Look at msp430-builtins.c in the implementation for delay cycles for examples of generating local labels, comparison, and jump instructions. You can probably use something like gen_rtx_MEM (Pmode, gen_rtx_SYMBOL_REF (HImode, "__stack_base")) and use that to compare against stack_pointer_rtx rather than having to push and pop r15.
The ICEs are likely if you don't set RTX_FRAME_RELATED_P on the prolog insns that access the stack pointer. If you get it working solidly, please file a tracker ticket on the mspgcc project and submit a patch for review and incorporation into the development branches. The final version should set a new frame flag in msp430_set_current_function, and implement it in expand_prologue, probably right before MSP430_FF_prologue_eint. You might also want to look at what -fstack-check really does and whether what you describe is compatible with its use on other platforms. A quick check suggests it's mostly implemented in the middle-end and may result in code generated that depends on some runtime support for stack probes. -mstack-check may be more appropriate for the approach you describe. Peter On Wed, Aug 24, 2011 at 3:27 PM, Radu Rendec <radu.ren...@cloudbit.ro> wrote: > Hello, > > I'm trying to implement a simple stack checking mechanism in an MSP430 > multi-task environment (task scheduling/switching is subject to a custom > "OS"). > > Each task has its own stack space, but since there's a single address > space and the memory is small, tasks can easily overflow their stack and > corrupt "foreign" data. > > The best idea I came up with for checking the overflow is adding some > explicit checking code at the beginning of each function - actually at > the end of the preamble, when SP is already lowered to reserve space for > local variables. > > It would be very nice though if the compiler could generate that code > itself, for instance when the "-fstack-check" is specified on the > command line. > > I tracked down the preamble generation to the expand_prologue() function > (file msp430-function.c). The global variable "flag_stack_check" is > visible and automatically set to 1 by (generic) gcc when "-fstack-check" > is specified on the command line. > > The best I could do was identify the code that generates "push" > instructions for all clobbered registers and the code that explicitly > lowers SP by the required amount. My code goes next to this. > > However, I'm new to gcc internals and most of my attempts to produce the > check code ended up with gcc reporting an internal bug and bailing out. > > I would really appreciate any suggestions or directions for implementing > this. The resulting code should be as simple as comparing R1 to a > variable (contents of a fixed memory location) that holds the stack base > address and conditionally jump to a handling routine. To my (limited) > MSP430 asm knowledge, it shoud look like this: > push r15 > mov #__stack_base, r15 > cmp r1, 0(r15) > jge .Lsomething > br #_handling_routine > .Lsomething > pop r15 > > Best regards, > > Radu Rendec > > > > ------------------------------------------------------------------------------ > EMC VNX: the world's simplest storage, starting under $10K > The only unified storage solution that offers unified management > Up to 160% more powerful than alternatives and 25% more efficient. > Guaranteed. http://p.sf.net/sfu/emc-vnx-dev2dev > _______________________________________________ > Mspgcc-users mailing list > Mspgcc-users@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/mspgcc-users > ------------------------------------------------------------------------------ EMC VNX: the world's simplest storage, starting under $10K The only unified storage solution that offers unified management Up to 160% more powerful than alternatives and 25% more efficient. Guaranteed. http://p.sf.net/sfu/emc-vnx-dev2dev _______________________________________________ Mspgcc-users mailing list Mspgcc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mspgcc-users
