enable tun/tap/iptables

hc

2024-05-10 ee930fffee469d076998274a2ca55e13dc1efb67

 kernel/drivers/misc/ocxl/link.c
..
..
@@ -6,6 +6,7 @@
6
6
 #include <linux/mmu_context.h>
7
7
 #include <asm/copro.h>
8
8
 #include <asm/pnv-ocxl.h>
9
+#include <asm/xive.h>
9
10
 #include <misc/ocxl.h>
10
11
 #include "ocxl_internal.h"
11
12
 #include "trace.h"
..
..
@@ -76,7 +77,7 @@
76
77
  * limited number of opencapi slots on a system and lookup is only
77
78
  * done when the device is probed
78
79
  */
79
-struct link {
80
+struct ocxl_link {
80
81
 	struct list_head list;
81
82
 	struct kref ref;
82
83
 	int domain;
..
..
@@ -163,7 +164,7 @@
163
164
 		if (fault->dsisr & SPA_XSL_S)
164
165
 			access |= _PAGE_WRITE;
165
166
 
166
-		if (REGION_ID(fault->dar) != USER_REGION_ID)
167
+		if (get_region_id(fault->dar) != USER_REGION_ID)
167
168
 			access |= _PAGE_PRIVILEGED;
168
169
 
169
170
 		local_irq_save(flags);
..
..
@@ -179,12 +180,12 @@
179
180
 
180
181
 static irqreturn_t xsl_fault_handler(int irq, void *data)
181
182
 {
182
-	struct link *link = (struct link *) data;
183
+	struct ocxl_link *link = (struct ocxl_link *) data;
183
184
 	struct spa *spa = link->spa;
184
185
 	u64 dsisr, dar, pe_handle;
185
186
 	struct pe_data *pe_data;
186
187
 	struct ocxl_process_element *pe;
187
-	int lpid, pid, tid;
188
+	int pid;
188
189
 	bool schedule = false;
189
190
 
190
191
 	read_irq(spa, &dsisr, &dar, &pe_handle);
..
..
@@ -192,9 +193,7 @@
192
193
 
193
194
 	WARN_ON(pe_handle > SPA_PE_MASK);
194
195
 	pe = spa->spa_mem + pe_handle;
195
-	lpid = be32_to_cpu(pe->lpid);
196
196
 	pid = be32_to_cpu(pe->pid);
197
-	tid = be32_to_cpu(pe->tid);
198
197
 	/* We could be reading all null values here if the PE is being
199
198
 	 * removed while an interrupt kicks in. It's not supposed to
200
199
 	 * happen if the driver notified the AFU to terminate the
..
..
@@ -223,6 +222,17 @@
223
222
 		 */
224
223
 		rcu_read_unlock();
225
224
 		pr_debug("Unknown mm context for xsl interrupt\n");
225
+		ack_irq(spa, ADDRESS_ERROR);
226
+		return IRQ_HANDLED;
227
+	}
228
+
229
+	if (!pe_data->mm) {
230
+		/*
231
+		 * translation fault from a kernel context - an OpenCAPI
232
+		 * device tried to access a bad kernel address
233
+		 */
234
+		rcu_read_unlock();
235
+		pr_warn("Unresolved OpenCAPI xsl fault in kernel context\n");
226
236
 		ack_irq(spa, ADDRESS_ERROR);
227
237
 		return IRQ_HANDLED;
228
238
 	}
..
..
@@ -256,7 +266,7 @@
256
266
 				&spa->reg_tfc, &spa->reg_pe_handle);
257
267
 }
258
268
 
259
-static int setup_xsl_irq(struct pci_dev *dev, struct link *link)
269
+static int setup_xsl_irq(struct pci_dev *dev, struct ocxl_link *link)
260
270
 {
261
271
 	struct spa *spa = link->spa;
262
272
 	int rc;
..
..
@@ -273,9 +283,9 @@
273
283
 	spa->irq_name = kasprintf(GFP_KERNEL, "ocxl-xsl-%x-%x-%x",
274
284
 				link->domain, link->bus, link->dev);
275
285
 	if (!spa->irq_name) {
276
-		unmap_irq_registers(spa);
277
286
 		dev_err(&dev->dev, "Can't allocate name for xsl interrupt\n");
278
-		return -ENOMEM;
287
+		rc = -ENOMEM;
288
+		goto err_xsl;
279
289
 	}
280
290
 	/*
281
291
 	 * At some point, we'll need to look into allowing a higher
..
..
@@ -283,11 +293,10 @@
283
293
 	 */
284
294
 	spa->virq = irq_create_mapping(NULL, hwirq);
285
295
 	if (!spa->virq) {
286
-		kfree(spa->irq_name);
287
-		unmap_irq_registers(spa);
288
296
 		dev_err(&dev->dev,
289
297
 			"irq_create_mapping failed for translation interrupt\n");
290
-		return -EINVAL;
298
+		rc = -EINVAL;
299
+		goto err_name;
291
300
 	}
292
301
 
