Boot mem handling #7039

Add -Wno-gnu-alignof-expression to the warnings flag for Clang in order to avoid warnings like: '_Alignof' applied to an expression is a GNU extension [-Werror,-Wgnu-alignof-expression] when alignof() is applied on an expression like dereferencing a pointer to get the alignment of type. Signed-off-by: Jens Wiklander <[email protected]> Reviewed-by: Jerome Forissier <[email protected]> Reviewed-by: Etienne Carriere <[email protected]>

Increase STACK_THREAD_SIZE when CFG_CORE_DEBUG_CHECK_STACKS=y. Signed-off-by: Jens Wiklander <[email protected]> Reviewed-by: Jerome Forissier <[email protected]> Reviewed-by: Etienne Carriere <[email protected]>

Add a wrapper function for virt_to_phys() using vaddr_t instead of a void pointer. Signed-off-by: Jens Wiklander <[email protected]> Reviewed-by: Jerome Forissier <[email protected]> Reviewed-by: Etienne Carriere <[email protected]>

The buffer attribute CORE_MEM_TA_RAM isn't used to query the status of a buffer anywhere. So remove the attribute to allow future simplifications. Signed-off-by: Jens Wiklander <[email protected]> Reviewed-by: Jerome Forissier <[email protected]> Reviewed-by: Etienne Carriere <[email protected]>

Add VCORE_FREE_{PA,SZ,END_PA} defines to identify the unused and free memory range at the end of TEE_RAM_START..(TEE_RAM_START + TEE_RAM_VA_SIZE). VCORE_FREE_SZ is 0 in a pager configuration since all the memory is used by the pager. The VCORE_FREE range is excluded from the TEE_RAM_RW area for CFG_NS_VIRTUALIZATION=y and instead put in a separate NEX_RAM_RW area. This makes each partition use a bit less memory and leaves the VCORE_FREE range available for the Nexus. The VCORE_FREE range is added to the TEE_RAM_RW area for the normal configuration with CFG_NS_VIRTUALIZATION=n and CFG_WITH_PAGER=n. It's in practice unchanged behaviour in this configuration. Signed-off-by: Jens Wiklander <[email protected]> Acked-by: Jerome Forissier <[email protected]> Reviewed-by: Etienne Carriere <[email protected]>

Allow unmapping core memory in the VCORE_FREE range when the original boot mapping isn't needed any more. Signed-off-by: Jens Wiklander <[email protected]> Reviewed-by: Jerome Forissier <[email protected]> Reviewed-by: Etienne Carriere <[email protected]>

Replace MEM_AREA_TA_RAM with MEM_AREA_SEC_RAM_OVERALL. All read/write secure memory is covered by MEM_AREA_SEC_RAM_OVERALL, sometimes using an aliased map. But secure read-only or execute core memory is not covered as that would defeat the purpose of CFG_CORE_RWDATA_NOEXEC. Since the partition TA memory isn't accessed via MEM_AREA_TA_RAM any longer, don't map it using the partition specific map. This is needed later where unification of OP-TEE core and physical TA memory is possible. Signed-off-by: Jens Wiklander <[email protected]>

In configurations where secure core and TA memory is allocated from the same contiguous physical memory block, carve out the memory needed by OP-TEE core and make the rest available as TA memory. This is needed by later patches where more core memory is allocated as needed from the pool of TA memory. Signed-off-by: Jens Wiklander <[email protected]> Reviewed-by: Etienne Carriere <[email protected]>

With CFG_NS_VIRTUALIZATION=y let phys_mem_core_alloc() allocate from both the core_pool and ta_pool since both pools keep equally secure memory. This is needed in later patches when some translation tables are dynamically allocated from spare physical core memory. Signed-off-by: Jens Wiklander <[email protected]> Reviewed-by: Etienne Carriere <[email protected]>

Increase MAX_XLAT_TABLES by 2 to be able to map all TEE memory with 4k pages. Signed-off-by: Jens Wiklander <[email protected]> Reviewed-by: Etienne Carriere <[email protected]>

