Skip to contentSkip to conversation

cmd: allow bootd without bootm

PATCH Superseded Mehmet Fide 3 participants 6 messages Jul 30 – Aug 6

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover cmd: allow bootd without bootm

From: Mehmet Fide <[email protected]>

The bootd command is a two line wrapper around run_command(env_get("bootcmd")), but its code lives in cmd/bootm.c, so it only links when CMD_BOOTM is enabled. Commit 1fd04cf46baf ("cmd: Correct dependencies for CMD_BOOTD") expressed that in Kconfig as "depends on CMD_BOOTM". The build reason is real, but the dependency reads as a functional one, and it leaves the boards that disable CMD_BOOTM without "boot" or "bootd" even though the implementation works there.

Patch 1 moves the command into cmd/bootd.c and drops the dependency, keeping the current default so that no board in tree changes. Patch 2 adds the test that bootd never had.

While testing I noticed something unrelated that this series does not touch:
on sandbox the "avoid bootd recursion" guard in cmd_process() never fires, so a bootcmd of "bootd" recurses until the stack is exhausted. That behaves identically before and after this series.

Mehmet Fide (2):
cmd: allow bootd without bootm
test: cmd: add a test for bootd

 cmd/Kconfig       |  3 +--
 cmd/Makefile      |  1 +
 cmd/bootd.c       | 30 ++++++++++++++++++++++++++++++
 cmd/bootm.c       | 24 ------------------------
 test/cmd/Makefile |  3 +++
 test/cmd/bootd.c  | 35 +++++++++++++++++++++++++++++++++++
 6 files changed, 70 insertions(+), 26 deletions(-)
 create mode 100644 cmd/bootd.c
 create mode 100644 test/cmd/bootd.c

base-commit: 94b349bd902d9e38e1846c157fadd9054c34680a

Signature · 1 line
2.54.0
#1 lore raw

PATCH 1/2 cmd: allow bootd without bootm

From: Mehmet Fide <[email protected]>

do_bootd() runs the "bootcmd" environment variable through run_command() and does nothing else. It uses no bootm functionality, and neither does the "bootd" recursion guard in cmd_process(). doc/usage/cmd/bootd.rst documents it that way, with an example that sets bootcmd to "echo Hello World".

Commit 1fd04cf46baf ("cmd: Correct dependencies for CMD_BOOTD") made CMD_BOOTD depend on CMD_BOOTM because the build fails otherwise:
common/command.c is compiled unconditionally and references do_bootd(), whose definition sits in cmd/bootm.c, which is only compiled when CMD_BOOTM=y. The dependency therefore describes where the code lives rather than what it needs, and the boards that disable CMD_BOOTM (colibri_vf, iot_devkit, mx6memcal and r8a78000_ironhide_cm33) cannot offer "boot" or "bootd" at all, although the implementation would work there.

Move do_bootd() and its two command registrations into cmd/bootd.c, compiled from CMD_BOOTD, and drop the artificial dependency. To leave every board in tree unchanged, CMD_BOOTD now defaults to y only when CMD_BOOTM is enabled, which is precisely the set of boards that have it today. Boards without bootm can enable it deliberately.

Tested on sandbox: bootd, the "boot" alias, the return value taken from a failing bootcmd and the recursion guard all behave as before. Also tested on a Colibri VF50 (vf610) board built with CMD_BOOTM=n and CMD_BOOTD=y, a combination that could not be selected before: "boot" and "bootd" run bootcmd and bring up the OS, and a bootcmd of "false" returns 1. A colibri_vf_defconfig build with CMD_BOOTD left off is byte identical to the previous one apart from the version string; enabling CMD_BOOTD costs 272 bytes.

