powerpc/MSI: Fix race condition in tearing down MSI interrupts
[firefly-linux-kernel-4.4.55.git] / arch / powerpc / platforms / powernv / pci.c
1
2 /*
3  * Support PCI/PCIe on PowerNV platforms
4  *
5  * Currently supports only P5IOC2
6  *
7  * Copyright 2011 Benjamin Herrenschmidt, IBM Corp.
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License
11  * as published by the Free Software Foundation; either version
12  * 2 of the License, or (at your option) any later version.
13  */
14
15 #include <linux/kernel.h>
16 #include <linux/pci.h>
17 #include <linux/delay.h>
18 #include <linux/string.h>
19 #include <linux/init.h>
20 #include <linux/bootmem.h>
21 #include <linux/irq.h>
22 #include <linux/io.h>
23 #include <linux/msi.h>
24
25 #include <asm/sections.h>
26 #include <asm/io.h>
27 #include <asm/prom.h>
28 #include <asm/pci-bridge.h>
29 #include <asm/machdep.h>
30 #include <asm/msi_bitmap.h>
31 #include <asm/ppc-pci.h>
32 #include <asm/opal.h>
33 #include <asm/iommu.h>
34 #include <asm/tce.h>
35 #include <asm/firmware.h>
36
37 #include "powernv.h"
38 #include "pci.h"
39
40 /* Delay in usec */
41 #define PCI_RESET_DELAY_US      3000000
42
43 #define cfg_dbg(fmt...) do { } while(0)
44 //#define cfg_dbg(fmt...)       printk(fmt)
45
46 #ifdef CONFIG_PCI_MSI
47 static int pnv_msi_check_device(struct pci_dev* pdev, int nvec, int type)
48 {
49         struct pci_controller *hose = pci_bus_to_host(pdev->bus);
50         struct pnv_phb *phb = hose->private_data;
51
52         if (pdev->no_64bit_msi && !phb->msi32_support)
53                 return -ENODEV;
54
55         return (phb && phb->msi_bmp.bitmap) ? 0 : -ENODEV;
56 }
57
58 static int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
59 {
60         struct pci_controller *hose = pci_bus_to_host(pdev->bus);
61         struct pnv_phb *phb = hose->private_data;
62         struct msi_desc *entry;
63         struct msi_msg msg;
64         int hwirq;
65         unsigned int virq;
66         int rc;
67
68         if (WARN_ON(!phb))
69                 return -ENODEV;
70
71         list_for_each_entry(entry, &pdev->msi_list, list) {
72                 if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
73                         pr_warn("%s: Supports only 64-bit MSIs\n",
74                                 pci_name(pdev));
75                         return -ENXIO;
76                 }
77                 hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
78                 if (hwirq < 0) {
79                         pr_warn("%s: Failed to find a free MSI\n",
80                                 pci_name(pdev));
81                         return -ENOSPC;
82                 }
83                 virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
84                 if (virq == NO_IRQ) {
85                         pr_warn("%s: Failed to map MSI to linux irq\n",
86                                 pci_name(pdev));
87                         msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
88                         return -ENOMEM;
89                 }
90                 rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
91                                     virq, entry->msi_attrib.is_64, &msg);
92                 if (rc) {
93                         pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
94                         irq_dispose_mapping(virq);
95                         msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
96                         return rc;
97                 }
98                 irq_set_msi_desc(virq, entry);
99                 write_msi_msg(virq, &msg);
100         }
101         return 0;
102 }
103
104 static void pnv_teardown_msi_irqs(struct pci_dev *pdev)
105 {
106         struct pci_controller *hose = pci_bus_to_host(pdev->bus);
107         struct pnv_phb *phb = hose->private_data;
108         struct msi_desc *entry;
109         irq_hw_number_t hwirq;
110
111         if (WARN_ON(!phb))
112                 return;
113
114         list_for_each_entry(entry, &pdev->msi_list, list) {
115                 if (entry->irq == NO_IRQ)
116                         continue;
117                 hwirq = virq_to_hw(entry->irq);
118                 irq_set_msi_desc(entry->irq, NULL);
119                 irq_dispose_mapping(entry->irq);
120                 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, 1);
121         }
122 }
123 #endif /* CONFIG_PCI_MSI */
124
125 static void pnv_pci_dump_p7ioc_diag_data(struct pnv_phb *phb)
