>The call is missing from the trap handler, probably because I was >looking at arm64 where it is missing as well. The result is that the >stack size accounting will be wrong.
Nice find. >In the diff below I only added the call to the "data" trap. That >means that an "instruction" trap will not run the accounting code. Is >that correct? The uvm_fault() call should never return success in >that case unless the stack has been mapped executable... I think both should have it. munmap and mprotect exist, and people can do strange things. >Index: arch/powerpc64/powerpc64/trap.c >=================================================================== >RCS file: /cvs/src/sys/arch/powerpc64/powerpc64/trap.c,v >retrieving revision 1.39 >diff -u -p -r1.39 trap.c >--- arch/powerpc64/powerpc64/trap.c 24 Sep 2020 20:22:15 -0000 1.39 >+++ arch/powerpc64/powerpc64/trap.c 24 Sep 2020 21:11:08 -0000 >@@ -181,6 +181,8 @@ trap(struct trapframe *frame) > ftype = PROT_READ; > KERNEL_LOCK(); > error = uvm_fault(map, trunc_page(va), 0, ftype); >+ if (error == 0) >+ uvm_grow(p, trunc_page(va)); > KERNEL_UNLOCK(); > if (error) { > #ifdef TRAP_DEBUG > >