diff options
Diffstat (limited to 'arch/arm/mach-tegra')
| -rw-r--r-- | arch/arm/mach-tegra/Kconfig | 6 | ||||
| -rw-r--r-- | arch/arm/mach-tegra/arm64-mmu.c | 3 | ||||
| -rw-r--r-- | arch/arm/mach-tegra/tegra186/Kconfig | 3 | ||||
| -rw-r--r-- | arch/arm/mach-tegra/tegra186/nvtboot_board.c | 265 | ||||
| -rw-r--r-- | arch/arm/mach-tegra/tegra186/nvtboot_mem.c | 144 | ||||
| -rw-r--r-- | arch/arm/mach-tegra/tegra210/Kconfig | 3 |
6 files changed, 392 insertions, 32 deletions
diff --git a/arch/arm/mach-tegra/Kconfig b/arch/arm/mach-tegra/Kconfig index 51d143687b..fd0082d22a 100644 --- a/arch/arm/mach-tegra/Kconfig +++ b/arch/arm/mach-tegra/Kconfig @@ -60,8 +60,14 @@ config TEGRA_ARMV7_COMMON config TEGRA_ARMV8_COMMON bool "Tegra 64-bit common options" select ARM64 + select LINUX_KERNEL_IMAGE_HEADER select TEGRA_COMMON +if TEGRA_ARMV8_COMMON +config LNX_KRNL_IMG_TEXT_OFFSET_BASE + default 0x80000000 +endif + choice prompt "Tegra SoC select" optional diff --git a/arch/arm/mach-tegra/arm64-mmu.c b/arch/arm/mach-tegra/arm64-mmu.c index a79a5192e0..3a126bdec4 100644 --- a/arch/arm/mach-tegra/arm64-mmu.c +++ b/arch/arm/mach-tegra/arm64-mmu.c @@ -12,7 +12,8 @@ #include <asm/system.h> #include <asm/armv8/mmu.h> -static struct mm_region tegra_mem_map[] = { +/* size: IO + NR_DRAM_BANKS + terminator */ +struct mm_region tegra_mem_map[1 + CONFIG_NR_DRAM_BANKS + 1] = { { .virt = 0x0UL, .phys = 0x0UL, diff --git a/arch/arm/mach-tegra/tegra186/Kconfig b/arch/arm/mach-tegra/tegra186/Kconfig index b2e53b58ca..479c0955ee 100644 --- a/arch/arm/mach-tegra/tegra186/Kconfig +++ b/arch/arm/mach-tegra/tegra186/Kconfig @@ -21,6 +21,9 @@ endchoice config SYS_SOC default "tegra186" +config SYS_INIT_SP_BSS_OFFSET + default 524288 + source "board/nvidia/p2771-0000/Kconfig" endif diff --git a/arch/arm/mach-tegra/tegra186/nvtboot_board.c b/arch/arm/mach-tegra/tegra186/nvtboot_board.c index b94eb424aa..8ecb454443 100644 --- a/arch/arm/mach-tegra/tegra186/nvtboot_board.c +++ b/arch/arm/mach-tegra/tegra186/nvtboot_board.c @@ -1,16 +1,278 @@ /* - * Copyright (c) 2016, NVIDIA CORPORATION. + * Copyright (c) 2016-2018, NVIDIA CORPORATION. * * SPDX-License-Identifier: GPL-2.0+ */ +#include <stdlib.h> #include <common.h> #include <fdt_support.h> #include <fdtdec.h> #include <asm/arch/tegra.h> +#include <asm/armv8/mmu.h> + +DECLARE_GLOBAL_DATA_PTR; extern unsigned long nvtboot_boot_x0; +/* + * The following few functions run late during the boot process and dynamically + * calculate the load address of various binaries. To keep track of multiple + * allocations, some writable list of RAM banks must be used. tegra_mem_map[] + * is used for this purpose to avoid making yet another copy of the list of RAM + * banks. This is safe because tegra_mem_map[] is only used once during very + * early boot to create U-Boot's page tables, long before this code runs. If + * this assumption becomes invalid later, we can just fix the code to copy the + * list of RAM banks into some private data structure before running. + */ + +extern struct mm_region tegra_mem_map[]; + +static char *gen_varname(const char *var, const char *ext) +{ + size_t len_var = strlen(var); + size_t len_ext = strlen(ext); + size_t len = len_var + len_ext + 1; + char *varext = malloc(len); + + if (!varext) + return 0; + strcpy(varext, var); + strcpy(varext + len_var, ext); + return varext; +} + +static void mark_ram_allocated(int bank, u64 allocated_start, u64 allocated_end) +{ + u64 bank_start = tegra_mem_map[bank].virt; + u64 bank_size = tegra_mem_map[bank].size; + u64 bank_end = bank_start + bank_size; + bool keep_front = allocated_start != bank_start; + bool keep_tail = allocated_end != bank_end; + + if (keep_front && keep_tail) { + /* + * There are CONFIG_NR_DRAM_BANKS DRAM entries in the array, + * starting at index 1 (index 0 is MMIO). So, we are at DRAM + * entry "bank" not "bank - 1" as for a typical 0-base array. + * The number of remaining DRAM entries is therefore + * "CONFIG_NR_DRAM_BANKS - bank". We want to duplicate the + * current entry and shift up the remaining entries, dropping + * the last one. Thus, we must copy one fewer entry than the + * number remaining. + */ + memmove(&tegra_mem_map[bank + 1], &tegra_mem_map[bank], + CONFIG_NR_DRAM_BANKS - bank - 1); + tegra_mem_map[bank].size = allocated_start - bank_start; + bank++; + tegra_mem_map[bank].virt = allocated_end; + tegra_mem_map[bank].phys = allocated_end; + tegra_mem_map[bank].size = bank_end - allocated_end; + } else if (keep_front) { + tegra_mem_map[bank].size = allocated_start - bank_start; + } else if (keep_tail) { + tegra_mem_map[bank].virt = allocated_end; + tegra_mem_map[bank].phys = allocated_end; + tegra_mem_map[bank].size = bank_end - allocated_end; + } else { + /* + * We could move all subsequent banks down in the array but + * that's not necessary for subsequent allocations to work, so + * we skip doing so. + */ + tegra_mem_map[bank].size = 0; + } +} + +static void reserve_ram(u64 start, u64 size) +{ + int bank; + u64 end = start + size; + + for (bank = 1; bank <= CONFIG_NR_DRAM_BANKS; bank++) { + u64 bank_start = tegra_mem_map[bank].virt; + u64 bank_size = tegra_mem_map[bank].size; + u64 bank_end = bank_start + bank_size; + + if (end <= bank_start || start > bank_end) + continue; + mark_ram_allocated(bank, start, end); + break; + } +} + +static u64 alloc_ram(u64 size, u64 align, u64 offset) +{ + int bank; + + for (bank = 1; bank <= CONFIG_NR_DRAM_BANKS; bank++) { + u64 bank_start = tegra_mem_map[bank].virt; + u64 bank_size = tegra_mem_map[bank].size; + u64 bank_end = bank_start + bank_size; + u64 allocated = ROUND(bank_start, align) + offset; + u64 allocated_end = allocated + size; + + if (allocated_end > bank_end) + continue; + mark_ram_allocated(bank, allocated, allocated_end); + return allocated; + } + return 0; +} + +static void set_calculated_aliases(char *aliases, u64 address) +{ + char *tmp, *alias; + int err; + + aliases = strdup(aliases); + if (!aliases) { + pr_err("strdup(aliases) failed"); + return; + } + + tmp = aliases; + while (true) { + alias = strsep(&tmp, " "); + if (!alias) + break; + debug("%s: alias: %s\n", __func__, alias); + err = env_set_hex(alias, address); + if (err) + pr_err("Could not set %s\n", alias); + } + + free(aliases); +} + +static void set_calculated_env_var(const char *var) +{ + char *var_size; + char *var_align; + char *var_offset; + char *var_aliases; + u64 size; + u64 align; + u64 offset; + char *aliases; + u64 address; + int err; + + var_size = gen_varname(var, "_size"); + if (!var_size) + return; + var_align = gen_varname(var, "_align"); + if (!var_align) + goto out_free_var_size; + var_offset = gen_varname(var, "_offset"); + if (!var_offset) + goto out_free_var_align; + var_aliases = gen_varname(var, "_aliases"); + if (!var_aliases) + goto out_free_var_offset; + + size = env_get_hex(var_size, 0); + if (!size) { + pr_err("%s not set or zero\n", var_size); + goto out_free_var_aliases; + } + align = env_get_hex(var_align, 1); + /* Handle extant variables, but with a value of 0 */ + if (!align) + align = 1; + offset = env_get_hex(var_offset, 0); + aliases = env_get(var_aliases); + + debug("%s: Calc var %s; size=%llx, align=%llx, offset=%llx\n", + __func__, var, size, align, offset); + if (aliases) + debug("%s: Aliases: %s\n", __func__, aliases); + + address = alloc_ram(size, align, offset); + if (!address) { + pr_err("Could not allocate %s\n", var); + goto out_free_var_aliases; + } + debug("%s: Address %llx\n", __func__, address); + + err = env_set_hex(var, address); + if (err) + pr_err("Could not set %s\n", var); + if (aliases) + set_calculated_aliases(aliases, address); + +out_free_var_aliases: + free(var_aliases); +out_free_var_offset: + free(var_offset); +out_free_var_align: + free(var_align); +out_free_var_size: + free(var_size); +} + +#ifdef DEBUG +static void dump_ram_banks(void) +{ + int bank; + + for (bank = 1; bank <= CONFIG_NR_DRAM_BANKS; bank++) { + u64 bank_start = tegra_mem_map[bank].virt; + u64 bank_size = tegra_mem_map[bank].size; + u64 bank_end = bank_start + bank_size; + + if (!bank_size) + continue; + printf("%d: %010llx..%010llx (+%010llx)\n", bank - 1, + bank_start, bank_end, bank_size); + } +} +#endif + +static void set_calculated_env_vars(void) +{ + char *vars, *tmp, *var; + +#ifdef DEBUG + printf("RAM banks before any calculated env. var.s:\n"); + dump_ram_banks(); +#endif + + reserve_ram(nvtboot_boot_x0, fdt_totalsize(nvtboot_boot_x0)); + +#ifdef DEBUG + printf("RAM after reserving cboot DTB:\n"); + dump_ram_banks(); +#endif + + vars = env_get("calculated_vars"); + if (!vars) { + debug("%s: No env var calculated_vars\n", __func__); + return; + } + + vars = strdup(vars); + if (!vars) { + pr_err("strdup(calculated_vars) failed"); + return; + } + + tmp = vars; + while (true) { + var = strsep(&tmp, " "); + if (!var) + break; + debug("%s: var: %s\n", __func__, var); + set_calculated_env_var(var); +#ifdef DEBUG + printf("RAM banks affter allocating %s:\n", var); + dump_ram_banks(); +#endif + } + + free(vars); +} + static int set_fdt_addr(void) { int ret; @@ -60,6 +322,7 @@ static int set_ethaddr_from_nvtboot(void) int tegra_soc_board_init_late(void) { + set_calculated_env_vars(); /* * Ignore errors here; the value may not be used depending on * extlinux.conf or boot script content. diff --git a/arch/arm/mach-tegra/tegra186/nvtboot_mem.c b/arch/arm/mach-tegra/tegra186/nvtboot_mem.c index 5224ef641c..2ca59747d4 100644 --- a/arch/arm/mach-tegra/tegra186/nvtboot_mem.c +++ b/arch/arm/mach-tegra/tegra186/nvtboot_mem.