Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / infiniband / hw / ipath / ipath_kernel.h
diff --git a/kernel/drivers/infiniband/hw/ipath/ipath_kernel.h b/kernel/drivers/infiniband/hw/ipath/ipath_kernel.h
new file mode 100644 (file)
index 0000000..e08db70
--- /dev/null
@@ -0,0 +1,1375 @@
+#ifndef _IPATH_KERNEL_H
+#define _IPATH_KERNEL_H
+/*
+ * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This header file is the base header file for infinipath kernel code
+ * ipath_user.h serves a similar purpose for user code.
+ */
+
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/dma-mapping.h>
+#include <linux/mutex.h>
+#include <linux/list.h>
+#include <linux/scatterlist.h>
+#include <asm/io.h>
+#include <rdma/ib_verbs.h>
+
+#include "ipath_common.h"
+#include "ipath_debug.h"
+#include "ipath_registers.h"
+
+/* only s/w major version of InfiniPath we can handle */
+#define IPATH_CHIP_VERS_MAJ 2U
+
+/* don't care about this except printing */
+#define IPATH_CHIP_VERS_MIN 0U
+
+/* temporary, maybe always */
+extern struct infinipath_stats ipath_stats;
+
+#define IPATH_CHIP_SWVERSION IPATH_CHIP_VERS_MAJ
+/*
+ * First-cut critierion for "device is active" is
+ * two thousand dwords combined Tx, Rx traffic per
+ * 5-second interval. SMA packets are 64 dwords,
+ * and occur "a few per second", presumably each way.
+ */
+#define IPATH_TRAFFIC_ACTIVE_THRESHOLD (2000)
+/*
+ * Struct used to indicate which errors are logged in each of the
+ * error-counters that are logged to EEPROM. A counter is incremented
+ * _once_ (saturating at 255) for each event with any bits set in
+ * the error or hwerror register masks below.
+ */
+#define IPATH_EEP_LOG_CNT (4)
+struct ipath_eep_log_mask {
+       u64 errs_to_log;
+       u64 hwerrs_to_log;
+};
+
+struct ipath_portdata {
+       void **port_rcvegrbuf;
+       dma_addr_t *port_rcvegrbuf_phys;
+       /* rcvhdrq base, needs mmap before useful */
+       void *port_rcvhdrq;
+       /* kernel virtual address where hdrqtail is updated */
+       void *port_rcvhdrtail_kvaddr;
+       /*
+        * temp buffer for expected send setup, allocated at open, instead
+        * of each setup call
+        */
+       void *port_tid_pg_list;
+       /* when waiting for rcv or pioavail */
+       wait_queue_head_t port_wait;
+       /*
+        * rcvegr bufs base, physical, must fit
+        * in 44 bits so 32 bit programs mmap64 44 bit works)
+        */
+       dma_addr_t port_rcvegr_phys;
+       /* mmap of hdrq, must fit in 44 bits */
+       dma_addr_t port_rcvhdrq_phys;
+       dma_addr_t port_rcvhdrqtailaddr_phys;
+       /*
+        * number of opens (including slave subports) on this instance
+        * (ignoring forks, dup, etc. for now)
+        */
+       int port_cnt;
+       /*
+        * how much space to leave at start of eager TID entries for
+        * protocol use, on each TID
+        */
+       /* instead of calculating it */
+       unsigned port_port;
+       /* non-zero if port is being shared. */
+       u16 port_subport_cnt;
+       /* non-zero if port is being shared. */
+       u16 port_subport_id;
+       /* number of pio bufs for this port (all procs, if shared) */
+       u32 port_piocnt;
+       /* first pio buffer for this port */
+       u32 port_pio_base;
+       /* chip offset of PIO buffers for this port */
+       u32 port_piobufs;
+       /* how many alloc_pages() chunks in port_rcvegrbuf_pages */
+       u32 port_rcvegrbuf_chunks;
+       /* how many egrbufs per chunk */
+       u32 port_rcvegrbufs_perchunk;
+       /* order for port_rcvegrbuf_pages */
+       size_t port_rcvegrbuf_size;
+       /* rcvhdrq size (for freeing) */
+       size_t port_rcvhdrq_size;
+       /* next expected TID to check when looking for free */
+       u32 port_tidcursor;
+       /* next expected TID to check */
+       unsigned long port_flag;
+       /* what happened */
+       unsigned long int_flag;
+       /* WAIT_RCV that timed out, no interrupt */
+       u32 port_rcvwait_to;
+       /* WAIT_PIO that timed out, no interrupt */
+       u32 port_piowait_to;
+       /* WAIT_RCV already happened, no wait */
+       u32 port_rcvnowait;
+       /* WAIT_PIO already happened, no wait */
+       u32 port_pionowait;
+       /* total number of rcvhdrqfull errors */
+       u32 port_hdrqfull;
+       /*
+        * Used to suppress multiple instances of same
+        * port staying stuck at same point.
+        */
+       u32 port_lastrcvhdrqtail;
+       /* saved total number of rcvhdrqfull errors for poll edge trigger */
+       u32 port_hdrqfull_poll;
+       /* total number of polled urgent packets */
+       u32 port_urgent;
+       /* saved total number of polled urgent packets for poll edge trigger */
+       u32 port_urgent_poll;
+       /* pid of process using this port */
+       struct pid *port_pid;
+       struct pid *port_subpid[INFINIPATH_MAX_SUBPORT];
+       /* same size as task_struct .comm[] */
+       char port_comm[16];
+       /* pkeys set by this use of this port */
+       u16 port_pkeys[4];
+       /* so file ops can get at unit */
+       struct ipath_devdata *port_dd;
+       /* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
+       void *subport_uregbase;
+       /* An array of pages for the eager receive buffers * N */
+       void *subport_rcvegrbuf;
+       /* An array of pages for the eager header queue entries * N */
+       void *subport_rcvhdr_base;
+       /* The version of the library which opened this port */
+       u32 userversion;
+       /* Bitmask of active slaves */
+       u32 active_slaves;
+       /* Type of packets or conditions we want to poll for */
+       u16 poll_type;
+       /* port rcvhdrq head offset */
+       u32 port_head;
+       /* receive packet sequence counter */
+       u32 port_seq_cnt;
+};
+
+struct sk_buff;
+struct ipath_sge_state;
+struct ipath_verbs_txreq;
+
+/*
+ * control information for layered drivers
+ */
+struct _ipath_layer {
+       void *l_arg;
+};
+
+struct ipath_skbinfo {
+       struct sk_buff *skb;
+       dma_addr_t phys;
+};
+
+struct ipath_sdma_txreq {
+       int                 flags;
+       int                 sg_count;
+       union {
+               struct scatterlist *sg;
+               void *map_addr;
+       };
+       void              (*callback)(void *, int);
+       void               *callback_cookie;
+       int                 callback_status;
+       u16                 start_idx;  /* sdma private */
+       u16                 next_descq_idx;  /* sdma private */
+       struct list_head    list;       /* sdma private */
+};
+
+struct ipath_sdma_desc {
+       __le64 qw[2];
