1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright 2014 IBM Corp.
4  */
5 
6 #include <linux/spinlock.h>
7 #include <linux/kernel.h>
8 #include <linux/module.h>
9 #include <linux/device.h>
10 #include <linux/mutex.h>
11 #include <linux/init.h>
12 #include <linux/list.h>
13 #include <linux/mm.h>
14 #include <linux/of.h>
15 #include <linux/slab.h>
16 #include <linux/idr.h>
17 #include <linux/pci.h>
18 #include <linux/platform_device.h>
19 #include <linux/sched/task.h>
20 
21 #include <asm/cputable.h>
22 #include <asm/mmu.h>
23 #include <misc/cxl-base.h>
24 
25 #include "cxl.h"
26 #include "trace.h"
27 
28 static DEFINE_SPINLOCK(adapter_idr_lock);
29 static DEFINE_IDR(cxl_adapter_idr);
30 
31 uint cxl_verbose;
32 module_param_named(verbose, cxl_verbose, uint, 0600);
33 MODULE_PARM_DESC(verbose, "Enable verbose dmesg output");
34 
35 const struct cxl_backend_ops *cxl_ops;
36 
cxl_afu_slbia(struct cxl_afu * afu)37 int cxl_afu_slbia(struct cxl_afu *afu)
38 {
39 	unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
40 
41 	pr_devel("cxl_afu_slbia issuing SLBIA command\n");
42 	cxl_p2n_write(afu, CXL_SLBIA_An, CXL_TLB_SLB_IQ_ALL);
43 	while (cxl_p2n_read(afu, CXL_SLBIA_An) & CXL_TLB_SLB_P) {
44 		if (time_after_eq(jiffies, timeout)) {
45 			dev_warn(&afu->dev, "WARNING: CXL AFU SLBIA timed out!\n");
46 			return -EBUSY;
47 		}
48 		/* If the adapter has gone down, we can assume that we
49 		 * will PERST it and that will invalidate everything.
50 		 */
51 		if (!cxl_ops->link_ok(afu->adapter, afu))
52 			return -EIO;
53 		cpu_relax();
54 	}
55 	return 0;
56 }
57 
_cxl_slbia(struct cxl_context * ctx,struct mm_struct * mm)58 static inline void _cxl_slbia(struct cxl_context *ctx, struct mm_struct *mm)
59 {
60 	unsigned long flags;
61 
62 	if (ctx->mm != mm)
63 		return;
64 
65 	pr_devel("%s matched mm - card: %i afu: %i pe: %i\n", __func__,
66 		 ctx->afu->adapter->adapter_num, ctx->afu->slice, ctx->pe);
67 
68 	spin_lock_irqsave(&ctx->sste_lock, flags);
69 	trace_cxl_slbia(ctx);
70 	memset(ctx->sstp, 0, ctx->sst_size);
71 	spin_unlock_irqrestore(&ctx->sste_lock, flags);
72 	mb();
73 	cxl_afu_slbia(ctx->afu);
74 }
75 
cxl_slbia_core(struct mm_struct * mm)76 static inline void cxl_slbia_core(struct mm_struct *mm)
77 {
78 	struct cxl *adapter;
79 	struct cxl_afu *afu;
80 	struct cxl_context *ctx;
81 	int card, slice, id;
82 
83 	pr_devel("%s called\n", __func__);
84 
85 	spin_lock(&adapter_idr_lock);
86 	idr_for_each_entry(&cxl_adapter_idr, adapter, card) {
87 		/* XXX: Make this lookup faster with link from mm to ctx */
88 		spin_lock(&adapter->afu_list_lock);
89 		for (slice = 0; slice < adapter->slices; slice++) {
90 			afu = adapter->afu[slice];
91 			if (!afu || !afu->enabled)
92 				continue;
93 			rcu_read_lock();
