Merge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6
[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
325 /**
326  * ipath_pe_handle_hwerrors - display hardware errors.
327  * @dd: the infinipath device
328  * @msg: the output buffer
329  * @msgl: the size of the output buffer
330  *
331  * Use same msg buffer as regular errors to avoid excessive stack
332  * use.  Most hardware errors are catastrophic, but for right now,
333  * we'll print them and continue.  We reuse the same message buffer as
334  * ipath_handle_errors() to avoid excessive stack usage.
335  */
336 static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
337                                      size_t msgl)
338 {
339         ipath_err_t hwerrs;
340         u32 bits, ctrl;
341         int isfatal = 0;
342         char bitsmsg[64];
343         int log_idx;
344
345         hwerrs = ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus);
346         if (!hwerrs) {
347                 /*
348                  * better than printing cofusing messages
349                  * This seems to be related to clearing the crc error, or
350                  * the pll error during init.
351                  */
352                 ipath_cdbg(VERBOSE, "Called but no hardware errors set\n");
353                 return;
354         } else if (hwerrs == ~0ULL) {
355                 ipath_dev_err(dd, "Read of hardware error status failed "
356                               "(all bits set); ignoring\n");
357                 return;
358         }
359         ipath_stats.sps_hwerrs++;
360
361         /* Always clear the error status register, except MEMBISTFAIL,
362          * regardless of whether we continue or stop using the chip.
363          * We want that set so we know it failed, even across driver reload.
364          * We'll still ignore it in the hwerrmask.  We do this partly for
365          * diagnostics, but also for support */
366         ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear,
367                          hwerrs&~INFINIPATH_HWE_MEMBISTFAILED);
368
369         hwerrs &= dd->ipath_hwerrmask;
370
371         /* We log some errors to EEPROM, check if we have any of those. */
372         for (log_idx = 0; log_idx < IPATH_EEP_LOG_CNT; ++log_idx)
373                 if (hwerrs & dd->ipath_eep_st_masks[log_idx].hwerrs_to_log)
374                         ipath_inc_eeprom_err(dd, log_idx, 1);
375
376         /*
377          * make sure we get this much out, unless told to be quiet,
378          * or it's occurred within the last 5 seconds
379          */
380         if ((hwerrs & ~(dd->ipath_lasthwerror |
381                         ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
382                           INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
383                          << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT))) ||
384             (ipath_debug & __IPATH_VERBDBG))
385                 dev_info(&dd->pcidev->dev, "Hardware error: hwerr=0x%llx "
386                          "(cleared)\n", (unsigned long long) hwerrs);
387         dd->ipath_lasthwerror |= hwerrs;
388
389         if (hwerrs & ~dd->ipath_hwe_bitsextant)
390                 ipath_dev_err(dd, "hwerror interrupt with unknown errors "
391                               "%llx set\n", (unsigned long long)
392                               (hwerrs & ~dd->ipath_hwe_bitsextant));
393
394         ctrl = ipath_read_kreg32(dd, dd->ipath_kregs->kr_control);
395         if (ctrl & INFINIPATH_C_FREEZEMODE) {
396                 /*
397                  * parity errors in send memory are recoverable,
398                  * just cancel the send (if indicated in * sendbuffererror),
399                  * count the occurrence, unfreeze (if no other handled
400                  * hardware error bits are set), and continue. They can
401                  * occur if a processor speculative read is done to the PIO
402                  * buffer while we are sending a packet, for example.
403                  */
404                 if ((hwerrs & TXE_PIO_PARITY) && ipath_pe_txe_recover(dd))
405                         hwerrs &= ~TXE_PIO_PARITY;
406                 if (hwerrs) {
407                         /*
408                          * if any set that we aren't ignoring only make the
409                          * complaint once, in case it's stuck or recurring,
410                          * and we get here multiple times
411                          * Force link down, so switch knows, and
412                          * LEDs are turned off
413                          */
414                         if (dd->ipath_flags & IPATH_INITTED) {
415                                 ipath_set_linkstate(dd, IPATH_IB_LINKDOWN);
416                                 ipath_setup_pe_setextled(dd,
417                                         INFINIPATH_IBCS_L_STATE_DOWN,
418                                         INFINIPATH_IBCS_LT_STATE_DISABLED);
419                                 ipath_dev_err(dd, "Fatal Hardware Error (freeze "
420                                               "mode), no longer usable, SN %.16s\n",
421                                                   dd->ipath_serial);
422                                 isfatal = 1;
423                         }
424                         /*
425                          * Mark as having had an error for driver, and also
426                          * for /sys and status word mapped to user programs.
