Skip to contentSkip to conversation

buildman: Fetch 12.2.0 toolchains by default

PATCH In review Tom Rini 4 participants 20 messages Nov 22, 2022 – Dec 7, 2022 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore

Tom Rini

author
#0 lore raw

PATCH 1/6 buildman: Fetch 12.2.0 toolchains by default

Update the toolchain list to be first 12.2.0 and second 11.1.0 and that's it.

---
 tools/buildman/toolchain.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/buildman/toolchain.py +1 -1
index fea40ba2151a..38b0dea8c709 100644
@@ -498,7 +498,7 @@ class Toolchains:
         if arch == 'aarch64':
             arch = 'arm64'
         base = 'https://www.kernel.org/pub/tools/crosstool/files/bin'
-        versions = ['11.1.0', '9.2.0', '7.3.0', '6.4.0', '4.9.4']
+        versions = ['12.2.0', '11.1.0']
         links = []
         for version in versions:
             url = '%s/%s/%s/' % (base, arch, version)
Signature · 1 line
2.25.1
author
#1 lore raw

PATCH 2/6 Makefile: Enforce DWARF4 output

At this point in time, using DWARF-5 format isn't easy to do by default with all toolchains that we support. And relying on the implicit default can lead to mixing 4 and 5 and then the debug info not being useful to tools. For now, enforce using DWARF-4 only.

---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Makefile +2 -2
index 47dbdcc6ae3d..eb7a8b90230e 100644
@@ -761,10 +761,10 @@ KBUILD_CFLAGS += $(call cc-disable-warning, maybe-uninitialized)
 # change __FILE__ to the relative path from the srctree
 KBUILD_CFLAGS	+= $(call cc-option,-fmacro-prefix-map=$(srctree)/=)
 
-KBUILD_CFLAGS	+= -g
+KBUILD_CFLAGS	+= -gdwarf-4
 # $(KBUILD_AFLAGS) sets -g, which causes gcc to pass a suitable -g<format>
 # option to the assembler.
-KBUILD_AFLAGS	+= -g
+KBUILD_AFLAGS	+= -gdwarf-4
 
 # Report stack usage if supported
 # ARC tools based on GCC 7.1 has an issue with stack usage
Signature · 1 line
2.25.1
Reviewed-by on v1 2/6
#2 lore raw

Re: [PATCH 2/6] Makefile: Enforce DWARF4 output

Quoted 10 lines · Tom Rini

At this point in time, using DWARF-5 format isn't easy to do by default with all toolchains that we support. And relying on the implicit default can lead to mixing 4 and 5 and then the debug info not being useful to tools. For now, enforce using DWARF-4 only.

---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
author
#3 lore raw

PATCH 3/6 event: Re-add file paths to the tests

In reply to Tom Rini

Now that we are enforcing dwarf-4 to be used we will have the full file paths present.

---
 test/py/tests/test_event_dump.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
test/py/tests/test_event_dump.py +3 -3
index 1a46ca30f42e..da196df4c3e6 100644
@@ -16,7 +16,7 @@ def test_event_dump(u_boot_console):
     out = util.run_and_log(cons, ['scripts/event_dump.py', sandbox])
     expect = '''.*Event type            Id                              Source location
 --------------------  ------------------------------  ------------------------------
-EVT_FT_FIXUP          bootmeth_vbe_ft_fixup           .*vbe_request.c:.*
-EVT_FT_FIXUP          bootmeth_vbe_simple_ft_fixup    .*vbe_simple_os.c:.*
-EVT_MISC_INIT_F       sandbox_misc_init_f             .*start.c:'''
+EVT_FT_FIXUP          bootmeth_vbe_ft_fixup           .*boot/vbe_request.c:.*
+EVT_FT_FIXUP          bootmeth_vbe_simple_ft_fixup    .*boot/vbe_simple_os.c:.*
+EVT_MISC_INIT_F       sandbox_misc_init_f             .*arch/sandbox/cpu/start.c:'''
     assert re.match(expect, out, re.MULTILINE) is not None
Signature · 1 line
2.25.1
author
#5 lore raw

PATCH 4/6 arm: Use the WEAK assembly entry point consistently

In reply to Tom Rini

It is a bad idea, and more modern toolchains will fail, if you declare an assembly function to be global and then weak, instead of declaring it weak to start with. Update assorted assembly files to use the WEAK macro directly.

---
 arch/arm/cpu/armv7/nonsec_virt.S |  3 +-
 arch/arm/cpu/armv7/psci.S        | 77 +++++++++++---------------------
 arch/arm/cpu/armv7/start.S       |  6 +--
 arch/arm/cpu/armv8/psci.S        | 12 ++---
 arch/arm/lib/relocate.S          |  3 +-
 5 files changed, 34 insertions(+), 67 deletions(-)
arch/arm/cpu/armv7/nonsec_virt.S +1 -2
index 39aeeb423f08..5ee6577c2c3c 100644
@@ -207,7 +207,7 @@ ENDPROC(_nonsec_init)
 
