[PATCH] rtems-tools/covoar: -f switch was not used to select coverage format

[Jiri Gaisler]

On 1/24/19 4:19 PM, Joel Sherrill wrote:
>
>
> On Thu, Jan 24, 2019 at 8:46 AM Jiri Gaisler <jiri at gaisler.se
> <mailto:jiri at gaisler.se>> wrote:
>
>
>     On 1/24/19 2:19 PM, Joel Sherrill wrote:
>>
>>
>>     On Thu, Jan 24, 2019, 6:38 AM Jiri Gaisler <jiri at gaisler.se
>>     <mailto:jiri at gaisler.se> wrote:
>>
>>         Small patch to fix covoar to work with TSIM coverage files.
>>         It should be noted that covoar erroneously marks some code as
>>         uncovered while it is marked as covered in the coverage file.
>>         This seems to be due to incorrect parsing of the symbol table
>>         in the exec file. I will file a ticket for this on trac ..
>>
>>
>>     Is this the entry and exit code of a method or something else?
>>     The switch to dwarf info seems to have caused that and not
>>     marking assembly from inlined methods.
>
>     I notice two problems: the first instruction in a range is
>     sometimes marked as uncovered:
>
>     4000807c <_Freechain_Get>:
>       Freechain_Control   *freechain,
>       Freechain_Allocator  allocator,
>       size_t               number_nodes_to_extend,
>       size_t               node_size
>     )
>     {
>     4000807c:       9d e3 bf a0     save  %sp, -96, %sp
>                                <== NOT EXECUTED
>     40008080:       ba 10 00 18     mov  %i0, %i5
>
>       return _Chain_Immutable_head( the_chain )->next;
>
>     40008084:       f0 06 00 00     ld  [ %i0 ], %i0
>
>     even though it is marked as executed in the coverage file. The
>     second problems is that the size of ranges is sometimes off by one
>     byte, e.g. 21 bytes instead of 20. This can only be seen when
>     debugging covoar in gdb.
>
>
> I recall that the CoverageReaderTSIM code did an offset of +1 to +4
> from the address. Perhaps it just needs to be 0-3. :)

CoverageReaderTSIM calls aCoverageMap->setWasExecuted() with correct 
addresses from the coverage file, I checked that through debug prints.
Checking the final annotated sources, I can see that the first
instruction in each range is always marked as uncovered, regardless of
coverage data. The remaining instructions seem to be annotated
correctly. For some reason, theCoverageMap->wasExecuted() in
ReportsBase.cc does not return the correct data for the first address in
each range. My C++ skills are rather poor, so I have a hard time
debugging all the object nonsense to understand why the coverage map is
wrong.

Maybe my mind will clear a bit after some food ... :-)



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20190124/e59bbfe0/attachment-0002.html>