427                          * This marks unit as not usable, until reset
428                          */
429                         *dd->ipath_statusp &= ~IPATH_STATUS_IB_READY;
430                         *dd->ipath_statusp |= IPATH_STATUS_HWERROR;
431                         dd->ipath_flags &= ~IPATH_INITTED;
432                 } else {
433                         static u32 freeze_cnt;
434
435                         freeze_cnt++;
436                         ipath_dbg("Clearing freezemode on ignored or recovered "
437                                   "hardware error (%u)\n", freeze_cnt);
438                         ipath_clear_freeze(dd);
439                 }
440         }
441
442         *msg = '\0';
443
444         if (hwerrs & INFINIPATH_HWE_MEMBISTFAILED) {
445                 strlcat(msg, "[Memory BIST test failed, InfiniPath hardware unusable]",
446                         msgl);
447                 /* ignore from now on, so disable until driver reloaded */
448                 *dd->ipath_statusp |= IPATH_STATUS_HWERROR;
449                 dd->ipath_hwerrmask &= ~INFINIPATH_HWE_MEMBISTFAILED;
450                 ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
451                                  dd->ipath_hwerrmask);
452         }
453
454         ipath_format_hwerrors(hwerrs,
455                               ipath_6120_hwerror_msgs,
456                               sizeof(ipath_6120_hwerror_msgs)/
457                               sizeof(ipath_6120_hwerror_msgs[0]),
458                               msg, msgl);
459
460         if (hwerrs & (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK
461                       << INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT)) {
462                 bits = (u32) ((hwerrs >>
463                                INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) &
464                               INFINIPATH_HWE_PCIEMEMPARITYERR_MASK);
465                 snprintf(bitsmsg, sizeof bitsmsg,
466                          "[PCIe Mem Parity Errs %x] ", bits);
467                 strlcat(msg, bitsmsg, msgl);
468         }
469
470 #define _IPATH_PLL_FAIL (INFINIPATH_HWE_COREPLL_FBSLIP |        \
471                          INFINIPATH_HWE_COREPLL_RFSLIP )
472
473         if (hwerrs & _IPATH_PLL_FAIL) {
474                 snprintf(bitsmsg, sizeof bitsmsg,
475                          "[PLL failed (%llx), InfiniPath hardware unusable]",
476                          (unsigned long long) hwerrs & _IPATH_PLL_FAIL);
477                 strlcat(msg, bitsmsg, msgl);
478                 /* ignore from now on, so disable until driver reloaded */
479                 dd->ipath_hwerrmask &= ~(hwerrs & _IPATH_PLL_FAIL);
480                 ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
481                                  dd->ipath_hwerrmask);
482         }
483
484         if (hwerrs & INFINIPATH_HWE_SERDESPLLFAILED) {
485                 /*
486                  * If it occurs, it is left masked since the eternal
487                  * interface is unused
488                  */
489                 dd->ipath_hwerrmask &= ~INFINIPATH_HWE_SERDESPLLFAILED;
490                 ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
491                                  dd->ipath_hwerrmask);
492         }
493
494         if (*msg)
495                 ipath_dev_err(dd, "%s hardware error\n", msg);
496         if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg) {
497                 /*
498                  * for /sys status file ; if no trailing } is copied, we'll
499                  * know it was truncated.
500                  */
501                 snprintf(dd->ipath_freezemsg, dd->ipath_freezelen,
502                          "{%s}", msg);
503         }
504 }
505
506 /**
507  * ipath_pe_boardname - fill in the board name
508  * @dd: the infinipath device
509  * @name: the output buffer
510  * @namelen: the size of the output buffer
511  *
512  * info is based on the board revision register
513  */
514 static int ipath_pe_boardname(struct ipath_devdata *dd, char *name,
515                               size_t namelen)
516 {
517         char *n = NULL;
518         u8 boardrev = dd->ipath_boardrev;
519         int ret;
520
521         switch (boardrev) {
522         case 0:
523                 n = "InfiniPath_Emulation";
524                 break;
525         case 1:
526                 n = "InfiniPath_QLE7140-Bringup";
527                 break;
528         case 2:
529                 n = "InfiniPath_QLE7140";
530                 break;
531         case 3:
532                 n = "InfiniPath_QMI7140";
533                 break;
534         case 4:
535                 n = "InfiniPath_QEM7140";
536                 break;
537         case 5:
538                 n = "InfiniPath_QMH7140";
539                 break;
540         case 6:
541                 n = "InfiniPath_QLE7142";
542                 break;
543         default:
544                 ipath_dev_err(dd,
545                               "Don't yet know about board with ID %u\n",
546                               boardrev);
547                 snprintf(name, namelen, "Unknown_InfiniPath_PCIe_%u",
548                          boardrev);
549                 break;
550         }
551         if (n)
552                 snprintf(name, namelen, "%s", n);
553
554         if (dd->ipath_majrev != 4 || !dd->ipath_minrev || dd->ipath_minrev>2) {
555                 ipath_dev_err(dd, "Unsupported InfiniPath hardware revision %u.%u!\n",
556                               dd->ipath_majrev, dd->ipath_minrev);
557                 ret = 1;
558         } else
559                 ret = 0;
560
561         return ret;
562 }
563
564 /**
565  * ipath_pe_init_hwerrors - enable hardware errors
566  * @dd: the infinipath device
567  *
568  * now that we have finished initializing everything that might reasonably
569  * cause a hardware error, and cleared those errors bits as they occur,
570  * we can enable hardware errors in the mask (potentially enabling
571  * freeze mode), and enable hardware errors as errors (along with
572  * everything else) in errormask
573  */
574 static void ipath_pe_init_hwerrors(struct ipath_devdata *dd)
575 {
576         ipath_err_t val;
577         u64 extsval;
578
579         extsval = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extstatus);
580
581         if (!(extsval & INFINIPATH_EXTS_MEMBIST_ENDTEST))
582                 ipath_dev_err(dd, "MemBIST did not complete!\n");
583         if (extsval & INFINIPATH_EXTS_MEMBIST_FOUND)
584                 ipath_dbg("MemBIST corrected\n");
585
586         val = ~0ULL;    /* barring bugs, all hwerrors become interrupts, */
587
588         if (!dd->ipath_boardrev)        // no PLL for Emulator
589                 val &= ~INFINIPATH_HWE_SERDESPLLFAILED;
590
591         if (dd->ipath_minrev < 2) {
592                 /* workaround bug 9460 in internal interface bus parity
593                  * checking. Fixed (HW bug 9490) in Rev2.
594                  */
595                 val &= ~INFINIPATH_HWE_PCIEBUSPARITYRADM;
596         }
597         dd->ipath_hwerrmask = val;
598 }
599
600 /**
601  * ipath_pe_bringup_serdes - bring up the serdes
602  * @dd: the infinipath device
603  */
604 static int ipath_pe_bringup_serdes(struct ipath_devdata *dd)
605 {
606         u64 val, config1, prev_val;
607         int ret = 0;
608
609         ipath_dbg("Trying to bringup serdes\n");
610
611         if (ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus) &
612             INFINIPATH_HWE_SERDESPLLFAILED) {
613                 ipath_dbg("At start, serdes PLL failed bit set "
614                           "in hwerrstatus, clearing and continuing\n");
615                 ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear,
616                                  INFINIPATH_HWE_SERDESPLLFAILED);
617         }
618
619         val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0);
620         config1 = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig1);
621
622         ipath_cdbg(VERBOSE, "SerDes status config0=%llx config1=%llx, "
623                    "xgxsconfig %llx\n", (unsigned long long) val,
624                    (unsigned long long) config1, (unsigned long long)
625                    ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig));
626
627         /*
628          * Force reset on, also set rxdetect enable.  Must do before reading
629          * serdesstatus at least for simulation, or some of the bits in
630          * serdes status will come back as undefined and cause simulation
631          * failures
632          */
633         val |= INFINIPATH_SERDC0_RESET_PLL | INFINIPATH_SERDC0_RXDETECT_EN
634                 | INFINIPATH_SERDC0_L1PWR_DN;
635         ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
636         /* be sure chip saw it */
637         ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
638         udelay(5);              /* need pll reset set at least for a bit */
639         /*
640          * after PLL is reset, set the per-lane Resets and TxIdle and
641          * clear the PLL reset and rxdetect (to get falling edge).