 #ifdef CONFIG_SMP_PEN_ADDR
 /* void __weak smp_waitloop(unsigned previous_address); */
-ENTRY(smp_waitloop)
+WEAK(smp_waitloop)
 	wfi
 	ldr	r1, =CONFIG_SMP_PEN_ADDR	@ load start address
 	ldr	r1, [r1]
@@ -219,7 +219,6 @@ ENTRY(smp_waitloop)
 	mov	r0, r1
 	b	_do_nonsec_entry
 ENDPROC(smp_waitloop)
-.weak smp_waitloop
 #endif
 
 	.popsection
arch/arm/cpu/armv7/psci.S +26 -51

134 lines not on this page · raw message

arch/arm/cpu/armv7/start.S +2 -4
index 4f6327fe3ab7..7d7aac021e22 100644
@@ -151,16 +151,14 @@ ENDPROC(c_runtime_cpu_setup)
  * Don't save anything to stack even if compiled with -O0
  *
  *************************************************************************/
-ENTRY(save_boot_params)
+WEAK(save_boot_params)
 	b	save_boot_params_ret		@ back to my caller
 ENDPROC(save_boot_params)
-	.weak	save_boot_params
 
 #ifdef CONFIG_ARMV7_LPAE
-ENTRY(switch_to_hypervisor)
+WEAK(switch_to_hypervisor)
 	b	switch_to_hypervisor_ret
 ENDPROC(switch_to_hypervisor)
-	.weak	switch_to_hypervisor
 #endif
 
 /*************************************************************************
arch/arm/cpu/armv8/psci.S +4 -8
index 7ffc8dbadbe0..6aece1198712 100644
@@ -12,11 +12,10 @@
 
 /* Default PSCI function, return -1, Not Implemented */
 #define PSCI_DEFAULT(__fn) \
-	ENTRY(__fn); \
+	WEAK(__fn); \
 	mov	w0, #ARM_PSCI_RET_NI; \
 	ret; \
 	ENDPROC(__fn); \
-	.weak __fn
 
 /* PSCI function and ID table definition*/
 #define PSCI_TABLE(__id, __fn) \
@@ -207,7 +206,7 @@ handle_smc64:
  * used for the return value, while in this PSCI environment, X0 usually holds
  * the SMC function identifier, so X0 should be saved by caller function.
  */
-ENTRY(psci_get_cpu_id)
+WEAK(psci_get_cpu_id)
 #ifdef CONFIG_ARMV8_PSCI_CPUS_PER_CLUSTER
 	mrs	x9, MPIDR_EL1
 	ubfx	x9, x9, #8, #8
@@ -221,7 +220,6 @@ ENTRY(psci_get_cpu_id)
 	add	x0, x10, x9
 	ret
 ENDPROC(psci_get_cpu_id)
-.weak psci_get_cpu_id
 
 /* CPU ID input in x0, stack top output in x0*/
 LENTRY(psci_get_cpu_stack_top)
@@ -261,10 +259,9 @@ handle_sync:
  * Override this function if custom error handling is
  * needed for asynchronous aborts
  */
-ENTRY(plat_error_handler)
+WEAK(plat_error_handler)
 	ret
 ENDPROC(plat_error_handler)
-.weak plat_error_handler
 
 handle_error:
 	bl	psci_get_cpu_id
@@ -323,9 +320,8 @@ ENTRY(psci_setup_vectors)
 	ret
 ENDPROC(psci_setup_vectors)
 
-ENTRY(psci_arch_init)
+WEAK(psci_arch_init)
 	ret
 ENDPROC(psci_arch_init)
-.weak psci_arch_init
 
 .popsection
arch/arm/lib/relocate.S +1 -2
index dd6f2e3bd5e0..345e282e3e65 100644
@@ -23,9 +23,8 @@
  */
 
 	.section	.text.relocate_vectors,"ax",%progbits
-	.weak		relocate_vectors
 
-ENTRY(relocate_vectors)
+WEAK(relocate_vectors)
 
 #ifdef CONFIG_CPU_V7M
 	/*
Signature · 1 line
2.25.1
Reviewed-by on v1 4/6
#6 lore raw

Re: [PATCH 4/6] arm: Use the WEAK assembly entry point consistently

Quoted 6 lines · Tom Rini

It is a bad idea, and more modern toolchains will fail, if you declare an assembly function to be global and then weak, instead of declaring it weak to start with. Update assorted assembly files to use the WEAK macro directly.

Quoted 9 lines
---
 arch/arm/cpu/armv7/nonsec_virt.S |  3 +-
 arch/arm/cpu/armv7/psci.S        | 77 +++++++++++---------------------
 arch/arm/cpu/armv7/start.S       |  6 +--
 arch/arm/cpu/armv8/psci.S        | 12 ++---
 arch/arm/lib/relocate.S          |  3 +-
 5 files changed, 34 insertions(+), 67 deletions(-)
Reviewed-by on v1 4/6
#7 lore raw

Re: [PATCH 4/6] arm: Use the WEAK assembly entry point consistently

In reply to Tom Rini
Quoted 6 lines · On Tuesday 22 November 2022 12:31:56 Tom Rini wrote:

It is a bad idea, and more modern toolchains will fail, if you declare an assembly function to be global and then weak, instead of declaring it weak to start with. Update assorted assembly files to use the WEAK macro directly.

During debugging of Nokia N900 code I was looking at this and I originally thought that this redefinition is the issue why N900 u-boot did not work... (but as we now know, the n900 issue was somewhere else).

