Hi Rob,
On 18 September 2017 at 11:03, Rob Clark robdclark@gmail.com wrote:
On Mon, Sep 18, 2017 at 11:07 AM, Rob Clark robdclark@gmail.com wrote:
On Mon, Sep 18, 2017 at 10:30 AM, Rob Clark robdclark@gmail.com wrote:
On Mon, Sep 18, 2017 at 9:31 AM, Rob Clark robdclark@gmail.com wrote:
On Mon, Sep 18, 2017 at 9:18 AM, Rob Clark robdclark@gmail.com wrote:
On Sun, Sep 17, 2017 at 11:48 PM, Heinrich Schuchardt xypron.glpk@gmx.de wrote:
On 09/18/2017 12:59 AM, Simon Glass wrote: > A limitation of the EFI loader at present is that it does not build with > sandbox. This makes it hard to write tests, since sandbox is used for most > testing in U-Boot. > > This series enables the EFI loader feature. It allows sandbox to build and > run a trivial function which calls the EFI API to output a message. > > Much work remains but this should serve as a basis for adding tests more > easily for EFI loader. > > This series sits on top of Heinrich's recent EFI test series. It is > available at u-boot-dm/efi-working > > > Simon Glass (16): > efi: Update efi_smbios_register() to return error code > efi: Move the init check inside efi_init_obj_list() > efi: Add error checking for efi_init_obj_list() > efi: Add a TODO to efi_init_obj_list() > efi: Correct header order in efi_memory > efi: sandbox: Adjust memory setup for sandbox > sandbox: smbios: Update to support sandbox > sandbox: Add a setjmp() implementation > efi: sandbox: Add required linker sections > efi: sandbox: Add distroboot support > Define board_quiesce_devices() in a shared location > Add a comment for board_quiesce_devices() > efi: sandbox: Add relocation constants > efi: Add a comment about duplicated ELF constants > efi: sandbox: Enable EFI loader builder for sandbox > efi: sandbox: Add a simple 'bootefi test' command > > arch/arm/include/asm/u-boot-arm.h | 1 - > arch/sandbox/cpu/cpu.c | 13 ++++++++++ > arch/sandbox/cpu/os.c | 17 ++++++++++++ > arch/sandbox/cpu/u-boot.lds | 29 +++++++++++++++++++++ > arch/sandbox/include/asm/setjmp.h | 21 +++++++++++++++ > arch/sandbox/lib/Makefile | 2 +- > arch/sandbox/lib/sections.c | 12 +++++++++ > arch/x86/include/asm/u-boot-x86.h | 1 - > arch/x86/lib/bootm.c | 4 --- > cmd/bootefi.c | 54 ++++++++++++++++++++++++++++++++++----- > common/bootm.c | 4 +++ > configs/sandbox_defconfig | 1 + > include/bootm.h | 8 ++++++ > include/config_distro_bootcmd.h | 2 +- > include/efi_loader.h | 13 ++++++++-- > include/os.h | 21 +++++++++++++++ > lib/efi_loader/Kconfig | 12 ++++++++- > lib/efi_loader/Makefile | 1 + > lib/efi_loader/efi_boottime.c | 4 +++ > lib/efi_loader/efi_memory.c | 33 +++++++++++++----------- > lib/efi_loader/efi_runtime.c | 7 +++++ > lib/efi_loader/efi_smbios.c | 6 +++-- > lib/efi_loader/efi_test.c | 17 ++++++++++++ > lib/smbios.c | 38 ++++++++++++++++++++------- > 24 files changed, 277 insertions(+), 44 deletions(-) > create mode 100644 arch/sandbox/include/asm/setjmp.h > create mode 100644 arch/sandbox/lib/sections.c > create mode 100644 lib/efi_loader/efi_test.c > Thanks for enabling efi_loader on sandbox. That will make many things easier.
Unfortunately efi_status_t EFIAPI efi_selftest(efi_handle_t image_handle, struct efi_system_table *systab) { ... boottime = systable->boottime; ... ret = boottime->allocate_pool(EFI_BOOT_SERVICES_DATA, map_size, (void **)&memory_map); leads to a segmentation fault:
I'm seeing something similar, because:
(gdb) print gd->bd->bi_dram[0] $2 = {start = 0, size = 134217728}
u-boot expects 1:1 phys:virt mapping, so that probably won't work.
