 0; /* span on given wct4xxp card */ spin_lock_irqsave(&synclock, flags); if (!wc->num) { /* If we're the first card, go through all the motions, up to 8 levels of sync source */ p = 1; while (p < 8) { nonzero = 0; for (x=0;cards[x];x++) { for (i = 0; i < cards[x]->numspans; i++) { if (cards[x]->tspans[i]->syncpos) { nonzero = 1; if ((cards[x]->tspans[i]->syncpos == p) && !(cards[x]->tspans[i]->span.alarms & (DAHDI_ALARM_RED | DAHDI_ALARM_BLUE | DAHDI_ALARM_LOOPBACK)) && (cards[x]->tspans[i]->span.flags & DAHDI_FLAG_RUNNING)) { /* This makes a good sync source */ newsyncsrc = cards[x]->tspans[i]->span.spanno; newsyncnum = x; newsyncspan = i + 1; /* Jump out */ goto found; } } } } if (nonzero) p++; else break; } found: if ((syncnum != newsyncnum) || (syncsrc != newsyncsrc) || (newsyncspan != syncspan)) { if (debug) dev_notice(&wc->dev->dev, "New syncnum: %d " "(was %d), syncsrc: %d (was %d), " "syncspan: %d (was %d)\n", newsyncnum, syncnum, newsyncsrc, syncsrc, newsyncspan, syncspan); syncnum = newsyncnum; syncsrc = newsyncsrc; syncspan = newsyncspan; for (x=0;cards[x];x++) { __t4_update_timing(cards[x]); } } } __t4_update_timing(wc); spin_unlock_irqrestore(&synclock, flags); return 0; } static void __t4_set_timing_source_auto(struct t4 *wc) { int x, i; int firstprio, secondprio; firstprio = secondprio = 4; if (debug) dev_info(&wc->dev->dev, "timing source auto\n"); clear_bit(T4_CHECK_TIMING, &wc->checkflag); if (timingcable) { __t4_findsync(wc); } else { if (debug) dev_info(&wc->dev->dev, "Evaluating spans for timing " "source\n"); for (x=0;xnumspans;x++) { if ((wc->tspans[x]->span.flags & DAHDI_FLAG_RUNNING) && !(wc->tspans[x]->span.alarms & (DAHDI_ALARM_RED | DAHDI_ALARM_BLUE))) { if (debug) dev_info(&wc->dev->dev, "span %d is " "green : syncpos %d\n", x+1, wc->tspans[x]->syncpos); if (wc->tspans[x]->syncpos) { /* Valid rsync source in recovered timing mode */ if (firstprio == 4) firstprio = x; else if (wc->tspans[x]->syncpos < wc->tspans[firstprio]->syncpos) firstprio = x; } else { /* Valid rsync source in system timing mode */ if (secondprio == 4) secondprio = x; } } } if (firstprio != 4) { wc->syncsrc = firstprio; __t4_set_rclk_src(wc, firstprio); __t4_set_sclk_src(wc, WC_RECOVER, 0, 0); dev_info(&wc->dev->dev, "Recovered timing mode, "\ "RCLK set to span %d\n", firstprio+1); } else if (secondprio != 4) { wc->syncsrc = -1; __t4_set_rclk_src(wc, secondprio); __t4_set_sclk_src(wc, WC_SELF, 0, 0); dev_info(&wc->dev->dev, "System timing mode, "\ "RCLK set to span %d\n", secondprio+1); } else { wc->syncsrc = -1; dev_info(&wc->dev->dev, "All spans in alarm : No valid"\ "span to source RCLK from\n"); /* Default rclk to lock with span 1 */ __t4_set_rclk_src(wc, 0); __t4_set_sclk_src(wc, WC_SELF, 0, 0); } /* Propagate sync selection to dahdi_span struct * this is read by dahdi_tool to display the span's * master/slave sync information */ for (i = 0; i < wc->numspans; i++) { wc->tspans[i]->span.syncsrc = wc->syncsrc + 1; } } } static void __t4_configure_t1(struct t4 *wc, int unit, int lineconfig, int txlevel) { unsigned int fmr4, fmr2, fmr1, fmr0, lim2; char *framing, *line; int mytxlevel; if ((txlevel > 7) || (txlevel < 4)) mytxlevel = 0; else mytxlevel = txlevel - 4; if (is_octal(wc)) fmr1 = 0x9c | 0x02; /* FMR1: Mode 1, T1 mode, CRC on for ESF, 8.192 Mhz system data rate, no XAIS */ else fmr1 = 0x9c; /* FMR1: Mode 1, T1 mode, CRC on for ESF, 8.192 Mhz system data rate, no XAIS */ fmr2 = 0x20; /* FMR2: no payload loopback, don't auto yellow */ fmr4 = 0x0c; /* FMR4: Lose sync on 2 out of 5 framing bits, auto resync */ lim2 = 0x21; /* LIM2: 50% peak is a "1", Advanced Loss recovery */ lim2 |= (mytxlevel << 6); /* LIM2: Add line buildout */ __t4_framer_out(wc, unit, 0x1d, fmr1); __t4_framer_out(wc, unit, 0x1e, fmr2); /* Configure line interface */ if (lineconfig & DAHDI_CONFIG_AMI) { line = "AMI"; /* workaround for errata #2 in ES v3 09-10-16 */ fmr0 = (wc->falc31) ? 0xb0 : 0xa0; } else { line = "B8ZS"; fmr0 = 0xf0; } if (lineconfig & DAHDI_CONFIG_D4) { framing = "D4"; } else { framing = "ESF"; fmr4 |= 0x2; fmr2 |= 0xc0; } __t4_framer_out(wc, unit, 0x1c, fmr0); __t4_framer_out(wc, unit, 0x20, fmr4); __t4_framer_out(wc, unit, FMR5, FMR5_EIBR); /* FMR5: Enable RBS mode */ __t4_framer_out(wc, unit, 0x37, 0xf0 ); /* LIM1: Clear data in case of LOS, Set receiver threshold (0.5V), No remote loop, no DRS */ __t4_framer_out(wc, unit, 0x36, 0x08); /* LIM0: Enable auto long haul mode, no local loop (must be after LIM1) */ __t4_framer_out(wc, unit, 0x02, 0x50); /* CMDR: Reset the receiver and transmitter line interface */ __t4_framer_out(wc, unit, 0x02, 0x00); /* CMDR: Reset the receiver and transmitter line interface */ if (wc->falc31) { if (debug) dev_info(&wc->dev->dev, "card %d span %d: setting Rtx " "to 0ohm for T1\n", wc->num, unit); __t4_framer_out(wc, unit, 0x86, 0x00); /* PC6: set Rtx to 0ohm for T1 */ // Hitting the bugfix register to fix errata #3 __t4_framer_out(wc, unit, 0xbd, 0x05); } __t4_framer_out(wc, unit, 0x3a, lim2); /* LIM2: 50% peak amplitude is a "1" */ __t4_framer_out(wc, unit, 0x38, 0x0a); /* PCD: LOS after 176 consecutive "zeros" */ __t4_framer_out(wc, unit, 0x39, 0x15); /* PCR: 22 "ones" clear LOS */ /* Generate pulse mask for T1 */ switch(mytxlevel) { case 3: __t4_framer_out(wc, unit, 0x26, 0x07); /* XPM0 */ __t4_framer_out(wc, unit, 0x27, 0x01); /* XPM1 */ __t4_framer_out(wc, unit, 0x28, 0x00); /* XPM2 */ break; case 2: __t4_framer_out(wc, unit, 0x26, 0x8c); /* XPM0 */ __t4_framer_out(wc, unit, 0x27, 0x11); /* XPM1 */ __t4_framer_out(wc, unit, 0x28, 0x01); /* XPM2 */ break; case 1: __t4_framer_out(wc, unit, 0x26, 0x8c); /* XPM0 */ __t4_framer_out(wc, unit, 0x27, 0x01); /* XPM1 */ __t4_framer_out(wc, unit, 0x28, 0x00); /* XPM2 */ break; case 0: default: __t4_framer_out(wc, unit, 0x26, 0xd7); /* XPM0 */ __t4_framer_out(wc, unit, 0x27, 0x22); /* XPM1 */ __t4_framer_out(wc, unit, 0x28, 0x01); /* XPM2 */ break; } /* Don't mask framer interrupts if hardware HDLC is in use */ __t4_framer_out(wc, unit, FRMR_IMR0, 0xff & ~((wc->tspans[unit]->sigchan) ? HDLC_IMR0_MASK : 0)); /* IMR0: We care about CAS changes, etc */ __t4_framer_out(wc, unit, FRMR_IMR1, 0xff & ~((wc->tspans[unit]->sigchan) ? HDLC_IMR1_MASK : 0)); /* IMR1: We care about nothing */ __t4_framer_out(wc, unit, 0x16, 0x00); /* IMR2: All the alarm stuff! */ __t4_framer_out(wc, unit, 0x17, 0x34); /* IMR3: AIS and friends */ __t4_framer_out(wc, unit, 0x18, 0x3f); /* IMR4: Slips on transmit */ dev_info(&wc->dev->dev, "Span %d configured for %s/%s\n", unit + 1, framing, line); } static void __t4_configure_e1(struct t4 *wc, int unit, int lineconfig) { unsigned int fmr2, fmr1, fmr0; unsigned int cas = 0; unsigned int imr3extra=0; char *crc4 = ""; char *framing, *line; if (is_octal(wc)) { /* 16 MHz */ fmr1 = 0x44 | 0x02; /* FMR1: E1 mode, Automatic force resync, PCM30 mode, 8.192 Mhz backplane, no XAIS */ } else { /* 8 MHz */ fmr1 = 0x44; /* FMR1: E1 mode, Automatic force resync, PCM30 mode, 8.192 Mhz backplane, no XAIS */ } fmr2 = 0x03; /* FMR2: Auto transmit remote alarm, auto loss of multiframe recovery, no payload loopback */ if (lineconfig & DAHDI_CONFIG_CRC4) { fmr1 |= 0x08; /* CRC4 transmit */ fmr2 |= 0xc0; /* CRC4 receive */ crc4 = "/CRC4"; } __t4_framer_out(wc, unit, 0x1d, fmr1); __t4_framer_out(wc, unit, 0x1e, fmr2); /* Configure line interface */ if (lineconfig & DAHDI_CONFIG_AMI) { line = "AMI"; /* workaround for errata #2 in ES v3 09-10-16 */ fmr0 = (wc->falc31) ? 0xb0 : 0xa0; } else { line = "HDB3"; fmr0 = 0xf0; } if (lineconfig & DAHDI_CONFIG_CCS) { framing = "CCS"; imr3extra = 0x28; } else { framing = "CAS"; cas = 0x40; } __t4_framer_out(wc, unit, 0x1c, fmr0); __t4_framer_out(wc, unit, 0x37, 0xf0 /*| 0x6 */ ); /* LIM1: Clear data in case of LOS, Set receiver threshold (0.5V), No remote loop, no DRS */ __t4_framer_out(wc, unit, 0x36, 0x08); /* LIM0: Enable auto long haul mode, no local loop (must be after LIM1) */ __t4_framer_out(wc, unit, 0x02, 0x50); /* CMDR: Reset the receiver and transmitter line interface */ __t4_framer_out(wc, unit, 0x02, 0x00); /* CMDR: Reset the receiver and transmitter line interface */ if (wc->falc31) { if (debug) dev_info(&wc->dev->dev, "setting Rtx to 7.5ohm for E1\n"); __t4_framer_out(wc, unit, 0x86, 0x40); /* PC6: turn on 7.5ohm Rtx for E1 */ } /* Condition receive line interface for E1 after reset */ __t4_framer_out(wc, unit, 0xbb, 0x17); __t4_framer_out(wc, unit, 0xbc, 0x55); __t4_framer_out(wc, unit, 0xbb, 0x97); __t4_framer_out(wc, unit, 0xbb, 0x11); __t4_framer_out(wc, unit, 0xbc, 0xaa); __t4_framer_out(wc, unit, 0xbb, 0x91); __t4_framer_out(wc, unit, 0xbb, 0x12); __t4_framer_out(wc, unit, 0xbc, 0x55); __t4_framer_out(wc, unit, 0xbb, 0x92); __t4_framer_out(wc, unit, 0xbb, 0x0c); __t4_framer_out(wc, unit, 0xbb, 0x00); __t4_framer_out(wc, unit, 0xbb, 0x8c); __t4_framer_out(wc, unit, 0x3a, 0x20); /* LIM2: 50% peak amplitude is a "1" */ __t4_framer_out(wc, unit, 0x38, 0x0a); /* PCD: LOS after 176 consecutive "zeros" */ __t4_framer_out(wc, unit, 0x39, 0x15); /* PCR: 22 "ones" clear LOS */ __t4_framer_out(wc, unit, 0x20, 0x9f); /* XSW: Spare bits all to 1 */ __t4_framer_out(wc, unit, 0x21, 0x1c|cas); /* XSP: E-bit set when async. AXS auto, XSIF to 1 */ /* Generate pulse mask for E1 */ __t4_framer_out(wc, unit, 0x26, 0x54); /* XPM0 */ __t4_framer_out(wc, unit, 0x27, 0x02); /* XPM1 */ __t4_framer_out(wc, unit, 0x28, 0x00); /* XPM2 */ /* Don't mask framer interrupts if hardware HDLC is in use */ __t4_framer_out(wc, unit, FRMR_IMR0, 0xff & ~((wc->tspans[unit]->sigchan) ? HDLC_IMR0_MASK : 0)); /* IMR0: We care about CRC errors, CAS changes, etc */ __t4_framer_out(wc, unit, FRMR_IMR1, 0x3f & ~((wc->tspans[unit]->sigchan) ? HDLC_IMR1_MASK : 0)); /* IMR1: We care about loopup / loopdown */ __t4_framer_out(wc, unit, 0x16, 0x00); /* IMR2: We care about all the alarm stuff! */ __t4_framer_out(wc, unit, 0x17, 0x04 | imr3extra); /* IMR3: AIS */ __t4_framer_out(wc, unit, 0x18, 0x3f); /* IMR4: We care about slips on transmit */ __t4_framer_out(wc, unit, 0x2f, 0x00); __t4_framer_out(wc, unit, 0x30, 0x00); __t4_framer_out(wc, unit, 0x31, 0x00); dev_info(&wc->dev->dev, "TE%dXXP: Span %d configured for %s/%s%s\n", wc->numspans, unit + 1, framing, line, crc4); } /** * t4_check_for_interrupts - Return 0 if the card is generating interrupts. * @wc: The card to check. * * If the card is not generating interrupts, this function will also place all * the spans on the card into red alarm. * */ static int t4_check_for_interrupts(struct t4 *wc) { unsigned int starting_intcount = wc->intcount; unsigned long stop_time = jiffies + HZ*2; unsigned long flags; int x; msleep(20); spin_lock_irqsave(&wc->reglock, flags); while (starting_intcount == wc->intcount) { spin_unlock_irqrestore(&wc->reglock, flags); if (time_after(jiffies, stop_time)) { for (x = 0; x < wc->numspans; x++) wc->tspans[x]->span.alarms = DAHDI_ALARM_RED; dev_err(&wc->dev->dev, "Interrupts not detected.