126 {
127         struct OpalIoP7IOCPhbErrorData *data = &phb->diag.p7ioc;
128         int i;
129
130         pr_info("PHB %d diagnostic data:\n", phb->hose->global_number);
131
132         pr_info("  brdgCtl              = 0x%08x\n", data->brdgCtl);
133
134         pr_info("  portStatusReg        = 0x%08x\n", data->portStatusReg);
135         pr_info("  rootCmplxStatus      = 0x%08x\n", data->rootCmplxStatus);
136         pr_info("  busAgentStatus       = 0x%08x\n", data->busAgentStatus);
137
138         pr_info("  deviceStatus         = 0x%08x\n", data->deviceStatus);
139         pr_info("  slotStatus           = 0x%08x\n", data->slotStatus);
140         pr_info("  linkStatus           = 0x%08x\n", data->linkStatus);
141         pr_info("  devCmdStatus         = 0x%08x\n", data->devCmdStatus);
142         pr_info("  devSecStatus         = 0x%08x\n", data->devSecStatus);
143
144         pr_info("  rootErrorStatus      = 0x%08x\n", data->rootErrorStatus);
145         pr_info("  uncorrErrorStatus    = 0x%08x\n", data->uncorrErrorStatus);
146         pr_info("  corrErrorStatus      = 0x%08x\n", data->corrErrorStatus);
147         pr_info("  tlpHdr1              = 0x%08x\n", data->tlpHdr1);
148         pr_info("  tlpHdr2              = 0x%08x\n", data->tlpHdr2);
149         pr_info("  tlpHdr3              = 0x%08x\n", data->tlpHdr3);
150         pr_info("  tlpHdr4              = 0x%08x\n", data->tlpHdr4);
151         pr_info("  sourceId             = 0x%08x\n", data->sourceId);
152
153         pr_info("  errorClass           = 0x%016llx\n", data->errorClass);
154         pr_info("  correlator           = 0x%016llx\n", data->correlator);
155
156         pr_info("  p7iocPlssr           = 0x%016llx\n", data->p7iocPlssr);
157         pr_info("  p7iocCsr             = 0x%016llx\n", data->p7iocCsr);
158         pr_info("  lemFir               = 0x%016llx\n", data->lemFir);
159         pr_info("  lemErrorMask         = 0x%016llx\n", data->lemErrorMask);
160         pr_info("  lemWOF               = 0x%016llx\n", data->lemWOF);
161         pr_info("  phbErrorStatus       = 0x%016llx\n", data->phbErrorStatus);
162         pr_info("  phbFirstErrorStatus  = 0x%016llx\n", data->phbFirstErrorStatus);
163         pr_info("  phbErrorLog0         = 0x%016llx\n", data->phbErrorLog0);
164         pr_info("  phbErrorLog1         = 0x%016llx\n", data->phbErrorLog1);
165         pr_info("  mmioErrorStatus      = 0x%016llx\n", data->mmioErrorStatus);
166         pr_info("  mmioFirstErrorStatus = 0x%016llx\n", data->mmioFirstErrorStatus);
167         pr_info("  mmioErrorLog0        = 0x%016llx\n", data->mmioErrorLog0);
168         pr_info("  mmioErrorLog1        = 0x%016llx\n", data->mmioErrorLog1);
169         pr_info("  dma0ErrorStatus      = 0x%016llx\n", data->dma0ErrorStatus);
170         pr_info("  dma0FirstErrorStatus = 0x%016llx\n", data->dma0FirstErrorStatus);
171         pr_info("  dma0ErrorLog0        = 0x%016llx\n", data->dma0ErrorLog0);
172         pr_info("  dma0ErrorLog1        = 0x%016llx\n", data->dma0ErrorLog1);
173         pr_info("  dma1ErrorStatus      = 0x%016llx\n", data->dma1ErrorStatus);
174         pr_info("  dma1FirstErrorStatus = 0x%016llx\n", data->dma1FirstErrorStatus);
175         pr_info("  dma1ErrorLog0        = 0x%016llx\n", data->dma1ErrorLog0);
176         pr_info("  dma1ErrorLog1        = 0x%016llx\n", data->dma1ErrorLog1);
177
178         for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
179                 if ((data->pestA[i] >> 63) == 0 &&
180                     (data->pestB[i] >> 63) == 0)
181                         continue;
182                 pr_info("  PE[%3d] PESTA        = 0x%016llx\n", i, data->pestA[i]);
183                 pr_info("          PESTB        = 0x%016llx\n", data->pestB[i]);
184         }
185 }
186
187 static void pnv_pci_dump_phb_diag_data(struct pnv_phb *phb)
188 {
189         switch(phb->model) {
190         case PNV_PHB_MODEL_P7IOC:
191                 pnv_pci_dump_p7ioc_diag_data(phb);
192                 break;
193         default:
194                 pr_warning("PCI %d: Can't decode this PHB diag data\n",
195                            phb->hose->global_number);
196         }
197 }
198
199 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
200 {
201         unsigned long flags, rc;
202         int has_diag;
203
204         spin_lock_irqsave(&phb->lock, flags);
205
206         rc = opal_pci_get_phb_diag_data(phb->opal_id, phb->diag.blob, PNV_PCI_DIAG_BUF_SIZE);
