| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * ZynqMP R5 Remote Processor driver |
| 4 | * |
| 5 | */ |
| 6 | |
| 7 | #include <dt-bindings/power/xlnx-zynqmp-power.h> |
| 8 | #include <linux/dma-mapping.h> |
| 9 | #include <linux/firmware/xlnx-zynqmp.h> |
| 10 | #include <linux/kernel.h> |
| 11 | #include <linux/mailbox_client.h> |
| 12 | #include <linux/mailbox/zynqmp-ipi-message.h> |
| 13 | #include <linux/module.h> |
| 14 | #include <linux/of_address.h> |
| 15 | #include <linux/of_platform.h> |
| 16 | #include <linux/of_reserved_mem.h> |
| 17 | #include <linux/platform_device.h> |
| 18 | #include <linux/remoteproc.h> |
| 19 | |
| 20 | #include "remoteproc_internal.h" |
| 21 | |
| 22 | /* IPI buffer MAX length */ |
| 23 | #define IPI_BUF_LEN_MAX 32U |
| 24 | |
| 25 | /* RX mailbox client buffer max length */ |
| 26 | #define MBOX_CLIENT_BUF_MAX (IPI_BUF_LEN_MAX + \ |
| 27 | sizeof(struct zynqmp_ipi_message)) |
| 28 | |
| 29 | #define RSC_TBL_XLNX_MAGIC ((uint32_t)'x' << 24 | (uint32_t)'a' << 16 | \ |
| 30 | (uint32_t)'m' << 8 | (uint32_t)'p') |
| 31 | |
| 32 | /* |
| 33 | * settings for RPU cluster mode which |
| 34 | * reflects possible values of xlnx,cluster-mode dt-property |
| 35 | */ |
| 36 | enum zynqmp_r5_cluster_mode { |
| 37 | SPLIT_MODE = 0, /* When cores run as separate processor */ |
| 38 | LOCKSTEP_MODE = 1, /* cores execute same code in lockstep,clk-for-clk */ |
| 39 | SINGLE_CPU_MODE = 2, /* core0 is held in reset and only core1 runs */ |
| 40 | }; |
| 41 | |
| 42 | /** |
| 43 | * struct mem_bank_data - Memory Bank description |
| 44 | * |
| 45 | * @addr: Start address of memory bank |
| 46 | * @da: device address |
| 47 | * @size: Size of Memory bank |
| 48 | * @pm_domain_id: Power-domains id of memory bank for firmware to turn on/off |
| 49 | * @bank_name: name of the bank for remoteproc framework |
| 50 | */ |
| 51 | struct mem_bank_data { |
| 52 | phys_addr_t addr; |
| 53 | u32 da; |
| 54 | size_t size; |
| 55 | u32 pm_domain_id; |
| 56 | char *bank_name; |
| 57 | }; |
| 58 | |
| 59 | /** |
| 60 | * struct zynqmp_sram_bank - sram bank description |
| 61 | * |
| 62 | * @sram_res: sram address region information |
| 63 | * @da: device address of sram |
| 64 | */ |
| 65 | struct zynqmp_sram_bank { |
| 66 | struct resource sram_res; |
| 67 | u32 da; |
| 68 | }; |
| 69 | |
| 70 | /** |
| 71 | * struct mbox_info |
| 72 | * |
| 73 | * @rx_mc_buf: to copy data from mailbox rx channel |
| 74 | * @tx_mc_buf: to copy data to mailbox tx channel |
| 75 | * @r5_core: this mailbox's corresponding r5_core pointer |
| 76 | * @mbox_work: schedule work after receiving data from mailbox |
| 77 | * @mbox_cl: mailbox client |
| 78 | * @tx_chan: mailbox tx channel |
| 79 | * @rx_chan: mailbox rx channel |
| 80 | */ |
| 81 | struct mbox_info { |
| 82 | unsigned char rx_mc_buf[MBOX_CLIENT_BUF_MAX]; |
| 83 | unsigned char tx_mc_buf[MBOX_CLIENT_BUF_MAX]; |
| 84 | struct zynqmp_r5_core *r5_core; |
| 85 | struct work_struct mbox_work; |
| 86 | struct mbox_client mbox_cl; |
| 87 | struct mbox_chan *tx_chan; |
| 88 | struct mbox_chan *rx_chan; |
| 89 | }; |
| 90 | |
| 91 | /** |
| 92 | * struct rsc_tbl_data |
| 93 | * |
| 94 | * Platform specific data structure used to sync resource table address. |
| 95 | * It's important to maintain order and size of each field on remote side. |
| 96 | * |
| 97 | * @version: version of data structure |
| 98 | * @magic_num: 32-bit magic number. |
| 99 | * @comp_magic_num: complement of above magic number |
| 100 | * @rsc_tbl_size: resource table size |
| 101 | * @rsc_tbl: resource table address |
| 102 | */ |
| 103 | struct rsc_tbl_data { |
| 104 | const int version; |
| 105 | const u32 magic_num; |
| 106 | const u32 comp_magic_num; |
| 107 | const u32 rsc_tbl_size; |
| 108 | const uintptr_t rsc_tbl; |
| 109 | } __packed; |
| 110 | |
| 111 | /* |
| 112 | * Hardcoded TCM bank values. This will stay in driver to maintain backward |
| 113 | * compatibility with device-tree that does not have TCM information. |
| 114 | */ |
| 115 | static const struct mem_bank_data zynqmp_tcm_banks_split[] = { |
| 116 | {0xffe00000UL, 0x0, 0x10000UL, PD_R5_0_ATCM, "atcm0" }, /* TCM 64KB each */ |
| 117 | {0xffe20000UL, 0x20000, 0x10000UL, PD_R5_0_BTCM, "btcm0" }, |
| 118 | {0xffe90000UL, 0x0, 0x10000UL, PD_R5_1_ATCM, "atcm1" }, |
| 119 | {0xffeb0000UL, 0x20000, 0x10000UL, PD_R5_1_BTCM, "btcm1" }, |
| 120 | }; |
| 121 | |
| 122 | /* In lockstep mode cluster uses each 64KB TCM from second core as well */ |
| 123 | static const struct mem_bank_data zynqmp_tcm_banks_lockstep[] = { |
| 124 | {0xffe00000UL, 0x0, 0x10000UL, PD_R5_0_ATCM, "atcm0" }, /* TCM 64KB each */ |
| 125 | {0xffe20000UL, 0x20000, 0x10000UL, PD_R5_0_BTCM, "btcm0" }, |
| 126 | {0xffe10000UL, 0x10000, 0x10000UL, PD_R5_1_ATCM, "atcm1" }, |
| 127 | {0xffe30000UL, 0x30000, 0x10000UL, PD_R5_1_BTCM, "btcm1" }, |
| 128 | }; |
| 129 | |
| 130 | /** |
| 131 | * struct zynqmp_r5_core |
| 132 | * |
| 133 | * @rsc_tbl_va: resource table virtual address |
| 134 | * @sram: Array of sram memories assigned to this core |
| 135 | * @num_sram: number of sram for this core |
| 136 | * @dev: device of RPU instance |
| 137 | * @np: device node of RPU instance |
| 138 | * @tcm_bank_count: number TCM banks accessible to this RPU |
| 139 | * @tcm_banks: array of each TCM bank data |
| 140 | * @rproc: rproc handle |
| 141 | * @rsc_tbl_size: resource table size retrieved from remote |
| 142 | * @pm_domain_id: RPU CPU power domain id |
| 143 | * @ipi: pointer to mailbox information |
| 144 | */ |
| 145 | struct zynqmp_r5_core { |
| 146 | void __iomem *rsc_tbl_va; |
| 147 | struct zynqmp_sram_bank *sram; |
| 148 | int num_sram; |
| 149 | struct device *dev; |
| 150 | struct device_node *np; |
| 151 | int tcm_bank_count; |
| 152 | struct mem_bank_data **tcm_banks; |
| 153 | struct rproc *rproc; |
| 154 | u32 rsc_tbl_size; |
| 155 | u32 pm_domain_id; |
| 156 | struct mbox_info *ipi; |
| 157 | }; |
| 158 | |
| 159 | /** |
| 160 | * struct zynqmp_r5_cluster |
| 161 | * |
| 162 | * @dev: r5f subsystem cluster device node |
| 163 | * @mode: cluster mode of type zynqmp_r5_cluster_mode |
| 164 | * @core_count: number of r5 cores used for this cluster mode |
| 165 | * @r5_cores: Array of pointers pointing to r5 core |
| 166 | */ |
| 167 | struct zynqmp_r5_cluster { |
| 168 | struct device *dev; |
| 169 | enum zynqmp_r5_cluster_mode mode; |
| 170 | int core_count; |
| 171 | struct zynqmp_r5_core **r5_cores; |
| 172 | }; |
| 173 | |
| 174 | /** |
| 175 | * event_notified_idr_cb() - callback for vq_interrupt per notifyid |
| 176 | * @id: rproc->notify id |
| 177 | * @ptr: pointer to idr private data |
| 178 | * @data: data passed to idr_for_each callback |
| 179 | * |
| 180 | * Pass notification to remoteproc virtio |
| 181 | * |
| 182 | * Return: 0. having return is to satisfy the idr_for_each() function |