\n"); return -EIO; } msleep(100); spin_lock_irqsave(&wc->reglock, flags); } spin_unlock_irqrestore(&wc->reglock, flags); return 0; } static int _t4_startup(struct file *file, struct dahdi_span *span) { #ifdef SUPPORT_GEN1 int i; #endif int tspan; unsigned long flags; int alreadyrunning; struct t4_span *ts = container_of(span, struct t4_span, span); struct t4 *wc = ts->owner; set_bit(T4_IGNORE_LATENCY, &wc->checkflag); if (debug) dev_info(&wc->dev->dev, "About to enter startup!\n"); tspan = span->offset + 1; if (tspan < 0) { dev_info(&wc->dev->dev, "TE%dXXP: Span '%d' isn't us?\n", wc->numspans, span->spanno); return -1; } spin_lock_irqsave(&wc->reglock, flags); alreadyrunning = span->flags & DAHDI_FLAG_RUNNING; #ifdef SUPPORT_GEN1 /* initialize the start value for the entire chunk of last ec buffer */ for(i = 0; i < span->channels; i++) { memset(ts->ec_chunk1[i], DAHDI_LIN2X(0,span->chans[i]),DAHDI_CHUNKSIZE); memset(ts->ec_chunk2[i], DAHDI_LIN2X(0,span->chans[i]),DAHDI_CHUNKSIZE); } #endif /* Force re-evaluation of timing source */ wc->syncsrc = -1; set_bit(T4_CHECK_TIMING, &wc->checkflag); if (E1 == ts->linemode) __t4_configure_e1(wc, span->offset, span->lineconfig); else __t4_configure_t1(wc, span->offset, span->lineconfig, span->txlevel); /* Note clear channel status */ wc->tspans[span->offset]->notclear = 0; __set_clear(wc, span->offset); if (!alreadyrunning) { span->flags |= DAHDI_FLAG_RUNNING; wc->spansstarted++; if (wc->devtype->flags & FLAG_5THGEN) __t4_pci_out(wc, 5, (ms_per_irq << 16) | wc->numbufs); else __t4_pci_out(wc, 5, (1 << 16) | 1); /* enable interrupts */ /* Start DMA, enabling DMA interrupts on read only */ wc->dmactrl |= (ts->spanflags & FLAG_2NDGEN) ? 0xc0000000 : 0xc0000003; #ifdef VPM_SUPPORT wc->dmactrl |= (wc->vpm) ? T4_VPM_PRESENT : 0; #endif /* Seed interrupt register */ __t4_pci_out(wc, WC_INTR, 0x0c); if (noburst || !(ts->spanflags & FLAG_BURST)) wc->dmactrl |= (1 << 26); else wc->dmactrl &= ~(1 << 26); __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); /* Startup HDLC controller too */ } if (ts->sigchan) { struct dahdi_chan *sigchan = ts->sigchan; spin_unlock_irqrestore(&wc->reglock, flags); if (hdlc_start(wc, span->offset, sigchan, ts->sigmode)) { dev_notice(&wc->dev->dev, "Error initializing " "signalling controller\n"); return -1; } spin_lock_irqsave(&wc->reglock, flags); } spin_unlock_irqrestore(&wc->reglock, flags); local_irq_save(flags); t4_check_alarms(wc, span->offset); t4_check_sigbits(wc, span->offset); local_irq_restore(flags); if (wc->tspans[0]->sync == span->spanno) dev_info(&wc->dev->dev, "SPAN %d: Primary Sync Source\n", span->spanno); if (wc->tspans[1]->sync == span->spanno) dev_info(&wc->dev->dev, "SPAN %d: Secondary Sync Source\n", span->spanno); if (wc->numspans >= 4) { if (wc->tspans[2]->sync == span->spanno) dev_info(&wc->dev->dev, "SPAN %d: Tertiary Sync Source" "\n", span->spanno); if (wc->tspans[3]->sync == span->spanno) dev_info(&wc->dev->dev, "SPAN %d: Quaternary Sync " "Source\n", span->spanno); } if (wc->numspans == 8) { if (wc->tspans[4]->sync == span->spanno) dev_info(&wc->dev->dev, "SPAN %d: Quinary Sync " "Source\n", span->spanno); if (wc->tspans[5]->sync == span->spanno) dev_info(&wc->dev->dev, "SPAN %d: Senary Sync " "Source\n", span->spanno); if (wc->tspans[6]->sync == span->spanno) dev_info(&wc->dev->dev, "SPAN %d: Septenary Sync " "Source\n", span->spanno); if (wc->tspans[7]->sync == span->spanno) dev_info(&wc->dev->dev, "SPAN %d: Octonary Sync " "Source\n", span->spanno); } if (!alreadyrunning) { if (t4_check_for_interrupts(wc)) return -EIO; } if (debug) dev_info(&wc->dev->dev, "Completed startup!\n"); clear_bit(T4_IGNORE_LATENCY, &wc->checkflag); return 0; } static int t4_startup(struct file *file, struct dahdi_span *span) { int ret; struct dahdi_device *const ddev = span->parent; struct dahdi_span *s; ret = _t4_startup(file, span); list_for_each_entry(s, &ddev->spans, device_node) { if (!test_bit(DAHDI_FLAGBIT_RUNNING, &s->flags)) { _t4_startup(file, s); } } return ret; } #ifdef SUPPORT_GEN1 static inline void e1_check(struct t4 *wc, int span, int val) { struct t4_span *ts = wc->tspans[span]; if ((ts->span.channels > 24) && (ts->span.flags & DAHDI_FLAG_RUNNING) && !(ts->span.alarms) && (!wc->e1recover)) { if (val != 0x1b) { ts->e1check++; } else ts->e1check = 0; if (ts->e1check > 100) { /* Wait 1000 ms */ wc->e1recover = 1000 * 8; wc->tspans[0]->e1check = wc->tspans[1]->e1check = 0; if (wc->numspans == 4) wc->tspans[2]->e1check = wc->tspans[3]->e1check = 0; if (debug & DEBUG_MAIN) dev_notice(&wc->dev->dev, "Detected loss of " "E1 alignment on span %d!\n", span); t4_reset_dma(wc); } } } static void t4_receiveprep(struct t4 *wc, int irq) { unsigned int *readchunk; int dbl = 0; int x,y,z; unsigned int tmp; int offset=0; if (!has_e1_span(wc)) offset = 4; if (irq & 1) { /* First part */ readchunk = wc->readchunk; if (!wc->last0) dbl = 1; wc->last0 = 0; } else { readchunk = wc->readchunk + DAHDI_CHUNKSIZE * 32; if (wc->last0) dbl = 1; wc->last0 = 1; } if (unlikely(dbl && (debug & DEBUG_MAIN))) dev_notice(&wc->dev->dev, "Double/missed interrupt detected\n"); for (x=0;xnumspans == 4) { wc->tspans[3]->span.chans[z]->readchunk[x] = tmp & 0xff; wc->tspans[2]->span.chans[z]->readchunk[x] = (tmp & 0xff00) >> 8; } wc->tspans[1]->span.chans[z]->readchunk[x] = (tmp & 0xff0000) >> 16; wc->tspans[0]->span.chans[z]->readchunk[x] = tmp >> 24; } if (has_e1_span(wc)) { if (wc->e1recover > 0) wc->e1recover--; tmp = readchunk[0]; if (wc->numspans == 4) { e1_check(wc, 3, (tmp & 0x7f)); e1_check(wc, 2, (tmp & 0x7f00) >> 8); } e1_check(wc, 1, (tmp & 0x7f0000) >> 16); e1_check(wc, 0, (tmp & 0x7f000000) >> 24); for (z=24;z<31;z++) { /* Only E1 channels now */ tmp = readchunk[z+1]; if (wc->numspans == 4) { if (wc->tspans[3]->span.channels > 24) wc->tspans[3]->span.chans[z]->readchunk[x] = tmp & 0xff; if (wc->tspans[2]->span.channels > 24) wc->tspans[2]->span.chans[z]->readchunk[x] = (tmp & 0xff00) >> 8; } if (wc->tspans[1]->span.channels > 24) wc->tspans[1]->span.chans[z]->readchunk[x] = (tmp & 0xff0000) >> 16; if (wc->tspans[0]->span.channels > 24) wc->tspans[0]->span.chans[z]->readchunk[x] = tmp >> 24; } } /* Advance pointer by 4 TDM frame lengths */ readchunk += 32; } for (x=0;xnumspans;x++) { if (wc->tspans[x]->span.flags & DAHDI_FLAG_RUNNING) { for (y=0;ytspans[x]->span.channels;y++) { /* Echo cancel double buffered data */ dahdi_ec_chunk(wc->tspans[x]->span.chans[y], wc->tspans[x]->span.chans[y]->readchunk, wc->tspans[x]->ec_chunk2[y]); memcpy(wc->tspans[x]->ec_chunk2[y],wc->tspans[x]->ec_chunk1[y], DAHDI_CHUNKSIZE); memcpy(wc->tspans[x]->ec_chunk1[y], wc->tspans[x]->span.chans[y]->writechunk, DAHDI_CHUNKSIZE); } _dahdi_receive(&wc->tspans[x]->span); } } } #endif #if (DAHDI_CHUNKSIZE != 8) #error Sorry, nextgen does not support chunksize != 8 #endif static void __receive_span(struct t4_span *ts) { #ifdef VPM_SUPPORT int y; unsigned long merged; merged = ts->dtmfactive & ts->dtmfmutemask; if (merged) { for (y=0;yspan.channels;y++) { /* Mute any DTMFs which are supposed to be muted */ if (test_bit(y, &merged)) { memset(ts->span.chans[y]->readchunk, DAHDI_XLAW(0, ts->span.chans[y]), DAHDI_CHUNKSIZE); } } } #endif _dahdi_ec_span(&ts->span); _dahdi_receive(&ts->span); } static inline void __transmit_span(struct t4_span *ts) { _dahdi_transmit(&ts->span); } #ifdef ENABLE_WORKQUEUES static void workq_handlespan(void *data) { struct t4_span *ts = data; struct t4 *wc = ts->owner; __receive_span(ts); __transmit_span(ts); atomic_dec(&wc->worklist); if (!atomic_read(&wc->worklist)) t4_pci_out(wc, WC_INTR, 0); } #else static void t4_prep_gen2(struct t4 *wc) { int x; for (x=0;xnumspans;x++) { if (wc->tspans[x]->span.flags & DAHDI_FLAG_RUNNING) { __receive_span(wc->tspans[x]); __transmit_span(wc->tspans[x]); } } } #endif #ifdef SUPPORT_GEN1 static void t4_transmitprep(struct t4 *wc, int irq) { u32 *writechunk; int x, y, z; unsigned int tmp; int offset = 0; if (!has_e1_span(wc)) offset = 4; if (irq & 1) { /* First part */ writechunk = wc->writechunk + 1; } else { writechunk = wc->writechunk + DAHDI_CHUNKSIZE * 32 + 1; } for (y=0;ynumspans;y++) { if (wc->tspans[y]->span.flags & DAHDI_FLAG_RUNNING) _dahdi_transmit(&wc->tspans[y]->span); } for (x=0;xtspans[3]->span.chans[z]->writechunk[x]) | (wc->tspans[2]->span.chans[z]->writechunk[x] << 8) | (wc->tspans[1]->span.chans[z]->writechunk[x] << 16) | (wc->tspans[0]->span.chans[z]->writechunk[x] << 24); writechunk[z+offset] = tmp; } if (has_e1_span(wc)) { for (z=24;z<31;z++) { /* Only E1 channels now */ tmp = 0; if (wc->numspans == 4) { if (wc->tspans[3]->span.channels > 24) tmp |= wc->tspans[3]->span.chans[z]->writechunk[x]; if (wc->tspans[2]->span.channels > 24) tmp |= (wc->tspans[2]->span.chans[z]->writechunk[x] << 8); } if (wc->tspans[1]->span.channels > 24) tmp |= (wc->tspans[1]->span.chans[z]->writechunk[x] << 16); if (wc->tspans[0]->span.channels > 24) tmp |= (wc->tspans[0]->span.chans[z]->writechunk[x] << 24); writechunk[z] = tmp; } } /* Advance pointer by 4 TDM frame lengths */ writechunk += 32; } } #endif static void t4_dahdi_rbsbits(struct dahdi_chan *const chan, int rxs) { if ((debug & DEBUG_RBS) && printk_ratelimit()) { const struct t4_span *tspan = container_of(chan->span, struct t4_span, span); const struct t4 *const wc = tspan->owner; dev_notice(&wc->dev->dev, "Detected sigbits change on " \ "channel %s to %04x\n", chan->name, rxs); } dahdi_rbsbits(chan, rxs); } static void t4_check_sigbits(struct t4 *wc, int span) { int a,i,rxs; struct t4_span *ts = wc->tspans[span]; if (debug & DEBUG_RBS) dev_notice(&wc->dev->dev, "Checking sigbits on span %d\n", span + 1); if (!