So I agree with this change, feel free to add my:

... but even after this change, linked u-boot.bin binary is not-so-correct. It works but has an issue: In final u-boot.bin binary there is both weak and non-weak version of every weak function. You can verify it for example by looking at "save_boot_params" code (really code, not just symbol) in u-boot ELF binary.

The reason for this is that linker (even LTO enabled) cannot eliminate code for weak version of function because it does not know how to "drop" it from binary/assembly code. So linker just set that non-weak version of function is active and let non-weak version present in binary as probably dead code.

This is affected only by assembly files, not by C files, because gcc is called with -ffunction-sections -fdata-sections flags which cause that every (weak) function is in its separate section (so C function "int abc() { ... }" is put into the section ".text.abc" instead of ".text") and linker's flag --gc-sections (or LTO optimization) then drop all unreferenced sections.

I do not know how fix this issue in assembly files. But cannot be WEAK macro modified to change section to ".text.<entry_name>" to mimic C compiler behavior? Would this cause any issues?

Quoted 271 lines
---
 arch/arm/cpu/armv7/nonsec_virt.S |  3 +-
 arch/arm/cpu/armv7/psci.S        | 77 +++++++++++---------------------
 arch/arm/cpu/armv7/start.S       |  6 +--
 arch/arm/cpu/armv8/psci.S        | 12 ++---
 arch/arm/lib/relocate.S          |  3 +-
 5 files changed, 34 insertions(+), 67 deletions(-)
arch/arm/cpu/armv7/nonsec_virt.S +1 -2
index 39aeeb423f08..5ee6577c2c3c 100644
@@ -207,7 +207,7 @@ ENDPROC(_nonsec_init)
 
 #ifdef CONFIG_SMP_PEN_ADDR
 /* void __weak smp_waitloop(unsigned previous_address); */
-ENTRY(smp_waitloop)
+WEAK(smp_waitloop)
 	wfi
 	ldr	r1, =CONFIG_SMP_PEN_ADDR	@ load start address
 	ldr	r1, [r1]
@@ -219,7 +219,6 @@ ENTRY(smp_waitloop)
 	mov	r0, r1
 	b	_do_nonsec_entry
 ENDPROC(smp_waitloop)
-.weak smp_waitloop
 #endif
 
 	.popsection
arch/arm/cpu/armv7/psci.S +26 -51
index 983cd9044295..6c066e50d91e 100644
@@ -36,34 +36,32 @@ _psci_vectors:
 	b	default_psci_vector	@ irq
 	b	psci_fiq_enter		@ fiq
 
-ENTRY(psci_fiq_enter)
+WEAK(psci_fiq_enter)
 	movs	pc, lr
 ENDPROC(psci_fiq_enter)
-.weak psci_fiq_enter
 
-ENTRY(default_psci_vector)
+WEAK(default_psci_vector)
 	movs	pc, lr
 ENDPROC(default_psci_vector)
-.weak default_psci_vector
-
-ENTRY(psci_version)
-ENTRY(psci_cpu_suspend)
-ENTRY(psci_cpu_off)
-ENTRY(psci_cpu_on)
-ENTRY(psci_affinity_info)
-ENTRY(psci_migrate)
-ENTRY(psci_migrate_info_type)
-ENTRY(psci_migrate_info_up_cpu)
-ENTRY(psci_system_off)
-ENTRY(psci_system_reset)
-ENTRY(psci_features)
-ENTRY(psci_cpu_freeze)
-ENTRY(psci_cpu_default_suspend)
-ENTRY(psci_node_hw_state)
-ENTRY(psci_system_suspend)
-ENTRY(psci_set_suspend_mode)
-ENTRY(psi_stat_residency)
-ENTRY(psci_stat_count)
+
+WEAK(psci_version)
+WEAK(psci_cpu_suspend)
+WEAK(psci_cpu_off)
+WEAK(psci_cpu_on)
+WEAK(psci_affinity_info)
+WEAK(psci_migrate)
+WEAK(psci_migrate_info_type)
+WEAK(psci_migrate_info_up_cpu)
+WEAK(psci_system_off)
+WEAK(psci_system_reset)
+WEAK(psci_features)
+WEAK(psci_cpu_freeze)
+WEAK(psci_cpu_default_suspend)
+WEAK(psci_node_hw_state)
+WEAK(psci_system_suspend)
+WEAK(psci_set_suspend_mode)
+WEAK(psi_stat_residency)
+WEAK(psci_stat_count)
 	mov	r0, #ARM_PSCI_RET_NI	@ Return -1 (Not Implemented)
 	mov	pc, lr
 ENDPROC(psci_stat_count)
@@ -84,24 +82,6 @@ ENDPROC(psci_cpu_on)
 ENDPROC(psci_cpu_off)
 ENDPROC(psci_cpu_suspend)
 ENDPROC(psci_version)
-.weak psci_version
-.weak psci_cpu_suspend
-.weak psci_cpu_off
-.weak psci_cpu_on
-.weak psci_affinity_info
-.weak psci_migrate
-.weak psci_migrate_info_type
-.weak psci_migrate_info_up_cpu
-.weak psci_system_off
-.weak psci_system_reset
-.weak psci_features
-.weak psci_cpu_freeze
-.weak psci_cpu_default_suspend
-.weak psci_node_hw_state
-.weak psci_system_suspend
-.weak psci_set_suspend_mode
-.weak psi_stat_residency
-.weak psci_stat_count
 