+};
+
+#define IPATH_SDMA_TXREQ_F_USELARGEBUF  0x1
+#define IPATH_SDMA_TXREQ_F_HEADTOHOST   0x2
+#define IPATH_SDMA_TXREQ_F_INTREQ       0x4
+#define IPATH_SDMA_TXREQ_F_FREEBUF      0x8
+#define IPATH_SDMA_TXREQ_F_FREEDESC     0x10
+#define IPATH_SDMA_TXREQ_F_VL15         0x20
+
+#define IPATH_SDMA_TXREQ_S_OK        0
+#define IPATH_SDMA_TXREQ_S_SENDERROR 1
+#define IPATH_SDMA_TXREQ_S_ABORTED   2
+#define IPATH_SDMA_TXREQ_S_SHUTDOWN  3
+
+#define IPATH_SDMA_STATUS_SCORE_BOARD_DRAIN_IN_PROG    (1ull << 63)
+#define IPATH_SDMA_STATUS_ABORT_IN_PROG                        (1ull << 62)
+#define IPATH_SDMA_STATUS_INTERNAL_SDMA_ENABLE         (1ull << 61)
+#define IPATH_SDMA_STATUS_SCB_EMPTY                    (1ull << 30)
+
+/* max dwords in small buffer packet */
+#define IPATH_SMALLBUF_DWORDS (dd->ipath_piosize2k >> 2)
+
+/*
+ * Possible IB config parameters for ipath_f_get/set_ib_cfg()
+ */
+#define IPATH_IB_CFG_LIDLMC 0 /* Get/set LID (LS16b) and Mask (MS16b) */
+#define IPATH_IB_CFG_HRTBT 1 /* Get/set Heartbeat off/enable/auto */
+#define IPATH_IB_HRTBT_ON 3 /* Heartbeat enabled, sent every 100msec */
+#define IPATH_IB_HRTBT_OFF 0 /* Heartbeat off */
+#define IPATH_IB_CFG_LWID_ENB 2 /* Get/set allowed Link-width */
+#define IPATH_IB_CFG_LWID 3 /* Get currently active Link-width */
+#define IPATH_IB_CFG_SPD_ENB 4 /* Get/set allowed Link speeds */
+#define IPATH_IB_CFG_SPD 5 /* Get current Link spd */
+#define IPATH_IB_CFG_RXPOL_ENB 6 /* Get/set Auto-RX-polarity enable */
+#define IPATH_IB_CFG_LREV_ENB 7 /* Get/set Auto-Lane-reversal enable */
+#define IPATH_IB_CFG_LINKLATENCY 8 /* Get Auto-Lane-reversal enable */
+
+
+struct ipath_devdata {
+       struct list_head ipath_list;
+
+       struct ipath_kregs const *ipath_kregs;
+       struct ipath_cregs const *ipath_cregs;
+
+       /* mem-mapped pointer to base of chip regs */
+       u64 __iomem *ipath_kregbase;
+       /* end of mem-mapped chip space; range checking */
+       u64 __iomem *ipath_kregend;
+       /* physical address of chip for io_remap, etc. */
+       unsigned long ipath_physaddr;
+       /* base of memory alloced for ipath_kregbase, for free */
+       u64 *ipath_kregalloc;
+       /* ipath_cfgports pointers */
+       struct ipath_portdata **ipath_pd;
+       /* sk_buffs used by port 0 eager receive queue */
+       struct ipath_skbinfo *ipath_port0_skbinfo;
+       /* kvirt address of 1st 2k pio buffer */
+       void __iomem *ipath_pio2kbase;
+       /* kvirt address of 1st 4k pio buffer */
+       void __iomem *ipath_pio4kbase;
+       /*
+        * points to area where PIOavail registers will be DMA'ed.
+        * Has to be on a page of it's own, because the page will be
+        * mapped into user program space.  This copy is *ONLY* ever
+        * written by DMA, not by the driver!  Need a copy per device
+        * when we get to multiple devices
+        */
+       volatile __le64 *ipath_pioavailregs_dma;
+       /* physical address where updates occur */
+       dma_addr_t ipath_pioavailregs_phys;
+       struct _ipath_layer ipath_layer;
+       /* setup intr */
+       int (*ipath_f_intrsetup)(struct ipath_devdata *);
+       /* fallback to alternate interrupt type if possible */
+       int (*ipath_f_intr_fallback)(struct ipath_devdata *);
+       /* setup on-chip bus config */
+       int (*ipath_f_bus)(struct ipath_devdata *, struct pci_dev *);
+       /* hard reset chip */
+       int (*ipath_f_reset)(struct ipath_devdata *);
+       int (*ipath_f_get_boardname)(struct ipath_devdata *, char *,
+                                    size_t);
+       void (*ipath_f_init_hwerrors)(struct ipath_devdata *);
+       void (*ipath_f_handle_hwerrors)(struct ipath_devdata *, char *,
+                                       size_t);
+       void (*ipath_f_quiet_serdes)(struct ipath_devdata *);
+       int (*ipath_f_bringup_serdes)(struct ipath_devdata *);
+       int (*ipath_f_early_init)(struct ipath_devdata *);
+       void (*ipath_f_clear_tids)(struct ipath_devdata *, unsigned);
+       void (*ipath_f_put_tid)(struct ipath_devdata *, u64 __iomem*,
+                               u32, unsigned long);
+       void (*ipath_f_tidtemplate)(struct ipath_devdata *);
+       void (*ipath_f_cleanup)(struct ipath_devdata *);
+       void (*ipath_f_setextled)(struct ipath_devdata *, u64, u64);
+       /* fill out chip-specific fields */
+       int (*ipath_f_get_base_info)(struct ipath_portdata *, void *);
+       /* free irq */
+       void (*ipath_f_free_irq)(struct ipath_devdata *);
+       struct ipath_message_header *(*ipath_f_get_msgheader)
+                                       (struct ipath_devdata *, __le32 *);
+       void (*ipath_f_config_ports)(struct ipath_devdata *, ushort);
+       int (*ipath_f_get_ib_cfg)(struct ipath_devdata *, int);
+       int (*ipath_f_set_ib_cfg)(struct ipath_devdata *, int, u32);
+       void (*ipath_f_config_jint)(struct ipath_devdata *, u16 , u16);
+       void (*ipath_f_read_counters)(struct ipath_devdata *,
+                                       struct infinipath_counters *);
+       void (*ipath_f_xgxs_reset)(struct ipath_devdata *);
+       /* per chip actions needed for IB Link up/down changes */
+       int (*ipath_f_ib_updown)(struct ipath_devdata *, int, u64);
+
+       unsigned ipath_lastegr_idx;
+       struct ipath_ibdev *verbs_dev;
+       struct timer_list verbs_timer;
+       /* total dwords sent (summed from counter) */
+       u64 ipath_sword;
+       /* total dwords rcvd (summed from counter) */
+       u64 ipath_rword;
+       /* total packets sent (summed from counter) */
+       u64 ipath_spkts;
+       /* total packets rcvd (summed from counter) */
+       u64 ipath_rpkts;
+       /* ipath_statusp initially points to this. */
+       u64 _ipath_status;
+       /* GUID for this interface, in network order */
+       __be64 ipath_guid;
+       /*
+        * aggregrate of error bits reported since last cleared, for
+        * limiting of error reporting
+        */
+       ipath_err_t ipath_lasterror;
+       /*
+        * aggregrate of error bits reported since last cleared, for
+        * limiting of hwerror reporting
+        */
+       ipath_err_t ipath_lasthwerror;
+       /* errors masked because they occur too fast */
+       ipath_err_t ipath_maskederrs;
+       u64 ipath_lastlinkrecov; /* link recoveries at last ACTIVE */
+       /* these 5 fields are used to establish deltas for IB Symbol
+        * errors and linkrecovery errors. They can be reported on
+        * some chips during link negotiation prior to INIT, and with
+        * DDR when faking DDR negotiations with non-IBTA switches.