642          * Leave L1PWR bits set (permanently)
643          */
644         val &= ~(INFINIPATH_SERDC0_RXDETECT_EN | INFINIPATH_SERDC0_RESET_PLL
645                  | INFINIPATH_SERDC0_L1PWR_DN);
646         val |= INFINIPATH_SERDC0_RESET_MASK | INFINIPATH_SERDC0_TXIDLE;
647         ipath_cdbg(VERBOSE, "Clearing pll reset and setting lane resets "
648                    "and txidle (%llx)\n", (unsigned long long) val);
649         ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
650         /* be sure chip saw it */
651         ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
652         /* need PLL reset clear for at least 11 usec before lane
653          * resets cleared; give it a few more to be sure */
654         udelay(15);
655         val &= ~(INFINIPATH_SERDC0_RESET_MASK | INFINIPATH_SERDC0_TXIDLE);
656
657         ipath_cdbg(VERBOSE, "Clearing lane resets and txidle "
658                    "(writing %llx)\n", (unsigned long long) val);
659         ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
660         /* be sure chip saw it */
661         val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
662
663         val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig);
664         prev_val = val;
665         if (((val >> INFINIPATH_XGXS_MDIOADDR_SHIFT) &
666              INFINIPATH_XGXS_MDIOADDR_MASK) != 3) {
667                 val &=
668                         ~(INFINIPATH_XGXS_MDIOADDR_MASK <<
669                           INFINIPATH_XGXS_MDIOADDR_SHIFT);
670                 /* MDIO address 3 */
671                 val |= 3ULL << INFINIPATH_XGXS_MDIOADDR_SHIFT;
672         }
673         if (val & INFINIPATH_XGXS_RESET) {
674                 val &= ~INFINIPATH_XGXS_RESET;
675         }
676         if (((val >> INFINIPATH_XGXS_RX_POL_SHIFT) &
677              INFINIPATH_XGXS_RX_POL_MASK) != dd->ipath_rx_pol_inv ) {
678                 /* need to compensate for Tx inversion in partner */
679                 val &= ~(INFINIPATH_XGXS_RX_POL_MASK <<
680                          INFINIPATH_XGXS_RX_POL_SHIFT);
681                 val |= dd->ipath_rx_pol_inv <<
682                         INFINIPATH_XGXS_RX_POL_SHIFT;
683         }
684         if (val != prev_val)
685                 ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val);
686
687         val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0);
688
689         /* clear current and de-emphasis bits */
690         config1 &= ~0x0ffffffff00ULL;
691         /* set current to 20ma */
692         config1 |= 0x00000000000ULL;
693         /* set de-emphasis to -5.68dB */
694         config1 |= 0x0cccc000000ULL;
695         ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig1, config1);
696
697         ipath_cdbg(VERBOSE, "done: SerDes status config0=%llx "
698                    "config1=%llx, sstatus=%llx xgxs=%llx\n",
699                    (unsigned long long) val, (unsigned long long) config1,
700                    (unsigned long long)
701                    ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesstatus),
702                    (unsigned long long)
703                    ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig));
704
705         if (!ipath_waitfor_mdio_cmdready(dd)) {
706                 ipath_write_kreg(
707                         dd, dd->ipath_kregs->kr_mdio,
708                         ipath_mdio_req(IPATH_MDIO_CMD_READ, 31,
709                                        IPATH_MDIO_CTRL_XGXS_REG_8, 0));
710                 if (ipath_waitfor_complete(dd, dd->ipath_kregs->kr_mdio,
711                                            IPATH_MDIO_DATAVALID, &val))
712                         ipath_dbg("Never got MDIO data for XGXS "
713                                   "status read\n");
714                 else
715                         ipath_cdbg(VERBOSE, "MDIO Read reg8, "
716                                    "'bank' 31 %x\n", (u32) val);
717         } else
718                 ipath_dbg("Never got MDIO cmdready for XGXS status read\n");
719
720         return ret;
721 }
722
723 /**
724  * ipath_pe_quiet_serdes - set serdes to txidle
725  * @dd: the infinipath device
726  * Called when driver is being unloaded
727  */
728 static void ipath_pe_quiet_serdes(struct ipath_devdata *dd)
729 {
730         u64 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0);
731
732         val |= INFINIPATH_SERDC0_TXIDLE;
733         ipath_dbg("Setting TxIdleEn on serdes (config0 = %llx)\n",
734                   (unsigned long long) val);
735         ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
736 }
737
738 static int ipath_pe_intconfig(struct ipath_devdata *dd)
739 {
740         u32 chiprev;
741
742         /*
743          * If the chip supports added error indication via GPIO pins,
744          * enable interrupts on those bits so the interrupt routine
745          * can count the events. Also set flag so interrupt routine
746          * can know they are expected.