293
302
 	dev_dbg(&dev->dev, "hwirq %d mapped to virq %d\n", hwirq, spa->virq);
..
..
@@ -295,18 +304,24 @@
295
304
 	rc = request_irq(spa->virq, xsl_fault_handler, 0, spa->irq_name,
296
305
 			link);
297
306
 	if (rc) {
298
-		irq_dispose_mapping(spa->virq);
299
-		kfree(spa->irq_name);
300
-		unmap_irq_registers(spa);
301
307
 		dev_err(&dev->dev,
302
308
 			"request_irq failed for translation interrupt: %d\n",
303
309
 			rc);
304
-		return -EINVAL;
310
+		rc = -EINVAL;
311
+		goto err_mapping;
305
312
 	}
306
313
 	return 0;
314
+
315
+err_mapping:
316
+	irq_dispose_mapping(spa->virq);
317
+err_name:
318
+	kfree(spa->irq_name);
319
+err_xsl:
320
+	unmap_irq_registers(spa);
321
+	return rc;
307
322
 }
308
323
 
309
-static void release_xsl_irq(struct link *link)
324
+static void release_xsl_irq(struct ocxl_link *link)
310
325
 {
311
326
 	struct spa *spa = link->spa;
312
327
 
..
..
@@ -318,7 +333,7 @@
318
333
 	unmap_irq_registers(spa);
319
334
 }
320
335
 
321
-static int alloc_spa(struct pci_dev *dev, struct link *link)
336
+static int alloc_spa(struct pci_dev *dev, struct ocxl_link *link)
322
337
 {
323
338
 	struct spa *spa;
324
339
 
..
..
@@ -345,7 +360,7 @@
345
360
 	return 0;
346
361
 }
347
362
 
348
-static void free_spa(struct link *link)
363
+static void free_spa(struct ocxl_link *link)
349
364
 {
350
365
 	struct spa *spa = link->spa;
351
366
 
..
..
@@ -359,12 +374,12 @@
359
374
 	}
360
375
 }
361
376
 
362
-static int alloc_link(struct pci_dev *dev, int PE_mask, struct link **out_link)
377
+static int alloc_link(struct pci_dev *dev, int PE_mask, struct ocxl_link **out_link)
363
378
 {
364
-	struct link *link;
379
+	struct ocxl_link *link;
365
380
 	int rc;
366
381
 
367
-	link = kzalloc(sizeof(struct link), GFP_KERNEL);
382
+	link = kzalloc(sizeof(struct ocxl_link), GFP_KERNEL);
368
383
 	if (!link)
369
384
 		return -ENOMEM;
370
385
 
..
..
@@ -400,7 +415,7 @@
400
415
 	return rc;
401
416
 }
402
417
 
403
-static void free_link(struct link *link)
418
+static void free_link(struct ocxl_link *link)
404
419
 {
405
420
 	release_xsl_irq(link);
406
421
 	free_spa(link);
..
..
@@ -410,7 +425,7 @@
410
425
 int ocxl_link_setup(struct pci_dev *dev, int PE_mask, void **link_handle)
411
426
 {
412
427
 	int rc = 0;
413
-	struct link *link;
428
+	struct ocxl_link *link;
414
429
 
415
430
 	mutex_lock(&links_list_lock);
416
431
 	list_for_each_entry(link, &links_list, list) {
..
..
@@ -437,7 +452,7 @@
437
452
 
438
453
 static void release_xsl(struct kref *ref)
439
454
 {
440
-	struct link *link = container_of(ref, struct link, ref);
455
+	struct ocxl_link *link = container_of(ref, struct ocxl_link, ref);
441
456
 
442
457
 	list_del(&link->list);
443
458
 	/* call platform code before releasing data */
..
..
@@ -447,7 +462,7 @@
447
462
 
448
463
 void ocxl_link_release(struct pci_dev *dev, void *link_handle)
449
464
 {
450
-	struct link *link = (struct link *) link_handle;
465
+	struct ocxl_link *link = (struct ocxl_link *) link_handle;
451
466
 
452
467
 	mutex_lock(&links_list_lock);
453
468
 	kref_put(&link->ref, release_xsl);
..
..
@@ -483,7 +498,7 @@
483
498
 		void (*xsl_err_cb)(void *data, u64 addr, u64 dsisr),
484
499
 		void *xsl_err_data)
485
500
 {
486
-	struct link *link = (struct link *) link_handle;
501
+	struct ocxl_link *link = (struct ocxl_link *) link_handle;
487
502
 	struct spa *spa = link->spa;
488
503
 	struct ocxl_process_element *pe;
489
504
 	int pe_handle, rc = 0;
..
..
@@ -520,7 +535,13 @@
520
535
 	pe->amr = cpu_to_be64(amr);
521
536
 	pe->software_state = cpu_to_be32(SPA_PE_VALID);
522
537
 
