void mmap_add_ctx(xlat_ctx_t *ctx, const mmap_region_t *mm)
{
	const mmap_region_t *mm_cursor = mm;

	while (mm_cursor->granularity != 0U) {
		mmap_add_region_ctx(ctx, mm_cursor);
		mm_cursor++;
	}
}
void mmap_add_region_alloc_va_ctx(xlat_ctx_t *ctx, mmap_region_t *mm)
{
	mm->base_va = ctx->max_va + 1UL;

	assert(mm->size > 0U);

	mmap_alloc_va_align_ctx(ctx, mm);

	/* Detect overflows. More checks are done in mmap_add_region_check(). */
	assert(mm->base_va > ctx->max_va);

	mmap_add_region_ctx(ctx, mm);
}
예제 #3
0
void sp_map_memory_regions(sp_context_t *sp_ctx)
{
	/* This region contains the exception vectors used at S-EL1. */
	const mmap_region_t sel1_exception_vectors =
		MAP_REGION_FLAT(SPM_SHIM_EXCEPTIONS_START,
				SPM_SHIM_EXCEPTIONS_SIZE,
				MT_CODE | MT_SECURE | MT_PRIVILEGED);

	mmap_add_region_ctx(sp_ctx->xlat_ctx_handle,
			    &sel1_exception_vectors);

	struct sp_rd_sect_mem_region *rdmem;

	for (rdmem = sp_ctx->rd.mem_region; rdmem != NULL; rdmem = rdmem->next) {
		map_rdmem(sp_ctx, rdmem);
	}

	init_xlat_tables_ctx(sp_ctx->xlat_ctx_handle);
}
예제 #4
0
/*
 * The data provided in the resource description structure is not directly
 * compatible with a mmap_region structure. This function handles the conversion
 * and maps it.
 */
static void map_rdmem(sp_context_t *sp_ctx, struct sp_rd_sect_mem_region *rdmem)
{
	int rc;
	mmap_region_t mmap;

	/* Location of the SP image */
	uintptr_t sp_size = sp_ctx->image_size;
	uintptr_t sp_base_va = sp_ctx->rd.attribute.load_address;
	unsigned long long sp_base_pa = sp_ctx->image_base;

	/* Location of the memory region to map */
	size_t rd_size = rdmem->size;
	uintptr_t rd_base_va = rdmem->base;
	unsigned long long rd_base_pa;

	unsigned int memtype = rdmem->attr & RD_MEM_MASK;

	VERBOSE("Adding memory region '%s'\n", rdmem->name);

	mmap.granularity = REGION_DEFAULT_GRANULARITY;

	/* Check if the RD region is inside of the SP image or not */
	int is_outside = (rd_base_va + rd_size <= sp_base_va) ||
			 (sp_base_va + sp_size <= rd_base_va);

	/* Set to 1 if it is needed to zero this region */
	int zero_region = 0;

	switch (memtype) {
	case RD_MEM_DEVICE:
		/* Device regions are mapped 1:1 */
		rd_base_pa = rd_base_va;
		break;

	case RD_MEM_NORMAL_CODE:
	case RD_MEM_NORMAL_RODATA:
	{
		if (is_outside == 1) {
			ERROR("Code and rodata sections must be fully contained in the image.");
			panic();
		}

		/* Get offset into the image */
		rd_base_pa = sp_base_pa + rd_base_va - sp_base_va;
		break;
	}
	case RD_MEM_NORMAL_DATA:
	{
		if (is_outside == 1) {
			ERROR("Data sections must be fully contained in the image.");
			panic();
		}

		rd_base_pa = spm_alloc_heap(rd_size);

		/* Get offset into the image */
		void *img_pa = (void *)(sp_base_pa + rd_base_va - sp_base_va);

		VERBOSE("  Copying data from %p to 0x%llx\n", img_pa, rd_base_pa);