747          */
748         chiprev = dd->ipath_revision >> INFINIPATH_R_CHIPREVMINOR_SHIFT;
749         if ((chiprev & INFINIPATH_R_CHIPREVMINOR_MASK) > 1) {
750                 /* Rev2+ reports extra errors via internal GPIO pins */
751                 dd->ipath_flags |= IPATH_GPIO_ERRINTRS;
752                 dd->ipath_gpio_mask |= IPATH_GPIO_ERRINTR_MASK;
753                 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask,
754                                  dd->ipath_gpio_mask);
755         }
756         return 0;
757 }
758
759 /**
760  * ipath_setup_pe_setextled - set the state of the two external LEDs
761  * @dd: the infinipath device
762  * @lst: the L state
763  * @ltst: the LT state
764
765  * These LEDs indicate the physical and logical state of IB link.
766  * For this chip (at least with recommended board pinouts), LED1
767  * is Yellow (logical state) and LED2 is Green (physical state),
768  *
769  * Note:  We try to match the Mellanox HCA LED behavior as best
770  * we can.  Green indicates physical link state is OK (something is
771  * plugged in, and we can train).
772  * Amber indicates the link is logically up (ACTIVE).
773  * Mellanox further blinks the amber LED to indicate data packet
774  * activity, but we have no hardware support for that, so it would
775  * require waking up every 10-20 msecs and checking the counters
776  * on the chip, and then turning the LED off if appropriate.  That's
777  * visible overhead, so not something we will do.
778  *
779  */
780 static void ipath_setup_pe_setextled(struct ipath_devdata *dd, u64 lst,
781                                      u64 ltst)
782 {
783         u64 extctl;
784         unsigned long flags = 0;
785
786         /* the diags use the LED to indicate diag info, so we leave
787          * the external LED alone when the diags are running */
788         if (ipath_diag_inuse)
789                 return;
790
791         /* Allow override of LED display for, e.g. Locating system in rack */
792         if (dd->ipath_led_override) {
793                 ltst = (dd->ipath_led_override & IPATH_LED_PHYS)
794                         ? INFINIPATH_IBCS_LT_STATE_LINKUP
795                         : INFINIPATH_IBCS_LT_STATE_DISABLED;
796                 lst = (dd->ipath_led_override & IPATH_LED_LOG)
797                         ? INFINIPATH_IBCS_L_STATE_ACTIVE
798                         : INFINIPATH_IBCS_L_STATE_DOWN;
799         }
800
801         spin_lock_irqsave(&dd->ipath_gpio_lock, flags);
802         extctl = dd->ipath_extctrl & ~(INFINIPATH_EXTC_LED1PRIPORT_ON |
803                                        INFINIPATH_EXTC_LED2PRIPORT_ON);
804
805         if (ltst & INFINIPATH_IBCS_LT_STATE_LINKUP)
806                 extctl |= INFINIPATH_EXTC_LED2PRIPORT_ON;
807         if (lst == INFINIPATH_IBCS_L_STATE_ACTIVE)
808                 extctl |= INFINIPATH_EXTC_LED1PRIPORT_ON;
809         dd->ipath_extctrl = extctl;
810         ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, extctl);
811         spin_unlock_irqrestore(&dd->ipath_gpio_lock, flags);
812 }
813
814 /**
815  * ipath_setup_pe_cleanup - clean up any per-chip chip-specific stuff
816  * @dd: the infinipath device
817  *
818  * This is called during driver unload.
819  * We do the pci_disable_msi here, not in generic code, because it
820  * isn't used for the HT chips. If we do end up needing pci_enable_msi
821  * at some point in the future for HT, we'll move the call back
822  * into the main init_one code.
823  */
824 static void ipath_setup_pe_cleanup(struct ipath_devdata *dd)
825 {
826         dd->ipath_msi_lo = 0;   /* just in case unload fails */
827         pci_disable_msi(dd->pcidev);
828 }
829
830 /**
831  * ipath_setup_pe_config - setup PCIe config related stuff
832  * @dd: the infinipath device
833  * @pdev: the PCI device
834  *
835  * The pci_enable_msi() call will fail on systems with MSI quirks
836  * such as those with AMD8131, even if the device of interest is not
837  * attached to that device, (in the 2.6.13 - 2.6.15 kernels, at least, fixed
838  * late in 2.6.16).
839  * All that can be done is to edit the kernel source to remove the quirk
840  * check until that is fixed.
841  * We do not need to call enable_msi() for our HyperTransport chip,
842  * even though it uses MSI, and we want to avoid the quirk warning, so
843  * So we call enable_msi only for PCIe.  If we do end up needing
844  * pci_enable_msi at some point in the future for HT, we'll move the
845  * call back into the main init_one code.
846  * We save the msi lo and hi values, so we can restore them after
847  * chip reset (the kernel PCI infrastructure doesn't yet handle that
848  * correctly).
