drivers/misc/cxl/main.c

   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>
  22 #include <linux/sched/task.h>
  23
  24 #include <asm/cputable.h>
  25 #include <misc/cxl-base.h>
  26
  27 #include "cxl.h"
  28 #include "trace.h"
  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
  37 const struct cxl_backend_ops *cxl_ops;
  38
  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                  */
  53                 if (!cxl_ops->link_ok(afu->adapter, afu))
  54                         return -EIO;
  55                 cpu_relax();
  56         }
  57         return 0;
  58 }
  59
  60 static inline void _cxl_slbia(struct cxl_context *ctx, struct mm_struct *mm)
  61 {
  62         unsigned long flags;
  63
  64         if (ctx->mm != mm)
  65                 return;
  66
  67         pr_devel("%s matched mm - card: %i afu: %i pe: %i\n", __func__,
  68                  ctx->afu->adapter->adapter_num, ctx->afu->slice, ctx->pe);
  69
  70         spin_lock_irqsave(&ctx->sste_lock, flags);
  71         trace_cxl_slbia(ctx);
  72         memset(ctx->sstp, 0, ctx->sst_size);
  73         spin_unlock_irqrestore(&ctx->sste_lock, flags);
  74         mb();
  75         cxl_afu_slbia(ctx->afu);
  76 }
  77
  78 static inline void cxl_slbia_core(struct mm_struct *mm)
  79 {
  80         struct cxl *adapter;
  81         struct cxl_afu *afu;
  82         struct cxl_context *ctx;
  83         int card, slice, id;
  84
  85         pr_devel("%s called\n", __func__);
  86
  87         spin_lock(&adapter_idr_lock);
  88         idr_for_each_entry(&cxl_adapter_idr, adapter, card) {
  89                 /* XXX: Make this lookup faster with link from mm to ctx */
  90                 spin_lock(&adapter->afu_list_lock);
  91                 for (slice = 0; slice < adapter->slices; slice++) {
  92                         afu = adapter->afu[slice];
  93                         if (!afu || !afu->enabled)
  94                                 continue;
  95                         rcu_read_lock();
  96                         idr_for_each_entry(&afu->contexts_idr, ctx, id)
  97                                 _cxl_slbia(ctx, mm);
  98                         rcu_read_unlock();
  99                 }
 100                 spin_unlock(&adapter->afu_list_lock);
 101         }
 102         spin_unlock(&adapter_idr_lock);
 103 }
 104
 105 static struct cxl_calls cxl_calls = {
 106         .cxl_slbia = cxl_slbia_core,
 107         .cxl_pci_associate_default_context = _cxl_pci_associate_default_context,
 108         .cxl_pci_disable_device = _cxl_pci_disable_device,
 109         .cxl_next_msi_hwirq = _cxl_next_msi_hwirq,
 110         .cxl_cx4_setup_msi_irqs = _cxl_cx4_setup_msi_irqs,
 111         .cxl_cx4_teardown_msi_irqs = _cxl_cx4_teardown_msi_irqs,
 112         .owner = THIS_MODULE,
 113 };
 114
 115 int cxl_alloc_sst(struct cxl_context *ctx)
 116 {
 117         unsigned long vsid;
 118         u64 ea_mask, size, sstp0, sstp1;
 119
 120         sstp0 = 0;
 121         sstp1 = 0;
 122
 123         ctx->sst_size = PAGE_SIZE;
 124         ctx->sst_lru = 0;
 125         ctx->sstp = (struct cxl_sste *)get_zeroed_page(GFP_KERNEL);
 126         if (!ctx->sstp) {
 127                 pr_err("cxl_alloc_sst: Unable to allocate segment table\n");
 128                 return -ENOMEM;
 129         }
 130         pr_devel("SSTP allocated at 0x%p\n", ctx->sstp);
 131
 132         vsid  = get_kernel_vsid((u64)ctx->sstp, mmu_kernel_ssize) << 12;
 133
 134         sstp0 |= (u64)mmu_kernel_ssize << CXL_SSTP0_An_B_SHIFT;
 135         sstp0 |= (SLB_VSID_KERNEL | mmu_psize_defs[mmu_linear_psize].sllp) << 50;
 136
 137         size = (((u64)ctx->sst_size >> 8) - 1) << CXL_SSTP0_An_SegTableSize_SHIFT;
 138         if (unlikely(size & ~CXL_SSTP0_An_SegTableSize_MASK)) {
 139                 WARN(1, "Impossible segment table size\n");
