Merge master.kernel.org:/pub/scm/linux/kernel/git/davej/cpufreq
[linux-2.6] / drivers / infiniband / hw / ipath / ipath_iba6120.c
1 /*
2  * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
3  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 /*
34  * This file contains all of the code that is specific to the
35  * InfiniPath PCIe chip.
36  */
37
38 #include <linux/interrupt.h>
39 #include <linux/pci.h>
40 #include <linux/delay.h>
41
42
43 #include "ipath_kernel.h"
44 #include "ipath_registers.h"
45
46 static void ipath_setup_pe_setextled(struct ipath_devdata *, u64, u64);
47
48 /*
49  * This file contains all the chip-specific register information and
50  * access functions for the QLogic InfiniPath PCI-Express chip.
51  *
52  * This lists the InfiniPath registers, in the actual chip layout.
53  * This structure should never be directly accessed.
54  */
55 struct _infinipath_do_not_use_kernel_regs {
56         unsigned long long Revision;
57         unsigned long long Control;
58         unsigned long long PageAlign;
59         unsigned long long PortCnt;
60         unsigned long long DebugPortSelect;
61         unsigned long long Reserved0;
62         unsigned long long SendRegBase;
63         unsigned long long UserRegBase;
64         unsigned long long CounterRegBase;
65         unsigned long long Scratch;
66         unsigned long long Reserved1;
67         unsigned long long Reserved2;
68         unsigned long long IntBlocked;
69         unsigned long long IntMask;
70         unsigned long long IntStatus;
71         unsigned long long IntClear;
72         unsigned long long ErrorMask;
73         unsigned long long ErrorStatus;
74         unsigned long long ErrorClear;
75         unsigned long long HwErrMask;
76         unsigned long long HwErrStatus;
77         unsigned long long HwErrClear;
78         unsigned long long HwDiagCtrl;
79         unsigned long long MDIO;
80         unsigned long long IBCStatus;
81         unsigned long long IBCCtrl;
82         unsigned long long ExtStatus;
83         unsigned long long ExtCtrl;
84         unsigned long long GPIOOut;
85         unsigned long long GPIOMask;
86         unsigned long long GPIOStatus;
87         unsigned long long GPIOClear;
88         unsigned long long RcvCtrl;
89         unsigned long long RcvBTHQP;
90         unsigned long long RcvHdrSize;
91         unsigned long long RcvHdrCnt;
92         unsigned long long RcvHdrEntSize;
93         unsigned long long RcvTIDBase;
94         unsigned long long RcvTIDCnt;
95         unsigned long long RcvEgrBase;
96         unsigned long long RcvEgrCnt;
97         unsigned long long RcvBufBase;
98         unsigned long long RcvBufSize;
99         unsigned long long RxIntMemBase;
100         unsigned long long RxIntMemSize;
101         unsigned long long RcvPartitionKey;
102         unsigned long long Reserved3;
103         unsigned long long RcvPktLEDCnt;
104         unsigned long long Reserved4[8];
105         unsigned long long SendCtrl;
106         unsigned long long SendPIOBufBase;
107         unsigned long long SendPIOSize;
108         unsigned long long SendPIOBufCnt;
109         unsigned long long SendPIOAvailAddr;
110         unsigned long long TxIntMemBase;
111         unsigned long long TxIntMemSize;
112         unsigned long long Reserved5;
113         unsigned long long PCIeRBufTestReg0;
114         unsigned long long PCIeRBufTestReg1;
115         unsigned long long Reserved51[6];
116         unsigned long long SendBufferError;
117         unsigned long long SendBufferErrorCONT1;
118         unsigned long long Reserved6SBE[6];
119         unsigned long long RcvHdrAddr0;
120         unsigned long long RcvHdrAddr1;
121         unsigned long long RcvHdrAddr2;
122         unsigned long long RcvHdrAddr3;
123         unsigned long long RcvHdrAddr4;
124         unsigned long long Reserved7RHA[11];
125         unsigned long long RcvHdrTailAddr0;
126         unsigned long long RcvHdrTailAddr1;
127         unsigned long long RcvHdrTailAddr2;
128         unsigned long long RcvHdrTailAddr3;
129         unsigned long long RcvHdrTailAddr4;
130         unsigned long long Reserved8RHTA[11];
131         unsigned long long Reserved9SW[8];
132         unsigned long long SerdesConfig0;
133         unsigned long long SerdesConfig1;
134         unsigned long long SerdesStatus;
135         unsigned long long XGXSConfig;
136         unsigned long long IBPLLCfg;
137         unsigned long long Reserved10SW2[3];
138         unsigned long long PCIEQ0SerdesConfig0;
139         unsigned long long PCIEQ0SerdesConfig1;
140         unsigned long long PCIEQ0SerdesStatus;
141         unsigned long long Reserved11;
142         unsigned long long PCIEQ1SerdesConfig0;
143         unsigned long long PCIEQ1SerdesConfig1;
144         unsigned long long PCIEQ1SerdesStatus;
145         unsigned long long Reserved12;
146 };
147
148 #define IPATH_KREG_OFFSET(field) (offsetof(struct \
149     _infinipath_do_not_use_kernel_regs, field) / sizeof(u64))
150 #define IPATH_CREG_OFFSET(field) (offsetof( \
151     struct infinipath_counters, field) / sizeof(u64))
152
153 static const struct ipath_kregs ipath_pe_kregs = {
154         .kr_control = IPATH_KREG_OFFSET(Control),
155         .kr_counterregbase = IPATH_KREG_OFFSET(CounterRegBase),
156         .kr_debugportselect = IPATH_KREG_OFFSET(DebugPortSelect),
157         .kr_errorclear = IPATH_KREG_OFFSET(ErrorClear),
158         .kr_errormask = IPATH_KREG_OFFSET(ErrorMask),
159         .kr_errorstatus = IPATH_KREG_OFFSET(ErrorStatus),
160         .kr_extctrl = IPATH_KREG_OFFSET(ExtCtrl),
161         .kr_extstatus = IPATH_KREG_OFFSET(ExtStatus),
162         .kr_gpio_clear = IPATH_KREG_OFFSET(GPIOClear),
163         .kr_gpio_mask = IPATH_KREG_OFFSET(GPIOMask),
164         .kr_gpio_out = IPATH_KREG_OFFSET(GPIOOut),
165         .kr_gpio_status = IPATH_KREG_OFFSET(GPIOStatus),
166         .kr_hwdiagctrl = IPATH_KREG_OFFSET(HwDiagCtrl),
167         .kr_hwerrclear = IPATH_KREG_OFFSET(HwErrClear),
168         .kr_hwerrmask = IPATH_KREG_OFFSET(HwErrMask),
169         .kr_hwerrstatus = IPATH_KREG_OFFSET(HwErrStatus),
170         .kr_ibcctrl = IPATH_KREG_OFFSET(IBCCtrl),
171         .kr_ibcstatus = IPATH_KREG_OFFSET(IBCStatus),
172         .kr_intblocked = IPATH_KREG_OFFSET(IntBlocked),
173         .kr_intclear = IPATH_KREG_OFFSET(IntClear),
174         .kr_intmask = IPATH_KREG_OFFSET(IntMask),
175         .kr_intstatus = IPATH_KREG_OFFSET(IntStatus),
176         .kr_mdio = IPATH_KREG_OFFSET(MDIO),
177         .kr_pagealign = IPATH_KREG_OFFSET(PageAlign),
178         .kr_partitionkey = IPATH_KREG_OFFSET(RcvPartitionKey),
179         .kr_portcnt = IPATH_KREG_OFFSET(PortCnt),
180         .kr_rcvbthqp = IPATH_KREG_OFFSET(RcvBTHQP),
181         .kr_rcvbufbase = IPATH_KREG_OFFSET(RcvBufBase),
182         .kr_rcvbufsize = IPATH_KREG_OFFSET(RcvBufSize),
183         .kr_rcvctrl = IPATH_KREG_OFFSET(RcvCtrl),
184         .kr_rcvegrbase = IPATH_KREG_OFFSET(RcvEgrBase),
185         .kr_rcvegrcnt = IPATH_KREG_OFFSET(RcvEgrCnt),