849  */
850 static int ipath_setup_pe_config(struct ipath_devdata *dd,
851                                  struct pci_dev *pdev)
852 {
853         int pos, ret;
854
855         dd->ipath_msi_lo = 0;   /* used as a flag during reset processing */
856         ret = pci_enable_msi(dd->pcidev);
857         if (ret)
858                 ipath_dev_err(dd, "pci_enable_msi failed: %d, "
859                               "interrupts may not work\n", ret);
860         /* continue even if it fails, we may still be OK... */
861         dd->ipath_irq = pdev->irq;
862
863         if ((pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI))) {
864                 u16 control;
865                 pci_read_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
866                                       &dd->ipath_msi_lo);
867                 pci_read_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
868                                       &dd->ipath_msi_hi);
869                 pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
870                                      &control);
871                 /* now save the data (vector) info */
872                 pci_read_config_word(dd->pcidev,
873                                      pos + ((control & PCI_MSI_FLAGS_64BIT)
874                                             ? 12 : 8),
875                                      &dd->ipath_msi_data);
876                 ipath_cdbg(VERBOSE, "Read msi data 0x%x from config offset "
877                            "0x%x, control=0x%x\n", dd->ipath_msi_data,
878                            pos + ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
879                            control);
880                 /* we save the cachelinesize also, although it doesn't
881                  * really matter */
882                 pci_read_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE,
883                                      &dd->ipath_pci_cacheline);
884         } else
885                 ipath_dev_err(dd, "Can't find MSI capability, "
886                               "can't save MSI settings for reset\n");
887         if ((pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP))) {
888                 u16 linkstat;
889                 pci_read_config_word(dd->pcidev, pos + PCI_EXP_LNKSTA,
890                                      &linkstat);
891                 linkstat >>= 4;
892                 linkstat &= 0x1f;
893                 if (linkstat != 8)
894                         ipath_dev_err(dd, "PCIe width %u, "
895                                       "performance reduced\n", linkstat);
896         }
897         else
898                 ipath_dev_err(dd, "Can't find PCI Express "
899                               "capability!\n");
900         return 0;
901 }
902
903 static void ipath_init_pe_variables(struct ipath_devdata *dd)
904 {
905         /*
906          * bits for selecting i2c direction and values,
907          * used for I2C serial flash
908          */
909         dd->ipath_gpio_sda_num = _IPATH_GPIO_SDA_NUM;
910         dd->ipath_gpio_scl_num = _IPATH_GPIO_SCL_NUM;
911         dd->ipath_gpio_sda = IPATH_GPIO_SDA;
912         dd->ipath_gpio_scl = IPATH_GPIO_SCL;
913
914         /* variables for sanity checking interrupt and errors */
915         dd->ipath_hwe_bitsextant =
916                 (INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
917                  INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) |
918                 (INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
919                  INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT) |
920                 (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK <<
921                  INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) |
922                 INFINIPATH_HWE_PCIE1PLLFAILED |
923                 INFINIPATH_HWE_PCIE0PLLFAILED |
924                 INFINIPATH_HWE_PCIEPOISONEDTLP |
925                 INFINIPATH_HWE_PCIECPLTIMEOUT |
926                 INFINIPATH_HWE_PCIEBUSPARITYXTLH |
927                 INFINIPATH_HWE_PCIEBUSPARITYXADM |
928                 INFINIPATH_HWE_PCIEBUSPARITYRADM |
929                 INFINIPATH_HWE_MEMBISTFAILED |
930                 INFINIPATH_HWE_COREPLL_FBSLIP |
931                 INFINIPATH_HWE_COREPLL_RFSLIP |
932                 INFINIPATH_HWE_SERDESPLLFAILED |
933                 INFINIPATH_HWE_IBCBUSTOSPCPARITYERR |
934                 INFINIPATH_HWE_IBCBUSFRSPCPARITYERR;
935         dd->ipath_i_bitsextant =
936                 (INFINIPATH_I_RCVURG_MASK << INFINIPATH_I_RCVURG_SHIFT) |
937                 (INFINIPATH_I_RCVAVAIL_MASK <<
938                  INFINIPATH_I_RCVAVAIL_SHIFT) |
939                 INFINIPATH_I_ERROR | INFINIPATH_I_SPIOSENT |
940                 INFINIPATH_I_SPIOBUFAVAIL | INFINIPATH_I_GPIO;
941         dd->ipath_e_bitsextant =
942                 INFINIPATH_E_RFORMATERR | INFINIPATH_E_RVCRC |
943                 INFINIPATH_E_RICRC | INFINIPATH_E_RMINPKTLEN |
944                 INFINIPATH_E_RMAXPKTLEN | INFINIPATH_E_RLONGPKTLEN |
945                 INFINIPATH_E_RSHORTPKTLEN | INFINIPATH_E_RUNEXPCHAR |
946                 INFINIPATH_E_RUNSUPVL | INFINIPATH_E_REBP |
947                 INFINIPATH_E_RIBFLOW | INFINIPATH_E_RBADVERSION |
948                 INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL |
949                 INFINIPATH_E_RBADTID | INFINIPATH_E_RHDRLEN |
950                 INFINIPATH_E_RHDR | INFINIPATH_E_RIBLOSTLINK |
951                 INFINIPATH_E_SMINPKTLEN | INFINIPATH_E_SMAXPKTLEN |
952                 INFINIPATH_E_SUNDERRUN | INFINIPATH_E_SPKTLEN |
953                 INFINIPATH_E_SDROPPEDSMPPKT | INFINIPATH_E_SDROPPEDDATAPKT |
954                 INFINIPATH_E_SPIOARMLAUNCH | INFINIPATH_E_SUNEXPERRPKTNUM |
955                 INFINIPATH_E_SUNSUPVL | INFINIPATH_E_IBSTATUSCHANGED |
956                 INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET |
957                 INFINIPATH_E_HARDWARE;
958
959         dd->ipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK;
960         dd->ipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK;
961
962         /*
963          * EEPROM error log 0 is TXE Parity errors. 1 is RXE Parity.
964          * 2 is Some Misc, 3 is reserved for future.
965          */
966         dd->ipath_eep_st_masks[0].hwerrs_to_log =
967                 INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
968                 INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT;
969
970         /* Ignore errors in PIO/PBC on systems with unordered write-combining */
971         if (ipath_unordered_wc())
972                 dd->ipath_eep_st_masks[0].hwerrs_to_log &= ~TXE_PIO_PARITY;
973
974         dd->ipath_eep_st_masks[1].hwerrs_to_log =
975                 INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
976                 INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT;
977
978         dd->ipath_eep_st_masks[2].errs_to_log =
979                 INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET;
980
981
982 }
983
984 /* setup the MSI stuff again after a reset.  I'd like to just call
985  * pci_enable_msi() and request_irq() again, but when I do that,
986  * the MSI enable bit doesn't get set in the command word, and
987  * we switch to to a different interrupt vector, which is confusing,
988  * so I instead just do it all inline.  Perhaps somehow can tie this
989  * into the PCIe hotplug support at some point
990  * Note, because I'm doing it all here, I don't call pci_disable_msi()
991  * or free_irq() at the start of ipath_setup_pe_reset().