+        * The chip counters are adjusted at driver unload if there is
+        * a non-zero delta.
+        */
+       u64 ibdeltainprog;
+       u64 ibsymdelta;
+       u64 ibsymsnap;
+       u64 iblnkerrdelta;
+       u64 iblnkerrsnap;
+
+       /* time in jiffies at which to re-enable maskederrs */
+       unsigned long ipath_unmasktime;
+       /* count of egrfull errors, combined for all ports */
+       u64 ipath_last_tidfull;
+       /* for ipath_qcheck() */
+       u64 ipath_lastport0rcv_cnt;
+       /* template for writing TIDs  */
+       u64 ipath_tidtemplate;
+       /* value to write to free TIDs */
+       u64 ipath_tidinvalid;
+       /* IBA6120 rcv interrupt setup */
+       u64 ipath_rhdrhead_intr_off;
+
+       /* size of memory at ipath_kregbase */
+       u32 ipath_kregsize;
+       /* number of registers used for pioavail */
+       u32 ipath_pioavregs;
+       /* IPATH_POLL, etc. */
+       u32 ipath_flags;
+       /* ipath_flags driver is waiting for */
+       u32 ipath_state_wanted;
+       /* last buffer for user use, first buf for kernel use is this
+        * index. */
+       u32 ipath_lastport_piobuf;
+       /* is a stats timer active */
+       u32 ipath_stats_timer_active;
+       /* number of interrupts for this device -- saturates... */
+       u32 ipath_int_counter;
+       /* dwords sent read from counter */
+       u32 ipath_lastsword;
+       /* dwords received read from counter */
+       u32 ipath_lastrword;
+       /* sent packets read from counter */
+       u32 ipath_lastspkts;
+       /* received packets read from counter */
+       u32 ipath_lastrpkts;
+       /* pio bufs allocated per port */
+       u32 ipath_pbufsport;
+       /* if remainder on bufs/port, ports < extrabuf get 1 extra */
+       u32 ipath_ports_extrabuf;
+       u32 ipath_pioupd_thresh; /* update threshold, some chips */
+       /*
+        * number of ports configured as max; zero is set to number chip
+        * supports, less gives more pio bufs/port, etc.
+        */
+       u32 ipath_cfgports;
+       /* count of port 0 hdrqfull errors */
+       u32 ipath_p0_hdrqfull;
+       /* port 0 number of receive eager buffers */
+       u32 ipath_p0_rcvegrcnt;
+
+       /*
+        * index of last piobuffer we used.  Speeds up searching, by
+        * starting at this point.  Doesn't matter if multiple cpu's use and
+        * update, last updater is only write that matters.  Whenever it
+        * wraps, we update shadow copies.  Need a copy per device when we
+        * get to multiple devices
+        */
+       u32 ipath_lastpioindex;
+       u32 ipath_lastpioindexl;
+       /* max length of freezemsg */
+       u32 ipath_freezelen;
+       /*
+        * consecutive times we wanted a PIO buffer but were unable to
+        * get one
+        */
+       u32 ipath_consec_nopiobuf;
+       /*
+        * hint that we should update ipath_pioavailshadow before
+        * looking for a PIO buffer
+        */
+       u32 ipath_upd_pio_shadow;
+       /* so we can rewrite it after a chip reset */
+       u32 ipath_pcibar0;
+       /* so we can rewrite it after a chip reset */
+       u32 ipath_pcibar1;
+       u32 ipath_x1_fix_tries;
+       u32 ipath_autoneg_tries;
+       u32 serdes_first_init_done;
+
+       struct ipath_relock {
+               atomic_t ipath_relock_timer_active;
+               struct timer_list ipath_relock_timer;
+               unsigned int ipath_relock_interval; /* in jiffies */
+       } ipath_relock_singleton;
+
+       /* interrupt number */
+       int ipath_irq;
+       /* HT/PCI Vendor ID (here for NodeInfo) */
+       u16 ipath_vendorid;
+       /* HT/PCI Device ID (here for NodeInfo) */
+       u16 ipath_deviceid;
+       /* offset in HT config space of slave/primary interface block */
+       u8 ipath_ht_slave_off;
+       /* for write combining settings */
+       unsigned long ipath_wc_cookie;
+       unsigned long ipath_wc_base;
+       unsigned long ipath_wc_len;
+       /* ref count for each pkey */
+       atomic_t ipath_pkeyrefs[4];
+       /* shadow copy of struct page *'s for exp tid pages */
+       struct page **ipath_pageshadow;
+       /* shadow copy of dma handles for exp tid pages */
+       dma_addr_t *ipath_physshadow;
+       u64 __iomem *ipath_egrtidbase;
+       /* lock to workaround chip bug 9437 and others */
+       spinlock_t ipath_kernel_tid_lock;
+       spinlock_t ipath_user_tid_lock;
+       spinlock_t ipath_sendctrl_lock;
+       /* around ipath_pd and (user ports) port_cnt use (intr vs free) */
+       spinlock_t ipath_uctxt_lock;
+
+       /*
+        * IPATH_STATUS_*,
+        * this address is mapped readonly into user processes so they can
+        * get status cheaply, whenever they want.
+        */
+       u64 *ipath_statusp;
+       /* freeze msg if hw error put chip in freeze */
+       char *ipath_freezemsg;
+       /* pci access data structure */
+       struct pci_dev *pcidev;
+       struct cdev *user_cdev;
+       struct cdev *diag_cdev;
+       struct device *user_dev;
+       struct device *diag_dev;
+       /* timer used to prevent stats overflow, error throttling, etc. */
+       struct timer_list ipath_stats_timer;
+       /* timer to verify interrupts work, and fallback if possible */
+       struct timer_list ipath_intrchk_timer;
+       void *ipath_dummy_hdrq; /* used after port close */
+       dma_addr_t ipath_dummy_hdrq_phys;
+
+       /* SendDMA related entries */
+       spinlock_t            ipath_sdma_lock;
+       unsigned long         ipath_sdma_status;
+       unsigned long         ipath_sdma_abort_jiffies;
+       unsigned long         ipath_sdma_abort_intr_timeout;
+       unsigned long         ipath_sdma_buf_jiffies;
+       struct ipath_sdma_desc *ipath_sdma_descq;
+       u64                   ipath_sdma_descq_added;
+       u64                   ipath_sdma_descq_removed;
+       int                   ipath_sdma_desc_nreserved;
+       u16                   ipath_sdma_descq_cnt;
+       u16                   ipath_sdma_descq_tail;
+       u16                   ipath_sdma_descq_head;
+       u16                   ipath_sdma_next_intr;
+       u16                   ipath_sdma_reset_wait;
+       u8                    ipath_sdma_generation;
+       struct tasklet_struct ipath_sdma_abort_task;
+       struct tasklet_struct ipath_sdma_notify_task;
+       struct list_head      ipath_sdma_activelist;
+       struct list_head      ipath_sdma_notifylist;
+       atomic_t              ipath_sdma_vl15_count;
+       struct timer_list     ipath_sdma_vl15_timer;
+
+       dma_addr_t       ipath_sdma_descq_phys;
+       volatile __le64 *ipath_sdma_head_dma;
+       dma_addr_t       ipath_sdma_head_phys;
+
+       unsigned long ipath_ureg_align; /* user register alignment */
+
+       struct delayed_work ipath_autoneg_work;
+       wait_queue_head_t ipath_autoneg_wait;
+
+       /* HoL blocking / user app forward-progress state */
+       unsigned          ipath_hol_state;
+       unsigned          ipath_hol_next;
+       struct timer_list ipath_hol_timer;
+
+       /*
+        * Shadow copies of registers; size indicates read access size.
