Skip to contentSkip to conversation

fwu: Call EFI stack initialization after preboot

PATCH Superseded Michal Simek 5 participants 12 messages Mar 18 – Mar 31

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover fwu: Call EFI stack initialization after preboot

Hi,

the whole series is trying to address issues we have on platforms where usb and ufs init are not done before efi initialization. This is happenin on platforms where A/B update is enabled. FWU code is called early and do EFI initialization before usb and ufs started that's why they are not visible.

We are workarounding it by starting usb/ufs from board file but this should be more generic approach.

Thanks,
Michal

Changes in v2:
- new patch in series
- check return value

Michal Simek (3):
event: Check return value from event_notify_null()
event: Introduce EVT_POST_PREBOOT event
fwu: Move boottime checks to EVT_POST_PREBOOT

 common/board_r.c      | 6 +++++-
 common/main.c         | 4 ++++
 include/event.h       | 9 +++++++++
 lib/fwu_updates/fwu.c | 2 +-
 test/dm/fwu_mdata.c   | 4 ++--
 5 files changed, 21 insertions(+), 4 deletions(-)
Signature · 4 lines
2.43.0

base-commit: 61ddd555f58ef5169c48b190423640d949e4aad1
branch: debian-sent3
#1 lore raw

PATCH v2 1/3 event: Check return value from event_notify_null()

event_notify_null() returns int but its return value is not checked in run_main_loop() and in fwu_mdata tests. Add proper error checking to all unchecked call sites.

---

Changes in v2:
- new patch in series

 common/board_r.c    | 6 +++++-
 test/dm/fwu_mdata.c | 4 ++--
 2 files changed, 7 insertions(+), 3 deletions(-)
common/board_r.c +5 -1
index 76f9fc090fbe..00f29385d80a 100644
@@ -579,11 +579,15 @@ static int dm_announce(void)
 
 static int run_main_loop(void)
 {
+	int ret;
+
 #ifdef CONFIG_SANDBOX
 	sandbox_main_loop_init();
 #endif
 
-	event_notify_null(EVT_MAIN_LOOP);
+	ret = event_notify_null(EVT_MAIN_LOOP);
+	if (ret)
+		return ret;
 
 	/* main_loop() can return to retry autoboot, if so just run it again */
 	for (;;)
test/dm/fwu_mdata.c +2 -2
index b7680632f95f..643b647af0a9 100644
@@ -100,7 +100,7 @@ static int dm_test_fwu_mdata_read(struct unit_test_state *uts)
 	 * Trigger lib/fwu_updates/fwu.c fwu_boottime_checks()
 	 * to populate g_dev global pointer in that library.
 	 */
-	event_notify_null(EVT_MAIN_LOOP);
+	ut_assertok(event_notify_null(EVT_MAIN_LOOP));
 
 	ut_assertok(uclass_first_device_err(UCLASS_FWU_MDATA, &dev));
 	ut_assertok(fwu_init());
@@ -127,7 +127,7 @@ static int dm_test_fwu_mdata_write(struct unit_test_state *uts)
 	 * Trigger lib/fwu_updates/fwu.c fwu_boottime_checks()
 	 * to populate g_dev global pointer in that library.
 	 */
-	event_notify_null(EVT_MAIN_LOOP);
+	ut_assertok(event_notify_null(EVT_MAIN_LOOP));
 
 	ut_assertok(uclass_first_device_err(UCLASS_FWU_MDATA, &dev));
 
Signature · 1 line
2.43.0
Reviewed-by on v2 1/3
#2 lore raw

Re: [PATCH v2 1/3] event: Check return value from event_notify_null()

Quoted 6 lines · Michal Simek

event_notify_null() returns int but its return value is not checked in run_main_loop() and in fwu_mdata tests. Add proper error checking to all unchecked call sites.

Quoted 8 lines

---

Changes in v2:
- new patch in series

 common/board_r.c    | 6 +++++-
 test/dm/fwu_mdata.c | 4 ++--
 2 files changed, 7 insertions(+), 3 deletions(-)
Quoted 47 lines
common/board_r.c +5 -1
index 76f9fc090fbe..00f29385d80a 100644
@@ -579,11 +579,15 @@ static int dm_announce(void)
 
 static int run_main_loop(void)
 {
+       int ret;
+
 #ifdef CONFIG_SANDBOX
        sandbox_main_loop_init();
 #endif
 
-       event_notify_null(EVT_MAIN_LOOP);
+       ret = event_notify_null(EVT_MAIN_LOOP);
+       if (ret)
+               return ret;
 