---

Notes:
checkpatch reports two things on this patch that I left as they are:
- "does MAINTAINERS need updating?": no MAINTAINERS entry covers cmd/bootm.c, so the new cmd/bootd.c does not need one either.
- "Lines should not end with a '('": that is the U_BOOT_CMD idiom, and both registrations are moved verbatim.

 cmd/Kconfig  |  3 +--
 cmd/Makefile |  1 +
 cmd/bootd.c  | 30 ++++++++++++++++++++++++++++++
 cmd/bootm.c  | 24 ------------------------
 4 files changed, 32 insertions(+), 26 deletions(-)
 create mode 100644 cmd/bootd.c
cmd/Kconfig +1 -2
index 35ce42f8397..f8eed5a5bdf 100644
@@ -300,8 +300,7 @@ menu "Boot commands"
 
 config CMD_BOOTD
 	bool "bootd"
-	depends on CMD_BOOTM
-	default y
+	default y if CMD_BOOTM
 	help
 	  Run the command stored in the environment "bootcmd", i.e.
 	  "bootd" does the same thing as "run bootcmd".
cmd/Makefile +1 -0
index ce772e5555b..7c3db81e524 100644
@@ -6,6 +6,7 @@
 ifndef CONFIG_XPL_BUILD
 # core command
 obj-y += boot.o
+obj-$(CONFIG_CMD_BOOTD) += bootd.o
 obj-$(CONFIG_CMD_BOOTM) += bootm.o
 obj-$(CONFIG_CMD_HELP) += help.o
 obj-y += panic.o
cmd/bootd.c new file +30 -0
new file mode 100644
index 00000000000..d39b1bfe888
@@ -0,0 +1,30 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * (C) Copyright 2000-2009
+ * Wolfgang Denk, DENX Software Engineering, [email protected].
+ */
+
+/*
+ * bootd - boot default, i.e. run the command in the "bootcmd" environment
+ * variable
+ */
+#include <command.h>
+#include <env.h>
+
+int do_bootd(struct cmd_tbl *cmdtp, int flag, int argc, char *const argv[])
+{
+	return run_command(env_get("bootcmd"), flag);
+}
+
+U_BOOT_CMD(
+	boot,	1,	1,	do_bootd,
+	"boot default, i.e., run 'bootcmd'",
+	""
+);
+
+/* keep old command name "bootd" for backward compatibility */
+U_BOOT_CMD(
+	bootd, 1,	1,	do_bootd,
+	"boot default, i.e., run 'bootcmd'",
+	""
+);
cmd/bootm.c +0 -24
index ca7cec91fad..07fd995ebed 100644
@@ -229,30 +229,6 @@ U_BOOT_CMD(
 	"boot application image from memory", bootm_help_text
 );
 
-/*******************************************************************/
-/* bootd - boot default image */
-/*******************************************************************/
-#if defined(CONFIG_CMD_BOOTD)
-int do_bootd(struct cmd_tbl *cmdtp, int flag, int argc, char *const argv[])
-{
-	return run_command(env_get("bootcmd"), flag);
-}
-
-U_BOOT_CMD(
-	boot,	1,	1,	do_bootd,
-	"boot default, i.e., run 'bootcmd'",
-	""
-);
-
-/* keep old command name "bootd" for backward compatibility */
-U_BOOT_CMD(
-	bootd, 1,	1,	do_bootd,
-	"boot default, i.e., run 'bootcmd'",
-	""
-);
-
-#endif
-
 /*******************************************************************/
 /* iminfo - print header info for a requested image */
 /*******************************************************************/
Signature · 1 line
2.54.0
Reviewed-by on v1 1/2
#2 lore raw
Quoted 33 lines · Mehmet Fide

From: Mehmet Fide <[email protected]>

do_bootd() runs the "bootcmd" environment variable through run_command() and does nothing else. It uses no bootm functionality, and neither does the "bootd" recursion guard in cmd_process(). doc/usage/cmd/bootd.rst documents it that way, with an example that sets bootcmd to "echo Hello World".

Commit 1fd04cf46baf ("cmd: Correct dependencies for CMD_BOOTD") made CMD_BOOTD depend on CMD_BOOTM because the build fails otherwise:
common/command.c is compiled unconditionally and references do_bootd(), whose definition sits in cmd/bootm.c, which is only compiled when CMD_BOOTM=y. The dependency therefore describes where the code lives rather than what it needs, and the boards that disable CMD_BOOTM (colibri_vf, iot_devkit, mx6memcal and r8a78000_ironhide_cm33) cannot offer "boot" or "bootd" at all, although the implementation would work there.

Move do_bootd() and its two command registrations into cmd/bootd.c, compiled from CMD_BOOTD, and drop the artificial dependency. To leave every board in tree unchanged, CMD_BOOTD now defaults to y only when CMD_BOOTM is enabled, which is precisely the set of boards that have it today. Boards without bootm can enable it deliberately.

