misc/cxl/main.c

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  * Copyright 2014 IBM Corp.
0004  */
0005
0006 #include <linux/spinlock.h>
0007 #include <linux/kernel.h>
0008 #include <linux/module.h>
0009 #include <linux/device.h>
0010 #include <linux/mutex.h>
0011 #include <linux/init.h>
0012 #include <linux/list.h>
0013 #include <linux/mm.h>
0014 #include <linux/of.h>
0015 #include <linux/slab.h>
0016 #include <linux/idr.h>
0017 #include <linux/pci.h>
0018 #include <linux/platform_device.h>
0019 #include <linux/sched/task.h>
0020
0021 #include <asm/cputable.h>
0022 #include <asm/mmu.h>
0023 #include <misc/cxl-base.h>
0024
0025 #include "cxl.h"
0026 #include "trace.h"
0027
0028 static DEFINE_SPINLOCK(adapter_idr_lock);
0029 static DEFINE_IDR(cxl_adapter_idr);
0030
0031 uint cxl_verbose;
0032 module_param_named(verbose, cxl_verbose, uint, 0600);
0033 MODULE_PARM_DESC(verbose, "Enable verbose dmesg output");
0034
0035 const struct cxl_backend_ops *cxl_ops;
0036
0037 int cxl_afu_slbia(struct cxl_afu *afu)
0038 {
0039     unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
0040
0041     pr_devel("cxl_afu_slbia issuing SLBIA command\n");
0042     cxl_p2n_write(afu, CXL_SLBIA_An, CXL_TLB_SLB_IQ_ALL);
0043     while (cxl_p2n_read(afu, CXL_SLBIA_An) & CXL_TLB_SLB_P) {
0044         if (time_after_eq(jiffies, timeout)) {
0045             dev_warn(&afu->dev, "WARNING: CXL AFU SLBIA timed out!\n");
0046             return -EBUSY;
0047         }
0048         /* If the adapter has gone down, we can assume that we
0049          * will PERST it and that will invalidate everything.
0050          */
0051         if (!cxl_ops->link_ok(afu->adapter, afu))
0052             return -EIO;
0053         cpu_relax();
0054     }
0055     return 0;
0056 }
0057
0058 static inline void _cxl_slbia(struct cxl_context *ctx, struct mm_struct *mm)
0059 {
0060     unsigned long flags;
0061
0062     if (ctx->mm != mm)
0063         return;
0064
0065     pr_devel("%s matched mm - card: %i afu: %i pe: %i\n", __func__,
0066          ctx->afu->adapter->adapter_num, ctx->afu->slice, ctx->pe);
0067
0068     spin_lock_irqsave(&ctx->sste_lock, flags);
0069     trace_cxl_slbia(ctx);
0070     memset(ctx->sstp, 0, ctx->sst_size);
0071     spin_unlock_irqrestore(&ctx->sste_lock, flags);
0072     mb();
0073     cxl_afu_slbia(ctx->afu);
0074 }
0075
0076 static inline void cxl_slbia_core(struct mm_struct *mm)
0077 {
0078     struct cxl *adapter;
0079     struct cxl_afu *afu;
0080     struct cxl_context *ctx;
0081     int card, slice, id;
0082
0083     pr_devel("%s called\n", __func__);
0084
0085     spin_lock(&adapter_idr_lock);
0086     idr_for_each_entry(&cxl_adapter_idr, adapter, card) {
0087         /* XXX: Make this lookup faster with link from mm to ctx */
0088         spin_lock(&adapter->afu_list_lock);
0089         for (slice = 0; slice < adapter->slices; slice++) {
0090             afu = adapter->afu[slice];
0091             if (!afu || !afu->enabled)
0092                 continue;
0093             rcu_read_lock();
0094             idr_for_each_entry(&afu->contexts_idr, ctx, id)
0095                 _cxl_slbia(ctx, mm);
0096             rcu_read_unlock();
0097         }
0098         spin_unlock(&adapter->afu_list_lock);
0099     }
0100     spin_unlock(&adapter_idr_lock);
0101 }
0102
0103 static struct cxl_calls cxl_calls = {
0104     .cxl_slbia = cxl_slbia_core,
0105     .owner = THIS_MODULE,
0106 };
0107
0108 int cxl_alloc_sst(struct cxl_context *ctx)
0109 {
0110     unsigned long vsid;
0111     u64 ea_mask, size, sstp0, sstp1;
0112
0113     sstp0 = 0;
0114     sstp1 = 0;
0115
0116     ctx->sst_size = PAGE_SIZE;
0117     ctx->sst_lru = 0;
0118     ctx->sstp = (struct cxl_sste *)get_zeroed_page(GFP_KERNEL);
0119     if (!ctx->sstp) {
0120         pr_err("cxl_alloc_sst: Unable to allocate segment table\n");
0121         return -ENOMEM;
0122     }
0123     pr_devel("SSTP allocated at 0x%p\n", ctx->sstp);