| 183 | * pointer input argument requirement. |
| 184 | **/ |
| 185 | static int event_notified_idr_cb(int id, void *ptr, void *data) |
| 186 | { |
| 187 | struct rproc *rproc = data; |
| 188 | |
| 189 | if (rproc_vq_interrupt(rproc, vq_id: id) == IRQ_NONE) |
| 190 | dev_dbg(&rproc->dev, "data not found for vqid=%d\n" , id); |
| 191 | |
| 192 | return 0; |
| 193 | } |
| 194 | |
| 195 | /** |
| 196 | * handle_event_notified() - remoteproc notification work function |
| 197 | * @work: pointer to the work structure |
| 198 | * |
| 199 | * It checks each registered remoteproc notify IDs. |
| 200 | */ |
| 201 | static void handle_event_notified(struct work_struct *work) |
| 202 | { |
| 203 | struct mbox_info *ipi; |
| 204 | struct rproc *rproc; |
| 205 | |
| 206 | ipi = container_of(work, struct mbox_info, mbox_work); |
| 207 | rproc = ipi->r5_core->rproc; |
| 208 | |
| 209 | /* |
| 210 | * We only use IPI for interrupt. The RPU firmware side may or may |
| 211 | * not write the notifyid when it trigger IPI. |
| 212 | * And thus, we scan through all the registered notifyids and |
| 213 | * find which one is valid to get the message. |
| 214 | * Even if message from firmware is NULL, we attempt to get vqid |
| 215 | */ |
| 216 | idr_for_each(&rproc->notifyids, fn: event_notified_idr_cb, data: rproc); |
| 217 | } |
| 218 | |
| 219 | /** |
| 220 | * zynqmp_r5_mb_rx_cb() - receive channel mailbox callback |
| 221 | * @cl: mailbox client |
| 222 | * @msg: message pointer |
| 223 | * |
| 224 | * Receive data from ipi buffer, ack interrupt and then |
| 225 | * it will schedule the R5 notification work. |
| 226 | */ |
| 227 | static void zynqmp_r5_mb_rx_cb(struct mbox_client *cl, void *msg) |
| 228 | { |
| 229 | struct zynqmp_ipi_message *ipi_msg, *buf_msg; |
| 230 | struct mbox_info *ipi; |
| 231 | size_t len; |
| 232 | |
| 233 | ipi = container_of(cl, struct mbox_info, mbox_cl); |
| 234 | |
| 235 | /* copy data from ipi buffer to r5_core */ |
| 236 | ipi_msg = (struct zynqmp_ipi_message *)msg; |
| 237 | buf_msg = (struct zynqmp_ipi_message *)ipi->rx_mc_buf; |
| 238 | len = ipi_msg->len; |
| 239 | if (len > IPI_BUF_LEN_MAX) { |
| 240 | dev_warn(cl->dev, "msg size exceeded than %d\n" , |
| 241 | IPI_BUF_LEN_MAX); |
| 242 | len = IPI_BUF_LEN_MAX; |
| 243 | } |
| 244 | buf_msg->len = len; |
| 245 | memcpy(buf_msg->data, ipi_msg->data, len); |
| 246 | |
| 247 | /* received and processed interrupt ack */ |
| 248 | if (mbox_send_message(chan: ipi->rx_chan, NULL) < 0) |
| 249 | dev_err(cl->dev, "ack failed to mbox rx_chan\n" ); |
| 250 | |
| 251 | schedule_work(work: &ipi->mbox_work); |
| 252 | } |
| 253 | |
| 254 | /** |
| 255 | * zynqmp_r5_setup_mbox() - Setup mailboxes related properties |
| 256 | * this is used for each individual R5 core |
| 257 | * |
| 258 | * @cdev: child node device |
| 259 | * |
| 260 | * Function to setup mailboxes related properties |
| 261 | * return : NULL if failed else pointer to mbox_info |
| 262 | */ |
| 263 | static struct mbox_info *zynqmp_r5_setup_mbox(struct device *cdev) |
| 264 | { |
| 265 | struct mbox_client *mbox_cl; |
| 266 | struct mbox_info *ipi; |
| 267 | |
| 268 | ipi = kzalloc(sizeof(*ipi), GFP_KERNEL); |
| 269 | if (!ipi) |
| 270 | return NULL; |
| 271 | |
| 272 | mbox_cl = &ipi->mbox_cl; |
| 273 | mbox_cl->rx_callback = zynqmp_r5_mb_rx_cb; |
| 274 | mbox_cl->tx_block = false; |
| 275 | mbox_cl->knows_txdone = false; |
| 276 | mbox_cl->tx_done = NULL; |
| 277 | mbox_cl->dev = cdev; |
| 278 | |
| 279 | /* Request TX and RX channels */ |
| 280 | ipi->tx_chan = mbox_request_channel_byname(cl: mbox_cl, name: "tx" ); |
| 281 | if (IS_ERR(ptr: ipi->tx_chan)) { |
| 282 | ipi->tx_chan = NULL; |
| 283 | kfree(objp: ipi); |
| 284 | dev_warn(cdev, "mbox tx channel request failed\n" ); |
| 285 | return NULL; |
| 286 | } |
| 287 | |
| 288 | ipi->rx_chan = mbox_request_channel_byname(cl: mbox_cl, name: "rx" ); |
| 289 | if (IS_ERR(ptr: ipi->rx_chan)) { |
| 290 | mbox_free_channel(chan: ipi->tx_chan); |
| 291 | ipi->rx_chan = NULL; |
| 292 | ipi->tx_chan = NULL; |
| 293 | kfree(objp: ipi); |
| 294 | dev_warn(cdev, "mbox rx channel request failed\n" ); |
| 295 | return NULL; |
| 296 | } |
| 297 | |
| 298 | INIT_WORK(&ipi->mbox_work, handle_event_notified); |
| 299 | |
| 300 | return ipi; |
| 301 | } |
| 302 | |
| 303 | static void zynqmp_r5_free_mbox(struct mbox_info *ipi) |
| 304 | { |
| 305 | if (!ipi) |
| 306 | return; |
| 307 | |
| 308 | if (ipi->tx_chan) { |
| 309 | mbox_free_channel(chan: ipi->tx_chan); |
| 310 | ipi->tx_chan = NULL; |
| 311 | } |
| 312 | |
| 313 | if (ipi->rx_chan) { |
| 314 | mbox_free_channel(chan: ipi->rx_chan); |
| 315 | ipi->rx_chan = NULL; |
| 316 | } |
| 317 | |
| 318 | kfree(objp: ipi); |
| 319 | } |
| 320 | |
| 321 | /* |
| 322 | * zynqmp_r5_core_kick() - kick a firmware if mbox is provided |
| 323 | * @rproc: r5 core's corresponding rproc structure |
| 324 | * @vqid: virtqueue ID |
| 325 | */ |
| 326 | static void zynqmp_r5_rproc_kick(struct rproc *rproc, int vqid) |
| 327 | { |
| 328 | struct zynqmp_r5_core *r5_core = rproc->priv; |
| 329 | struct device *dev = r5_core->dev; |
| 330 | struct zynqmp_ipi_message *mb_msg; |
| 331 | struct mbox_info *ipi; |
| 332 | int ret; |
| 333 | |
| 334 | ipi = r5_core->ipi; |
| 335 | if (!ipi) |
| 336 | return; |
| 337 | |
| 338 | mb_msg = (struct zynqmp_ipi_message *)ipi->tx_mc_buf; |
| 339 | memcpy(mb_msg->data, &vqid, sizeof(vqid)); |
| 340 | mb_msg->len = sizeof(vqid); |
| 341 | ret = mbox_send_message(chan: ipi->tx_chan, mssg: mb_msg); |
| 342 | if (ret < 0) |
| 343 | dev_warn(dev, "failed to send message\n" ); |
| 344 | } |
| 345 | |
| 346 | /* |
| 347 | * zynqmp_r5_rproc_start() |
| 348 | * @rproc: single R5 core's corresponding rproc instance |
| 349 | * |
| 350 | * Start R5 Core from designated boot address. |
| 351 | * |
| 352 | * return 0 on success, otherwise non-zero value on failure |
| 353 | */ |
| 354 | static int zynqmp_r5_rproc_start(struct rproc *rproc) |
| 355 | { |
| 356 | struct zynqmp_r5_core *r5_core = rproc->priv; |
| 357 | enum rpu_boot_mem bootmem; |
| 358 | int ret; |
| 359 | |
| 360 | /* |
| 361 | * The exception vector pointers (EVP) refer to the base-address of |
| 362 | * exception vectors (for reset, IRQ, FIQ, etc). The reset-vector |
| 363 | * starts at the base-address and subsequent vectors are on 4-byte |
| 364 | * boundaries. |
| 365 | * |
| 366 | * Exception vectors can start either from 0x0000_0000 (LOVEC) or |
| 367 | * from 0xFFFF_0000 (HIVEC) which is mapped in the OCM (On-Chip Memory) |
| 368 | * |
| 369 | * Usually firmware will put Exception vectors at LOVEC. |
| 370 | * |
| 371 | * It is not recommend that you change the exception vector. |
| 372 | * Changing the EVP to HIVEC will result in increased interrupt latency |
| 373 | * and jitter. Also, if the OCM is secured and the Cortex-R5F processor |
| 374 | * is non-secured, then the Cortex-R5F processor cannot access the |