Tested on sandbox: bootd, the "boot" alias, the return value taken from a failing bootcmd and the recursion guard all behave as before. Also tested on a Colibri VF50 (vf610) board built with CMD_BOOTM=n and CMD_BOOTD=y, a combination that could not be selected before: "boot" and "bootd" run bootcmd and bring up the OS, and a bootcmd of "false" returns 1. A colibri_vf_defconfig build with CMD_BOOTD left off is byte identical to the previous one apart from the version string; enabling CMD_BOOTD costs 272 bytes.

Quoted 1 line
Signature · 1 line
Tom
Reviewed-by on v1 1/2
#3 lore raw
In reply to Mehmet Fide
Quoted 33 lines · Mehmet Fide

cmd: allow bootd without bootm

do_bootd() runs the "bootcmd" environment variable through run_command() and does nothing else. It uses no bootm functionality, and neither does the "bootd" recursion guard in cmd_process(). doc/usage/cmd/bootd.rst documents it that way, with an example that sets bootcmd to "echo Hello World".

Commit 1fd04cf46baf ("cmd: Correct dependencies for CMD_BOOTD") made CMD_BOOTD depend on CMD_BOOTM because the build fails otherwise:
common/command.c is compiled unconditionally and references do_bootd(), whose definition sits in cmd/bootm.c, which is only compiled when CMD_BOOTM=y. The dependency therefore describes where the code lives rather than what it needs, and the boards that disable CMD_BOOTM (colibri_vf, iot_devkit, mx6memcal and r8a78000_ironhide_cm33) cannot offer "boot" or "bootd" at all, although the implementation would work there.

Move do_bootd() and its two command registrations into cmd/bootd.c, compiled from CMD_BOOTD, and drop the artificial dependency. To leave every board in tree unchanged, CMD_BOOTD now defaults to y only when CMD_BOOTM is enabled, which is precisely the set of boards that have it today. Boards without bootm can enable it deliberately.

Tested on sandbox: bootd, the "boot" alias, the return value taken from a failing bootcmd and the recursion guard all behave as before. Also tested on a Colibri VF50 (vf610) board built with CMD_BOOTM=n and CMD_BOOTD=y, a combination that could not be selected before: "boot" and "bootd" run bootcmd and bring up the OS, and a bootcmd of "false" returns 1. A colibri_vf_defconfig build with CMD_BOOTD left off is byte identical to the previous one apart from the version string; enabling CMD_BOOTD costs 272 bytes.

Quoted 8 lines

cmd/Kconfig | 3 +--

 cmd/Makefile |  1 +
 cmd/bootd.c  | 30 ++++++++++++++++++++++++++++++
 cmd/bootm.c  | 24 ------------------------
 4 files changed, 32 insertions(+), 26 deletions(-)
#4 lore raw

PATCH 2/2 test: cmd: add a test for bootd

In reply to Mehmet Fide

From: Mehmet Fide <[email protected]>

There is no test for the bootd command. Add one to the cmd suite that covers the documented behaviour: bootd and its "boot" alias run the command held in the bootcmd environment variable, and the return value of bootd is the one of that command.

---

Notes:
checkpatch asks for a blank line before CMD_TEST(); test/cmd/command.c and test/cmd/exit.c place the macro directly after the closing brace, so this follows the surrounding tests.

