]>
Commit | Line | Data |
---|---|---|
f204e0b8 IM |
1 | /* |
2 | * Copyright 2014 IBM Corp. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | */ | |
9 | ||
10 | #include <linux/spinlock.h> | |
11 | #include <linux/kernel.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/device.h> | |
14 | #include <linux/mutex.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/list.h> | |
17 | #include <linux/mm.h> | |
18 | #include <linux/of.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/idr.h> | |
21 | #include <linux/pci.h> | |
0881e7bd IM |
22 | #include <linux/sched/task.h> |
23 | ||
f204e0b8 | 24 | #include <asm/cputable.h> |
ec249dd8 | 25 | #include <misc/cxl-base.h> |
f204e0b8 IM |
26 | |
27 | #include "cxl.h" | |
9bcf28cd | 28 | #include "trace.h" |
f204e0b8 IM |
29 | |
30 | static DEFINE_SPINLOCK(adapter_idr_lock); | |
31 | static DEFINE_IDR(cxl_adapter_idr); | |
32 | ||
33 | uint cxl_verbose; | |
34 | module_param_named(verbose, cxl_verbose, uint, 0600); | |
35 | MODULE_PARM_DESC(verbose, "Enable verbose dmesg output"); | |
36 | ||
5be587b1 FB |
37 | const struct cxl_backend_ops *cxl_ops; |
38 | ||
86331862 CL |
39 | int cxl_afu_slbia(struct cxl_afu *afu) |
40 | { | |
41 | unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT); | |
42 | ||
43 | pr_devel("cxl_afu_slbia issuing SLBIA command\n"); | |
44 | cxl_p2n_write(afu, CXL_SLBIA_An, CXL_TLB_SLB_IQ_ALL); | |
45 | while (cxl_p2n_read(afu, CXL_SLBIA_An) & CXL_TLB_SLB_P) { | |
46 | if (time_after_eq(jiffies, timeout)) { | |
47 | dev_warn(&afu->dev, "WARNING: CXL AFU SLBIA timed out!\n"); | |
48 | return -EBUSY; | |
49 | } | |
50 | /* If the adapter has gone down, we can assume that we | |
51 | * will PERST it and that will invalidate everything. | |
52 | */ | |
0d400f77 | 53 | if (!cxl_ops->link_ok(afu->adapter, afu)) |
86331862 CL |
54 | return -EIO; |
55 | cpu_relax(); | |
56 | } | |
57 | return 0; | |
58 | } | |
59 | ||
f204e0b8 IM |
60 | static inline void _cxl_slbia(struct cxl_context *ctx, struct mm_struct *mm) |
61 | { | |
62 | struct task_struct *task; | |
63 | unsigned long flags; | |
64 | if (!(task = get_pid_task(ctx->pid, PIDTYPE_PID))) { | |
65 | pr_devel("%s unable to get task %i\n", | |
66 | __func__, pid_nr(ctx->pid)); | |
67 | return; | |
68 | } | |
69 | ||
70 | if (task->mm != mm) | |
71 | goto out_put; | |
72 | ||
73 | pr_devel("%s matched mm - card: %i afu: %i pe: %i\n", __func__, | |
74 | ctx->afu->adapter->adapter_num, ctx->afu->slice, ctx->pe); | |
75 | ||
76 | spin_lock_irqsave(&ctx->sste_lock, flags); | |
9bcf28cd | 77 | trace_cxl_slbia(ctx); |
f204e0b8 IM |
78 | memset(ctx->sstp, 0, ctx->sst_size); |
79 | spin_unlock_irqrestore(&ctx->sste_lock, flags); | |
80 | mb(); | |
81 | cxl_afu_slbia(ctx->afu); | |
82 | out_put: | |
83 | put_task_struct(task); | |
84 | } | |
85 | ||
86 | static inline void cxl_slbia_core(struct mm_struct *mm) | |
87 | { | |
88 | struct cxl *adapter; | |
89 | struct cxl_afu *afu; | |
90 | struct cxl_context *ctx; | |
91 | int card, slice, id; | |
92 | ||
93 | pr_devel("%s called\n", __func__); | |
94 | ||
95 | spin_lock(&adapter_idr_lock); | |
96 | idr_for_each_entry(&cxl_adapter_idr, adapter, card) { | |
97 | /* XXX: Make this lookup faster with link from mm to ctx */ | |