| 375 | * HIVEC exception vectors in the OCM. |
| 376 | */ |
| 377 | bootmem = (rproc->bootaddr >= 0xFFFC0000) ? |
| 378 | PM_RPU_BOOTMEM_HIVEC : PM_RPU_BOOTMEM_LOVEC; |
| 379 | |
| 380 | dev_dbg(r5_core->dev, "RPU boot addr 0x%llx from %s." , rproc->bootaddr, |
| 381 | bootmem == PM_RPU_BOOTMEM_HIVEC ? "OCM" : "TCM" ); |
| 382 | |
| 383 | /* Request node before starting RPU core if new version of API is supported */ |
| 384 | if (zynqmp_pm_feature(api_id: PM_REQUEST_NODE) > 1) { |
| 385 | ret = zynqmp_pm_request_node(node: r5_core->pm_domain_id, |
| 386 | ZYNQMP_PM_CAPABILITY_ACCESS, qos: 0, |
| 387 | ack: ZYNQMP_PM_REQUEST_ACK_BLOCKING); |
| 388 | if (ret < 0) { |
| 389 | dev_err(r5_core->dev, "failed to request 0x%x" , |
| 390 | r5_core->pm_domain_id); |
| 391 | return ret; |
| 392 | } |
| 393 | } |
| 394 | |
| 395 | ret = zynqmp_pm_request_wake(node: r5_core->pm_domain_id, set_addr: 1, |
| 396 | address: bootmem, ack: ZYNQMP_PM_REQUEST_ACK_NO); |
| 397 | if (ret) |
| 398 | dev_err(r5_core->dev, |
| 399 | "failed to start RPU = 0x%x\n" , r5_core->pm_domain_id); |
| 400 | return ret; |
| 401 | } |
| 402 | |
| 403 | /* |
| 404 | * zynqmp_r5_rproc_stop() |
| 405 | * @rproc: single R5 core's corresponding rproc instance |
| 406 | * |
| 407 | * Power down R5 Core. |
| 408 | * |
| 409 | * return 0 on success, otherwise non-zero value on failure |
| 410 | */ |
| 411 | static int zynqmp_r5_rproc_stop(struct rproc *rproc) |
| 412 | { |
| 413 | struct zynqmp_r5_core *r5_core = rproc->priv; |
| 414 | int ret; |
| 415 | |
| 416 | /* Use release node API to stop core if new version of API is supported */ |
| 417 | if (zynqmp_pm_feature(api_id: PM_RELEASE_NODE) > 1) { |
| 418 | ret = zynqmp_pm_release_node(node: r5_core->pm_domain_id); |
| 419 | if (ret) |
| 420 | dev_err(r5_core->dev, "failed to stop remoteproc RPU %d\n" , ret); |
| 421 | return ret; |
| 422 | } |
| 423 | |
| 424 | /* |
| 425 | * Check expected version of EEMI call before calling it. This avoids |
| 426 | * any error or warning prints from firmware as it is expected that fw |
| 427 | * doesn't support it. |
| 428 | */ |
| 429 | if (zynqmp_pm_feature(api_id: PM_FORCE_POWERDOWN) != 1) { |
| 430 | dev_dbg(r5_core->dev, "EEMI interface %d ver 1 not supported\n" , |
| 431 | PM_FORCE_POWERDOWN); |
| 432 | return -EOPNOTSUPP; |
| 433 | } |
| 434 | |
| 435 | /* maintain force pwr down for backward compatibility */ |
| 436 | ret = zynqmp_pm_force_pwrdwn(target: r5_core->pm_domain_id, |
| 437 | ack: ZYNQMP_PM_REQUEST_ACK_BLOCKING); |
| 438 | if (ret) |
| 439 | dev_err(r5_core->dev, "core force power down failed\n" ); |
| 440 | |
| 441 | return ret; |
| 442 | } |
| 443 | |
| 444 | /* |
| 445 | * zynqmp_r5_mem_region_map() |
| 446 | * @rproc: single R5 core's corresponding rproc instance |
| 447 | * @mem: mem descriptor to map reserved memory-regions |
| 448 | * |
| 449 | * Callback to map va for memory-region's carveout. |
| 450 | * |
| 451 | * return 0 on success, otherwise non-zero value on failure |
| 452 | */ |
| 453 | static int zynqmp_r5_mem_region_map(struct rproc *rproc, |
| 454 | struct rproc_mem_entry *mem) |
| 455 | { |
| 456 | void __iomem *va; |
| 457 | |
| 458 | va = ioremap_wc(offset: mem->dma, size: mem->len); |
| 459 | if (IS_ERR_OR_NULL(ptr: va)) |
| 460 | return -ENOMEM; |
| 461 | |
| 462 | mem->va = (void *)va; |
| 463 | |
| 464 | return 0; |
| 465 | } |
| 466 | |
| 467 | /* |
| 468 | * zynqmp_r5_rproc_mem_unmap |
| 469 | * @rproc: single R5 core's corresponding rproc instance |
| 470 | * @mem: mem entry to unmap |
| 471 | * |
| 472 | * Unmap memory-region carveout |
| 473 | * |
| 474 | * return: always returns 0 |
| 475 | */ |
| 476 | static int zynqmp_r5_mem_region_unmap(struct rproc *rproc, |
| 477 | struct rproc_mem_entry *mem) |
| 478 | { |
| 479 | iounmap(addr: (void __iomem *)mem->va); |
| 480 | return 0; |
| 481 | } |
| 482 | |
| 483 | /* |
| 484 | * add_mem_regions_carveout() |
| 485 | * @rproc: single R5 core's corresponding rproc instance |
| 486 | * |
| 487 | * Construct rproc mem carveouts from memory-region property nodes |
| 488 | * |
| 489 | * return 0 on success, otherwise non-zero value on failure |
| 490 | */ |
| 491 | static int add_mem_regions_carveout(struct rproc *rproc) |
| 492 | { |
| 493 | struct rproc_mem_entry *rproc_mem; |
| 494 | struct zynqmp_r5_core *r5_core; |
| 495 | struct of_phandle_iterator it; |
| 496 | struct reserved_mem *rmem; |
| 497 | int i = 0; |
| 498 | |
| 499 | r5_core = rproc->priv; |
| 500 | |
| 501 | /* Register associated reserved memory regions */ |
| 502 | of_phandle_iterator_init(it: &it, np: r5_core->np, list_name: "memory-region" , NULL, cell_count: 0); |
| 503 | |
| 504 | while (of_phandle_iterator_next(it: &it) == 0) { |
| 505 | rmem = of_reserved_mem_lookup(np: it.node); |
| 506 | if (!rmem) { |
| 507 | of_node_put(node: it.node); |
| 508 | dev_err(&rproc->dev, "unable to acquire memory-region\n" ); |
| 509 | return -EINVAL; |
| 510 | } |
| 511 | |
| 512 | if (!strcmp(it.node->name, "vdev0buffer" )) { |
| 513 | /* Init reserved memory for vdev buffer */ |
| 514 | rproc_mem = rproc_of_resm_mem_entry_init(dev: &rproc->dev, of_resm_idx: i, |
| 515 | len: rmem->size, |
| 516 | da: rmem->base, |
| 517 | name: it.node->name); |
| 518 | } else { |
| 519 | /* Register associated reserved memory regions */ |
| 520 | rproc_mem = rproc_mem_entry_init(dev: &rproc->dev, NULL, |
| 521 | dma: (dma_addr_t)rmem->base, |
| 522 | len: rmem->size, da: rmem->base, |
| 523 | alloc: zynqmp_r5_mem_region_map, |
| 524 | release: zynqmp_r5_mem_region_unmap, |
| 525 | name: it.node->name); |
| 526 | } |
| 527 | |
| 528 | if (!rproc_mem) { |
| 529 | of_node_put(node: it.node); |
| 530 | return -ENOMEM; |
| 531 | } |
| 532 | |
| 533 | rproc_add_carveout(rproc, mem: rproc_mem); |
| 534 | rproc_coredump_add_segment(rproc, da: rmem->base, size: rmem->size); |
| 535 | |
| 536 | dev_dbg(&rproc->dev, "reserved mem carveout %s addr=%llx, size=0x%llx" , |
| 537 | it.node->name, rmem->base, rmem->size); |
| 538 | i++; |
| 539 | } |
| 540 | |
| 541 | return 0; |
| 542 | } |
| 543 | |
| 544 | static int add_sram_carveouts(struct rproc *rproc) |
| 545 | { |
| 546 | struct zynqmp_r5_core *r5_core = rproc->priv; |
| 547 | struct rproc_mem_entry *rproc_mem; |
| 548 | struct zynqmp_sram_bank *sram; |
| 549 | dma_addr_t dma_addr; |
| 550 | size_t len; |
| 551 | int da, i; |
| 552 | |
| 553 | for (i = 0; i < r5_core->num_sram; i++) { |
| 554 | sram = &r5_core->sram[i]; |
| 555 | |
| 556 | dma_addr = (dma_addr_t)sram->sram_res.start; |
| 557 | |
| 558 | len = resource_size(res: &sram->sram_res); |
| 559 | da = sram->da; |
| 560 | |
| 561 | rproc_mem = rproc_mem_entry_init(dev: &rproc->dev, NULL, |
| 562 | dma: dma_addr, |
| 563 | len, da, |
| 564 | alloc: zynqmp_r5_mem_region_map, |
| 565 | release: zynqmp_r5_mem_region_unmap, |
| 566 | name: sram->sram_res.name); |