992  */
993 static int ipath_reinit_msi(struct ipath_devdata *dd)
994 {
995         int pos;
996         u16 control;
997         int ret;
998
999         if (!dd->ipath_msi_lo) {
1000                 dev_info(&dd->pcidev->dev, "Can't restore MSI config, "
1001                          "initial setup failed?\n");
1002                 ret = 0;
1003                 goto bail;
1004         }
1005
1006         if (!(pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI))) {
1007                 ipath_dev_err(dd, "Can't find MSI capability, "
1008                               "can't restore MSI settings\n");
1009                 ret = 0;
1010                 goto bail;
1011         }
1012         ipath_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n",
1013                    dd->ipath_msi_lo, pos + PCI_MSI_ADDRESS_LO);
1014         pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
1015                                dd->ipath_msi_lo);
1016         ipath_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n",
1017                    dd->ipath_msi_hi, pos + PCI_MSI_ADDRESS_HI);
1018         pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
1019                                dd->ipath_msi_hi);
1020         pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control);
1021         if (!(control & PCI_MSI_FLAGS_ENABLE)) {
1022                 ipath_cdbg(VERBOSE, "MSI control at off %x was %x, "
1023                            "setting MSI enable (%x)\n", pos + PCI_MSI_FLAGS,
1024                            control, control | PCI_MSI_FLAGS_ENABLE);
1025                 control |= PCI_MSI_FLAGS_ENABLE;
1026                 pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
1027                                       control);
1028         }
1029         /* now rewrite the data (vector) info */
1030         pci_write_config_word(dd->pcidev, pos +
1031                               ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
1032                               dd->ipath_msi_data);
1033         /* we restore the cachelinesize also, although it doesn't really
1034          * matter */
1035         pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE,
1036                               dd->ipath_pci_cacheline);
1037         /* and now set the pci master bit again */
1038         pci_set_master(dd->pcidev);
1039         ret = 1;
1040
1041 bail:
1042         return ret;
1043 }
1044
1045 /* This routine sleeps, so it can only be called from user context, not
1046  * from interrupt context.  If we need interrupt context, we can split
1047  * it into two routines.
1048 */
1049 static int ipath_setup_pe_reset(struct ipath_devdata *dd)
1050 {
1051         u64 val;
1052         int i;
1053         int ret;
1054
1055         /* Use ERROR so it shows up in logs, etc. */
1056         ipath_dev_err(dd, "Resetting InfiniPath unit %u\n", dd->ipath_unit);
1057         /* keep chip from being accessed in a few places */
1058         dd->ipath_flags &= ~(IPATH_INITTED|IPATH_PRESENT);
1059         val = dd->ipath_control | INFINIPATH_C_RESET;
1060         ipath_write_kreg(dd, dd->ipath_kregs->kr_control, val);
1061         mb();
1062
1063         for (i = 1; i <= 5; i++) {
1064                 int r;
1065                 /* allow MBIST, etc. to complete; longer on each retry.
1066                  * We sometimes get machine checks from bus timeout if no
1067                  * response, so for now, make it *really* long.
1068                  */
1069                 msleep(1000 + (1 + i) * 2000);
1070                 if ((r =
1071                      pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
1072                                             dd->ipath_pcibar0)))
1073                         ipath_dev_err(dd, "rewrite of BAR0 failed: %d\n",
1074                                       r);
1075                 if ((r =
1076                      pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
1077                                             dd->ipath_pcibar1)))
1078                         ipath_dev_err(dd, "rewrite of BAR1 failed: %d\n",
1079                                       r);
1080                 /* now re-enable memory access */
1081                 if ((r = pci_enable_device(dd->pcidev)))
1082                         ipath_dev_err(dd, "pci_enable_device failed after "
1083                                       "reset: %d\n", r);
1084                 /* whether it worked or not, mark as present, again */
1085                 dd->ipath_flags |= IPATH_PRESENT;
1086                 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_revision);
1087                 if (val == dd->ipath_revision) {
1088                         ipath_cdbg(VERBOSE, "Got matching revision "
1089                                    "register %llx on try %d\n",
1090                                    (unsigned long long) val, i);
1091                         ret = ipath_reinit_msi(dd);
1092                         goto bail;
1093                 }
1094                 /* Probably getting -1 back */
1095                 ipath_dbg("Didn't get expected revision register, "
1096                           "got %llx, try %d\n", (unsigned long long) val,
1097                           i + 1);
1098         }
1099         ret = 0; /* failed */
1100
1101 bail:
1102         return ret;
1103 }
1104
1105 /**
1106  * ipath_pe_put_tid - write a TID in chip
1107  * @dd: the infinipath device
1108  * @tidptr: pointer to the expected TID (in chip) to udpate
1109  * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0) for expected
1110  * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
1111  *
1112  * This exists as a separate routine to allow for special locking etc.
1113  * It's used for both the full cleanup on exit, as well as the normal
1114  * setup and teardown.
1115  */
1116 static void ipath_pe_put_tid(struct ipath_devdata *dd, u64 __iomem *tidptr,
1117                              u32 type, unsigned long pa)
1118 {
1119         u32 __iomem *tidp32 = (u32 __iomem *)tidptr;
1120         unsigned long flags = 0; /* keep gcc quiet */
1121
1122         if (pa != dd->ipath_tidinvalid) {
1123                 if (pa & ((1U << 11) - 1)) {
1124                         dev_info(&dd->pcidev->dev, "BUG: physaddr %lx "
1125                                  "not 4KB aligned!\n", pa);
1126                         return;
1127                 }
1128                 pa >>= 11;
1129                 /* paranoia check */
1130                 if (pa & (7<<29))
1131                         ipath_dev_err(dd,
1132                                       "BUG: Physical page address 0x%lx "
1133                                       "has bits set in 31-29\n", pa);
1134
1135                 if (type == RCVHQ_RCV_TYPE_EAGER)
1136                         pa |= dd->ipath_tidtemplate;
1137                 else /* for now, always full 4KB page */
1138                         pa |= 2 << 29;
1139         }
1140
1141         /* workaround chip bug 9437 by writing each TID twice
1142          * and holding a spinlock around the writes, so they don't
1143          * intermix with other TID (eager or expected) writes
1144          * Unfortunately, this call can be done from interrupt level
1145          * for the port 0 eager TIDs, so we have to use irqsave
1146          */
1147         spin_lock_irqsave(&dd->ipath_tid_lock, flags);
1148         ipath_write_kreg(dd, dd->ipath_kregs->kr_scratch, 0xfeeddeaf);
1149         if (dd->ipath_kregbase)
1150                 writel(pa, tidp32);
1151         ipath_write_kreg(dd, dd->ipath_kregs->kr_scratch, 0xdeadbeef);
1152         mmiowb();
1153         spin_unlock_irqrestore(&dd->ipath_tid_lock, flags);
1154 }
1155 /**
1156  * ipath_pe_put_tid_2 - write a TID in chip, Revision 2 or higher
1157  * @dd: the infinipath device
1158  * @tidptr: pointer to the expected TID (in chip) to udpate
1159  * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0) for expected
1160  * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
1161  *
1162  * This exists as a separate routine to allow for selection of the
1163  * appropriate "flavor". The static calls in cleanup just use the
1164  * revision-agnostic form, as they are not performance critical.