 _psci_table:
 	.word	ARM_PSCI_FN_CPU_SUSPEND
@@ -179,12 +159,11 @@ _smc_psci:
 	movs	pc, lr			@ Return to the kernel
 
 @ Requires dense and single-cluster CPU ID space
-ENTRY(psci_get_cpu_id)
+WEAK(psci_get_cpu_id)
 	mrc	p15, 0, r0, c0, c0, 5	/* read MPIDR */
 	and	r0, r0, #0xff		/* return CPU ID in cluster */
 	bx	lr
 ENDPROC(psci_get_cpu_id)
-.weak psci_get_cpu_id
 
 /* Imported from Linux kernel */
 ENTRY(psci_v7_flush_dcache_all)
@@ -236,7 +215,7 @@ finished:
 	bx	lr
 ENDPROC(psci_v7_flush_dcache_all)
 
-ENTRY(psci_disable_smp)
+WEAK(psci_disable_smp)
 	mrc	p15, 0, r0, c1, c0, 1		@ ACTLR
 	bic	r0, r0, #(1 << 6)		@ Clear SMP bit
 	mcr	p15, 0, r0, c1, c0, 1		@ ACTLR
@@ -244,16 +223,14 @@ ENTRY(psci_disable_smp)
 	dsb
 	bx	lr
 ENDPROC(psci_disable_smp)
-.weak psci_disable_smp
 
-ENTRY(psci_enable_smp)
+WEAK(psci_enable_smp)
 	mrc	p15, 0, r0, c1, c0, 1		@ ACTLR
 	orr	r0, r0, #(1 << 6)		@ Set SMP bit
 	mcr	p15, 0, r0, c1, c0, 1		@ ACTLR
 	isb
 	bx	lr
 ENDPROC(psci_enable_smp)
-.weak psci_enable_smp
 
 ENTRY(psci_cpu_off_common)
 	push	{lr}
@@ -316,15 +293,13 @@ ENTRY(psci_stack_setup)
 	bx	r6
 ENDPROC(psci_stack_setup)
 
-ENTRY(psci_arch_init)
+WEAK(psci_arch_init)
 	mov	pc, lr
 ENDPROC(psci_arch_init)
-.weak psci_arch_init
 
-ENTRY(psci_arch_cpu_entry)
+WEAK(psci_arch_cpu_entry)
 	mov	pc, lr
 ENDPROC(psci_arch_cpu_entry)
-.weak psci_arch_cpu_entry
 
 ENTRY(psci_cpu_entry)
 	bl	psci_enable_smp
arch/arm/cpu/armv7/start.S +2 -4
index 4f6327fe3ab7..7d7aac021e22 100644
@@ -151,16 +151,14 @@ ENDPROC(c_runtime_cpu_setup)
  * Don't save anything to stack even if compiled with -O0
  *
  *************************************************************************/
-ENTRY(save_boot_params)
+WEAK(save_boot_params)
 	b	save_boot_params_ret		@ back to my caller
 ENDPROC(save_boot_params)
-	.weak	save_boot_params
 
 #ifdef CONFIG_ARMV7_LPAE
-ENTRY(switch_to_hypervisor)
+WEAK(switch_to_hypervisor)
 	b	switch_to_hypervisor_ret
 ENDPROC(switch_to_hypervisor)
-	.weak	switch_to_hypervisor
 #endif
 
 /*************************************************************************
arch/arm/cpu/armv8/psci.S +4 -8
index 7ffc8dbadbe0..6aece1198712 100644
@@ -12,11 +12,10 @@
 
 /* Default PSCI function, return -1, Not Implemented */
 #define PSCI_DEFAULT(__fn) \
-	ENTRY(__fn); \
+	WEAK(__fn); \
 	mov	w0, #ARM_PSCI_RET_NI; \
 	ret; \
 	ENDPROC(__fn); \
-	.weak __fn
 
 /* PSCI function and ID table definition*/
 #define PSCI_TABLE(__id, __fn) \
@@ -207,7 +206,7 @@ handle_smc64:
  * used for the return value, while in this PSCI environment, X0 usually holds
  * the SMC function identifier, so X0 should be saved by caller function.
  */
-ENTRY(psci_get_cpu_id)
+WEAK(psci_get_cpu_id)
 #ifdef CONFIG_ARMV8_PSCI_CPUS_PER_CLUSTER
 	mrs	x9, MPIDR_EL1
 	ubfx	x9, x9, #8, #8
@@ -221,7 +220,6 @@ ENTRY(psci_get_cpu_id)
 	add	x0, x10, x9
 	ret
 ENDPROC(psci_get_cpu_id)
-.weak psci_get_cpu_id
 
 /* CPU ID input in x0, stack top output in x0*/
 LENTRY(psci_get_cpu_stack_top)
@@ -261,10 +259,9 @@ handle_sync:
  * Override this function if custom error handling is
  * needed for asynchronous aborts
  */
-ENTRY(plat_error_handler)
+WEAK(plat_error_handler)
 	ret
 ENDPROC(plat_error_handler)
-.weak plat_error_handler
 
 handle_error:
 	bl	psci_get_cpu_id
@@ -323,9 +320,8 @@ ENTRY(psci_setup_vectors)
 	ret
 ENDPROC(psci_setup_vectors)
 