(ts->span.flags & DAHDI_FLAG_RUNNING)) return; if (E1 == ts->linemode) { for (i = 0; i < 15; i++) { a = t4_framer_in(wc, span, 0x71 + i); /* Get high channel in low bits */ rxs = (a & 0xf); if (!(ts->span.chans[i+16]->sig & DAHDI_SIG_CLEAR)) { if (ts->span.chans[i+16]->rxsig != rxs) t4_dahdi_rbsbits(ts->span.chans[i+16], rxs); } rxs = (a >> 4) & 0xf; if (!(ts->span.chans[i]->sig & DAHDI_SIG_CLEAR)) { if (ts->span.chans[i]->rxsig != rxs) t4_dahdi_rbsbits(ts->span.chans[i], rxs); } } } else if (ts->span.lineconfig & DAHDI_CONFIG_D4) { for (i = 0; i < 24; i+=4) { a = t4_framer_in(wc, span, 0x70 + (i>>2)); /* Get high channel in low bits */ rxs = (a & 0x3) << 2; if (!(ts->span.chans[i+3]->sig & DAHDI_SIG_CLEAR)) { if (ts->span.chans[i+3]->rxsig != rxs) t4_dahdi_rbsbits(ts->span.chans[i+3], rxs); } rxs = (a & 0xc); if (!(ts->span.chans[i+2]->sig & DAHDI_SIG_CLEAR)) { if (ts->span.chans[i+2]->rxsig != rxs) t4_dahdi_rbsbits(ts->span.chans[i+2], rxs); } rxs = (a >> 2) & 0xc; if (!(ts->span.chans[i+1]->sig & DAHDI_SIG_CLEAR)) { if (ts->span.chans[i+1]->rxsig != rxs) t4_dahdi_rbsbits(ts->span.chans[i+1], rxs); } rxs = (a >> 4) & 0xc; if (!(ts->span.chans[i]->sig & DAHDI_SIG_CLEAR)) { if (ts->span.chans[i]->rxsig != rxs) t4_dahdi_rbsbits(ts->span.chans[i], rxs); } } } else { for (i = 0; i < 24; i+=2) { a = t4_framer_in(wc, span, 0x70 + (i>>1)); /* Get high channel in low bits */ rxs = (a & 0xf); if (!(ts->span.chans[i+1]->sig & DAHDI_SIG_CLEAR)) { /* XXX Not really reset on every trans! XXX */ if (ts->span.chans[i+1]->rxsig != rxs) { t4_dahdi_rbsbits(ts->span.chans[i+1], rxs); } } rxs = (a >> 4) & 0xf; if (!(ts->span.chans[i]->sig & DAHDI_SIG_CLEAR)) { /* XXX Not really reset on every trans! XXX */ if (ts->span.chans[i]->rxsig != rxs) { t4_dahdi_rbsbits(ts->span.chans[i], rxs); } } } } } /* Must be called from within hardirq context. */ static void t4_check_alarms(struct t4 *wc, int span) { unsigned char c, d, e; int alarms; int x,j; struct t4_span *ts = wc->tspans[span]; if (time_before(jiffies, ts->alarmcheck_time)) return; if (!(ts->span.flags & DAHDI_FLAG_RUNNING)) return; spin_lock(&wc->reglock); c = __t4_framer_in(wc, span, 0x4c); d = __t4_framer_in(wc, span, 0x4d); /* Assume no alarms */ alarms = 0; /* And consider only carrier alarms */ ts->span.alarms &= (DAHDI_ALARM_RED | DAHDI_ALARM_BLUE | DAHDI_ALARM_NOTOPEN); if (E1 == ts->linemode) { if (c & 0x04) { /* No multiframe found, force RAI high after 400ms only if we haven't found a multiframe since last loss of frame */ if (!(ts->spanflags & FLAG_NMF)) { __t4_framer_out(wc, span, 0x20, 0x9f | 0x20); /* LIM0: Force RAI High */ ts->spanflags |= FLAG_NMF; dev_notice(&wc->dev->dev, "Lost crc4-multiframe alignment\n"); } __t4_framer_out(wc, span, 0x1e, 0xc3); /* Reset to CRC4 mode */ __t4_framer_out(wc, span, 0x1c, 0xf2); /* Force Resync */ __t4_framer_out(wc, span, 0x1c, 0xf0); /* Force Resync */ } else if (!(c & 0x02)) { if ((ts->spanflags & FLAG_NMF)) { __t4_framer_out(wc, span, 0x20, 0x9f); /* LIM0: Clear forced RAI */ ts->spanflags &= ~FLAG_NMF; dev_notice(&wc->dev->dev, "Obtained crc4-multiframe alignment\n"); } } } else { /* Detect loopup code if we're not sending one */ if ((!ts->span.mainttimer) && (d & 0x08)) { /* Loop-up code detected */ if ((ts->loopupcnt++ > 80) && (ts->span.maintstat != DAHDI_MAINT_REMOTELOOP)) { dev_notice(&wc->dev->dev, "span %d: Loopup detected,"\ " enabling remote loop\n", span+1); __t4_framer_out(wc, span, 0x36, 0x08); /* LIM0: Disable any local loop */ __t4_framer_out(wc, span, 0x37, 0xf6 ); /* LIM1: Enable remote loop */ ts->span.maintstat = DAHDI_MAINT_REMOTELOOP; } } else ts->loopupcnt = 0; /* Same for loopdown code */ if ((!ts->span.mainttimer) && (d & 0x10)) { /* Loop-down code detected */ if ((ts->loopdowncnt++ > 80) && (ts->span.maintstat == DAHDI_MAINT_REMOTELOOP)) { dev_notice(&wc->dev->dev, "span %d: Loopdown detected,"\ " disabling remote loop\n", span+1); __t4_framer_out(wc, span, 0x36, 0x08); /* LIM0: Disable any local loop */ __t4_framer_out(wc, span, 0x37, 0xf0 ); /* LIM1: Disable remote loop */ ts->span.maintstat = DAHDI_MAINT_NONE; } } else ts->loopdowncnt = 0; } if (ts->span.lineconfig & DAHDI_CONFIG_NOTOPEN) { for (x=0,j=0;x < ts->span.channels;x++) if ((ts->span.chans[x]->flags & DAHDI_FLAG_OPEN) || dahdi_have_netdev(ts->span.chans[x])) j++; if (!j) alarms |= DAHDI_ALARM_NOTOPEN; } /* Loss of Frame Alignment */ if (c & 0x20) { if (!ts->alarm_time) { if (unlikely(debug)) { /* starting to debounce LOF/LFA */ dev_info(&wc->dev->dev, "wct%dxxp: LOF/LFA " "detected on span %d but debouncing " "for %d ms\n", wc->numspans, span + 1, alarmdebounce); } ts->alarm_time = jiffies + msecs_to_jiffies(alarmdebounce); } else if (time_after(jiffies, ts->alarm_time)) { /* Disable Slip Interrupts */ e = __t4_framer_in(wc, span, 0x17); __t4_framer_out(wc, span, 0x17, (e|0x03)); alarms |= DAHDI_ALARM_RED; } } else { ts->alarm_time = 0; } /* Loss of Signal */ if (c & 0x80) { if (!ts->losalarm_time) { if (unlikely(debug)) { /* starting to debounce LOS */ dev_info(&wc->dev->dev, "wct%dxxp: LOS " "detected on span %d but debouncing " "for %d ms\n", wc->numspans, span + 1, losalarmdebounce); } ts->losalarm_time = jiffies + msecs_to_jiffies(losalarmdebounce); } else if (time_after(jiffies, ts->losalarm_time)) { /* Disable Slip Interrupts */ e = __t4_framer_in(wc, span, 0x17); __t4_framer_out(wc, span, 0x17, (e|0x03)); alarms |= DAHDI_ALARM_RED; } } else { ts->losalarm_time = 0; } /* Alarm Indication Signal */ if (c & 0x40) { if (!ts->aisalarm_time) { if (unlikely(debug)) { /* starting to debounce AIS */ dev_info(&wc->dev->dev, "wct%dxxp: AIS " "detected on span %d but debouncing " "for %d ms\n", wc->numspans, span + 1, aisalarmdebounce); } ts->aisalarm_time = jiffies + msecs_to_jiffies(aisalarmdebounce); } else if (time_after(jiffies, ts->aisalarm_time)) { alarms |= DAHDI_ALARM_BLUE; } } else { ts->aisalarm_time = 0; } /* Add detailed alarm status information to a red alarm state */ if (alarms & DAHDI_ALARM_RED) { if (c & FRS0_LOS) alarms |= DAHDI_ALARM_LOS; if (c & FRS0_LFA) alarms |= DAHDI_ALARM_LFA; if (c & FRS0_LMFA) alarms |= DAHDI_ALARM_LMFA; } if (unlikely(debug)) { /* Check to ensure the xmit line isn't shorted */ if (unlikely(d & FRS1_XLS)) { dev_info(&wc->dev->dev, "Detected a possible hardware malfunction"\ " this card may need servicing\n"); } } if (((!ts->span.alarms) && alarms) || (ts->span.alarms && (!alarms))) set_bit(T4_CHECK_TIMING, &wc->checkflag); /* Keep track of recovering */ if ((!alarms) && ts->span.alarms) ts->alarmtimer = DAHDI_ALARMSETTLE_TIME; if (ts->alarmtimer) alarms |= DAHDI_ALARM_RECOVER; /* If receiving alarms, go into Yellow alarm state */ if (alarms && !(ts->spanflags & FLAG_SENDINGYELLOW)) { /* We manually do yellow alarm to handle RECOVER and NOTOPEN, otherwise it's auto anyway */ unsigned char fmr4; fmr4 = __t4_framer_in(wc, span, 0x20); __t4_framer_out(wc, span, 0x20, fmr4 | 0x20); dev_info(&wc->dev->dev, "Setting yellow alarm span %d\n", span+1); ts->spanflags |= FLAG_SENDINGYELLOW; } else if ((!alarms) && (ts->spanflags & FLAG_SENDINGYELLOW)) { unsigned char fmr4; /* We manually do yellow alarm to handle RECOVER */ fmr4 = __t4_framer_in(wc, span, 0x20); __t4_framer_out(wc, span, 0x20, fmr4 & ~0x20); dev_info(&wc->dev->dev, "Clearing yellow alarm span %d\n", span+1); /* Re-enable timing slip interrupts */ e = __t4_framer_in(wc, span, 0x17); __t4_framer_out(wc, span, 0x17, (e & ~(0x03))); ts->spanflags &= ~FLAG_SENDINGYELLOW; } /* Re-check the timing source when we enter/leave alarm, not withstanding yellow alarm */ if (c & 0x10) { /* receiving yellow (RAI) */ if (!ts->yelalarm_time) { if (unlikely(debug)) { /* starting to debounce AIS */ dev_info(&wc->dev->dev, "wct%dxxp: yellow " "(RAI) detected on span %d but " "debouncing for %d ms\n", wc->numspans, span + 1, yelalarmdebounce); } ts->yelalarm_time = jiffies + msecs_to_jiffies(yelalarmdebounce); } else if (time_after(jiffies, ts->yelalarm_time)) { alarms |= DAHDI_ALARM_YELLOW; } } else { ts->yelalarm_time = 0; } if (alarms) ts->alarmcheck_time = jiffies + msecs_to_jiffies(100); else ts->alarmcheck_time = jiffies + msecs_to_jiffies(50); if (ts->span.mainttimer || ts->span.maintstat) alarms |= DAHDI_ALARM_LOOPBACK; ts->span.alarms = alarms; spin_unlock(&wc->reglock); dahdi_alarm_notify(&ts->span); } static void t4_do_counters(struct t4 *wc) { int span; for (span = 0; span < wc->numspans; span++) { struct t4_span *ts = wc->tspans[span]; spin_lock(&wc->reglock); if (ts->alarmtimer && (0 == (--ts->alarmtimer))) ts->span.alarms &= ~(DAHDI_ALARM_RECOVER); spin_unlock(&wc->reglock); t4_check_alarms(wc, span); } } static inline void __handle_leds(struct t4 *wc) { int x; wc->blinktimer++; for (x=0;xnumspans;x++) { struct t4_span *ts = wc->tspans[x]; if (ts->span.flags & DAHDI_FLAG_RUNNING) { if ((ts->span.alarms & (DAHDI_ALARM_RED | DAHDI_ALARM_BLUE)) || ts->losalarm_time) { #ifdef FANCY_ALARM if (wc->blinktimer == (altab[wc->alarmpos] >> 1)) { __t4_set_led(wc, x, WC_RED); } if (wc->blinktimer == 0xf) { __t4_set_led(wc, x, WC_OFF); } #else if (wc->blinktimer == 160) { __t4_set_led(wc, x, WC_RED); } else if (wc->blinktimer == 480) { __t4_set_led(wc, x, WC_OFF); } #endif } else if (ts->span.alarms & DAHDI_ALARM_YELLOW) { /* Yellow Alarm */ __t4_set_led(wc, x, WC_YELLOW); } else if (ts->span.mainttimer || ts->span.maintstat) { #ifdef FANCY_ALARM if (wc->blinktimer == (altab[wc->alarmpos] >> 1)) { __t4_set_led(wc, x, WC_GREEN); } if (wc->blinktimer == 0xf) { __t4_set_led(wc, x, WC_OFF); } #else if (wc->blinktimer == 160) { __t4_set_led(wc, x, WC_GREEN); } else if (wc->blinktimer == 480) { __t4_set_led(wc, x, WC_OFF); } #endif } else { /* No Alarm */ __t4_set_led(wc, x, WC_GREEN); } } else __t4_set_led(wc, x, WC_OFF); } #ifdef FANCY_ALARM if (wc->blinktimer == 0xf) { wc->blinktimer = -1; wc->alarmpos++; if (wc->alarmpos >= ARRAY_SIZE(altab)) wc->alarmpos = 0; } #else if (wc->blinktimer == 480) wc->blinktimer = 0; #endif } static inline void t4_framer_interrupt(struct t4 *wc, int span) { /* Check interrupts for a given span */ unsigned char gis, isr0, isr1, isr2, isr3, isr4; int readsize = -1; struct t4_span *ts = wc->tspans[span]; struct dahdi_chan *sigchan; unsigned long flags; /* 1st gen cards isn't used interrupts */ spin_lock_irqsave(&wc->reglock, flags); gis = __t4_framer_in(wc, span, FRMR_GIS); isr0 = (gis & FRMR_GIS_ISR0) ? __t4_framer_in(wc, span, FRMR_ISR0) : 0; isr1 = (gis & FRMR_GIS_ISR1) ? __t4_framer_in(wc, span, FRMR_ISR1) : 0; isr2 = (gis & FRMR_GIS_ISR2) ? __t4_framer_in(wc, span, FRMR_ISR2) : 0; isr3 = (gis & FRMR_GIS_ISR3) ? __t4_framer_in(wc, span, FRMR_ISR3) : 0; isr4 = (gis & FRMR_GIS_ISR4) ? __t4_framer_in(wc, span, FRMR_ISR4) : 0; if ((debug & DEBUG_FRAMER) && !(isr3 & ISR3_SEC)) { dev_info(&wc->dev->dev, "gis: %02x, isr0: %02x, isr1: %02x, "\ "isr2: %02x, isr3: %08x, isr4: %02x, intcount=%u\n", gis, isr0, isr1, isr2, isr3, isr4, wc->intcount); } /* Collect performance counters once per second */ if (isr3 & ISR3_SEC) { ts->span.count.fe += __t4_framer_in(wc, span, FECL_T); ts->span.count.crc4 += __t4_framer_in(wc, span, CEC1L_T); ts->span.count.cv += __t4_framer_in(wc, span, CVCL_T); ts->span.count.ebit += __t4_framer_in(wc, span, EBCL_T); ts->span.count.be += __t4_framer_in(wc, span, BECL_T); ts->span.count.prbs = __t4_framer_in(wc, span, FRS1_T); } spin_unlock_irqrestore(&wc->reglock, flags); /* Collect errored second counter once per second */ if (isr3 & ISR3_ES) { ts->span.count.errsec += 1; } if (isr0) t4_check_sigbits(wc, span); if (E1 == ts->linemode) { /* E1 checks */ if ((isr3 & 0x38) || isr2 || isr1) t4_check_alarms(wc, span); } else { /* T1 checks */ if (isr2 || (isr3 & 0x08)) t4_check_alarms(wc, span); } if (!ts->span.alarms) { if ((isr3 & 0x3) || (isr4 & 0xc0)) ts->span.count.timingslips++; if (debug & DEBUG_MAIN) { if (isr3 & 0x02) dev_notice(&wc->dev->dev, "TE%d10P: RECEIVE " "slip NEGATIVE on span %d\n", wc->numspans, span + 1); if (isr3 & 0x01) dev_notice(&wc->dev->dev, "TE%d10P: RECEIVE " "slip POSITIVE on span %d\n", wc->numspans, span + 1); if (isr4 & 0x80) dev_notice(&wc->dev->dev, "TE%dXXP: TRANSMIT " "slip POSITIVE on span %d\n", wc->numspans, span + 1); if (isr4 & 0x40) dev_notice(&wc->dev->dev, "TE%d10P: TRANSMIT " "slip NEGATIVE on span %d\n", wc->numspans, span + 1); } } else ts->span.count.timingslips = 0; spin_lock_irqsave(&wc->reglock, flags); /* HDLC controller checks - receive side */ if (!ts->sigchan) { spin_unlock_irqrestore(&wc->reglock, flags); return; } sigchan = ts->sigchan; spin_unlock_irqrestore(&wc->reglock, flags); if (isr0 & FRMR_ISR0_RME) { readsize = (t4_framer_in(wc, span, FRMR_RBCH) << 8) | t4_framer_in(wc, span, FRMR_RBCL); if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "Received data length is %d " "(%d)\n", readsize, readsize & FRMR_RBCL_MAX_SIZE); /* RPF isn't set on last part of frame */ if ((readsize > 0) && ((readsize &= FRMR_RBCL_MAX_SIZE) == 0)) readsize = FRMR_RBCL_MAX_SIZE + 1; } else if (isr0 & FRMR_ISR0_RPF) readsize = FRMR_RBCL_MAX_SIZE + 1; if (readsize > 0) { int i; unsigned char readbuf[FRMR_RBCL_MAX_SIZE + 1]; if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "Framer %d: Got RPF/RME! " "readsize is %d\n", sigchan->span->offset, readsize); for (i = 0; i < readsize; i++) readbuf[i] = t4_framer_in(wc, span, FRMR_RXFIFO); /* Tell the framer to clear the RFIFO */ t4_framer_cmd_wait(wc, span, FRMR_CMDR_RMC); if (debug & DEBUG_FRAMER) { dev_notice(&wc->dev->dev, "RX("); for (i = 0; i < readsize; i++) dev_notice(&wc->dev->dev, "%s%02x", (i ? " " : ""), readbuf[i]); dev_notice(&wc->dev->dev, ")\n"); } if (isr0 & FRMR_ISR0_RME) { /* Do checks for HDLC problems */ unsigned char rsis = readbuf[readsize-1]; unsigned char rsis_reg = t4_framer_in(wc, span, FRMR_RSIS); ++ts->frames_in; if ((debug & DEBUG_FRAMER) && !(ts->frames_in & 0x0f)) dev_notice(&wc->dev->dev, "Received %d frames " "on span %d\n", ts->frames_in, span); if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "Received HDLC frame" " %d. RSIS = 0x%x (%x)\n", ts->frames_in, rsis, rsis_reg); if (!(rsis & FRMR_RSIS_CRC16)) { if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "CRC check " "failed %d\n", span); dahdi_hdlc_abort(sigchan, DAHDI_EVENT_BADFCS); } else if (rsis & FRMR_RSIS_RAB) { if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "ABORT of " "current frame due to " "overflow %d\n", span); dahdi_hdlc_abort(sigchan, DAHDI_EVENT_ABORT); } else if (rsis & FRMR_RSIS_RDO) { if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "HDLC " "overflow occured %d\n", span); dahdi_hdlc_abort(sigchan, DAHDI_EVENT_OVERRUN); } else if (!(rsis & FRMR_RSIS_VFR)) { if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "Valid Frame" " check failed on span %d\n", span); dahdi_hdlc_abort(sigchan, DAHDI_EVENT_ABORT); } else { dahdi_hdlc_putbuf(sigchan, readbuf, readsize - 1); dahdi_hdlc_finish(sigchan); if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "Received " "valid HDLC frame on span %d" "\n", span); } } else if (isr0 & FRMR_ISR0_RPF) dahdi_hdlc_putbuf(sigchan, readbuf, readsize); } /* Transmit side */ if (isr1 & FRMR_ISR1_XDU) { if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "XDU: Resetting signal " "controller!\n"); t4_framer_cmd_wait(wc, span, FRMR_CMDR_SRES); } else if (isr1 & FRMR_ISR1_XPR) { if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "Sigchan %d is %p\n", sigchan->chanpos, sigchan); if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "Framer %d: Got XPR!\n", sigchan->span->offset); t4_hdlc_xmit_fifo(wc, span, ts); } if (isr1 & FRMR_ISR1_ALLS) { if (debug & DEBUG_FRAMER) dev_notice(&wc->dev->dev, "ALLS received\n"); } } #ifdef SUPPORT_GEN1 static irqreturn_t _t4_interrupt(int irq, void *dev_id) { struct t4 *wc = dev_id; unsigned long flags; int x; unsigned int status; unsigned int status2; /* Make sure it's really for us */ status = __t4_pci_in(wc, WC_INTR); /* Process framer interrupts */ status2 = t4_framer_in(wc, 0, FRMR_CIS); if (status2 & 0x0f) { for (x = 0; x < wc->numspans; ++x) { if (status2 & (1 << x)) t4_framer_interrupt(wc, x); } } /* Ignore if it's not for us */ if (!status) return IRQ_NONE; __t4_pci_out(wc, WC_INTR, 0); if (!wc->spansstarted) { dev_notice(&wc->dev->dev, "Not prepped yet!\n"); return IRQ_NONE; } wc->intcount++; if (status & 0x3) { t4_receiveprep(wc, status); t4_transmitprep(wc, status); } t4_do_counters(wc); x = wc->intcount & 15 /* 63 */; switch(x) { case 0: case 1: case 2: case 3: t4_check_sigbits(wc, x); break; case 4: case 5: case 6: case 7: t4_check_alarms(wc, x - 4); break; } spin_lock_irqsave(&wc->reglock, flags); __handle_leds(wc); if (test_bit(T4_CHECK_TIMING, &wc->checkflag)) __t4_set_timing_source_auto(wc); spin_unlock_irqrestore(&wc->reglock, flags); return IRQ_RETVAL(1); } DAHDI_IRQ_HANDLER(t4_interrupt) { irqreturn_t ret; unsigned long flags; local_irq_save(flags); ret = _t4_interrupt(irq, dev_id); local_irq_restore(flags); return ret; } #endif static int t4_allocate_buffers(struct t4 *wc, int numbufs, void **oldalloc, dma_addr_t *oldwritedma) { void *alloc; dma_addr_t writedma; /* 32 channels, Double-buffer, Read/Write, 4 spans */ alloc = pci_alloc_consistent(wc->dev, numbufs * T4_BASE_SIZE(wc) * 2, &writedma); if (!alloc) { dev_notice(&wc->dev->dev, "wct%dxxp: Unable to allocate " "DMA-able memory\n", wc->numspans); return -ENOMEM; } if (oldwritedma) *oldwritedma = wc->writedma; if (oldalloc) *oldalloc = wc->writechunk; wc->writechunk = alloc; wc->writedma = writedma; /* Read is after the whole write piece (in words) */ wc->readchunk = wc->writechunk + (T4_BASE_SIZE(wc) * numbufs) / 4; /* Same thing but in bytes... */ wc->readdma = wc->writedma + (T4_BASE_SIZE(wc) * numbufs); wc->numbufs = numbufs; /* Initialize Write/Buffers to all blank data */ memset(wc->writechunk, 0x00, T4_BASE_SIZE(wc) * numbufs); memset(wc->readchunk, 0xff, T4_BASE_SIZE(wc) * numbufs); if (debug) { dev_notice(&wc->dev->dev, "DMA