From b337ee2436f8070661e07a453aec0aafdf0cf7e5 Mon Sep 17 00:00:00 2001 From: Keerthy Date: Wed, 1 Sep 2021 15:09:33 +0200 Subject: remoteproc: ipu: Add driver to bring up ipu The driver enables IPU support. Basically enables the clocks, timers, watchdog timers and bare minimal MMU and supports loading the firmware from mmc. Signed-off-by: Keerthy [Amjad: fix compile warnings] Signed-off-by: Amjad Ouled-Ameur --- drivers/remoteproc/ipu_rproc.c | 759 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 759 insertions(+) create mode 100644 drivers/remoteproc/ipu_rproc.c (limited to 'drivers/remoteproc/ipu_rproc.c') diff --git a/drivers/remoteproc/ipu_rproc.c b/drivers/remoteproc/ipu_rproc.c new file mode 100644 index 00000000000..141583dec28 --- /dev/null +++ b/drivers/remoteproc/ipu_rproc.c @@ -0,0 +1,759 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * IPU remoteproc driver for various SoCs + * + * Copyright (C) 2019 Texas Instruments Incorporated - http://www.ti.com/ + * Angela Stegmaier + * Venkateswara Rao Mandela + * Keerthy + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define IPU1_LOAD_ADDR (0xa17ff000) +#define MAX_REMOTECORE_BIN_SIZE (8 * 0x100000) + +enum ipu_num { + IPU1 = 0, + IPU2, + RPROC_END_ENUMS, +}; + +#define IPU2_LOAD_ADDR (IPU1_LOAD_ADDR + MAX_REMOTECORE_BIN_SIZE) + +#define PAGE_SHIFT 12 +#define PAGESIZE_1M 0x0 +#define PAGESIZE_64K 0x1 +#define PAGESIZE_4K 0x2 +#define PAGESIZE_16M 0x3 +#define LE 0 +#define BE 1 +#define ELEMSIZE_8 0x0 +#define ELEMSIZE_16 0x1 +#define ELEMSIZE_32 0x2 +#define MIXED_TLB 0x0 +#define MIXED_CPU 0x1 + +#define PGT_SMALLPAGE_SIZE 0x00001000 +#define PGT_LARGEPAGE_SIZE 0x00010000 +#define PGT_SECTION_SIZE 0x00100000 +#define PGT_SUPERSECTION_SIZE 0x01000000 + +#define PGT_L1_DESC_PAGE 0x00001 +#define PGT_L1_DESC_SECTION 0x00002 +#define PGT_L1_DESC_SUPERSECTION 0x40002 + +#define PGT_L1_DESC_PAGE_MASK 0xfffffC00 +#define PGT_L1_DESC_SECTION_MASK 0xfff00000 +#define PGT_L1_DESC_SUPERSECTION_MASK 0xff000000 + +#define PGT_L1_DESC_SMALLPAGE_INDEX_SHIFT 12 +#define PGT_L1_DESC_LARGEPAGE_INDEX_SHIFT 16 +#define PGT_L1_DESC_SECTION_INDEX_SHIFT 20 +#define PGT_L1_DESC_SUPERSECTION_INDEX_SHIFT 24 + +#define PGT_L2_DESC_SMALLPAGE 0x02 +#define PGT_L2_DESC_LARGEPAGE 0x01 + +#define PGT_L2_DESC_SMALLPAGE_MASK 0xfffff000 +#define PGT_L2_DESC_LARGEPAGE_MASK 0xffff0000 + +/* + * The memory for the page tables (256 KB per IPU) is placed just before + * the carveout memories for the remote processors. 16 KB of memory is + * needed for the L1 page table (4096 entries * 4 bytes per 1 MB section). + * Any smaller page (64 KB or 4 KB) entries are supported through L2 page + * tables (1 KB per table). The remaining 240 KB can provide support for + * 240 L2 page tables. Any remoteproc firmware image requiring more than + * 240 L2 page table entries would need more memory to be reserved. + */ +#define PAGE_TABLE_SIZE_L1 (0x00004000) +#define PAGE_TABLE_SIZE_L2 (0x400) +#define MAX_NUM_L2_PAGE_TABLES (240) +#define