| 567 | if (!rproc_mem) { |
| 568 | dev_err(&rproc->dev, "failed to add sram %s da=0x%x, size=0x%lx" , |
| 569 | sram->sram_res.name, da, len); |
| 570 | return -ENOMEM; |
| 571 | } |
| 572 | |
| 573 | rproc_add_carveout(rproc, mem: rproc_mem); |
| 574 | rproc_coredump_add_segment(rproc, da, size: len); |
| 575 | |
| 576 | dev_dbg(&rproc->dev, "sram carveout %s addr=%llx, da=0x%x, size=0x%lx" , |
| 577 | sram->sram_res.name, dma_addr, da, len); |
| 578 | } |
| 579 | |
| 580 | return 0; |
| 581 | } |
| 582 | |
| 583 | /* |
| 584 | * tcm_mem_unmap() |
| 585 | * @rproc: single R5 core's corresponding rproc instance |
| 586 | * @mem: tcm mem entry to unmap |
| 587 | * |
| 588 | * Unmap TCM banks when powering down R5 core. |
| 589 | * |
| 590 | * return always 0 |
| 591 | */ |
| 592 | static int tcm_mem_unmap(struct rproc *rproc, struct rproc_mem_entry *mem) |
| 593 | { |
| 594 | iounmap(addr: (void __iomem *)mem->va); |
| 595 | |
| 596 | return 0; |
| 597 | } |
| 598 | |
| 599 | /* |
| 600 | * tcm_mem_map() |
| 601 | * @rproc: single R5 core's corresponding rproc instance |
| 602 | * @mem: tcm memory entry descriptor |
| 603 | * |
| 604 | * Given TCM bank entry, this func setup virtual address for TCM bank |
| 605 | * remoteproc carveout. It also takes care of va to da address translation |
| 606 | * |
| 607 | * return 0 on success, otherwise non-zero value on failure |
| 608 | */ |
| 609 | static int tcm_mem_map(struct rproc *rproc, |
| 610 | struct rproc_mem_entry *mem) |
| 611 | { |
| 612 | void __iomem *va; |
| 613 | |
| 614 | va = ioremap_wc(offset: mem->dma, size: mem->len); |
| 615 | if (IS_ERR_OR_NULL(ptr: va)) |
| 616 | return -ENOMEM; |
| 617 | |
| 618 | /* Update memory entry va */ |
| 619 | mem->va = (void *)va; |
| 620 | |
| 621 | /* clear TCMs */ |
| 622 | memset_io(va, 0, mem->len); |
| 623 | |
| 624 | return 0; |
| 625 | } |
| 626 | |
| 627 | /* |
| 628 | * add_tcm_banks() |
| 629 | * @rproc: single R5 core's corresponding rproc instance |
| 630 | * |
| 631 | * allocate and add remoteproc carveout for TCM memory |
| 632 | * |
| 633 | * return 0 on success, otherwise non-zero value on failure |
| 634 | */ |
| 635 | static int add_tcm_banks(struct rproc *rproc) |
| 636 | { |
| 637 | struct rproc_mem_entry *rproc_mem; |
| 638 | struct zynqmp_r5_core *r5_core; |
| 639 | int i, num_banks, ret; |
| 640 | phys_addr_t bank_addr; |
| 641 | struct device *dev; |
| 642 | u32 pm_domain_id; |
| 643 | size_t bank_size; |
| 644 | char *bank_name; |
| 645 | u32 da; |
| 646 | |
| 647 | r5_core = rproc->priv; |
| 648 | dev = r5_core->dev; |
| 649 | num_banks = r5_core->tcm_bank_count; |
| 650 | |
| 651 | /* |
| 652 | * Power-on Each 64KB TCM, |
| 653 | * register its address space, map and unmap functions |
| 654 | * and add carveouts accordingly |
| 655 | */ |
| 656 | for (i = 0; i < num_banks; i++) { |
| 657 | bank_addr = r5_core->tcm_banks[i]->addr; |
| 658 | da = r5_core->tcm_banks[i]->da; |
| 659 | bank_name = r5_core->tcm_banks[i]->bank_name; |
| 660 | bank_size = r5_core->tcm_banks[i]->size; |
| 661 | pm_domain_id = r5_core->tcm_banks[i]->pm_domain_id; |
| 662 | |
| 663 | ret = zynqmp_pm_request_node(node: pm_domain_id, |
| 664 | ZYNQMP_PM_CAPABILITY_ACCESS, qos: 0, |
| 665 | ack: ZYNQMP_PM_REQUEST_ACK_BLOCKING); |
| 666 | if (ret < 0) { |
| 667 | dev_err(dev, "failed to turn on TCM 0x%x" , pm_domain_id); |
| 668 | goto release_tcm; |
| 669 | } |
| 670 | |
| 671 | dev_dbg(dev, "TCM carveout %s addr=%llx, da=0x%x, size=0x%lx" , |
| 672 | bank_name, bank_addr, da, bank_size); |
| 673 | |
| 674 | /* |
| 675 | * In DETACHED state firmware is already running so no need to |
| 676 | * request add TCM registers. However, request TCM PD node to let |
| 677 | * platform management firmware know that TCM is in use. |
| 678 | */ |
| 679 | if (rproc->state == RPROC_DETACHED) |
| 680 | continue; |
| 681 | |
| 682 | rproc_mem = rproc_mem_entry_init(dev, NULL, dma: bank_addr, |
| 683 | len: bank_size, da, |
| 684 | alloc: tcm_mem_map, release: tcm_mem_unmap, |
| 685 | name: bank_name); |
| 686 | if (!rproc_mem) { |
| 687 | ret = -ENOMEM; |
| 688 | zynqmp_pm_release_node(node: pm_domain_id); |
| 689 | goto release_tcm; |
| 690 | } |
| 691 | |
| 692 | rproc_add_carveout(rproc, mem: rproc_mem); |
| 693 | rproc_coredump_add_segment(rproc, da, size: bank_size); |
| 694 | } |
| 695 | |
| 696 | return 0; |
| 697 | |
| 698 | release_tcm: |
| 699 | /* If failed, Turn off all TCM banks turned on before */ |
| 700 | for (i--; i >= 0; i--) { |
| 701 | pm_domain_id = r5_core->tcm_banks[i]->pm_domain_id; |
| 702 | zynqmp_pm_release_node(node: pm_domain_id); |
| 703 | } |
| 704 | return ret; |
| 705 | } |
| 706 | |
| 707 | /* |
| 708 | * zynqmp_r5_parse_fw() |
| 709 | * @rproc: single R5 core's corresponding rproc instance |
| 710 | * @fw: ptr to firmware to be loaded onto r5 core |
| 711 | * |
| 712 | * get resource table if available |
| 713 | * |
| 714 | * return 0 on success, otherwise non-zero value on failure |
| 715 | */ |
| 716 | static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct firmware *fw) |
| 717 | { |
| 718 | int ret; |
| 719 | |
| 720 | ret = rproc_elf_load_rsc_table(rproc, fw); |
| 721 | if (ret == -EINVAL) { |
| 722 | /* |
| 723 | * resource table only required for IPC. |
| 724 | * if not present, this is not necessarily an error; |
| 725 | * for example, loading r5 hello world application |
| 726 | * so simply inform user and keep going. |
| 727 | */ |
| 728 | dev_info(&rproc->dev, "no resource table found.\n" ); |
| 729 | ret = 0; |
| 730 | } |
| 731 | return ret; |
| 732 | } |
| 733 | |
| 734 | /** |
| 735 | * zynqmp_r5_rproc_prepare() |
| 736 | * adds carveouts for TCM bank and reserved memory regions |
| 737 | * |
| 738 | * @rproc: Device node of each rproc |
| 739 | * |
| 740 | * Return: 0 for success else < 0 error code |
| 741 | */ |
| 742 | static int zynqmp_r5_rproc_prepare(struct rproc *rproc) |
| 743 | { |
| 744 | int ret; |
| 745 | |
| 746 | ret = add_tcm_banks(rproc); |
| 747 | if (ret) { |
| 748 | dev_err(&rproc->dev, "failed to get TCM banks, err %d\n" , ret); |
| 749 | return ret; |
| 750 | } |
| 751 | |
| 752 | ret = add_mem_regions_carveout(rproc); |
| 753 | if (ret) { |
| 754 | dev_err(&rproc->dev, "failed to get reserve mem regions %d\n" , ret); |
| 755 | return ret; |
| 756 | } |
| 757 | |
| 758 | ret = add_sram_carveouts(rproc); |
| 759 | if (ret) { |
| 760 | dev_err(&rproc->dev, "failed to get sram carveout %d\n" , ret); |
| 761 | return ret; |
| 762 | } |
| 763 | |
| 764 | return 0; |
| 765 | } |
| 766 | |
| 767 | /** |
| 768 | * zynqmp_r5_rproc_unprepare() |
| 769 | * Turns off TCM banks using power-domain id |
| 770 | * |
| 771 | * @rproc: Device node of each rproc |
| 772 | * |
| 773 | * Return: always 0 |
| 774 | */ |
| 775 | static int zynqmp_r5_rproc_unprepare(struct rproc *rproc) |
| 776 | { |
| 777 | struct zynqmp_r5_core *r5_core; |
