linux-um.lists.infradead.org archive mirror
 help / color / mirror / Atom feed
* [uml-devel] [PATCH v3 0/3] um: fix and extend FPU support
@ 2016-03-19 16:58 Eli Cooper
  2016-03-19 16:58 ` [uml-devel] [PATCH v3 1/3] um: fix FPU state preservation around signal handlers Eli Cooper
                   ` (3 more replies)
  0 siblings, 4 replies; 6+ messages in thread
From: Eli Cooper @ 2016-03-19 16:58 UTC (permalink / raw)
  To: user-mode-linux-devel; +Cc: Richard Weinberger, Jeff Dike

This series first fixes a bug that results in corrupted FPU state after
invoking signal handlers. It also adds support for the extended processor
state (XSTATE) for x86_64 UML, especially the YMM registers used by AVX(2)
instructions.

Tested with a minimal multi-threaded FPU-intensive test program (see below).
This series supersedes the previous sigreturn fix as that one is incorrect
when the process is multi-threaded.

Changes since v2:
 - Add an improved sigreturn fix to this series
 - Merge the ptrace changes into the last commit
 - Make the selftest program multi-threaded

Changes since v1:
 - Refactor functions with oversized stack frame
 - Add a tiny selftest program to the cover letter

Eli Cooper (3):
  um: fix FPU state preservation around signal handlers
  um: extend fpstate to _xstate to support YMM registers
  um: add extended processor state save/restore support

 arch/um/include/shared/registers.h    |  2 ++
 arch/um/kernel/process.c              |  2 +-
 arch/um/os-Linux/signal.c             | 28 ++++++++++++++------
 arch/x86/um/os-Linux/registers.c      | 49 +++++++++++++++++++++++++++++++++--
 arch/x86/um/ptrace_32.c               |  5 ++--
 arch/x86/um/ptrace_64.c               | 16 ++++++------
 arch/x86/um/shared/sysdep/ptrace_64.h |  4 +--
 arch/x86/um/signal.c                  | 37 +++++++++-----------------
 arch/x86/um/user-offsets.c            |  2 +-
 9 files changed, 95 insertions(+), 50 deletions(-)

--
/* Test if context switches preserve YMM registers, multi-threaded version
 * The main function, threads and the signal handler all have their unique
 * ymm0 value, and constantly detect if someone steps on their toes.
 * Should loop forever.
 */
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <signal.h>
#include <unistd.h>
#include <pthread.h>
#include <sys/time.h>
#include <immintrin.h>

#define N 10

void sighandler(int signum)
{
	int n = 0xffff;
	register uint32_t eax asm("eax");

	__m256i m0 = _mm256_set_epi64x(0x1234L << 32, 0, 0, 0);
	while (n--) {
		asm("vextracti128 $1,%ymm0,%xmm1");
		asm("vpextrd $3,%xmm1,%eax");
		if (eax != 0x1234)
			exit(3);
	}
}

void thread(void *arg)
{
	long n = (long)arg;
	register uint32_t eax asm("eax");

	__m256i m0 = _mm256_set_epi64x(n << 32, 0, 0, 0);
	do {
		asm("vextracti128 $1,%ymm0,%xmm1");
		asm("vpextrd $3,%xmm1,%eax");
	} while (eax == n);
	exit(2);
}

int main()
{
	register uint32_t eax asm("eax");
	pthread_t threads[N];
	struct itimerval itv;
	struct timeval tv;
	struct sigaction act;

	tv.tv_sec = 0;
	tv.tv_usec = 100000;
	itv.it_interval = tv;
	itv.it_value = tv;

	act.sa_handler = sighandler;
	act.sa_flags = 0;
	sigemptyset(&act.sa_mask);
	sigaction(SIGALRM, &act, NULL);

	setitimer(ITIMER_REAL, &itv, NULL);

	for (long i = 0; i < N; i++)
		pthread_create(threads + i, NULL, (void *)thread, (void *)i);

	__m256i m0 = _mm256_set_epi64x(0xabcdL << 32, 0, 0, 0);
	do {
		asm("vextracti128 $1,%ymm0,%xmm1");
		asm("vpextrd $3,%xmm1,%eax");
	} while (eax == 0xabcd);
	printf("%lx\n", eax);

	return 1;
}

------------------------------------------------------------------------------
Transform Data into Opportunity.
Accelerate data analysis in your applications with
Intel Data Analytics Acceleration Library.
Click to learn more.
http://pubads.g.doubleclick.net/gampad/clk?id=278785231&iu=/4140
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel


^ permalink raw reply	[flat|nested] 6+ messages in thread
end of thread, other threads:[~2016-05-20 15:29 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2016-03-19 16:58 [uml-devel] [PATCH v3 0/3] um: fix and extend FPU support Eli Cooper
2016-03-19 16:58 ` [uml-devel] [PATCH v3 1/3] um: fix FPU state preservation around signal handlers Eli Cooper
2016-03-19 16:58 ` [uml-devel] [PATCH v3 2/3] um: extend fpstate to _xstate to support YMM registers Eli Cooper
2016-03-19 16:58 ` [uml-devel] [PATCH v3 3/3] um: add extended processor state save/restore support Eli Cooper
2016-04-04 21:42 ` [uml-devel] [PATCH v3 0/3] um: fix and extend FPU support Richard Weinberger
2016-05-20 15:31   ` Eli Cooper

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).