+        * Most of them are readonly, but some are write-only register,
+        * where we manipulate the bits in the shadow copy, and then write
+        * the shadow copy to infinipath.
+        *
+        * We deliberately make most of these 32 bits, since they have
+        * restricted range.  For any that we read, we won't to generate 32
+        * bit accesses, since Opteron will generate 2 separate 32 bit HT
+        * transactions for a 64 bit read, and we want to avoid unnecessary
+        * HT transactions.
+        */
+
+       /* This is the 64 bit group */
+
+       /*
+        * shadow of pioavail, check to be sure it's large enough at
+        * init time.
+        */
+       unsigned long ipath_pioavailshadow[8];
+       /* bitmap of send buffers available for the kernel to use with PIO. */
+       unsigned long ipath_pioavailkernel[8];
+       /* shadow of kr_gpio_out, for rmw ops */
+       u64 ipath_gpio_out;
+       /* shadow the gpio mask register */
+       u64 ipath_gpio_mask;
+       /* shadow the gpio output enable, etc... */
+       u64 ipath_extctrl;
+       /* kr_revision shadow */
+       u64 ipath_revision;
+       /*
+        * shadow of ibcctrl, for interrupt handling of link changes,
+        * etc.
+        */
+       u64 ipath_ibcctrl;
+       /*
+        * last ibcstatus, to suppress "duplicate" status change messages,
+        * mostly from 2 to 3
+        */
+       u64 ipath_lastibcstat;
+       /* hwerrmask shadow */
+       ipath_err_t ipath_hwerrmask;
+       ipath_err_t ipath_errormask; /* errormask shadow */
+       /* interrupt config reg shadow */
+       u64 ipath_intconfig;
+       /* kr_sendpiobufbase value */
+       u64 ipath_piobufbase;
+       /* kr_ibcddrctrl shadow */
+       u64 ipath_ibcddrctrl;
+
+       /* these are the "32 bit" regs */
+
+       /*
+        * number of GUIDs in the flash for this interface; may need some
+        * rethinking for setting on other ifaces
+        */
+       u32 ipath_nguid;
+       /*
+        * the following two are 32-bit bitmasks, but {test,clear,set}_bit
+        * all expect bit fields to be "unsigned long"
+        */
+       /* shadow kr_rcvctrl */
+       unsigned long ipath_rcvctrl;
+       /* shadow kr_sendctrl */
+       unsigned long ipath_sendctrl;
+       /* to not count armlaunch after cancel */
+       unsigned long ipath_lastcancel;
+       /* count cases where special trigger was needed (double write) */
+       unsigned long ipath_spectriggerhit;
+
+       /* value we put in kr_rcvhdrcnt */
+       u32 ipath_rcvhdrcnt;
+       /* value we put in kr_rcvhdrsize */
+       u32 ipath_rcvhdrsize;
+       /* value we put in kr_rcvhdrentsize */
+       u32 ipath_rcvhdrentsize;
+       /* offset of last entry in rcvhdrq */
+       u32 ipath_hdrqlast;
+       /* kr_portcnt value */
+       u32 ipath_portcnt;
+       /* kr_pagealign value */
+       u32 ipath_palign;
+       /* number of "2KB" PIO buffers */
+       u32 ipath_piobcnt2k;
+       /* size in bytes of "2KB" PIO buffers */
+       u32 ipath_piosize2k;
+       /* number of "4KB" PIO buffers */
+       u32 ipath_piobcnt4k;
+       /* size in bytes of "4KB" PIO buffers */
+       u32 ipath_piosize4k;
+       u32 ipath_pioreserved; /* reserved special-inkernel; */
+       /* kr_rcvegrbase value */
+       u32 ipath_rcvegrbase;
+       /* kr_rcvegrcnt value */
+       u32 ipath_rcvegrcnt;
+       /* kr_rcvtidbase value */
+       u32 ipath_rcvtidbase;
+       /* kr_rcvtidcnt value */
+       u32 ipath_rcvtidcnt;
+       /* kr_sendregbase */
+       u32 ipath_sregbase;
+       /* kr_userregbase */
+       u32 ipath_uregbase;
+       /* kr_counterregbase */
+       u32 ipath_cregbase;
+       /* shadow the control register contents */
+       u32 ipath_control;
+       /* PCI revision register (HTC rev on FPGA) */
+       u32 ipath_pcirev;
+
+       /* chip address space used by 4k pio buffers */
+       u32 ipath_4kalign;
+       /* The MTU programmed for this unit */
+       u32 ipath_ibmtu;
+       /*
+        * The max size IB packet, included IB headers that we can send.
+        * Starts same as ipath_piosize, but is affected when ibmtu is
+        * changed, or by size of eager buffers
+        */
+       u32 ipath_ibmaxlen;
+       /*
+        * ibmaxlen at init time, limited by chip and by receive buffer
+        * size.  Not changed after init.
+        */
+       u32 ipath_init_ibmaxlen;
+       /* size of each rcvegrbuffer */
+       u32 ipath_rcvegrbufsize;
+       /* localbus width (1, 2,4,8,16,32) from config space  */
+       u32 ipath_lbus_width;
+       /* localbus speed (HT: 200,400,800,1000; PCIe 2500) */
+       u32 ipath_lbus_speed;
+       /*
+        * number of sequential ibcstatus change for polling active/quiet
+        * (i.e., link not coming up).
+        */
+       u32 ipath_ibpollcnt;
+       /* low and high portions of MSI capability/vector */
+       u32 ipath_msi_lo;
+       /* saved after PCIe init for restore after reset */
+       u32 ipath_msi_hi;
+       /* MSI data (vector) saved for restore */
+       u16 ipath_msi_data;
+       /* MLID programmed for this instance */
+       u16 ipath_mlid;
+       /* LID programmed for this instance */
+       u16 ipath_lid;
+       /* list of pkeys programmed; 0 if not set */
+       u16 ipath_pkeys[4];
+       /*
+        * ASCII serial number, from flash, large enough for original
+        * all digit strings, and longer QLogic serial number format
+        */
+       u8 ipath_serial[16];
+       /* human readable board version */
+       u8 ipath_boardversion[96];
+       u8 ipath_lbus_info[32]; /* human readable localbus info */
+       /* chip major rev, from ipath_revision */
+       u8 ipath_majrev;
+       /* chip minor rev, from ipath_revision */
+       u8 ipath_minrev;
+       /* board rev, from ipath_revision */
+       u8 ipath_boardrev;
+       /* saved for restore after reset */
+       u8 ipath_pci_cacheline;
+       /* LID mask control */
+       u8 ipath_lmc;
+       /* link width supported */
+       u8 ipath_link_width_supported;
+       /* link speed supported */
+       u8 ipath_link_speed_supported;
+       u8 ipath_link_width_enabled;
+       u8 ipath_link_speed_enabled;
+       u8 ipath_link_width_active;
+       u8 ipath_link_speed_active;
+       /* Rx Polarity inversion (compensate for ~tx on partner) */
+       u8 ipath_rx_pol_inv;
+
+       u8 ipath_r_portenable_shift;
+       u8 ipath_r_intravail_shift;
+       u8 ipath_r_tailupd_shift;
+       u8 ipath_r_portcfg_shift;
+
+       /* unit # of this chip, if present */
+       int ipath_unit;
+
+       /* local link integrity counter */
+       u32 ipath_lli_counter;
+       /* local link integrity errors */
+       u32 ipath_lli_errors;
+       /*
+        * Above counts only cases where _successive_ LocalLinkIntegrity
+        * errors were seen in the receive headers of kern-packets.