| 778 | u32 pm_domain_id; |
| 779 | int i; |
| 780 | |
| 781 | r5_core = rproc->priv; |
| 782 | |
| 783 | for (i = 0; i < r5_core->tcm_bank_count; i++) { |
| 784 | pm_domain_id = r5_core->tcm_banks[i]->pm_domain_id; |
| 785 | if (zynqmp_pm_release_node(node: pm_domain_id)) |
| 786 | dev_warn(r5_core->dev, |
| 787 | "can't turn off TCM bank 0x%x" , pm_domain_id); |
| 788 | } |
| 789 | |
| 790 | return 0; |
| 791 | } |
| 792 | |
| 793 | static struct resource_table *zynqmp_r5_get_loaded_rsc_table(struct rproc *rproc, |
| 794 | size_t *size) |
| 795 | { |
| 796 | struct zynqmp_r5_core *r5_core; |
| 797 | |
| 798 | r5_core = rproc->priv; |
| 799 | |
| 800 | *size = r5_core->rsc_tbl_size; |
| 801 | |
| 802 | return (struct resource_table *)r5_core->rsc_tbl_va; |
| 803 | } |
| 804 | |
| 805 | static int zynqmp_r5_get_rsc_table_va(struct zynqmp_r5_core *r5_core) |
| 806 | { |
| 807 | struct resource_table *rsc_tbl_addr; |
| 808 | struct device *dev = r5_core->dev; |
| 809 | struct rsc_tbl_data *rsc_data_va; |
| 810 | struct resource res_mem; |
| 811 | struct device_node *np; |
| 812 | int ret; |
| 813 | |
| 814 | /* |
| 815 | * It is expected from remote processor firmware to provide resource |
| 816 | * table address via struct rsc_tbl_data data structure. |
| 817 | * Start address of first entry under "memory-region" property list |
| 818 | * contains that data structure which holds resource table address, size |
| 819 | * and some magic number to validate correct resource table entry. |
| 820 | */ |
| 821 | np = of_parse_phandle(np: r5_core->np, phandle_name: "memory-region" , index: 0); |
| 822 | if (!np) { |
| 823 | dev_err(dev, "failed to get memory region dev node\n" ); |
| 824 | return -EINVAL; |
| 825 | } |
| 826 | |
| 827 | ret = of_address_to_resource(dev: np, index: 0, r: &res_mem); |
| 828 | of_node_put(node: np); |
| 829 | if (ret) { |
| 830 | dev_err(dev, "failed to get memory-region resource addr\n" ); |
| 831 | return -EINVAL; |
| 832 | } |
| 833 | |
| 834 | rsc_data_va = (struct rsc_tbl_data *)ioremap_wc(offset: res_mem.start, |
| 835 | size: sizeof(struct rsc_tbl_data)); |
| 836 | if (!rsc_data_va) { |
| 837 | dev_err(dev, "failed to map resource table data address\n" ); |
| 838 | return -EIO; |
| 839 | } |
| 840 | |
| 841 | /* |
| 842 | * If RSC_TBL_XLNX_MAGIC number and its complement isn't found then |
| 843 | * do not consider resource table address valid and don't attach |
| 844 | */ |
| 845 | if (rsc_data_va->magic_num != RSC_TBL_XLNX_MAGIC || |
| 846 | rsc_data_va->comp_magic_num != ~RSC_TBL_XLNX_MAGIC) { |
| 847 | dev_dbg(dev, "invalid magic number, won't attach\n" ); |
| 848 | return -EINVAL; |
| 849 | } |
| 850 | |
| 851 | r5_core->rsc_tbl_va = ioremap_wc(offset: rsc_data_va->rsc_tbl, |
| 852 | size: rsc_data_va->rsc_tbl_size); |
| 853 | if (!r5_core->rsc_tbl_va) { |
| 854 | dev_err(dev, "failed to get resource table va\n" ); |
| 855 | return -EINVAL; |
| 856 | } |
| 857 | |
| 858 | rsc_tbl_addr = (struct resource_table *)r5_core->rsc_tbl_va; |
| 859 | |
| 860 | /* |
| 861 | * As of now resource table version 1 is expected. Don't fail to attach |
| 862 | * but warn users about it. |
| 863 | */ |
| 864 | if (rsc_tbl_addr->ver != 1) |
| 865 | dev_warn(dev, "unexpected resource table version %d\n" , |
| 866 | rsc_tbl_addr->ver); |
| 867 | |
| 868 | r5_core->rsc_tbl_size = rsc_data_va->rsc_tbl_size; |
| 869 | |
| 870 | iounmap(addr: (void __iomem *)rsc_data_va); |
| 871 | |
| 872 | return 0; |
| 873 | } |
| 874 | |
| 875 | static int zynqmp_r5_attach(struct rproc *rproc) |
| 876 | { |
| 877 | dev_dbg(&rproc->dev, "rproc %d attached\n" , rproc->index); |
| 878 | |
| 879 | return 0; |
| 880 | } |
| 881 | |
| 882 | static int zynqmp_r5_detach(struct rproc *rproc) |
| 883 | { |
| 884 | /* |
| 885 | * Generate last notification to remote after clearing virtio flag. |
| 886 | * Remote can avoid polling on virtio reset flag if kick is generated |
| 887 | * during detach by host and check virtio reset flag on kick interrupt. |
| 888 | */ |
| 889 | zynqmp_r5_rproc_kick(rproc, vqid: 0); |
| 890 | |
| 891 | return 0; |
| 892 | } |
| 893 | |
| 894 | static const struct rproc_ops zynqmp_r5_rproc_ops = { |
| 895 | .prepare = zynqmp_r5_rproc_prepare, |
| 896 | .unprepare = zynqmp_r5_rproc_unprepare, |
| 897 | .start = zynqmp_r5_rproc_start, |
| 898 | .stop = zynqmp_r5_rproc_stop, |
| 899 | .load = rproc_elf_load_segments, |
| 900 | .parse_fw = zynqmp_r5_parse_fw, |
| 901 | .find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table, |
| 902 | .sanity_check = rproc_elf_sanity_check, |
| 903 | .get_boot_addr = rproc_elf_get_boot_addr, |
| 904 | .kick = zynqmp_r5_rproc_kick, |
| 905 | .get_loaded_rsc_table = zynqmp_r5_get_loaded_rsc_table, |
| 906 | .attach = zynqmp_r5_attach, |
| 907 | .detach = zynqmp_r5_detach, |
| 908 | }; |
| 909 | |
| 910 | /** |
| 911 | * zynqmp_r5_add_rproc_core() |
| 912 | * Allocate and add struct rproc object for each r5f core |
| 913 | * This is called for each individual r5f core |
| 914 | * |
| 915 | * @cdev: Device node of each r5 core |
| 916 | * |
| 917 | * Return: zynqmp_r5_core object for success else error code pointer |
| 918 | */ |
| 919 | static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev) |
| 920 | { |
| 921 | struct zynqmp_r5_core *r5_core; |
| 922 | struct rproc *r5_rproc; |
| 923 | int ret; |
| 924 | |
| 925 | /* Set up DMA mask */ |
| 926 | ret = dma_set_coherent_mask(dev: cdev, DMA_BIT_MASK(32)); |
| 927 | if (ret) |
| 928 | return ERR_PTR(error: ret); |
| 929 | |
| 930 | /* Allocate remoteproc instance */ |
| 931 | r5_rproc = rproc_alloc(dev: cdev, name: dev_name(dev: cdev), |
| 932 | ops: &zynqmp_r5_rproc_ops, |
| 933 | NULL, len: sizeof(struct zynqmp_r5_core)); |
| 934 | if (!r5_rproc) { |
| 935 | dev_err(cdev, "failed to allocate memory for rproc instance\n" ); |
| 936 | return ERR_PTR(error: -ENOMEM); |
| 937 | } |
| 938 | |
| 939 | rproc_coredump_set_elf_info(rproc: r5_rproc, ELFCLASS32, EM_ARM); |
| 940 | |
| 941 | r5_rproc->auto_boot = false; |
| 942 | r5_core = r5_rproc->priv; |
| 943 | r5_core->dev = cdev; |
| 944 | r5_core->np = dev_of_node(dev: cdev); |
| 945 | if (!r5_core->np) { |
| 946 | dev_err(cdev, "can't get device node for r5 core\n" ); |
| 947 | ret = -EINVAL; |
| 948 | goto free_rproc; |
| 949 | } |
| 950 | |
| 951 | /* Add R5 remoteproc core */ |
| 952 | ret = rproc_add(rproc: r5_rproc); |
| 953 | if (ret) { |
| 954 | dev_err(cdev, "failed to add r5 remoteproc\n" ); |
| 955 | goto free_rproc; |
| 956 | } |
| 957 | |
| 958 | /* |
| 959 | * If firmware is already available in the memory then move rproc state |
| 960 | * to DETACHED. Firmware can be preloaded via debugger or by any other |
| 961 | * agent (processors) in the system. |
| 962 | * If firmware isn't available in the memory and resource table isn't |
| 963 | * found, then rproc state remains OFFLINE. |
| 964 | */ |
