Update to 2-region model for HEAP and Stack Memory #9571

mprse · 2019-01-31T12:48:50Z

Description

The following commits: #8039, #9092 added Boot/ISR stack definition to all scatter files (ARM_LIB_STACK).

This has changed memory model for RTOS-less builds to 2-region memory model and caused failure in case of rtos less builds.
This PR defines valid heap/stack regions for rtos-less builds.

Verified that blinky example works on rtos-less build with this fix.

Pull request type

[ ] Fix
[x] Refactor
[ ] Target update
[ ] Functionality change
[ ] Docs update
[ ] Test update
[ ] Breaking change

Reviewers

@deepikabhavnani Please review.

Release Notes

RAM memory model is updated to 2-region model. RTOS and non-RTOS versions of Mbed OS across all targets and toolchains will have separate stack and heap sections.
ISR stack size is unified across all targets and is set as 1K for Mbed OS with RTOS. In case of no RTOS, stack size is also the main thread size and is set as 4K.

ciarmcom · 2019-01-31T14:00:38Z

@mprse, thank you for your changes.
@deepikabhavnani @ARMmbed/mbed-os-core @ARMmbed/mbed-os-maintainers please review.

0xc0170 · 2019-01-31T14:10:04Z

Adding @ARMmbed/team-nuvoton and @ARMmbed/team-renesas for review

ccli8 · 2019-02-01T03:29:07Z

platform/mbed_retarget.cpp

@@ -904,21 +904,68 @@ extern "C" long PREFIX(_flen)(FILEHANDLE fh)
    return size;
 }

-extern "C" char Image$$RW_IRAM1$$ZI$$Limit[];
+#if defined(TARGET_NUMAKER_IOT_M487) || defined(TARGET_NUMAKER_PFM_M2351) || defined(TARGET_NUMAKER_PFM_M453) || defined(TARGET_NUMAKER_PFM_M487) || defined(TARGET_NUMAKER_PFM_NANO130) || defined(TARGET_NUMAKER_PFM_NUC472) || defined(TARGET_GR_LYCHEE) || defined(TARGET_RZ_A1H) || defined(TARGET_RZ_A1XX)


All Nuvoton targets adopt two-region model, so this list of Nuvoton target defines can simplify to TARGET_NUVOTON.

ccli8 · 2019-02-01T03:53:25Z

platform/mbed_retarget.cpp

-    r.heap_base = zi_limit;
-    r.heap_limit = sp_limit;
+    r.heap_base = (uint32_t)HEAP_BASE;
+    r.heap_limit = (uint32_t)STACK_BASE;


Why assign STACK_BASE to r.heap_limit rather than e.g. Image$$ARM_LIB_HEAP$$ZI$$Limit? Heap and stack may arrange non-consecutive or stack in lower memory/heap in higher memory (like Nuvoton targets)? The same question to $Sub$$__rt_lib_init.

So yeah it looks strange. The problem is that currently on most targets ARM_LIB_HEAP is not defined, but ARM_LIB_STACK is. This should be changed in the future.
Since it is misleading I think I add another macro HEAP_LIMIT.

ccli8 · 2019-02-01T12:33:11Z

platform/mbed_retarget.cpp

+#if defined(TARGET_NUVOTON) || defined(TARGET_RZ_A1XX)
+extern char Image$$ARM_LIB_HEAP$$Base[];
+extern char Image$$ARM_LIB_STACK$$Base[];
+extern char Image$$ARM_LIB_HEAP$$Limit[];


According to Image$$ execution region symbols, Image$$ARM_LIB_HEAP$$Limit doesn't include ZI section. Change to Image$$ARM_LIB_HEAP$$ZI$$Limit?

ccli8 · 2019-02-01T13:03:06Z

platform/mbed_retarget.cpp

+
+extern "C" __value_in_regs struct __argc_argv $Sub$$__rt_lib_init(unsigned heapbase, unsigned heaptop)
+{
+    return $Super$$__rt_lib_init((unsigned)HEAP_BASE, (unsigned)STACK_BASE);


Change STACK_BASE to HEAP_LIMIT?

Missed that. Sorry. Should be fixed now.

0xc0170 · 2019-02-01T13:27:45Z

Thanks @ccli8 for review !

deepikabhavnani

Design and docs changes are needed for new memory model, I will add them next week.

deepikabhavnani · 2019-02-04T21:13:06Z

@mprse - I cannot add my commits to this PR, please cherry pick below listed commits, else I can create another PR (if that is preferred).

https://github.com/deepikabhavnani/mbed-os/commits/2-region

ccli8 · 2019-02-05T02:33:09Z

platform/mbed_retarget.cpp

+#endif
+
+#if !defined(HEAP_START)
+#if defined(Image$$ARM_LIB_HEAP$$ZI$$Base) && defined(Image$$ARM_LIB_HEAP$$ZI$$Length)


@deepikabhavnani Image$$ARM_LIB_HEAP$$ZI$$Base and Image$$ARM_LIB_HEAP$$ZI$$Length are linker-generated symbols. Can they be used for preprocessor defines?