186         .kr_rcvhdrcnt = IPATH_KREG_OFFSET(RcvHdrCnt),
187         .kr_rcvhdrentsize = IPATH_KREG_OFFSET(RcvHdrEntSize),
188         .kr_rcvhdrsize = IPATH_KREG_OFFSET(RcvHdrSize),
189         .kr_rcvintmembase = IPATH_KREG_OFFSET(RxIntMemBase),
190         .kr_rcvintmemsize = IPATH_KREG_OFFSET(RxIntMemSize),
191         .kr_rcvtidbase = IPATH_KREG_OFFSET(RcvTIDBase),
192         .kr_rcvtidcnt = IPATH_KREG_OFFSET(RcvTIDCnt),
193         .kr_revision = IPATH_KREG_OFFSET(Revision),
194         .kr_scratch = IPATH_KREG_OFFSET(Scratch),
195         .kr_sendbuffererror = IPATH_KREG_OFFSET(SendBufferError),
196         .kr_sendctrl = IPATH_KREG_OFFSET(SendCtrl),
197         .kr_sendpioavailaddr = IPATH_KREG_OFFSET(SendPIOAvailAddr),
198         .kr_sendpiobufbase = IPATH_KREG_OFFSET(SendPIOBufBase),
199         .kr_sendpiobufcnt = IPATH_KREG_OFFSET(SendPIOBufCnt),
200         .kr_sendpiosize = IPATH_KREG_OFFSET(SendPIOSize),
201         .kr_sendregbase = IPATH_KREG_OFFSET(SendRegBase),
202         .kr_txintmembase = IPATH_KREG_OFFSET(TxIntMemBase),
203         .kr_txintmemsize = IPATH_KREG_OFFSET(TxIntMemSize),
204         .kr_userregbase = IPATH_KREG_OFFSET(UserRegBase),
205         .kr_serdesconfig0 = IPATH_KREG_OFFSET(SerdesConfig0),
206         .kr_serdesconfig1 = IPATH_KREG_OFFSET(SerdesConfig1),
207         .kr_serdesstatus = IPATH_KREG_OFFSET(SerdesStatus),
208         .kr_xgxsconfig = IPATH_KREG_OFFSET(XGXSConfig),
209         .kr_ibpllcfg = IPATH_KREG_OFFSET(IBPLLCfg),
210
211         /*
212          * These should not be used directly via ipath_write_kreg64(),
213          * use them with ipath_write_kreg64_port(),
214          */
215         .kr_rcvhdraddr = IPATH_KREG_OFFSET(RcvHdrAddr0),
216         .kr_rcvhdrtailaddr = IPATH_KREG_OFFSET(RcvHdrTailAddr0),
217
218         /* The rcvpktled register controls one of the debug port signals, so
219          * a packet activity LED can be connected to it. */
220         .kr_rcvpktledcnt = IPATH_KREG_OFFSET(RcvPktLEDCnt),
221         .kr_pcierbuftestreg0 = IPATH_KREG_OFFSET(PCIeRBufTestReg0),
222         .kr_pcierbuftestreg1 = IPATH_KREG_OFFSET(PCIeRBufTestReg1),
223         .kr_pcieq0serdesconfig0 = IPATH_KREG_OFFSET(PCIEQ0SerdesConfig0),
224         .kr_pcieq0serdesconfig1 = IPATH_KREG_OFFSET(PCIEQ0SerdesConfig1),
225         .kr_pcieq0serdesstatus = IPATH_KREG_OFFSET(PCIEQ0SerdesStatus),
226         .kr_pcieq1serdesconfig0 = IPATH_KREG_OFFSET(PCIEQ1SerdesConfig0),
227         .kr_pcieq1serdesconfig1 = IPATH_KREG_OFFSET(PCIEQ1SerdesConfig1),
228         .kr_pcieq1serdesstatus = IPATH_KREG_OFFSET(PCIEQ1SerdesStatus)
229 };
230
231 static const struct ipath_cregs ipath_pe_cregs = {
232         .cr_badformatcnt = IPATH_CREG_OFFSET(RxBadFormatCnt),
233         .cr_erricrccnt = IPATH_CREG_OFFSET(RxICRCErrCnt),
234         .cr_errlinkcnt = IPATH_CREG_OFFSET(RxLinkProblemCnt),
235         .cr_errlpcrccnt = IPATH_CREG_OFFSET(RxLPCRCErrCnt),
236         .cr_errpkey = IPATH_CREG_OFFSET(RxPKeyMismatchCnt),
237         .cr_errrcvflowctrlcnt = IPATH_CREG_OFFSET(RxFlowCtrlErrCnt),
238         .cr_err_rlencnt = IPATH_CREG_OFFSET(RxLenErrCnt),
239         .cr_errslencnt = IPATH_CREG_OFFSET(TxLenErrCnt),
240         .cr_errtidfull = IPATH_CREG_OFFSET(RxTIDFullErrCnt),
241         .cr_errtidvalid = IPATH_CREG_OFFSET(RxTIDValidErrCnt),
242         .cr_errvcrccnt = IPATH_CREG_OFFSET(RxVCRCErrCnt),
243         .cr_ibstatuschange = IPATH_CREG_OFFSET(IBStatusChangeCnt),
244         .cr_intcnt = IPATH_CREG_OFFSET(LBIntCnt),
245         .cr_invalidrlencnt = IPATH_CREG_OFFSET(RxMaxMinLenErrCnt),
246         .cr_invalidslencnt = IPATH_CREG_OFFSET(TxMaxMinLenErrCnt),
247         .cr_lbflowstallcnt = IPATH_CREG_OFFSET(LBFlowStallCnt),
248         .cr_pktrcvcnt = IPATH_CREG_OFFSET(RxDataPktCnt),
249         .cr_pktrcvflowctrlcnt = IPATH_CREG_OFFSET(RxFlowPktCnt),
250         .cr_pktsendcnt = IPATH_CREG_OFFSET(TxDataPktCnt),
251         .cr_pktsendflowcnt = IPATH_CREG_OFFSET(TxFlowPktCnt),
252         .cr_portovflcnt = IPATH_CREG_OFFSET(RxP0HdrEgrOvflCnt),
253         .cr_rcvebpcnt = IPATH_CREG_OFFSET(RxEBPCnt),
254         .cr_rcvovflcnt = IPATH_CREG_OFFSET(RxBufOvflCnt),
255         .cr_senddropped = IPATH_CREG_OFFSET(TxDroppedPktCnt),
256         .cr_sendstallcnt = IPATH_CREG_OFFSET(TxFlowStallCnt),
257         .cr_sendunderruncnt = IPATH_CREG_OFFSET(TxUnderrunCnt),
258         .cr_wordrcvcnt = IPATH_CREG_OFFSET(RxDwordCnt),
259         .cr_wordsendcnt = IPATH_CREG_OFFSET(TxDwordCnt),
260         .cr_unsupvlcnt = IPATH_CREG_OFFSET(TxUnsupVLErrCnt),
261         .cr_rxdroppktcnt = IPATH_CREG_OFFSET(RxDroppedPktCnt),
262         .cr_iblinkerrrecovcnt = IPATH_CREG_OFFSET(IBLinkErrRecoveryCnt),
263         .cr_iblinkdowncnt = IPATH_CREG_OFFSET(IBLinkDownedCnt),
264         .cr_ibsymbolerrcnt = IPATH_CREG_OFFSET(IBSymbolErrCnt)
265 };
266
267 /* kr_intstatus, kr_intclear, kr_intmask bits */
268 #define INFINIPATH_I_RCVURG_MASK ((1U<<5)-1)
269 #define INFINIPATH_I_RCVAVAIL_MASK ((1U<<5)-1)
270
271 /* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */
272 #define INFINIPATH_HWE_PCIEMEMPARITYERR_MASK  0x000000000000003fULL
273 #define INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT 0
274 #define INFINIPATH_HWE_PCIEPOISONEDTLP      0x0000000010000000ULL
275 #define INFINIPATH_HWE_PCIECPLTIMEOUT       0x0000000020000000ULL
276 #define INFINIPATH_HWE_PCIEBUSPARITYXTLH    0x0000000040000000ULL
277 #define INFINIPATH_HWE_PCIEBUSPARITYXADM    0x0000000080000000ULL
278 #define INFINIPATH_HWE_PCIEBUSPARITYRADM    0x0000000100000000ULL
279 #define INFINIPATH_HWE_COREPLL_FBSLIP       0x0080000000000000ULL
280 #define INFINIPATH_HWE_COREPLL_RFSLIP       0x0100000000000000ULL
281 #define INFINIPATH_HWE_PCIE1PLLFAILED       0x0400000000000000ULL
282 #define INFINIPATH_HWE_PCIE0PLLFAILED       0x0800000000000000ULL
283 #define INFINIPATH_HWE_SERDESPLLFAILED      0x1000000000000000ULL
284
285 /* kr_extstatus bits */
286 #define INFINIPATH_EXTS_FREQSEL 0x2
287 #define INFINIPATH_EXTS_SERDESSEL 0x4
288 #define INFINIPATH_EXTS_MEMBIST_ENDTEST     0x0000000000004000
289 #define INFINIPATH_EXTS_MEMBIST_FOUND       0x0000000000008000
290
291 #define _IPATH_GPIO_SDA_NUM 1
292 #define _IPATH_GPIO_SCL_NUM 0
293
294 #define IPATH_GPIO_SDA (1ULL << \
295         (_IPATH_GPIO_SDA_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT))
296 #define IPATH_GPIO_SCL (1ULL << \
297         (_IPATH_GPIO_SCL_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT))
298
299 /* 6120 specific hardware errors... */
300 static const struct ipath_hwerror_msgs ipath_6120_hwerror_msgs[] = {
301         INFINIPATH_HWE_MSG(PCIEPOISONEDTLP, "PCIe Poisoned TLP"),
302         INFINIPATH_HWE_MSG(PCIECPLTIMEOUT, "PCIe completion timeout"),
303         /*
304          * In practice, it's unlikely wthat we'll see PCIe PLL, or bus
305          * parity or memory parity error failures, because most likely we
306          * won't be able to talk to the core of the chip.  Nonetheless, we
307          * might see them, if they are in parts of the PCIe core that aren't
308          * essential.
