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 <asm/cputable.h> 23 #include <misc/cxl.h> 24 25 #include "cxl.h" 26 27 static DEFINE_SPINLOCK(adapter_idr_lock); 28 static DEFINE_IDR(cxl_adapter_idr); 29 30 uint cxl_verbose; 31 module_param_named(verbose, cxl_verbose, uint, 0600); 32 MODULE_PARM_DESC(verbose, "Enable verbose dmesg output"); 33 34 static inline void _cxl_slbia(struct cxl_context *ctx, struct mm_struct *mm) 35 { 36 struct task_struct *task; 37 unsigned long flags; 38 if (!(task = get_pid_task(ctx->pid, PIDTYPE_PID))) { 39 pr_devel("%s unable to get task %i\n", 40 __func__, pid_nr(ctx->pid)); 41 return; 42 } 43 44 if (task->mm != mm) 45 goto out_put; 46 47 pr_devel("%s matched mm - card: %i afu: %i pe: %i\n", __func__, 48 ctx->afu->adapter->adapter_num, ctx->afu->slice, ctx->pe); 49 50 spin_lock_irqsave(&ctx->sste_lock, flags); 51 memset(ctx->sstp, 0, ctx->sst_size); 52 spin_unlock_irqrestore(&ctx->sste_lock, flags); 53 mb(); 54 cxl_afu_slbia(ctx->afu); 55 out_put: 56 put_task_struct(task); 57 } 58 59 static inline void cxl_slbia_core(struct mm_struct *mm) 60 { 61 struct cxl *adapter; 62 struct cxl_afu *afu; 63 struct cxl_context *ctx; 64 int card, slice, id; 65 66 pr_devel("%s called\n", __func__); 67 68 spin_lock(&adapter_idr_lock); 69 idr_for_each_entry(&cxl_adapter_idr, adapter, card) { 70 /* XXX: Make this lookup faster with link from mm to ctx */ 71 spin_lock(&adapter->afu_list_lock); 72 for (slice = 0; slice < adapter->slices; slice++) { 73 afu = adapter->afu[slice]; 74 if (!afu->enabled) 75 continue; 76 rcu_read_lock(); 77 idr_for_each_entry(&afu->contexts_idr, ctx, id) 78 _cxl_slbia(ctx, mm); 79 rcu_read_unlock(); 80 } 81 spin_unlock(&adapter->afu_list_lock); 82 } 83 spin_unlock(&adapter_idr_lock); 84 } 85 86 static struct cxl_calls cxl_calls = { 87 .cxl_slbia = cxl_slbia_core, 88 .owner = THIS_MODULE, 89 }; 90 91 int cxl_alloc_sst(struct cxl_context *ctx) 92 { 93 unsigned long vsid; 94 u64 ea_mask, size, sstp0, sstp1; 95 96 sstp0 = 0; 97 sstp1 = 0; 98 99 ctx->sst_size = PAGE_SIZE; 100 ctx->sst_lru = 0; 101 ctx->sstp = (struct cxl_sste *)get_zeroed_page(GFP_KERNEL); 102 if (!ctx->sstp) { 103 pr_err("cxl_alloc_sst: Unable to allocate segment table\n"); 104 return -ENOMEM; 105 } 106 pr_devel("SSTP allocated at 0x%p\n", ctx->sstp); 107 108 vsid = get_kernel_vsid((u64)ctx->sstp, mmu_kernel_ssize) << 12; 109 110 sstp0 |= (u64)mmu_kernel_ssize << CXL_SSTP0_An_B_SHIFT; 111 sstp0 |= (SLB_VSID_KERNEL | mmu_psize_defs[mmu_linear_psize].sllp) << 50; 112 113 size = (((u64)ctx->sst_size >> 8) - 1) << CXL_SSTP0_An_SegTableSize_SHIFT; 114 if (unlikely(size & ~CXL_SSTP0_An_SegTableSize_MASK)) { 115 WARN(1, "Impossible segment table size\n"); 116 return -EINVAL; 117 } 118 sstp0 |= size; 