207         has_diag = (rc == OPAL_SUCCESS);
208
209         rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no,
210                                        OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
211         if (rc) {
212                 pr_warning("PCI %d: Failed to clear EEH freeze state"
213                            " for PE#%d, err %ld\n",
214                            phb->hose->global_number, pe_no, rc);
215
216                 /* For now, let's only display the diag buffer when we fail to clear
217                  * the EEH status. We'll do more sensible things later when we have
218                  * proper EEH support. We need to make sure we don't pollute ourselves
219                  * with the normal errors generated when probing empty slots
220                  */
221                 if (has_diag)
222                         pnv_pci_dump_phb_diag_data(phb);
223                 else
224                         pr_warning("PCI %d: No diag data available\n",
225                                    phb->hose->global_number);
226         }
227
228         spin_unlock_irqrestore(&phb->lock, flags);
229 }
230
231 static void pnv_pci_config_check_eeh(struct pnv_phb *phb, struct pci_bus *bus,
232                                      u32 bdfn)
233 {
234         s64     rc;
235         u8      fstate;
236         u16     pcierr;
237         u32     pe_no;
238
239         /* Get PE# if we support IODA */
240         pe_no = phb->bdfn_to_pe ? phb->bdfn_to_pe(phb, bus, bdfn & 0xff) : 0;
241
242         /* Read freeze status */
243         rc = opal_pci_eeh_freeze_status(phb->opal_id, pe_no, &fstate, &pcierr,
244                                         NULL);
245         if (rc) {
246                 pr_warning("PCI %d: Failed to read EEH status for PE#%d,"
247                            " err %lld\n", phb->hose->global_number, pe_no, rc);
248                 return;
249         }
250         cfg_dbg(" -> EEH check, bdfn=%04x PE%d fstate=%x\n",
251                 bdfn, pe_no, fstate);
252         if (fstate != 0)
253                 pnv_pci_handle_eeh_config(phb, pe_no);
254 }
255
256 static int pnv_pci_read_config(struct pci_bus *bus,
257                                unsigned int devfn,
258                                int where, int size, u32 *val)
259 {
260         struct pci_controller *hose = pci_bus_to_host(bus);
261         struct pnv_phb *phb = hose->private_data;
262         u32 bdfn = (((uint64_t)bus->number) << 8) | devfn;
263         s64 rc;
264
265         if (hose == NULL)
266                 return PCIBIOS_DEVICE_NOT_FOUND;
267
268         switch (size) {
269         case 1: {
270                 u8 v8;
271                 rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
272                 *val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
273                 break;
274         }
275         case 2: {
276                 u16 v16;
277                 rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
278                                                    &v16);
279                 *val = (rc == OPAL_SUCCESS) ? v16 : 0xffff;
280                 break;
281         }
282         case 4: {
283                 u32 v32;
284                 rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
285                 *val = (rc == OPAL_SUCCESS) ? v32 : 0xffffffff;
286                 break;
287         }
288         default:
289                 return PCIBIOS_FUNC_NOT_SUPPORTED;
290         }
291         cfg_dbg("pnv_pci_read_config bus: %x devfn: %x +%x/%x -> %08x\n",
292                 bus->number, devfn, where, size, *val);
293
294         /* Check if the PHB got frozen due to an error (no response) */
295         pnv_pci_config_check_eeh(phb, bus, bdfn);
296
297         return PCIBIOS_SUCCESSFUL;
298 }
299
300 static int pnv_pci_write_config(struct pci_bus *bus,
301                                 unsigned int devfn,
302                                 int where, int size, u32 val)
303 {
304         struct pci_controller *hose = pci_bus_to_host(bus);
305         struct pnv_phb *phb = hose->private_data;
306         u32 bdfn = (((uint64_t)bus->number) << 8) | devfn;
307
308         if (hose == NULL)
309                 return PCIBIOS_DEVICE_NOT_FOUND;
310
311         cfg_dbg("pnv_pci_write_config bus: %x devfn: %x +%x/%x -> %08x\n",
312                 bus->number, devfn, where, size, val);
313         switch (size) {
314         case 1:
315                 opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
316                 break;
317         case 2:
318                 opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
319                 break;
320         case 4:
321                 opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
322                 break;
323         default:
324                 return PCIBIOS_FUNC_NOT_SUPPORTED;
325         }
326         /* Check if the PHB got frozen due to an error (no response) */
327         pnv_pci_config_check_eeh(phb, bus, bdfn);
328
329         return PCIBIOS_SUCCESSFUL;
330 }
331
332 struct pci_ops pnv_pci_ops = {
333         .read = pnv_pci_read_config,
334         .write = pnv_pci_write_config,
335 };
336
337 static int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
338                          unsigned long uaddr, enum dma_data_direction direction,
339                          struct dma_attrs *attrs)
340 {
341         u64 proto_tce;
342         u64 *tcep, *tces;
343         u64 rpn;
344
345         proto_tce = TCE_PCI_READ; // Read allowed
346
347         if (direction != DMA_TO_DEVICE)
348                 proto_tce |= TCE_PCI_WRITE;
349
350         tces = tcep = ((u64 *)tbl->it_base) + index - tbl->it_offset;
351         rpn = __pa(uaddr) >> TCE_SHIFT;
352
353         while (npages--)
354                 *(tcep++) = proto_tce | (rpn++ << TCE_RPN_SHIFT);
355
356         /* Some implementations won't cache invalid TCEs and thus may not
357          * need that flush. We'll probably turn it_type into a bit mask
358          * of flags if that becomes the case
359          */
360         if (tbl->it_type & TCE_PCI_SWINV_CREATE)
361                 pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1);
362
363         return 0;
364 }
365
366 static void pnv_tce_free(struct iommu_table *tbl, long index, long npages)
367 {
368         u64 *tcep, *tces;
369
370         tces = tcep = ((u64 *)tbl->it_base) + index - tbl->it_offset;
371
372         while (npages--)
373                 *(tcep++) = 0;
374
375         if (tbl->it_type & TCE_PCI_SWINV_FREE)
376                 pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1);
377 }
378
379 static unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
380 {
381         return ((u64 *)tbl->it_base)[index - tbl->it_offset];
382 }
383
384 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
385                                void *tce_mem, u64 tce_size,
386                                u64 dma_offset)
387 {
388         tbl->it_blocksize = 16;
389         tbl->it_base = (unsigned long)tce_mem;
390         tbl->it_offset = dma_offset >> IOMMU_PAGE_SHIFT;
391         tbl->it_index = 0;
392         tbl->it_size = tce_size >> 3;
393         tbl->it_busno = 0;
394         tbl->it_type = TCE_PCI;
395 }
396
397 static struct iommu_table *pnv_pci_setup_bml_iommu(struct pci_controller *hose)
398 {
399         struct iommu_table *tbl;
400         const __be64 *basep, *swinvp;
401         const __be32 *sizep;
402
403         basep = of_get_property(hose->dn, "linux,tce-base", NULL);
404         sizep = of_get_property(hose->dn, "linux,tce-size", NULL);
405         if (basep == NULL || sizep == NULL) {
406                 pr_err("PCI: %s has missing tce entries !\n",
407                        hose->dn->full_name);
408                 return NULL;
409         }
410         tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, hose->node);
411         if (WARN_ON(!tbl))
412                 return NULL;
413         pnv_pci_setup_iommu_table(tbl, __va(be64_to_cpup(basep)),
414                                   be32_to_cpup(sizep), 0);
415         iommu_init_table(tbl, hose->node);
416
417         /* Deal with SW invalidated TCEs when needed (BML way) */
418         swinvp = of_get_property(hose->dn, "linux,tce-sw-invalidate-info",
419                                  NULL);
420         if (swinvp) {
421                 tbl->it_busno = swinvp[1];
422                 tbl->it_index = (unsigned long)ioremap(swinvp[0], 8);
423                 tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE;
424         }
425         return tbl;
426 }
427
428 static void pnv_pci_dma_fallback_setup(struct pci_controller *hose,
429                                        struct pci_dev *pdev)
430 {
431         struct device_node *np = pci_bus_to_OF_node(hose->bus);
432         struct pci_dn *pdn;
433
434         if (np == NULL)
435                 return;
436         pdn = PCI_DN(np);