 140                 return -EINVAL;
 141         }
 142         sstp0 |= size;
 143
 144         if (mmu_kernel_ssize == MMU_SEGSIZE_256M)
 145                 ea_mask = 0xfffff00ULL;
 146         else
 147                 ea_mask = 0xffffffff00ULL;
 148
 149         sstp0 |=  vsid >>     (50-14);  /*   Top 14 bits of VSID */
 150         sstp1 |= (vsid << (64-(50-14))) & ~ea_mask;
 151         sstp1 |= (u64)ctx->sstp & ea_mask;
 152         sstp1 |= CXL_SSTP1_An_V;
 153
 154         pr_devel("Looked up %#llx: slbfee. %#llx (ssize: %x, vsid: %#lx), copied to SSTP0: %#llx, SSTP1: %#llx\n",
 155                         (u64)ctx->sstp, (u64)ctx->sstp & ESID_MASK, mmu_kernel_ssize, vsid, sstp0, sstp1);
 156
 157         /* Store calculated sstp hardware points for use later */
 158         ctx->sstp0 = sstp0;
 159         ctx->sstp1 = sstp1;
 160
 161         return 0;
 162 }
 163
 164 /* print buffer content as integers when debugging */
 165 void cxl_dump_debug_buffer(void *buf, size_t buf_len)
 166 {
 167 #ifdef DEBUG
 168         int i, *ptr;
 169
 170         /*
 171          * We want to regroup up to 4 integers per line, which means they
 172          * need to be in the same pr_devel() statement
 173          */
 174         ptr = (int *) buf;
 175         for (i = 0; i * 4 < buf_len; i += 4) {
 176                 if ((i + 3) * 4 < buf_len)
 177                         pr_devel("%.8x %.8x %.8x %.8x\n", ptr[i], ptr[i + 1],
 178                                 ptr[i + 2], ptr[i + 3]);
 179                 else if ((i + 2) * 4 < buf_len)
 180                         pr_devel("%.8x %.8x %.8x\n", ptr[i], ptr[i + 1],
 181                                 ptr[i + 2]);
 182                 else if ((i + 1) * 4 < buf_len)
 183                         pr_devel("%.8x %.8x\n", ptr[i], ptr[i + 1]);
 184                 else
 185                         pr_devel("%.8x\n", ptr[i]);
 186         }
 187 #endif /* DEBUG */
 188 }
 189
 190 /* Find a CXL adapter by it's number and increase it's refcount */
 191 struct cxl *get_cxl_adapter(int num)
 192 {
 193         struct cxl *adapter;
 194
 195         spin_lock(&adapter_idr_lock);
 196         if ((adapter = idr_find(&cxl_adapter_idr, num)))
 197                 get_device(&adapter->dev);
 198         spin_unlock(&adapter_idr_lock);
 199
 200         return adapter;
 201 }
 202
 203 static int cxl_alloc_adapter_nr(struct cxl *adapter)
 204 {
 205         int i;
 206
 207         idr_preload(GFP_KERNEL);
 208         spin_lock(&adapter_idr_lock);
 209         i = idr_alloc(&cxl_adapter_idr, adapter, 0, 0, GFP_NOWAIT);
 210         spin_unlock(&adapter_idr_lock);
 211         idr_preload_end();
 212         if (i < 0)
 213                 return i;
 214
 215         adapter->adapter_num = i;
 216
 217         return 0;
 218 }
 219
 220 void cxl_remove_adapter_nr(struct cxl *adapter)
 221 {
 222         idr_remove(&cxl_adapter_idr, adapter->adapter_num);
 223 }
 224
 225 struct cxl *cxl_alloc_adapter(void)
 226 {
 227         struct cxl *adapter;
 228
 229         if (!(adapter = kzalloc(sizeof(struct cxl), GFP_KERNEL)))
 230                 return NULL;
 231
 232         spin_lock_init(&adapter->afu_list_lock);
 233
 234         if (cxl_alloc_adapter_nr(adapter))
 235                 goto err1;
 236
 237         if (dev_set_name(&adapter->dev, "card%i", adapter->adapter_num))
 238                 goto err2;
 239
 240         /* start with context lock taken */
 241         atomic_set(&adapter->contexts_num, -1);
 242
 243         return adapter;
 244 err2:
 245         cxl_remove_adapter_nr(adapter);
 246 err1:
 247         kfree(adapter);
 248         return NULL;
 249 }
 250
 251 struct cxl_afu *cxl_alloc_afu(struct cxl *adapter, int slice)
 252 {
 253         struct cxl_afu *afu;
 254
 255         if (!(afu = kzalloc(sizeof(struct cxl_afu), GFP_KERNEL)))