309          */
310         INFINIPATH_HWE_MSG(PCIE1PLLFAILED, "PCIePLL1"),
311         INFINIPATH_HWE_MSG(PCIE0PLLFAILED, "PCIePLL0"),
312         INFINIPATH_HWE_MSG(PCIEBUSPARITYXTLH, "PCIe XTLH core parity"),
313         INFINIPATH_HWE_MSG(PCIEBUSPARITYXADM, "PCIe ADM TX core parity"),
314         INFINIPATH_HWE_MSG(PCIEBUSPARITYRADM, "PCIe ADM RX core parity"),
315         INFINIPATH_HWE_MSG(RXDSYNCMEMPARITYERR, "Rx Dsync"),
316         INFINIPATH_HWE_MSG(SERDESPLLFAILED, "SerDes PLL"),
317 };
318
319 #define TXE_PIO_PARITY ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF | \
320                         INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC) \
321                         << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT)
322
323 static int ipath_pe_txe_recover(struct ipath_devdata *);
324 static void ipath_pe_put_tid_2(struct ipath_devdata *, u64 __iomem *,
325                                u32, unsigned long);
326
327 /**
328  * ipath_pe_handle_hwerrors - display hardware errors.
329  * @dd: the infinipath device
330  * @msg: the output buffer
331  * @msgl: the size of the output buffer
332  *
333  * Use same msg buffer as regular errors to avoid excessive stack
334  * use.  Most hardware errors are catastrophic, but for right now,
335  * we'll print them and continue.  We reuse the same message buffer as
336  * ipath_handle_errors() to avoid excessive stack usage.
337  */
338 static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
339                                      size_t msgl)
340 {
341         ipath_err_t hwerrs;
342         u32 bits, ctrl;
343         int isfatal = 0;
344         char bitsmsg[64];
345         int log_idx;
346
347         hwerrs = ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus);
348         if (!hwerrs) {
349                 /*
350                  * better than printing cofusing messages
351                  * This seems to be related to clearing the crc error, or
352                  * the pll error during init.
353                  */
354                 ipath_cdbg(VERBOSE, "Called but no hardware errors set\n");
355                 return;
356         } else if (hwerrs == ~0ULL) {
357                 ipath_dev_err(dd, "Read of hardware error status failed "
358                               "(all bits set); ignoring\n");
359                 return;
360         }
361         ipath_stats.sps_hwerrs++;
362
363         /* Always clear the error status register, except MEMBISTFAIL,
364          * regardless of whether we continue or stop using the chip.
365          * We want that set so we know it failed, even across driver reload.
366          * We'll still ignore it in the hwerrmask.  We do this partly for
367          * diagnostics, but also for support */
368         ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear,
369                          hwerrs&~INFINIPATH_HWE_MEMBISTFAILED);
370
371         hwerrs &= dd->ipath_hwerrmask;
372
373         /* We log some errors to EEPROM, check if we have any of those. */
374         for (log_idx = 0; log_idx < IPATH_EEP_LOG_CNT; ++log_idx)
375                 if (hwerrs & dd->ipath_eep_st_masks[log_idx].hwerrs_to_log)
376                         ipath_inc_eeprom_err(dd, log_idx, 1);
377
378         /*
379          * make sure we get this much out, unless told to be quiet,
380          * or it's occurred within the last 5 seconds
381          */
382         if ((hwerrs & ~(dd->ipath_lasthwerror |
383                         ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
384                           INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
385                          << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT))) ||
386             (ipath_debug & __IPATH_VERBDBG))
387                 dev_info(&dd->pcidev->dev, "Hardware error: hwerr=0x%llx "
388                          "(cleared)\n", (unsigned long long) hwerrs);
389         dd->ipath_lasthwerror |= hwerrs;
390
391         if (hwerrs & ~dd->ipath_hwe_bitsextant)
392                 ipath_dev_err(dd, "hwerror interrupt with unknown errors "
393                               "%llx set\n", (unsigned long long)
394                               (hwerrs & ~dd->ipath_hwe_bitsextant));
395
396         ctrl = ipath_read_kreg32(dd, dd->ipath_kregs->kr_control);
397         if (ctrl & INFINIPATH_C_FREEZEMODE) {
398                 /*
399                  * parity errors in send memory are recoverable,
400                  * just cancel the send (if indicated in * sendbuffererror),
401                  * count the occurrence, unfreeze (if no other handled
402                  * hardware error bits are set), and continue. They can
403                  * occur if a processor speculative read is done to the PIO
404                  * buffer while we are sending a packet, for example.
405                  */
406                 if ((hwerrs & TXE_PIO_PARITY) && ipath_pe_txe_recover(dd))
407                         hwerrs &= ~TXE_PIO_PARITY;
408                 if (hwerrs) {
409                         /*
410                          * if any set that we aren't ignoring only make the
411                          * complaint once, in case it's stuck or recurring,
412                          * and we get here multiple times
413                          * Force link down, so switch knows, and
414                          * LEDs are turned off
415                          */
416                         if (dd->ipath_flags & IPATH_INITTED) {
417                                 ipath_set_linkstate(dd, IPATH_IB_LINKDOWN);
418                                 ipath_setup_pe_setextled(dd,
419                                         INFINIPATH_IBCS_L_STATE_DOWN,
420                                         INFINIPATH_IBCS_LT_STATE_DISABLED);
421                                 ipath_dev_err(dd, "Fatal Hardware Error (freeze "
422                                               "mode), no longer usable, SN %.16s\n",
423                                                   dd->ipath_serial);
424                                 isfatal = 1;
425                         }
426                         /*
427                          * Mark as having had an error for driver, and also
428                          * for /sys and status word mapped to user programs.
