Context switching for protected stacks

[Utkarsh Rai]

On Fri, Jun 19, 2020 at 3:04 AM Gedare Bloom <gedare at rtems.org> wrote:

> On Thu, Jun 18, 2020 at 8:03 AM Utkarsh Rai <utkarsh.rai60 at gmail.com>
> wrote:
> >
> >
> >
> > On Thu, Jun 18, 2020 at 6:34 PM Gedare Bloom <gedare at rtems.org> wrote:
> >>
> >> On Wed, Jun 17, 2020 at 11:17 PM Utkarsh Rai <utkarsh.rai60 at gmail.com>
> wrote:
> >> >
> >> > Hello,
> >> > For my GSoC project, I need to set/unset the memory attributes of the
> thread stacks on each context switch.
> >> > Right now I am making changes to the CPU-specific context switch
> assembly code. The arm/../cpu.h file has the Context_Control structure,
> which is used to store the relevant registers at each context switch.
> >> >  I have been trying to add the stack attributes to this structure
> from the 'rtems/score/stackmanagement.h' header (This header has been
> defined by me and has the stack attributes structure).
> >>
> >> When you change the context structure, there might be other places you
> >> need to update besides the context switch itself. One that comes to
> >> mind is _CPU_Context_validate. Maybe chase down where the context is
> >> referenced.
> >>
> >> >  After that, I would initialize the stack attributes in
> _CPU_Context_Initialize.
> >> >  For unsetting stack memories, I plan on making changes to
> _CPU_Context_Switch().
> >> > The problem is when I try to include the header in cpu.h it breaks
> the build with the following error -
> >> >
> >> > '.......
> >> > ........
> >> >
> /home/utkarsh/sandbox/rtems/5/arm-rtems5/include/machine/_default_types.h:
> Assembler messages:
> >> >
> /home/utkarsh/sandbox/rtems/cpukit/include/rtems/score/stackmanagement.h:42:
> Error: junk at end of line, first unrecognized character is `}'
> >> >
> /home/utkarsh/sandbox/rtems/cpukit/include/rtems/score/stackmanagement.h:44:
> Error: bad instruction `typedef struct stack_attr_prot'
> >> >
> /home/utkarsh/sandbox/rtems/cpukit/include/rtems/score/stackmanagement.h:45:
> Error: junk at end of line, first unrecognized character is `{'
> >> >
> /home/utkarsh/sandbox/rtems/cpukit/include/rtems/score/stackmanagement.h:46:
> Error: bad instruction `stack_attr_shared *shared_stacks'
> >> >
> /home/utkarsh/sandbox/rtems/cpukit/include/rtems/score/stackmanagement.h:47:
> Error: bad instruction `stack_attr Base'
> >> >
> /home/utkarsh/sandbox/rtems/cpukit/include/rtems/score/stackmanagement.h:48:
> Error: bad instruction `_bool current_stack'
> >> > {standard input}: Error: invalid operands (*UND* and *ABS* sections)
> for `*' when setting `iter'
> >> > Makefile:11563: recipe for target
> 'score/cpu/arm/arm-context-validate.o' failed
> >> > make[4]: *** [score/cpu/arm/arm-context-validate.o] Error 1
> >> > '
> >> > I have defined the stackmanagement.h header in the cpukit/headers.am
> and this is obviously an assembler error.  Can someone point out what is it
> that I am doing wrong?
> >> >
> >>
> >> When a .h file can be included in both C and ASM, you need to separate
> >> some of it, and disable some of it in ASM scope that doesn't
> >> understand C data types. See, for example,
> >> cpukit/include/rtems/score/percpu.h
> >>
>

Thank you, this solves the problem.


> >> > I also would like to verify if the above way of handling context
> switching is the appropriate way to proceed.
> >>
> >> Seems like the right general idea. We probably need a CPP switch to
> >> enable/disable this stuff though.
> >
> >
> > Can you please elaborate on CPP switch? I am not sure if I am clear
> about this.
>
> Yeah, my reply was a little short. CPP for C Preprocessor. A CPP
> switch is used for conditional compilation of code, something like
>
> #if defined(USE_MMU)
> ... /* some MMU related code */
> #else
> ... /* something without using MMU */
> #endif
>
> This way high-level features can control which code gets compiled into
> the binary image. This works pretty well for coarse-grained features.
> It is not that useful for finer-grained features, like you might not
> want to have USE_MMU_2K_PAGES and USE_MMU_4K_PAGES and
> USE_MMU_4M_PAGES instead it would be better to have USE_MMU and within
> that scope provide some way to distinguish the page sizes. I hope that
> makes some sense.
>
> Another example is ASM. We have #if ASM to control whether a header
> has been included in an assembler source file.
>

Thank you, this makes it clear. Also, if we are using CPP switches for
switching to MMU related code, I wonder what should be the high-level
configuration option that should be provided to the user. A very naive way
can be if the user defines something like-
 #define USE_THREAD_STACK_PROTECTION

and then the places where we need to handle for thread stack protection we
can proceed as you illustrate above.

Gedare
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20200619/e1664d8f/attachment-0001.html>