+        * Below are the three (monotonically increasing) counters
+        * maintained via GPIO interrupts on iba6120-rev2.
+        */
+       u32 ipath_rxfc_unsupvl_errs;
+       u32 ipath_overrun_thresh_errs;
+       u32 ipath_lli_errs;
+
+       /*
+        * Not all devices managed by a driver instance are the same
+        * type, so these fields must be per-device.
+        */
+       u64 ipath_i_bitsextant;
+       ipath_err_t ipath_e_bitsextant;
+       ipath_err_t ipath_hwe_bitsextant;
+
+       /*
+        * Below should be computable from number of ports,
+        * since they are never modified.
+        */
+       u64 ipath_i_rcvavail_mask;
+       u64 ipath_i_rcvurg_mask;
+       u16 ipath_i_rcvurg_shift;
+       u16 ipath_i_rcvavail_shift;
+
+       /*
+        * Register bits for selecting i2c direction and values, used for
+        * I2C serial flash.
+        */
+       u8 ipath_gpio_sda_num;
+       u8 ipath_gpio_scl_num;
+       u8 ipath_i2c_chain_type;
+       u64 ipath_gpio_sda;
+       u64 ipath_gpio_scl;
+
+       /* lock for doing RMW of shadows/regs for ExtCtrl and GPIO */
+       spinlock_t ipath_gpio_lock;
+
+       /*
+        * IB link and linktraining states and masks that vary per chip in
+        * some way.  Set at init, to avoid each IB status change interrupt
+        */
+       u8 ibcs_ls_shift;
+       u8 ibcs_lts_mask;
+       u32 ibcs_mask;
+       u32 ib_init;
+       u32 ib_arm;
+       u32 ib_active;
+
+       u16 ipath_rhf_offset; /* offset of RHF within receive header entry */
+
+       /*
+        * shift/mask for linkcmd, linkinitcmd, maxpktlen in ibccontol
+        * reg. Changes for IBA7220
+        */
+       u8 ibcc_lic_mask; /* LinkInitCmd */
+       u8 ibcc_lc_shift; /* LinkCmd */
+       u8 ibcc_mpl_shift; /* Maxpktlen */
+
+       u8 delay_mult;
+
+       /* used to override LED behavior */
+       u8 ipath_led_override;  /* Substituted for normal value, if non-zero */
+       u16 ipath_led_override_timeoff; /* delta to next timer event */
+       u8 ipath_led_override_vals[2]; /* Alternates per blink-frame */
+       u8 ipath_led_override_phase; /* Just counts, LSB picks from vals[] */
+       atomic_t ipath_led_override_timer_active;
+       /* Used to flash LEDs in override mode */
+       struct timer_list ipath_led_override_timer;
+
+       /* Support (including locks) for EEPROM logging of errors and time */
+       /* control access to actual counters, timer */
+       spinlock_t ipath_eep_st_lock;
+       /* control high-level access to EEPROM */
+       struct mutex ipath_eep_lock;
+       /* Below inc'd by ipath_snap_cntrs(), locked by ipath_eep_st_lock */
+       uint64_t ipath_traffic_wds;
+       /* active time is kept in seconds, but logged in hours */
+       atomic_t ipath_active_time;
+       /* Below are nominal shadow of EEPROM, new since last EEPROM update */
+       uint8_t ipath_eep_st_errs[IPATH_EEP_LOG_CNT];
+       uint8_t ipath_eep_st_new_errs[IPATH_EEP_LOG_CNT];
+       uint16_t ipath_eep_hrs;
+       /*
+        * masks for which bits of errs, hwerrs that cause
+        * each of the counters to increment.
+        */
+       struct ipath_eep_log_mask ipath_eep_st_masks[IPATH_EEP_LOG_CNT];
+
+       /* interrupt mitigation reload register info */
+       u16 ipath_jint_idle_ticks;      /* idle clock ticks */
+       u16 ipath_jint_max_packets;     /* max packets across all ports */
+
+       /*
+        * lock for access to SerDes, and flags to sequence preset
+        * versus steady-state. 7220-only at the moment.
+        */
+       spinlock_t ipath_sdepb_lock;
+       u8 ipath_presets_needed; /* Set if presets to be restored next DOWN */
+};
+
+/* ipath_hol_state values (stopping/starting user proc, send flushing) */
+#define IPATH_HOL_UP       0
+#define IPATH_HOL_DOWN     1
+/* ipath_hol_next toggle values, used when hol_state IPATH_HOL_DOWN */
+#define IPATH_HOL_DOWNSTOP 0
+#define IPATH_HOL_DOWNCONT 1
+
+/* bit positions for sdma_status */
+#define IPATH_SDMA_ABORTING  0
+#define IPATH_SDMA_DISARMED  1
+#define IPATH_SDMA_DISABLED  2
+#define IPATH_SDMA_LAYERBUF  3
+#define IPATH_SDMA_RUNNING  30
+#define IPATH_SDMA_SHUTDOWN 31
+
+/* bit combinations that correspond to abort states */
+#define IPATH_SDMA_ABORT_NONE 0
+#define IPATH_SDMA_ABORT_ABORTING (1UL << IPATH_SDMA_ABORTING)
+#define IPATH_SDMA_ABORT_DISARMED ((1UL << IPATH_SDMA_ABORTING) | \
+       (1UL << IPATH_SDMA_DISARMED))
+#define IPATH_SDMA_ABORT_DISABLED ((1UL << IPATH_SDMA_ABORTING) | \
+       (1UL << IPATH_SDMA_DISABLED))
+#define IPATH_SDMA_ABORT_ABORTED ((1UL << IPATH_SDMA_ABORTING) | \
+       (1UL << IPATH_SDMA_DISARMED) | (1UL << IPATH_SDMA_DISABLED))
+#define IPATH_SDMA_ABORT_MASK ((1UL<<IPATH_SDMA_ABORTING) | \
+       (1UL << IPATH_SDMA_DISARMED) | (1UL << IPATH_SDMA_DISABLED))
+
+#define IPATH_SDMA_BUF_NONE 0
+#define IPATH_SDMA_BUF_MASK (1UL<<IPATH_SDMA_LAYERBUF)
+
+/* Private data for file operations */
+struct ipath_filedata {
+       struct ipath_portdata *pd;
+       unsigned subport;
+       unsigned tidcursor;
+       struct ipath_user_sdma_queue *pq;
+};
+extern struct list_head ipath_dev_list;
+extern spinlock_t ipath_devs_lock;
+extern struct ipath_devdata *ipath_lookup(int unit);
+
+int ipath_init_chip(struct ipath_devdata *, int);
+int ipath_enable_wc(struct ipath_devdata *dd);
+void ipath_disable_wc(struct ipath_devdata *dd);
+int ipath_count_units(int *npresentp, int *nupp, int *maxportsp);
+void ipath_shutdown_device(struct ipath_devdata *);
+void ipath_clear_freeze(struct ipath_devdata *);
+
+struct file_operations;
+int ipath_cdev_init(int minor, char *name, const struct file_operations *fops,
+                   struct cdev **cdevp, struct device **devp);
+void ipath_cdev_cleanup(struct cdev **cdevp,
+                       struct device **devp);
+
+int ipath_diag_add(struct ipath_devdata *);
+void ipath_diag_remove(struct ipath_devdata *);
+
+extern wait_queue_head_t ipath_state_wait;
+
+int ipath_user_add(struct ipath_devdata *dd);
+void ipath_user_remove(struct ipath_devdata *dd);
+
+struct sk_buff *ipath_alloc_skb(struct ipath_devdata *dd, gfp_t);
+
+extern int ipath_diag_inuse;
+
+irqreturn_t ipath_intr(int irq, void *devid);
+int ipath_decode_err(struct ipath_devdata *dd, char *buf, size_t blen,