-ENTRY(psci_arch_init)
+WEAK(psci_arch_init)
 	ret
 ENDPROC(psci_arch_init)
-.weak psci_arch_init
 
 .popsection
arch/arm/lib/relocate.S +1 -2
index dd6f2e3bd5e0..345e282e3e65 100644
@@ -23,9 +23,8 @@
  */
 
 	.section	.text.relocate_vectors,"ax",%progbits
-	.weak		relocate_vectors
 
-ENTRY(relocate_vectors)
+WEAK(relocate_vectors)
 
 #ifdef CONFIG_CPU_V7M
 	/*
Signature · 1 line
2.25.1
author
#8 lore raw

Re: [PATCH 4/6] arm: Use the WEAK assembly entry point consistently

Quoted 35 lines · Pali Rohár
Quoted 6 lines · On Tuesday 22 November 2022 12:31:56 Tom Rini wrote:

It is a bad idea, and more modern toolchains will fail, if you declare an assembly function to be global and then weak, instead of declaring it weak to start with. Update assorted assembly files to use the WEAK macro directly.

During debugging of Nokia N900 code I was looking at this and I originally thought that this redefinition is the issue why N900 u-boot did not work... (but as we now know, the n900 issue was somewhere else).

So I agree with this change, feel free to add my:

... but even after this change, linked u-boot.bin binary is not-so-correct. It works but has an issue: In final u-boot.bin binary there is both weak and non-weak version of every weak function. You can verify it for example by looking at "save_boot_params" code (really code, not just symbol) in u-boot ELF binary.

The reason for this is that linker (even LTO enabled) cannot eliminate code for weak version of function because it does not know how to "drop" it from binary/assembly code. So linker just set that non-weak version of function is active and let non-weak version present in binary as probably dead code.

This is affected only by assembly files, not by C files, because gcc is called with -ffunction-sections -fdata-sections flags which cause that every (weak) function is in its separate section (so C function "int abc() { ... }" is put into the section ".text.abc" instead of ".text") and linker's flag --gc-sections (or LTO optimization) then drop all unreferenced sections.

Quoted 3 lines

I do not know how fix this issue in assembly files. But cannot be WEAK macro modified to change section to ".text.<entry_name>" to mimic C compiler behavior? Would this cause any issues?

Yes, you're right about the cause, and potential solution. If you can come up with a way to get each assembly function put in to a separate .text.funcname section, that would be great and much appreciated. I think I tried this at one point a long long time ago and it did work, but I didn't have something clean, either. I think I was hoping that the linux kernel folks would solve it in time, but they decided the time/effort for --gc-sections wasn't worth it, in the end.

Signature · 1 line
Tom
#9 lore raw

Re: [PATCH 4/6] arm: Use the WEAK assembly entry point consistently

Quoted 40 lines · On Wednesday 23 November 2022 18:15:17 Tom Rini wrote:
Quoted 35 lines · Pali Rohár
Quoted 6 lines · On Tuesday 22 November 2022 12:31:56 Tom Rini wrote:

It is a bad idea, and more modern toolchains will fail, if you declare an assembly function to be global and then weak, instead of declaring it weak to start with. Update assorted assembly files to use the WEAK macro directly.

During debugging of Nokia N900 code I was looking at this and I originally thought that this redefinition is the issue why N900 u-boot did not work... (but as we now know, the n900 issue was somewhere else).

So I agree with this change, feel free to add my:

... but even after this change, linked u-boot.bin binary is not-so-correct. It works but has an issue: In final u-boot.bin binary there is both weak and non-weak version of every weak function. You can verify it for example by looking at "save_boot_params" code (really code, not just symbol) in u-boot ELF binary.

The reason for this is that linker (even LTO enabled) cannot eliminate code for weak version of function because it does not know how to "drop" it from binary/assembly code. So linker just set that non-weak version of function is active and let non-weak version present in binary as probably dead code.

This is affected only by assembly files, not by C files, because gcc is called with -ffunction-sections -fdata-sections flags which cause that every (weak) function is in its separate section (so C function "int abc() { ... }" is put into the section ".text.abc" instead of ".text") and linker's flag --gc-sections (or LTO optimization) then drop all unreferenced sections.

Quoted 3 lines

I do not know how fix this issue in assembly files. But cannot be WEAK macro modified to change section to ".text.<entry_name>" to mimic C compiler behavior? Would this cause any issues?

Quoted 7 lines

Yes, you're right about the cause, and potential solution. If you can come up with a way to get each assembly function put in to a separate .text.funcname section, that would be great and much appreciated. I think I tried this at one point a long long time ago and it did work, but I didn't have something clean, either. I think I was hoping that the linux kernel folks would solve it in time, but they decided the time/effort for --gc-sections wasn't worth it, in the end.

