[U-Boot] [PATCH v5 1/7] cmd: add efidebug command

[AKASHI Takahiro <takahiro.akashi@linaro.org>]

On Tue, Jan 22, 2019 at 10:18:58AM +0100, Alexander Graf wrote:
> 
> 
> On 22.01.19 02:02, AKASHI Takahiro wrote:
> > Alex,
> > 
> > On Mon, Jan 21, 2019 at 02:07:31PM +0100, Alexander Graf wrote:
> >> On 01/21/2019 08:49 AM, AKASHI Takahiro wrote:
> >>> Currently, there is no easy way to add or modify UEFI variables.
> >>> In particular, bootmgr supports BootOrder/BootXXXX variables, it is
> >>> quite hard to define them as u-boot variables because they are represented
> >>> in a complicated and encoded format.
> >>>
> >>> The new command, efidebug, helps address these issues and give us
> >>> more friendly interfaces:
> >>>  * efidebug boot add: add BootXXXX variable
> >>>  * efidebug boot rm: remove BootXXXX variable
> >>>  * efidebug boot dump: display all BootXXXX variables
> >>>  * efidebug boot next: set BootNext variable
> >>>  * efidebug boot order: set/display a boot order (BootOrder)
> >>>  * efidebug setvar: set an UEFI variable (with limited functionality)
> >>>  * efidebug dumpvar: display all UEFI variables
> >>>
> >>> Please note that the file, efidebug.c, will be compiled under
> >>> CONFIG_EFI_LOADER because some helper functions will be used
> >>> to enable "env -e" command in a later patch whether or not
> >>> the command is compiled in.
> >>>
> >>> Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
> >>> ---
> >>>  MAINTAINERS       |   1 +
> >>>  cmd/Kconfig       |  10 +
> >>>  cmd/Makefile      |   1 +
> >>>  cmd/efidebug.c    | 755 ++++++++++++++++++++++++++++++++++++++++++++++
> >>>  include/command.h |   6 +
> >>>  5 files changed, 773 insertions(+)
> >>>  create mode 100644 cmd/efidebug.c
> >>>
> >>> diff --git a/MAINTAINERS b/MAINTAINERS
> >>> index ae825014bda9..301c5c69ea25 100644
> >>> --- a/MAINTAINERS
> >>> +++ b/MAINTAINERS
> >>> @@ -438,6 +438,7 @@ F:	lib/efi*/
> >>>  F:	test/py/tests/test_efi*
> >>>  F:	test/unicode_ut.c
> >>>  F:	cmd/bootefi.c
> >>> +F:	cmd/efidebug.c
> >>>  F:	tools/file2include.c
> >>>  FPGA
> >>> diff --git a/cmd/Kconfig b/cmd/Kconfig
> >>> index ea1a325eb301..d9cab3cc0c49 100644
> >>> --- a/cmd/Kconfig
> >>> +++ b/cmd/Kconfig
> >>> @@ -1397,6 +1397,16 @@ config CMD_DISPLAY
> >>>  	  displayed on a simple board-specific display. Implement
> >>>  	  display_putc() to use it.
> >>> +config CMD_EFIDEBUG
> >>> +	bool "efidebug - display/customize UEFI environment"
> >>> +	depends on EFI_LOADER
> >>> +	default n
> >>> +	help
> >>> +	  Enable the 'efidebug' command which provides a subset of UEFI
> >>> +	  shell utility with simplified functionality. It will be useful
> >>> +	  particularly for managing boot parameters as  well as examining
> >>> +	  various EFI status for debugging.
> >>> +
> >>>  config CMD_LED
> >>>  	bool "led"
> >>>  	default y if LED
> >>> diff --git a/cmd/Makefile b/cmd/Makefile
> >>> index 15ae4d250f50..e48d34c394ee 100644
> >>> --- a/cmd/Makefile
> >>> +++ b/cmd/Makefile
> >>> @@ -51,6 +51,7 @@ obj-$(CONFIG_CMD_ECHO) += echo.o
> >>>  obj-$(CONFIG_ENV_IS_IN_EEPROM) += eeprom.o
> >>>  obj-$(CONFIG_CMD_EEPROM) += eeprom.o
> >>>  obj-$(CONFIG_EFI_STUB) += efi.o
> >>> +obj-$(CONFIG_EFI_LOADER) += efidebug.o
> >>>  obj-$(CONFIG_CMD_ELF) += elf.o
> >>>  obj-$(CONFIG_HUSH_PARSER) += exit.o
> >>>  obj-$(CONFIG_CMD_EXT4) += ext4.o
> >>> diff --git a/cmd/efidebug.c b/cmd/efidebug.c
> >>> new file mode 100644
> >>> index 000000000000..c54fb6cfa101
> >>> --- /dev/null
> >>> +++ b/cmd/efidebug.c