        /* main_loop() can return to retry autoboot, if so just run it again */
        for (;;)
test/dm/fwu_mdata.c +2 -3
index b7680632f95f..643b647af0a9 100644
@@ -100,7 +100,7 @@ static int dm_test_fwu_mdata_read(struct unit_test_state *uts)
         * Trigger lib/fwu_updates/fwu.c fwu_boottime_checks()
         * to populate g_dev global pointer in that library.
         */
-       event_notify_null(EVT_MAIN_LOOP);
+       ut_assertok(event_notify_null(EVT_MAIN_LOOP));
 
        ut_assertok(uclass_first_device_err(UCLASS_FWU_MDATA, &dev));
        ut_assertok(fwu_init());
@@ -127,7 +127,7 @@ static int dm_test_fwu_mdata_write(struct unit_test_state *uts)
         * Trigger lib/fwu_updates/fwu.c fwu_boottime_checks()
         * to populate g_dev global pointer in that library.
         */
-       event_notify_null(EVT_MAIN_LOOP);
+       ut_assertok(event_notify_null(EVT_MAIN_LOOP));
 
        ut_assertok(uclass_first_device_err(UCLASS_FWU_MDATA, &dev));
 
--

2.43.0

#3 lore raw

PATCH v2 2/3 event: Introduce EVT_POST_PREBOOT event

In reply to Michal Simek

Add a new EVT_POST_PREBOOT event type which is fired in main_loop() after the preboot command has been executed.

---

Changes in v2:
- check return value

 common/main.c   | 4 ++++
 include/event.h | 9 +++++++++
 2 files changed, 13 insertions(+)
common/main.c +4 -0
index b0b6e74f5d3d..4b4504557f6e 100644
@@ -19,6 +19,7 @@
 #include <net.h>
 #include <version_string.h>
 #include <efi_loader.h>
+#include <event.h>
 
 static void run_preboot_environment_command(void)
 {
@@ -53,6 +54,9 @@ void main_loop(void)
 	if (IS_ENABLED(CONFIG_USE_PREBOOT))
 		run_preboot_environment_command();
 
+	if (event_notify_null(EVT_POST_PREBOOT))
+		return;
+
 	if (IS_ENABLED(CONFIG_UPDATE_TFTP))
 		update_tftp(0UL, NULL, NULL);
 
include/event.h +9 -0
index 1d267f1d1054..3ce5f992b04e 100644
@@ -153,6 +153,15 @@ enum event_t {
 	 */
 	EVT_MAIN_LOOP,
 
+	/**
+	 * @EVT_POST_PREBOOT:
+	 * This event is triggered in main_loop() after the preboot command
+	 * has run, so that devices initialised by preboot (e.g. USB, UFS)
+	 * are available to event handlers. Its parameter is NULL.
+	 * A non-zero return value causes the boot to fail.
+	 */
+	EVT_POST_PREBOOT,
+
 	/**
 	 * @EVT_OF_LIVE_BUILT:
 	 * This event is triggered immediately after the live device tree has been
Signature · 1 line
2.43.0
Reviewed-by on v2 2/3
#4 lore raw

Re: [PATCH v2 2/3] event: Introduce EVT_POST_PREBOOT event

Quoted 5 lines · Michal Simek

Add a new EVT_POST_PREBOOT event type which is fired in main_loop() after the preboot command has been executed.

Quoted 8 lines

---

Changes in v2:
- check return value

 common/main.c   | 4 ++++
 include/event.h | 9 +++++++++
 2 files changed, 13 insertions(+)
Quoted 46 lines
common/main.c +4 -0
index b0b6e74f5d3d..4b4504557f6e 100644
@@ -19,6 +19,7 @@
 #include <net.h>
 #include <version_string.h>
 #include <efi_loader.h>
+#include <event.h>
 
 static void run_preboot_environment_command(void)
 {
@@ -53,6 +54,9 @@ void main_loop(void)
        if (IS_ENABLED(CONFIG_USE_PREBOOT))
                run_preboot_environment_command();
 
+       if (event_notify_null(EVT_POST_PREBOOT))
+               return;
+
        if (IS_ENABLED(CONFIG_UPDATE_TFTP))
                update_tftp(0UL, NULL, NULL);
 
include/event.h +9 -1
index 1d267f1d1054..3ce5f992b04e 100644
@@ -153,6 +153,15 @@ enum event_t {
         */
        EVT_MAIN_LOOP,
 
+       /**
+        * @EVT_POST_PREBOOT:
+        * This event is triggered in main_loop() after the preboot command
+        * has run, so that devices initialised by preboot (e.g. USB, UFS)
+        * are available to event handlers. Its parameter is NULL.
+        * A non-zero return value causes the boot to fail.
+        */
+       EVT_POST_PREBOOT,
+
        /**
         * @EVT_OF_LIVE_BUILT:
         * This event is triggered immediately after the live device tree has been
--

2.43.0

#5 lore raw

PATCH v2 3/3 fwu: Move boottime checks to EVT_POST_PREBOOT

In reply to Michal Simek

Switch fwu_boottime_checks() from EVT_MAIN_LOOP to EVT_POST_PREBOOT because there is no reason to call FWU so early. FWU triggers EFI stack initialization before all devices are visible which prevents the EFI stack from scanning these devices and adding them to EFI variables.

