bpo-46968: Fix faulthandler for Sapphire Rapids Xeon (GH-31789) (GH-31830)

In Linux kernel 5.14 one can dynamically request size of altstacksize
based on hardware capabilities with getauxval(AT_MINSIGSTKSZ).

This changes allows for Python extension's request to Linux kernel
to use AMX_TILE instruction set on Sapphire Rapids Xeon processor
to succeed, unblocking use of the ISA in frameworks.

Introduced HAVE_LINUX_AUXVEC_H in configure.ac and pyconfig.h.in
Used cpython_autoconf:269 docker container to generate configure.

(cherry picked from commit 3b128c0548)

Co-authored-by: Oleksandr Pavlyk <oleksandr.pavlyk@intel.com>

Misc/NEWS.d/next/Library/2022-03-10-14-51-11.bpo-46968.ym2QxL.rst

@@ -0,0 +1,5 @@
+:mod:`faulthandler`: On Linux 5.14 and newer, dynamically determine size of
+signal handler stack size CPython allocates using ``getauxval(AT_MINSIGSTKSZ)``.
+This changes allows for Python extension's request to Linux kernel to use
+AMX_TILE instruction set on Sapphire Rapids Xeon processor to succeed,
+unblocking use of the ISA in frameworks.

Modules/faulthandler.c

@@ -17,6 +17,17 @@
 #  include <sys/resource.h>
 #endif
 
+/* Using an alternative stack requires sigaltstack()
+   and sigaction() SA_ONSTACK */
+#if defined(HAVE_SIGALTSTACK) && defined(HAVE_SIGACTION)
+#  define FAULTHANDLER_USE_ALT_STACK
+#endif
+
+#if defined(FAULTHANDLER_USE_ALT_STACK) && defined(HAVE_LINUX_AUXVEC_H)
+#  include <linux/auxvec.h>
+#  include <sys/auxv.h>
+#endif
+
 /* Allocate at maximum 100 MiB of the stack to raise the stack overflow */
 #define STACK_OVERFLOW_MAX_SIZE (100 * 1024 * 1024)
 
@@ -139,12 +150,6 @@ static fault_handler_t faulthandler_handlers[] = {
 static const size_t faulthandler_nsignals = \
     Py_ARRAY_LENGTH(faulthandler_handlers);
 
-/* Using an alternative stack requires sigaltstack()
-   and sigaction() SA_ONSTACK */
-#if defined(HAVE_SIGALTSTACK) && defined(HAVE_SIGACTION)
-#  define FAULTHANDLER_USE_ALT_STACK
-#endif
-
 #ifdef FAULTHANDLER_USE_ALT_STACK
 static stack_t stack;
 static stack_t old_stack;
@@ -1374,6 +1379,15 @@ _PyFaulthandler_Init(int enable)
        signal handler uses more than SIGSTKSZ bytes of stack memory on some
        platforms. */
     stack.ss_size = SIGSTKSZ * 2;
+#ifdef AT_MINSIGSTKSZ
+    /* bpo-46968: Query Linux for minimal stack size to ensure signal delivery
+       for the hardware running CPython. This OS feature is available in
+       Linux kernel version >= 5.14 */
+    unsigned long at_minstack_size = getauxval(AT_MINSIGSTKSZ);
+    if (at_minstack_size != 0) {
+        stack.ss_size = SIGSTKSZ + at_minstack_size;
+    }
+#endif
 #endif
 
     memset(&thread, 0, sizeof(thread));

configure

@@ -8101,7 +8101,7 @@ sys/stat.h sys/syscall.h sys/sys_domain.h sys/termio.h sys/time.h \
 sys/times.h sys/types.h sys/uio.h sys/un.h sys/utsname.h sys/wait.h pty.h \
 libutil.h sys/resource.h netpacket/packet.h sysexits.h bluetooth.h \
 linux/tipc.h linux/random.h spawn.h util.h alloca.h endian.h \
-sys/endian.h sys/sysmacros.h linux/memfd.h linux/wait.h sys/memfd.h \
+sys/endian.h sys/sysmacros.h linux/auxvec.h linux/memfd.h linux/wait.h sys/memfd.h \
 sys/mman.h sys/eventfd.h
 do :
   as_ac_Header=`$as_echo "ac_cv_header_$ac_header" | $as_tr_sh`

configure.ac

@@ -2225,7 +2225,7 @@ sys/stat.h sys/syscall.h sys/sys_domain.h sys/termio.h sys/time.h \
 sys/times.h sys/types.h sys/uio.h sys/un.h sys/utsname.h sys/wait.h pty.h \
 libutil.h sys/resource.h netpacket/packet.h sysexits.h bluetooth.h \
 linux/tipc.h linux/random.h spawn.h util.h alloca.h endian.h \
-sys/endian.h sys/sysmacros.h linux/memfd.h linux/wait.h sys/memfd.h \
+sys/endian.h sys/sysmacros.h linux/auxvec.h linux/memfd.h linux/wait.h sys/memfd.h \
 sys/mman.h sys/eventfd.h)
 AC_HEADER_DIRENT
 AC_HEADER_MAJOR

pyconfig.h.in

@@ -637,6 +637,9 @@
 /* Define to 1 if you have the `linkat' function. */
 #undef HAVE_LINKAT
 
+/* Define to 1 if you have the <linux/auxvec.h> header file. */
+#undef HAVE_LINUX_AUXVEC_H
+
 /* Define to 1 if you have the <linux/can/bcm.h> header file. */
 #undef HAVE_LINUX_CAN_BCM_H