[Ovmsdev] version 3.2.007: Crash on boot

Michael Balzer dexter at expeedo.de
Thu Dec 12 18:12:22 HKT 2019


Mark,

1) maybe I missed posting my later version of a2l, which automatically strips the stack pointers from the ":" address syntax so you can copy&paste the USB
output. Here it is:

#!/bin/bash
elf=~/esp/Open-Vehicle-Monitoring-System-3/vehicle/OVMS.V3/build/ovms3.elf
for adr in $* ; do
  if [[ "$adr" =~ "elf" ]] ; then
    elf="$adr"
  else
    cmd+=" -ex 'l *${adr/:*/}'"
  fi
done
cmd+=" -ex 'q'"
echo "Using elf file: $elf"


2) You're right, the bug is tx_frame with null origin overwriting body.bus in the union. I didn't notice that when checking Marko's submission.

tx_frame is a copy of the last frame given to the bus for transmission. The queue msg is gone when the TX done IRQ comes in, and Marko needed a copy of the
frame the TX IRQ relates to. I asked him (see PR discussion), he checked and confirmed that all TX is done sequentially, so a single buffer is sufficient.

Swapping the lines would work. The frame.origin also shouldn't be null, but the handler should tolerate that. …oops, tx_frame also doesn't get initialized in
the canbus constructor, so there's also potentially garbage in tx_frame if due to some bug a TX IRQ is generated or processed without a previous tx.

I'll do the fixes… and also rename tx_frame to m_tx_frame for consistency.