TEE memory is always supposed to be mapped with 4k pages for maximum flexibility, but can_map_at_level() doesn't check the requested block size for a region, so fix that. However, assign_mem_granularity() assigns smaller than necessary block sizes on page aligned regions, so fix that by only requesting 4k granularity for TEE memory and PGDIR granularity for the rest. This is needed in later patches where some TEE memory is unmapped. Signed-off-by: Jens Wiklander <[email protected]>

This concerns configurations with CFG_WITH_PAGER=y. Until this patch, even if __vcore_init_ro_size (VCORE_INIT_RO_SZ) is 0 for CFG_CORE_RODATA_NOEXEC=n, __vcore_init_ro_start was using some value smaller than __vcore_init_rx_end. To simplify code trying to find the end of VCORE_INIT_RX and VCORE_INIT_RO parts of the binary, make sure that __vcore_init_ro_start follows right after __vcore_init_rx_end. Signed-off-by: Jens Wiklander <[email protected]> Reviewed-by: Etienne Carriere <[email protected]>

For CFG_WITH_PAGER=y map the remaining memory following the VCORE_INIT_RO memory to make sure that all physical TEE memory is mapped even if VCORE_INIT_RO doesn't cover it entirely. This will be used in later patches to use the temporarily unused memory while booting. Signed-off-by: Jens Wiklander <[email protected]>

Adds CFG_BOOT_MEM to support stack-like memory allocations during boot before a heap has been configured. Signed-off-by: Jens Wiklander <[email protected]>

Add boot_cached_mem_end in C code, replacing the previous read-only mapped cached_mem_end. This allows updates to boot_cached_mem_end after MMU has been enabled. Signed-off-by: Jens Wiklander <[email protected]>

Enable CFG_BOOT_MEM unconditionally and call the boot_mem_*() functions as needed from entry_*.S and boot.c. The pager will reuse all boot_mem memory internally when configured. The non-pager configuration will unmap the memory and make it available for TAs if needed. __FLATMAP_PAGER_TRAILING_SPACE is removed from the link script, collect_mem_ranges() in core/mm/core_mmu.c maps the memory following VCORE_INIT_RO automatically. Signed-off-by: Jens Wiklander <[email protected]>

With CFG_BOOT_MEM enabled, allocate a temporary memory map array using boot_mem_alloc_tmp() instead of using the global static_mmap_regions[]. core_mmu_save_mem_map() is added and called from boot_init_primary_late() before the temporary memory is reused. Signed-off-by: Jens Wiklander <[email protected]>

Initialize guest physical memory in virt_guest_created() before the first entry into the guest from normal world. This replaces the call to core_mmu_init_phys_mem() in init_tee_runtime(). Remove unused code in core_mmu_init_phys_mem() and the now unused functions core_mmu_get_ta_range() and virt_get_ta_ram(). Signed-off-by: Jens Wiklander <[email protected]>

Moves the implementation of core_mmu_init_virtualization() into core_mmu_init_phys_mem(). This simplifies init_primary() in core/arch/arm/kernel/boot.c. Signed-off-by: Jens Wiklander <[email protected]>

Add a log entry when CFG_NS_VIRTUALIZATION is enabled, for example: D/TC:0 0 boot_init_primary_late:1028 NS-Virtualization enabled, supporting 2 guests Signed-off-by: Jens Wiklander <[email protected]> Reviewed-by: Jerome Forissier <[email protected]>

dump_mmap_table() is updated to use map->va directly as it's always the start of the mapping regardless of map->region_size. Signed-off-by: Jens Wiklander <[email protected]>

Add missing ADD_PHYS_MEM(MEM_AREA_SEC_RAM_OVERALL,...) for CFG_CORE_RWDATA_NOEXEC=n. Signed-off-by: Jens Wiklander <[email protected]>

Addressing a comment. Signed-off-by: Jens Wiklander <[email protected]>

Remove the read-only mapping between TEE_RAM_START and TEE_LOAD_ADDR. Signed-off-by: Jens Wiklander <[email protected]>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Boot mem handling #7039

Boot mem handling #7039

Commits on Sep 30, 2024

Boot mem handling #7039

Are you sure you want to change the base?

Boot mem handling #7039

Commits on Sep 30, 2024