0124
0125     vsid  = get_kernel_vsid((u64)ctx->sstp, mmu_kernel_ssize) << 12;
0126
0127     sstp0 |= (u64)mmu_kernel_ssize << CXL_SSTP0_An_B_SHIFT;
0128     sstp0 |= (SLB_VSID_KERNEL | mmu_psize_defs[mmu_linear_psize].sllp) << 50;
0129
0130     size = (((u64)ctx->sst_size >> 8) - 1) << CXL_SSTP0_An_SegTableSize_SHIFT;
0131     if (unlikely(size & ~CXL_SSTP0_An_SegTableSize_MASK)) {
0132         WARN(1, "Impossible segment table size\n");
0133         return -EINVAL;
0134     }
0135     sstp0 |= size;
0136
0137     if (mmu_kernel_ssize == MMU_SEGSIZE_256M)
0138         ea_mask = 0xfffff00ULL;
0139     else
0140         ea_mask = 0xffffffff00ULL;
0141
0142     sstp0 |=  vsid >>     (50-14);  /*   Top 14 bits of VSID */
0143     sstp1 |= (vsid << (64-(50-14))) & ~ea_mask;
0144     sstp1 |= (u64)ctx->sstp & ea_mask;
0145     sstp1 |= CXL_SSTP1_An_V;
0146
0147     pr_devel("Looked up %#llx: slbfee. %#llx (ssize: %x, vsid: %#lx), copied to SSTP0: %#llx, SSTP1: %#llx\n",
0148             (u64)ctx->sstp, (u64)ctx->sstp & ESID_MASK, mmu_kernel_ssize, vsid, sstp0, sstp1);
0149
0150     /* Store calculated sstp hardware points for use later */
0151     ctx->sstp0 = sstp0;
0152     ctx->sstp1 = sstp1;
0153
0154     return 0;
0155 }
0156
0157 /* print buffer content as integers when debugging */
0158 void cxl_dump_debug_buffer(void *buf, size_t buf_len)
0159 {
0160 #ifdef DEBUG
0161     int i, *ptr;
0162
0163     /*
0164      * We want to regroup up to 4 integers per line, which means they
0165      * need to be in the same pr_devel() statement
0166      */
0167     ptr = (int *) buf;
0168     for (i = 0; i * 4 < buf_len; i += 4) {
0169         if ((i + 3) * 4 < buf_len)
0170             pr_devel("%.8x %.8x %.8x %.8x\n", ptr[i], ptr[i + 1],
0171                 ptr[i + 2], ptr[i + 3]);
0172         else if ((i + 2) * 4 < buf_len)
0173             pr_devel("%.8x %.8x %.8x\n", ptr[i], ptr[i + 1],
0174                 ptr[i + 2]);
0175         else if ((i + 1) * 4 < buf_len)
0176             pr_devel("%.8x %.8x\n", ptr[i], ptr[i + 1]);
0177         else
0178             pr_devel("%.8x\n", ptr[i]);
0179     }
0180 #endif /* DEBUG */
0181 }
0182
0183 /* Find a CXL adapter by it's number and increase it's refcount */
0184 struct cxl *get_cxl_adapter(int num)
0185 {
0186     struct cxl *adapter;
0187
0188     spin_lock(&adapter_idr_lock);
0189     if ((adapter = idr_find(&cxl_adapter_idr, num)))
0190         get_device(&adapter->dev);
0191     spin_unlock(&adapter_idr_lock);
0192
0193     return adapter;
0194 }
0195
0196 static int cxl_alloc_adapter_nr(struct cxl *adapter)
0197 {
0198     int i;
0199
0200     idr_preload(GFP_KERNEL);
0201     spin_lock(&adapter_idr_lock);
0202     i = idr_alloc(&cxl_adapter_idr, adapter, 0, 0, GFP_NOWAIT);
0203     spin_unlock(&adapter_idr_lock);
0204     idr_preload_end();
0205     if (i < 0)
0206         return i;
0207
0208     adapter->adapter_num = i;
0209
0210     return 0;
0211 }
0212
0213 void cxl_remove_adapter_nr(struct cxl *adapter)
0214 {
0215     idr_remove(&cxl_adapter_idr, adapter->adapter_num);
0216 }
0217
0218 struct cxl *cxl_alloc_adapter(void)
0219 {
0220     struct cxl *adapter;
0221
0222     if (!(adapter = kzalloc(sizeof(struct cxl), GFP_KERNEL)))
0223         return NULL;
0224
0225     spin_lock_init(&adapter->afu_list_lock);
0226
0227     if (cxl_alloc_adapter_nr(adapter))
0228         goto err1;
0229
0230     if (dev_set_name(&adapter->dev, "card%i", adapter->adapter_num))
0231         goto err2;
0232
0233     /* start with context lock taken */
0234     atomic_set(&adapter->contexts_num, -1);
0235
0236     return adapter;
0237 err2:
0238     cxl_remove_adapter_nr(adapter);
0239 err1:
0240     kfree(adapter);
0241     return NULL;
0242 }
0243
0244 struct cxl_afu *cxl_alloc_afu(struct cxl *adapter, int slice)
0245 {
0246     struct cxl_afu *afu;
0247
0248     if (!(afu = kzalloc(sizeof(struct cxl_afu), GFP_KERNEL)))