94 			idr_for_each_entry(&afu->contexts_idr, ctx, id)
95 				_cxl_slbia(ctx, mm);
96 			rcu_read_unlock();
97 		}
98 		spin_unlock(&adapter->afu_list_lock);
99 	}
100 	spin_unlock(&adapter_idr_lock);
101 }
102 
103 static struct cxl_calls cxl_calls = {
104 	.cxl_slbia = cxl_slbia_core,
105 	.owner = THIS_MODULE,
106 };
107 
cxl_alloc_sst(struct cxl_context * ctx)108 int cxl_alloc_sst(struct cxl_context *ctx)
109 {
110 	unsigned long vsid;
111 	u64 ea_mask, size, sstp0, sstp1;
112 
113 	sstp0 = 0;
114 	sstp1 = 0;
115 
116 	ctx->sst_size = PAGE_SIZE;
117 	ctx->sst_lru = 0;
118 	ctx->sstp = (struct cxl_sste *)get_zeroed_page(GFP_KERNEL);
119 	if (!ctx->sstp) {
120 		pr_err("cxl_alloc_sst: Unable to allocate segment table\n");
121 		return -ENOMEM;
122 	}
123 	pr_devel("SSTP allocated at 0x%p\n", ctx->sstp);
124 
125 	vsid  = get_kernel_vsid((u64)ctx->sstp, mmu_kernel_ssize) << 12;
126 
127 	sstp0 |= (u64)mmu_kernel_ssize << CXL_SSTP0_An_B_SHIFT;
128 	sstp0 |= (SLB_VSID_KERNEL | mmu_psize_defs[mmu_linear_psize].sllp) << 50;
129 
130 	size = (((u64)ctx->sst_size >> 8) - 1) << CXL_SSTP0_An_SegTableSize_SHIFT;
131 	if (unlikely(size & ~CXL_SSTP0_An_SegTableSize_MASK)) {
132 		WARN(1, "Impossible segment table size\n");
133 		return -EINVAL;
134 	}
135 	sstp0 |= size;
136 
137 	if (mmu_kernel_ssize == MMU_SEGSIZE_256M)
138 		ea_mask = 0xfffff00ULL;
139 	else
140 		ea_mask = 0xffffffff00ULL;
141 
142 	sstp0 |=  vsid >>     (50-14);  /*   Top 14 bits of VSID */
143 	sstp1 |= (vsid << (64-(50-14))) & ~ea_mask;
144 	sstp1 |= (u64)ctx->sstp & ea_mask;
145 	sstp1 |= CXL_SSTP1_An_V;
146 
147 	pr_devel("Looked up %#llx: slbfee. %#llx (ssize: %x, vsid: %#lx), copied to SSTP0: %#llx, SSTP1: %#llx\n",
148 			(u64)ctx->sstp, (u64)ctx->sstp & ESID_MASK, mmu_kernel_ssize, vsid, sstp0, sstp1);
149 
150 	/* Store calculated sstp hardware points for use later */
151 	ctx->sstp0 = sstp0;
152 	ctx->sstp1 = sstp1;
153 
154 	return 0;
155 }
156 
157 /* print buffer content as integers when debugging */
cxl_dump_debug_buffer(void * buf,size_t buf_len)158 void cxl_dump_debug_buffer(void *buf, size_t buf_len)
159 {
160 #ifdef DEBUG
161 	int i, *ptr;
162 
163 	/*
164 	 * We want to regroup up to 4 integers per line, which means they
165 	 * need to be in the same pr_devel() statement
166 	 */
167 	ptr = (int *) buf;
168 	for (i = 0; i * 4 < buf_len; i += 4) {
169 		if ((i + 3) * 4 < buf_len)
170 			pr_devel("%.8x %.8x %.8x %.8x\n", ptr[i], ptr[i + 1],
171 				ptr[i + 2], ptr[i + 3]);
172 		else if ((i + 2) * 4 < buf_len)
173 			pr_devel("%.8x %.8x %.8x\n", ptr[i], ptr[i + 1],
174 				ptr[i + 2]);
175 		else if ((i + 1) * 4 < buf_len)
176 			pr_devel("%.8x %.8x\n", ptr[i], ptr[i + 1]);
177 		else
178 			pr_devel("%.8x\n", ptr[i]);
179 	}
180 #endif /* DEBUG */