I quickly looked at this. If "as" is invoked with --sectname-subst flag then it is possible to use '.section %S.<func_name>' and '.previous' directives. See documentation where is example of that:
https://sourceware.org/binutils/docs/as/Section.html#ELF-Version

I experimented with adding into include/linux/linkage.h:

  #define WEAKSECT(name) \
    .section .text.name ASM_NL \
    WEAK(name)

  #define ENDPROCSECT(name) \
    ENDPROC(name) ASM_NL \
    .previous

And then defining in arch/arm/cpu/armv7/start.S:

WEAKSECT(save_boot_params)
b save_boot_params_ret
ENDPROCSECT(save_boot_params)

(Note that n900 has custom non-weak save_boot_params)

Then I run:

make CROSS_COMPILE=arm-linux-gnueabi- nokia_rx51_defconfig u-boot.bin KBUILD_LDFLAGS="--print-gc-sections"

And it showed me:

ld: removing unused section '.text.save_boot_params' in file 'arch/arm/cpu/armv7/start.o'

So seems that it is working.

For proper integration it would be needed to integrate --sectname-subst flag support and then replace all usage by new macros.

author
#10 lore raw

Re: [PATCH 4/6] arm: Use the WEAK assembly entry point consistently

Quoted 76 lines · Pali Rohár
Quoted 40 lines · On Wednesday 23 November 2022 18:15:17 Tom Rini wrote:
Quoted 35 lines · Pali Rohár
Quoted 6 lines · On Tuesday 22 November 2022 12:31:56 Tom Rini wrote:

It is a bad idea, and more modern toolchains will fail, if you declare an assembly function to be global and then weak, instead of declaring it weak to start with. Update assorted assembly files to use the WEAK macro directly.

During debugging of Nokia N900 code I was looking at this and I originally thought that this redefinition is the issue why N900 u-boot did not work... (but as we now know, the n900 issue was somewhere else).

So I agree with this change, feel free to add my:

... but even after this change, linked u-boot.bin binary is not-so-correct. It works but has an issue: In final u-boot.bin binary there is both weak and non-weak version of every weak function. You can verify it for example by looking at "save_boot_params" code (really code, not just symbol) in u-boot ELF binary.

The reason for this is that linker (even LTO enabled) cannot eliminate code for weak version of function because it does not know how to "drop" it from binary/assembly code. So linker just set that non-weak version of function is active and let non-weak version present in binary as probably dead code.

This is affected only by assembly files, not by C files, because gcc is called with -ffunction-sections -fdata-sections flags which cause that every (weak) function is in its separate section (so C function "int abc() { ... }" is put into the section ".text.abc" instead of ".text") and linker's flag --gc-sections (or LTO optimization) then drop all unreferenced sections.

Quoted 3 lines

I do not know how fix this issue in assembly files. But cannot be WEAK macro modified to change section to ".text.<entry_name>" to mimic C compiler behavior? Would this cause any issues?

Quoted 7 lines

Yes, you're right about the cause, and potential solution. If you can come up with a way to get each assembly function put in to a separate .text.funcname section, that would be great and much appreciated. I think I tried this at one point a long long time ago and it did work, but I didn't have something clean, either. I think I was hoping that the linux kernel folks would solve it in time, but they decided the time/effort for --gc-sections wasn't worth it, in the end.

I quickly looked at this. If "as" is invoked with --sectname-subst flag then it is possible to use '.section %S.<func_name>' and '.previous' directives. See documentation where is example of that:
https://sourceware.org/binutils/docs/as/Section.html#ELF-Version

I experimented with adding into include/linux/linkage.h:

  #define WEAKSECT(name) \
    .section .text.name ASM_NL \
    WEAK(name)

  #define ENDPROCSECT(name) \
    ENDPROC(name) ASM_NL \
    .previous

And then defining in arch/arm/cpu/armv7/start.S:

WEAKSECT(save_boot_params)
b save_boot_params_ret
ENDPROCSECT(save_boot_params)

(Note that n900 has custom non-weak save_boot_params)

Then I run:

make CROSS_COMPILE=arm-linux-gnueabi- nokia_rx51_defconfig u-boot.bin KBUILD_LDFLAGS="--print-gc-sections"

Quoted 8 lines

And it showed me:

ld: removing unused section '.text.save_boot_params' in file 'arch/arm/cpu/armv7/start.o'

So seems that it is working.

For proper integration it would be needed to integrate --sectname-subst flag support and then replace all usage by new macros.

That's all good to know, thanks for digging more. It would be good to know if a quick and dirty experimental patch showed enough size savings to be worth a more full pursuit or if we really don't have many / any unused assembly functions or what we do have unused could be more easily handled with an ifdef or refactoring into multiple files.

Signature · 1 line
Tom
author
#11 lore raw

PATCH 5/6 pogo_v4: Disable LTO

In reply to Tom Rini

With gcc-12.2 we now get:
lib/zlib/inflate.c:360: undefined reference to `__gnu_thumb1_case_si' when building this platform. This seems like some odd problem with LTO and Thumb, but since the platform continues to link, I assume it's within size constraints, so lets just disable LTO for now.

---
 configs/pogo_v4_defconfig | 1 -
 1 file changed, 1 deletion(-)
configs/pogo_v4_defconfig +0 -1
index 3b08cb7b1d14..c4b7f7c31471 100644
@@ -17,7 +17,6 @@ CONFIG_DEFAULT_DEVICE_TREE="kirkwood-pogoplug-series-4"
 CONFIG_SYS_PROMPT="Pogo_V4> "
 CONFIG_IDENT_STRING="\nPogoplug V4"
 CONFIG_SYS_LOAD_ADDR=0x800000
-CONFIG_LTO=y
 CONFIG_DISTRO_DEFAULTS=y
 CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y
 CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0xc8012000
Signature · 1 line
2.25.1

A message that is not in the archive (sent off-list or not yet received): CAJaLiFwY3eyeca4eOEW0XRF0NdaLvFAgi_BLTD-v7U=vjTemKg@mail.gmail.com

#13 lore raw

Re: [PATCH 5/6] pogo_v4: Disable LTO

Resend to include the mailing list!

