| 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
| 2 | /* |
| 3 | * AMD Node helper functions and common defines |
| 4 | * |
| 5 | * Copyright (c) 2024, Advanced Micro Devices, Inc. |
| 6 | * All Rights Reserved. |
| 7 | * |
| 8 | * Author: Yazen Ghannam <Yazen.Ghannam@amd.com> |
| 9 | */ |
| 10 | |
| 11 | #include <linux/debugfs.h> |
| 12 | #include <asm/amd/node.h> |
| 13 | |
| 14 | /* |
| 15 | * AMD Nodes are a physical collection of I/O devices within an SoC. There can be one |
| 16 | * or more nodes per package. |
| 17 | * |
| 18 | * The nodes are software-visible through PCI config space. All nodes are enumerated |
| 19 | * on segment 0 bus 0. The device (slot) numbers range from 0x18 to 0x1F (maximum 8 |
| 20 | * nodes) with 0x18 corresponding to node 0, 0x19 to node 1, etc. Each node can be a |
| 21 | * multi-function device. |
| 22 | * |
| 23 | * On legacy systems, these node devices represent integrated Northbridge functionality. |
| 24 | * On Zen-based systems, these node devices represent Data Fabric functionality. |
| 25 | * |
| 26 | * See "Configuration Space Accesses" section in BKDGs or |
| 27 | * "Processor x86 Core" -> "Configuration Space" section in PPRs. |
| 28 | */ |
| 29 | struct pci_dev *amd_node_get_func(u16 node, u8 func) |
| 30 | { |
| 31 | if (node >= MAX_AMD_NUM_NODES) |
| 32 | return NULL; |
| 33 | |
| 34 | return pci_get_domain_bus_and_slot(domain: 0, bus: 0, PCI_DEVFN(AMD_NODE0_PCI_SLOT + node, func)); |
| 35 | } |
| 36 | |
| 37 | static struct pci_dev **amd_roots; |
| 38 | |
| 39 | /* Protect the PCI config register pairs used for SMN. */ |
| 40 | static DEFINE_MUTEX(smn_mutex); |
| 41 | static bool smn_exclusive; |
| 42 | |
| 43 | #define SMN_INDEX_OFFSET 0x60 |
| 44 | #define SMN_DATA_OFFSET 0x64 |
| 45 | |
| 46 | #define HSMP_INDEX_OFFSET 0xc4 |
| 47 | #define HSMP_DATA_OFFSET 0xc8 |
| 48 | |
| 49 | /* |
| 50 | * SMN accesses may fail in ways that are difficult to detect here in the called |
| 51 | * functions amd_smn_read() and amd_smn_write(). Therefore, callers must do |
| 52 | * their own checking based on what behavior they expect. |
| 53 | * |
| 54 | * For SMN reads, the returned value may be zero if the register is Read-as-Zero. |
| 55 | * Or it may be a "PCI Error Response", e.g. all 0xFFs. The "PCI Error Response" |
| 56 | * can be checked here, and a proper error code can be returned. |
| 57 | * |
| 58 | * But the Read-as-Zero response cannot be verified here. A value of 0 may be |
| 59 | * correct in some cases, so callers must check that this correct is for the |
| 60 | * register/fields they need. |
| 61 | * |
| 62 | * For SMN writes, success can be determined through a "write and read back" |
| 63 | * However, this is not robust when done here. |
| 64 | * |
| 65 | * Possible issues: |
| 66 | * |
| 67 | * 1) Bits that are "Write-1-to-Clear". In this case, the read value should |
| 68 | * *not* match the write value. |
| 69 | * |
| 70 | * 2) Bits that are "Read-as-Zero"/"Writes-Ignored". This information cannot be |
| 71 | * known here. |
| 72 | * |
| 73 | * 3) Bits that are "Reserved / Set to 1". Ditto above. |
| 74 | * |
| 75 | * Callers of amd_smn_write() should do the "write and read back" check |
| 76 | * themselves, if needed. |
| 77 | * |
| 78 | * For #1, they can see if their target bits got cleared. |
| 79 | * |
| 80 | * For #2 and #3, they can check if their target bits got set as intended. |
| 81 | * |
| 82 | * This matches what is done for RDMSR/WRMSR. As long as there's no #GP, then |
| 83 | * the operation is considered a success, and the caller does their own |
| 84 | * checking. |
| 85 | */ |
