🚨🔊 Add a better Data Abort Handler #112

edjubuh · 2019-02-17T00:55:05Z

Summary:

Dunno how to be concise yet. But this makes the data abort handler do something interesting.

We have our own error box on the V5 that reports the proper PC and also current task name (because we can)
We still background process and printfs work, so I print registers of the state that caused the abort and stack trace of the running task
- This required adding unwind tables. Some additional adjustments will need to be made once ✨ Add 🔥/❄ linking #89 lands to ensure that the unwinder can find both hot and cold tables.

There will need to be additional work in the CLI to create a frontend for doing addr2line and objdump to make it a bit easier to see where things went wrong.

Motivation:

I'm tired of guessing why the MEMORY ERROR 0000000 box pops up because the 0000000 is meaningless.

References (optional):

Closes #150

Test Plan:

int x = *(int*)0; printf("%d\n" ,x); and get this output

Octogonapus · 2019-02-17T01:01:02Z

I have been seeing a lot of people ask for memory usage. Is it possible to get the current heap/task stack memory usage? That might also be something we want to display.

edjubuh · 2019-02-17T01:08:11Z

I have been seeing a lot of people ask for memory usage. Is it possible to get the current heap/task stack memory usage? That might also be something we want to display.

Stack usage is a bit philosophical. Do you want to see current task stack usage or peak stack usage or perhaps average stack usage? Current task stack usage at the time of the abort is easy since we know the current SP and where the top of the stack is (since we have access to the TCB). Peak stack usage is a more expensive to compute and perfect if you're trying to fool it, but FreeRTOS does have a facility. We don't currently measure any sort of average stack usage. Average stack usage could be probed periodically on every context switch and do a rolling average on that, but that seems like a lot of work for little tangible benefit.

edjubuh · 2019-02-17T01:12:44Z

I have been seeing a lot of people ask for memory usage. Is it possible to get the current heap/task stack memory usage? That might also be something we want to display.

Heap usage can be done by users today with mallinfo. I'll add that to the printf'd abort report

Octogonapus · 2019-02-17T01:18:58Z

that seems like a lot of work for little tangible benefit

I agree. Just reporting whatever the stack usage happens to be at abort time is probably good enough for now.

edjubuh · 2019-03-07T21:10:55Z

From some initial testing from @baylessj, we've discovered that many (if not all) of the FILE* family of functions (fopen, fputs) are marked as cantunwind. Investigating further.

edjubuh · 2019-03-07T21:34:53Z

Turns out there's no unwinding information in newlib/libc.a that ships in the GNU Toolchain whatsoever... I'm trying to figure out if that was intentional.

arm-none-eabi-readelf -u /usr/arm-none-eabi/lib/libc.a | less

src/system/hot.c

@@ -71,6 +76,7 @@ void invoke_install_hot_table() {
  // printf("%s %p %p %x %x\n", __FUNCTION__, (void*)install_hot_table, (void*)HOT_TABLE, MAGIC_ADDR[0], MAGIC_ADDR[1]);
  if(vexSystemLinkAddrGet() == (uint32_t)0x03800000 && MAGIC_ADDR[0] == MAGIC0 && MAGIC_ADDR[1] == MAGIC1) {
    install_hot_table(HOT_TABLE);
+


missed this in #155, but these two functions are moved to their own file for the same reason as __malloc_lock/__malloc_unlock were-- to follow the override scheme that newlib expects. found the right filename from here: https://newlib.sourceware.narkive.com/aXWplROp/queries-concerning-rtos-protection-in-newllib