| 965 | if (!zynqmp_r5_get_rsc_table_va(r5_core)) |
| 966 | r5_rproc->state = RPROC_DETACHED; |
| 967 | |
| 968 | r5_core->rproc = r5_rproc; |
| 969 | return r5_core; |
| 970 | |
| 971 | free_rproc: |
| 972 | rproc_free(rproc: r5_rproc); |
| 973 | return ERR_PTR(error: ret); |
| 974 | } |
| 975 | |
| 976 | static int zynqmp_r5_get_sram_banks(struct zynqmp_r5_core *r5_core) |
| 977 | { |
| 978 | struct device_node *np = r5_core->np; |
| 979 | struct device *dev = r5_core->dev; |
| 980 | struct zynqmp_sram_bank *sram; |
| 981 | struct device_node *sram_np; |
| 982 | int num_sram, i, ret; |
| 983 | u64 abs_addr, size; |
| 984 | |
| 985 | /* "sram" is optional property. Do not fail, if unavailable. */ |
| 986 | if (!of_property_present(np: r5_core->np, propname: "sram" )) |
| 987 | return 0; |
| 988 | |
| 989 | num_sram = of_property_count_elems_of_size(np, propname: "sram" , elem_size: sizeof(phandle)); |
| 990 | if (num_sram <= 0) { |
| 991 | dev_err(dev, "Invalid sram property, ret = %d\n" , |
| 992 | num_sram); |
| 993 | return -EINVAL; |
| 994 | } |
| 995 | |
| 996 | sram = devm_kcalloc(dev, n: num_sram, |
| 997 | size: sizeof(struct zynqmp_sram_bank), GFP_KERNEL); |
| 998 | if (!sram) |
| 999 | return -ENOMEM; |
| 1000 | |
| 1001 | for (i = 0; i < num_sram; i++) { |
| 1002 | sram_np = of_parse_phandle(np, phandle_name: "sram" , index: i); |
| 1003 | if (!sram_np) { |
| 1004 | dev_err(dev, "failed to get sram %d phandle\n" , i); |
| 1005 | return -EINVAL; |
| 1006 | } |
| 1007 | |
| 1008 | if (!of_device_is_available(device: sram_np)) { |
| 1009 | dev_err(dev, "sram device not available\n" ); |
| 1010 | ret = -EINVAL; |
| 1011 | goto fail_sram_get; |
| 1012 | } |
| 1013 | |
| 1014 | ret = of_address_to_resource(dev: sram_np, index: 0, r: &sram[i].sram_res); |
| 1015 | if (ret) { |
| 1016 | dev_err(dev, "addr to res failed\n" ); |
| 1017 | goto fail_sram_get; |
| 1018 | } |
| 1019 | |
| 1020 | /* Get SRAM device address */ |
| 1021 | ret = of_property_read_reg(np: sram_np, idx: i, addr: &abs_addr, size: &size); |
| 1022 | if (ret) { |
| 1023 | dev_err(dev, "failed to get reg property\n" ); |
| 1024 | goto fail_sram_get; |
| 1025 | } |
| 1026 | |
| 1027 | sram[i].da = (u32)abs_addr; |
| 1028 | |
| 1029 | of_node_put(node: sram_np); |
| 1030 | |
| 1031 | dev_dbg(dev, "sram %d: name=%s, addr=0x%llx, da=0x%x, size=0x%llx\n" , |
| 1032 | i, sram[i].sram_res.name, sram[i].sram_res.start, |
| 1033 | sram[i].da, resource_size(&sram[i].sram_res)); |
| 1034 | } |
| 1035 | |
| 1036 | r5_core->sram = sram; |
| 1037 | r5_core->num_sram = num_sram; |
| 1038 | |
| 1039 | return 0; |
| 1040 | |
| 1041 | fail_sram_get: |
| 1042 | of_node_put(node: sram_np); |
| 1043 | |
| 1044 | return ret; |
| 1045 | } |
| 1046 | |
| 1047 | static int zynqmp_r5_get_tcm_node_from_dt(struct zynqmp_r5_cluster *cluster) |
| 1048 | { |
| 1049 | int i, j, tcm_bank_count, ret, tcm_pd_idx, pd_count; |
| 1050 | struct of_phandle_args out_args; |
| 1051 | struct zynqmp_r5_core *r5_core; |
| 1052 | struct platform_device *cpdev; |
| 1053 | struct mem_bank_data *tcm; |
| 1054 | struct device_node *np; |
| 1055 | struct resource *res; |
| 1056 | u64 abs_addr, size; |
| 1057 | struct device *dev; |
| 1058 | |
| 1059 | for (i = 0; i < cluster->core_count; i++) { |
| 1060 | r5_core = cluster->r5_cores[i]; |
| 1061 | dev = r5_core->dev; |
| 1062 | np = r5_core->np; |
| 1063 | |
| 1064 | pd_count = of_count_phandle_with_args(np, list_name: "power-domains" , |
| 1065 | cells_name: "#power-domain-cells" ); |
| 1066 | |
| 1067 | if (pd_count <= 0) { |
| 1068 | dev_err(dev, "invalid power-domains property, %d\n" , pd_count); |
| 1069 | return -EINVAL; |
| 1070 | } |
| 1071 | |
| 1072 | /* First entry in power-domains list is for r5 core, rest for TCM. */ |
| 1073 | tcm_bank_count = pd_count - 1; |
| 1074 | |
| 1075 | if (tcm_bank_count <= 0) { |
| 1076 | dev_err(dev, "invalid TCM count %d\n" , tcm_bank_count); |
| 1077 | return -EINVAL; |
| 1078 | } |
| 1079 | |
| 1080 | r5_core->tcm_banks = devm_kcalloc(dev, n: tcm_bank_count, |
| 1081 | size: sizeof(struct mem_bank_data *), |
| 1082 | GFP_KERNEL); |
| 1083 | if (!r5_core->tcm_banks) |
| 1084 | return -ENOMEM; |
| 1085 | |
| 1086 | r5_core->tcm_bank_count = tcm_bank_count; |
| 1087 | for (j = 0, tcm_pd_idx = 1; j < tcm_bank_count; j++, tcm_pd_idx++) { |
| 1088 | tcm = devm_kzalloc(dev, size: sizeof(struct mem_bank_data), |
| 1089 | GFP_KERNEL); |
| 1090 | if (!tcm) |
| 1091 | return -ENOMEM; |
| 1092 | |
| 1093 | r5_core->tcm_banks[j] = tcm; |
| 1094 | |
| 1095 | /* Get power-domains id of TCM. */ |
| 1096 | ret = of_parse_phandle_with_args(np, list_name: "power-domains" , |
| 1097 | cells_name: "#power-domain-cells" , |
| 1098 | index: tcm_pd_idx, out_args: &out_args); |
| 1099 | if (ret) { |
| 1100 | dev_err(r5_core->dev, |
| 1101 | "failed to get tcm %d pm domain, ret %d\n" , |
| 1102 | tcm_pd_idx, ret); |
| 1103 | return ret; |
| 1104 | } |
| 1105 | tcm->pm_domain_id = out_args.args[0]; |
| 1106 | of_node_put(node: out_args.np); |
| 1107 | |
| 1108 | /* Get TCM address without translation. */ |
| 1109 | ret = of_property_read_reg(np, idx: j, addr: &abs_addr, size: &size); |
| 1110 | if (ret) { |
| 1111 | dev_err(dev, "failed to get reg property\n" ); |
| 1112 | return ret; |
| 1113 | } |
| 1114 | |
| 1115 | /* |
| 1116 | * Remote processor can address only 32 bits |
| 1117 | * so convert 64-bits into 32-bits. This will discard |
| 1118 | * any unwanted upper 32-bits. |
| 1119 | */ |
| 1120 | tcm->da = (u32)abs_addr; |
| 1121 | tcm->size = (u32)size; |
| 1122 | |
| 1123 | cpdev = to_platform_device(dev); |
| 1124 | res = platform_get_resource(cpdev, IORESOURCE_MEM, j); |
| 1125 | if (!res) { |
| 1126 | dev_err(dev, "failed to get tcm resource\n" ); |
| 1127 | return -EINVAL; |
| 1128 | } |
| 1129 | |
| 1130 | tcm->addr = (u32)res->start; |
| 1131 | tcm->bank_name = (char *)res->name; |
| 1132 | res = devm_request_mem_region(dev, tcm->addr, tcm->size, |
| 1133 | tcm->bank_name); |
| 1134 | if (!res) { |
| 1135 | dev_err(dev, "failed to request tcm resource\n" ); |
| 1136 | return -EINVAL; |
| 1137 | } |
| 1138 | } |
| 1139 | } |
| 1140 | |
| 1141 | return 0; |
| 1142 | } |
| 1143 | |
| 1144 | /** |
| 1145 | * zynqmp_r5_get_tcm_node() |
| 1146 | * Ideally this function should parse tcm node and store information |
| 1147 | * in r5_core instance. For now, Hardcoded TCM information is used. |
| 1148 | * This approach is used as TCM bindings for system-dt is being developed |
| 1149 | * |
| 1150 | * @cluster: pointer to zynqmp_r5_cluster type object |
| 1151 | * |
| 1152 | * Return: 0 for success and < 0 error code for failure. |
| 1153 | */ |
| 1154 | static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster) |
| 1155 | { |
| 1156 | const struct mem_bank_data *zynqmp_tcm_banks; |
| 1157 | struct device *dev = cluster->dev; |
| 1158 | struct zynqmp_r5_core *r5_core; |