429                          * This marks unit as not usable, until reset
430                          */
431                         *dd->ipath_statusp &= ~IPATH_STATUS_IB_READY;
432                         *dd->ipath_statusp |= IPATH_STATUS_HWERROR;
433                         dd->ipath_flags &= ~IPATH_INITTED;
434                 } else {
435                         static u32 freeze_cnt;
436
437                         freeze_cnt++;
438                         ipath_dbg("Clearing freezemode on ignored or recovered "
439                                   "hardware error (%u)\n", freeze_cnt);
440                         ipath_clear_freeze(dd);
441                 }
442         }
443
444         *msg = '\0';
445
446         if (hwerrs & INFINIPATH_HWE_MEMBISTFAILED) {
447                 strlcat(msg, "[Memory BIST test failed, InfiniPath hardware unusable]",
448                         msgl);
449                 /* ignore from now on, so disable until driver reloaded */
450                 *dd->ipath_statusp |= IPATH_STATUS_HWERROR;
451                 dd->ipath_hwerrmask &= ~INFINIPATH_HWE_MEMBISTFAILED;
452                 ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
453                                  dd->ipath_hwerrmask);
454         }
455
456         ipath_format_hwerrors(hwerrs,
457                               ipath_6120_hwerror_msgs,
458                               sizeof(ipath_6120_hwerror_msgs)/
459                               sizeof(ipath_6120_hwerror_msgs[0]),
460                               msg, msgl);
461
462         if (hwerrs & (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK
463                       << INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT)) {
464                 bits = (u32) ((hwerrs >>
465                                INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) &
466                               INFINIPATH_HWE_PCIEMEMPARITYERR_MASK);
467                 snprintf(bitsmsg, sizeof bitsmsg,
468                          "[PCIe Mem Parity Errs %x] ", bits);
469                 strlcat(msg, bitsmsg, msgl);
470         }
471
472 #define _IPATH_PLL_FAIL (INFINIPATH_HWE_COREPLL_FBSLIP |        \
473                          INFINIPATH_HWE_COREPLL_RFSLIP )
474
475         if (hwerrs & _IPATH_PLL_FAIL) {
476                 snprintf(bitsmsg, sizeof bitsmsg,
477                          "[PLL failed (%llx), InfiniPath hardware unusable]",
478                          (unsigned long long) hwerrs & _IPATH_PLL_FAIL);
479                 strlcat(msg, bitsmsg, msgl);
480                 /* ignore from now on, so disable until driver reloaded */
481                 dd->ipath_hwerrmask &= ~(hwerrs & _IPATH_PLL_FAIL);
482                 ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
483                                  dd->ipath_hwerrmask);
484         }
485
486         if (hwerrs & INFINIPATH_HWE_SERDESPLLFAILED) {
487                 /*
488                  * If it occurs, it is left masked since the eternal
489                  * interface is unused
490                  */
491                 dd->ipath_hwerrmask &= ~INFINIPATH_HWE_SERDESPLLFAILED;
492                 ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
493                                  dd->ipath_hwerrmask);
494         }
495
496         if (*msg)
497                 ipath_dev_err(dd, "%s hardware error\n", msg);
498         if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg) {
499                 /*
500                  * for /sys status file ; if no trailing } is copied, we'll
501                  * know it was truncated.
502                  */
503                 snprintf(dd->ipath_freezemsg, dd->ipath_freezelen,
504                          "{%s}", msg);
505         }
506 }
507
508 /**
509  * ipath_pe_boardname - fill in the board name
510  * @dd: the infinipath device
511  * @name: the output buffer
512  * @namelen: the size of the output buffer
513  *
514  * info is based on the board revision register
515  */
516 static int ipath_pe_boardname(struct ipath_devdata *dd, char *name,
517                               size_t namelen)
518 {
519         char *n = NULL;
520         u8 boardrev = dd->ipath_boardrev;
521         int ret;
522
523         switch (boardrev) {
524         case 0:
525                 n = "InfiniPath_Emulation";
526                 break;
527         case 1:
528                 n = "InfiniPath_QLE7140-Bringup";
529                 break;
530         case 2:
531                 n = "InfiniPath_QLE7140";
532                 break;
533         case 3:
534                 n = "InfiniPath_QMI7140";
535                 break;
536         case 4:
537                 n = "InfiniPath_QEM7140";
538                 break;
539         case 5:
540                 n = "InfiniPath_QMH7140";
541                 break;
542         case 6:
543                 n = "InfiniPath_QLE7142";
544                 break;
545         default:
546                 ipath_dev_err(dd,
547                               "Don't yet know about board with ID %u\n",
548                               boardrev);
549                 snprintf(name, namelen, "Unknown_InfiniPath_PCIe_%u",
550                          boardrev);
551                 break;
552         }
553         if (n)
554                 snprintf(name, namelen, "%s", n);
555
556         if (dd->ipath_majrev != 4 || !dd->ipath_minrev || dd->ipath_minrev>2) {
557                 ipath_dev_err(dd, "Unsupported InfiniPath hardware revision %u.%u!\n",
558                               dd->ipath_majrev, dd->ipath_minrev);
559                 ret = 1;
560         } else {
561                 ret = 0;
562                 if (dd->ipath_minrev >= 2)
563                         dd->ipath_f_put_tid = ipath_pe_put_tid_2;
564         }
565
566         return ret;
567 }
568
569 /**
570  * ipath_pe_init_hwerrors - enable hardware errors
571  * @dd: the infinipath device
572  *
573  * now that we have finished initializing everything that might reasonably
574  * cause a hardware error, and cleared those errors bits as they occur,
575  * we can enable hardware errors in the mask (potentially enabling
576  * freeze mode), and enable hardware errors as errors (along with
577  * everything else) in errormask
578  */
579 static void ipath_pe_init_hwerrors(struct ipath_devdata *dd)
580 {
581         ipath_err_t val;
582         u64 extsval;
583
584         extsval = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extstatus);
585
586         if (!(extsval & INFINIPATH_EXTS_MEMBIST_ENDTEST))
587                 ipath_dev_err(dd, "MemBIST did not complete!\n");
588         if (extsval & INFINIPATH_EXTS_MEMBIST_FOUND)
589                 ipath_dbg("MemBIST corrected\n");
590
591         val = ~0ULL;    /* barring bugs, all hwerrors become interrupts, */
592
593         if (!dd->ipath_boardrev)        // no PLL for Emulator
594                 val &= ~INFINIPATH_HWE_SERDESPLLFAILED;
595
596         if (dd->ipath_minrev < 2) {
597                 /* workaround bug 9460 in internal interface bus parity
598                  * checking. Fixed (HW bug 9490) in Rev2.
599                  */
600                 val &= ~INFINIPATH_HWE_PCIEBUSPARITYRADM;
601         }
602         dd->ipath_hwerrmask = val;
603 }
604
605 /**
606  * ipath_pe_bringup_serdes - bring up the serdes
607  * @dd: the infinipath device
608  */
609 static int ipath_pe_bringup_serdes(struct ipath_devdata *dd)
610 {
611         u64 val, config1, prev_val;
612         int ret = 0;
613
614         ipath_dbg("Trying to bringup serdes\n");
615
616         if (ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus) &
617             INFINIPATH_HWE_SERDESPLLFAILED) {
618                 ipath_dbg("At start, serdes PLL failed bit set "
619                           "in hwerrstatus, clearing and continuing\n");
620                 ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear,
621                                  INFINIPATH_HWE_SERDESPLLFAILED);
622         }
623
624         val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0);
625         config1 = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig1);
626
627         ipath_cdbg(VERBOSE, "SerDes status config0=%llx config1=%llx, "
628                    "xgxsconfig %llx\n", (unsigned long long) val,
629                    (unsigned long long) config1, (unsigned long long)
630                    ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig));
631
632         /*
633          * Force reset on, also set rxdetect enable.  Must do before reading
634          * serdesstatus at least for simulation, or some of the bits in
635          * serdes status will come back as undefined and cause simulation
636          * failures
637          */
638         val |= INFINIPATH_SERDC0_RESET_PLL | INFINIPATH_SERDC0_RXDETECT_EN
639                 | INFINIPATH_SERDC0_L1PWR_DN;
640         ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
641         /* be sure chip saw it */
642         ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
643         udelay(5);              /* need pll reset set at least for a bit */
644         /*
645          * after PLL is reset, set the per-lane Resets and TxIdle and
646          * clear the PLL reset and rxdetect (to get falling edge).