| 86 | static int __amd_smn_rw(u8 i_off, u8 d_off, u16 node, u32 address, u32 *value, bool write) |
| 87 | { |
| 88 | struct pci_dev *root; |
| 89 | int err = -ENODEV; |
| 90 | |
| 91 | if (node >= amd_num_nodes()) |
| 92 | return err; |
| 93 | |
| 94 | root = amd_roots[node]; |
| 95 | if (!root) |
| 96 | return err; |
| 97 | |
| 98 | if (!smn_exclusive) |
| 99 | return err; |
| 100 | |
| 101 | guard(mutex)(T: &smn_mutex); |
| 102 | |
| 103 | err = pci_write_config_dword(dev: root, where: i_off, val: address); |
| 104 | if (err) { |
| 105 | pr_warn("Error programming SMN address 0x%x.\n" , address); |
| 106 | return pcibios_err_to_errno(err); |
| 107 | } |
| 108 | |
| 109 | err = (write ? pci_write_config_dword(dev: root, where: d_off, val: *value) |
| 110 | : pci_read_config_dword(dev: root, where: d_off, val: value)); |
| 111 | |
| 112 | return pcibios_err_to_errno(err); |
| 113 | } |
| 114 | |
| 115 | int __must_check amd_smn_read(u16 node, u32 address, u32 *value) |
| 116 | { |
| 117 | int err = __amd_smn_rw(SMN_INDEX_OFFSET, SMN_DATA_OFFSET, node, address, value, write: false); |
| 118 | |
| 119 | if (PCI_POSSIBLE_ERROR(*value)) { |
| 120 | err = -ENODEV; |
| 121 | *value = 0; |
| 122 | } |
| 123 | |
| 124 | return err; |
| 125 | } |
| 126 | EXPORT_SYMBOL_GPL(amd_smn_read); |
| 127 | |
| 128 | int __must_check amd_smn_write(u16 node, u32 address, u32 value) |
| 129 | { |
| 130 | return __amd_smn_rw(SMN_INDEX_OFFSET, SMN_DATA_OFFSET, node, address, value: &value, write: true); |
| 131 | } |
| 132 | EXPORT_SYMBOL_GPL(amd_smn_write); |
| 133 | |
| 134 | int __must_check amd_smn_hsmp_rdwr(u16 node, u32 address, u32 *value, bool write) |
| 135 | { |
| 136 | return __amd_smn_rw(HSMP_INDEX_OFFSET, HSMP_DATA_OFFSET, node, address, value, write); |
| 137 | } |
| 138 | EXPORT_SYMBOL_GPL(amd_smn_hsmp_rdwr); |
| 139 | |
| 140 | static struct dentry *debugfs_dir; |
| 141 | static u16 debug_node; |
| 142 | static u32 debug_address; |
| 143 | |
| 144 | static ssize_t smn_node_write(struct file *file, const char __user *userbuf, |
| 145 | size_t count, loff_t *ppos) |
| 146 | { |
| 147 | u16 node; |
| 148 | int ret; |
| 149 | |
| 150 | ret = kstrtou16_from_user(s: userbuf, count, base: 0, res: &node); |
| 151 | if (ret) |
| 152 | return ret; |
| 153 | |
| 154 | if (node >= amd_num_nodes()) |
| 155 | return -ENODEV; |
| 156 | |
| 157 | debug_node = node; |
| 158 | return count; |
| 159 | } |
| 160 | |
| 161 | static int smn_node_show(struct seq_file *m, void *v) |
| 162 | { |
| 163 | seq_printf(m, fmt: "0x%08x\n" , debug_node); |
| 164 | return 0; |
| 165 | } |
| 166 | |
| 167 | static ssize_t smn_address_write(struct file *file, const char __user *userbuf, |
| 168 | size_t count, loff_t *ppos) |
| 169 | { |
| 170 | int ret; |
| 171 | |
| 172 | ret = kstrtouint_from_user(s: userbuf, count, base: 0, res: &debug_address); |
| 173 | if (ret) |
| 174 | return ret; |
| 175 | |
| 176 | return count; |
| 177 | } |
| 178 | |
| 179 | static int smn_address_show(struct seq_file *m, void *v) |
| 180 | { |
| 181 | seq_printf(m, fmt: "0x%08x\n" , debug_address); |
| 182 | return 0; |
| 183 | } |
| 184 | |
| 185 | static int smn_value_show(struct seq_file *m, void *v) |
| 186 | { |
| 187 | u32 val; |
| 188 | int ret; |
| 189 | |
| 190 | ret = amd_smn_read(debug_node, debug_address, &val); |
| 191 | if (ret) |
| 192 | return ret; |
| 193 | |
| 194 | seq_printf(m, fmt: "0x%08x\n" , val); |
| 195 | return 0; |
| 196 | } |
| 197 | |
| 198 | static ssize_t smn_value_write(struct file *file, const char __user *userbuf, |
| 199 | size_t count, loff_t *ppos) |
| 200 | { |
| 201 | u32 val; |