Tony

Quoted 40 lines · Tony Dinh

Hi Tom,

Quoted 6 lines · Tom Rini

With gcc-12.2 we now get:
lib/zlib/inflate.c:360: undefined reference to `__gnu_thumb1_case_si' when building this platform. This seems like some odd problem with LTO and Thumb, but since the platform continues to link, I assume it's within size constraints, so lets just disable LTO for now.

We don't have a hard size constraint for this board. Thumb and LTO were to help keep the image 512K or less (by convention, for some downstream scripts).

Thanks,

Quoted 22 lines
---
 configs/pogo_v4_defconfig | 1 -
 1 file changed, 1 deletion(-)
configs/pogo_v4_defconfig +0 -2
index 3b08cb7b1d14..c4b7f7c31471 100644
@@ -17,7 +17,6 @@ CONFIG_DEFAULT_DEVICE_TREE="kirkwood-pogoplug-series-4"
 CONFIG_SYS_PROMPT="Pogo_V4> "
 CONFIG_IDENT_STRING="\nPogoplug V4"
 CONFIG_SYS_LOAD_ADDR=0x800000
-CONFIG_LTO=y
 CONFIG_DISTRO_DEFAULTS=y
 CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y
 CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0xc8012000
--

2.25.1

author
#14 lore raw

Re: [PATCH 5/6] pogo_v4: Disable LTO

Quoted 19 lines · Tony Dinh

Resend to include the mailing list!

Tony

Quoted 14 lines · Tony Dinh

Hi Tom,

Quoted 6 lines · Tom Rini

With gcc-12.2 we now get:
lib/zlib/inflate.c:360: undefined reference to `__gnu_thumb1_case_si' when building this platform. This seems like some odd problem with LTO and Thumb, but since the platform continues to link, I assume it's within size constraints, so lets just disable LTO for now.

We don't have a hard size constraint for this board. Thumb and LTO were to help keep the image 512K or less (by convention, for some downstream scripts).

Ah, well, uh-oh. Then we needed to add:
CONFIG_HAS_BOARD_SIZE_LIMIT=y
CONFIG_BOARD_SIZE_LIMIT=524288
to the config. Which is fine now, but without both LTO and thumb, we now are too big. So this means digging harder at that error message. Which in turn shows that 806f86bd826e ("arm: lib: Import Thumb1 functions") is where we imported the __gnu_thumb1_case_* functions we have now, and need to do similar work to import (and then do the follow-up improvements to) __gnu_thumb1_case_si. Can you please take a look at that? FWIW, the gcc-12.2.0 toolchain I'm using is just
https://mirrors.edge.kernel.org/pub/tools/crosstool/files/bin/x86_64/12.2.0/x86_64-gcc-12.2.0-nolibc-arm-linux-gnueabi.tar.xz

Signature · 1 line
Tom
#15 lore raw

Re: [PATCH 5/6] pogo_v4: Disable LTO

Hi Tom,

Quoted 22 lines · Tom Rini
Quoted 19 lines · Tony Dinh

Resend to include the mailing list!

Tony

Quoted 14 lines · Tony Dinh

Hi Tom,

Quoted 6 lines · Tom Rini

With gcc-12.2 we now get:
lib/zlib/inflate.c:360: undefined reference to `__gnu_thumb1_case_si' when building this platform. This seems like some odd problem with LTO and Thumb, but since the platform continues to link, I assume it's within size constraints, so lets just disable LTO for now.

We don't have a hard size constraint for this board. Thumb and LTO were to help keep the image 512K or less (by convention, for some downstream scripts).

Quoted 11 lines

Ah, well, uh-oh. Then we needed to add:
CONFIG_HAS_BOARD_SIZE_LIMIT=y
CONFIG_BOARD_SIZE_LIMIT=524288
to the config. Which is fine now, but without both LTO and thumb, we now are too big. So this means digging harder at that error message. Which in turn shows that 806f86bd826e ("arm: lib: Import Thumb1 functions") is where we imported the __gnu_thumb1_case_* functions we have now, and need to do similar work to import (and then do the follow-up improvements to) __gnu_thumb1_case_si. Can you please take a look at that? FWIW, the gcc-12.2.0 toolchain I'm using is just
https://mirrors.edge.kernel.org/pub/tools/crosstool/files/bin/x86_64/12.2.0/x86_64-gcc-12.2.0-nolibc-arm-linux-gnueabi.tar.xz

I think I was not clear in my previous response. We did not implement a hard size constraint, but only tried to keep it small at 512K. If it grows larger, it's OK. Others and I just need to modify the flashing script, which is not a big deal. So removing LTO and Thumb is fine to get through this gcc build problem. Nothing will break because it is still small enough at 670K (built using gcc 11).