647          * Leave L1PWR bits set (permanently)
648          */
649         val &= ~(INFINIPATH_SERDC0_RXDETECT_EN | INFINIPATH_SERDC0_RESET_PLL
650                  | INFINIPATH_SERDC0_L1PWR_DN);
651         val |= INFINIPATH_SERDC0_RESET_MASK | INFINIPATH_SERDC0_TXIDLE;
652         ipath_cdbg(VERBOSE, "Clearing pll reset and setting lane resets "
653                    "and txidle (%llx)\n", (unsigned long long) val);
654         ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
655         /* be sure chip saw it */
656         ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
657         /* need PLL reset clear for at least 11 usec before lane
658          * resets cleared; give it a few more to be sure */
659         udelay(15);
660         val &= ~(INFINIPATH_SERDC0_RESET_MASK | INFINIPATH_SERDC0_TXIDLE);
661
662         ipath_cdbg(VERBOSE, "Clearing lane resets and txidle "
663                    "(writing %llx)\n", (unsigned long long) val);
664         ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
665         /* be sure chip saw it */
666         val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
667
668         val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig);
669         prev_val = val;
670         if (((val >> INFINIPATH_XGXS_MDIOADDR_SHIFT) &
671              INFINIPATH_XGXS_MDIOADDR_MASK) != 3) {
672                 val &=
673                         ~(INFINIPATH_XGXS_MDIOADDR_MASK <<
674                           INFINIPATH_XGXS_MDIOADDR_SHIFT);
675                 /* MDIO address 3 */
676                 val |= 3ULL << INFINIPATH_XGXS_MDIOADDR_SHIFT;
677         }
678         if (val & INFINIPATH_XGXS_RESET) {
679                 val &= ~INFINIPATH_XGXS_RESET;
680         }
681         if (((val >> INFINIPATH_XGXS_RX_POL_SHIFT) &
682              INFINIPATH_XGXS_RX_POL_MASK) != dd->ipath_rx_pol_inv ) {
683                 /* need to compensate for Tx inversion in partner */
684                 val &= ~(INFINIPATH_XGXS_RX_POL_MASK <<
685                          INFINIPATH_XGXS_RX_POL_SHIFT);
686                 val |= dd->ipath_rx_pol_inv <<
687                         INFINIPATH_XGXS_RX_POL_SHIFT;
688         }
689         if (val != prev_val)
690                 ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val);
691
692         val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0);
693
694         /* clear current and de-emphasis bits */
695         config1 &= ~0x0ffffffff00ULL;
696         /* set current to 20ma */
697         config1 |= 0x00000000000ULL;
698         /* set de-emphasis to -5.68dB */
699         config1 |= 0x0cccc000000ULL;
700         ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig1, config1);
701
702         ipath_cdbg(VERBOSE, "done: SerDes status config0=%llx "
703                    "config1=%llx, sstatus=%llx xgxs=%llx\n",
704                    (unsigned long long) val, (unsigned long long) config1,
705                    (unsigned long long)
706                    ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesstatus),
707                    (unsigned long long)
708                    ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig));
709
710         if (!ipath_waitfor_mdio_cmdready(dd)) {
711                 ipath_write_kreg(
712                         dd, dd->ipath_kregs->kr_mdio,
713                         ipath_mdio_req(IPATH_MDIO_CMD_READ, 31,
714                                        IPATH_MDIO_CTRL_XGXS_REG_8, 0));
715                 if (ipath_waitfor_complete(dd, dd->ipath_kregs->kr_mdio,
716                                            IPATH_MDIO_DATAVALID, &val))
717                         ipath_dbg("Never got MDIO data for XGXS "
718                                   "status read\n");
719                 else
720                         ipath_cdbg(VERBOSE, "MDIO Read reg8, "
721                                    "'bank' 31 %x\n", (u32) val);
722         } else
723                 ipath_dbg("Never got MDIO cmdready for XGXS status read\n");
724
725         return ret;
726 }
727
728 /**
729  * ipath_pe_quiet_serdes - set serdes to txidle
730  * @dd: the infinipath device
731  * Called when driver is being unloaded
732  */
733 static void ipath_pe_quiet_serdes(struct ipath_devdata *dd)
734 {
735         u64 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0);
736
737         val |= INFINIPATH_SERDC0_TXIDLE;
738         ipath_dbg("Setting TxIdleEn on serdes (config0 = %llx)\n",
739                   (unsigned long long) val);
740         ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
741 }
742
743 static int ipath_pe_intconfig(struct ipath_devdata *dd)
744 {
745         u32 chiprev;
746
747         /*
748          * If the chip supports added error indication via GPIO pins,
749          * enable interrupts on those bits so the interrupt routine
750          * can count the events. Also set flag so interrupt routine
751          * can know they are expected.
752          */
753         chiprev = dd->ipath_revision >> INFINIPATH_R_CHIPREVMINOR_SHIFT;
754         if ((chiprev & INFINIPATH_R_CHIPREVMINOR_MASK) > 1) {
755                 /* Rev2+ reports extra errors via internal GPIO pins */
756                 dd->ipath_flags |= IPATH_GPIO_ERRINTRS;
757                 dd->ipath_gpio_mask |= IPATH_GPIO_ERRINTR_MASK;
758                 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask,
759                                  dd->ipath_gpio_mask);
760         }
761         return 0;
762 }
763
764 /**
765  * ipath_setup_pe_setextled - set the state of the two external LEDs
766  * @dd: the infinipath device
767  * @lst: the L state
768  * @ltst: the LT state
769
770  * These LEDs indicate the physical and logical state of IB link.
771  * For this chip (at least with recommended board pinouts), LED1
772  * is Yellow (logical state) and LED2 is Green (physical state),
773  *
774  * Note:  We try to match the Mellanox HCA LED behavior as best
775  * we can.  Green indicates physical link state is OK (something is
776  * plugged in, and we can train).
777  * Amber indicates the link is logically up (ACTIVE).
778  * Mellanox further blinks the amber LED to indicate data packet
779  * activity, but we have no hardware support for that, so it would
780  * require waking up every 10-20 msecs and checking the counters
781  * on the chip, and then turning the LED off if appropriate.  That's
782  * visible overhead, so not something we will do.
783  *
784  */
785 static void ipath_setup_pe_setextled(struct ipath_devdata *dd, u64 lst,
786                                      u64 ltst)
787 {
788         u64 extctl;
789         unsigned long flags = 0;
790
791         /* the diags use the LED to indicate diag info, so we leave
792          * the external LED alone when the diags are running */
793         if (ipath_diag_inuse)
794                 return;
795
796         /* Allow override of LED display for, e.g. Locating system in rack */
797         if (dd->ipath_led_override) {
798                 ltst = (dd->ipath_led_override & IPATH_LED_PHYS)
799                         ? INFINIPATH_IBCS_LT_STATE_LINKUP
800                         : INFINIPATH_IBCS_LT_STATE_DISABLED;
801                 lst = (dd->ipath_led_override & IPATH_LED_LOG)
802                         ? INFINIPATH_IBCS_L_STATE_ACTIVE
803                         : INFINIPATH_IBCS_L_STATE_DOWN;
804         }
805
806         spin_lock_irqsave(&dd->ipath_gpio_lock, flags);
807         extctl = dd->ipath_extctrl & ~(INFINIPATH_EXTC_LED1PRIPORT_ON |
808                                        INFINIPATH_EXTC_LED2PRIPORT_ON);
809
810         if (ltst & INFINIPATH_IBCS_LT_STATE_LINKUP)
811                 extctl |= INFINIPATH_EXTC_LED2PRIPORT_ON;
812         if (lst == INFINIPATH_IBCS_L_STATE_ACTIVE)
813                 extctl |= INFINIPATH_EXTC_LED1PRIPORT_ON;
814         dd->ipath_extctrl = extctl;
815         ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, extctl);
816         spin_unlock_irqrestore(&dd->ipath_gpio_lock, flags);
817 }
818
819 /**
820  * ipath_setup_pe_cleanup - clean up any per-chip chip-specific stuff
821  * @dd: the infinipath device
822  *
823  * This is called during driver unload.
824  * We do the pci_disable_msi here, not in generic code, because it
825  * isn't used for the HT chips. If we do end up needing pci_enable_msi
826  * at some point in the future for HT, we'll move the call back
827  * into the main init_one code.
828  */
829 static void ipath_setup_pe_cleanup(struct ipath_devdata *dd)
830 {
831         dd->ipath_msi_lo = 0;   /* just in case unload fails */
832         pci_disable_msi(dd->pcidev);
833 }
834
835 /**
836  * ipath_setup_pe_config - setup PCIe config related stuff
837  * @dd: the infinipath device
838  * @pdev: the PCI device
839  *
840  * The pci_enable_msi() call will fail on systems with MSI quirks
841  * such as those with AMD8131, even if the device of interest is not
842  * attached to that device, (in the 2.6.13 - 2.6.15 kernels, at least, fixed
843  * late in 2.6.16).
844  * All that can be done is to edit the kernel source to remove the quirk
845  * check until that is fixed.
846  * We do not need to call enable_msi() for our HyperTransport chip,
847  * even though it uses MSI, and we want to avoid the quirk warning, so
848  * So we call enable_msi only for PCIe.  If we do end up needing
849  * pci_enable_msi at some point in the future for HT, we'll move the
850  * call back into the main init_one code.
851  * We save the msi lo and hi values, so we can restore them after
852  * chip reset (the kernel PCI infrastructure doesn't yet handle that
853  * correctly).