		/* Map destination */
		rc = mmap_add_dynamic_region(rd_base_pa, rd_base_pa,
				rd_size, MT_MEMORY | MT_RW | MT_SECURE);
		if (rc != 0) {
			ERROR("Unable to map data region at EL3: %d\n", rc);
			panic();
		}

		/* Copy original data to destination */
		memcpy((void *)rd_base_pa, img_pa, rd_size);

		/* Unmap destination region */
		rc = mmap_remove_dynamic_region(rd_base_pa, rd_size);
		if (rc != 0) {
			ERROR("Unable to remove data region at EL3: %d\n", rc);
			panic();
		}

		break;
	}
	case RD_MEM_NORMAL_MISCELLANEOUS:
		/* Allow SPM to change the attributes of the region. */
		mmap.granularity = PAGE_SIZE;
		rd_base_pa = spm_alloc_heap(rd_size);
		zero_region = 1;
		break;

	case RD_MEM_NORMAL_SPM_SP_SHARED_MEM:
		if ((sp_ctx->spm_sp_buffer_base != 0) ||
		    (sp_ctx->spm_sp_buffer_size != 0)) {
			ERROR("A partition must have only one SPM<->SP buffer.\n");
			panic();
		}
		rd_base_pa = spm_alloc_heap(rd_size);
		zero_region = 1;
		/* Save location of this buffer, it is needed by SPM */
		sp_ctx->spm_sp_buffer_base = rd_base_pa;
		sp_ctx->spm_sp_buffer_size = rd_size;
		break;

	case RD_MEM_NORMAL_CLIENT_SHARED_MEM:
		/* Fallthrough */
	case RD_MEM_NORMAL_BSS:
		rd_base_pa = spm_alloc_heap(rd_size);
		zero_region = 1;
		break;

	default:
		panic();
	}

	mmap.base_pa = rd_base_pa;
	mmap.base_va = rd_base_va;
	mmap.size = rd_size;

	/* Only S-EL0 mappings supported for now */
	mmap.attr = rdmem_attr_to_mmap_attr(rdmem->attr) | MT_USER;

	VERBOSE("  VA: 0x%lx PA: 0x%llx (0x%lx, attr: 0x%x)\n",
		mmap.base_va, mmap.base_pa, mmap.size, mmap.attr);

	/* Map region in the context of the Secure Partition */
	mmap_add_region_ctx(sp_ctx->xlat_ctx_handle, &mmap);

	if (zero_region == 1) {
		VERBOSE("  Zeroing region...\n");

		rc = mmap_add_dynamic_region(mmap.base_pa, mmap.base_pa,
				mmap.size, MT_MEMORY | MT_RW | MT_SECURE);
		if (rc != 0) {
			ERROR("Unable to map memory at EL3 to zero: %d\n",
			      rc);
			panic();
		}

		zeromem((void *)mmap.base_pa, mmap.size);

		/*
		 * Unmap destination region unless it is the SPM<->SP buffer,
		 * which must be used by SPM.
		 */
		if (memtype != RD_MEM_NORMAL_SPM_SP_SHARED_MEM) {
			rc = mmap_remove_dynamic_region(rd_base_pa, rd_size);
			if (rc != 0) {
				ERROR("Unable to remove region at EL3: %d\n", rc);
				panic();
			}
		}
	}
}
/* Setup context of the Secure Partition */
void secure_partition_setup(void)
{
	VERBOSE("S-EL1/S-EL0 context setup start...\n");

	cpu_context_t *ctx = cm_get_context(SECURE);

	/* Make sure that we got a Secure context. */
	assert(ctx != NULL);

	/* Assert we are in Secure state. */
	assert((read_scr_el3() & SCR_NS_BIT) == 0);

	/* Disable MMU at EL1. */
	disable_mmu_icache_el1();

	/* Invalidate TLBs at EL1. */
	tlbivmalle1();