.arclint

HotelCalifornia

this looks good to me, now that it works and compiles and all that 😄

@kdcrafter

* Better errno for registry validation failure, update vision sensor errnos Summary: - Changed port number validation to return ENXIO when outside the domain of port numbers - Changed port validation to return ENODEV when the requested device doesn't match the actual device - Vision Sensor read/get functions consistently use EAGAIN when the libv5rts tells us it couldn't get data - Remove docstrings for EACCES since it is practically impossible to get that errno (we wait for mutex indefinitely) Test Plan: TBD * ✨ Add Makefile-variable for excluding libraries from cold image (#121) #### Summary: Someone want to enable 🔥/❄ function but he doesn't want to include all libraries in the cold image. So he adds this feature here. you can edit `EXCLUDE_COLD_LIBRARIES` to exclude the library. The library will be compiled every time you use `make` command. (🔥🔥🔥) #### Motivation: Add it. ##### References (optional): lazy to make an issue. 404 not found. #### Test Plan: - [x] Make a library for testing - [x] Make a new project and apply the testing library - [x] Compile with 🔥/❄ function (use `make all`) - [x] Upload and test if the library is running? - [x] Exclude the library - [x] Compile with 🔥/❄ function (use `make all`) - [x] Upload and test if the library is running? - [x] Change something in the project - [x] Compile with 🔥/❄ function (use `make`) - [x] Upload and test if the library is running? - [x] delete variable `EXCLUDE_COLD_LIBRARIES` - [x] Compile with 🔥/❄ function (use `make all`) - [x] Upload and test if the library is running? * Update vdml_vision.c * Update vdml_vision.c * Update vdml_vision.c * Update Copyright info to current year * 🐛 Fix cold libraries variable assignment (#129) * Apply suggestions from code review Co-Authored-By: edjubuh <edjubuh@elliotjb.com> * Fix whitespace * Fix up errno when mismatch/not connected * Fix up errno when mismatch/not connected for ADI * Fix up errno when mismatched/not connected for ADI round 2 * Fix up errno when mismatched/not connected for ADI round 3 * ✨ Improve Makefile compile speed (#123) #### Summary: Someone doesn't want to waste his time. So he adds this code to make compile-job faster. The compiler now doesn't need to remove and create `temp.log` every time it compiles a file. #### Motivation: Add it. ##### References (optional): close #103 #### Test Plan: - [x] Call `make all` - [x] Call `make` - [x] Try to compile with no error - [x] Try to compile with warning - [x] Try to compile with error #### Commits: * ✨ Improve Makefile compile speed * Use only variables; use printf * Keep changes from #119 * Do not treat strings as escaped * Migate kernel makefile to test_output_2 (#130) #### Summary: Migrate kernel makefile to `test_output_2` and make make make again. #### Motivation: It's broken because `@cd` isn't a command. `@` was supposed to suppress echoing the whole command, but `test_output` has changed. ##### References (optional): #123 #### Test Plan: - [X] It build * Fix incorrect port type (#139) #### Summary: Fix incorrect port type in some motor functions #### Motivation: They were different from everywhere else with seemingly no reason. #### Test Plan: - [x] Still compiles * Add alternate generic serial API and fix generic serial filesystem driver (#141) * Add alternate serial API * Fix generic serial files * Add serial tests * Fix dev_read_r and dev_write_r starving tasks * Revert DEVCTL_SET_BAUDRATE to maintain compatibility * Update generic serial test * Modify blocking serial sleep time * Move serial include * Change serial telemetry function names * 🐛 Update vision_read_by_sig to be less funky (#124) * Update vision_read_by_sig to properly transform all coordinates