Ideally, after kernel assumes control of the platform, firmware shouldn't access EFI boot services code/data regions. But, it's noticed that this is not so true in many x86 platforms. Hence, during boot, kernel reserves EFI boot services code/data regions [1] and maps [2] them to efi_pgd so that call to set_virtual_address_map() doesn't fail. After returning from set_virtual_address_map(), kernel frees the reserved regions [3] but they still remain mapped.
This means that any code that's running in efi_pgd address space (e.g: any EFI runtime service) would still be able to access EFI boot services code/data regions but the contents of these regions would have long been over written by someone else as they are freed by efi_free_boot_services(). So, it's important to unmap these regions. After unmapping EFI boot services code/data regions, any illegal access by buggy firmware to these regions would result in page fault which will be handled by efi specific fault handler. Unmapping EFI boot services code/data regions will result in clearing PAGE_PRESENT bit and it shouldn't bother L1TF cases because it's already handled by protnone_mask() at arch/x86/include/asm/pgtable-invert.h. [1] Please see efi_reserve_boot_services() [2] Please see efi_map_region() -> __map_region() [3] Please see efi_free_boot_services() Signed-off-by: Sai Praneeth Prakhya <[email protected]> Cc: Borislav Petkov <[email protected]> Cc: Ingo Molnar <[email protected]> Cc: Andy Lutomirski <[email protected]> Cc: Dave Hansen <[email protected]> Cc: Bhupesh Sharma <[email protected]> Cc: Thomas Gleixner <[email protected]> Cc: Peter Zijlstra <[email protected]> Cc: Ard Biesheuvel <[email protected]> --- arch/x86/include/asm/pgtable_types.h | 2 ++ arch/x86/mm/pageattr.c | 26 ++++++++++++++++++++++++++ arch/x86/platform/efi/quirks.c | 25 +++++++++++++++++++++++++ 3 files changed, 53 insertions(+) diff --git a/arch/x86/include/asm/pgtable_types.h b/arch/x86/include/asm/pgtable_types.h index b64acb08a62b..cda04ecf5432 100644 --- a/arch/x86/include/asm/pgtable_types.h +++ b/arch/x86/include/asm/pgtable_types.h @@ -566,6 +566,8 @@ extern pmd_t *lookup_pmd_address(unsigned long address); extern phys_addr_t slow_virt_to_phys(void *__address); extern int kernel_map_pages_in_pgd(pgd_t *pgd, u64 pfn, unsigned long address, unsigned numpages, unsigned long page_flags); +extern int kernel_unmap_pages_in_pgd(pgd_t *pgd, unsigned long address, + unsigned long numpages); #endif /* !__ASSEMBLY__ */ #endif /* _ASM_X86_PGTABLE_DEFS_H */ diff --git a/arch/x86/mm/pageattr.c b/arch/x86/mm/pageattr.c index 51a5a69ecac9..248f16181bed 100644 --- a/arch/x86/mm/pageattr.c +++ b/arch/x86/mm/pageattr.c @@ -2147,6 +2147,32 @@ int kernel_map_pages_in_pgd(pgd_t *pgd, u64 pfn, unsigned long address, return retval; } +int kernel_unmap_pages_in_pgd(pgd_t *pgd, unsigned long address, + unsigned long numpages) +{ + int retval; + + /* + * The typical sequence for unmapping is to find a pte through + * lookup_address_in_pgd() (ideally, it should never return NULL because + * the address is already mapped) and change it's protections. + * As pfn is the *target* of a mapping, it's not useful while unmapping. + */ + struct cpa_data cpa = { + .vaddr = &address, + .pgd = pgd, + .numpages = numpages, + .mask_set = __pgprot(0), + .mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW), + .flags = 0, + }; + + retval = __change_page_attr_set_clr(&cpa, 0); + __flush_tlb_all(); + + return retval; +} + /* * The testcases use internal knowledge of the implementation that shouldn't * be exposed to the rest of the kernel. Include these directly here. diff --git a/arch/x86/platform/efi/quirks.c b/arch/x86/platform/efi/quirks.c index 669babcaf245..fb1c44b11235 100644 --- a/arch/x86/platform/efi/quirks.c +++ b/arch/x86/platform/efi/quirks.c @@ -370,6 +370,24 @@ void __init efi_reserve_boot_services(void) } } +/* + * Apart from having VA mappings for EFI boot services code/data regions, + * (duplicate) 1:1 mappings were also created as a quirk for buggy firmware. So, + * unmap both 1:1 and VA mappings. + */ +static void __init efi_unmap_pages(efi_memory_desc_t *md) +{ + pgd_t *pgd = efi_mm.pgd; + u64 pa = md->phys_addr; + u64 va = md->virt_addr; + + if (kernel_unmap_pages_in_pgd(pgd, pa, md->num_pages)) + pr_err("Failed to unmap 1:1 mapping for 0x%llx\n", pa); + + if (kernel_unmap_pages_in_pgd(pgd, va, md->num_pages)) + pr_err("Failed to unmap VA mapping for 0x%llx\n", va); +} + void __init efi_free_boot_services(void) { phys_addr_t new_phys, new_size; @@ -394,6 +412,13 @@ void __init efi_free_boot_services(void) continue; } + /* + * Before calling set_virtual_address_map(), EFI boot services + * code/data regions were mapped as a quirk for buggy firmware. + * Unmap them from efi_pgd before freeing them up. + */ + efi_unmap_pages(md); + /* * Nasty quirk: if all sub-1MB memory is used for boot * services, we can get here without having allocated the -- 2.19.1