854  */
855 static int ipath_setup_pe_config(struct ipath_devdata *dd,
856                                  struct pci_dev *pdev)
857 {
858         int pos, ret;
859
860         dd->ipath_msi_lo = 0;   /* used as a flag during reset processing */
861         ret = pci_enable_msi(dd->pcidev);
862         if (ret)
863                 ipath_dev_err(dd, "pci_enable_msi failed: %d, "
864                               "interrupts may not work\n", ret);
865         /* continue even if it fails, we may still be OK... */
866         dd->ipath_irq = pdev->irq;
867
868         if ((pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI))) {
869                 u16 control;
870                 pci_read_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
871                                       &dd->ipath_msi_lo);
872                 pci_read_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
873                                       &dd->ipath_msi_hi);
874                 pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
875                                      &control);
876                 /* now save the data (vector) info */
877                 pci_read_config_word(dd->pcidev,
878                                      pos + ((control & PCI_MSI_FLAGS_64BIT)
879                                             ? 12 : 8),
880                                      &dd->ipath_msi_data);
881                 ipath_cdbg(VERBOSE, "Read msi data 0x%x from config offset "
882                            "0x%x, control=0x%x\n", dd->ipath_msi_data,
883                            pos + ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
884                            control);
885                 /* we save the cachelinesize also, although it doesn't
886                  * really matter */
887                 pci_read_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE,
888                                      &dd->ipath_pci_cacheline);
889         } else
890                 ipath_dev_err(dd, "Can't find MSI capability, "
891                               "can't save MSI settings for reset\n");
892         if ((pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP))) {
893                 u16 linkstat;
894                 pci_read_config_word(dd->pcidev, pos + PCI_EXP_LNKSTA,
895                                      &linkstat);
896                 linkstat >>= 4;
897                 linkstat &= 0x1f;
898                 if (linkstat != 8)
899                         ipath_dev_err(dd, "PCIe width %u, "
900                                       "performance reduced\n", linkstat);
901         }
902         else
903                 ipath_dev_err(dd, "Can't find PCI Express "
904                               "capability!\n");
905         return 0;
906 }
907
908 static void ipath_init_pe_variables(struct ipath_devdata *dd)
909 {
910         /*
911          * bits for selecting i2c direction and values,
912          * used for I2C serial flash
913          */
914         dd->ipath_gpio_sda_num = _IPATH_GPIO_SDA_NUM;
915         dd->ipath_gpio_scl_num = _IPATH_GPIO_SCL_NUM;
916         dd->ipath_gpio_sda = IPATH_GPIO_SDA;
917         dd->ipath_gpio_scl = IPATH_GPIO_SCL;
918
919         /* variables for sanity checking interrupt and errors */
920         dd->ipath_hwe_bitsextant =
921                 (INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
922                  INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) |
923                 (INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
924                  INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT) |
925                 (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK <<
926                  INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) |
927                 INFINIPATH_HWE_PCIE1PLLFAILED |
928                 INFINIPATH_HWE_PCIE0PLLFAILED |
929                 INFINIPATH_HWE_PCIEPOISONEDTLP |
930                 INFINIPATH_HWE_PCIECPLTIMEOUT |
931                 INFINIPATH_HWE_PCIEBUSPARITYXTLH |
932                 INFINIPATH_HWE_PCIEBUSPARITYXADM |
933                 INFINIPATH_HWE_PCIEBUSPARITYRADM |
934                 INFINIPATH_HWE_MEMBISTFAILED |
935                 INFINIPATH_HWE_COREPLL_FBSLIP |
936                 INFINIPATH_HWE_COREPLL_RFSLIP |
937                 INFINIPATH_HWE_SERDESPLLFAILED |
938                 INFINIPATH_HWE_IBCBUSTOSPCPARITYERR |
939                 INFINIPATH_HWE_IBCBUSFRSPCPARITYERR;
940         dd->ipath_i_bitsextant =
941                 (INFINIPATH_I_RCVURG_MASK << INFINIPATH_I_RCVURG_SHIFT) |
942                 (INFINIPATH_I_RCVAVAIL_MASK <<
943                  INFINIPATH_I_RCVAVAIL_SHIFT) |
944                 INFINIPATH_I_ERROR | INFINIPATH_I_SPIOSENT |
945                 INFINIPATH_I_SPIOBUFAVAIL | INFINIPATH_I_GPIO;
946         dd->ipath_e_bitsextant =
947                 INFINIPATH_E_RFORMATERR | INFINIPATH_E_RVCRC |
948                 INFINIPATH_E_RICRC | INFINIPATH_E_RMINPKTLEN |
949                 INFINIPATH_E_RMAXPKTLEN | INFINIPATH_E_RLONGPKTLEN |
950                 INFINIPATH_E_RSHORTPKTLEN | INFINIPATH_E_RUNEXPCHAR |
951                 INFINIPATH_E_RUNSUPVL | INFINIPATH_E_REBP |
952                 INFINIPATH_E_RIBFLOW | INFINIPATH_E_RBADVERSION |
953                 INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL |
954                 INFINIPATH_E_RBADTID | INFINIPATH_E_RHDRLEN |
955                 INFINIPATH_E_RHDR | INFINIPATH_E_RIBLOSTLINK |
956                 INFINIPATH_E_SMINPKTLEN | INFINIPATH_E_SMAXPKTLEN |
957                 INFINIPATH_E_SUNDERRUN | INFINIPATH_E_SPKTLEN |
958                 INFINIPATH_E_SDROPPEDSMPPKT | INFINIPATH_E_SDROPPEDDATAPKT |
959                 INFINIPATH_E_SPIOARMLAUNCH | INFINIPATH_E_SUNEXPERRPKTNUM |
960                 INFINIPATH_E_SUNSUPVL | INFINIPATH_E_IBSTATUSCHANGED |
961                 INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET |
962                 INFINIPATH_E_HARDWARE;
963
964         dd->ipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK;
965         dd->ipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK;
966
967         /*
968          * EEPROM error log 0 is TXE Parity errors. 1 is RXE Parity.
969          * 2 is Some Misc, 3 is reserved for future.
970          */
971         dd->ipath_eep_st_masks[0].hwerrs_to_log =
972                 INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
973                 INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT;
974
975         /* Ignore errors in PIO/PBC on systems with unordered write-combining */
976         if (ipath_unordered_wc())
977                 dd->ipath_eep_st_masks[0].hwerrs_to_log &= ~TXE_PIO_PARITY;
978
979         dd->ipath_eep_st_masks[1].hwerrs_to_log =
980                 INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
981                 INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT;
982
983         dd->ipath_eep_st_masks[2].errs_to_log =
984                 INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET;
985
986
987 }
988
989 /* setup the MSI stuff again after a reset.  I'd like to just call
990  * pci_enable_msi() and request_irq() again, but when I do that,
991  * the MSI enable bit doesn't get set in the command word, and
992  * we switch to to a different interrupt vector, which is confusing,
993  * so I instead just do it all inline.  Perhaps somehow can tie this
994  * into the PCIe hotplug support at some point
995  * Note, because I'm doing it all here, I don't call pci_disable_msi()
996  * or free_irq() at the start of ipath_setup_pe_reset().
997  */
998 static int ipath_reinit_msi(struct ipath_devdata *dd)
999 {
1000         int pos;
1001         u16 control;
1002         int ret;
1003
1004         if (!dd->ipath_msi_lo) {
1005                 dev_info(&dd->pcidev->dev, "Can't restore MSI config, "
1006                          "initial setup failed?\n");
1007                 ret = 0;
1008                 goto bail;
1009         }
1010
1011         if (!(pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI))) {
1012                 ipath_dev_err(dd, "Can't find MSI capability, "
1013                               "can't restore MSI settings\n");
1014                 ret = 0;
1015                 goto bail;
1016         }
1017         ipath_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n",
1018                    dd->ipath_msi_lo, pos + PCI_MSI_ADDRESS_LO);
1019         pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
1020                                dd->ipath_msi_lo);
1021         ipath_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n",
1022                    dd->ipath_msi_hi, pos + PCI_MSI_ADDRESS_HI);
1023         pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
1024                                dd->ipath_msi_hi);
1025         pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control);
1026         if (!(control & PCI_MSI_FLAGS_ENABLE)) {
1027                 ipath_cdbg(VERBOSE, "MSI control at off %x was %x, "
1028                            "setting MSI enable (%x)\n", pos + PCI_MSI_FLAGS,
1029                            control, control | PCI_MSI_FLAGS_ENABLE);
1030                 control |= PCI_MSI_FLAGS_ENABLE;
1031                 pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
1032                                       control);
1033         }
1034         /* now rewrite the data (vector) info */
1035         pci_write_config_word(dd->pcidev, pos +
1036                               ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
1037                               dd->ipath_msi_data);
1038         /* we restore the cachelinesize also, although it doesn't really
1039          * matter */
1040         pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE,
1041                               dd->ipath_pci_cacheline);
1042         /* and now set the pci master bit again */
1043         pci_set_master(dd->pcidev);
1044         ret = 1;
1045
1046 bail:
1047         return ret;
1048 }
1049
1050 /* This routine sleeps, so it can only be called from user context, not
1051  * from interrupt context.  If we need interrupt context, we can split
1052  * it into two routines.