	/*
	 * General-Purpose registers
	 * -------------------------
	 */

	/*
	 * X0: Virtual address of a buffer shared between EL3 and Secure EL0.
	 *     The buffer will be mapped in the Secure EL1 translation regime
	 *     with Normal IS WBWA attributes and RO data and Execute Never
	 *     instruction access permissions.
	 *
	 * X1: Size of the buffer in bytes
	 *
	 * X2: cookie value (Implementation Defined)
	 *
	 * X3: cookie value (Implementation Defined)
	 *
	 * X4 to X30 = 0 (already done by cm_init_my_context())
	 */
	write_ctx_reg(get_gpregs_ctx(ctx), CTX_GPREG_X0, PLAT_SPM_BUF_BASE);
	write_ctx_reg(get_gpregs_ctx(ctx), CTX_GPREG_X1, PLAT_SPM_BUF_SIZE);
	write_ctx_reg(get_gpregs_ctx(ctx), CTX_GPREG_X2, PLAT_SPM_COOKIE_0);
	write_ctx_reg(get_gpregs_ctx(ctx), CTX_GPREG_X3, PLAT_SPM_COOKIE_1);

	/*
	 * SP_EL0: A non-zero value will indicate to the SP that the SPM has
	 * initialized the stack pointer for the current CPU through
	 * implementation defined means. The value will be 0 otherwise.
	 */
	write_ctx_reg(get_gpregs_ctx(ctx), CTX_GPREG_SP_EL0,
			PLAT_SP_IMAGE_STACK_BASE + PLAT_SP_IMAGE_STACK_PCPU_SIZE);

	/*
	 * Setup translation tables
	 * ------------------------
	 */

#if ENABLE_ASSERTIONS

	/* Get max granularity supported by the platform. */

	u_register_t id_aa64prf0_el1 = read_id_aa64pfr0_el1();

	int tgran64_supported =
		((id_aa64prf0_el1 >> ID_AA64MMFR0_EL1_TGRAN64_SHIFT) &
		 ID_AA64MMFR0_EL1_TGRAN64_MASK) ==
		 ID_AA64MMFR0_EL1_TGRAN64_SUPPORTED;

	int tgran16_supported =
		((id_aa64prf0_el1 >> ID_AA64MMFR0_EL1_TGRAN16_SHIFT) &
		 ID_AA64MMFR0_EL1_TGRAN16_MASK) ==
		 ID_AA64MMFR0_EL1_TGRAN16_SUPPORTED;

	int tgran4_supported =
		((id_aa64prf0_el1 >> ID_AA64MMFR0_EL1_TGRAN4_SHIFT) &
		 ID_AA64MMFR0_EL1_TGRAN4_MASK) ==
		 ID_AA64MMFR0_EL1_TGRAN4_SUPPORTED;

	uintptr_t max_granule_size;

	if (tgran64_supported) {
		max_granule_size = 64 * 1024;
	} else if (tgran16_supported) {
		max_granule_size = 16 * 1024;
	} else {
		assert(tgran4_supported);
		max_granule_size = 4 * 1024;
	}

	VERBOSE("Max translation granule supported: %lu KiB\n",
		max_granule_size);

	uintptr_t max_granule_size_mask = max_granule_size - 1;

	/* Base must be aligned to the max granularity */
	assert((ARM_SP_IMAGE_NS_BUF_BASE & max_granule_size_mask) == 0);

	/* Size must be a multiple of the max granularity */
	assert((ARM_SP_IMAGE_NS_BUF_SIZE & max_granule_size_mask) == 0);

#endif /* ENABLE_ASSERTIONS */

	/* This region contains the exception vectors used at S-EL1. */
	const mmap_region_t sel1_exception_vectors =
		MAP_REGION_FLAT(SPM_SHIM_EXCEPTIONS_START,
				SPM_SHIM_EXCEPTIONS_SIZE,
				MT_CODE | MT_SECURE | MT_PRIVILEGED);
	mmap_add_region_ctx(&secure_partition_xlat_ctx,
			    &sel1_exception_vectors);

	mmap_add_ctx(&secure_partition_xlat_ctx,
		     plat_get_secure_partition_mmap(NULL));

	init_xlat_tables_ctx(&secure_partition_xlat_ctx);

