linker script: Vector table regression due to change in definition of _vector_end

[ABOSTM]

Describe the bug
A series of tests are now failed on STMicroelectronics test bench on STM32F0 series (Cortex M0):

• tests/arch/arm/arm_interrupt/arch.interrupt.arm
• tests/arch/arm/arm_interrupt/arch.interrupt.extra_exception_info
• tests/arch/arm/arm_interrupt/arch.interrupt.no_optimizations
• tests/arch/arm/arm_irq_advanced_features/arch.arm.irq_advanced_features
• tests/arch/arm/arm_irq_vector_table/arch.arm.irq_vector_table
• tests/drivers/adc/adc_api/drivers.adc
• tests/drivers/counter/counter_basic_api/drivers.counter.basic_api
• tests/drivers/dma/chan_blen_transfer/drivers.dma.chan_blen_transfer
• tests/drivers/dma/loop_transfer/drivers.dma.loop_transfer
• tests/drivers/gpio/gpio_api_1pin/peripheral.gpio.1pin
• tests/kernel/interrupt/arch.interrupt
• tests/kernel/profiling/profiling_api/kernel.common.profiling
• tests/kernel/tickless/tickless_concept/kernel.tickless.concept
• tests/kernel/timer/timer_api/kernel.timer.tickless
• tests/subsys/pm/device_runtime_api/pm.device_runtime.api

Test/MCU is stuck when handling interrupts.

This is a regression introduced since merge of PR: #46769 "Try to fix IRQ vector table relocation"
and especially the move of line
KEEP(*(_IRQ_VECTOR_TABLE_SECTION_SYMS))
from arch/arm/core/aarch32/vector_table.ld to tests/arch/arm/arm_irq_vector_table/irq-vector-table.ld

See detailed analysis below.

To Reproduce
Steps to reproduce the behavior:

1. west build -p auto -b nucleo_f091rc tests/arch/arm/arm_interrupt/
2. west flash
3. See error

Logs and console output
*** Booting Zephyr OS build zephyr-v3.1.0-985-gd497cca845e8 ***

Running TESTSUITE arm_interrupt
===================================================================
START - test_arm_null_pointer_exception
Skipped
 PASS - test_arm_null_pointer_exception in 0.1 seconds
===================================================================
START - test_arm_interrupt
Available IRQ line: 30
E: ***** HARD FAULT *****

Environment (please complete the following information):

• OS: . Linux,
• Toolchain Zephyr SDK
• Commit SHA or Version used

[ABOSTM]

Detailed analysis

Before the PR #46769, arch/arc/core/vector_table.ld , defines

_vector_start = .;
KEEP(*(.exc_vector_table))
KEEP(*(".exc_vector_table.*"))

KEEP(*(_IRQ_VECTOR_TABLE_SECTION_SYMS))

KEEP(*(.vectors))

_vector_end = .;

Thus _vector_end includes exception vector table as well as IRQ vector table.
This is confirmed by zephyr.map file that show
0x080000bc _vector_end = .
(0x080000bc is effectively the end of the whole vector table)

On the contrary, after the PR #46769,

_vector_start = .;
KEEP(*(.exc_vector_table))
KEEP(*(".exc_vector_table.*"))
KEEP(*(.vectors))

_vector_end = .;

Thus _vector_end doesn't include IRQ table, but only exception table,
this is confirmed by zephyr.map file that show
0x08000040 _vector_end = .
(0x08000040 is effectively the end of the exception vector table)

In other words, it is like patch changes the definition of symbol _vector_end
from whole table end to exception table end only.
And because this symbol is used in specific STM32F0 linker script

zephyr/soc/arm/st_stm32/stm32f0/sram_vector_table.ld

Line 11 in 7a11d88

. += _vector_end - _vector_start;

this causes tests failures due to bad vector table.

@carlocaione
Is it intentional to change the definition of _vector_end ? (It may impact other code)
Shall we keep the new definition or revert to the old one ?

