From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-lf1-f67.google.com (mail-lf1-f67.google.com [209.85.167.67]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by dev.tarantool.org (Postfix) with ESMTPS id AB207469719 for ; Thu, 15 Oct 2020 19:28:09 +0300 (MSK) Received: by mail-lf1-f67.google.com with SMTP id c141so4310462lfg.5 for ; Thu, 15 Oct 2020 09:28:09 -0700 (PDT) Date: Thu, 15 Oct 2020 19:28:07 +0300 From: Cyrill Gorcunov Message-ID: <20201015162807.GF173841@grain> References: <20201015142827.386709-1-gorcunov@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20201015142827.386709-1-gorcunov@gmail.com> Subject: Re: [Tarantool-patches] [PATCH] test/uint: fiber List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: tml On Thu, Oct 15, 2020 at 05:28:27PM +0300, Cyrill Gorcunov wrote: > When optimization level increased we've found that > memset calls might be optimized, moreover there > were no check if calls themselves would override > the stack. > > Lets rework and use number of calls instead. Strictly > speaking the test is still a bit fragile and precise > testing would rather require asm level code. > > Repored-by: "Alexander V. Tikhonov" > Signed-off-by: Cyrill Gorcunov > --- > > branch gorcunov/unit-fiber I've force update the patch to match results file. --- From: Cyrill Gorcunov Date: Thu, 15 Oct 2020 17:25:55 +0300 Subject: [PATCH] test/uint: fiber When optimization level increased we've found that memset calls might be optimized, moreover there were no check if calls themselves would override the stack. Lets rework and use number of calls instead. Strictly speaking the test is still a bit fragile and precise testing would rather require asm level code. Repored-by: "Alexander V. Tikhonov" Signed-off-by: Cyrill Gorcunov --- test/unit/fiber.cc | 47 +++++++++++++++++++++++++++++++++++----------- 1 file changed, 36 insertions(+), 11 deletions(-) diff --git a/test/unit/fiber.cc b/test/unit/fiber.cc index b0dfc9b14..9c1a23bdd 100644 --- a/test/unit/fiber.cc +++ b/test/unit/fiber.cc @@ -1,10 +1,16 @@ +#include + #include "memory.h" #include "fiber.h" #include "unit.h" #include "trivia/util.h" +static size_t stack_expand_limit; static struct fiber_attr default_attr; +static unsigned long page_size; +#define PAGE_4K 4096 + static int noop_f(va_list ap) { @@ -49,24 +55,38 @@ cancel_dead_f(va_list ap) return 0; } -static size_t stack_expand_limit; - static void NOINLINE -stack_expand(void *ptr) +stack_expand(unsigned long *ret, unsigned long nr_calls) { - char buf[2048]; - memset(buf, 0x45, 2048); - ptrdiff_t stack_diff = (buf - (char *)ptr); - stack_diff = stack_diff >= 0 ? stack_diff : -stack_diff; - if (stack_diff < (ptrdiff_t)stack_expand_limit) - stack_expand(ptr); + char volatile fill[PAGE_4K]; + char volatile *p; + + memset((void *)fill, (unsigned char)nr_calls, sizeof(fill)); + p = fill; + p[PAGE_4K / 2] = (unsigned char)nr_calls; + + if (nr_calls != 0) { + stack_expand(ret, nr_calls-1); + } else { + *ret = (unsigned long)&fill[0]; + } } static int test_stack_f(va_list ap) { - char s; - stack_expand(&s); + unsigned long ret = 0; + unsigned long ret_addr = (unsigned long)&ret; + + /* + * We can't just dirty the stack in precise + * way without using assembly. Thus lets do + * the following trick: + * - assume 8K will be enough to carry all + * arguments passed for all calls, still + * we might need to adjust this value + */ + stack_expand(&ret, (stack_expand_limit - 2 * page_size) / page_size); return 0; } @@ -193,6 +213,11 @@ main_f(va_list ap) int main() { + page_size = sysconf(_SC_PAGESIZE); + + /* Page should be at least 4K */ + assert(page_size >= PAGE_4K); + memory_init(); fiber_init(fiber_cxx_invoke); fiber_attr_create(&default_attr); -- 2.26.2