---

(no changes since v1)

 lib/fwu_updates/fwu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/fwu_updates/fwu.c +1 -1
index 37c613014d18..e82600a29a40 100644
@@ -796,4 +796,4 @@ static int fwu_boottime_checks(void)
 
 	return 0;
 }
-EVENT_SPY_SIMPLE(EVT_MAIN_LOOP, fwu_boottime_checks);
+EVENT_SPY_SIMPLE(EVT_POST_PREBOOT, fwu_boottime_checks);
Signature · 1 line
2.43.0
#6 lore raw
In reply to Michal Simek

Hi Michal,

Quoted 6 lines · Michal Simek

Hi,

the whole series is trying to address issues we have on platforms where usb and ufs init are not done before efi initialization. This is happenin on platforms where A/B update is enabled. FWU code is called early and do EFI initialization before usb and ufs started that's why they are not visible.

I guess I'm missing something here (but I'd very much appreciate an explanation!) but isn't it actually necessary to set up the EFI subsystem /before/ register block devices?

Since EFI sets up an event listener to react to new UCLASS_BLK devices and create EFI device handles to represent them, it doesn't create device handles for block devices that existed before the EFI subsystem is up.

I ran into this issue when I wrote this patch:
https://gitlab.com/LinaroLtd/qcom-eco/u-boot/-/commit/e3616009a441358bf96bd2885d77fd46df710a9a

Kind regards,

Quoted 23 lines

We are workarounding it by starting usb/ufs from board file but this should be more generic approach.

Thanks,
Michal

Changes in v2:
- new patch in series
- check return value

Michal Simek (3):
event: Check return value from event_notify_null()
event: Introduce EVT_POST_PREBOOT event
fwu: Move boottime checks to EVT_POST_PREBOOT

 common/board_r.c      | 6 +++++-
 common/main.c         | 4 ++++
 include/event.h       | 9 +++++++++
 lib/fwu_updates/fwu.c | 2 +-
 test/dm/fwu_mdata.c   | 4 ++--
 5 files changed, 21 insertions(+), 4 deletions(-)
Signature · 1 line
// Casey (she/her)
#7 lore raw
Quoted 10 lines · Casey Connolly

Hi Michal,

Quoted 6 lines · Michal Simek

Hi,

the whole series is trying to address issues we have on platforms where usb and ufs init are not done before efi initialization. This is happenin on platforms where A/B update is enabled. FWU code is called early and do EFI initialization before usb and ufs started that's why they are not visible.

Quoted 8 lines

I guess I'm missing something here (but I'd very much appreciate an explanation!) but isn't it actually necessary to set up the EFI subsystem /before/ register block devices?

Since EFI sets up an event listener to react to new UCLASS_BLK devices and create EFI device handles to represent them, it doesn't create device handles for block devices that existed before the EFI subsystem is up.

This is solving different problem. This is related to FWU code/update capsule itself.

There are two options for automatic capsule applying. The first one is in main_loop()

62 if (IS_ENABLED(CONFIG_EFI_CAPSULE_ON_DISK_EARLY)) {
63 /* efi_init_early() already called */
64 if (efi_init_obj_list() == EFI_SUCCESS)
65 efi_launch_capsules();
66 }

And second is in efi_init_obj_list()

385 /* Execute capsules after reboot */
386 if (IS_ENABLED(CONFIG_EFI_CAPSULE_ON_DISK) &&
387 !IS_ENABLED(CONFIG_EFI_CAPSULE_ON_DISK_EARLY))
388 ret = efi_launch_capsules();

And because everything passed on the first run when efi_init_obj_list() is called again it never calls efi_launch_capsules().

240 /* Initialize only once, but start every time if correctly initialized*/
241 if (efi_obj_list_initialized == OBJ_LIST_INITIALIZED)
242 return efi_start_obj_list();

It means when you plug things later capsule update is never automatically called again.

These 3 patches happened based on my discussion with Ilias where we agreed that there is actually no reason to call FWU (which calls efi_init_obj_list() first) that early. Calling it later postpone efi initialization which give us a room to do additional steps via preboot.