| 202 | int ret; |
| 203 | |
| 204 | ret = kstrtouint_from_user(s: userbuf, count, base: 0, res: &val); |
| 205 | if (ret) |
| 206 | return ret; |
| 207 | |
| 208 | add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_STILL_OK); |
| 209 | |
| 210 | ret = amd_smn_write(debug_node, debug_address, val); |
| 211 | if (ret) |
| 212 | return ret; |
| 213 | |
| 214 | return count; |
| 215 | } |
| 216 | |
| 217 | DEFINE_SHOW_STORE_ATTRIBUTE(smn_node); |
| 218 | DEFINE_SHOW_STORE_ATTRIBUTE(smn_address); |
| 219 | DEFINE_SHOW_STORE_ATTRIBUTE(smn_value); |
| 220 | |
| 221 | static struct pci_dev *get_next_root(struct pci_dev *root) |
| 222 | { |
| 223 | while ((root = pci_get_class(PCI_CLASS_BRIDGE_HOST << 8, from: root))) { |
| 224 | /* Root device is Device 0 Function 0. */ |
| 225 | if (root->devfn) |
| 226 | continue; |
| 227 | |
| 228 | if (root->vendor != PCI_VENDOR_ID_AMD && |
| 229 | root->vendor != PCI_VENDOR_ID_HYGON) |
| 230 | continue; |
| 231 | |
| 232 | break; |
| 233 | } |
| 234 | |
| 235 | return root; |
| 236 | } |
| 237 | |
| 238 | static bool enable_dfs; |
| 239 | |
| 240 | static int __init amd_smn_enable_dfs(char *str) |
| 241 | { |
| 242 | enable_dfs = true; |
| 243 | return 1; |
| 244 | } |
| 245 | __setup("amd_smn_debugfs_enable" , amd_smn_enable_dfs); |
| 246 | |
| 247 | static int __init amd_smn_init(void) |
| 248 | { |
| 249 | u16 count, num_roots, roots_per_node, node, num_nodes; |
| 250 | struct pci_dev *root; |
| 251 | |
| 252 | if (!cpu_feature_enabled(X86_FEATURE_ZEN)) |
| 253 | return 0; |
| 254 | |
| 255 | guard(mutex)(T: &smn_mutex); |
| 256 | |
| 257 | if (amd_roots) |
| 258 | return 0; |
| 259 | |
| 260 | num_roots = 0; |
| 261 | root = NULL; |
| 262 | while ((root = get_next_root(root))) { |
| 263 | pci_dbg(root, "Reserving PCI config space\n" ); |
| 264 | |
| 265 | /* |
| 266 | * There are a few SMN index/data pairs and other registers |
| 267 | * that shouldn't be accessed by user space. So reserve the |
| 268 | * entire PCI config space for simplicity rather than covering |
| 269 | * specific registers piecemeal. |
| 270 | */ |
| 271 | if (!pci_request_config_region_exclusive(pdev: root, offset: 0, PCI_CFG_SPACE_SIZE, NULL)) { |
| 272 | pci_err(root, "Failed to reserve config space\n" ); |
| 273 | return -EEXIST; |
| 274 | } |
| 275 | |
| 276 | num_roots++; |
| 277 | } |
| 278 | |
| 279 | pr_debug("Found %d AMD root devices\n" , num_roots); |
| 280 | |
| 281 | if (!num_roots) |
| 282 | return -ENODEV; |
| 283 | |
| 284 | num_nodes = amd_num_nodes(); |
| 285 | amd_roots = kcalloc(num_nodes, sizeof(*amd_roots), GFP_KERNEL); |
| 286 | if (!amd_roots) |
| 287 | return -ENOMEM; |
| 288 | |
| 289 | roots_per_node = num_roots / num_nodes; |
| 290 | |
| 291 | count = 0; |
| 292 | node = 0; |
| 293 | root = NULL; |
| 294 | while (node < num_nodes && (root = get_next_root(root))) { |
| 295 | /* Use one root for each node and skip the rest. */ |
| 296 | if (count++ % roots_per_node) |
| 297 | continue; |
| 298 | |
| 299 | pci_dbg(root, "is root for AMD node %u\n" , node); |
| 300 | amd_roots[node++] = root; |
| 301 | } |
| 302 | |
| 303 | if (enable_dfs) { |
| 304 | debugfs_dir = debugfs_create_dir(name: "amd_smn" , parent: arch_debugfs_dir); |
| 305 | |
| 306 | debugfs_create_file("node" , 0600, debugfs_dir, NULL, &smn_node_fops); |
| 307 | debugfs_create_file("address" , 0600, debugfs_dir, NULL, &smn_address_fops); |
| 308 | debugfs_create_file("value" , 0600, debugfs_dir, NULL, &smn_value_fops); |
| 309 | } |
| 310 | |
| 311 | smn_exclusive = true; |
| 312 | |
| 313 | return 0; |
| 314 | } |
| 315 | |
| 316 | fs_initcall(amd_smn_init); |
| 317 | |