Regards,
Michael


Am 12.12.19 um 01:20 schrieb Mark Webb-Johnson:
> Two issues:
>
> _*1] A2L*_
>
> My a2l is this:
>
>     #!/bin/bash
>     elf=3.2.007.ovms3.elf
>     for adr in $* ; do
>       if [[ "$adr" =~ "elf" ]] ; then
>         elf="$adr"
>       else
>         cmd+=" -ex 'l *$adr'"
>       fi
>     done
>     cmd+=" -ex 'q'"
>     echo "Using elf file: $elf"
>     echo "xtensa-esp32-elf-gdb -batch $elf $cmd"
>     eval xtensa-esp32-elf-gdb -batch $elf $cmd 2>/dev/null #| grep "^0x.* is in "
>
>
> When I run it, I get:
>
>     $ a2l 3.2.007.ovms3.elf 0x400d5e4c:0x3ffc5c40 0x7ffffffd:0x3ffc5c90
>     Using elf file: 3.2.007.ovms3.elf
>     xtensa-esp32-elf-gdb -batch 3.2.007.ovms3.elf  -ex 'l *0x400d5e4c:0x3ffc5c40' -ex 'l *0x7ffffffd:0x3ffc5c90' -ex ‘q'
>
>
> And a manual run gives:
>
>     $ xtensa-esp32-elf-gdb -batch 3.2.007.ovms3.elf  -ex 'l *0x400d5e4c:0x3ffc5c40' -ex 'l *0x7ffffffd:0x3ffc5c90' -ex 'q'
>     Junk at end of line specification.
>
>     $ xtensa-esp32-elf-gdb 3.2.007.ovms3.elf
>     GNU gdb (crosstool-NG crosstool-ng-1.22.0-80-g6c4433a) 7.10
>     (gdb) l *0x400d5e4c:0x3ffc5c40
>     Junk at end of line specification.
>     (gdb) l *0x7ffffffd:0x3ffc5c90
>     (gdb) quit
>
>
> _*2] Frame origin*_
>
> Regarding the CAN_txcallback, I think you are correct. And both generators of that message set the frame correctly. So my initial thought was that it is
> either a memory corruption, or somewhere else sending a frame with garbage data.
>
> I do see this technique used in both the mcp2515 and esp32can drivers:
>
>     msg.body.bus = me;
>     msg.body.frame = me->tx_frame;
>
>
> I don’t normally just copy structures over like that. I memcpy() them, but I guess it must work. However, as our CAN_queue_msg_t is a union of CAN_frame_t
> (frame, with first member of the structure a canbus*) and 'canbus* bus’, that is a little worrying. It seems that the msg.body.bus will get overwritten with
> whatever is in msg.body.frame.origin. I can’t see anywhere that tx_frame.origin is set - which is scary because that would mean it is always random junk.
>
> Maybe it works if tx_frame.origin is set to the bus before anything else, but not in Greg’s circumstances where something else arrives first. Perhaps related
> to obd2ecu? I only see tx_frame set in can.cpp canbus::Write. I don’t really understand the tx_frame approach at all, and why the frame is just not passed on
> the queue.
>
>     // CAN Frame
>     // Note: Take care changing this structure, as it is a union with
>     // CAN_log_message_t and position of 'origin' is fixed.
>     struct CAN_frame_t
>       {
>       canbus*     origin;                   // Origin of the frame
>       CanFrameCallback * callback;          // Frame-specific callback. Is called when this frame is successfully sent (or sending failed)
>       CAN_FIR_t   FIR;                      // Frame information record
>       uint32_t    MsgID;                    // Message ID
>       union
>         {
>         uint8_t   u8[8];                    // Payload byte access
>         uint32_t  u32[2];                   // Payload u32 access (Att: little endian!)
>         uint64_t  u64;                      // Payload u64 access (Att: little endian!)
>         } data;
>
>       esp_err_t Write(canbus* bus=NULL, TickType_t maxqueuewait=0);  // bus: NULL=origin
>       };
>
>     // CAN message
>     typedef struct
>       {
>       CAN_queue_type_t type;
>       union
>         {
>         CAN_frame_t frame;  // CAN_frame
>         canbus* bus;
>         } body;
>       } CAN_queue_msg_t;
>
>
> This approach seems to date back to the swcan support merge (f94ae5a1b).
>
> Should we swap around, and set the msg.body.bus after msg.body.frame? Or am I missing something…
>
> Regards, Mark.
>
>> On 12 Dec 2019, at 2:29 AM, Michael Balzer <dexter at expeedo.de <mailto:dexter at expeedo.de>> wrote:
>>
>> Mark,
>>
>> good example why not to use addr2line: I think that result is wrong. a2l uses gdb which gives:
>>
>> balzer at leela:~/esp/Open-Vehicle-Monitoring-System-3/vehicle/OVMS.V3> a2l tmp/3.2.007.ovms3.elf 0x400d5e4c:0x3ffc5c40 0x7ffffffd:0x3ffc5c90
>> Using elf file: tmp/3.2.007.ovms3.elf
>> 0x400d5e4c is in CAN_rxtask(void*) (/home/openvehicles/build/Open-Vehicle-Monitoring-System-3.1/vehicle/OVMS.V3/components/can/src/can.cpp:730).