1165  */
1166 static void ipath_pe_put_tid_2(struct ipath_devdata *dd, u64 __iomem *tidptr,
1167                              u32 type, unsigned long pa)
1168 {
1169         u32 __iomem *tidp32 = (u32 __iomem *)tidptr;
1170
1171         if (pa != dd->ipath_tidinvalid) {
1172                 if (pa & ((1U << 11) - 1)) {
1173                         dev_info(&dd->pcidev->dev, "BUG: physaddr %lx "
1174                                  "not 2KB aligned!\n", pa);
1175                         return;
1176                 }
1177                 pa >>= 11;
1178                 /* paranoia check */
1179                 if (pa & (7<<29))
1180                         ipath_dev_err(dd,
1181                                       "BUG: Physical page address 0x%lx "
1182                                       "has bits set in 31-29\n", pa);
1183
1184                 if (type == RCVHQ_RCV_TYPE_EAGER)
1185                         pa |= dd->ipath_tidtemplate;
1186                 else /* for now, always full 4KB page */
1187                         pa |= 2 << 29;
1188         }
1189         if (dd->ipath_kregbase)
1190                 writel(pa, tidp32);
1191         mmiowb();
1192 }
1193
1194
1195 /**
1196  * ipath_pe_clear_tid - clear all TID entries for a port, expected and eager
1197  * @dd: the infinipath device
1198  * @port: the port
1199  *
1200  * clear all TID entries for a port, expected and eager.
1201  * Used from ipath_close().  On this chip, TIDs are only 32 bits,
1202  * not 64, but they are still on 64 bit boundaries, so tidbase
1203  * is declared as u64 * for the pointer math, even though we write 32 bits
1204  */
1205 static void ipath_pe_clear_tids(struct ipath_devdata *dd, unsigned port)
1206 {
1207         u64 __iomem *tidbase;
1208         unsigned long tidinv;
1209         int i;
1210
1211         if (!dd->ipath_kregbase)
1212                 return;
1213
1214         ipath_cdbg(VERBOSE, "Invalidate TIDs for port %u\n", port);
1215
1216         tidinv = dd->ipath_tidinvalid;
1217         tidbase = (u64 __iomem *)
1218                 ((char __iomem *)(dd->ipath_kregbase) +
1219                  dd->ipath_rcvtidbase +
1220                  port * dd->ipath_rcvtidcnt * sizeof(*tidbase));
1221
1222         for (i = 0; i < dd->ipath_rcvtidcnt; i++)
1223                 ipath_pe_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EXPECTED,
1224                                  tidinv);
1225
1226         tidbase = (u64 __iomem *)
1227                 ((char __iomem *)(dd->ipath_kregbase) +
1228                  dd->ipath_rcvegrbase +
1229                  port * dd->ipath_rcvegrcnt * sizeof(*tidbase));
1230
1231         for (i = 0; i < dd->ipath_rcvegrcnt; i++)
1232                 ipath_pe_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EAGER,
1233                                  tidinv);
1234 }
1235
1236 /**
1237  * ipath_pe_tidtemplate - setup constants for TID updates
1238  * @dd: the infinipath device
1239  *
1240  * We setup stuff that we use a lot, to avoid calculating each time
1241  */
1242 static void ipath_pe_tidtemplate(struct ipath_devdata *dd)
1243 {
1244         u32 egrsize = dd->ipath_rcvegrbufsize;
1245
1246         /* For now, we always allocate 4KB buffers (at init) so we can
1247          * receive max size packets.  We may want a module parameter to
1248          * specify 2KB or 4KB and/or make be per port instead of per device
1249          * for those who want to reduce memory footprint.  Note that the
1250          * ipath_rcvhdrentsize size must be large enough to hold the largest
1251          * IB header (currently 96 bytes) that we expect to handle (plus of
1252          * course the 2 dwords of RHF).
1253          */
1254         if (egrsize == 2048)
1255                 dd->ipath_tidtemplate = 1U << 29;
1256         else if (egrsize == 4096)
1257                 dd->ipath_tidtemplate = 2U << 29;
1258         else {
1259                 egrsize = 4096;
1260                 dev_info(&dd->pcidev->dev, "BUG: unsupported egrbufsize "
1261                          "%u, using %u\n", dd->ipath_rcvegrbufsize,
1262                          egrsize);
1263                 dd->ipath_tidtemplate = 2U << 29;
1264         }
1265         dd->ipath_tidinvalid = 0;
1266 }
1267
1268 static int ipath_pe_early_init(struct ipath_devdata *dd)
1269 {
1270         dd->ipath_flags |= IPATH_4BYTE_TID;
1271
1272         /*
1273          * For openfabrics, we need to be able to handle an IB header of
1274          * 24 dwords.  HT chip has arbitrary sized receive buffers, so we
1275          * made them the same size as the PIO buffers.  This chip does not
1276          * handle arbitrary size buffers, so we need the header large enough
1277          * to handle largest IB header, but still have room for a 2KB MTU
1278          * standard IB packet.