+                    ipath_err_t err);
+#if __IPATH_INFO || __IPATH_DBG
+extern const char *ipath_ibcstatus_str[];
+#endif
+
+/* clean up any per-chip chip-specific stuff */
+void ipath_chip_cleanup(struct ipath_devdata *);
+/* clean up any chip type-specific stuff */
+void ipath_chip_done(void);
+
+void ipath_disarm_piobufs(struct ipath_devdata *, unsigned first,
+                         unsigned cnt);
+void ipath_cancel_sends(struct ipath_devdata *, int);
+
+int ipath_create_rcvhdrq(struct ipath_devdata *, struct ipath_portdata *);
+void ipath_free_pddata(struct ipath_devdata *, struct ipath_portdata *);
+
+int ipath_parse_ushort(const char *str, unsigned short *valp);
+
+void ipath_kreceive(struct ipath_portdata *);
+int ipath_setrcvhdrsize(struct ipath_devdata *, unsigned);
+int ipath_reset_device(int);
+void ipath_get_faststats(unsigned long);
+int ipath_wait_linkstate(struct ipath_devdata *, u32, int);
+int ipath_set_linkstate(struct ipath_devdata *, u8);
+int ipath_set_mtu(struct ipath_devdata *, u16);
+int ipath_set_lid(struct ipath_devdata *, u32, u8);
+int ipath_set_rx_pol_inv(struct ipath_devdata *dd, u8 new_pol_inv);
+void ipath_enable_armlaunch(struct ipath_devdata *);
+void ipath_disable_armlaunch(struct ipath_devdata *);
+void ipath_hol_down(struct ipath_devdata *);
+void ipath_hol_up(struct ipath_devdata *);
+void ipath_hol_event(unsigned long);
+void ipath_toggle_rclkrls(struct ipath_devdata *);
+void ipath_sd7220_clr_ibpar(struct ipath_devdata *);
+void ipath_set_relock_poll(struct ipath_devdata *, int);
+void ipath_shutdown_relock_poll(struct ipath_devdata *);
+
+/* for use in system calls, where we want to know device type, etc. */
+#define port_fp(fp) ((struct ipath_filedata *)(fp)->private_data)->pd
+#define subport_fp(fp) \
+       ((struct ipath_filedata *)(fp)->private_data)->subport
+#define tidcursor_fp(fp) \
+       ((struct ipath_filedata *)(fp)->private_data)->tidcursor
+#define user_sdma_queue_fp(fp) \
+       ((struct ipath_filedata *)(fp)->private_data)->pq
+
+/*
+ * values for ipath_flags
+ */
+               /* chip can report link latency (IB 1.2) */
+#define IPATH_HAS_LINK_LATENCY 0x1
+               /* The chip is up and initted */
+#define IPATH_INITTED       0x2
+               /* set if any user code has set kr_rcvhdrsize */
+#define IPATH_RCVHDRSZ_SET  0x4
+               /* The chip is present and valid for accesses */
+#define IPATH_PRESENT       0x8
+               /* HT link0 is only 8 bits wide, ignore upper byte crc
+                * errors, etc. */
+#define IPATH_8BIT_IN_HT0   0x10
+               /* HT link1 is only 8 bits wide, ignore upper byte crc
+                * errors, etc. */
+#define IPATH_8BIT_IN_HT1   0x20
+               /* The link is down */
+#define IPATH_LINKDOWN      0x40
+               /* The link level is up (0x11) */
+#define IPATH_LINKINIT      0x80
+               /* The link is in the armed (0x21) state */
+#define IPATH_LINKARMED     0x100
+               /* The link is in the active (0x31) state */
+#define IPATH_LINKACTIVE    0x200
+               /* link current state is unknown */
+#define IPATH_LINKUNK       0x400
+               /* Write combining flush needed for PIO */
+#define IPATH_PIO_FLUSH_WC  0x1000
+               /* DMA Receive tail pointer */
+#define IPATH_NODMA_RTAIL   0x2000
+               /* no IB cable, or no device on IB cable */
+#define IPATH_NOCABLE       0x4000
+               /* Supports port zero per packet receive interrupts via
+                * GPIO */
+#define IPATH_GPIO_INTR     0x8000
+               /* uses the coded 4byte TID, not 8 byte */
+#define IPATH_4BYTE_TID     0x10000
+               /* packet/word counters are 32 bit, else those 4 counters
+                * are 64bit */
+#define IPATH_32BITCOUNTERS 0x20000
+               /* Interrupt register is 64 bits */
+#define IPATH_INTREG_64     0x40000
+               /* can miss port0 rx interrupts */
+#define IPATH_DISABLED      0x80000 /* administratively disabled */
+               /* Use GPIO interrupts for new counters */
+#define IPATH_GPIO_ERRINTRS 0x100000
+#define IPATH_SWAP_PIOBUFS  0x200000
+               /* Supports Send DMA */
+#define IPATH_HAS_SEND_DMA  0x400000
+               /* Supports Send Count (not just word count) in PBC */
+#define IPATH_HAS_PBC_CNT   0x800000
+               /* Suppress heartbeat, even if turning off loopback */
+#define IPATH_NO_HRTBT      0x1000000
+#define IPATH_HAS_THRESH_UPDATE 0x4000000
+#define IPATH_HAS_MULT_IB_SPEED 0x8000000
+#define IPATH_IB_AUTONEG_INPROG 0x10000000
+#define IPATH_IB_AUTONEG_FAILED 0x20000000
+               /* Linkdown-disable intentionally, Do not attempt to bring up */
+#define IPATH_IB_LINK_DISABLED 0x40000000
+#define IPATH_IB_FORCE_NOTIFY 0x80000000 /* force notify on next ib change */
+
+/* Bits in GPIO for the added interrupts */
+#define IPATH_GPIO_PORT0_BIT 2
+#define IPATH_GPIO_RXUVL_BIT 3
+#define IPATH_GPIO_OVRUN_BIT 4
+#define IPATH_GPIO_LLI_BIT 5
+#define IPATH_GPIO_ERRINTR_MASK 0x38
+
+/* portdata flag bit offsets */
+               /* waiting for a packet to arrive */
+#define IPATH_PORT_WAITING_RCV   2
+               /* master has not finished initializing */
+#define IPATH_PORT_MASTER_UNINIT 4
+               /* waiting for an urgent packet to arrive */
+#define IPATH_PORT_WAITING_URG 5
+
+/* free up any allocated data at closes */
+void ipath_free_data(struct ipath_portdata *dd);
+u32 __iomem *ipath_getpiobuf(struct ipath_devdata *, u32, u32 *);
+void ipath_chg_pioavailkernel(struct ipath_devdata *dd, unsigned start,
+                               unsigned len, int avail);
+void ipath_init_iba6110_funcs(struct ipath_devdata *);
+void ipath_get_eeprom_info(struct ipath_devdata *);
+int ipath_update_eeprom_log(struct ipath_devdata *dd);
+void ipath_inc_eeprom_err(struct ipath_devdata *dd, u32 eidx, u32 incr);
+u64 ipath_snap_cntr(struct ipath_devdata *, ipath_creg);
+void ipath_disarm_senderrbufs(struct ipath_devdata *);
+void ipath_force_pio_avail_update(struct ipath_devdata *);
+void signal_ib_event(struct ipath_devdata *dd, enum ib_event_type ev);
+
+/*
+ * Set LED override, only the two LSBs have "public" meaning, but
+ * any non-zero value substitutes them for the Link and LinkTrain
+ * LED states.
