[U-Boot] bootefi failures on armv7
Ard Biesheuvel
ard.biesheuvel at linaro.org
Thu Apr 11 18:15:13 UTC 2019
On Wed, 10 Apr 2019 at 09:57, Ilias Apalodimas
<ilias.apalodimas at linaro.org> wrote:
>
> Hi Alexander, Heinrich,
>
> (Resending since i was too quick on the trigger forgot to cc the public
> mailing list. Sorry for the mess)
>
> I can't get an armv7 board (STM32MP157C-DK2) to boot with bootefi on
> specific kernel/uboot configurations.
> u-boot version: U-Boot 2019.04-rc3
> kernel version: 5.0.0-rc3
>
> What works:
> - bootefi + 'dtb=' on kernel cmdline regardless of kernel config
> - bootefi + fdtcontroladdr if CONFIG_ARM_LPAE=y on my kernel config
>
> What doesn't work:
> bootefi + fdtcontroladdr if CONFIG_ARM_LPAE is not set.
>
> Once this config is selected the kernel hangs on a BUG_ON() in arm_pte_alloc()
>
> Adding a dump_stack() there throws the relevant information.
> [ 0.000000] CPU: 0 PID: 0 Comm: swapper Not tainted 5.0.0-rc3-29427-g769f1f8f9b56-dirty #127
> [ 0.000000] Hardware name: STM32 (Device Tree Support)
> [ 0.000000] [<c03123ec>] (unwind_backtrace) from [<c030ce0c>] (show_stack+0x10/0x14)
> [ 0.000000] [<c030ce0c>] (show_stack) from [<c0e27250>] (dump_stack+0x8c/0xa0)
> [ 0.000000] [<c0e27250>] (dump_stack) from [<c1507e9c>] (arm_pte_alloc+0x74/0xb8)
> [ 0.000000] [<c1507e9c>] (arm_pte_alloc) from [<c15081ec>] (__create_mapping+0x30c/0x36c)
> [ 0.000000] [<c15081ec>] (__create_mapping) from [<c15089f8>] (paging_init+0x234/0x648)
> [ 0.000000] [<c15089f8>] (paging_init) from [<c1504950>] (setup_arch+0x660/0xcac)
> [ 0.000000] [<c1504950>] (setup_arch) from [<c1500a4c>] (start_kernel+0x70/0x458)
> [ 0.000000] [<c1500a4c>] (start_kernel) from [<00000000>] ( (null))
>
> Kernel's EFI map:
> [ 0.000000] efi: 0x0000c0000000-0x0000c1ffffff [Boot Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c2000000-0x0000c2860fff [Loader Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c2861000-0x0000c7cfffff [Conventional Memory| | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c7d00000-0x0000c7efffff [Loader Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c7f00000-0x0000c7f0bfff [Runtime Data |RUN| | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c7f0c000-0x0000dc705fff [Conventional Memory| | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dc706000-0x0000dc709fff [Loader Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dc70a000-0x0000dcf6afff [Loader Code | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dcf6b000-0x0000dcf6bfff [Runtime Data |RUN| | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dcf6c000-0x0000dcf72fff [Reserved | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dcf73000-0x0000dcf73fff [Runtime Data |RUN| | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dcf74000-0x0000dcf75fff [Reserved | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dcf76000-0x0000dff80fff [Loader Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dff81000-0x0000dff81fff [Runtime Code |RUN| | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dff82000-0x0000dfffffff [Loader Data | | | | | | | | |WB| | | ]
>
> I've found some hacky workarounds on that and i'd like some input on what's
> actually wrong
>
> - Workaround 1, add a bogus bootefi cmd
>
> instead of
> load mmc 0:5 $kernel_addr_r efi/BOOT/zImage;
> bootefi $kernel_addr_r $fdtcontroladdr
>
> doing
> bootefi $kernel_addr_r $fdtcontroladdr
> load mmc 0:5 $kernel_addr_r efi/BOOT/zImage
> bootefi $kernel_addr_r $fdtcontroladdr' works
>
> The first bootefi command will fail since there's nothing loaded on
> that address yet. What happens is that u-boot adds an extra
> memory_map, via efi_add_memory_map() and the kernel boots
>
> Kernel's EFI map:
> [ 0.000000] efi: Processing EFI memory map:
> [ 0.000000] efi: 0x0000c0000000-0x0000c1ffffff [Boot Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c2000000-0x0000c2860fff [Loader Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c2861000-0x0000c7cf3fff [Conventional Memory| | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c7cf4000-0x0000c7ef3fff [Loader Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c7ef4000-0x0000c7f0bfff [Runtime Data |RUN| | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c7f0c000-0x0000dc704fff [Conventional Memory| | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dc705000-0x0000dc708fff [Loader Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dc709000-0x0000dcf69fff [Loader Code | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dcf6a000-0x0000dcf6cfff [Reserved | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dcf6d000-0x0000dcf6dfff [Runtime Data |RUN| | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dcf6e000-0x0000dcf74fff [Reserved | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dcf75000-0x0000dcf75fff [Runtime Data |RUN| | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dcf76000-0x0000dff80fff [Loader Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dff81000-0x0000dff81fff [Runtime Code |RUN| | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000dff82000-0x0000dfffffff [Loader Data | | | | | | | | |WB| | | ]
>
>
> - Workaround 2, copy fdt one page before the 127mb limit
>
> diff --git a/cmd/bootefi.c b/cmd/bootefi.c
> index 3619a20e6433..a0b6156845da 100644
> --- a/cmd/bootefi.c
> +++ b/cmd/bootefi.c
> @@ -108,11 +108,13 @@ static efi_status_t copy_fdt(void **fdtp)
> * Safe fdt location is at 127 MiB.
> * On the sandbox convert from the sandbox address space.
> */
> - new_fdt_addr = (uintptr_t)map_sysmem(fdt_ram_start + 0x7f00000 +
> - fdt_size, 0);
> + //new_fdt_addr = (uintptr_t)map_sysmem(fdt_ram_start + 0x7f00000 +
> + //fdt_size, 0);
> + new_fdt_addr = (uintptr_t)map_sysmem(fdt_ram_start + 0x7f00000, fdt_size);
> ret = efi_allocate_pages(EFI_ALLOCATE_MAX_ADDRESS,
> EFI_RUNTIME_SERVICES_DATA, fdt_pages,
> &new_fdt_addr);
> + printf("FDTADDR: %llx pages %lu sz %x\n", new_fdt_addr, fdt_pages, fdt_size);
> if (ret != EFI_SUCCESS) {
> /* If we can't put it there, put it somewhere */
> new_fdt_addr = (ulong)memalign(EFI_PAGE_SIZE, fdt_size);
>
> This loads the fdt on c7ef4000 (which is more than a page). Changing
> the address from 0x7f00000 to 7EFF000, on the original code, works as
> well
>
> Kernel's EFI map (with the patch) :
> [ 0.000000] efi: Processing EFI memory map:
> [ 0.000000] efi: 0x0000c0000000-0x0000c1ffffff [Boot Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c2000000-0x0000c2860fff [Loader Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c2861000-0x0000c7cf3fff [Conventional Memory| | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c7cf4000-0x0000c7ef3fff [Loader Data | | | | | | | | |WB| | | ]
> [ 0.000000] efi: 0x0000c7ef4000-0x0000c7efffff [Runtime Data |RUN| | | | | | | |WB| | | ]
As an aside, putting the FDT in runtime data is not the right thing to do.
Runtime data sections are intended for data that is used by the
runtime services implementations themselves, which is why they
automatically get the EFI_MEMORY_RUNTIME attribute as well, and get
mapped into the EFI runtime address space. They also get flagged as
NOMAP regions, which means they get omitted from the linear map, which
causes unnecessary page table fragmentation leading to more TLB
pressure.
I recommend using boot services data here, or acpi reclaim if you are
concerned about the OS not reserving the region correctly.
More information about the U-Boot
mailing list