I have not tested it, yes you are right linker symbols will not be valid during pre-processing step... will have to find another condition or define heap_start in target specific file mbed_rtx.h as it was done earlier

d-kato

Sorry for the delay of the review. I have no problem with that.
And thanks @ccli8. Your code was helpful.

mprse · 2019-02-05T07:56:49Z

rtos/TARGET_CORTEX/TOOLCHAIN_ARM_STD/mbed_boot_arm_std.c

+
+extern uint32_t Image$$RW_IRAM1$$ZI$$Limit[];
+
+extern uint32_t Image$$ARM_LIB_HEAP$$ZI$$Base[];


Currently ARM_LIB_HEAP is specified only for few targets (ARM_LIB_STACK for all), this means that the following change will cause build failure for most of the targets.

To fully move to the 2 region memory model we need to specify ARM_LIB_HEAP in all scatter files and __HeapLimit for GCC linker scripts. Then we will be able to completely remove ISR_STACK_SIZE, ISR_STACK_START, ISR_HEAP_SIZE, ISR_HEAP_START macros and relay only on the linker script symbols to define heap and stack.

I already tried that, but I took a step back since there was too much changes and we have discussed this here (and few next posts):
#9092 (comment)

Agree on __HeapLimit, will be needed for all GCC linker files, for now exploring the options to not have ARM_LIB_HEAP in all targets, moving it to mbed_rtx.h back seems reasonable.

Also I will not remove existing symbols for HEAP and Stack, since we have few targets who use completely different ram bank for heap, or want to expand heap to multiple banks. Symbols will be needed for those cases.

Added __HeapLimit for all targets in GCC linker scripts.

deepikabhavnani · 2019-02-26T10:51:22Z

@c1728p9 - Please review

felser

Looks good for MultiTech.

bulislaw · 2019-02-27T10:50:30Z

@deepikabhavnani @mprse please remember to update handbook, there's a page about memory layout (not 100% sure it needs updating, but that needs to be confirmed)

bulislaw · 2019-02-27T10:51:19Z

Also this change was in works for longer time did you guys remembered to update the new targets like PSoC6 based ones?

0xc0170 · 2019-02-27T10:57:34Z

Also this change was in works for longer time did you guys remembered to update the new targets like PSoC6 based ones?

@mprse Can you please verify all targets on master are covered? I'll review now any outstanding targets in PRs to follow this as well.

mprse · 2019-02-27T12:22:27Z

I'll check if all targets are covered.

mprse · 2019-02-27T13:10:15Z

I checked all ARM_STD scatter files and GCC linker scripts and it looks like all are ok. Do we have successful CI run on master after this has been merged?

0xc0170 · 2019-02-27T13:34:33Z

I've referenced this one to all current opened PR for new targets. See above

I checked all ARM_STD scatter files and GCC linker scripts and it looks like all are ok. Do we have successful CI run on master after this has been merged?

I can't access logs for the latest nightly. One job was triggered now, so might have results later today afternoon.

deepikabhavnani · 2019-02-27T15:25:36Z

Also this change was in works for longer time did you guys remembered to update the new targets like PSoC6 based ones?

I did check the linker scripts of PSoC6 and they were compatible. Also Musca was checked and I did shared the updated linker scripts with Musca

deepikabhavnani · 2019-02-27T15:34:28Z

please remember to update handbook, there's a page about memory layout (not 100% sure it needs updating, but that needs to be confirmed)

@bulislaw Bootstrap document is updated, please review ARMmbed/mbed-os-5-docs#951

deepikabhavnani · 2019-02-27T15:36:31Z

there's a page about memory layout

@AnotherButler - Do we have any other doc in handbook describing memory layout (apart from one updated in ARMmbed/mbed-os-5-docs#951)

AnotherButler · 2019-02-27T15:40:20Z

We have this one: https://github.com/ARMmbed/mbed-os-5-docs/blob/development/docs/reference/runtime/Memory.md

deepikabhavnani · 2019-02-27T15:53:53Z

All, We have not changed the location of stack and heap, just separated them so the memory model is still the same.

@AnotherButler - Thanks, I have another concern related to bootstrap.md - This document is updated by this is not visible on mbed.com website.

AnotherButler · 2019-02-27T15:56:26Z

It's not this one? https://os.mbed.com/docs/mbed-os/v5.11/porting/porting-bootstrap.html

deepikabhavnani · 2019-02-27T16:25:41Z

It's not this one? https://os.mbed.com/docs/mbed-os/v5.11/porting/porting-bootstrap.html

👍

LMESTM · 2019-02-27T16:53:18Z

Has this been tested with uARM toolchain ?
When on top of master I see errors like:

$ python tools/singletest.py --auto --tc uARM  -n MBED_23 -v -V
MBEDLS: Detecting connected mbed-enabled devices...
MBEDLS: Detected DISCO_L496AG[FC3B], port: COM3, mounted: D:
[...]
Link: ticker_2
Error: L6218E: Undefined symbol $Super$$__rt_lib_init (referred from mbed_retarget.o).
Error: L6218E: Undefined symbol __use_two_region_memory (referred from mbed_retarget.o).
Finished: 0 information, 0 warning and 2 error messages.