119 120 if (mmu_kernel_ssize == MMU_SEGSIZE_256M) 121 ea_mask = 0xfffff00ULL; 122 else 123 ea_mask = 0xffffffff00ULL; 124 125 sstp0 |= vsid >> (50-14); /* Top 14 bits of VSID */ 126 sstp1 |= (vsid << (64-(50-14))) & ~ea_mask; 127 sstp1 |= (u64)ctx->sstp & ea_mask; 128 sstp1 |= CXL_SSTP1_An_V; 129 130 pr_devel("Looked up %#llx: slbfee. %#llx (ssize: %x, vsid: %#lx), copied to SSTP0: %#llx, SSTP1: %#llx\n", 131 (u64)ctx->sstp, (u64)ctx->sstp & ESID_MASK, mmu_kernel_ssize, vsid, sstp0, sstp1); 132 133 /* Store calculated sstp hardware points for use later */ 134 ctx->sstp0 = sstp0; 135 ctx->sstp1 = sstp1; 136 137 return 0; 138 } 139 140 /* Find a CXL adapter by it's number and increase it's refcount */ 141 struct cxl *get_cxl_adapter(int num) 142 { 143 struct cxl *adapter; 144 145 spin_lock(&adapter_idr_lock); 146 if ((adapter = idr_find(&cxl_adapter_idr, num))) 147 get_device(&adapter->dev); 148 spin_unlock(&adapter_idr_lock); 149 150 return adapter; 151 } 152 153 int cxl_alloc_adapter_nr(struct cxl *adapter) 154 { 155 int i; 156 157 idr_preload(GFP_KERNEL); 158 spin_lock(&adapter_idr_lock); 159 i = idr_alloc(&cxl_adapter_idr, adapter, 0, 0, GFP_NOWAIT); 160 spin_unlock(&adapter_idr_lock); 161 idr_preload_end(); 162 if (i < 0) 163 return i; 164 165 adapter->adapter_num = i; 166 167 return 0; 168 } 169 170 void cxl_remove_adapter_nr(struct cxl *adapter) 171 { 172 idr_remove(&cxl_adapter_idr, adapter->adapter_num); 173 } 174 175 int cxl_afu_select_best_mode(struct cxl_afu *afu) 176 { 177 if (afu->modes_supported & CXL_MODE_DIRECTED) 178 return cxl_afu_activate_mode(afu, CXL_MODE_DIRECTED); 179 180 if (afu->modes_supported & CXL_MODE_DEDICATED) 181 return cxl_afu_activate_mode(afu, CXL_MODE_DEDICATED); 182 183 dev_warn(&afu->dev, "No supported programming modes available\n"); 184 /* We don't fail this so the user can inspect sysfs */ 185 return 0; 186 } 187 188 static int __init init_cxl(void) 189 { 190 int rc = 0; 191 192 if (!cpu_has_feature(CPU_FTR_HVMODE)) 193 return -EPERM; 194 195 if ((rc = cxl_file_init())) 196 return rc; 197 198 cxl_debugfs_init(); 199 200 if ((rc = register_cxl_calls(&cxl_calls))) 201 goto err; 202 203 if ((rc = pci_register_driver(&cxl_pci_driver))) 204 goto err1; 205 206 return 0; 207 err1: 208 unregister_cxl_calls(&cxl_calls); 209 err: 210 cxl_debugfs_exit(); 211 cxl_file_exit(); 212 213 return rc; 214 } 215 216 static void exit_cxl(void) 217 { 218 pci_unregister_driver(&cxl_pci_driver); 219 220 cxl_debugfs_exit(); 221 cxl_file_exit(); 222 unregister_cxl_calls(&cxl_calls); 223 } 224 225 module_init(init_cxl); 226 module_exit(exit_cxl); 227 228 MODULE_DESCRIPTION("IBM Coherent Accelerator"); 229 MODULE_AUTHOR("Ian Munsie <imunsie@au1.ibm.com>"); 230 MODULE_LICENSE("GPL"); 231