523
-	mm_context_add_copro(mm);
538
+	/*
539
+	 * For user contexts, register a copro so that TLBIs are seen
540
+	 * by the nest MMU. If we have a kernel context, TLBIs are
541
+	 * already global.
542
+	 */
543
+	if (mm)
544
+		mm_context_add_copro(mm);
524
545
 	/*
525
546
 	 * Barrier is to make sure PE is visible in the SPA before it
526
547
 	 * is used by the device. It also helps with the global TLBI
..
..
@@ -543,7 +564,8 @@
543
564
 	 * have a reference on mm_users. Incrementing mm_count solves
544
565
 	 * the problem.
545
566
 	 */
546
-	mmgrab(mm);
567
+	if (mm)
568
+		mmgrab(mm);
547
569
 	trace_ocxl_context_add(current->pid, spa->spa_mem, pasid, pidr, tidr);
548
570
 unlock:
549
571
 	mutex_unlock(&spa->spa_lock);
..
..
@@ -553,7 +575,7 @@
553
575
 
554
576
 int ocxl_link_update_pe(void *link_handle, int pasid, __u16 tid)
555
577
 {
556
-	struct link *link = (struct link *) link_handle;
578
+	struct ocxl_link *link = (struct ocxl_link *) link_handle;
557
579
 	struct spa *spa = link->spa;
558
580
 	struct ocxl_process_element *pe;
559
581
 	int pe_handle, rc;
..
..
@@ -589,7 +611,7 @@
589
611
 
590
612
 int ocxl_link_remove_pe(void *link_handle, int pasid)
591
613
 {
592
-	struct link *link = (struct link *) link_handle;
614
+	struct ocxl_link *link = (struct ocxl_link *) link_handle;
593
615
 	struct spa *spa = link->spa;
594
616
 	struct ocxl_process_element *pe;
595
617
 	struct pe_data *pe_data;
..
..
@@ -649,8 +671,10 @@
649
671
 	if (!pe_data) {
650
672
 		WARN(1, "Couldn't find pe data when removing PE\n");
651
673
 	} else {
652
-		mm_context_remove_copro(pe_data->mm);
653
-		mmdrop(pe_data->mm);
674
+		if (pe_data->mm) {
675
+			mm_context_remove_copro(pe_data->mm);
676
+			mmdrop(pe_data->mm);
677
+		}
654
678
 		kfree_rcu(pe_data, rcu);
655
679
 	}
656
680
 unlock:
..
..
@@ -659,32 +683,30 @@
659
683
 }
660
684
 EXPORT_SYMBOL_GPL(ocxl_link_remove_pe);
661
685
 
662
-int ocxl_link_irq_alloc(void *link_handle, int *hw_irq, u64 *trigger_addr)
686
+int ocxl_link_irq_alloc(void *link_handle, int *hw_irq)
663
687
 {
664
-	struct link *link = (struct link *) link_handle;
665
-	int rc, irq;
666
-	u64 addr;
688
+	struct ocxl_link *link = (struct ocxl_link *) link_handle;
689
+	int irq;
667
690
 
668
691
 	if (atomic_dec_if_positive(&link->irq_available) < 0)
669
692
 		return -ENOSPC;
670
693
 
671
-	rc = pnv_ocxl_alloc_xive_irq(&irq, &addr);
672
-	if (rc) {
694
+	irq = xive_native_alloc_irq();
695
+	if (!irq) {
673
696
 		atomic_inc(&link->irq_available);
674
-		return rc;
697
+		return -ENXIO;
675
698
 	}
676
699
 
677
700
 	*hw_irq = irq;
678
-	*trigger_addr = addr;
679
701
 	return 0;
680
702
 }
681
703
 EXPORT_SYMBOL_GPL(ocxl_link_irq_alloc);
682
704
 
683
705
 void ocxl_link_free_irq(void *link_handle, int hw_irq)
684
706
 {
685
-	struct link *link = (struct link *) link_handle;
707
+	struct ocxl_link *link = (struct ocxl_link *) link_handle;
686
708
 
687
-	pnv_ocxl_free_xive_irq(hw_irq);
709
+	xive_native_free_irq(hw_irq);
688
710
 	atomic_inc(&link->irq_available);
689
711
 }
690
712
 EXPORT_SYMBOL_GPL(ocxl_link_free_irq);