whereas this works ok with ARM tc ...

deepikabhavnani · 2019-02-27T20:04:13Z

@LMESTM - #9766 should fix uARM

mprse mentioned this pull request Jan 31, 2019

Non Rtos build fails for ARM compiler #9523

Closed

ciarmcom requested review from deepikabhavnani and a team January 31, 2019 14:00

ciarmcom added the needs: review label Jan 31, 2019

0xc0170 requested review from a team January 31, 2019 14:09

ccli8 reviewed Feb 1, 2019

View reviewed changes

mprse force-pushed the fix_9523_rtos_less_issue branch from 434794f to 57dd1f4 Compare February 1, 2019 10:35

ccli8 reviewed Feb 1, 2019

View reviewed changes

mprse force-pushed the fix_9523_rtos_less_issue branch 2 times, most recently from 1f22408 to d2d91d5 Compare February 1, 2019 12:57

ccli8 reviewed Feb 1, 2019

View reviewed changes

mprse force-pushed the fix_9523_rtos_less_issue branch 2 times, most recently from 52bee5b to ec86ee1 Compare February 1, 2019 13:15

ccli8 approved these changes Feb 1, 2019

View reviewed changes

0xc0170 approved these changes Feb 1, 2019

View reviewed changes

deepikabhavnani added the needs: work label Feb 1, 2019

deepikabhavnani suggested changes Feb 1, 2019

View reviewed changes

cmonr removed the needs: review label Feb 2, 2019

ccli8 reviewed Feb 5, 2019

View reviewed changes

d-kato approved these changes Feb 5, 2019

View reviewed changes

mprse changed the title ~~Fix for issue #9523 (Non Rtos build fails for ARM compiler)~~ Update to 2-region model for HEAP and Stack Memory Feb 5, 2019

mprse commented Feb 5, 2019

View reviewed changes

deepikabhavnani force-pushed the fix_9523_rtos_less_issue branch from 7319b2f to 70c6057 Compare February 5, 2019 16:48

cmonr removed the needs: review label Feb 26, 2019

c1728p9 approved these changes Feb 26, 2019

View reviewed changes

felser approved these changes Feb 26, 2019

View reviewed changes

cmonr merged commit e1736cd into ARMmbed:master Feb 27, 2019

cmonr removed the ready for merge label Feb 27, 2019

This was referenced Feb 27, 2019

Adding NUCLEO_WB55RG support #9814

Merged

Uhuru RAVEN: Adding platform HAL #9787

Closed

Add TARGET_STMBLUE #9491

Closed

Update SAML21J18A target to Mbed OS 5.12 #8658

Closed

Add target RTL8710AF #7965

Closed

deepikabhavnani deleted the fix_9523_rtos_less_issue branch February 27, 2019 15:26

jeromecoutant mentioned this pull request Mar 6, 2019

GCC - Add support to split heap across 2-RAM banks #9944

Merged

stevew817 mentioned this pull request Mar 8, 2019

Green tea test failing on EFM32GG11_STK3701 #9933

Closed

kjbracey mentioned this pull request Apr 18, 2019

Stream obj failure on two STM32L4 custom targets #10425

Closed


		extern uint32_t Image$$RW_IRAM1$$ZI$$Limit[];

		extern uint32_t Image$$ARM_LIB_HEAP$$ZI$$Base[];

Update to 2-region model for HEAP and Stack Memory #9571

Update to 2-region model for HEAP and Stack Memory #9571

Conversation

mprse commented Jan 31, 2019 • edited by cmonr Loading

Description

Pull request type

Reviewers

Release Notes

ciarmcom commented Jan 31, 2019

0xc0170 commented Jan 31, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

0xc0170 commented Feb 1, 2019

deepikabhavnani left a comment

Choose a reason for hiding this comment

deepikabhavnani commented Feb 4, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

d-kato left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

deepikabhavnani commented Feb 26, 2019

felser left a comment

Choose a reason for hiding this comment

bulislaw commented Feb 27, 2019

bulislaw commented Feb 27, 2019

0xc0170 commented Feb 27, 2019

mprse commented Feb 27, 2019

mprse commented Feb 27, 2019

0xc0170 commented Feb 27, 2019 • edited Loading

deepikabhavnani commented Feb 27, 2019

deepikabhavnani commented Feb 27, 2019

deepikabhavnani commented Feb 27, 2019

AnotherButler commented Feb 27, 2019

deepikabhavnani commented Feb 27, 2019

AnotherButler commented Feb 27, 2019

deepikabhavnani commented Feb 27, 2019

LMESTM commented Feb 27, 2019

deepikabhavnani commented Feb 27, 2019

mprse commented Jan 31, 2019 •

edited by cmonr

Loading

0xc0170 commented Feb 27, 2019 •

edited

Loading