The x86 backend defines a PROBE_INTERVAL which is supposed to be used by the -fstack-check= mechanisms.
Some stack-clash code was using PROBE_INTERVAL rather than querying the PARAM system for the right value. If the former is larger than the latter and we allocate a large stack, then the loop to probe the stack space may not terminate (PR82788)
Rather than playing wack-a-mole on this problem I decided to just create a little helper that would return the right probing interval for whatever option was active, then changed all the x86 code to use that new function. The patch is mostly a search/replace and looks much bigger than it really is.
Bootstrapped and regression tested on x86_64. OK for the trunk? Jeff
commit 2939aa3d6f49bd2b4376f23ebbff4bea4c9afa12 Author: Jeff Law <l...@tor.usersys.redhat.com> Date: Thu Nov 2 20:15:11 2017 -0400 FIx bz diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c index fc43962..672a085 100644 --- a/gcc/config/i386/i386.c +++ b/gcc/config/i386/i386.c @@ -12083,7 +12083,17 @@ release_scratch_register_on_entry (struct scratch_reg *sr) } } -#define PROBE_INTERVAL (1 << STACK_CHECK_PROBE_INTERVAL_EXP) +/* Return the probing interval for -fstack-clash-protection. */ + +static HOST_WIDE_INT +get_probe_interval (void) +{ + if (flag_stack_clash_protection) + return (HOST_WIDE_INT_1U + << PARAM_VALUE (PARAM_STACK_CLASH_PROTECTION_PROBE_INTERVAL)); + else + return (HOST_WIDE_INT_1U << STACK_CHECK_PROBE_INTERVAL_EXP); +} /* Emit code to adjust the stack pointer by SIZE bytes while probing it. @@ -12147,8 +12157,7 @@ ix86_adjust_stack_and_probe_stack_clash (const HOST_WIDE_INT size) /* We're allocating a large enough stack frame that we need to emit probes. Either emit them inline or in a loop depending on the size. */ - HOST_WIDE_INT probe_interval - = 1 << PARAM_VALUE (PARAM_STACK_CLASH_PROTECTION_PROBE_INTERVAL); + HOST_WIDE_INT probe_interval = get_probe_interval (); if (size <= 4 * probe_interval) { HOST_WIDE_INT i; @@ -12157,7 +12166,7 @@ ix86_adjust_stack_and_probe_stack_clash (const HOST_WIDE_INT size) /* Allocate PROBE_INTERVAL bytes. */ rtx insn = pro_epilogue_adjust_stack (stack_pointer_rtx, stack_pointer_rtx, - GEN_INT (-PROBE_INTERVAL), -1, + GEN_INT (-probe_interval), -1, m->fs.cfa_reg == stack_pointer_rtx); add_reg_note (insn, REG_STACK_CHECK, const0_rtx); @@ -12250,7 +12259,7 @@ ix86_adjust_stack_and_probe (const HOST_WIDE_INT size) that's the easy case. The run-time loop is made up of 9 insns in the generic case while the compile-time loop is made up of 3+2*(n-1) insns for n # of intervals. */ - if (size <= 4 * PROBE_INTERVAL) + if (size <= 4 * get_probe_interval ()) { HOST_WIDE_INT i, adjust; bool first_probe = true; @@ -12259,15 +12268,15 @@ ix86_adjust_stack_and_probe (const HOST_WIDE_INT size) values of N from 1 until it exceeds SIZE. If only one probe is needed, this will not generate any code. Then adjust and probe to PROBE_INTERVAL + SIZE. */ - for (i = PROBE_INTERVAL; i < size; i += PROBE_INTERVAL) + for (i = get_probe_interval (); i < size; i += get_probe_interval ()) { if (first_probe) { - adjust = 2 * PROBE_INTERVAL + dope; + adjust = 2 * get_probe_interval () + dope; first_probe = false; } else - adjust = PROBE_INTERVAL; + adjust = get_probe_interval (); emit_insn (gen_rtx_SET (stack_pointer_rtx, plus_constant (Pmode, stack_pointer_rtx, @@ -12276,9 +12285,9 @@ ix86_adjust_stack_and_probe (const HOST_WIDE_INT size) } if (first_probe) - adjust = size + PROBE_INTERVAL + dope; + adjust = size + get_probe_interval () + dope; else - adjust = size + PROBE_INTERVAL - i; + adjust = size + get_probe_interval () - i; emit_insn (gen_rtx_SET (stack_pointer_rtx, plus_constant (Pmode, stack_pointer_rtx, @@ -12288,7 +12297,8 @@ ix86_adjust_stack_and_probe (const HOST_WIDE_INT size) /* Adjust back to account for the additional first interval. */ last = emit_insn (gen_rtx_SET (stack_pointer_rtx, plus_constant (Pmode, stack_pointer_rtx, - PROBE_INTERVAL + dope))); + (get_probe_interval () + + dope)))); } /* Otherwise, do the same as above, but in a loop. Note that we must be @@ -12306,7 +12316,7 @@ ix86_adjust_stack_and_probe (const HOST_WIDE_INT size) /* Step 1: round SIZE to the previous multiple of the interval. */ - rounded_size = ROUND_DOWN (size, PROBE_INTERVAL); + rounded_size = ROUND_DOWN (size, get_probe_interval ()); /* Step 2: compute initial and final value of the loop counter. */ @@ -12314,7 +12324,7 @@ ix86_adjust_stack_and_probe (const HOST_WIDE_INT size) /* SP = SP_0 + PROBE_INTERVAL. */ emit_insn (gen_rtx_SET (stack_pointer_rtx, plus_constant (Pmode, stack_pointer_rtx, - - (PROBE_INTERVAL + dope)))); + - (get_probe_interval ()+ dope)))); /* LAST_ADDR = SP_0 + PROBE_INTERVAL + ROUNDED_SIZE. */ if (rounded_size <= (HOST_WIDE_INT_1 << 31)) @@ -12359,7 +12369,8 @@ ix86_adjust_stack_and_probe (const HOST_WIDE_INT size) /* Adjust back to account for the additional first interval. */ last = emit_insn (gen_rtx_SET (stack_pointer_rtx, plus_constant (Pmode, stack_pointer_rtx, - PROBE_INTERVAL + dope))); + (get_probe_interval () + + dope)))); release_scratch_register_on_entry (&sr); } @@ -12376,7 +12387,7 @@ ix86_adjust_stack_and_probe (const HOST_WIDE_INT size) XVECEXP (expr, 0, 1) = gen_rtx_SET (stack_pointer_rtx, plus_constant (Pmode, stack_pointer_rtx, - PROBE_INTERVAL + dope + size)); + get_probe_interval () + dope + size)); add_reg_note (last, REG_FRAME_RELATED_EXPR, expr); RTX_FRAME_RELATED_P (last) = 1; @@ -12403,7 +12414,7 @@ output_adjust_stack_and_probe (rtx reg) /* SP = SP + PROBE_INTERVAL. */ xops[0] = stack_pointer_rtx; - xops[1] = GEN_INT (PROBE_INTERVAL); + xops[1] = GEN_INT (get_probe_interval ()); output_asm_insn ("sub%z0\t{%1, %0|%0, %1}", xops); /* Probe at SP. */ @@ -12433,14 +12444,14 @@ ix86_emit_probe_stack_range (HOST_WIDE_INT first, HOST_WIDE_INT size) that's the easy case. The run-time loop is made up of 6 insns in the generic case while the compile-time loop is made up of n insns for n # of intervals. */ - if (size <= 6 * PROBE_INTERVAL) + if (size <= 6 * get_probe_interval ()) { HOST_WIDE_INT i; /* Probe at FIRST + N * PROBE_INTERVAL for values of N from 1 until it exceeds SIZE. If only one probe is needed, this will not generate any code. Then probe at FIRST + SIZE. */ - for (i = PROBE_INTERVAL; i < size; i += PROBE_INTERVAL) + for (i = get_probe_interval (); i < size; i += get_probe_interval ()) emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, -(first + i))); @@ -12463,7 +12474,7 @@ ix86_emit_probe_stack_range (HOST_WIDE_INT first, HOST_WIDE_INT size) /* Step 1: round SIZE to the previous multiple of the interval. */ - rounded_size = ROUND_DOWN (size, PROBE_INTERVAL); + rounded_size = ROUND_DOWN (size, get_probe_interval ()); /* Step 2: compute initial and final value of the loop counter. */ @@ -12524,7 +12535,7 @@ output_probe_stack_range (rtx reg, rtx end) /* TEST_ADDR = TEST_ADDR + PROBE_INTERVAL. */ xops[0] = reg; - xops[1] = GEN_INT (PROBE_INTERVAL); + xops[1] = GEN_INT (get_probe_interval ()); output_asm_insn ("sub%z0\t{%1, %0|%0, %1}", xops); /* Probe at TEST_ADDR. */ @@ -13182,7 +13193,7 @@ ix86_expand_prologue (void) else if (STACK_CHECK_MOVING_SP) { if (!(crtl->is_leaf && !cfun->calls_alloca - && allocate <= PROBE_INTERVAL)) + && allocate <= get_probe_interval ())) { ix86_adjust_stack_and_probe (allocate); allocate = 0; @@ -13199,7 +13210,7 @@ ix86_expand_prologue (void) { if (crtl->is_leaf && !cfun->calls_alloca) { - if (size > PROBE_INTERVAL) + if (size > get_probe_interval ()) ix86_emit_probe_stack_range (0, size); } else @@ -13210,7 +13221,7 @@ ix86_expand_prologue (void) { if (crtl->is_leaf && !cfun->calls_alloca) { - if (size > PROBE_INTERVAL + if (size > get_probe_interval () && size > get_stack_check_protect ()) ix86_emit_probe_stack_range (get_stack_check_protect (), size - get_stack_check_protect ()); diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index ea0c0e0..f0eda16 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,6 +1,8 @@ 2017-11-02 Jeff Law <l...@redhat.com> - * gcc.target/i386/stack-check-12.c: New test + * gcc.c-torture/execute/pr82788.c: New test. + + * gcc.target/i386/stack-check-12.c: New test. 2017-11-02 Nathan Sidwell <nat...@acm.org> diff --git a/gcc/testsuite/gcc.c-torture/execute/pr82788.c b/gcc/testsuite/gcc.c-torture/execute/pr82788.c new file mode 100644 index 0000000..ceaa25f --- /dev/null +++ b/gcc/testsuite/gcc.c-torture/execute/pr82788.c @@ -0,0 +1,2 @@ + +int main() { int a[1442]; return 0;}