[carlocaione]

Is it intentional to change the definition of _vector_end ? (It may impact other code)

Not really. Why CI didn't catch this?

Shall we keep the new definition or revert to the old one ?

I don't want to revert that PR. Let me see if I can fix this in a proper way.

[carlocaione]

@ABOSTM can you try this patch since I don't have the hardware?

diff --git a/include/zephyr/linker/irq-vector-table-section.ld b/include/zephyr/linker/irq-vector-table-section.ld
index 6a07abc34e..a1947ef74d 100644
--- a/include/zephyr/linker/irq-vector-table-section.ld
+++ b/include/zephyr/linker/irq-vector-table-section.ld
@@ -2,3 +2,6 @@
 
 . = ALIGN(CONFIG_ARCH_IRQ_VECTOR_TABLE_ALIGN);
 KEEP(*(_IRQ_VECTOR_TABLE_SECTION_SYMS))
+
+/* On ARM this symbol must be after the IRQ vector table */
+_vector_end = .;

[ABOSTM]

@carlocaione thanks for your quick response.
Yes, 1st patch works, moving _vector_end from vector_table.ld to irq-vector-table-section.ld

About your 2nd post, I was not sure you request to test only the addition of _vector_end in irq-vector-table-section.ld
In doubts I tests it and it is working too.

[ABOSTM]

@carlocaione ,
I reopen this issue because despite the fix proposed in #47309
there is still one test failed: tests/arch/arm/arm_irq_vector_table
(all other tests are passed).
I am confused, I should have check all tests instead of only one 😕

With a passed tests, let's say tests/arch/arm/arm_interrupt , the map file shows:
0x080000bc _vector_end = .
fine, it looks like it uses definition from irq-vector-table-section.ld

But with failed test : tests/arch/arm/arm_irq_vector_table the map file shows:
0x08000040 _vector_end = .
This is not good, it looks like it uses definition from vector_table.ld

I tried to remove _vector_end from vector_table.ld and keep only the one from irq-vector-table-section.ld
but in this case, there is a linker error _vector_end is undefined, which confirms that in case of tests/arch/arm/arm_irq_vector_table the definition used is the one from vector_table.ld

I noticed that tests/arch/arm/arm_irq_vector_table has its own linker file
https://github.com/zephyrproject-rtos/zephyr/blob/main/tests/arch/arm/arm_irq_vector_table/irq-vector-table.ld
shall we add _vector_end in this file ?
But it doesn't work, map file show 0x0800004c _vector_end = .

[carlocaione]

there is still one test failed: tests/arch/arm/arm_irq_vector_table

That test is kind special in the sense that it is defining by itself the IRQ vector table.

With a passed tests, let's say tests/arch/arm/arm_interrupt , the map file shows:
0x080000bc _vector_end = .
fine, it looks like it uses definition from irq-vector-table-section.ld

But with failed test : tests/arch/arm/arm_irq_vector_table the map file shows:
0x08000040 _vector_end = .
This is not good, it looks like it uses definition from vector_table.ld

That's fine. _vector_end must not always have the same value, this is not a problem at all.

I noticed that tests/arch/arm/arm_irq_vector_table has its own linker file
https://github.com/zephyrproject-rtos/zephyr/blob/main/tests/arch/arm/arm_irq_vector_table/irq-vector-table.ld
shall we add _vector_end in this file ?

Yes, the correct fix here is to add _vector_end = .; at the end of the tests/arch/arm/arm_irq_vector_table/irq-vector-table.ld file.

But it doesn't work, map file show 0x0800004c _vector_end = .

What do you mean it doesn't work? 0x0800004c is the correct address for this test.

[ABOSTM]

What do you mean it doesn't work? 0x0800004c is the correct address for this test.

I was expecting _vector_end to be always the size of the hardware vectors,
but ok, I understand that for tests purpose this is not the case for this specific test.
Then test is passed with the new fix ! !
I will submit a PR. Thanks