and respect array boundaries Summary: - We weren't properly respecting array boundaries on read_by_sig, now we are Test Plan: - [x] @kdcrafter tested in T659 and throughout Worlds * Update vdml_vision.c * 🚨 Make things work with GCC 8.3 (#155) * add __sync_synchronize stub * catch exception by reference to suppress new warning * move mlock functions to their own file to fix newlib * 🔧 Enable hot/cold by default (#154) * 🏗 Combine template source files (#152) * Rename main.c to startup.c * Combine initialize, autonomous, and opcontrol into main * Patch Makefile * Patch template-Makefile * Don't change USE_PACKAGE in this PR * 🐛 Fix multiple ADI bugs and inconsistencies (#151) * Fix IntelliSense error * Note possible hack with registry pad * Note possible issue with claim_port * Cleanup ADI C API * Remove config clear from ADI destructor * Move ADI config to initilizer list * Fix double normalization of ultrasonic port * Partially Revert "Remove config clear from ADI destructor" This partially reverts commit 6a562ef. * Declare ADIPort destructor as default * Revert "Note possible hack with registry pad" This reverts commit 866067b. * Change claim_port note * ⬆️ Upgrade LVGL to 5.3 (#144) * [wip] rollback point * finish upgrade\? doubt it * update lv_conf * make everything compile - still need to test for some regressions - still need to test some 5.3 features * initialize LVGL before system daemon * make non-monospaced font the default again * limit LLEMU screen to 33 monospace chars * clean versions * add comment about system_daemon_initialize * 🤫 Suppress ignored qualifiers when including lvgl (#147) * 🔧 Add default `.gitignore` to template (#156) * 🚸 Add alternate constructor for pros::Task (#158) * Alternate Task contructor: ommission of prio and stack_depth while still allowing access to name * Delegation of truncated constructor to the normal constructor * 💚 Update toolchain installation for azure (#163) * update toolchain installation for azure the PPA was hanging, and it installs an old version anyway * Update azure-pipelines.yml Co-Authored-By: Elliot Berman <berman5@purdue.edu> * 🚨🔊 Add a better Data Abort Handler (#112) * Add WIP of a better DataAbortHandler * More WIP * Cleanup * refactor some things * Refactor, clean up, and documentation * More comments * Update unwind.c * Tweak printfs * Add custom build of newlib * Include libc/libm in template * Exclude libc/libm from being included in cold image * check whether pc is in the hot zone * test unwind with hot/cold linking * Add support for unwinding through hot code * Clean up for PR * Undo USE_PACKAGE:=1 * remove leftover include * move __env_lock/__env_unlock to src/system/envlock.c missed this in #155, but these two functions are moved to their own file for the same reason as __malloc_lock/__malloc_unlock were-- to follow the override scheme that newlib expects. found the right filename from here: https://newlib.sourceware.narkive.com/aXWplROp/queries-concerning-rtos-protection-in-newllib * lint files * Update .arclint whitespace * 🐛 Add default LLEMU init code back to main.cpp (#165) * 🐛 adddefault LLEMU init code back to main.cpp * add documentation comment for callback function * Bump to 3.2.0 * 🔧 Recompile on header change (#167) * implemented header dependency tracking * Added dependency folder to the makefile * fix bug in making c folders * Renamed POSTCOMPILE * Undo unwanted changes * Modify template gitignore * Ignore clean errors * 🐛 Return proper error code for VDML functions that return floating point numbers (#160) * 🐛 Fix problem of return PROS_ERR_F error code * 🐛 Improve condition statement and comment VDML function 🤫 forgot to change * 🐛 Undo claim_port_try code to avoid warning 🐛 Undo claim_port_try code to avoid warning * 🐛 Fix motor PID functions bug (#162)