181 }
182 
183 /* Find a CXL adapter by it's number and increase it's refcount */
get_cxl_adapter(int num)184 struct cxl *get_cxl_adapter(int num)
185 {
186 	struct cxl *adapter;
187 
188 	spin_lock(&adapter_idr_lock);
189 	if ((adapter = idr_find(&cxl_adapter_idr, num)))
190 		get_device(&adapter->dev);
191 	spin_unlock(&adapter_idr_lock);
192 
193 	return adapter;
194 }
195 
cxl_alloc_adapter_nr(struct cxl * adapter)196 static int cxl_alloc_adapter_nr(struct cxl *adapter)
197 {
198 	int i;
199 
200 	idr_preload(GFP_KERNEL);
201 	spin_lock(&adapter_idr_lock);
202 	i = idr_alloc(&cxl_adapter_idr, adapter, 0, 0, GFP_NOWAIT);
203 	spin_unlock(&adapter_idr_lock);
204 	idr_preload_end();
205 	if (i < 0)
206 		return i;
207 
208 	adapter->adapter_num = i;
209 
210 	return 0;
211 }
212 
cxl_remove_adapter_nr(struct cxl * adapter)213 void cxl_remove_adapter_nr(struct cxl *adapter)
214 {
215 	idr_remove(&cxl_adapter_idr, adapter->adapter_num);
216 }
217 
cxl_alloc_adapter(void)218 struct cxl *cxl_alloc_adapter(void)
219 {
220 	struct cxl *adapter;
221 
222 	if (!(adapter = kzalloc(sizeof(struct cxl), GFP_KERNEL)))
223 		return NULL;
224 
225 	spin_lock_init(&adapter->afu_list_lock);
226 
227 	if (cxl_alloc_adapter_nr(adapter))
228 		goto err1;
229 
230 	if (dev_set_name(&adapter->dev, "card%i", adapter->adapter_num))
231 		goto err2;
232 
233 	/* start with context lock taken */
234 	atomic_set(&adapter->contexts_num, -1);
235 
236 	return adapter;
237 err2:
238 	cxl_remove_adapter_nr(adapter);
239 err1:
240 	kfree(adapter);
241 	return NULL;
242 }
243 
cxl_alloc_afu(struct cxl * adapter,int slice)244 struct cxl_afu *cxl_alloc_afu(struct cxl *adapter, int slice)
245 {
246 	struct cxl_afu *afu;
247 
248 	if (!(afu = kzalloc(sizeof(struct cxl_afu), GFP_KERNEL)))
249 		return NULL;
250 
251 	afu->adapter = adapter;
252 	afu->dev.parent = &adapter->dev;
253 	afu->dev.release = cxl_ops->release_afu;
254 	afu->slice = slice;
255 	idr_init(&afu->contexts_idr);
256 	mutex_init(&afu->contexts_lock);
257 	spin_lock_init(&afu->afu_cntl_lock);
258 	atomic_set(&afu->configured_state, -1);
259 	afu->prefault_mode = CXL_PREFAULT_NONE;
260 	afu->irqs_max = afu->adapter->user_irqs;
261 
262 	return afu;
263 }
264 
cxl_afu_select_best_mode(struct cxl_afu * afu)265 int cxl_afu_select_best_mode(struct cxl_afu *afu)
266 {
267 	if (afu->modes_supported & CXL_MODE_DIRECTED)
268 		return cxl_ops->afu_activate_mode(afu, CXL_MODE_DIRECTED);
269 
270 	if (afu->modes_supported & CXL_MODE_DEDICATED)
271 		return cxl_ops->afu_activate_mode(afu, CXL_MODE_DEDICATED);
272 
273 	dev_warn(&afu->dev, "No supported programming modes available\n");
274 	/* We don't fail this so the user can inspect sysfs */
275 	return 0;
276 }
277 