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016, NVIDIA CORPORATION. + * Copyright (c) 2016-2018, NVIDIA CORPORATION. * * SPDX-License-Identifier: GPL-2.0+ */ @@ -8,28 +8,48 @@ #include <fdt_support.h> #include <fdtdec.h> #include <asm/arch/tegra.h> +#include <asm/armv8/mmu.h> + +#define SZ_4G 0x100000000ULL + +/* + * Size of a region that's large enough to hold the relocated U-Boot and all + * other allocations made around it (stack, heap, page tables, etc.) + * In practice, running "bdinfo" at the shell prompt, the stack reaches about + * 5MB from the address selected for ram_top as of the time of writing, + * so a 16MB region should be plenty. + */ +#define MIN_USABLE_RAM_SIZE SZ_16M +/* + * The amount of space we expect to require for stack usage. Used to validate + * that all reservations fit into the region selected for the relocation target + */ +#define MIN_USABLE_STACK_SIZE SZ_1M DECLARE_GLOBAL_DATA_PTR; extern unsigned long nvtboot_boot_x0; +extern struct mm_region tegra_mem_map[]; /* - * A parsed version of /memory/reg from the DTB that is passed to U-Boot in x0. - * - * We only support up to two banks since that's all the binary bootloader - * ever sets. We assume bank 0 is RAM below 4G and bank 1 is RAM above 4G. - * This is all a fairly safe assumption, since the L4T kernel makes the same - * assumptions, so the bootloader is unlikely to change. - * - * This is written to before relocation, and hence cannot be in .bss, since - * .bss overlaps the DTB that's appended to the U-Boot binary. The initializer - * forces this into .data and avoids this issue. This also has the nice side- - * effect of the content being valid after relocation. + * These variables are written to before relocation, and hence cannot be + * in.bss, since .bss overlaps the DTB that's appended to the U-Boot binary. + * The section attribute forces this into .data and avoids this issue. This + * also has the nice side-effect of the content being valid after relocation. + */ + +/* The number of valid entries in ram_banks[] */ +static int ram_bank_count __attribute__((section(".data"))); + +/* + * The usable top-of-RAM for U-Boot. This is both: + * a) Below 4GB to avoid issues with peripherals that use 32-bit addressing. + * b) At the end of a region that has enough space to hold the relocated U-Boot + * and all other allocations made around it (stack, heap, page tables, etc.) */ -static struct { - u64 start; - u64 size; -} ram_banks[2] = {{1}}; +static u64 ram_top __attribute__((section(".data"))); +/* The base address of the region of RAM that ends at ram_top */ +static u64 region_base __attribute__((section(".data"))); int dram_init(void) { @@ -38,8 +58,6 @@ int dram_init(void) int node, len, i; const u32 *prop; - memset(ram_banks, 0, sizeof(ram_banks)); - na = fdtdec_get_uint(nvtboot_blob, 0, "#address-cells", 2); ns = fdtdec_get_uint(nvtboot_blob, 0, "#size-cells", 2); @@ -54,37 +72,103 @@ int dram_init(void) hang(); } - len /= (na + ns); - if (len > ARRAY_SIZE(ram_banks)) - len = ARRAY_SIZE(ram_banks); + /* Calculate the true # of base/size pairs to read */ + len /= 4; /* Convert bytes to number of cells */ + len /= (na + ns); /* Convert cells to number of banks */ + if (len > CONFIG_NR_DRAM_BANKS) + len = CONFIG_NR_DRAM_BANKS; + /* Parse the /memory node, and save useful entries */ gd->ram_size = 0; + ram_bank_count = 0; for (i = 0; i < len; i++) { - ram_banks[i].start = fdt_read_number(prop, na); + u64 bank_start, bank_end, bank_size, usable_bank_size; + + /* Extract raw memory region data from DTB */ + bank_start = fdt_read_number(prop, na); prop += na; - ram_banks[i].size = fdt_read_number(prop, ns); + bank_size = fdt_read_number(prop, ns); prop += ns; - gd->ram_size += ram_banks[i].size; + gd->ram_size += bank_size; + bank_end = bank_start + bank_size; + debug("Bank %d: %llx..