1053 */
1054 static int ipath_setup_pe_reset(struct ipath_devdata *dd)
1055 {
1056         u64 val;
1057         int i;
1058         int ret;
1059
1060         /* Use ERROR so it shows up in logs, etc. */
1061         ipath_dev_err(dd, "Resetting InfiniPath unit %u\n", dd->ipath_unit);
1062         /* keep chip from being accessed in a few places */
1063         dd->ipath_flags &= ~(IPATH_INITTED|IPATH_PRESENT);
1064         val = dd->ipath_control | INFINIPATH_C_RESET;
1065         ipath_write_kreg(dd, dd->ipath_kregs->kr_control, val);
1066         mb();
1067
1068         for (i = 1; i <= 5; i++) {
1069                 int r;
1070                 /* allow MBIST, etc. to complete; longer on each retry.
1071                  * We sometimes get machine checks from bus timeout if no
1072                  * response, so for now, make it *really* long.
1073                  */
1074                 msleep(1000 + (1 + i) * 2000);
1075                 if ((r =
1076                      pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
1077                                             dd->ipath_pcibar0)))
1078                         ipath_dev_err(dd, "rewrite of BAR0 failed: %d\n",
1079                                       r);
1080                 if ((r =
1081                      pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
1082                                             dd->ipath_pcibar1)))
1083                         ipath_dev_err(dd, "rewrite of BAR1 failed: %d\n",
1084                                       r);
1085                 /* now re-enable memory access */
1086                 if ((r = pci_enable_device(dd->pcidev)))
1087                         ipath_dev_err(dd, "pci_enable_device failed after "
1088                                       "reset: %d\n", r);
1089                 /* whether it worked or not, mark as present, again */
1090                 dd->ipath_flags |= IPATH_PRESENT;
1091                 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_revision);
1092                 if (val == dd->ipath_revision) {
1093                         ipath_cdbg(VERBOSE, "Got matching revision "
1094                                    "register %llx on try %d\n",
1095                                    (unsigned long long) val, i);
1096                         ret = ipath_reinit_msi(dd);
1097                         goto bail;
1098                 }
1099                 /* Probably getting -1 back */
1100                 ipath_dbg("Didn't get expected revision register, "
1101                           "got %llx, try %d\n", (unsigned long long) val,
1102                           i + 1);
1103         }
1104         ret = 0; /* failed */
1105
1106 bail:
1107         return ret;
1108 }
1109
1110 /**
1111  * ipath_pe_put_tid - write a TID in chip
1112  * @dd: the infinipath device
1113  * @tidptr: pointer to the expected TID (in chip) to udpate
1114  * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0) for expected
1115  * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
1116  *
1117  * This exists as a separate routine to allow for special locking etc.
1118  * It's used for both the full cleanup on exit, as well as the normal
1119  * setup and teardown.
1120  */
1121 static void ipath_pe_put_tid(struct ipath_devdata *dd, u64 __iomem *tidptr,
1122                              u32 type, unsigned long pa)
1123 {
1124         u32 __iomem *tidp32 = (u32 __iomem *)tidptr;
1125         unsigned long flags = 0; /* keep gcc quiet */
1126
1127         if (pa != dd->ipath_tidinvalid) {
1128                 if (pa & ((1U << 11) - 1)) {
1129                         dev_info(&dd->pcidev->dev, "BUG: physaddr %lx "
1130                                  "not 4KB aligned!\n", pa);
1131                         return;
1132                 }
1133                 pa >>= 11;
1134                 /* paranoia check */
1135                 if (pa & (7<<29))
1136                         ipath_dev_err(dd,
1137                                       "BUG: Physical page address 0x%lx "
1138                                       "has bits set in 31-29\n", pa);
1139
1140                 if (type == RCVHQ_RCV_TYPE_EAGER)
1141                         pa |= dd->ipath_tidtemplate;
1142                 else /* for now, always full 4KB page */
1143                         pa |= 2 << 29;
1144         }
1145
1146         /*
1147          * Workaround chip bug 9437 by writing the scratch register
1148          * before and after the TID, and with an io write barrier.
1149          * We use a spinlock around the writes, so they can't intermix
1150          * with other TID (eager or expected) writes (the chip bug
1151          * is triggered by back to back TID writes). Unfortunately, this
1152          * call can be done from interrupt level for the port 0 eager TIDs,
1153          * so we have to use irqsave locks.
1154          */
1155         spin_lock_irqsave(&dd->ipath_tid_lock, flags);
1156         ipath_write_kreg(dd, dd->ipath_kregs->kr_scratch, 0xfeeddeaf);
1157         if (dd->ipath_kregbase)
1158                 writel(pa, tidp32);
1159         ipath_write_kreg(dd, dd->ipath_kregs->kr_scratch, 0xdeadbeef);
1160         mmiowb();
1161         spin_unlock_irqrestore(&dd->ipath_tid_lock, flags);
1162 }
1163 /**
1164  * ipath_pe_put_tid_2 - write a TID in chip, Revision 2 or higher
1165  * @dd: the infinipath device
1166  * @tidptr: pointer to the expected TID (in chip) to udpate
1167  * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0) for expected
1168  * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
1169  *
1170  * This exists as a separate routine to allow for selection of the
1171  * appropriate "flavor". The static calls in cleanup just use the
1172  * revision-agnostic form, as they are not performance critical.
1173  */
1174 static void ipath_pe_put_tid_2(struct ipath_devdata *dd, u64 __iomem *tidptr,
1175                              u32 type, unsigned long pa)
1176 {
1177         u32 __iomem *tidp32 = (u32 __iomem *)tidptr;
1178
1179         if (pa != dd->ipath_tidinvalid) {
1180                 if (pa & ((1U << 11) - 1)) {
1181                         dev_info(&dd->pcidev->dev, "BUG: physaddr %lx "
1182                                  "not 2KB aligned!\n", pa);
1183                         return;
1184                 }
1185                 pa >>= 11;
1186                 /* paranoia check */
1187                 if (pa & (7<<29))
1188                         ipath_dev_err(dd,
1189                                       "BUG: Physical page address 0x%lx "
1190                                       "has bits set in 31-29\n", pa);
1191
1192                 if (type == RCVHQ_RCV_TYPE_EAGER)
1193                         pa |= dd->ipath_tidtemplate;
1194                 else /* for now, always full 4KB page */
1195                         pa |= 2 << 29;
1196         }
1197         if (dd->ipath_kregbase)
1198                 writel(pa, tidp32);
1199         mmiowb();
1200 }
1201
1202
1203 /**
1204  * ipath_pe_clear_tid - clear all TID entries for a port, expected and eager
1205  * @dd: the infinipath device
1206  * @port: the port
1207  *
1208  * clear all TID entries for a port, expected and eager.
1209  * Used from ipath_close().  On this chip, TIDs are only 32 bits,
1210  * not 64, but they are still on 64 bit boundaries, so tidbase
1211  * is declared as u64 * for the pointer math, even though we write 32 bits
1212  */
1213 static void ipath_pe_clear_tids(struct ipath_devdata *dd, unsigned port)
1214 {
1215         u64 __iomem *tidbase;
1216         unsigned long tidinv;
1217         int i;
1218
1219         if (!dd->ipath_kregbase)
1220                 return;
1221
1222         ipath_cdbg(VERBOSE, "Invalidate TIDs for port %u\n", port);
1223
1224         tidinv = dd->ipath_tidinvalid;
1225         tidbase = (u64 __iomem *)
1226                 ((char __iomem *)(dd->ipath_kregbase) +
1227                  dd->ipath_rcvtidbase +
1228                  port * dd->ipath_rcvtidcnt * sizeof(*tidbase));
1229
1230         for (i = 0; i < dd->ipath_rcvtidcnt; i++)
1231                 dd->ipath_f_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EXPECTED,
1232                                  tidinv);
1233
1234         tidbase = (u64 __iomem *)
1235                 ((char __iomem *)(dd->ipath_kregbase) +
1236                  dd->ipath_rcvegrbase +
1237                  port * dd->ipath_rcvegrcnt * sizeof(*tidbase));
1238
1239         for (i = 0; i < dd->ipath_rcvegrcnt; i++)
1240                 dd->ipath_f_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EAGER,
1241                                  tidinv);
1242 }
1243
1244 /**
1245  * ipath_pe_tidtemplate - setup constants for TID updates
1246  * @dd: the infinipath device
1247  *
1248  * We setup stuff that we use a lot, to avoid calculating each time
1249  */
1250 static void ipath_pe_tidtemplate(struct ipath_devdata *dd)
1251 {
1252         u32 egrsize = dd->ipath_rcvegrbufsize;
1253
1254         /* For now, we always allocate 4KB buffers (at init) so we can
1255          * receive max size packets.  We may want a module parameter to
1256          * specify 2KB or 4KB and/or make be per port instead of per device
1257          * for those who want to reduce memory footprint.  Note that the
1258          * ipath_rcvhdrentsize size must be large enough to hold the largest
1259          * IB header (currently 96 bytes) that we expect to handle (plus of
1260          * course the 2 dwords of RHF).