+ */
+#define IPATH_LED_PHYS 1 /* Physical (linktraining) GREEN LED */
+#define IPATH_LED_LOG 2  /* Logical (link) YELLOW LED */
+void ipath_set_led_override(struct ipath_devdata *dd, unsigned int val);
+
+/* send dma routines */
+int setup_sdma(struct ipath_devdata *);
+void teardown_sdma(struct ipath_devdata *);
+void ipath_restart_sdma(struct ipath_devdata *);
+void ipath_sdma_intr(struct ipath_devdata *);
+int ipath_sdma_verbs_send(struct ipath_devdata *, struct ipath_sge_state *,
+                         u32, struct ipath_verbs_txreq *);
+/* ipath_sdma_lock should be locked before calling this. */
+int ipath_sdma_make_progress(struct ipath_devdata *dd);
+
+/* must be called under ipath_sdma_lock */
+static inline u16 ipath_sdma_descq_freecnt(const struct ipath_devdata *dd)
+{
+       return dd->ipath_sdma_descq_cnt -
+               (dd->ipath_sdma_descq_added - dd->ipath_sdma_descq_removed) -
+               1 - dd->ipath_sdma_desc_nreserved;
+}
+
+static inline void ipath_sdma_desc_reserve(struct ipath_devdata *dd, u16 cnt)
+{
+       dd->ipath_sdma_desc_nreserved += cnt;
+}
+
+static inline void ipath_sdma_desc_unreserve(struct ipath_devdata *dd, u16 cnt)
+{
+       dd->ipath_sdma_desc_nreserved -= cnt;
+}
+
+/*
+ * number of words used for protocol header if not set by ipath_userinit();
+ */
+#define IPATH_DFLT_RCVHDRSIZE 9
+
+int ipath_get_user_pages(unsigned long, size_t, struct page **);
+void ipath_release_user_pages(struct page **, size_t);
+void ipath_release_user_pages_on_close(struct page **, size_t);
+int ipath_eeprom_read(struct ipath_devdata *, u8, void *, int);
+int ipath_eeprom_write(struct ipath_devdata *, u8, const void *, int);
+int ipath_tempsense_read(struct ipath_devdata *, u8 regnum);
+int ipath_tempsense_write(struct ipath_devdata *, u8 regnum, u8 data);
+
+/* these are used for the registers that vary with port */
+void ipath_write_kreg_port(const struct ipath_devdata *, ipath_kreg,
+                          unsigned, u64);
+
+/*
+ * We could have a single register get/put routine, that takes a group type,
+ * but this is somewhat clearer and cleaner.  It also gives us some error
+ * checking.  64 bit register reads should always work, but are inefficient
+ * on opteron (the northbridge always generates 2 separate HT 32 bit reads),
+ * so we use kreg32 wherever possible.  User register and counter register
+ * reads are always 32 bit reads, so only one form of those routines.
+ */
+
+/*
+ * At the moment, none of the s-registers are writable, so no
+ * ipath_write_sreg().
+ */
+
+/**
+ * ipath_read_ureg32 - read 32-bit virtualized per-port register
+ * @dd: device
+ * @regno: register number
+ * @port: port number
+ *
+ * Return the contents of a register that is virtualized to be per port.
+ * Returns -1 on errors (not distinguishable from valid contents at
+ * runtime; we may add a separate error variable at some point).
+ */
+static inline u32 ipath_read_ureg32(const struct ipath_devdata *dd,
+                                   ipath_ureg regno, int port)
+{
+       if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT))
+               return 0;
+
+       return readl(regno + (u64 __iomem *)
+                    (dd->ipath_uregbase +
+                     (char __iomem *)dd->ipath_kregbase +
+                     dd->ipath_ureg_align * port));
+}
+
+/**
+ * ipath_write_ureg - write 32-bit virtualized per-port register
+ * @dd: device
+ * @regno: register number
+ * @value: value
+ * @port: port
+ *
+ * Write the contents of a register that is virtualized to be per port.