98 | spin_lock(&adapter->afu_list_lock); | |
99 | for (slice = 0; slice < adapter->slices; slice++) { | |
100 | afu = adapter->afu[slice]; | |
2c069a11 | 101 | if (!afu || !afu->enabled) |
f204e0b8 IM |
102 | continue; |
103 | rcu_read_lock(); | |
104 | idr_for_each_entry(&afu->contexts_idr, ctx, id) | |
105 | _cxl_slbia(ctx, mm); | |
106 | rcu_read_unlock(); | |
107 | } | |
108 | spin_unlock(&adapter->afu_list_lock); | |
109 | } | |
110 | spin_unlock(&adapter_idr_lock); | |
111 | } | |
112 | ||
113 | static struct cxl_calls cxl_calls = { | |
114 | .cxl_slbia = cxl_slbia_core, | |
a19bd79e IM |
115 | .cxl_pci_associate_default_context = _cxl_pci_associate_default_context, |
116 | .cxl_pci_disable_device = _cxl_pci_disable_device, | |
cbce0917 | 117 | .cxl_next_msi_hwirq = _cxl_next_msi_hwirq, |
a2f67d5e IM |
118 | .cxl_cx4_setup_msi_irqs = _cxl_cx4_setup_msi_irqs, |
119 | .cxl_cx4_teardown_msi_irqs = _cxl_cx4_teardown_msi_irqs, | |
f204e0b8 IM |
120 | .owner = THIS_MODULE, |
121 | }; | |
122 | ||
123 | int cxl_alloc_sst(struct cxl_context *ctx) | |
124 | { | |
125 | unsigned long vsid; | |
126 | u64 ea_mask, size, sstp0, sstp1; | |
127 | ||
128 | sstp0 = 0; | |
129 | sstp1 = 0; | |
130 | ||
131 | ctx->sst_size = PAGE_SIZE; | |
132 | ctx->sst_lru = 0; | |
133 | ctx->sstp = (struct cxl_sste *)get_zeroed_page(GFP_KERNEL); | |
134 | if (!ctx->sstp) { | |
135 | pr_err("cxl_alloc_sst: Unable to allocate segment table\n"); | |
136 | return -ENOMEM; | |
137 | } | |
138 | pr_devel("SSTP allocated at 0x%p\n", ctx->sstp); | |
139 | ||
140 | vsid = get_kernel_vsid((u64)ctx->sstp, mmu_kernel_ssize) << 12; | |
141 | ||
142 | sstp0 |= (u64)mmu_kernel_ssize << CXL_SSTP0_An_B_SHIFT; | |
143 | sstp0 |= (SLB_VSID_KERNEL | mmu_psize_defs[mmu_linear_psize].sllp) << 50; | |
144 | ||
145 | size = (((u64)ctx->sst_size >> 8) - 1) << CXL_SSTP0_An_SegTableSize_SHIFT; | |
146 | if (unlikely(size & ~CXL_SSTP0_An_SegTableSize_MASK)) { | |
147 | WARN(1, "Impossible segment table size\n"); | |
148 | return -EINVAL; | |
149 | } | |
150 | sstp0 |= size; | |
151 | ||
152 | if (mmu_kernel_ssize == MMU_SEGSIZE_256M) | |
153 | ea_mask = 0xfffff00ULL; | |
154 | else | |
155 | ea_mask = 0xffffffff00ULL; | |
156 | ||
157 | sstp0 |= vsid >> (50-14); /* Top 14 bits of VSID */ | |
158 | sstp1 |= (vsid << (64-(50-14))) & ~ea_mask; | |
159 | sstp1 |= (u64)ctx->sstp & ea_mask; | |
160 | sstp1 |= CXL_SSTP1_An_V; | |
161 | ||
162 | pr_devel("Looked up %#llx: slbfee. %#llx (ssize: %x, vsid: %#lx), copied to SSTP0: %#llx, SSTP1: %#llx\n", | |
163 | (u64)ctx->sstp, (u64)ctx->sstp & ESID_MASK, mmu_kernel_ssize, vsid, sstp0, sstp1); | |
164 | ||
165 | /* Store calculated sstp hardware points for use later */ | |
166 | ctx->sstp0 = sstp0; | |
167 | ctx->sstp1 = sstp1; | |
168 | ||
169 | return 0; | |
170 | } | |
171 | ||
444c4ba4 CL |
172 | /* print buffer content as integers when debugging */ |
173 | void cxl_dump_debug_buffer(void *buf, size_t buf_len) | |
174 | { | |
175 | #ifdef DEBUG | |
176 | int i, *ptr; | |
177 | ||
178 | /* | |
179 | * We want to regroup up to 4 integers per line, which means they | |
180 | * need to be in the same pr_devel() statement | |
181 | */ | |
182 | ptr = (int *) buf; | |
183 | for (i = 0; i * 4 < buf_len; i += 4) { | |