 256                 return NULL;
 257
 258         afu->adapter = adapter;
 259         afu->dev.parent = &adapter->dev;
 260         afu->dev.release = cxl_ops->release_afu;
 261         afu->slice = slice;
 262         idr_init(&afu->contexts_idr);
 263         mutex_init(&afu->contexts_lock);
 264         spin_lock_init(&afu->afu_cntl_lock);
 265         atomic_set(&afu->configured_state, -1);
 266         afu->prefault_mode = CXL_PREFAULT_NONE;
 267         afu->irqs_max = afu->adapter->user_irqs;
 268
 269         return afu;
 270 }
 271
 272 int cxl_afu_select_best_mode(struct cxl_afu *afu)
 273 {
 274         if (afu->modes_supported & CXL_MODE_DIRECTED)
 275                 return cxl_ops->afu_activate_mode(afu, CXL_MODE_DIRECTED);
 276
 277         if (afu->modes_supported & CXL_MODE_DEDICATED)
 278                 return cxl_ops->afu_activate_mode(afu, CXL_MODE_DEDICATED);
 279
 280         dev_warn(&afu->dev, "No supported programming modes available\n");
 281         /* We don't fail this so the user can inspect sysfs */
 282         return 0;
 283 }
 284
 285 int cxl_adapter_context_get(struct cxl *adapter)
 286 {
 287         int rc;
 288
 289         rc = atomic_inc_unless_negative(&adapter->contexts_num);
 290         return rc >= 0 ? 0 : -EBUSY;
 291 }
 292
 293 void cxl_adapter_context_put(struct cxl *adapter)
 294 {
 295         atomic_dec_if_positive(&adapter->contexts_num);
 296 }
 297
 298 int cxl_adapter_context_lock(struct cxl *adapter)
 299 {
 300         int rc;
 301         /* no active contexts -> contexts_num == 0 */
 302         rc = atomic_cmpxchg(&adapter->contexts_num, 0, -1);
 303         return rc ? -EBUSY : 0;
 304 }
 305
 306 void cxl_adapter_context_unlock(struct cxl *adapter)
 307 {
 308         int val = atomic_cmpxchg(&adapter->contexts_num, -1, 0);
 309
 310         /*
 311          * contexts lock taken -> contexts_num == -1
 312          * If not true then show a warning and force reset the lock.
 313          * This will happen when context_unlock was requested without
 314          * doing a context_lock.
 315          */
 316         if (val != -1) {
 317                 atomic_set(&adapter->contexts_num, 0);
 318                 WARN(1, "Adapter context unlocked with %d active contexts",
 319                      val);
 320         }
 321 }
 322
 323 static int __init init_cxl(void)
 324 {
 325         int rc = 0;
 326
 327         if ((rc = cxl_file_init()))
 328                 return rc;
 329
 330         cxl_debugfs_init();
 331
 332         /*
 333          * we don't register the callback on P9. slb callack is only
 334          * used for the PSL8 MMU and CX4.
 335          */
 336         if (cxl_is_power8()) {
 337                 rc = register_cxl_calls(&cxl_calls);
 338                 if (rc)
 339                         goto err;
 340         }
 341
 342         if (cpu_has_feature(CPU_FTR_HVMODE)) {
 343                 cxl_ops = &cxl_native_ops;
 344                 rc = pci_register_driver(&cxl_pci_driver);
 345         }
 346 #ifdef CONFIG_PPC_PSERIES
 347         else {
 348                 cxl_ops = &cxl_guest_ops;
 349                 rc = platform_driver_register(&cxl_of_driver);
 350         }
 351 #endif
 352         if (rc)
 353                 goto err1;
 354
 355         return 0;
 356 err1:
 357         if (cxl_is_power8())
 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 {
 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
 374
 375         cxl_debugfs_exit();
 376         cxl_file_exit();
 377         if (cxl_is_power8())
 378                 unregister_cxl_calls(&cxl_calls);
 379         idr_destroy(&cxl_adapter_idr);
 380 }
 381
 382 module_init(init_cxl);
 383 module_exit(exit_cxl);
 384
 385 MODULE_DESCRIPTION("IBM Coherent Accelerator");
 386 MODULE_AUTHOR("Ian Munsie <imunsie@au1.ibm.com>");
 387 MODULE_LICENSE("GPL");