cxl_adapter_context_get(struct cxl * adapter)278 int cxl_adapter_context_get(struct cxl *adapter)
279 {
280 	int rc;
281 
282 	rc = atomic_inc_unless_negative(&adapter->contexts_num);
283 	return rc ? 0 : -EBUSY;
284 }
285 
cxl_adapter_context_put(struct cxl * adapter)286 void cxl_adapter_context_put(struct cxl *adapter)
287 {
288 	atomic_dec_if_positive(&adapter->contexts_num);
289 }
290 
cxl_adapter_context_lock(struct cxl * adapter)291 int cxl_adapter_context_lock(struct cxl *adapter)
292 {
293 	int rc;
294 	/* no active contexts -> contexts_num == 0 */
295 	rc = atomic_cmpxchg(&adapter->contexts_num, 0, -1);
296 	return rc ? -EBUSY : 0;
297 }
298 
cxl_adapter_context_unlock(struct cxl * adapter)299 void cxl_adapter_context_unlock(struct cxl *adapter)
300 {
301 	int val = atomic_cmpxchg(&adapter->contexts_num, -1, 0);
302 
303 	/*
304 	 * contexts lock taken -> contexts_num == -1
305 	 * If not true then show a warning and force reset the lock.
306 	 * This will happen when context_unlock was requested without
307 	 * doing a context_lock.
308 	 */
309 	if (val != -1) {
310 		atomic_set(&adapter->contexts_num, 0);
311 		WARN(1, "Adapter context unlocked with %d active contexts",
312 		     val);
313 	}
314 }
315 
init_cxl(void)316 static int __init init_cxl(void)
317 {
318 	int rc = 0;
319 
320 	if (!tlbie_capable)
321 		return -EINVAL;
322 
323 	if ((rc = cxl_file_init()))
324 		return rc;
325 
326 	cxl_debugfs_init();
327 
328 	/*
329 	 * we don't register the callback on P9. slb callack is only
330 	 * used for the PSL8 MMU and CX4.
331 	 */
332 	if (cxl_is_power8()) {
333 		rc = register_cxl_calls(&cxl_calls);
334 		if (rc)
335 			goto err;
336 	}
337 
338 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
339 		cxl_ops = &cxl_native_ops;
340 		rc = pci_register_driver(&cxl_pci_driver);
341 	}
342 #ifdef CONFIG_PPC_PSERIES
343 	else {
344 		cxl_ops = &cxl_guest_ops;
345 		rc = platform_driver_register(&cxl_of_driver);
346 	}
347 #endif
348 	if (rc)
349 		goto err1;
350 
351 	return 0;
352 err1:
353 	if (cxl_is_power8())
354 		unregister_cxl_calls(&cxl_calls);
355 err:
356 	cxl_debugfs_exit();
357 	cxl_file_exit();
358 
359 	return rc;
360 }
361 
exit_cxl(void)362 static void exit_cxl(void)
363 {
364 	if (cpu_has_feature(CPU_FTR_HVMODE))
365 		pci_unregister_driver(&cxl_pci_driver);
366 #ifdef CONFIG_PPC_PSERIES
367 	else
368 		platform_driver_unregister(&cxl_of_driver);
369 #endif
370 
371 	cxl_debugfs_exit();
372 	cxl_file_exit();
373 	if (cxl_is_power8())
374 		unregister_cxl_calls(&cxl_calls);
375 	idr_destroy(&cxl_adapter_idr);
376 }
377 
378 module_init(init_cxl);
379 module_exit(exit_cxl);
380 
381 MODULE_DESCRIPTION("IBM Coherent Accelerator");
382 MODULE_AUTHOR("Ian Munsie <imunsie@au1.ibm.com>");
383 MODULE_LICENSE("GPL");
384