	/*
	 * MMU-related registers
	 * ---------------------
	 */

	/* Set attributes in the right indices of the MAIR */
	u_register_t mair_el1 =
		MAIR_ATTR_SET(ATTR_DEVICE, ATTR_DEVICE_INDEX) |
		MAIR_ATTR_SET(ATTR_IWBWA_OWBWA_NTR, ATTR_IWBWA_OWBWA_NTR_INDEX) |
		MAIR_ATTR_SET(ATTR_NON_CACHEABLE, ATTR_NON_CACHEABLE_INDEX);

	write_ctx_reg(get_sysregs_ctx(ctx), CTX_MAIR_EL1, mair_el1);

	/* Setup TCR_EL1. */
	u_register_t tcr_ps_bits = tcr_physical_addr_size_bits(PLAT_PHY_ADDR_SPACE_SIZE);

	u_register_t tcr_el1 =
		/* Size of region addressed by TTBR0_EL1 = 2^(64-T0SZ) bytes. */
		(64 - __builtin_ctzl(PLAT_VIRT_ADDR_SPACE_SIZE))		|
		/* Inner and outer WBWA, shareable. */
		TCR_SH_INNER_SHAREABLE | TCR_RGN_OUTER_WBA | TCR_RGN_INNER_WBA	|
		/* Set the granularity to 4KB. */
		TCR_TG0_4K							|
		/* Limit Intermediate Physical Address Size. */
		tcr_ps_bits << TCR_EL1_IPS_SHIFT				|
		/* Disable translations using TBBR1_EL1. */
		TCR_EPD1_BIT
		/* The remaining fields related to TBBR1_EL1 are left as zero. */
	;

	tcr_el1 &= ~(
		/* Enable translations using TBBR0_EL1 */
		TCR_EPD0_BIT
	);

	write_ctx_reg(get_sysregs_ctx(ctx), CTX_TCR_EL1, tcr_el1);

	/* Setup SCTLR_EL1 */
	u_register_t sctlr_el1 = read_ctx_reg(get_sysregs_ctx(ctx), CTX_SCTLR_EL1);

	sctlr_el1 |=
		/*SCTLR_EL1_RES1 |*/
		/* Don't trap DC CVAU, DC CIVAC, DC CVAC, DC CVAP, or IC IVAU */
		SCTLR_UCI_BIT							|
		/* RW regions at xlat regime EL1&0 are forced to be XN. */
		SCTLR_WXN_BIT							|
		/* Don't trap to EL1 execution of WFI or WFE at EL0. */
		SCTLR_NTWI_BIT | SCTLR_NTWE_BIT					|
		/* Don't trap to EL1 accesses to CTR_EL0 from EL0. */
		SCTLR_UCT_BIT							|
		/* Don't trap to EL1 execution of DZ ZVA at EL0. */
		SCTLR_DZE_BIT							|
		/* Enable SP Alignment check for EL0 */
		SCTLR_SA0_BIT							|
		/* Allow cacheable data and instr. accesses to normal memory. */
		SCTLR_C_BIT | SCTLR_I_BIT					|
		/* Alignment fault checking enabled when at EL1 and EL0. */
		SCTLR_A_BIT							|
		/* Enable MMU. */
		SCTLR_M_BIT
	;

	sctlr_el1 &= ~(
		/* Explicit data accesses at EL0 are little-endian. */
		SCTLR_E0E_BIT							|
		/* Accesses to DAIF from EL0 are trapped to EL1. */
		SCTLR_UMA_BIT
	);

	write_ctx_reg(get_sysregs_ctx(ctx), CTX_SCTLR_EL1, sctlr_el1);