edjubuh added 2 commits February 15, 2019 23:52

Add WIP of a better DataAbortHandler

065dc55

More WIP

bf074e1

edjubuh added p: normal Normal priority enhancement This builds on top of an existing feature labels Feb 17, 2019

edjubuh added this to the 3.1.7 milestone Feb 17, 2019

edjubuh changed the title ~~Add a better~~ Add a better Data Abort Handler Feb 17, 2019

edjubuh added 2 commits February 16, 2019 20:23

Cleanup

f1e47cd

refactor some things

0742952

edjubuh changed the title ~~Add a better Data Abort Handler~~ 🚨🔊 Add a better Data Abort Handler Feb 17, 2019

edjubuh added 3 commits February 17, 2019 13:41

Refactor, clean up, and documentation

bd02ca4

Merge branch 'develop' into feature/better-abort-handler

56d3875

More comments

7f34e55

edjubuh added this to In progress in 3.1.7 Feb 18, 2019

edjubuh mentioned this pull request Feb 18, 2019

Add interactive window for analyzing stack traces purduesigbots/pros-cli#76

Open

edjubuh added this to In Progress in PROS Non-Versioned Kernel Development Feb 18, 2019

HotelCalifornia added the in progress This is currently being worked on label Feb 18, 2019

edjubuh marked this pull request as ready for review February 20, 2019 00:39

edjubuh added 2 commits February 25, 2019 17:46

Update unwind.c

b92431b

Tweak printfs

2c96190

edjubuh self-assigned this Mar 7, 2019

edjubuh added 4 commits March 10, 2019 15:44

Add custom build of newlib

6c5e72e

Include libc/libm in template

d2840a6

Merge branch 'develop' into feature/better-abort-handler

8712d1e

Merge branch 'develop' into feature/better-abort-handler

cacb282

test unwind with hot/cold linking

8eb3c3c

HotelCalifornia dismissed baylessj’s stale review via 8eb3c3c August 8, 2019 02:21

PROS Non-Versioned Kernel Development automation moved this from Reviewer Approved to In Review Aug 8, 2019

Merge branch 'develop' into feature/better-abort-handler

3806d04

HotelCalifornia moved this from In Review to In Progress in PROS Non-Versioned Kernel Development Aug 20, 2019

edjubuh added 2 commits August 31, 2019 16:06

Merge branch 'develop' into feature/better-abort-handler

4c2166e

Add support for unwinding through hot code

b38b510

edjubuh commented Aug 31, 2019

View reviewed changes

edjubuh and others added 4 commits August 31, 2019 17:03

Clean up for PR

588a18c

Undo USE_PACKAGE:=1

b547956

Merge branch 'develop' into feature/better-abort-handler

9e211e4

remove leftover include

e76d9e7

HotelCalifornia added enhancement This builds on top of an existing feature feature A brand new feature and removed enhancement This builds on top of an existing feature labels Sep 2, 2019

HotelCalifornia added 2 commits September 2, 2019 18:58

lint files

5dc60f0

HotelCalifornia reviewed Sep 2, 2019

View reviewed changes

.arclint Outdated Show resolved Hide resolved

Update .arclint whitespace

f719774

HotelCalifornia requested review from HotelCalifornia and baylessj September 2, 2019 23:23

PROS Non-Versioned Kernel Development automation moved this from In Progress to Reviewer Approved Sep 2, 2019

HotelCalifornia approved these changes Sep 2, 2019

View reviewed changes

HotelCalifornia merged commit 6394b23 into develop Sep 2, 2019

PROS Non-Versioned Kernel Development automation moved this from Reviewer Approved to Done Sep 2, 2019

HotelCalifornia mentioned this pull request Sep 3, 2019

🔖 Release 3.2.0 #166

Merged

1 task

HotelCalifornia deleted the feature/better-abort-handler branch September 13, 2020 07:38

Sleuth56 mentioned this pull request Dec 17, 2020

Add interactive stack trace #90 to develope!! Team20610N/pros-cli#6

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

🚨🔊 Add a better Data Abort Handler #112

🚨🔊 Add a better Data Abort Handler #112

edjubuh commented Feb 17, 2019 •

edited

Loading

Octogonapus commented Feb 17, 2019 •

edited

Loading

edjubuh commented Feb 17, 2019

edjubuh commented Feb 17, 2019

Octogonapus commented Feb 17, 2019

edjubuh commented Mar 7, 2019

edjubuh commented Mar 7, 2019

This comment was marked as resolved.

HotelCalifornia left a comment

🚨🔊 Add a better Data Abort Handler #112

🚨🔊 Add a better Data Abort Handler #112

Conversation

edjubuh commented Feb 17, 2019 • edited Loading

Summary:

Motivation:

References (optional):

Test Plan:

Octogonapus commented Feb 17, 2019 • edited Loading

edjubuh commented Feb 17, 2019

edjubuh commented Feb 17, 2019

Octogonapus commented Feb 17, 2019

edjubuh commented Mar 7, 2019

edjubuh commented Mar 7, 2019

This comment was marked as resolved.

HotelCalifornia left a comment

Choose a reason for hiding this comment

edjubuh commented Feb 17, 2019 •

edited

Loading

Octogonapus commented Feb 17, 2019 •

edited

Loading