memory base of size %d at " \ "%p. Read: %p and Write %p\n", numbufs * T4_BASE_SIZE(wc) * 2, wc->writechunk, wc->readchunk, wc->writechunk); } return 0; } static void t4_increase_latency(struct t4 *wc, int newlatency) { unsigned long flags; void *oldalloc; dma_addr_t oldaddr; int oldbufs; spin_lock_irqsave(&wc->reglock, flags); __t4_pci_out(wc, WC_DMACTRL, 0x00000000); /* Acknowledge any pending interrupts */ __t4_pci_out(wc, WC_INTR, 0x00000000); __t4_pci_in(wc, WC_VERSION); oldbufs = wc->numbufs; if (t4_allocate_buffers(wc, newlatency, &oldalloc, &oldaddr)) { dev_info(&wc->dev->dev, "Error allocating latency buffers for " "latency of %d\n", newlatency); __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); spin_unlock_irqrestore(&wc->reglock, flags); return; } __t4_pci_out(wc, WC_RDADDR, wc->readdma); __t4_pci_out(wc, WC_WRADDR, wc->writedma); __t4_pci_in(wc, WC_VERSION); __t4_pci_out(wc, 5, (ms_per_irq << 16) | newlatency); __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); __t4_pci_in(wc, WC_VERSION); wc->rxident = 0; wc->lastindex = 0; spin_unlock_irqrestore(&wc->reglock, flags); pci_free_consistent(wc->dev, T4_BASE_SIZE(wc) * oldbufs * 2, oldalloc, oldaddr); dev_info(&wc->dev->dev, "Increased latency to %d\n", newlatency); } static void t4_isr_bh(unsigned long data) { struct t4 *wc = (struct t4 *)data; if (test_bit(T4_CHANGE_LATENCY, &wc->checkflag)) { if (wc->needed_latency != wc->numbufs) { t4_increase_latency(wc, wc->needed_latency); clear_bit(T4_CHANGE_LATENCY, &wc->checkflag); } } #ifdef VPM_SUPPORT if (wc->vpm) { if (test_and_clear_bit(T4_CHECK_VPM, &wc->checkflag)) { /* How stupid is it that the octasic can't generate an * interrupt when there's a tone, in spite of what * their documentation says? */ t4_check_vpm(wc); } } #endif } static irqreturn_t _t4_interrupt_gen2(int irq, void *dev_id) { struct t4 *wc = dev_id; unsigned int status; unsigned char rxident, expected; /* Check this first in case we get a spurious interrupt */ if (unlikely(test_bit(T4_STOP_DMA, &wc->checkflag))) { /* Stop DMA cleanly if requested */ wc->dmactrl = 0x0; t4_pci_out(wc, WC_DMACTRL, 0x00000000); /* Acknowledge any pending interrupts */ t4_pci_out(wc, WC_INTR, 0x00000000); spin_lock(&wc->reglock); __t4_set_sclk_src(wc, WC_SELF, 0, 0); spin_unlock(&wc->reglock); return IRQ_RETVAL(1); } /* Make sure it's really for us */ status = __t4_pci_in(wc, WC_INTR); /* Ignore if it's not for us */ if (!(status & 0x7)) { return IRQ_NONE; } #ifdef ENABLE_WORKQUEUES __t4_pci_out(wc, WC_INTR, status & 0x00000008); #endif if (unlikely(!wc->spansstarted)) { dev_info(&wc->dev->dev, "Not prepped yet!\n"); return IRQ_NONE; } wc->intcount++; if ((wc->devtype->flags & FLAG_5THGEN) && (status & 0x2)) { rxident = (status >> 16) & 0x7f; expected = (wc->rxident + ms_per_irq) % 128; if ((rxident != expected) && !test_bit(T4_IGNORE_LATENCY, &wc->checkflag)) { int needed_latency; int smallest_max; if (debug & DEBUG_MAIN) dev_warn(&wc->dev->dev, "Missed interrupt. " "Expected ident of %d and got ident " "of %d\n", expected, rxident); if (test_bit(T4_IGNORE_LATENCY, &wc->checkflag)) { dev_info(&wc->dev->dev, "Should have ignored latency\n"); } if (rxident > wc->rxident) { needed_latency = rxident - wc->rxident; } else { needed_latency = (128 - wc->rxident) + rxident; } needed_latency += 1; smallest_max = (max_latency >= GEN5_MAX_LATENCY) ? GEN5_MAX_LATENCY : max_latency; if (needed_latency > smallest_max) { dev_info(&wc->dev->dev, "Truncating latency " "request to %d instead of %d\n", smallest_max, needed_latency); needed_latency = smallest_max; } if (needed_latency > wc->numbufs) { dev_info(&wc->dev->dev, "Need to increase " "latency. Estimated latency should " "be %d\n", needed_latency); wc->ddev->irqmisses++; wc->needed_latency = needed_latency; __t4_pci_out(wc, WC_DMACTRL, 0x00000000); set_bit(T4_CHANGE_LATENCY, &wc->checkflag); goto out; } } wc->rxident = rxident; } #ifdef DEBUG if (unlikely((wc->intcount < 20))) dev_dbg(&wc->dev->dev, "2G: Got interrupt, status = %08x, " "CIS = %04x\n", status, t4_framer_in(wc, 0, FRMR_CIS)); #endif if (likely(status & 0x2)) { #ifdef ENABLE_WORKQUEUES int cpus = num_online_cpus(); atomic_set(&wc->worklist, wc->numspans); if (wc->tspans[0]->span.flags & DAHDI_FLAG_RUNNING) t4_queue_work(wc->workq, &wc->tspans[0]->swork, 0); else atomic_dec(&wc->worklist); if (wc->tspans[1]->span.flags & DAHDI_FLAG_RUNNING) t4_queue_work(wc->workq, &wc->tspans[1]->swork, 1 % cpus); else atomic_dec(&wc->worklist); if (wc->numspans == 4) { if (wc->tspans[2]->span.flags & DAHDI_FLAG_RUNNING) t4_queue_work(wc->workq, &wc->tspans[2]->swork, 2 % cpus); else atomic_dec(&wc->worklist); if (wc->tspans[3]->span.flags & DAHDI_FLAG_RUNNING) t4_queue_work(wc->workq, &wc->tspans[3]->swork, 3 % cpus); else atomic_dec(&wc->worklist); } #else unsigned int reg5 = __t4_pci_in(wc, 5); #ifdef DEBUG if (wc->intcount < 20) dev_info(&wc->dev->dev, "Reg 5 is %08x\n", reg5); #endif if (wc->devtype->flags & FLAG_5THGEN) { unsigned int current_index = (reg5 >> 8) & 0x7f; while (((wc->lastindex + 1) % wc->numbufs) != current_index) { wc->lastindex = (wc->lastindex + 1) % wc->numbufs; setup_chunks(wc, wc->lastindex); t4_prep_gen2(wc); } } else { t4_prep_gen2(wc); } #endif t4_do_counters(wc); spin_lock(&wc->reglock); __handle_leds(wc); spin_unlock(&wc->reglock); } if (unlikely(status & 0x1)) { unsigned char cis; cis = t4_framer_in(wc, 0, FRMR_CIS); if (cis & FRMR_CIS_GIS1) t4_framer_interrupt(wc, 0); if (cis & FRMR_CIS_GIS2) t4_framer_interrupt(wc, 1); if (cis & FRMR_CIS_GIS3) t4_framer_interrupt(wc, 2); if (cis & FRMR_CIS_GIS4) t4_framer_interrupt(wc, 3); if (is_octal(wc)) { if (cis & FRMR_CIS_GIS5) t4_framer_interrupt(wc, 4); if (cis & FRMR_CIS_GIS6) t4_framer_interrupt(wc, 5); if (cis & FRMR_CIS_GIS7) t4_framer_interrupt(wc, 6); if (cis & FRMR_CIS_GIS8) t4_framer_interrupt(wc, 7); } } #ifdef VPM_SUPPORT if (wc->vpm && vpmdtmfsupport) { /* How stupid is it that the octasic can't generate an * interrupt when there's a tone, in spite of what their * documentation says? */ if (!(wc->intcount & 0xf)) set_bit(T4_CHECK_VPM, &wc->checkflag); } #endif spin_lock(&wc->reglock); if (unlikely(test_bit(T4_CHECK_TIMING, &wc->checkflag))) { __t4_set_timing_source_auto(wc); } spin_unlock(&wc->reglock); out: if (unlikely(test_bit(T4_CHANGE_LATENCY, &wc->checkflag) || test_bit(T4_CHECK_VPM, &wc->checkflag))) tasklet_schedule(&wc->t4_tlet); #ifndef ENABLE_WORKQUEUES __t4_pci_out(wc, WC_INTR, 0); #endif return IRQ_RETVAL(1); } DAHDI_IRQ_HANDLER(t4_interrupt_gen2) { irqreturn_t ret; unsigned long flags; local_irq_save(flags); ret = _t4_interrupt_gen2(irq, dev_id); local_irq_restore(flags); return ret; } #ifdef SUPPORT_GEN1 static int t4_reset_dma(struct t4 *wc) { /* Turn off DMA and such */ wc->dmactrl = 0x0; t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); t4_pci_out(wc, WC_COUNT, 0); t4_pci_out(wc, WC_RDADDR, 0); t4_pci_out(wc, WC_WRADDR, 0); t4_pci_out(wc, WC_INTR, 0); /* Turn it all back on */ t4_pci_out(wc, WC_RDADDR, wc->readdma); t4_pci_out(wc, WC_WRADDR, wc->writedma); t4_pci_out(wc, WC_COUNT, ((DAHDI_MAX_CHUNKSIZE * 2 * 32 - 1) << 18) | ((DAHDI_MAX_CHUNKSIZE * 2 * 32 - 1) << 2)); t4_pci_out(wc, WC_INTR, 0); #ifdef VPM_SUPPORT wc->dmactrl = 0xc0000000 | (1 << 29) | ((wc->vpm) ? T4_VPM_PRESENT : 0); #else wc->dmactrl = 0xc0000000 | (1 << 29); #endif if (noburst) wc->dmactrl |= (1 << 26); t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); return 0; } #endif #ifdef VPM_SUPPORT static void t4_vpm_init(struct t4 *wc) { int laws[8] = { 0, }; int x; unsigned int vpm_capacity; struct firmware embedded_firmware; const struct firmware *firmware = &embedded_firmware; #if !defined(HOTPLUG_FIRMWARE) extern void _binary_dahdi_fw_oct6114_064_bin_size; extern void _binary_dahdi_fw_oct6114_128_bin_size; extern void _binary_dahdi_fw_oct6114_256_bin_size; extern u8 _binary_dahdi_fw_oct6114_064_bin_start[]; extern u8 _binary_dahdi_fw_oct6114_128_bin_start[]; extern u8 _binary_dahdi_fw_oct6114_256_bin_start[]; #else static const char oct064_firmware[] = "dahdi-fw-oct6114-064.bin"; static const char oct128_firmware[] = "dahdi-fw-oct6114-128.bin"; static const char oct256_firmware[] = "dahdi-fw-oct6114-256.bin"; #endif if (!vpmsupport) { dev_info(&wc->dev->dev, "VPM450: Support Disabled\n"); return; } /* Turn on GPIO/DATA mux if supported */ t4_gpio_setdir(wc, (1 << 24), (1 << 24)); __t4_raw_oct_out(wc, 0x000a, 0x5678); __t4_raw_oct_out(wc, 0x0004, 0x1234); __t4_raw_oct_in(wc, 0x0004); __t4_raw_oct_in(wc, 0x000a); if (debug) dev_notice(&wc->dev->dev, "OCT Result: %04x/%04x\n", __t4_raw_oct_in(wc, 0x0004), __t4_raw_oct_in(wc, 0x000a)); if (__t4_raw_oct_in(wc, 0x0004) != 0x1234) { dev_notice(&wc->dev->dev, "VPM450: Not Present\n"); return; } /* Setup alaw vs ulaw rules */ for (x = 0;x < wc->numspans; x++) { if (wc->tspans[x]->span.channels > 24) laws[x] = 1; } vpm_capacity = get_vpm450m_capacity(&wc->dev->dev); if (vpm_capacity != wc->numspans * 32) { dev_info(&wc->dev->dev, "Disabling VPMOCT%03d. TE%dXXP"\ " requires a VPMOCT%03d", vpm_capacity, wc->numspans, wc->numspans*32); return; } switch (vpm_capacity) { case 64: #if defined(HOTPLUG_FIRMWARE) if ((request_firmware(&firmware, oct064_firmware, &wc->dev->dev) != 0) || !firmware) { dev_notice(&wc->dev->dev, "VPM450: firmware %s not " "available from userspace\n", oct064_firmware); return; } #else embedded_firmware.data = _binary_dahdi_fw_oct6114_064_bin_start; /* Yes... this is weird. objcopy gives us a symbol containing the size of the firmware, not a pointer a variable containing the size. The only way we can get the value of the symbol