437         if (!pdn->iommu_table)
438                 pdn->iommu_table = pnv_pci_setup_bml_iommu(hose);
439         if (!pdn->iommu_table)
440                 return;
441         set_iommu_table_base(&pdev->dev, pdn->iommu_table);
442 }
443
444 static void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
445 {
446         struct pci_controller *hose = pci_bus_to_host(pdev->bus);
447         struct pnv_phb *phb = hose->private_data;
448
449         /* If we have no phb structure, try to setup a fallback based on
450          * the device-tree (RTAS PCI for example)
451          */
452         if (phb && phb->dma_dev_setup)
453                 phb->dma_dev_setup(phb, pdev);
454         else
455                 pnv_pci_dma_fallback_setup(hose, pdev);
456 }
457
458 void pnv_pci_shutdown(void)
459 {
460         struct pci_controller *hose;
461
462         list_for_each_entry(hose, &hose_list, list_node) {
463                 struct pnv_phb *phb = hose->private_data;
464
465                 if (phb && phb->shutdown)
466                         phb->shutdown(phb);
467         }
468 }
469
470 /* Fixup wrong class code in p7ioc and p8 root complex */
471 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
472 {
473         dev->class = PCI_CLASS_BRIDGE_PCI << 8;
474 }
475 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
476
477 static int pnv_pci_probe_mode(struct pci_bus *bus)
478 {
479         struct pci_controller *hose = pci_bus_to_host(bus);
480         const __be64 *tstamp;
481         u64 now, target;
482
483
484         /* We hijack this as a way to ensure we have waited long
485          * enough since the reset was lifted on the PCI bus
486          */
487         if (bus != hose->bus)
488                 return PCI_PROBE_NORMAL;
489         tstamp = of_get_property(hose->dn, "reset-clear-timestamp", NULL);
490         if (!tstamp || !*tstamp)
491                 return PCI_PROBE_NORMAL;
492
493         now = mftb() / tb_ticks_per_usec;
494         target = (be64_to_cpup(tstamp) / tb_ticks_per_usec)
495                 + PCI_RESET_DELAY_US;
496
497         pr_devel("pci %04d: Reset target: 0x%llx now: 0x%llx\n",
498                  hose->global_number, target, now);
499
500         if (now < target)
501                 msleep((target - now + 999) / 1000);
502
503         return PCI_PROBE_NORMAL;
504 }
505
506 void __init pnv_pci_init(void)
507 {
508         struct device_node *np;
509
510         pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
511
512         /* OPAL absent, try POPAL first then RTAS detection of PHBs */
513         if (!firmware_has_feature(FW_FEATURE_OPAL)) {
514 #ifdef CONFIG_PPC_POWERNV_RTAS
515                 init_pci_config_tokens();
516                 find_and_init_phbs();
517 #endif /* CONFIG_PPC_POWERNV_RTAS */
518         }
519         /* OPAL is here, do our normal stuff */
520         else {
521                 int found_ioda = 0;
522
523                 /* Look for IODA IO-Hubs. We don't support mixing IODA
524                  * and p5ioc2 due to the need to change some global
525                  * probing flags
526                  */
527                 for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
528                         pnv_pci_init_ioda_hub(np);
529                         found_ioda = 1;
530                 }
531
532                 /* Look for p5ioc2 IO-Hubs */
533                 if (!found_ioda)
534                         for_each_compatible_node(np, NULL, "ibm,p5ioc2")
535                                 pnv_pci_init_p5ioc2_hub(np);
536
537                 /* Look for ioda2 built-in PHB3's */
538                 for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
539                         pnv_pci_init_ioda2_phb(np);
540         }
541
542         /* Setup the linkage between OF nodes and PHBs */
543         pci_devs_phb_init();
544
545         /* Configure IOMMU DMA hooks */
546         ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
547         ppc_md.tce_build = pnv_tce_build;
548         ppc_md.tce_free = pnv_tce_free;
549         ppc_md.tce_get = pnv_tce_get;
550         ppc_md.pci_probe_mode = pnv_pci_probe_mode;
551         set_pci_dma_ops(&dma_iommu_ops);
552
553         /* Configure MSIs */
554 #ifdef CONFIG_PCI_MSI
555         ppc_md.msi_check_device = pnv_msi_check_device;
556         ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
557         ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
558 #endif
559 }