1279          */
1280         dd->ipath_rcvhdrentsize = 24;
1281         dd->ipath_rcvhdrsize = IPATH_DFLT_RCVHDRSIZE;
1282
1283         /*
1284          * To truly support a 4KB MTU (for usermode), we need to
1285          * bump this to a larger value.  For now, we use them for
1286          * the kernel only.
1287          */
1288         dd->ipath_rcvegrbufsize = 2048;
1289         /*
1290          * the min() check here is currently a nop, but it may not always
1291          * be, depending on just how we do ipath_rcvegrbufsize
1292          */
1293         dd->ipath_ibmaxlen = min(dd->ipath_piosize2k,
1294                                  dd->ipath_rcvegrbufsize +
1295                                  (dd->ipath_rcvhdrentsize << 2));
1296         dd->ipath_init_ibmaxlen = dd->ipath_ibmaxlen;
1297
1298         /*
1299          * We can request a receive interrupt for 1 or
1300          * more packets from current offset.  For now, we set this
1301          * up for a single packet.
1302          */
1303         dd->ipath_rhdrhead_intr_off = 1ULL<<32;
1304
1305         ipath_get_eeprom_info(dd);
1306
1307         return 0;
1308 }
1309
1310 int __attribute__((weak)) ipath_unordered_wc(void)
1311 {
1312         return 0;
1313 }
1314
1315 /**
1316  * ipath_init_pe_get_base_info - set chip-specific flags for user code
1317  * @pd: the infinipath port
1318  * @kbase: ipath_base_info pointer
1319  *
1320  * We set the PCIE flag because the lower bandwidth on PCIe vs
1321  * HyperTransport can affect some user packet algorithms.
1322  */
1323 static int ipath_pe_get_base_info(struct ipath_portdata *pd, void *kbase)
1324 {
1325         struct ipath_base_info *kinfo = kbase;
1326         struct ipath_devdata *dd;
1327
1328         if (ipath_unordered_wc()) {
1329                 kinfo->spi_runtime_flags |= IPATH_RUNTIME_FORCE_WC_ORDER;
1330                 ipath_cdbg(PROC, "Intel processor, forcing WC order\n");
1331         }
1332         else
1333                 ipath_cdbg(PROC, "Not Intel processor, WC ordered\n");
1334
1335         if (pd == NULL)
1336                 goto done;
1337
1338         dd = pd->port_dd;
1339
1340 done:
1341         kinfo->spi_runtime_flags |= IPATH_RUNTIME_PCIE;
1342         return 0;
1343 }
1344
1345 static void ipath_pe_free_irq(struct ipath_devdata *dd)
1346 {
1347         free_irq(dd->ipath_irq, dd);
1348         dd->ipath_irq = 0;
1349 }
1350
1351 /*
1352  * On platforms using this chip, and not having ordered WC stores, we
1353  * can get TXE parity errors due to speculative reads to the PIO buffers,
1354  * and this, due to a chip bug can result in (many) false parity error
1355  * reports.  So it's a debug print on those, and an info print on systems
1356  * where the speculative reads don't occur.
1357  * Because we can get lots of false errors, we have no upper limit
1358  * on recovery attempts on those platforms.
1359  */
1360 static int ipath_pe_txe_recover(struct ipath_devdata *dd)
1361 {
1362         if (ipath_unordered_wc())
1363                 ipath_dbg("Recovering from TXE PIO parity error\n");
1364         else {
1365                 int cnt = ++ipath_stats.sps_txeparity;
1366                 if (cnt >= IPATH_MAX_PARITY_ATTEMPTS)  {
1367                         if (cnt == IPATH_MAX_PARITY_ATTEMPTS)
1368                                 ipath_dev_err(dd,
1369                                         "Too many attempts to recover from "
1370                                         "TXE parity, giving up\n");
1371                         return 0;
1372                 }
1373                 dev_info(&dd->pcidev->dev,
1374                         "Recovering from TXE PIO parity error\n");
1375         }
1376         return 1;
1377 }
1378
1379 /**
1380  * ipath_init_iba6120_funcs - set up the chip-specific function pointers
1381  * @dd: the infinipath device
1382  *
1383  * This is global, and is called directly at init to set up the
1384  * chip-specific function pointers for later use.
1385  */
1386 void ipath_init_iba6120_funcs(struct ipath_devdata *dd)
1387 {
1388         dd->ipath_f_intrsetup = ipath_pe_intconfig;
1389         dd->ipath_f_bus = ipath_setup_pe_config;
1390         dd->ipath_f_reset = ipath_setup_pe_reset;
1391         dd->ipath_f_get_boardname = ipath_pe_boardname;
1392         dd->ipath_f_init_hwerrors = ipath_pe_init_hwerrors;
1393         dd->ipath_f_early_init = ipath_pe_early_init;
1394         dd->ipath_f_handle_hwerrors = ipath_pe_handle_hwerrors;
1395         dd->ipath_f_quiet_serdes = ipath_pe_quiet_serdes;
1396         dd->ipath_f_bringup_serdes = ipath_pe_bringup_serdes;
1397         dd->ipath_f_clear_tids = ipath_pe_clear_tids;
1398         if (dd->ipath_minrev >= 2)
1399                 dd->ipath_f_put_tid = ipath_pe_put_tid_2;
1400         else
1401                 dd->ipath_f_put_tid = ipath_pe_put_tid;
1402         dd->ipath_f_cleanup = ipath_setup_pe_cleanup;
1403         dd->ipath_f_setextled = ipath_setup_pe_setextled;
1404         dd->ipath_f_get_base_info = ipath_pe_get_base_info;
1405         dd->ipath_f_free_irq = ipath_pe_free_irq;
1406
1407         /* initialize chip-specific variables */
1408         dd->ipath_f_tidtemplate = ipath_pe_tidtemplate;
1409
1410         /*
1411          * setup the register offsets, since they are different for each
1412          * chip
1413          */
1414         dd->ipath_kregs = &ipath_pe_kregs;
1415         dd->ipath_cregs = &ipath_pe_cregs;
1416
1417         ipath_init_pe_variables(dd);
1418 }
1419