| 1159 | int tcm_bank_count, tcm_node; |
| 1160 | int i, j; |
| 1161 | |
| 1162 | if (cluster->mode == SPLIT_MODE) { |
| 1163 | zynqmp_tcm_banks = zynqmp_tcm_banks_split; |
| 1164 | tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks_split); |
| 1165 | } else { |
| 1166 | zynqmp_tcm_banks = zynqmp_tcm_banks_lockstep; |
| 1167 | tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks_lockstep); |
| 1168 | } |
| 1169 | |
| 1170 | /* count per core tcm banks */ |
| 1171 | tcm_bank_count = tcm_bank_count / cluster->core_count; |
| 1172 | |
| 1173 | /* |
| 1174 | * r5 core 0 will use all of TCM banks in lockstep mode. |
| 1175 | * In split mode, r5 core0 will use 128k and r5 core1 will use another |
| 1176 | * 128k. Assign TCM banks to each core accordingly |
| 1177 | */ |
| 1178 | tcm_node = 0; |
| 1179 | for (i = 0; i < cluster->core_count; i++) { |
| 1180 | r5_core = cluster->r5_cores[i]; |
| 1181 | r5_core->tcm_banks = devm_kcalloc(dev, n: tcm_bank_count, |
| 1182 | size: sizeof(struct mem_bank_data *), |
| 1183 | GFP_KERNEL); |
| 1184 | if (!r5_core->tcm_banks) |
| 1185 | return -ENOMEM; |
| 1186 | |
| 1187 | for (j = 0; j < tcm_bank_count; j++) { |
| 1188 | /* |
| 1189 | * Use pre-defined TCM reg values. |
| 1190 | * Eventually this should be replaced by values |
| 1191 | * parsed from dts. |
| 1192 | */ |
| 1193 | r5_core->tcm_banks[j] = |
| 1194 | (struct mem_bank_data *)&zynqmp_tcm_banks[tcm_node]; |
| 1195 | tcm_node++; |
| 1196 | } |
| 1197 | |
| 1198 | r5_core->tcm_bank_count = tcm_bank_count; |
| 1199 | } |
| 1200 | |
| 1201 | return 0; |
| 1202 | } |
| 1203 | |
| 1204 | /* |
| 1205 | * zynqmp_r5_core_init() |
| 1206 | * Create and initialize zynqmp_r5_core type object |
| 1207 | * |
| 1208 | * @cluster: pointer to zynqmp_r5_cluster type object |
| 1209 | * @fw_reg_val: value expected by firmware to configure RPU cluster mode |
| 1210 | * @tcm_mode: value expected by fw to configure TCM mode (lockstep or split) |
| 1211 | * |
| 1212 | * Return: 0 for success and error code for failure. |
| 1213 | */ |
| 1214 | static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster, |
| 1215 | enum rpu_oper_mode fw_reg_val, |
| 1216 | enum rpu_tcm_comb tcm_mode) |
| 1217 | { |
| 1218 | struct device *dev = cluster->dev; |
| 1219 | struct zynqmp_r5_core *r5_core; |
| 1220 | int ret = -EINVAL, i; |
| 1221 | |
| 1222 | r5_core = cluster->r5_cores[0]; |
| 1223 | |
| 1224 | /* Maintain backward compatibility for zynqmp by using hardcode TCM address. */ |
| 1225 | if (of_property_present(np: r5_core->np, propname: "reg" )) |
| 1226 | ret = zynqmp_r5_get_tcm_node_from_dt(cluster); |
| 1227 | else if (device_is_compatible(dev, compat: "xlnx,zynqmp-r5fss" )) |
| 1228 | ret = zynqmp_r5_get_tcm_node(cluster); |
| 1229 | |
| 1230 | if (ret) { |
| 1231 | dev_err(dev, "can't get tcm, err %d\n" , ret); |
| 1232 | return ret; |
| 1233 | } |
| 1234 | |
| 1235 | for (i = 0; i < cluster->core_count; i++) { |
| 1236 | r5_core = cluster->r5_cores[i]; |
| 1237 | |
| 1238 | /* Initialize r5 cores with power-domains parsed from dts */ |
| 1239 | ret = of_property_read_u32_index(np: r5_core->np, propname: "power-domains" , |
| 1240 | index: 1, out_value: &r5_core->pm_domain_id); |
| 1241 | if (ret) { |
| 1242 | dev_err(dev, "failed to get power-domains property\n" ); |
| 1243 | return ret; |
| 1244 | } |
| 1245 | |
| 1246 | ret = zynqmp_pm_set_rpu_mode(node_id: r5_core->pm_domain_id, rpu_mode: fw_reg_val); |
| 1247 | if (ret < 0) { |
| 1248 | dev_err(r5_core->dev, "failed to set RPU mode\n" ); |
| 1249 | return ret; |
| 1250 | } |
| 1251 | |
| 1252 | if (of_property_present(np: dev_of_node(dev), propname: "xlnx,tcm-mode" ) || |
| 1253 | device_is_compatible(dev, compat: "xlnx,zynqmp-r5fss" )) { |
| 1254 | ret = zynqmp_pm_set_tcm_config(node_id: r5_core->pm_domain_id, |
| 1255 | tcm_mode); |
| 1256 | if (ret < 0) { |
| 1257 | dev_err(r5_core->dev, "failed to configure TCM\n" ); |
| 1258 | return ret; |
| 1259 | } |
| 1260 | } |
| 1261 | |
| 1262 | ret = zynqmp_r5_get_sram_banks(r5_core); |
| 1263 | if (ret) |
| 1264 | return ret; |
| 1265 | } |
| 1266 | |
| 1267 | return 0; |
| 1268 | } |
| 1269 | |
| 1270 | /* |
| 1271 | * zynqmp_r5_cluster_init() |
| 1272 | * Create and initialize zynqmp_r5_cluster type object |
| 1273 | * |
| 1274 | * @cluster: pointer to zynqmp_r5_cluster type object |
| 1275 | * |
| 1276 | * Return: 0 for success and error code for failure. |
| 1277 | */ |
| 1278 | static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster) |
| 1279 | { |
| 1280 | enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE; |
| 1281 | struct device *dev = cluster->dev; |
| 1282 | struct device_node *dev_node = dev_of_node(dev); |
| 1283 | struct platform_device *child_pdev; |
| 1284 | struct zynqmp_r5_core **r5_cores; |
| 1285 | enum rpu_oper_mode fw_reg_val; |
| 1286 | struct device **child_devs; |
| 1287 | struct device_node *child; |
| 1288 | enum rpu_tcm_comb tcm_mode; |
| 1289 | int core_count, ret, i; |
| 1290 | struct mbox_info *ipi; |
| 1291 | |
| 1292 | ret = of_property_read_u32(np: dev_node, propname: "xlnx,cluster-mode" , out_value: &cluster_mode); |
| 1293 | |
| 1294 | /* |
| 1295 | * on success returns 0, if not defined then returns -EINVAL, |
| 1296 | * In that case, default is LOCKSTEP mode. Other than that |
| 1297 | * returns relative error code < 0. |
| 1298 | */ |
| 1299 | if (ret != -EINVAL && ret != 0) { |
| 1300 | dev_err(dev, "Invalid xlnx,cluster-mode property\n" ); |
| 1301 | return ret; |
| 1302 | } |
| 1303 | |
| 1304 | /* |
| 1305 | * For now driver only supports split mode and lockstep mode. |
| 1306 | * fail driver probe if either of that is not set in dts. |
| 1307 | */ |
| 1308 | if (cluster_mode == LOCKSTEP_MODE) { |
| 1309 | fw_reg_val = PM_RPU_MODE_LOCKSTEP; |
| 1310 | } else if (cluster_mode == SPLIT_MODE) { |
| 1311 | fw_reg_val = PM_RPU_MODE_SPLIT; |
| 1312 | } else { |
| 1313 | dev_err(dev, "driver does not support cluster mode %d\n" , cluster_mode); |
| 1314 | return -EINVAL; |
| 1315 | } |
| 1316 | |
| 1317 | if (of_property_present(np: dev_node, propname: "xlnx,tcm-mode" )) { |
| 1318 | ret = of_property_read_u32(np: dev_node, propname: "xlnx,tcm-mode" , out_value: (u32 *)&tcm_mode); |
| 1319 | if (ret) |
| 1320 | return ret; |
| 1321 | } else if (device_is_compatible(dev, compat: "xlnx,zynqmp-r5fss" )) { |
| 1322 | if (cluster_mode == LOCKSTEP_MODE) |
| 1323 | tcm_mode = PM_RPU_TCM_COMB; |
| 1324 | else |
| 1325 | tcm_mode = PM_RPU_TCM_SPLIT; |
| 1326 | } else { |
| 1327 | tcm_mode = PM_RPU_TCM_COMB; |
| 1328 | } |
| 1329 | |
| 1330 | /* |
| 1331 | * Number of cores is decided by number of child nodes of |
| 1332 | * r5f subsystem node in dts. If Split mode is used in dts |
| 1333 | * 2 child nodes are expected. |
| 1334 | * In lockstep mode if two child nodes are available, |
| 1335 | * only use first child node and consider it as core0 |