%llx (+%llx)\n", i, + bank_start, bank_end, bank_size); + + /* + * Align the bank to MMU section size. This is not strictly + * necessary, since the translation table construction code + * handles page granularity without issue. However, aligning + * the MMU entries reduces the size and number of levels in the + * page table, so is worth it. + */ + bank_start = ROUND(bank_start, SZ_2M); + bank_end = bank_end & ~(SZ_2M - 1); + bank_size = bank_end - bank_start; + debug(" aligned: %llx..%llx (+%llx)\n", + bank_start, bank_end, bank_size); + if (bank_end <= bank_start) + continue; + + /* Record data used to create MMU translation tables */ + ram_bank_count++; + /* Index below is deliberately 1-based to skip MMIO entry */ + tegra_mem_map[ram_bank_count].virt = bank_start; + tegra_mem_map[ram_bank_count].phys = bank_start; + tegra_mem_map[ram_bank_count].size = bank_size; + tegra_mem_map[ram_bank_count].attrs = + PTE_BLOCK_MEMTYPE(MT_NORMAL) | PTE_BLOCK_INNER_SHARE; + + /* Determine best bank to relocate U-Boot into */ + if (bank_end > SZ_4G) + bank_end = SZ_4G; + debug(" end %llx (usable)\n", bank_end); + usable_bank_size = bank_end - bank_start; + debug(" size %llx (usable)\n", usable_bank_size); + if ((usable_bank_size >= MIN_USABLE_RAM_SIZE) && + (bank_end > ram_top)) { + ram_top = bank_end; + region_base = bank_start; + debug("ram top now %llx\n", ram_top); + } + } + + /* Ensure memory map contains the desired sentinel entry */ + tegra_mem_map[ram_bank_count + 1].virt = 0; + tegra_mem_map[ram_bank_count + 1].phys = 0; + tegra_mem_map[ram_bank_count + 1].size = 0; + tegra_mem_map[ram_bank_count + 1].attrs = 0; + + /* Error out if a relocation target couldn't be found */ + if (!ram_top) { + pr_err("Can't find a usable RAM top"); + hang(); } return 0; } -extern unsigned long nvtboot_boot_x0; - int dram_init_banksize(void) { int i; - for (i = 0; i < 2; i++) { - gd->bd->bi_dram[i].start = ram_banks[i].start; - gd->bd->bi_dram[i].size = ram_banks[i].size; + if ((gd->start_addr_sp - region_base) < MIN_USABLE_STACK_SIZE) { + pr_err("Reservations exceed chosen region size"); + hang(); + } + + for (i = 0; i < ram_bank_count; i++) { + gd->bd->bi_dram[i].start = tegra_mem_map[1 + i].virt; + gd->bd->bi_dram[i].size = tegra_mem_map[1 + i].size; } +#ifdef CONFIG_PCI + gd->pci_ram_top = ram_top; +#endif + return 0; } ulong board_get_usable_ram_top(ulong total_size) { - return ram_banks[0].start + ram_banks[0].size; + return ram_top; } diff --git a/arch/arm/mach-tegra/tegra210/Kconfig b/arch/arm/mach-tegra/tegra210/Kconfig index 3637473051..250738aed3 100644 --- a/arch/arm/mach-tegra/tegra210/Kconfig +++ b/arch/arm/mach-tegra/tegra210/Kconfig @@ -40,6 +40,9 @@ endchoice config SYS_SOC default "tegra210" +config SYS_INIT_SP_BSS_OFFSET + default 524288 + source "board/nvidia/e2220-1170/Kconfig" source "board/nvidia/p2371-0000/Kconfig" source "board/nvidia/p2371-2180/Kconfig" |