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