PAGE_TABLE_SIZE_L2_TOTAL (MAX_NUM_L2_PAGE_TABLES * PAGE_TABLE_SIZE_L2) +#define PAGE_TABLE_SIZE (PAGE_TABLE_SIZE_L1 + (PAGE_TABLE_SIZE_L2_TOTAL)) + +/** + * struct omap_rproc_mem - internal memory structure + * @cpu_addr: MPU virtual address of the memory region + * @bus_addr: bus address used to access the memory region + * @dev_addr: device address of the memory region from DSP view + * @size: size of the memory region + */ +struct omap_rproc_mem { + void __iomem *cpu_addr; + phys_addr_t bus_addr; + u32 dev_addr; + size_t size; +}; + +struct ipu_privdata { + struct omap_rproc_mem mem; + struct list_head mappings; + const char *fw_name; + u32 bootaddr; + int id; + struct udevice *rdev; +}; + +typedef int (*handle_resource_t) (void *, int offset, int avail); + +unsigned int *page_table_l1 = (unsigned int *)0x0; +unsigned int *page_table_l2 = (unsigned int *)0x0; + +/* + * Set maximum carveout size to 96 MB + */ +#define DRA7_RPROC_MAX_CO_SIZE (96 * 0x100000) + +/* + * These global variables are used for deriving the MMU page tables. They + * are initialized for each core with the appropriate values. The length + * of the array mem_bitmap is set as per a 96 MB carveout which the + * maximum set aside in the current memory map. + */ +unsigned long mem_base; +unsigned long mem_size; +unsigned long + +mem_bitmap[BITS_TO_LONGS(DRA7_RPROC_MAX_CO_SIZE >> PAGE_SHIFT)]; +unsigned long mem_count; + +unsigned int pgtable_l2_map[MAX_NUM_L2_PAGE_TABLES]; +unsigned int pgtable_l2_cnt; + +void *ipu_alloc_mem(struct udevice *dev, unsigned long len, unsigned long align) +{ + unsigned long mask; + unsigned long pageno; + int count; + + count = ((len + (PAGE_SIZE - 1)) & ~(PAGE_SIZE - 1)) >> PAGE_SHIFT; + mask = (1 << align) - 1; + pageno = + bitmap_find_next_zero_area(mem_bitmap, mem_count, 0, count, mask); + debug("%s: count %d mask %#lx pageno %#lx\n", __func__, count, mask, + pageno); + + if (pageno >= mem_count) { + debug("%s: %s Error allocating memory; " + "Please check carveout size\n", __FILE__, __func__); + return NULL; + } + + bitmap_set(mem_bitmap, pageno, count); + return (void *)(mem_base + (pageno << PAGE_SHIFT)); +} + +int find_pagesz(unsigned int virt, unsigned int phys, unsigned int len) +{ + int pg_sz_ind = -1; + unsigned int min_align = __ffs(virt); + + if (min_align > __ffs(phys)) + min_align = __ffs(phys); + + if (min_align >= PGT_L1_DESC_SUPERSECTION_INDEX_SHIFT && + len >= 0x1000000) { + pg_sz_ind = PAGESIZE_16M; + goto ret_block; + } + if (min_align >= PGT_L1_DESC_SECTION_INDEX_SHIFT && + len >= 0x100000) { + pg_sz_ind = PAGESIZE_1M; + goto ret_block; + } + if (min_align >= PGT_L1_DESC_LARGEPAGE_INDEX_SHIFT && + len >= 0x10000) { + pg_sz_ind = PAGESIZE_64K; + goto ret_block; + } + if (min_align >= PGT_L1_DESC_SMALLPAGE_INDEX_SHIFT && + len >= 0x1000) { + pg_sz_ind = PAGESIZE_4K; + goto ret_block; + } + + ret_block: + return pg_sz_ind; +} + +int get_l2_pg_tbl_addr(unsigned int virt, unsigned int *pg_tbl_addr) +{ + int ret = -1; + int i = 0; + int match_found = 0; + unsigned int tag = (virt & PGT_L1_DESC_SECTION_MASK); + + *pg_tbl_addr = 0; + for (i = 0; (i < pgtable_l2_cnt) && (match_found == 0); i++) { + if (tag == pgtable_l2_map[i]) { + *pg_tbl_addr = + ((unsigned int)page_table_l2) + + (i * PAGE_TABLE_SIZE_L2); + match_found = 1; + ret = 0; + } + } + + if (match_found == 0 && i < MAX_NUM_L2_PAGE_TABLES) { + pgtable_l2_map[i] = tag; + pgtable_l2_cnt++; + *pg_tbl_addr = + ((unsigned int)page_table_l2) + (i * PAGE_TABLE_SIZE_L2); + ret = 0; + } + + return ret; +} + +int +config_l2_pagetable(unsigned int virt, unsigned int phys, + unsigned int pg_sz, unsigned int pg_tbl_addr) +{ + int ret = -1; + unsigned int desc = 0; + int i = 0; + unsigned int *pg_tbl = (unsigned int *)pg_tbl_addr; + + /* + * Pick bit 19:12 of the virtual address as index + */ + unsigned int index = (virt & (~PGT_L1_DESC_SECTION_MASK)) >> PAGE_SHIFT; + + switch (pg_sz) { + case PAGESIZE_64K: + desc = + (phys & PGT_L2_DESC_LARGEPAGE_MASK) | PGT_L2_DESC_LARGEPAGE; + for (i = 0; i < 16; i++) + pg_tbl[index + i] = desc; + ret = 0; + break; + case PAGESIZE_4K: + desc = + (phys & PGT_L2_DESC_SMALLPAGE_MASK) | PGT_L2_DESC_SMALLPAGE; + pg_tbl[index] = desc; + ret = 0; + break; + default: + break; + } + + return ret; +} + +unsigned int +ipu_config_pagetable(struct udevice *dev, unsigned int virt, unsigned int phys, + unsigned int len) +{ + unsigned int index; + unsigned int l = len; + unsigned int desc; + int pg_sz = 0; + int i = 0, err = 0; + unsigned int pg_tbl_l2_addr = 0; + unsigned int tmp_pgsz; + + if ((len & 0x0FFF) != 0) + return 0; + + while (l > 0) { + pg_sz = find_pagesz(virt, phys, l); + index = virt >> PGT_L1_DESC_SECTION_INDEX_SHIFT; + switch (pg_sz) { + /* + * 16 MB super section + */ + case PAGESIZE_16M: + /* + * Program the next 16 descriptors + */ + desc = + (phys & PGT_L1_DESC_SUPERSECTION_MASK) | + PGT_L1_DESC_SUPERSECTION; + for (i = 0; i < 16; i++) + page_table_l1[index + i] = desc; + l -= PGT_SUPERSECTION_SIZE; + phys += PGT_SUPERSECTION_SIZE; + virt += PGT_SUPERSECTION_SIZE; + break; + /* + * 1 MB section + */ + case PAGESIZE_1M: + desc = + (phys & PGT_L1_DESC_SECTION_MASK) | + PGT_L1_DESC_SECTION; + page_table_l1[index] = desc; + l -= PGT_SECTION_SIZE; + phys += PGT_SECTION_SIZE; + virt += PGT_SECTION_SIZE; + break; + /* + * 64 KB large page + */ + case PAGESIZE_64K: + case PAGESIZE_4K: + if (pg_sz == PAGESIZE_64K) + tmp_pgsz = 0x10000; + else + tmp_pgsz = 0x1000; + + err = get_l2_pg_tbl_addr(virt, &pg_tbl_l2_addr); + if (err != 0) { + debug + ("Unable to get level 2 PT address\n"); + hang(); + } + err = + config_l2_pagetable(virt, phys, pg_sz, + pg_tbl_l2_addr); + desc = + (pg_tbl_l2_addr & PGT_L1_DESC_PAGE_MASK) | + PGT_L1_DESC_PAGE; + page_table_l1[index] = desc; + l -= tmp_pgsz; + phys += tmp_pgsz; + virt += tmp_pgsz; + break; + case -1: + default: + return 0; + } + } + + return len; +} + +int da_to_pa(struct udevice *dev, int da) +{ + struct rproc_mem_entry *maps = NULL; + struct ipu_privdata *priv = dev_get_priv(dev); + + list_for_each_entry(maps, &priv->mappings, node) { + if (da >= maps->da && da < (maps->da + maps->len)) + return maps->dma + (da - maps->da); + } + + return 0; +} + +u32 ipu_config_mmu(u32 core_id, struct rproc *cfg) +{ + u32 i = 0; + u32 reg = 0; + + /* + * Clear the entire pagetable location before programming the + * address into the MMU + */ + memset((void *)cfg->page_table_addr, 0x00, PAGE_TABLE_SIZE); + + for (i = 0; i < cfg->num_iommus; i++) { + u32 mmu_base = cfg->mmu_base_addr[i]; + + __raw_writel((int)cfg->page_table_addr, mmu_base + 0x4c); + reg = __raw_readl(mmu_base + 0x88); + + /* + * enable bus-error back + */ + __raw_writel(reg | 0x1, mmu_base + 0x88); + + /* + * Enable the MMU IRQs during MMU programming for the + * late attachcase. This is to allow the MMU fault to be + * detected by the kernel. + * + * MULTIHITFAULT|EMMUMISS|TRANSLATIONFAULT|TABLEWALKFAULT + */ + __raw_writel(0x1E, mmu_base + 0x1c); + + /* + * emutlbupdate|TWLENABLE|MMUENABLE + */ + __raw_writel(0x6, mmu_base + 0x44); + } + + return 0; +} + +/** + * enum ipu_mem - PRU core memory range identifiers + */ +enum ipu_mem { + PRU_MEM_IRAM = 0, + PRU_MEM_CTRL, + PRU_MEM_DEBUG, + PRU_MEM_MAX, +}; + +static int ipu_start(struct udevice *dev) +{ + struct ipu_privdata *priv; + struct reset_ctl reset; + struct rproc *cfg = NULL; + int ret; + + priv = dev_get_priv(dev); + + cfg = rproc_cfg_arr[priv->id]; + if (cfg->config_peripherals) + cfg->config_peripherals(priv->id, cfg); + + /* + * Start running the remote core + */ + ret = reset_get_by_index(dev, 0, &reset); + if (ret < 0) { + dev_err(dev, "%s: error getting reset index %d\n", __func__, 0); + return ret; + } + + ret = reset_deassert(&reset); + if (ret < 0) { + dev_err(dev, "%s: error deasserting reset %d\n", __func__, 0); + return ret; + } + + ret = reset_get_by_index(dev, 1, &reset); + if (ret < 0) { + dev_err(dev, "%s: error getting reset index %d\n", __func__, 1); + return ret; + } + + ret = reset_deassert(&reset); + if (ret < 0) { + dev_err(dev, "%s: error deasserting reset %d\n", __func__, 1); + return ret; + } + + return 0; +} + +static int ipu_stop(struct udevice *dev) +{ + return 0; +} + +/** + * ipu_init() - Initialize the remote processor + * @dev: rproc device pointer + * + * Return: 0 if all went ok, else return appropriate error + */ +static int ipu_init(struct udevice *dev) +{ + return 0; +} + +static int ipu_add_res(struct udevice *dev, struct rproc_mem_entry *mapping) +{ + struct ipu_privdata *priv = dev_get_priv(dev); + + list_add_tail(&mapping->node, &priv->mappings); + return 0; +} + +static int ipu_load(struct udevice *dev, ulong addr, ulong size) +{ + Elf32_Ehdr *ehdr; /* Elf header structure pointer */ + Elf32_Phdr *phdr; /* Program header structure pointer */ + Elf32_Phdr proghdr; + int va; + int pa; + int i; + + ehdr = (Elf32_Ehdr *)addr; + phdr = (Elf32_Phdr *)(addr + ehdr->e_phoff); + /* + * Load each program header + */ + for (i = 0; i < ehdr->e_phnum; ++i) { + memcpy(&proghdr, phdr, sizeof(Elf32_Phdr)); + + if (proghdr.p_type != PT_LOAD) { + ++phdr; + continue; + } + + va = proghdr.p_paddr; + pa = da_to_pa(dev, va); + if (pa) + proghdr.p_paddr = pa; + + void *dst = (void *)(uintptr_t)proghdr.p_paddr; + void *src = (void *)addr + proghdr.p_offset; + + debug("Loading phdr %i to 0x%p (%i bytes)\n", i, dst, + proghdr.p_filesz); + if (proghdr.p_filesz) + memcpy(dst, src, proghdr.p_filesz); + + flush_cache((unsigned long)dst, proghdr.p_memsz); + + ++phdr; + } + + return 0; +} + +static const struct dm_rproc_ops ipu_ops = { + .init = ipu_init, + .start = ipu_start, + .stop = ipu_stop, + .load = ipu_load, + .add_res = ipu_add_res, + .config_pagetable = ipu_config_pagetable, + .alloc_mem = ipu_alloc_mem, +}; + +/* + * If the remotecore binary expects any peripherals to be setup before it has + * booted, configure them here. + * + * These functions are left empty by default as their operation is usecase + * specific. + */ + +u32 ipu1_config_peripherals(u32 core_id, struct rproc *cfg) +{ + return 0; +} + +u32 ipu2_config_peripherals(u32 core_id, struct rproc *cfg) +{ + return 0; +} + +struct rproc_intmem_to_l3_mapping ipu1_intmem_to_l3_mapping = { + .num_entries = 1, + .mappings = { + /* + * L2 SRAM + */ + { + .priv_addr = 0x55020000, + .l3_addr = 0x58820000, + .len = (64 * 1024)}, + } +}; + +struct rproc_intmem_to_l3_mapping ipu2_intmem_to_l3_mapping = { + .num_entries = 1, + .mappings = { + /* + * L2 SRAM + */ + { + .priv_addr = 0x55020000, + .l3_addr = 0x55020000, + .len = (64 * 1024)}, + } +}; + +struct rproc ipu1_config = { + .num_iommus = 1, + .mmu_base_addr = {0x58882000, 0}, + .load_addr = IPU1_LOAD_ADDR, + .core_name = "IPU1", + .firmware_name = "dra7-ipu1-fw.xem4", + .config_mmu = ipu_config_mmu, + .config_peripherals = ipu1_config_peripherals, + .intmem_to_l3_mapping = &ipu1_intmem_to_l3_mapping +}; + +struct rproc ipu2_config = { + .num_iommus = 1, + .mmu_base_addr = {0x55082000, 0}, + .load_addr = IPU2_LOAD_ADDR, + .core_name = "IPU2", + .firmware_name = "dra7-ipu2-fw.xem4", + .config_mmu = ipu_config_mmu, + .config_peripherals = ipu2_config_peripherals, + .intmem_to_l3_mapping = &ipu2_intmem_to_l3_mapping +}; + +struct rproc *rproc_cfg_arr[2] = { + [IPU2] = &ipu2_config, + [IPU1] = &ipu1_config, +}; + +u32 spl_pre_boot_core(struct udevice *dev, u32 core_id) +{ + struct rproc *cfg = NULL; + unsigned long load_elf_status = 0; + int tablesz; + + cfg = rproc_cfg_arr[core_id]; + /* + * Check for valid elf image + */ + if (!valid_elf_image(cfg->load_addr)) + return 1; + + if (rproc_find_resource_table(dev, cfg->load_addr, &tablesz)) + cfg->has_rsc_table = 1; + else + cfg->has_rsc_table = 0; + + /* + * Configure the MMU + */ + if (cfg->config_mmu && cfg->has_rsc_table) + cfg->config_mmu(core_id, cfg); + + /* + * Load the remote core. Fill the page table of the first(possibly + * only) IOMMU during ELF loading. Copy the page table to the second + * IOMMU before running the remote core. + */ + + page_table_l1 = (unsigned int *)cfg->page_table_addr; + page_table_l2 = + (unsigned int *)(cfg->page_table_addr + PAGE_TABLE_SIZE_L1); + mem_base = cfg->cma_base; + mem_size = cfg->cma_size; + memset(mem_bitmap, 0x00, sizeof(mem_bitmap)); + mem_count = (cfg->cma_size >> PAGE_SHIFT); + + /* + * Clear variables used for level 2 page table allocation + */ + memset(pgtable_l2_map, 0x00, sizeof(pgtable_l2_map)); + pgtable_l2_cnt = 0; + + load_elf_status = rproc_parse_resource_table(dev, cfg); + if (load_elf_status == 0) { + debug("load_elf_image_phdr returned error for core %s\n", + cfg->core_name); + return 1; + } + + flush_cache(cfg->page_table_addr, PAGE_TABLE_SIZE); + + return 0; +} + +static fdt_addr_t ipu_parse_mem_nodes(struct udevice *dev, char *name, + int privid, fdt_size_t *sizep) +{ + int ret; + u32 sp; + ofnode mem_node; + + ret = ofnode_read_u32(dev_ofnode(dev), name, &sp); + if (ret) { + dev_err(dev, "memory-region node fetch failed %d\n", ret); + return ret; + } + + mem_node = ofnode_get_by_phandle(sp); + if (!ofnode_valid(mem_node)) + return -EINVAL; + + return ofnode_get_addr_size_index(mem_node, 0, sizep); +} + +/** + * ipu_probe() - Basic probe + * @dev: corresponding k3 remote processor device + * + * Return: 0 if all goes good, else appropriate error message. + */ +static int ipu_probe(struct udevice *dev) +{ + struct ipu_privdata *priv; + struct rproc *cfg = NULL; + struct reset_ctl reset; + static const char *const ipu_mem_names[] = { "l2ram" }; + int ret; + fdt_size_t sizep; + + priv = dev_get_priv(dev); + + priv->mem.bus_addr = + devfdt_get_addr_size_name(dev, + ipu_mem_names[0], + (fdt_addr_t *)&priv->mem.size); + + ret = reset_get_by_index(dev, 2, &reset); + if (ret < 0) { + dev_err(dev, "%s: error getting reset index %d\n", __func__, 2); + return ret; + } + + ret = reset_deassert(&reset); + if (ret < 0) { + dev_err(dev, "%s: error deasserting reset %d\n", __func__, 2); + return ret; + } + + if (priv->mem.bus_addr == FDT_ADDR_T_NONE) { + dev_err(dev, "%s bus address not found\n", ipu_mem_names[0]); + return -EINVAL; + } + priv->mem.cpu_addr = map_physmem(priv->mem.bus_addr, + priv->mem.size, MAP_NOCACHE); + + if (devfdt_get_addr(dev) == 0x58820000) + priv->id = 0; + else + priv->id = 1; + + cfg = rproc_cfg_arr[priv->id]; + cfg->cma_base = ipu_parse_mem_nodes(dev, "memory-region", priv->id, + &sizep); + cfg->cma_size = sizep; + + cfg->page_table_addr = ipu_parse_mem_nodes(dev, "pg-tbl", priv->id, + &sizep); + + dev_info(dev, + "ID %d memory %8s: bus addr %pa size 0x%zx va %p da 0x%x\n", + priv->id, ipu_mem_names[0], &priv->mem.bus_addr, + priv->mem.size, priv->mem.cpu_addr, priv->mem.dev_addr); + + INIT_LIST_HEAD(&priv->mappings); + if (spl_pre_boot_core(dev, priv->id)) + return -EINVAL; + + return 0; +} + +static const struct udevice_id ipu_ids[] = { + {.compatible = "ti,dra7-ipu"}, + {} +}; + +U_BOOT_DRIVER(ipu) = { + .name = "ipu", + .of_match = ipu_ids, + .id = UCLASS_REMOTEPROC, + .ops = &ipu_ops, + .probe = ipu_probe, + .priv_auto_alloc_size = sizeof(struct ipu_privdata), +}; -- cgit v1.2.3