1261          */
1262         if (egrsize == 2048)
1263                 dd->ipath_tidtemplate = 1U << 29;
1264         else if (egrsize == 4096)
1265                 dd->ipath_tidtemplate = 2U << 29;
1266         else {
1267                 egrsize = 4096;
1268                 dev_info(&dd->pcidev->dev, "BUG: unsupported egrbufsize "
1269                          "%u, using %u\n", dd->ipath_rcvegrbufsize,
1270                          egrsize);
1271                 dd->ipath_tidtemplate = 2U << 29;
1272         }
1273         dd->ipath_tidinvalid = 0;
1274 }
1275
1276 static int ipath_pe_early_init(struct ipath_devdata *dd)
1277 {
1278         dd->ipath_flags |= IPATH_4BYTE_TID;
1279         if (ipath_unordered_wc())
1280                 dd->ipath_flags |= IPATH_PIO_FLUSH_WC;
1281
1282         /*
1283          * For openfabrics, we need to be able to handle an IB header of
1284          * 24 dwords.  HT chip has arbitrary sized receive buffers, so we
1285          * made them the same size as the PIO buffers.  This chip does not
1286          * handle arbitrary size buffers, so we need the header large enough
1287          * to handle largest IB header, but still have room for a 2KB MTU
1288          * standard IB packet.
1289          */
1290         dd->ipath_rcvhdrentsize = 24;
1291         dd->ipath_rcvhdrsize = IPATH_DFLT_RCVHDRSIZE;
1292
1293         /*
1294          * To truly support a 4KB MTU (for usermode), we need to
1295          * bump this to a larger value.  For now, we use them for
1296          * the kernel only.
1297          */
1298         dd->ipath_rcvegrbufsize = 2048;
1299         /*
1300          * the min() check here is currently a nop, but it may not always
1301          * be, depending on just how we do ipath_rcvegrbufsize
1302          */
1303         dd->ipath_ibmaxlen = min(dd->ipath_piosize2k,
1304                                  dd->ipath_rcvegrbufsize +
1305                                  (dd->ipath_rcvhdrentsize << 2));
1306         dd->ipath_init_ibmaxlen = dd->ipath_ibmaxlen;
1307
1308         /*
1309          * We can request a receive interrupt for 1 or
1310          * more packets from current offset.  For now, we set this
1311          * up for a single packet.
1312          */
1313         dd->ipath_rhdrhead_intr_off = 1ULL<<32;
1314
1315         ipath_get_eeprom_info(dd);
1316
1317         return 0;
1318 }
1319
1320 int __attribute__((weak)) ipath_unordered_wc(void)
1321 {
1322         return 0;
1323 }
1324
1325 /**
1326  * ipath_init_pe_get_base_info - set chip-specific flags for user code
1327  * @pd: the infinipath port
1328  * @kbase: ipath_base_info pointer
1329  *
1330  * We set the PCIE flag because the lower bandwidth on PCIe vs
1331  * HyperTransport can affect some user packet algorithms.
1332  */
1333 static int ipath_pe_get_base_info(struct ipath_portdata *pd, void *kbase)
1334 {
1335         struct ipath_base_info *kinfo = kbase;
1336         struct ipath_devdata *dd;
1337
1338         if (ipath_unordered_wc()) {
1339                 kinfo->spi_runtime_flags |= IPATH_RUNTIME_FORCE_WC_ORDER;
1340                 ipath_cdbg(PROC, "Intel processor, forcing WC order\n");
1341         }
1342         else
1343                 ipath_cdbg(PROC, "Not Intel processor, WC ordered\n");
1344
1345         if (pd == NULL)
1346                 goto done;
1347
1348         dd = pd->port_dd;
1349
1350 done:
1351         kinfo->spi_runtime_flags |= IPATH_RUNTIME_PCIE |
1352                 IPATH_RUNTIME_FORCE_PIOAVAIL | IPATH_RUNTIME_PIO_REGSWAPPED;
1353         return 0;
1354 }
1355
1356 static void ipath_pe_free_irq(struct ipath_devdata *dd)
1357 {
1358         free_irq(dd->ipath_irq, dd);
1359         dd->ipath_irq = 0;
1360 }
1361
1362 /*
1363  * On platforms using this chip, and not having ordered WC stores, we
1364  * can get TXE parity errors due to speculative reads to the PIO buffers,
1365  * and this, due to a chip bug can result in (many) false parity error
1366  * reports.  So it's a debug print on those, and an info print on systems
1367  * where the speculative reads don't occur.
1368  * Because we can get lots of false errors, we have no upper limit
1369  * on recovery attempts on those platforms.
1370  */
1371 static int ipath_pe_txe_recover(struct ipath_devdata *dd)
1372 {
1373         if (ipath_unordered_wc())
1374                 ipath_dbg("Recovering from TXE PIO parity error\n");
1375         else {
1376                 int cnt = ++ipath_stats.sps_txeparity;
1377                 if (cnt >= IPATH_MAX_PARITY_ATTEMPTS)  {
1378                         if (cnt == IPATH_MAX_PARITY_ATTEMPTS)
1379                                 ipath_dev_err(dd,
1380                                         "Too many attempts to recover from "
1381                                         "TXE parity, giving up\n");
1382                         return 0;
1383                 }
1384                 dev_info(&dd->pcidev->dev,
1385                         "Recovering from TXE PIO parity error\n");
1386         }
1387         return 1;
1388 }
1389
1390 /**
1391  * ipath_init_iba6120_funcs - set up the chip-specific function pointers
1392  * @dd: the infinipath device
1393  *
1394  * This is global, and is called directly at init to set up the
1395  * chip-specific function pointers for later use.
1396  */
1397 void ipath_init_iba6120_funcs(struct ipath_devdata *dd)
1398 {
1399         dd->ipath_f_intrsetup = ipath_pe_intconfig;
1400         dd->ipath_f_bus = ipath_setup_pe_config;
1401         dd->ipath_f_reset = ipath_setup_pe_reset;
1402         dd->ipath_f_get_boardname = ipath_pe_boardname;
1403         dd->ipath_f_init_hwerrors = ipath_pe_init_hwerrors;
1404         dd->ipath_f_early_init = ipath_pe_early_init;
1405         dd->ipath_f_handle_hwerrors = ipath_pe_handle_hwerrors;
1406         dd->ipath_f_quiet_serdes = ipath_pe_quiet_serdes;
1407         dd->ipath_f_bringup_serdes = ipath_pe_bringup_serdes;
1408         dd->ipath_f_clear_tids = ipath_pe_clear_tids;
1409         /*
1410          * this may get changed after we read the chip revision,
1411          * but we start with the safe version for all revs
1412          */
1413         dd->ipath_f_put_tid = ipath_pe_put_tid;
1414         dd->ipath_f_cleanup = ipath_setup_pe_cleanup;
1415         dd->ipath_f_setextled = ipath_setup_pe_setextled;
1416         dd->ipath_f_get_base_info = ipath_pe_get_base_info;
1417         dd->ipath_f_free_irq = ipath_pe_free_irq;
1418
1419         /* initialize chip-specific variables */
1420         dd->ipath_f_tidtemplate = ipath_pe_tidtemplate;
1421
1422         /*
1423          * setup the register offsets, since they are different for each
1424          * chip
1425          */
1426         dd->ipath_kregs = &ipath_pe_kregs;
1427         dd->ipath_cregs = &ipath_pe_cregs;
1428
1429         ipath_init_pe_variables(dd);
1430 }
1431