| 1336 | * and ignore core1 dt node. |
| 1337 | */ |
| 1338 | core_count = of_get_available_child_count(np: dev_node); |
| 1339 | if (core_count == 0) { |
| 1340 | dev_err(dev, "Invalid number of r5 cores %d" , core_count); |
| 1341 | return -EINVAL; |
| 1342 | } else if (cluster_mode == SPLIT_MODE && core_count != 2) { |
| 1343 | dev_err(dev, "Invalid number of r5 cores for split mode\n" ); |
| 1344 | return -EINVAL; |
| 1345 | } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) { |
| 1346 | dev_warn(dev, "Only r5 core0 will be used\n" ); |
| 1347 | core_count = 1; |
| 1348 | } |
| 1349 | |
| 1350 | child_devs = kcalloc(core_count, sizeof(struct device *), GFP_KERNEL); |
| 1351 | if (!child_devs) |
| 1352 | return -ENOMEM; |
| 1353 | |
| 1354 | r5_cores = kcalloc(core_count, |
| 1355 | sizeof(struct zynqmp_r5_core *), GFP_KERNEL); |
| 1356 | if (!r5_cores) { |
| 1357 | kfree(objp: child_devs); |
| 1358 | return -ENOMEM; |
| 1359 | } |
| 1360 | |
| 1361 | i = 0; |
| 1362 | for_each_available_child_of_node(dev_node, child) { |
| 1363 | child_pdev = of_find_device_by_node(np: child); |
| 1364 | if (!child_pdev) { |
| 1365 | of_node_put(node: child); |
| 1366 | ret = -ENODEV; |
| 1367 | goto release_r5_cores; |
| 1368 | } |
| 1369 | |
| 1370 | child_devs[i] = &child_pdev->dev; |
| 1371 | |
| 1372 | /* create and add remoteproc instance of type struct rproc */ |
| 1373 | r5_cores[i] = zynqmp_r5_add_rproc_core(cdev: &child_pdev->dev); |
| 1374 | if (IS_ERR(ptr: r5_cores[i])) { |
| 1375 | of_node_put(node: child); |
| 1376 | ret = PTR_ERR(ptr: r5_cores[i]); |
| 1377 | r5_cores[i] = NULL; |
| 1378 | goto release_r5_cores; |
| 1379 | } |
| 1380 | |
| 1381 | /* |
| 1382 | * If mailbox nodes are disabled using "status" property then |
| 1383 | * setting up mailbox channels will fail. |
| 1384 | */ |
| 1385 | ipi = zynqmp_r5_setup_mbox(cdev: &child_pdev->dev); |
| 1386 | if (ipi) { |
| 1387 | r5_cores[i]->ipi = ipi; |
| 1388 | ipi->r5_core = r5_cores[i]; |
| 1389 | } |
| 1390 | |
| 1391 | /* |
| 1392 | * If two child nodes are available in dts in lockstep mode, |
| 1393 | * then ignore second child node. |
| 1394 | */ |
| 1395 | if (cluster_mode == LOCKSTEP_MODE) { |
| 1396 | of_node_put(node: child); |
| 1397 | break; |
| 1398 | } |
| 1399 | |
| 1400 | i++; |
| 1401 | } |
| 1402 | |
| 1403 | cluster->mode = cluster_mode; |
| 1404 | cluster->core_count = core_count; |
| 1405 | cluster->r5_cores = r5_cores; |
| 1406 | |
| 1407 | ret = zynqmp_r5_core_init(cluster, fw_reg_val, tcm_mode); |
| 1408 | if (ret < 0) { |
| 1409 | dev_err(dev, "failed to init r5 core err %d\n" , ret); |
| 1410 | cluster->core_count = 0; |
| 1411 | cluster->r5_cores = NULL; |
| 1412 | |
| 1413 | /* |
| 1414 | * at this point rproc resources for each core are allocated. |
| 1415 | * adjust index to free resources in reverse order |
| 1416 | */ |
| 1417 | i = core_count - 1; |
| 1418 | goto release_r5_cores; |
| 1419 | } |
| 1420 | |
| 1421 | kfree(objp: child_devs); |
| 1422 | return 0; |
| 1423 | |
| 1424 | release_r5_cores: |
| 1425 | while (i >= 0) { |
| 1426 | put_device(dev: child_devs[i]); |
| 1427 | if (r5_cores[i]) { |
| 1428 | zynqmp_r5_free_mbox(ipi: r5_cores[i]->ipi); |
| 1429 | of_reserved_mem_device_release(dev: r5_cores[i]->dev); |
| 1430 | rproc_del(rproc: r5_cores[i]->rproc); |
| 1431 | rproc_free(rproc: r5_cores[i]->rproc); |
| 1432 | } |
| 1433 | i--; |
| 1434 | } |
| 1435 | kfree(objp: r5_cores); |
| 1436 | kfree(objp: child_devs); |
| 1437 | return ret; |
| 1438 | } |
| 1439 | |
| 1440 | static void zynqmp_r5_cluster_exit(void *data) |
| 1441 | { |
| 1442 | struct platform_device *pdev = data; |
| 1443 | struct zynqmp_r5_cluster *cluster; |
| 1444 | struct zynqmp_r5_core *r5_core; |
| 1445 | int i; |
| 1446 | |
| 1447 | cluster = platform_get_drvdata(pdev); |
| 1448 | if (!cluster) |
| 1449 | return; |
| 1450 | |
| 1451 | for (i = 0; i < cluster->core_count; i++) { |
| 1452 | r5_core = cluster->r5_cores[i]; |
| 1453 | zynqmp_r5_free_mbox(ipi: r5_core->ipi); |
| 1454 | iounmap(addr: r5_core->rsc_tbl_va); |
| 1455 | of_reserved_mem_device_release(dev: r5_core->dev); |
| 1456 | put_device(dev: r5_core->dev); |
| 1457 | rproc_del(rproc: r5_core->rproc); |
| 1458 | rproc_free(rproc: r5_core->rproc); |
| 1459 | } |
| 1460 | |
| 1461 | kfree(objp: cluster->r5_cores); |
| 1462 | kfree(objp: cluster); |
| 1463 | platform_set_drvdata(pdev, NULL); |
| 1464 | } |
| 1465 | |
| 1466 | /* |
| 1467 | * zynqmp_r5_remoteproc_probe() |
| 1468 | * parse device-tree, initialize hardware and allocate required resources |
| 1469 | * and remoteproc ops |
| 1470 | * |
| 1471 | * @pdev: domain platform device for R5 cluster |
| 1472 | * |
| 1473 | * Return: 0 for success and < 0 for failure. |
| 1474 | */ |
| 1475 | static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev) |
| 1476 | { |
| 1477 | struct zynqmp_r5_cluster *cluster; |
| 1478 | struct device *dev = &pdev->dev; |
| 1479 | int ret; |
| 1480 | |
| 1481 | cluster = kzalloc(sizeof(*cluster), GFP_KERNEL); |
| 1482 | if (!cluster) |
| 1483 | return -ENOMEM; |
| 1484 | |
| 1485 | cluster->dev = dev; |
| 1486 | |
| 1487 | ret = devm_of_platform_populate(dev); |
| 1488 | if (ret) { |
| 1489 | dev_err_probe(dev, err: ret, fmt: "failed to populate platform dev\n" ); |
| 1490 | kfree(objp: cluster); |
| 1491 | return ret; |
| 1492 | } |
| 1493 | |
| 1494 | /* wire in so each core can be cleaned up at driver remove */ |
| 1495 | platform_set_drvdata(pdev, data: cluster); |
| 1496 | |
| 1497 | ret = zynqmp_r5_cluster_init(cluster); |
| 1498 | if (ret) { |
| 1499 | kfree(objp: cluster); |
| 1500 | platform_set_drvdata(pdev, NULL); |
| 1501 | dev_err_probe(dev, err: ret, fmt: "Invalid r5f subsystem device tree\n" ); |
| 1502 | return ret; |
| 1503 | } |
| 1504 | |
| 1505 | ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev); |
| 1506 | if (ret) |
| 1507 | return ret; |
| 1508 | |
| 1509 | return 0; |
| 1510 | } |
| 1511 | |
| 1512 | /* Match table for OF platform binding */ |
| 1513 | static const struct of_device_id zynqmp_r5_remoteproc_match[] = { |
| 1514 | { .compatible = "xlnx,versal-net-r52fss" , }, |
| 1515 | { .compatible = "xlnx,versal-r5fss" , }, |
| 1516 | { .compatible = "xlnx,zynqmp-r5fss" , }, |
| 1517 | { /* end of list */ }, |
| 1518 | }; |
| 1519 | MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match); |
| 1520 | |
| 1521 | static struct platform_driver zynqmp_r5_remoteproc_driver = { |
| 1522 | .probe = zynqmp_r5_remoteproc_probe, |
| 1523 | .driver = { |
| 1524 | .name = "zynqmp_r5_remoteproc" , |
| 1525 | .of_match_table = zynqmp_r5_remoteproc_match, |
| 1526 | }, |
| 1527 | }; |
| 1528 | module_platform_driver(zynqmp_r5_remoteproc_driver); |
| 1529 | |
| 1530 | MODULE_DESCRIPTION("Xilinx R5F remote processor driver" ); |
| 1531 | MODULE_AUTHOR("Xilinx Inc." ); |
| 1532 | MODULE_LICENSE("GPL" ); |
| 1533 | |