[ros-dev] Re: [ros-svn] [hbirr] 13570: - Started the rewriting of the cache manger.

[Filip Navara]

Hartmut Birr wrote:

>> [cc02.diff] Don't try to map data past the end of file in VFAT.
>
> I think that this patch isn't necessary, because the real problem is 
> in pin.c.

At first I did the same thing as you did in the attached patch. Then I 
actually followed it by some testing on Windows and _it is possible to 
map data past the end of file_ with CcMapData. I haven't tried to access 
the mapped memory though...

> [cc03.diff] Fix incorrect check in CcMapData.
>
> The check which you have changed was correct. Currently the file size 
> is limited to the number of cache view entries. I will implement a 
> x-level tree structure in the future. But some other checks are 
> incorrect. I've add a patch.

Oh, sorry, I didn't realize it's checked because of multi-level VACB 
tables. A comment would have been helpful ;-)

>> Also I've patch which fixes fast mutex implementation and it's usage, 
>> which depends on these cache manager changes.
>
[snip]

To clarify this, the fix I have fixes some small issues with FAST_MUTEX 
implementation and changes it to actually raise to APC_LEVEL instead of 
entring critical section. Many people (especially Mike Nordell ;-) have 
been complaining about it for some time and all previous attempts to fix 
it failed because of the broken locking in cache manager. Now that you 
have done the dirty work, it's finally possible to correctly fix this 
issue... ;-)

Regards,
Filip