	/* Point TTBR0_EL1 at the tables of the context created for the SP. */
	write_ctx_reg(get_sysregs_ctx(ctx), CTX_TTBR0_EL1,
			(u_register_t)secure_partition_base_xlat_table);

	/*
	 * Setup other system registers
	 * ----------------------------
	 */

	/* Shim Exception Vector Base Address */
	write_ctx_reg(get_sysregs_ctx(ctx), CTX_VBAR_EL1,
			SPM_SHIM_EXCEPTIONS_PTR);

	/*
	 * FPEN: Forbid the Secure Partition to access FP/SIMD registers.
	 * TTA: Enable access to trace registers.
	 * ZEN (v8.2): Trap SVE instructions and access to SVE registers.
	 */
	write_ctx_reg(get_sysregs_ctx(ctx), CTX_CPACR_EL1,
			CPACR_EL1_FPEN(CPACR_EL1_FP_TRAP_ALL));

	/*
	 * Prepare information in buffer shared between EL3 and S-EL0
	 * ----------------------------------------------------------
	 */

	void *shared_buf_ptr = (void *) PLAT_SPM_BUF_BASE;

	/* Copy the boot information into the shared buffer with the SP. */
	assert((uintptr_t)shared_buf_ptr + sizeof(secure_partition_boot_info_t)
	       <= (PLAT_SPM_BUF_BASE + PLAT_SPM_BUF_SIZE));

	assert(PLAT_SPM_BUF_BASE <= (UINTPTR_MAX - PLAT_SPM_BUF_SIZE + 1));

	const secure_partition_boot_info_t *sp_boot_info =
			plat_get_secure_partition_boot_info(NULL);

	assert(sp_boot_info != NULL);

	memcpy((void *) shared_buf_ptr, (const void *) sp_boot_info,
	       sizeof(secure_partition_boot_info_t));

	/* Pointer to the MP information from the platform port. */
	secure_partition_mp_info_t *sp_mp_info =
		((secure_partition_boot_info_t *) shared_buf_ptr)->mp_info;

	assert(sp_mp_info != NULL);

	/*
	 * Point the shared buffer MP information pointer to where the info will
	 * be populated, just after the boot info.
	 */
	((secure_partition_boot_info_t *) shared_buf_ptr)->mp_info =
		(secure_partition_mp_info_t *) ((uintptr_t)shared_buf_ptr
				+ sizeof(secure_partition_boot_info_t));

	/*
	 * Update the shared buffer pointer to where the MP information for the
	 * payload will be populated
	 */
	shared_buf_ptr = ((secure_partition_boot_info_t *) shared_buf_ptr)->mp_info;

	/*
	 * Copy the cpu information into the shared buffer area after the boot
	 * information.
	 */
	assert(sp_boot_info->num_cpus <= PLATFORM_CORE_COUNT);

	assert((uintptr_t)shared_buf_ptr
	       <= (PLAT_SPM_BUF_BASE + PLAT_SPM_BUF_SIZE -
		       (sp_boot_info->num_cpus * sizeof(*sp_mp_info))));

	memcpy(shared_buf_ptr, (const void *) sp_mp_info,
		sp_boot_info->num_cpus * sizeof(*sp_mp_info));

	/*
	 * Calculate the linear indices of cores in boot information for the
	 * secure partition and flag the primary CPU
	 */
	sp_mp_info = (secure_partition_mp_info_t *) shared_buf_ptr;

	for (unsigned int index = 0; index < sp_boot_info->num_cpus; index++) {
		u_register_t mpidr = sp_mp_info[index].mpidr;

		sp_mp_info[index].linear_id = plat_core_pos_by_mpidr(mpidr);
		if (plat_my_core_pos() == sp_mp_info[index].linear_id)
			sp_mp_info[index].flags |= MP_INFO_FLAG_PRIMARY_CPU;
	}

	VERBOSE("S-EL1/S-EL0 context setup end.\n");
}