184 | if ((i + 3) * 4 < buf_len) | |
185 | pr_devel("%.8x %.8x %.8x %.8x\n", ptr[i], ptr[i + 1], | |
186 | ptr[i + 2], ptr[i + 3]); | |
187 | else if ((i + 2) * 4 < buf_len) | |
188 | pr_devel("%.8x %.8x %.8x\n", ptr[i], ptr[i + 1], | |
189 | ptr[i + 2]); | |
190 | else if ((i + 1) * 4 < buf_len) | |
191 | pr_devel("%.8x %.8x\n", ptr[i], ptr[i + 1]); | |
192 | else | |
193 | pr_devel("%.8x\n", ptr[i]); | |
194 | } | |
195 | #endif /* DEBUG */ | |
196 | } | |
197 | ||
f204e0b8 IM |
198 | /* Find a CXL adapter by it's number and increase it's refcount */ |
199 | struct cxl *get_cxl_adapter(int num) | |
200 | { | |
201 | struct cxl *adapter; | |
202 | ||
203 | spin_lock(&adapter_idr_lock); | |
204 | if ((adapter = idr_find(&cxl_adapter_idr, num))) | |
205 | get_device(&adapter->dev); | |
206 | spin_unlock(&adapter_idr_lock); | |
207 | ||
208 | return adapter; | |
209 | } | |
210 | ||
d56d301b | 211 | static int cxl_alloc_adapter_nr(struct cxl *adapter) |
f204e0b8 IM |
212 | { |
213 | int i; | |
214 | ||
215 | idr_preload(GFP_KERNEL); | |
216 | spin_lock(&adapter_idr_lock); | |
217 | i = idr_alloc(&cxl_adapter_idr, adapter, 0, 0, GFP_NOWAIT); | |
218 | spin_unlock(&adapter_idr_lock); | |
219 | idr_preload_end(); | |
220 | if (i < 0) | |
221 | return i; | |
222 | ||
223 | adapter->adapter_num = i; | |
224 | ||
225 | return 0; | |
226 | } | |
227 | ||
228 | void cxl_remove_adapter_nr(struct cxl *adapter) | |
229 | { | |
230 | idr_remove(&cxl_adapter_idr, adapter->adapter_num); | |
231 | } | |
232 | ||
86331862 CL |
233 | struct cxl *cxl_alloc_adapter(void) |
234 | { | |
235 | struct cxl *adapter; | |
236 | ||
237 | if (!(adapter = kzalloc(sizeof(struct cxl), GFP_KERNEL))) | |
238 | return NULL; | |
239 | ||
240 | spin_lock_init(&adapter->afu_list_lock); | |
241 | ||
242 | if (cxl_alloc_adapter_nr(adapter)) | |
243 | goto err1; | |
244 | ||
245 | if (dev_set_name(&adapter->dev, "card%i", adapter->adapter_num)) | |
246 | goto err2; | |
247 | ||
70b565bb VJ |
248 | /* start with context lock taken */ |
249 | atomic_set(&adapter->contexts_num, -1); | |
86331862 | 250 | |
70b565bb | 251 | return adapter; |
86331862 CL |
252 | err2: |
253 | cxl_remove_adapter_nr(adapter); | |
254 | err1: | |
255 | kfree(adapter); | |
256 | return NULL; | |
257 | } | |
258 | ||
259 | struct cxl_afu *cxl_alloc_afu(struct cxl *adapter, int slice) | |
260 | { | |
261 | struct cxl_afu *afu; | |
262 | ||
263 | if (!(afu = kzalloc(sizeof(struct cxl_afu), GFP_KERNEL))) | |
264 | return NULL; | |
265 | ||
266 | afu->adapter = adapter; | |
267 | afu->dev.parent = &adapter->dev; | |
5be587b1 | 268 | afu->dev.release = cxl_ops->release_afu; |
86331862 CL |
269 | afu->slice = slice; |
270 | idr_init(&afu->contexts_idr); | |
271 | mutex_init(&afu->contexts_lock); | |
272 | spin_lock_init(&afu->afu_cntl_lock); | |
171ed0fc | 273 | atomic_set(&afu->configured_state, -1); |
86331862 CL |
274 | afu->prefault_mode = CXL_PREFAULT_NONE; |
275 | afu->irqs_max = afu->adapter->user_irqs; | |
276 | ||
277 | return afu; | |
278 | } | |
279 | ||
f204e0b8 IM |
280 | int cxl_afu_select_best_mode(struct cxl_afu *afu) |
281 | { | |
282 | if (afu->modes_supported & CXL_MODE_DIRECTED) | |
5be587b1 | 283 | return cxl_ops->afu_activate_mode(afu, CXL_MODE_DIRECTED); |