 test/cmd/Makefile |  3 +++
 test/cmd/bootd.c  | 35 +++++++++++++++++++++++++++++++++++
 2 files changed, 38 insertions(+)
 create mode 100644 test/cmd/bootd.c
test/cmd/Makefile +3 -0
index 8d36463879d..757aefa3e02 100644
@@ -17,6 +17,9 @@ ifdef CONFIG_CONSOLE_RECORD
 obj-$(CONFIG_CMD_ACPI) += acpi.o
 endif
 obj-$(CONFIG_CMD_BDI) += bdinfo.o
+ifdef CONFIG_CONSOLE_RECORD
+obj-$(CONFIG_CMD_BOOTD) += bootd.o
+endif
 obj-$(CONFIG_CMD_CONFIG) += config.o
 obj-$(CONFIG_COREBOOT_SYSINFO) += coreboot.o
 obj-$(CONFIG_CMD_FDT) += fdt.o
test/cmd/bootd.c new file +35 -0
new file mode 100644
index 00000000000..83dc9ab6ecc
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * Tests for the bootd command
+ *
+ * Copyright 2026 Mehmet Fide <[email protected]>
+ */
+
+#include <command.h>
+#include <env.h>
+#include <test/cmd.h>
+#include <test/ut.h>
+
+/* Check that bootd and its "boot" alias run the bootcmd variable */
+static int cmd_bootd_test(struct unit_test_state *uts)
+{
+	ut_assertok(env_set("bootcmd", "echo hello bootd"));
+
+	ut_assertok(run_command("bootd", 0));
+	ut_assert_nextline("hello bootd");
+	ut_assert_console_end();
+
+	ut_assertok(run_command("boot", 0));
+	ut_assert_nextline("hello bootd");
+	ut_assert_console_end();
+
+	/* the return value is the one of the command in bootcmd */
+	ut_assertok(env_set("bootcmd", "false"));
+	ut_asserteq(1, run_command("bootd", 0));
+	ut_assert_console_end();
+
+	ut_assertok(env_set("bootcmd", NULL));
+
+	return 0;
+}
+CMD_TEST(cmd_bootd_test, UTF_CONSOLE);
Signature · 1 line
2.54.0
#5 lore raw

Re: [PATCH 2/2] test: cmd: add a test for bootd

Hi Mehmet,

Quoted 15 lines · Mehmet Fide

test: cmd: add a test for bootd

There is no test for the bootd command. Add one to the cmd suite that covers the documented behaviour: bootd and its "boot" alias run the command held in the bootcmd environment variable, and the return value of bootd is the one of that command.

test/cmd/Makefile | 3 +++

 test/cmd/bootd.c  | 35 +++++++++++++++++++++++++++++++++++
 2 files changed, 38 insertions(+)

test/cmd/Makefile ifdef CONFIG_CONSOLE_RECORD · line 22

Quoted 6 lines
test/cmd/Makefile +3 -0
@@ -17,6 +17,9 @@ ifdef CONFIG_CONSOLE_RECORD
 obj-$(CONFIG_CMD_ACPI) += acpi.o
 endif
 obj-$(CONFIG_CMD_BDI) += bdinfo.o
+ifdef CONFIG_CONSOLE_RECORD
+obj-$(CONFIG_CMD_BOOTD) += bootd.o
+endif

The CONFIG_CONSOLE_RECORD ifdef is not needed. UTF_CONSOLE means the runner skips the test when console recording is unavailable, and bdinfo.o just above has no such guard. Please add 'obj-$(CONFIG_CMD_BOOTD) += bootd.o' next to the bdinfo line.

test/cmd/bootd.c line 4

Quoted 6 lines
test/cmd/bootd.c +4 -0
@@ -0,0 +1,35 @@
+     /* the return value is the one of the command in bootcmd */
+     ut_assertok(env_set("bootcmd", "false"));
+     ut_asserteq(1, run_command("bootd", 0));
+     ut_assert_console_end();

'false' is a hush builtin, so this leg only works when the sandbox build uses that parser. Please either note the assumption in a comment or pick something that does not depend on the shell.

test/cmd/bootd.c line 3

Quoted 5 lines
test/cmd/bootd.c +3 -0
@@ -0,0 +1,35 @@
+static int cmd_bootd_test(struct unit_test_state *uts)
+{
+     ut_assertok(env_set("bootcmd", "echo hello bootd"));

The test clobbers 'bootcmd' and unsets it at the end without preserving the previous value. Please save it with env_get() and restore it in cleanup, so a later test that relies on the environment is not affected by ordering.

test/cmd/bootd.c line 5

Quoted 7 lines
test/cmd/bootd.c +5 -0
@@ -0,0 +1,35 @@
+     ut_assertok(env_set("bootcmd", NULL));
+
+     return 0;
+}
+CMD_TEST(cmd_bootd_test, UTF_CONSOLE);

The cover letter mentions the recursion guard in cmd_process() never fires on sandbox, so I won't ask for a test of that here, but it would be nice to assert that do_bootd() itself, when invoked with CMD_FLAG_BOOTD and bootcmd='bootd', does not recurse - that part of the guard lives in the code you moved, and you can exercise it directly with cmd_process_error() or by calling do_bootd() with the flag set. What do you think?

Regards,
Simon