Data abort on Sansa Clip+ with latest daily build

[Carlo]

After updating Rockbox on my Sansa Clip+ to the 20230208 daily build I'm frequently experiencing "Data abort" errors when switching tracks or fast forwarding. They started to happen right after updating from a last december build. A sample of the error summary:

Data abort at 30063908
FSR 0x8
(domain 0, fault
address 0x57ee557a
pc:30063908
sp:300d47a8
bt end

The MP3 files I'm listening to have been transferred on the Clip+ several months ago, so no recent additions, and the error manifest itself on average after 15 minutes, sometimes just after two or three minutes from booting. It cannot be reproduced, but it's triggered by switching tracks and by fast forward/reverse. Daily build from today shows the same behavior.

I've performed a r/w test on the internal memory and there are no reported issues.

Any suggestion about how to further diagnose and solve the issue? Thanks.

Edit: it seems it could be related to this bug report, https://www.rockbox.org/tracker/task/13388 . I tried to register on the bug tracker but I receive the error "Fatal error: Array and string offset access syntax with curly braces is no longer supported in /home/rockbox/flyspray/vendor/dapphp/securimage/securimage.php on line 2209".

Edit 2: it happens with the build from 20230113 too, so the bug was introduced earlier.

[Bilgus]

I have a clip+ as well and I can't repro this so far

if you want to give me the dataabort on this build we can try to trace it
I've included the .map file which should show us where the program counter is pointing

https://www.mediafire.com/file/sxd4vdznpuj8f0c/ClipPlus_3bb75e0039-230213_rockbox-full.zip/file

be sure to rename your old .rockbox folder and do a clean 'install'

it doesn't mean there is or isn't an issue with your sd card though but i'd try a different one anyway

[Carlo]

It's unlikely that's an hardware issue, as I've compiled a build from 20221230 and it seems fine so far. I'll give your build a try and report back.

[Carlo]

The error showed up again when using your build, this is the data:

Data abort at 30062974
FSR 0x8
(domain 0, fault 8 )
address 0xfffffef4
pc:30062974
sp:300d4740

It happened right after booting and switching no more than 6 or 7 tracks. A second one, after a few minutes:

Data abort at 30062318
FSR 0x8
(domain 0, fault 8 )
address 0xac5fdf74
pc:30062318
sp:300d4740
bt end

It seems the only common address between the last two crashes is the stack pointer one. A third one, also happened while cycling through the same directory as the other two crashes:

Data abort at 30063908
FSR 0x8
(domain 0, fault 8 )
address: 0x55555561
pc:30063908
sp:300d47a8
bt end

[Bilgus]

The first two are in the font system and the third appears to be in LCD init_viewport are you using any themes or is it a bone stock install? Also do the songs it crashes on crash everytime or just randomly? any special chars in those filenames ETC?

[Carlo]

1) I'm using the Informative Beauty theme, as I've done for years on all my Clip+ players without any issue.

2) The crashes are random, I could cycle 5-6 times through the files in the same directory before getting one.

3) No special characters. Again, last time I added new files on this device was a few months ago, and it's my daily driver - so I noticed the crashes immediately after updating, as until that moment it was working perfectly.

I can confirm beyond any reasonable doubt that the 20221230 build is NOT affected by the issue, while 20230113 is affected.

Thank you for your help!

[Bilgus]

Try to repro the issue using the default theme cabbie the closer you can get me to the issue the less time it'll take to fix it if i can reproduce it it'll be even faster otherwise we will have to bisect through 10-100 builds one at a time

[Carlo]

Could the bug be related to the "[BUGFix] ClipPlus move picker below yellow screen split" from 20230113? It wouldn't be that much of a hassle for me to bisect the two weeks between 20221230 and 20230113 until I can pinpoint a bug free build.

I've prepared a Debian vm with the proper compiler toolchain last night, and it doesn't take that long to checkout a specific commit, build and test.

[Bilgus]

not unless you are using the keyboard in that run..

git bisect is your next course of action if you can compile builds or I can for you given about 30 mins to an hour
see us on IRC in that case

[Carlo]

I'm able and willing to compile builds, will start right now with 20230106 and see how it goes.

[Bilgus]

https://gerrit.rockbox.org/r/c/rockbox/+/5112
parent 78718aa7eb < here is where it works is my guess

[Carlo]

It seems that change was committed on Jan 29, but the crashes already happen with the daily Jan 13 build.

[Bilgus]

I think I found it its been merged so try the latest daily
https://gerrit.rockbox.org/r/c/rockbox/+/5123

[Carlo]

Thank you very much, I will try the latest daily for a couple of hours later today and report back.

[Carlo]

Sadly I have bad news: I had an hard freeze with the Informative Beauty theme a couple of minutes after booting, but when I switched to the default Cabbiev2 theme it worked fine for more than two hours. I'll do more testing tonight.

Edit: as puzzling as it sounds, after performing a clean install and manually changing all the relevant settings I didn't have any crashes for more than one hour.