is to take its address, so we define it as a pointer and then cast that value to the proper type. */ embedded_firmware.size = (size_t) &_binary_dahdi_fw_oct6114_064_bin_size; #endif break; case 128: #if defined(HOTPLUG_FIRMWARE) if ((request_firmware(&firmware, oct128_firmware, &wc->dev->dev) != 0) || !firmware) { dev_notice(&wc->dev->dev, "VPM450: firmware %s not " "available from userspace\n", oct128_firmware); return; } #else embedded_firmware.data = _binary_dahdi_fw_oct6114_128_bin_start; /* Yes... this is weird. objcopy gives us a symbol containing the size of the firmware, not a pointer a variable containing the size. The only way we can get the value of the symbol is to take its address, so we define it as a pointer and then cast that value to the proper type. */ embedded_firmware.size = (size_t) &_binary_dahdi_fw_oct6114_128_bin_size; #endif break; case 256: #if defined(HOTPLUG_FIRMWARE) if ((request_firmware(&firmware, oct256_firmware, &wc->dev->dev) != 0) || !firmware) { dev_notice(&wc->dev->dev, "VPM450: firmware %s not " "available from userspace\n", oct256_firmware); return; } #else embedded_firmware.data = _binary_dahdi_fw_oct6114_256_bin_start; /* Yes... this is weird. objcopy gives us a symbol containing the size of the firmware, not a pointer a variable containing the size. The only way we can get the value of the symbol is to take its address, so we define it as a pointer and then cast that value to the proper type. */ embedded_firmware.size = (size_t) &_binary_dahdi_fw_oct6114_256_bin_size; #endif break; default: dev_notice(&wc->dev->dev, "Unsupported channel capacity found " "on VPM module (%d).\n", vpm_capacity); return; } wc->vpm = init_vpm450m(&wc->dev->dev, laws, wc->numspans, firmware); if (!wc->vpm) { dev_notice(&wc->dev->dev, "VPM450: Failed to initialize\n"); if (firmware != &embedded_firmware) release_firmware(firmware); return; } if (firmware != &embedded_firmware) release_firmware(firmware); if (vpmdtmfsupport == -1) { dev_info(&wc->dev->dev, "VPM450: hardware DTMF disabled.\n"); vpmdtmfsupport = 0; } dev_info(&wc->dev->dev, "VPM450: Present and operational servicing %d " "span(s)\n", wc->numspans); } #endif /* VPM_SUPPORT */ static void t4_tsi_reset(struct t4 *wc) { int x; if (is_octal(wc)) { for (x = 0; x < 256; x++) { wc->dmactrl &= ~0x0001ffff; wc->dmactrl |= (0x00004000 | ((x & 0x7f) << 7) | ((x >> 7) << 15)); t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); } wc->dmactrl &= ~0x0001ffff; } else { for (x = 0; x < 128; x++) { wc->dmactrl &= ~0x00007fff; wc->dmactrl |= (0x00004000 | (x << 7)); t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); } wc->dmactrl &= ~0x00007fff; } t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); } /* Note that channels here start from 1 */ static void t4_tsi_assign(struct t4 *wc, int fromspan, int fromchan, int tospan, int tochan) { unsigned long flags; int fromts, tots; fromts = (fromspan << 5) |(fromchan); tots = (tospan << 5) | (tochan); if (!has_e1_span(wc)) { fromts += 4; tots += 4; } spin_lock_irqsave(&wc->reglock, flags); if (is_octal(wc)) { int fromts_b = fromts & 0x7f; int fromts_t = fromts >> 7; int tots_b = tots & 0x7f; int tots_t = tots >> 7; wc->dmactrl &= ~0x0001ffff; wc->dmactrl |= ((fromts_t << 16) | (tots_t << 15) | 0x00004000 | (tots_b << 7) | (fromts_b)); __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); wc->dmactrl &= ~0x0001ffff; __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); } else { wc->dmactrl &= ~0x00007fff; wc->dmactrl |= (0x00004000 | (tots << 7) | (fromts)); __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); wc->dmactrl &= ~0x00007fff; __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); } spin_unlock_irqrestore(&wc->reglock, flags); } static void t4_tsi_unassign(struct t4 *wc, int tospan, int tochan) { unsigned long flags; int tots; tots = (tospan << 5) | (tochan); if (!has_e1_span(wc)) tots += 4; spin_lock_irqsave(&wc->reglock, flags); if (is_octal(wc)) { int tots_b = tots & 0x7f; int tots_t = tots >> 7; wc->dmactrl &= ~0x0001ffff; wc->dmactrl |= ((tots_t << 15) | 0x00004000 | (tots_b << 7)); __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); if (debug & DEBUG_TSI) dev_notice(&wc->dev->dev, "Sending '%08x\n", wc->dmactrl); wc->dmactrl &= ~0x0001ffff; __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); } else { wc->dmactrl &= ~0x00007fff; wc->dmactrl |= (0x00004000 | (tots << 7)); __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); if (debug & DEBUG_TSI) dev_notice(&wc->dev->dev, "Sending '%08x\n", wc->dmactrl); wc->dmactrl &= ~0x00007fff; __t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); } spin_unlock_irqrestore(&wc->reglock, flags); } #ifndef CONFIG_NOEXTENDED_RESET static void t4_extended_reset(struct t4 *wc) { unsigned int oldreg = t4_pci_in(wc, 0x4); udelay(1000); t4_pci_out(wc, 0x4, 0x42000000); t4_pci_out(wc, 0xa, 0x42000000); t4_pci_out(wc, 0xa, 0x00080000); t4_pci_out(wc, 0xa, 0x00080000); t4_pci_out(wc, 0xa, 0x00080000); t4_pci_out(wc, 0xa, 0x00180000); t4_pci_out(wc, 0xa, 0x00080000); t4_pci_out(wc, 0xa, 0x00180000); t4_pci_out(wc, 0xa, 0x00080000); t4_pci_out(wc, 0xa, 0x00180000); t4_pci_out(wc, 0xa, 0x00080000); t4_pci_out(wc, 0xa, 0x00180000); t4_pci_out(wc, 0xa, 0x00080000); t4_pci_out(wc, 0xa, 0x00180000); t4_pci_out(wc, 0xa, 0x00080000); t4_pci_out(wc, 0xa, 0x00180000); t4_pci_out(wc, 0x4, oldreg); udelay(1000); } #endif #define SPI_CS (0) #define SPI_CLK (1) #define SPI_IO0 (2) #define SPI_IO1 (3) #define SPI_IO3 (4) #define SPI_IO2 (5) #define SPI_IN (0) #define SPI_OUT (1) #define ESPI_REG 13 static void t8_clear_bit(struct t4 *wc, int whichb) { wc->st.wrreg &= ~(1 << whichb); } static void t8_set_bit(struct t4 *wc, int whichb, int val) { t8_clear_bit(wc, whichb); wc->st.wrreg |= (val << whichb); } static int t8_get_bit(struct t4 *wc, int whichb) { return (wc->st.rdreg >> whichb) & 1; } static void set_iodir(struct t4 *wc, int whichb, int dir) { whichb += 16; t8_clear_bit(wc, whichb); t8_set_bit(wc, whichb, dir); } static void write_hwreg(struct t4 *wc) { t4_pci_out(wc, ESPI_REG, wc->st.wrreg); } static void read_hwreg(struct t4 *wc) { wc->st.rdreg = t4_pci_in(wc, ESPI_REG); } static void set_cs(struct t4 *wc, int state) { t8_set_bit(wc, SPI_CS, state); write_hwreg(wc); } static void set_clk(struct t4 *wc, int clk) { t8_set_bit(wc, SPI_CLK, clk); write_hwreg(wc); } static void clk_bit_out(struct t4 *wc, int val) { t8_set_bit(wc, SPI_IO0, val & 1); set_clk(wc, 0); set_clk(wc, 1); } static void shift_out(struct t4 *wc, int val) { int i; for (i = 7; i >= 0; i--) clk_bit_out(wc, (val >> i) & 1); } static int clk_bit_in(struct t4 *wc) { int ret; set_clk(wc, 0); read_hwreg(wc); ret = t8_get_bit(wc, SPI_IO1); set_clk(wc, 1); return ret; } static int shift_in(struct t4 *wc) { int ret = 0; int i; int bit; for (i = 7; i >= 0; i--) { bit = clk_bit_in(wc); ret |= ((bit & 1) << i); } return ret; } static void write_enable(struct t4 *wc) { int cmd = 0x06; set_cs(wc, 0); shift_out(wc, cmd); set_cs(wc, 1); } static int read_sr1(struct t4 *wc) { int cmd = 0x05; int ret; set_cs(wc, 0); shift_out(wc, cmd); ret = shift_in(wc); set_cs(wc, 1); return ret; } static void clear_busy(struct t4 *wc) { static const int SR1_BUSY = (1 << 0); unsigned long stop; stop = jiffies + 2*HZ; while (read_sr1(wc) & SR1_BUSY) { if (time_after(jiffies, stop)) { if (printk_ratelimit()) { dev_err(&wc->dev->dev, "Lockup in %s\n", __func__); } break; } cond_resched(); } } static void sector_erase(struct t4 *wc, uint32_t addr) { int cmd = 0x20; write_enable(wc); set_cs(wc, 0); shift_out(wc, cmd); shift_out(wc, (addr >> 16) & 0xff); shift_out(wc, (addr >> 8) & 0xff); shift_out(wc, (addr >> 0) & 0xff); set_cs(wc, 1); clear_busy(wc); } static void erase_half(struct t4 *wc) { uint32_t addr = 0x00080000; uint32_t i; dev_info(&wc->dev->dev, "Erasing octal firmware\n"); for (i = addr; i < (addr + 0x80000); i += 4096) sector_erase(wc, i); } #define T8_FLASH_PAGE_SIZE 256UL static void t8_update_firmware_page(struct t4 *wc, u32 address, const u8 *page_data, size_t size) { int i; write_enable(wc); set_cs(wc, 0); shift_out(wc, 0x02); shift_out(wc, (address >> 16) & 0xff); shift_out(wc, (address >> 8) & 0xff); shift_out(wc, (address >> 0) & 0xff); for (i = 0; i < size; ++i) shift_out(wc, page_data[i]); set_cs(wc, 1); clear_busy(wc); } static int t8_update_firmware(struct t4 *wc, const struct firmware *fw, const char *t8_firmware) { int res; size_t offset = 0; const u32 BASE_ADDRESS = 0x00080000; const u8 *data, *end; size_t size = 0; /* Erase flash */ erase_half(wc); dev_info(&wc->dev->dev, "Uploading %s. This can take up to 30 seconds.\n", t8_firmware); data = &fw->data[sizeof(struct t8_firm_header)]; end = &fw->data[fw->size]; while (data < end) { /* Calculate the tail end of data that's shorter than a page */ size = min(T8_FLASH_PAGE_SIZE, (unsigned long)(end - data)); t8_update_firmware_page(wc, BASE_ADDRESS + offset, data, size); data += T8_FLASH_PAGE_SIZE; offset += T8_FLASH_PAGE_SIZE; cond_resched(); } /* Reset te820 fpga after loading firmware */ dev_info(&wc->dev->dev, "Firmware load complete. Reseting device.