0249         return NULL;
0250
0251     afu->adapter = adapter;
0252     afu->dev.parent = &adapter->dev;
0253     afu->dev.release = cxl_ops->release_afu;
0254     afu->slice = slice;
0255     idr_init(&afu->contexts_idr);
0256     mutex_init(&afu->contexts_lock);
0257     spin_lock_init(&afu->afu_cntl_lock);
0258     atomic_set(&afu->configured_state, -1);
0259     afu->prefault_mode = CXL_PREFAULT_NONE;
0260     afu->irqs_max = afu->adapter->user_irqs;
0261
0262     return afu;
0263 }
0264
0265 int cxl_afu_select_best_mode(struct cxl_afu *afu)
0266 {
0267     if (afu->modes_supported & CXL_MODE_DIRECTED)
0268         return cxl_ops->afu_activate_mode(afu, CXL_MODE_DIRECTED);
0269
0270     if (afu->modes_supported & CXL_MODE_DEDICATED)
0271         return cxl_ops->afu_activate_mode(afu, CXL_MODE_DEDICATED);
0272
0273     dev_warn(&afu->dev, "No supported programming modes available\n");
0274     /* We don't fail this so the user can inspect sysfs */
0275     return 0;
0276 }
0277
0278 int cxl_adapter_context_get(struct cxl *adapter)
0279 {
0280     int rc;
0281
0282     rc = atomic_inc_unless_negative(&adapter->contexts_num);
0283     return rc ? 0 : -EBUSY;
0284 }
0285
0286 void cxl_adapter_context_put(struct cxl *adapter)
0287 {
0288     atomic_dec_if_positive(&adapter->contexts_num);
0289 }
0290
0291 int cxl_adapter_context_lock(struct cxl *adapter)
0292 {
0293     int rc;
0294     /* no active contexts -> contexts_num == 0 */
0295     rc = atomic_cmpxchg(&adapter->contexts_num, 0, -1);
0296     return rc ? -EBUSY : 0;
0297 }
0298
0299 void cxl_adapter_context_unlock(struct cxl *adapter)
0300 {
0301     int val = atomic_cmpxchg(&adapter->contexts_num, -1, 0);
0302
0303     /*
0304      * contexts lock taken -> contexts_num == -1
0305      * If not true then show a warning and force reset the lock.
0306      * This will happen when context_unlock was requested without
0307      * doing a context_lock.
0308      */
0309     if (val != -1) {
0310         atomic_set(&adapter->contexts_num, 0);
0311         WARN(1, "Adapter context unlocked with %d active contexts",
0312              val);
0313     }
0314 }
0315
0316 static int __init init_cxl(void)
0317 {
0318     int rc = 0;
0319
0320     if (!tlbie_capable)
0321         return -EINVAL;
0322
0323     if ((rc = cxl_file_init()))
0324         return rc;
0325
0326     cxl_debugfs_init();
0327
0328     /*
0329      * we don't register the callback on P9. slb callack is only
0330      * used for the PSL8 MMU and CX4.
0331      */
0332     if (cxl_is_power8()) {
0333         rc = register_cxl_calls(&cxl_calls);
0334         if (rc)
0335             goto err;
0336     }
0337
0338     if (cpu_has_feature(CPU_FTR_HVMODE)) {
0339         cxl_ops = &cxl_native_ops;
0340         rc = pci_register_driver(&cxl_pci_driver);
0341     }
0342 #ifdef CONFIG_PPC_PSERIES
0343     else {
0344         cxl_ops = &cxl_guest_ops;
0345         rc = platform_driver_register(&cxl_of_driver);
0346     }
0347 #endif
0348     if (rc)
0349         goto err1;
0350
0351     return 0;
0352 err1:
0353     if (cxl_is_power8())
0354         unregister_cxl_calls(&cxl_calls);
0355 err:
0356     cxl_debugfs_exit();
0357     cxl_file_exit();
0358
0359     return rc;
0360 }
0361
0362 static void exit_cxl(void)
0363 {
0364     if (cpu_has_feature(CPU_FTR_HVMODE))
0365         pci_unregister_driver(&cxl_pci_driver);
0366 #ifdef CONFIG_PPC_PSERIES
0367     else
0368         platform_driver_unregister(&cxl_of_driver);
0369 #endif
0370
0371     cxl_debugfs_exit();
0372     cxl_file_exit();
0373     if (cxl_is_power8())
0374         unregister_cxl_calls(&cxl_calls);
0375     idr_destroy(&cxl_adapter_idr);
0376 }
0377
0378 module_init(init_cxl);
0379 module_exit(exit_cxl);
0380
0381 MODULE_DESCRIPTION("IBM Coherent Accelerator");
0382 MODULE_AUTHOR("Ian Munsie <imunsie@au1.ibm.com>");
0383 MODULE_LICENSE("GPL");