Quoted 2 lines

I ran into this issue when I wrote this patch:
https://gitlab.com/LinaroLtd/qcom-eco/u-boot/-/commit/e3616009a441358bf96bd2885d77fd46df710a9a

Then please send this patch out and your patch is likely solving different problem.

Thanks,
Michal

#8 lore raw
In reply to Michal Simek
Quoted 9 lines · Michal Simek

Hi,

the whole series is trying to address issues we have on platforms where usb and ufs init are not done before efi initialization. This is happenin on platforms where A/B update is enabled. FWU code is called early and do EFI initialization before usb and ufs started that's why they are not visible.

We are workarounding it by starting usb/ufs from board file but this should be more generic approach.

This doesn't build on current next:
https://source.denx.de/u-boot/u-boot/-/jobs/1416493#L1001
common/event.c:56:1: error: static assertion failed: "event type_name size"

   56 | _Static_assert(ARRAY_SIZE(type_name) == EVT_COUNT, "event type_name size");
      | ^~~~~~~~~~~~~~

make[2]: *** [scripts/Makefile.build:271: common/event.o] Error 1 make[1]: *** [Makefile:2201: common] Error 2
make[1]: *** Waiting for unfinished jobs....

Signature · 1 line
Tom
#9 lore raw
Quoted 12 lines · Tom Rini
Quoted 9 lines · Michal Simek

Hi,

the whole series is trying to address issues we have on platforms where usb and ufs init are not done before efi initialization. This is happenin on platforms where A/B update is enabled. FWU code is called early and do EFI initialization before usb and ufs started that's why they are not visible.

We are workarounding it by starting usb/ufs from board file but this should be more generic approach.

Quoted 9 lines

This doesn't build on current next:
https://source.denx.de/u-boot/u-boot/-/jobs/1416493#L1001
common/event.c:56:1: error: static assertion failed: "event type_name size"

    56 | _Static_assert(ARRAY_SIZE(type_name) == EVT_COUNT, "event type_name size");
       | ^~~~~~~~~~~~~~

make[2]: *** [scripts/Makefile.build:271: common/event.o] Error 1 make[1]: *** [Makefile:2201: common] Error 2
make[1]: *** Waiting for unfinished jobs....

I am able to replicate. Will send v3 with fixing this.

common/event.c +3 -0
index 8d7513eb10b6..936d5e36d691 100644
@@ -49,6 +49,9 @@ const char *const type_name[] = {
        /* main loop events */
        "main_loop",
 
+       /* post preboot events */
+       "post preboot",
+
        /* livetree has been built */
        "of_live_init",
 };

Thanks,
Michal

#10 lore raw
In reply to Tom Rini

Am 31. März 2026 16:28:06 MESZ schrieb Tom Rini <[email protected]>:

Quoted 11 lines · Michal Simek
Quoted 10 lines

Hi,

the whole series is trying to address issues we have on platforms where usb and ufs init are not done before efi initialization. This is happenin on platforms where A/B update is enabled. FWU code is called early and do EFI initialization before usb and ufs started that's why they are not visible.

We are workarounding it by starting usb/ufs from board file but this should be more generic approach.

If a block device is created after EFI initialization an event handler creates an EFI handle.

Do have an example for QEMU to reproduce the issue.

I am on vacation till April 11th. But maybe Ilias could look at it.

Best regards

Heinrich

Quoted 10 lines

This doesn't build on current next:
https://source.denx.de/u-boot/u-boot/-/jobs/1416493#L1001
common/event.c:56:1: error: static assertion failed: "event type_name size"

  56 | _Static_assert(ARRAY_SIZE(type_name) == EVT_COUNT, "event type_name size");
     | ^~~~~~~~~~~~~~

make[2]: *** [scripts/Makefile.build:271: common/event.o] Error 1 make[1]: *** [Makefile:2201: common] Error 2
make[1]: *** Waiting for unfinished jobs....

#11 lore raw
Quoted 10 lines · Heinrich Schuchardt

Am 31. März 2026 16:28:06 MESZ schrieb Tom Rini <[email protected]>:

Quoted 9 lines · Michal Simek
Quoted 8 lines

Hi,

the whole series is trying to address issues we have on platforms where usb and ufs init are not done before efi initialization. This is happenin on platforms where A/B update is enabled. FWU code is called early and do EFI initialization before usb and ufs started that's why they are not visible.

Quoted 4 lines
Quoted 2 lines
Quoted 2 lines

We are workarounding it by starting usb/ufs from board file but this should be more generic approach.

If a block device is created after EFI initialization an event handler creates an EFI handle.

But capsules are not automatically taken.

Quoted 1 line

Do have an example for QEMU to reproduce the issue.

Nope. Only on our HW.

Quoted 1 line

I am on vacation till April 11th. But maybe Ilias could look at it.

Enjoy.

M