\n"); res = pci_save_state(wc->dev); if (res) goto error_exit; /* Set the fpga reset bits and clobber the remainder of the * register, device will be reset anyway */ t4_pci_out(wc, WC_LEDS, 0xe0000000); msleep(1000); pci_restore_state(wc->dev); /* Signal the driver to restart initialization. * This will back out all initialization so far and * restart the driver load process */ return -EAGAIN; error_exit: return res; } static char read_flash_byte(struct t4 *wc, unsigned int addr) { int cmd = 0x03; char data; set_cs(wc, 0); shift_out(wc, cmd); shift_out(wc, (addr >> 16) & 0xff); shift_out(wc, (addr >> 8) & 0xff); shift_out(wc, (addr >> 0) & 0xff); data = shift_in(wc) & 0xff; set_cs(wc, 1); return data; } /** * t8_read_serial - Returns the serial number of the board. * @wc: The board whos serial number we are reading. * * The buffer returned is dynamically allocated and must be kfree'd by the * caller. If memory could not be allocated, NULL is returned. * * Must be called in process context. * */ static char *t8_read_serial(struct t4 *wc) { int i; static const int MAX_SERIAL = 14; static const u32 base_addr = 0x00080000-256; unsigned char c; unsigned char *serial = kzalloc(MAX_SERIAL + 1, GFP_KERNEL); if (!serial) return NULL; for (i = 0; i < MAX_SERIAL; ++i) { c = read_flash_byte(wc, base_addr+i); if ((c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || (c >= '0' && c <= '9')) serial[i] = c; else break; } if (!i) { kfree(serial); serial = NULL; } return serial; } static void setup_spi(struct t4 *wc) { wc->st.rdreg = wc->st.wrreg = 0; set_iodir(wc, SPI_IO0, SPI_OUT); set_iodir(wc, SPI_IO1, SPI_IN); set_iodir(wc, SPI_CS, SPI_OUT); set_iodir(wc, SPI_CLK, SPI_OUT); t8_set_bit(wc, SPI_CS, 1); t8_set_bit(wc, SPI_CLK, 1); write_hwreg(wc); } static int t8_check_firmware(struct t4 *wc, unsigned int version) { const struct firmware *fw; static const char t8_firmware[] = "dahdi-fw-te820.bin"; const struct t8_firm_header *header; int res = 0; u32 crc; res = request_firmware(&fw, t8_firmware, &wc->dev->dev); if (res) { dev_info(&wc->dev->dev, "firmware %s not " "available from userspace\n", t8_firmware); goto cleanup; } header = (const struct t8_firm_header *)fw->data; /* Check the crc before anything else */ crc = crc32(~0, &fw->data[10], fw->size - 10) ^ ~0; if (memcmp("DIGIUM", header->header, sizeof(header->header)) || (le32_to_cpu(header->chksum) != crc)) { dev_info(&wc->dev->dev, "%s is invalid. Please reinstall.\n", t8_firmware); goto cleanup; } /* Spi struct must be setup before any access to flash memory */ setup_spi(wc); /* Check the two firmware versions */ if (le32_to_cpu(header->version) == version) goto cleanup; dev_info(&wc->dev->dev, "%s Version: %08x available for flash\n", t8_firmware, header->version); res = t8_update_firmware(wc, fw, t8_firmware); if (res && res != -EAGAIN) { dev_info(&wc->dev->dev, "Failed to load firmware %s\n", t8_firmware); } cleanup: release_firmware(fw); return res; } static int __t4_hardware_init_1(struct t4 *wc, unsigned int cardflags, bool first_time) { unsigned int version; int res; version = t4_pci_in(wc, WC_VERSION); if (is_octal(wc) && first_time) { dev_info(&wc->dev->dev, "Firmware Version: %01x.%02x\n", (version & 0xf00) >> 8, version & 0xff); } else if (first_time) { dev_info(&wc->dev->dev, "Firmware Version: %08x\n", version); } if (debug) { dev_info(&wc->dev->dev, "Burst Mode: %s\n", (!(cardflags & FLAG_BURST) && noburst) ? "Off" : "On"); #ifdef ENABLE_WORKQUEUES dev_info(&wc->dev->dev, "Work Queues: Enabled\n"); #endif } /* Check the field updatable firmware for the wcte820 */ if (is_octal(wc)) { res = t8_check_firmware(wc, version); if (res) return res; wc->ddev->hardware_id = t8_read_serial(wc); } #if defined(CONFIG_FORCE_EXTENDED_RESET) t4_extended_reset(wc); #elif !defined(CONFIG_NOEXTENDED_RESET) if (wc->devtype->flags & FLAG_EXPRESS) t4_extended_reset(wc); #endif /* Make sure DMA engine is not running and interrupts are acknowledged */ wc->dmactrl = 0x0; t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); /* Reset Framer and friends */ t4_pci_out(wc, WC_LEDS, 0x00000000); /* Set DMA addresses */ t4_pci_out(wc, WC_RDADDR, wc->readdma); t4_pci_out(wc, WC_WRADDR, wc->writedma); /* Setup counters, interrupt flags (ignored in Gen2) */ if (cardflags & FLAG_2NDGEN) { t4_tsi_reset(wc); } else { t4_pci_out(wc, WC_COUNT, ((DAHDI_MAX_CHUNKSIZE * 2 * 32 - 1) << 18) | ((DAHDI_MAX_CHUNKSIZE * 2 * 32 - 1) << 2)); } /* Reset pending interrupts */ t4_pci_out(wc, WC_INTR, 0x00000000); /* Read T1/E1 status */ if (first_time) { if (!strcasecmp("auto", default_linemode)) { if (-1 == t1e1override) { wc->t1e1 = (((t4_pci_in(wc, WC_LEDS)) & 0x0f00) >> 8); wc->t1e1 &= 0xf; if (is_octal(wc)) { wc->t1e1 |= ((t4_pci_in(wc, WC_LEDS2)) & 0x0f00) >> 4; } } else { dev_warn(&wc->dev->dev, "'t1e1override' is deprecated. Please use 'default_linemode'.\n"); wc->t1e1 = t1e1override & 0xf; } } else if (!strcasecmp("t1", default_linemode)) { wc->t1e1 = 0; } else if (!strcasecmp("e1", default_linemode)) { wc->t1e1 = 0xff; } else if (!strcasecmp("j1", default_linemode)) { wc->t1e1 = 0; j1mode = 1; } else { dev_err(&wc->dev->dev, "'%s' is an unknown linemode.\n", default_linemode); wc->t1e1 = 0; } } wc->order = ((t4_pci_in(wc, WC_LEDS)) & 0xf0000000) >> 28; order_index[wc->order]++; /* TE820 Auth Check */ if (is_octal(wc)) { unsigned long stop = jiffies + HZ; uint32_t donebit; t4_pci_out(wc, WC_LEDS2, WC_SET_AUTH); donebit = t4_pci_in(wc, WC_LEDS2); while (!(donebit & WC_GET_AUTH)) { if (time_after(jiffies, stop)) { /* Encryption check failed, stop operation */ dev_info(&wc->dev->dev, "Failed encryption check. " "Unloading driver.\n"); return -EIO; } msleep(20); donebit = t4_pci_in(wc, WC_LEDS2); } } return 0; } static int t4_hardware_init_1(struct t4 *wc, unsigned int cardflags) { return __t4_hardware_init_1(wc, cardflags, true); } static int __t4_hardware_init_2(struct t4 *wc, bool first_time) { int x; unsigned int regval; unsigned long flags; if (t4_pci_in(wc, WC_VERSION) >= 0xc01a0165) { wc->tspans[0]->spanflags |= FLAG_OCTOPT; } /* Setup LEDS, take out of reset */ t4_pci_out(wc, WC_LEDS, 0x000000ff); udelay(100); t4_activate(wc); udelay(100); /* In order to find out the QFALC framer version, we have to * temporarily turn off compat mode and take a peak at VSTR. We turn * compat back on when we are done. * */ spin_lock_irqsave(&wc->reglock, flags); regval = __t4_framer_in(wc, 0, 0xd6); if (is_octal(wc)) regval |= (1 << 4); /* SSI16 - For 16 MHz multiplex mode with comp = 1 */ else regval |= (1 << 5); /* set COMP_DIS*/ __t4_framer_out(wc, 0, 0xd6, regval); spin_unlock_irqrestore(&wc->reglock, flags); if (!is_octal(wc)) { regval = t4_framer_in(wc, 0, 0x4a); if (first_time && regval == 0x05) { dev_info(&wc->dev->dev, "FALC Framer Version: 2.1 or " "earlier\n"); } else if (regval == 0x20) { if (first_time) dev_info(&wc->dev->dev, "FALC Framer Version: 3.1\n"); wc->falc31 = 1; } else if (first_time) { dev_info(&wc->dev->dev, "FALC Framer Version: Unknown " "(VSTR = 0x%02x)\n", regval); } } spin_lock_irqsave(&wc->reglock, flags); regval = __t4_framer_in(wc, 0, 0xd6); regval &= ~(1 << 5); /* clear COMP_DIS*/ __t4_framer_out(wc, 0, 0xd6, regval); __t4_framer_out(wc, 0, 0x4a, 0xaa); spin_unlock_irqrestore(&wc->reglock, flags); if (debug) { dev_info(&wc->dev->dev, "Board ID: %02x\n", wc->order); for (x = 0; x < 11; x++) { dev_info(&wc->dev->dev, "Reg %d: 0x%08x\n", x, t4_pci_in(wc, x)); } } wc->gpio = 0x00000000; t4_pci_out(wc, WC_GPIO, wc->gpio); t4_gpio_setdir(wc, (1 << 17), (1 << 17)); t4_gpio_setdir(wc, (0xff), (0xff)); return 0; } static int t4_hardware_init_2(struct t4 *wc) { return __t4_hardware_init_2(wc, true); } static int __devinit t4_launch(struct t4 *wc) { int x; int res; if (test_bit(DAHDI_FLAGBIT_REGISTERED, &wc->tspans[0]->span.flags)) return 0; if (debug) { dev_info(&wc->dev->dev, "TE%dXXP: Launching card: %d\n", wc->numspans, wc->order); } wc->ddev->manufacturer = "Digium"; if (!ignore_rotary && (1 == order_index[wc->order])) { wc->ddev->location = kasprintf(GFP_KERNEL, "Board ID Switch %d", wc->order); } else { bool express = ((wc->tspans[0]->spanflags & FLAG_EXPRESS) > 0); wc->ddev->location = kasprintf(GFP_KERNEL, "PCI%s Bus %02d Slot %02d", (express) ? " Express" : "", wc->dev->bus->number, PCI_SLOT(wc->dev->devfn) + 1); } if (!wc->ddev->location) return -ENOMEM; for (x = 0; x < wc->numspans; ++x) { list_add_tail(&wc->tspans[x]->span.device_node, &wc->ddev->spans); } tasklet_init(&wc->t4_tlet, t4_isr_bh, (unsigned long)wc); res = dahdi_register_device(wc->ddev, &wc->dev->dev); if (res) { dev_err(&wc->dev->dev, "Failed to register with DAHDI.\n"); return res; } return 0; } /** * wct4xxp_sort_cards - Sort the cards in card array by rotary switch settings. * */ static void wct4xxp_sort_cards(void) { int x; /* get the current number of probed cards and run a slice of a tail * insertion sort */ for (x = 0; x < MAX_T4_CARDS; x++) { if (!cards[x+1]) break; } for ( ; x > 0; x--) { if (cards[x]->order < cards[x-1]->order) { struct t4 *tmp = cards[x]; cards[x] = cards[x-1]; cards[x-1] = tmp; } else { /* if we're not moving it, we won't move any more * since all cards are sorted on addition */ break; } } } static int __devinit t4_init_one(struct pci_dev *pdev, const struct pci_device_id *ent) { int res; struct t4 *wc; unsigned int x; int init_latency; if (pci_enable_device(pdev)) { return -EIO; } wc = kzalloc(sizeof(*wc), GFP_KERNEL); if (!wc) return -ENOMEM; wc->ddev = dahdi_create_device(); if (!wc->ddev) { kfree(wc); return -ENOMEM; } spin_lock_init(&wc->reglock); wc->devtype = (const struct devtype *)(ent->driver_data); #ifdef CONFIG_WCT4XXP_DISABLE_ASPM if (is_pcie(wc)) { pci_disable_link_state(pdev->bus->self, PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1 | PCIE_LINK_STATE_CLKPM); }; #endif if (is_octal(wc)) wc->numspans = 8; else if (wc->devtype->flags & FLAG_2PORT) wc->numspans = 2; else wc->numspans = 4; wc->membase = pci_iomap(pdev, 0, 0); /* This rids of the Double missed interrupt message after loading */ wc->last0 = 1; if (pci_request_regions(pdev, wc->devtype->desc)) dev_info(&pdev->dev, "wct%dxxp: Unable to request regions\n", wc->numspans); if (debug) dev_info(&pdev->dev, "Found TE%dXXP\n", wc->numspans); wc->dev = pdev; /* Enable bus mastering */ pci_set_master(pdev); /* Keep track of which device we are */ pci_set_drvdata(pdev, wc); if (wc->devtype->flags & FLAG_5THGEN) { if ((ms_per_irq > 1) && (latency <= ((ms_per_irq) << 1))) { init_latency = ms_per_irq << 1; } else { if (latency > 2) init_latency = latency; else init_latency = 2; } dev_info(&wc->dev->dev, "5th gen card with initial latency of " "%d and %d ms per IRQ\n", init_latency, ms_per_irq); } else { if (wc->devtype->flags & FLAG_2NDGEN) init_latency = 1; else init_latency = 2; } if (max_latency < init_latency) { printk(KERN_INFO "maxlatency must be set to something greater than %d ms, increasing it to %d\n", init_latency, init_latency); max_latency = init_latency; } if (t4_allocate_buffers(wc, init_latency, NULL, NULL)) { return -ENOMEM; } /* Initialize hardware */ res = t4_hardware_init_1(wc, wc->devtype->flags); if (res) { /* If this function returns -EAGAIN, we expect * to attempt another driver load. Clean everything * up first */ pci_iounmap(wc->dev, wc->membase); pci_release_regions(wc->dev); pci_free_consistent(wc->dev, T4_BASE_SIZE(wc) * wc->numbufs * 2, wc->writechunk, wc->writedma); pci_set_drvdata(wc->dev, NULL); free_wc(wc); return res; } for(x = 0; x < MAX_T4_CARDS; x++) { if (!cards[x]) break; } if (x >= MAX_T4_CARDS) { dev_notice(&wc->dev->dev, "No cards[] slot available!!\n"); kfree(wc); return -ENOMEM; } wc->num = x; cards[x] = wc; #ifdef ENABLE_WORKQUEUES if (wc->devtype->flags & FLAG_2NDGEN) { char tmp[20]; sprintf(tmp, "te%dxxp[%d]", wc->numspans, wc->num); wc->workq = create_workqueue(tmp); } #endif /* Allocate pieces we need here */ for (x = 0; x < ports_on_framer(wc); x++) { struct t4_span *ts; enum linemode linemode; ts = kzalloc(sizeof(*ts), GFP_KERNEL); if (!ts) { free_wc(wc); return -ENOMEM; } wc->tspans[x] = ts; #ifdef ENABLE_WORKQUEUES INIT_WORK(&ts->swork, workq_handlespan, ts); #endif ts->spanflags |= wc->devtype->flags; linemode = (wc->t1e1 & (1 << x)) ? E1 : ((j1mode) ? J1 : T1); t4_alloc_channels(wc, wc->tspans[x], linemode); } /* Continue hardware intiialization */ t4_hardware_init_2(wc); #ifdef SUPPORT_GEN1 if (request_irq(pdev->irq, (wc->devtype->flags & FLAG_2NDGEN) ? t4_interrupt_gen2 : t4_interrupt, DAHDI_IRQ_SHARED, (wc->numspans == 8) ? "wct8xxp" : (wc->numspans == 2) ? "wct2xxp" : "wct4xxp", wc)) { #else if (!(wc->tspans[0]->spanflags & FLAG_2NDGEN)) { dev_notice(&wc->dev->dev, "This driver does not " "support 1st gen modules\n"); free_wc(wc); return -ENODEV; } if (request_irq(pdev->irq, t4_interrupt_gen2, DAHDI_IRQ_SHARED, "t4xxp", wc)) { #endif dev_notice(&wc->dev->dev, "Unable to request IRQ %d\n", pdev->irq); free_wc(wc); return -EIO; } init_spans(wc); if (!ignore_rotary) wct4xxp_sort_cards(); if (wc->ddev->hardware_id) { dev_info(&wc->dev->dev, "Found a Wildcard: %s (SN: %s)\n", wc->devtype->desc, wc->ddev->hardware_id); } else { dev_info(&wc->dev->dev, "Found a Wildcard: %s\n", wc->devtype->desc); } #ifdef VPM_SUPPORT if (!wc->vpm) { t4_vpm_init(wc); wc->dmactrl |= (wc->vpm) ? T4_VPM_PRESENT : 0; t4_pci_out(wc, WC_DMACTRL, wc->dmactrl); if (wc->vpm) set_span_devicetype(wc); } #endif create_sysfs_files(wc); res = 0; if (ignore_rotary) res = t4_launch(wc); return res; } static int t4_hardware_stop(struct t4 *wc) { /* Turn off DMA, leave interrupts enabled */ set_bit(T4_STOP_DMA, &wc->checkflag); /* Wait for interrupts to stop */ msleep(25); /* Turn off counter, address, etc */ if (wc->tspans[0]->spanflags & FLAG_2NDGEN) { t4_tsi_reset(wc); } else { t4_pci_out(wc, WC_COUNT, 0x000000); } t4_pci_out(wc, WC_RDADDR, 0x0000000); t4_pci_out(wc, WC_WRADDR, 0x0000000); wc->gpio = 0x00000000; t4_pci_out(wc, WC_GPIO, wc->gpio); t4_pci_out(wc, WC_LEDS, 0x00000000); if (debug) { dev_notice(&wc->dev->dev, "Stopped TE%dXXP, Turned off DMA\n", wc->numspans); } return 0; } static int __devinit t4_init_one_retry(struct pci_dev *pdev, const struct pci_device_id *ent) { int res; res = t4_init_one(pdev, ent); /* If the driver was reset by a firmware load, * try to load once again */ if (-EAGAIN == res) { res = t4_init_one(pdev, ent); if (-EAGAIN == res) { dev_err(&pdev->dev, "Failed to update firmware.\n"); res = -EIO; } } return res; } static void _t4_remove_one(struct t4 *wc) { int basesize; if (!wc) return; dahdi_unregister_device(wc->ddev); remove_sysfs_files(wc); /* Stop hardware */ t4_hardware_stop(wc); #ifdef VPM_SUPPORT /* Release vpm */ if (wc->vpm) release_vpm450m(wc->vpm); wc->vpm = NULL; #endif /* Unregister spans */ basesize = DAHDI_MAX_CHUNKSIZE * 32 * 4; if (!(wc->tspans[0]->spanflags & FLAG_2NDGEN)) basesize = basesize * 2; #ifdef ENABLE_WORKQUEUES if (wc->workq) { flush_workqueue(wc->workq); destroy_workqueue(wc->workq); } #endif free_irq(wc->dev->irq, wc); if (wc->membase) pci_iounmap(wc->dev, wc->membase); pci_release_regions(wc->dev); /* Immediately free resources */ pci_free_consistent(wc->dev, T4_BASE_SIZE(wc) * wc->numbufs * 2, wc->writechunk, wc->writedma); order_index[wc->order]--; cards[wc->num] = NULL; pci_set_drvdata(wc->dev, NULL); free_wc(wc); } static void __devexit t4_remove_one(struct pci_dev *pdev) { struct t4 *wc = pci_get_drvdata(pdev); if (!wc) return; _t4_remove_one(wc); } static DEFINE_PCI_DEVICE_TABLE(t4_pci_tbl) = { { 0x10ee, 0x0314, PCI_ANY_ID, PCI_ANY_ID, 0, 0, (unsigned long)&wct4xxp }, { 0xd161, 0x1820, PCI_ANY_ID, PCI_ANY_ID, 0, 0, (unsigned long)&wct820p5 }, { 0xd161, 0x1420, 0x0005, PCI_ANY_ID, 0, 0, (unsigned long)&wct420p5 }, { 0xd161, 0x1410, 0x0005, PCI_ANY_ID, 0, 0, (unsigned long)&wct410p5 }, { 0xd161, 0x1405, 0x0005, PCI_ANY_ID, 0, 0, (unsigned long)&wct405p5 }, { 0xd161, 0x0420, 0x0004, PCI_ANY_ID, 0, 0, (unsigned long)&wct420p4 }, { 0xd161, 0x0410, 0x0004, PCI_ANY_ID, 0, 0, (unsigned long)&wct410p4 }, { 0xd161, 0x0405, 0x0004, PCI_ANY_ID, 0, 0, (unsigned long)&wct405p4 }, { 0xd161, 0x0410, 0x0003, PCI_ANY_ID, 0, 0, (unsigned long)&wct410p3 }, { 0xd161, 0x0405, 0x0003, PCI_ANY_ID, 0, 0, (unsigned long)&wct405p3 }, { 0xd161, 0x0410, PCI_ANY_ID, PCI_ANY_ID, 0, 0, (unsigned long)&wct410p2 }, { 0xd161, 0x0405, PCI_ANY_ID, PCI_ANY_ID, 0, 0, (unsigned long)&wct405p2 }, { 0xd161, 0x1220, 0x0005, PCI_ANY_ID, 0, 0, (unsigned long)&wct220p5 }, { 0xd161, 0x1205, 0x0005, PCI_ANY_ID, 0, 0, (unsigned long)&wct205p5 }, { 0xd161, 0x1210, 0x0005, PCI_ANY_ID, 0, 0, (unsigned long)&wct210p5 }, { 0xd161, 0x0220, 0x0004, PCI_ANY_ID, 0, 0, (unsigned long)&wct220p4 }, { 0xd161, 0x0205, 0x0004, PCI_ANY_ID, 0, 0, (unsigned long)&wct205p4 }, { 0xd161, 0x0210, 0x0004, PCI_ANY_ID, 0, 0, (unsigned long)&wct210p4 }, { 0xd161, 0x0205, 0x0003, PCI_ANY_ID, 0, 0, (unsigned long)&wct205p3 }, { 0xd161, 0x0210, 0x0003, PCI_ANY_ID, 0, 0, (unsigned long)&wct210p3 }, { 0xd161, 0x0205, PCI_ANY_ID, PCI_ANY_ID, 0, 0, (unsigned long)&wct205 }, { 0xd161, 0x0210, PCI_ANY_ID, PCI_ANY_ID, 0, 0, (unsigned long)&wct210 }, { 0, } }; static void _t4_shutdown(struct pci_dev *pdev) { struct t4 *wc = pci_get_drvdata(pdev); t4_hardware_stop(wc); } static int t4_suspend(struct pci_dev *pdev, pm_message_t state) { return -ENOSYS; } static struct pci_driver t4_driver = { .name = "wct4xxp", .probe = t4_init_one_retry, .remove = __devexit_p(t4_remove_one), .shutdown = _t4_shutdown, .suspend = t4_suspend, .id_table = t4_pci_tbl, }; static int __init t4_init(void) { int i; int res; if (-1 != t1e1override) { pr_info("'t1e1override' module parameter is deprecated. " "Please use 'default_linemode' instead.\n"); } res = dahdi_pci_module(&t4_driver); if (res) return -ENODEV; /* If we're ignoring the rotary switch settings, then we've already * registered in the context of .probe */ if (!ignore_rotary) { /* Initialize cards since we have all of them. Warn for * missing zero and duplicate numbers. */ if (cards[0] && cards[0]->order != 0) { printk(KERN_NOTICE "wct4xxp: Ident of first card is not zero (%d)\n", cards[0]->order); } for (i = 0; cards[i]; i++) { /* warn the user of duplicate ident values it is * probably unintended */ if (debug && res < 15 && cards[i+1] && cards[res]->order == cards[i+1]->order) { printk(KERN_NOTICE "wct4xxp: Duplicate ident " "value found (%d)\n", cards[i]->order); } res = t4_launch(cards[i]); if (res) { int j; for (j = 0; j < i; ++j) _t4_remove_one(cards[j]); break; } } } return res; } static void __exit t4_cleanup(void) { pci_unregister_driver(&t4_driver); } MODULE_AUTHOR("Digium Incorporated "); MODULE_DESCRIPTION("Wildcard Dual/Quad/Octal-port Digital Card Driver"); MODULE_ALIAS("wct2xxp"); MODULE_LICENSE("GPL v2"); module_param(debug, int, 0600); module_param(noburst, int, 0600); module_param(timingcable, int, 0600); module_param(t1e1override, int, 0400); module_param(default_linemode, charp, S_IRUGO); MODULE_PARM_DESC(default_linemode, "\"auto\"(default), \"e1\", \"t1\", " "or \"j1\". \"auto\" will use the value from any hardware " "jumpers."); module_param(alarmdebounce, int, 0600); module_param(losalarmdebounce, int, 0600); module_param(aisalarmdebounce, int, 0600); module_param(yelalarmdebounce, int, 0600); module_param(max_latency, int, 0600); module_param(j1mode, int, 0600); module_param(sigmode, int, 0600); module_param(latency, int, 0600); module_param(ms_per_irq, int, 0600); module_param(ignore_rotary, int, 0400); MODULE_PARM_DESC(ignore_rotary, "Set to > 0 to ignore the rotary switch when " \ "registering with DAHDI."); #ifdef VPM_SUPPORT module_param(vpmsupport, int, 0600); module_param(vpmdtmfsupport, int, 0600); #endif MODULE_DEVICE_TABLE(pci, t4_pci_tbl); module_init(t4_init); module_exit(t4_cleanup);