f204e0b8 IM |
284 | |
285 | if (afu->modes_supported & CXL_MODE_DEDICATED) | |
5be587b1 | 286 | return cxl_ops->afu_activate_mode(afu, CXL_MODE_DEDICATED); |
f204e0b8 IM |
287 | |
288 | dev_warn(&afu->dev, "No supported programming modes available\n"); | |
289 | /* We don't fail this so the user can inspect sysfs */ | |
290 | return 0; | |
291 | } | |
292 | ||
70b565bb VJ |
293 | int cxl_adapter_context_get(struct cxl *adapter) |
294 | { | |
295 | int rc; | |
296 | ||
297 | rc = atomic_inc_unless_negative(&adapter->contexts_num); | |
298 | return rc >= 0 ? 0 : -EBUSY; | |
299 | } | |
300 | ||
301 | void cxl_adapter_context_put(struct cxl *adapter) | |
302 | { | |
303 | atomic_dec_if_positive(&adapter->contexts_num); | |
304 | } | |
305 | ||
306 | int cxl_adapter_context_lock(struct cxl *adapter) | |
307 | { | |
308 | int rc; | |
309 | /* no active contexts -> contexts_num == 0 */ | |
310 | rc = atomic_cmpxchg(&adapter->contexts_num, 0, -1); | |
311 | return rc ? -EBUSY : 0; | |
312 | } | |
313 | ||
314 | void cxl_adapter_context_unlock(struct cxl *adapter) | |
315 | { | |
316 | int val = atomic_cmpxchg(&adapter->contexts_num, -1, 0); | |
317 | ||
318 | /* | |
319 | * contexts lock taken -> contexts_num == -1 | |
320 | * If not true then show a warning and force reset the lock. | |
321 | * This will happen when context_unlock was requested without | |
322 | * doing a context_lock. | |
323 | */ | |
324 | if (val != -1) { | |
325 | atomic_set(&adapter->contexts_num, 0); | |
326 | WARN(1, "Adapter context unlocked with %d active contexts", | |
327 | val); | |
328 | } | |
329 | } | |
330 | ||
f204e0b8 IM |
331 | static int __init init_cxl(void) |
332 | { | |
333 | int rc = 0; | |
334 | ||
f204e0b8 IM |
335 | if ((rc = cxl_file_init())) |
336 | return rc; | |
337 | ||
338 | cxl_debugfs_init(); | |
339 | ||
340 | if ((rc = register_cxl_calls(&cxl_calls))) | |
341 | goto err; | |
342 | ||
14baf4d9 CL |
343 | if (cpu_has_feature(CPU_FTR_HVMODE)) { |
344 | cxl_ops = &cxl_native_ops; | |
345 | rc = pci_register_driver(&cxl_pci_driver); | |
346 | } | |
347 | #ifdef CONFIG_PPC_PSERIES | |
348 | else { | |
349 | cxl_ops = &cxl_guest_ops; | |
350 | rc = platform_driver_register(&cxl_of_driver); | |
351 | } | |
352 | #endif | |
353 | if (rc) | |
f204e0b8 IM |
354 | goto err1; |
355 | ||
356 | return 0; | |
357 | err1: | |
358 | unregister_cxl_calls(&cxl_calls); | |
359 | err: | |
360 | cxl_debugfs_exit(); | |
361 | cxl_file_exit(); | |
362 | ||
363 | return rc; | |
364 | } | |
365 | ||
366 | static void exit_cxl(void) | |
367 | { | |
14baf4d9 CL |
368 | if (cpu_has_feature(CPU_FTR_HVMODE)) |
369 | pci_unregister_driver(&cxl_pci_driver); | |
370 | #ifdef CONFIG_PPC_PSERIES | |
371 | else | |
372 | platform_driver_unregister(&cxl_of_driver); | |
373 | #endif | |
f204e0b8 IM |
374 | |
375 | cxl_debugfs_exit(); | |
376 | cxl_file_exit(); | |
377 | unregister_cxl_calls(&cxl_calls); | |
b2a02ac6 | 378 | idr_destroy(&cxl_adapter_idr); |
f204e0b8 IM |
379 | } |
380 | ||
381 | module_init(init_cxl); | |
382 | module_exit(exit_cxl); | |
383 | ||
384 | MODULE_DESCRIPTION("IBM Coherent Accelerator"); | |
385 | MODULE_AUTHOR("Ian Munsie <imunsie@au1.ibm.com>"); | |
386 | MODULE_LICENSE("GPL"); |