I'm not well versed in troubleshooting this type of compiler problem. I could try, but I think someone like Simon would be able to solve it quicker. Please let me know what you think.

All the best,
Tony

Quoted 3 lines

--
Tom

Reviewed-by on v1 5/6
#16 lore raw

Re: [PATCH 5/6] pogo_v4: Disable LTO

In reply to Tom Rini
Quoted 7 lines · Tom Rini

With gcc-12.2 we now get:
lib/zlib/inflate.c:360: undefined reference to `__gnu_thumb1_case_si' when building this platform. This seems like some odd problem with LTO and Thumb, but since the platform continues to link, I assume it's within size constraints, so lets just disable LTO for now.

Quoted 5 lines
---
 configs/pogo_v4_defconfig | 1 -
 1 file changed, 1 deletion(-)
author
#17 lore raw

PATCH 6/6 CI: Update to gcc-12.2

In reply to Tom Rini

- Update to gcc-12.2, and cherry-pick a fix in grub for risc-v

---
 .azure-pipelines.yml    |  8 ++---
 .gitlab-ci.yml          |  6 ++--
 tools/docker/Dockerfile | 65 +++++++++++++++++++++--------------------
 3 files changed, 41 insertions(+), 38 deletions(-)
.azure-pipelines.yml +4 -4
index 43ebcb6a346f..c37a876f08f5 100644
@@ -2,7 +2,7 @@ variables:
   windows_vm: windows-2019
   ubuntu_vm: ubuntu-22.04
   macos_vm: macOS-12
-  ci_runner_image: trini/u-boot-gitlab-ci-runner:jammy-20221003-17Oct2022
+  ci_runner_image: trini/u-boot-gitlab-ci-runner:jammy-20221101-22Nov2022
   # Add '-u 0' options for Azure pipelines, otherwise we get "permission
   # denied" error when it tries to "useradd -m -u 1001 vsts_azpcontainer",
   # since our $(ci_runner_image) user is not root.
@@ -204,7 +204,7 @@ stages:
       options: $(container_option)
     steps:
       - script: |
-          export PATH=/opt/gcc-11.1.0-nolibc/arm-linux-gnueabi/bin:$PATH
+          export PATH=/opt/gcc-12.2.0-nolibc/arm-linux-gnueabi/bin:$PATH
           test/nokia_rx51_test.sh
 
   - job: pylint
@@ -242,7 +242,7 @@ stages:
           TEST_PY_BD: "sandbox"
         sandbox_clang:
           TEST_PY_BD: "sandbox"
-          OVERRIDE: "-O clang-13"
+          OVERRIDE: "-O clang-14"
         sandbox_nolto:
           TEST_PY_BD: "sandbox"
           BUILD_ENV: "NO_LTO=1"
@@ -486,7 +486,7 @@ stages:
           OVERRIDE: "-a ASAN"
         sandbox_clang_asan:
           BUILDMAN: "sandbox"
-          OVERRIDE: "-O clang-13 -a ASAN"
+          OVERRIDE: "-O clang-14 -a ASAN"
         samsung_socfpga:
           BUILDMAN: "samsung socfpga"
         sun4i:
.gitlab-ci.yml +3 -3
index df224527c680..9eeed0b25fde 100644
@@ -2,7 +2,7 @@
 
 # Grab our configured image.  The source for this is found
 # in the u-boot tree at tools/docker/Dockerfile
-image: trini/u-boot-gitlab-ci-runner:jammy-20221003-17Oct2022
+image: trini/u-boot-gitlab-ci-runner:jammy-20221101-22Nov2022
 
 # We run some tests in different order, to catch some failures quicker.
 stages:
@@ -225,7 +225,7 @@ Run binman, buildman, dtoc, Kconfig and patman testsuites:
 Run tests for Nokia RX-51 (aka N900):
   stage: testsuites
   script:
-    - export PATH=/opt/gcc-11.1.0-nolibc/arm-linux-gnueabi/bin:$PATH;
+    - export PATH=/opt/gcc-12.2.0-nolibc/arm-linux-gnueabi/bin:$PATH;
       test/nokia_rx51_test.sh
 
 # Check for any pylint regressions
@@ -256,7 +256,7 @@ sandbox test.py:
 sandbox with clang test.py:
   variables:
     TEST_PY_BD: "sandbox"
-    OVERRIDE: "-O clang-13"
+    OVERRIDE: "-O clang-14"
   <<: *buildman_and_testpy_dfn
 
 sandbox without LTO test.py:
tools/docker/Dockerfile +34 -31

112 lines not on this page · raw message

Signature · 1 line
2.25.1
Reviewed-by on v1 6/6
#18 lore raw

Re: [PATCH 6/6] CI: Update to gcc-12.2

Quoted 9 lines · Tom Rini

- Update to gcc-12.2, and cherry-pick a fix in grub for risc-v

---
 .azure-pipelines.yml    |  8 ++---
 .gitlab-ci.yml          |  6 ++--
 tools/docker/Dockerfile | 65 +++++++++++++++++++++--------------------
 3 files changed, 41 insertions(+), 38 deletions(-)
author
#20 lore raw
In reply to Tom Rini
Quoted 4 lines · Tom Rini

Update the toolchain list to be first 12.2.0 and second 11.1.0 and that's it.

Applied to u-boot/next, thanks!

Signature · 1 line
Tom