>> 725                  me->IncomingFrame(&msg.body.frame);
>> 726                } while (loop);
>> 727              break;
>> 728              }
>> 729            case CAN_txcallback:
>> 730              msg.body.bus->TxCallback(&msg.body.frame, true);
>> 731              break;
>> 732            case CAN_txfailedcallback:
>> 733              msg.body.bus->TxCallback(&msg.body.frame, false);
>> 734              msg.body.bus->LogStatus(CAN_LogStatus_Error);
>>
>> …and that actually makes sense and matches the register dump.
>>
>> If I read the gdb disassembly correctly, A10 = msg.body.bus, so Greg's got a CAN_txcallback msg without a bus.
>>
>> Hardening the rxtask against null here would probably avoid the crash, but I don't see yet how that could be possible.
>> Both esp32can and mcp2515 set the bus field to their object addresses, which cannot be null.
>>
>> Regards,
>> Michael
>>
>>
>> Am 11.12.19 um 13:45 schrieb Mark Webb-Johnson:
>>> Can’t get a2l working at the moment. The addr2line gives:
>>>
>>>     addr2line -e 3.2.007.ovms3.elf 0x400d5e4c:0x3ffc5c40 0x7ffffffd:0x3ffc5c90
>>>     /home/openvehicles/build/Open-Vehicle-Monitoring-System-3.1/vehicle/OVMS.V3/components/can/src/can.cpp:551
>>>
>>>
>>> That is:
>>>
>>>     void canbus::LogInfo(CAN_log_type_t type, const char* text)
>>>       {
>>>       MyCan.LogInfo(this, type, text);    <—— HERE
>>>       }
>>>
>>>
>>> ELF is at:
>>>
>>>     http://api.openvehicles.com/firmware/ota/v3.2/main/3.2.007.ovms3.elf
>>>
>>>
>>> Regards, Mark.
>>>
>>>> On 11 Dec 2019, at 5:20 PM, Michael Balzer <dexter at expeedo.de <mailto:dexter at expeedo.de>> wrote:
>>>>
>>>> Mark,
>>>>
>>>> Greg uses your build, the crash point seems to be consistent, can you post the a2l on this?
>>>>
>>>> Regards,
>>>> Michael
>>>>
>>>> PS: Greg, would you mind switching to EAP to beta test future releases?
>>>>
>>>>
>>>> Am 11.12.19 um 04:33 schrieb Greg D.:
>>>>> Hi folks,
>>>>>
>>>>> Well, the module updated to 3.2.007 last night.  I just checked on it,
>>>>> and it appears that it didn't exactly survive.  Crashed while running
>>>>> the autoconfig script.  Log with two cycles attached.
>>>>>
>>>>> I tried renaming the /store/events directory to /store/was_events since
>>>>> it seems like Duktape was getting in the way, but that didn't resolve
>>>>> the crash.  I manually enabled wifi so I could manage the module, and
>>>>> moved it back to 3.2.005 from the other partition.  Seems stable again.
>>>>>
>>>>> The car is going into the Service Center tomorrow (the perennial issue
>>>>> with 1146 alerts), so I need to have the module stable so that I can
>>>>> keep an eye on it.  Going to leave it on 3.2.005 for now, unless someone
>>>>> has a quick fix in the next few hours...
>>>>>
>>>>> Otherwise, any ideas for troubleshooting after I get the car back back
>>>>> (hopefully end of day)?
>>>>>
>>>>> Greg
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> OvmsDev mailing list
>>>>> OvmsDev at lists.openvehicles.com
>>>>> http://lists.openvehicles.com/mailman/listinfo/ovmsdev
>>>>
>>>>
>>>> -- 
>>>> Michael Balzer * Helkenberger Weg 9 * D-58256 Ennepetal
>>>> Fon 02333 / 833 5735 * Handy 0176 / 206 989 26
>>>> _______________________________________________
>>>> OvmsDev mailing list
>>>> OvmsDev at lists.openvehicles.com <mailto:OvmsDev at lists.openvehicles.com>
>>>> http://lists.openvehicles.com/mailman/listinfo/ovmsdev
>>>
>>>
>>> _______________________________________________
>>> OvmsDev mailing list
>>> OvmsDev at lists.openvehicles.com
>>> http://lists.openvehicles.com/mailman/listinfo/ovmsdev
>>
>> -- 
>> Michael Balzer * Helkenberger Weg 9 * D-58256 Ennepetal
>> Fon 02333 / 833 5735 * Handy 0176 / 206 989 26
>> _______________________________________________
>> OvmsDev mailing list
>> OvmsDev at lists.openvehicles.com <mailto:OvmsDev at lists.openvehicles.com>
>> http://lists.openvehicles.com/mailman/listinfo/ovmsdev
>
>
> _______________________________________________
> OvmsDev mailing list
> OvmsDev at lists.openvehicles.com
> http://lists.openvehicles.com/mailman/listinfo/ovmsdev

-- 
Michael Balzer * Helkenberger Weg 9 * D-58256 Ennepetal
Fon 02333 / 833 5735 * Handy 0176 / 206 989 26

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openvehicles.com/pipermail/ovmsdev/attachments/20191212/c316f73c/attachment.htm>


More information about the OvmsDev mailing list