> >>> @@ -0,0 +1,755 @@
> >>> +// SPDX-License-Identifier: GPL-2.0+
> >>> +/*
> >>> + *  UEFI Shell-like command
> >>> + *
> >>> + *  Copyright (c) 2018 AKASHI Takahiro, Linaro Limited
> >>> + */
> >>> +
> >>> +#include <charset.h>
> >>> +#include <common.h>
> >>> +#include <command.h>
> >>> +#include <efi_loader.h>
> >>> +#include <environment.h>
> >>> +#include <errno.h>
> >>> +#include <exports.h>
> >>> +#include <hexdump.h>
> >>> +#include <malloc.h>
> >>> +#include <search.h>
> >>> +#include <linux/ctype.h>
> >>> +#include <asm/global_data.h>
> >>> +
> >>> +static void dump_var_data(char *data, unsigned long len)
> >>> +{
> >>> +	char *start, *end, *p;
> >>> +	unsigned long pos, count;
> >>> +	char hex[3], line[9];
> >>> +	int i;
> >>> +
> >>> +	end = data + len;
> >>> +	for (start = data, pos = 0; start < end; start += count, pos += count) {
> >>> +		count = end - start;
> >>> +		if (count > 16)
> >>> +			count = 16;
> >>> +
> >>> +		/* count should be multiple of two */
> >>> +		printf("%08lx: ", pos);
> >>> +
> >>> +		/* in hex format */
> >>> +		p = start;
> >>> +		for (i = 0; i < count / 2; p += 2, i++)
> >>> +			printf(" %c%c", *p, *(p + 1));
> >>> +		for (; i < 8; i++)
> >>> +			printf("   ");
> >>> +
> >>> +		/* in character format */
> >>> +		p = start;
> >>> +		hex[2] = '\0';
> >>> +		for (i = 0; i < count / 2; i++) {
> >>> +			hex[0] = *p++;
> >>> +			hex[1] = *p++;
> >>> +			line[i] = (char)simple_strtoul(hex, 0, 16);
> >>> +			if (line[i] < 0x20 || line[i] > 0x7f)
> >>> +				line[i] = '.';
> >>> +		}
> >>> +		line[i] = '\0';
> >>> +		printf("  %s\n", line);
> >>> +	}
> >>> +}
> >>
> >> Is this print_hex_dump() reimplemented?
> > 
> > Actually, no.
> > A UEFI variable on u-boot is encoded as ascii representation of binary data.
> > That means, for example,
> > 
> >         => env set -e PlatformLang en
> >         => env print -e
> >         PlatformLang: {boot,run}(blob)
> >         00000000:  65 6e                    en
> >         => env print
> >         ...
> >         efi_8be4df61-93ca-11d2-aa0d-00e098032b8c_PlatformLang={boot,run}(blob)656e
> >         ...
> > 
> > the value of "PlatformLang" as a u-boot variable here is "656e", not "en."
> > So if we want to use print_hex_dump(), we first have to convert that
> > string to a binary. But then print_hex_dump() converts the binary to
> > a string. It's just rediculuous, isn't it?
> 
> I actually think I would prefer that. This way we consolidate everything
> through a single API which means we then know that the results are
> always the same. If we implement parsing multiple times, there's a good
> chance we'll have a bug in one of them which gives us different results
> which then again makes debugging harder rather than easier.

Agree, but please note we have to add an apparently weird dependency of
CONFIG_HEXDUMP to EFI_LOADER as do_efi_dump_var_ext() needs to be compiled in
whether CMD_EFIDEBUG is enabled or not.

> > You might think that the value in this case should be
> >         {boot, run}(utf8)en
> >                     ^^^^
> > It's possible, but it depends on a variable and
> > currently my do_set_efi_var() doesn't support it anyway.
> 
> One more reason to use only a single path to funnel things through :).
> 
> So yes, can you maybe reuse RTS->get_variable() here? Same question on
> writing variables I suppose.

Well, we need GetNextVariableName API if we want to use GetVariable
(for listing). I have held off merging my another patch for this.
Is it time to do it?

-Takahiro Akashi

> 
> Alex