The following quick hack works.. something similar could probably be smashed in to ""
diff --git a/lib/efi_loader/efi_memory.c b/lib/efi_loader/efi_memory.c index cddafe2d43..da2079a4b1 100644 --- a/lib/efi_loader/efi_memory.c +++ b/lib/efi_loader/efi_memory.c @@ -459,9 +459,10 @@ int efi_memory_init(void) unsigned long uboot_start, uboot_pages; unsigned long uboot_stack_size = 16 * 1024 * 1024;
efi_add_known_memory();
if (!IS_ENABLED(CONFIG_SANDBOX)) {
efi_add_known_memory();/* Add U-Boot */ uboot_start = (gd->start_addr_sp - uboot_stack_size) & ~EFI_PAGE_MASK;@@ -476,6 +477,12 @@ int efi_memory_init(void) runtime_pages = (runtime_end - runtime_start) >> EFI_PAGE_SHIFT; efi_add_memory_map(runtime_start, runtime_pages, EFI_RUNTIME_SERVICES_CODE, false);
- } else {
+#define SZ_256M 0x10000000
size_t sz = SZ_256M;void *ram = os_malloc(sz);efi_add_memory_map((uintptr_t)ram, sz >> EFI_PAGE_SHIFT,EFI_CONVENTIONAL_MEMORY, false);#ifdef CONFIG_EFI_LOADER_BOUNCE_BUFFER
With that I'm at least getting further.. efi_allocate_pool() eventually fails, possibly making every small memory allocation page aligned means that 256m isn't enough..
Ok, still just as hacky, but works a bit better:
diff --git a/lib/efi_loader/efi_memory.c b/lib/efi_loader/efi_memory.c index cddafe2d43..b546b5e35d 100644 --- a/lib/efi_loader/efi_memory.c +++ b/lib/efi_loader/efi_memory.c @@ -14,6 +14,7 @@ #include <linux/list_sort.h> #include <inttypes.h> #include <watchdog.h> +#include <os.h>
DECLARE_GLOBAL_DATA_PTR;
@@ -459,9 +460,9 @@ int efi_memory_init(void) unsigned long uboot_start, uboot_pages; unsigned long uboot_stack_size = 16 * 1024 * 1024;
- efi_add_known_memory();
- if (!IS_ENABLED(CONFIG_SANDBOX)) {
efi_add_known_memory();/* Add U-Boot */ uboot_start = (gd->start_addr_sp - uboot_stack_size) & ~EFI_PAGE_MASK;@@ -476,6 +477,14 @@ int efi_memory_init(void) runtime_pages = (runtime_end - runtime_start) >> EFI_PAGE_SHIFT; efi_add_memory_map(runtime_start, runtime_pages, EFI_RUNTIME_SERVICES_CODE, false);
- } else {
+#define SZ_4K 0x00001000 +#define SZ_256M 0x10000000
size_t sz = SZ_256M;uintptr_t ram = (uintptr_t)os_malloc(sz + SZ_4K) + SZ_4K;efi_add_memory_map(ram & ~EFI_PAGE_MASK, sz >> EFI_PAGE_SHIFT,EFI_CONVENTIONAL_MEMORY, false);gd->start_addr_sp = ~0;#ifdef CONFIG_EFI_LOADER_BOUNCE_BUFFER
At this point it crashes in efi_load_pe() when it first tries to dereference the address of the image passed in, ie. I'm running:
host bind 0 x86_64-sct.img load host 0:1 0x01000000 /efi/boot/shell.efi bootefi 0x01000000
Not sure if there is a better way to pick an address to load into. Or maybe just assuming that PA==VA isn't a good idea in sandbox?
Ok, I realized there is map_sysmem().. which gets me further.. efi_loader really expects identity mapping (PA==VA), and iirc this is what UEFI spec expects too so I wouldn't necessarily call it a bug in efi_loader.
So, I don't know x86(_64) asm or calling conventions as well as arm.. but I wonder if we are screwing up something long those lines:
0000000000000280 <.text>: 280: 48 89 5c 24 08 mov %rbx,0x8(%rsp) 285: 57 push %rdi 286: 48 83 ec 20 sub $0x20,%rsp 28a: 48 8b f9 mov %rcx,%rdi
28d: e8 1e 00 00 00 callq 0x2b0
this jump is taken to 0x2b0
292: e8 2d 06 00 00 callq 0x8c4 297: 48 8b cf mov %rdi,%rcx 29a: 48 8b d8 mov %rax,%rbx 29d: e8 ea 01 00 00 callq 0x48c 2a2: 48 8b c3 mov %rbx,%rax 2a5: 48 8b 5c 24 30 mov 0x30(%rsp),%rbx 2aa: 48 83 c4 20 add $0x20,%rsp 2ae: 5f pop %rdi 2af: c3 retq2b0: 40 53 rex push %rbx
2b2: 48 83 ec 20 sub $0x20,%rsp 2b6: 48 89 0d e3 b9 05 00 mov %rcx,0x5b9e3(%rip)# 0x5bca0 2bd: 4c 8d 05 f4 b9 05 00 lea 0x5b9f4(%rip),%r8 # 0x5bcb8
2c4: 48 8b 42 60 mov 0x60(%rdx),%rax
and at 0x2c4 %rdx is 0x2.. I always thought x86 asm syntax strange, but I assume that is trying to write to value of %rdx + offset of 0x60?? But this is a register never written, so I assume it is expected to be passed from efi_loader?
From https://en.wikipedia.org/wiki/X86_calling_conventions it seems that MS calling convention expects 2nd arg in %rdx, but linux/gcc calling convention expects 3rd arg in %rdx (there is no 3rd arg)..
I don't know it well either. But so long as functions are properly declared in header files I don't really understand how we can have a mismatch.
Regards, Simon