+ */
+static inline void ipath_write_ureg(const struct ipath_devdata *dd,
+                                   ipath_ureg regno, u64 value, int port)
+{
+       u64 __iomem *ubase = (u64 __iomem *)
+               (dd->ipath_uregbase + (char __iomem *) dd->ipath_kregbase +
+                dd->ipath_ureg_align * port);
+       if (dd->ipath_kregbase)
+               writeq(value, &ubase[regno]);
+}
+
+static inline u32 ipath_read_kreg32(const struct ipath_devdata *dd,
+                                   ipath_kreg regno)
+{
+       if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT))
+               return -1;
+       return readl((u32 __iomem *) & dd->ipath_kregbase[regno]);
+}
+
+static inline u64 ipath_read_kreg64(const struct ipath_devdata *dd,
+                                   ipath_kreg regno)
+{
+       if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT))
+               return -1;
+
+       return readq(&dd->ipath_kregbase[regno]);
+}
+
+static inline void ipath_write_kreg(const struct ipath_devdata *dd,
+                                   ipath_kreg regno, u64 value)
+{
+       if (dd->ipath_kregbase)
+               writeq(value, &dd->ipath_kregbase[regno]);
+}
+
+static inline u64 ipath_read_creg(const struct ipath_devdata *dd,
+                                 ipath_sreg regno)
+{
+       if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT))
+               return 0;
+
+       return readq(regno + (u64 __iomem *)
+                    (dd->ipath_cregbase +
+                     (char __iomem *)dd->ipath_kregbase));
+}
+
+static inline u32 ipath_read_creg32(const struct ipath_devdata *dd,
+                                        ipath_sreg regno)
+{
+       if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT))
+               return 0;
+       return readl(regno + (u64 __iomem *)
+                    (dd->ipath_cregbase +
+                     (char __iomem *)dd->ipath_kregbase));
+}
+
+static inline void ipath_write_creg(const struct ipath_devdata *dd,
+                                   ipath_creg regno, u64 value)
+{
+       if (dd->ipath_kregbase)
+               writeq(value, regno + (u64 __iomem *)
+                      (dd->ipath_cregbase +
+                       (char __iomem *)dd->ipath_kregbase));
+}
+
+static inline void ipath_clear_rcvhdrtail(const struct ipath_portdata *pd)
+{
+       *((u64 *) pd->port_rcvhdrtail_kvaddr) = 0ULL;
+}
+
+static inline u32 ipath_get_rcvhdrtail(const struct ipath_portdata *pd)
+{
+       return (u32) le64_to_cpu(*((volatile __le64 *)
+                               pd->port_rcvhdrtail_kvaddr));
+}
+
+static inline u32 ipath_get_hdrqtail(const struct ipath_portdata *pd)
+{
+       const struct ipath_devdata *dd = pd->port_dd;
+       u32 hdrqtail;
+
+       if (dd->ipath_flags & IPATH_NODMA_RTAIL) {
+               __le32 *rhf_addr;
+               u32 seq;
+
+               rhf_addr = (__le32 *) pd->port_rcvhdrq +
+                       pd->port_head + dd->ipath_rhf_offset;
+               seq = ipath_hdrget_seq(rhf_addr);
+               hdrqtail = pd->port_head;
+               if (seq == pd->port_seq_cnt)
+                       hdrqtail++;
+       } else
+               hdrqtail = ipath_get_rcvhdrtail(pd);
+
+       return hdrqtail;
+}
+
+static inline u64 ipath_read_ireg(const struct ipath_devdata *dd, ipath_kreg r)
+{
+       return (dd->ipath_flags & IPATH_INTREG_64) ?
+               ipath_read_kreg64(dd, r) : ipath_read_kreg32(dd, r);
+}
+
+/*
+ * from contents of IBCStatus (or a saved copy), return linkstate
+ * Report ACTIVE_DEFER as ACTIVE, because we treat them the same
+ * everywhere, anyway (and should be, for almost all purposes).
+ */
+static inline u32 ipath_ib_linkstate(struct ipath_devdata *dd, u64 ibcs)
+{
+       u32 state = (u32)(ibcs >> dd->ibcs_ls_shift) &
+               INFINIPATH_IBCS_LINKSTATE_MASK;
+       if (state == INFINIPATH_IBCS_L_STATE_ACT_DEFER)
+               state = INFINIPATH_IBCS_L_STATE_ACTIVE;
+       return state;
+}
+
+/* from contents of IBCStatus (or a saved copy), return linktrainingstate */
+static inline u32 ipath_ib_linktrstate(struct ipath_devdata *dd, u64 ibcs)
+{
+       return (u32)(ibcs >> INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) &
+               dd->ibcs_lts_mask;
+}
+
+/*
+ * from contents of IBCStatus (or a saved copy), return logical link state
+ * combination of link state and linktraining state (down, active, init,
+ * arm, etc.
+ */
+static inline u32 ipath_ib_state(struct ipath_devdata *dd, u64 ibcs)
+{
+       u32 ibs;
+       ibs = (u32)(ibcs >> INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) &
+               dd->ibcs_lts_mask;
+       ibs |= (u32)(ibcs &
+               (INFINIPATH_IBCS_LINKSTATE_MASK << dd->ibcs_ls_shift));
+       return ibs;
+}
+
+/*
+ * sysfs interface.
+ */
+
+struct device_driver;
+
+extern const char ib_ipath_version[];
+
+extern const struct attribute_group *ipath_driver_attr_groups[];
+
+int ipath_device_create_group(struct device *, struct ipath_devdata *);
+void ipath_device_remove_group(struct device *, struct ipath_devdata *);
+int ipath_expose_reset(struct device *);
+
+int ipath_init_ipathfs(void);
+void ipath_exit_ipathfs(void);
+int ipathfs_add_device(struct ipath_devdata *);
+int ipathfs_remove_device(struct ipath_devdata *);
+
+/*
+ * dma_addr wrappers - all 0's invalid for hw
+ */
+dma_addr_t ipath_map_page(struct pci_dev *, struct page *, unsigned long,
+                         size_t, int);
+dma_addr_t ipath_map_single(struct pci_dev *, void *, size_t, int);
+const char *ipath_get_unit_name(int unit);
+
+/*
+ * Flush write combining store buffers (if present) and perform a write
+ * barrier.
+ */
+#if defined(CONFIG_X86_64)
+#define ipath_flush_wc() asm volatile("sfence" ::: "memory")
+#else
+#define ipath_flush_wc() wmb()
+#endif
+
+extern unsigned ipath_debug; /* debugging bit mask */
+extern unsigned ipath_linkrecovery;
+extern unsigned ipath_mtu4096;
+extern struct mutex ipath_mutex;
+
+#define IPATH_DRV_NAME         "ib_ipath"
+#define IPATH_MAJOR            233
+#define IPATH_USER_MINOR_BASE  0
+#define IPATH_DIAGPKT_MINOR    127
+#define IPATH_DIAG_MINOR_BASE  129
+#define IPATH_NMINORS          255
+
+#define ipath_dev_err(dd,fmt,...) \
+       do { \
+               const struct ipath_devdata *__dd = (dd); \
+               if (__dd->pcidev) \
+                       dev_err(&__dd->pcidev->dev, "%s: " fmt, \
+                               ipath_get_unit_name(__dd->ipath_unit), \
+                               ##__VA_ARGS__); \
+               else \
+                       printk(KERN_ERR IPATH_DRV_NAME ": %s: " fmt, \
+                              ipath_get_unit_name(__dd->ipath_unit), \
+                              ##__VA_ARGS__); \
+       } while (0)
+
+#if _IPATH_DEBUGGING
+
+# define __IPATH_DBG_WHICH(which,fmt,...) \
+       do { \
+               if (unlikely(ipath_debug & (which))) \
+                       printk(KERN_DEBUG IPATH_DRV_NAME ": %s: " fmt, \
+                              __func__,##__VA_ARGS__); \
+       } while(0)
+
+# define ipath_dbg(fmt,...) \
+       __IPATH_DBG_WHICH(__IPATH_DBG,fmt,##__VA_ARGS__)
+# define ipath_cdbg(which,fmt,...) \
+       __IPATH_DBG_WHICH(__IPATH_##which##DBG,fmt,##__VA_ARGS__)
+
+#else /* ! _IPATH_DEBUGGING */
+
+# define ipath_dbg(fmt,...)
+# define ipath_cdbg(which,fmt,...)
+
+#endif /* _IPATH_DEBUGGING */
+
+/*
+ * this is used for formatting hw error messages...
+ */
+struct ipath_hwerror_msgs {
+       u64 mask;
+       const char *msg;
+};
+
+#define INFINIPATH_HWE_MSG(a, b) { .mask = INFINIPATH_HWE_##a, .msg = b }
+
+/* in ipath_intr.c... */
+void ipath_format_hwerrors(u64 hwerrs,
+                          const struct ipath_hwerror_msgs *hwerrmsgs,
+                          size_t nhwerrmsgs,
+                          char *msg, size_t lmsg);
+
+#endif                         /* _IPATH_KERNEL_H */