From: Andy Chiu <tchiu@tenstorrent.com>
To: Nickolai Zeldovich <nickolai@csail.mit.edu>
Cc: Paul Walmsley <pjw@kernel.org>,
Palmer Dabbelt <palmer@dabbelt.com>,
Albert Ou <aou@eecs.berkeley.edu>,
Alexandre Ghiti <alex@ghiti.fr>, Andy Chiu <andybnac@gmail.com>,
Guobin Zhang <guobin.zhang@intel.com>,
Deepak Gupta <debug@rivosinc.com>, Shuah Khan <shuah@kernel.org>,
linux-riscv@lists.infradead.org, linux-kselftest@vger.kernel.org,
linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2 2/2] selftests: riscv: test FP state restore when rt_sigreturn() faults
Date: Thu, 5 Nov 2026 11:59:04 -0600 [thread overview]
Message-ID: <auzD6JgYSnW186Bb@Tao-Chiu's-Mac> (raw)
In-Reply-To: <20261031161826.858908-3-nickolai@csail.mit.edu>
Hi Nickolai,
On Sat, Oct 31, 2026 at 12:18:26PM -0400, Nickolai Zeldovich wrote:
> Add a test for the previous patch: the signal frame sits on a
> userfaultfd page that the handler drops after dirtying the FP
> registers, so rt_sigreturn() sleeps in the middle of copying the FP
> registers from the frame. The interrupted code must get its own
> registers back.
>
> The test is skipped without the D extension or when userfaultfd is not
> available to the caller.
>
The test is failing on qemu, with or without the patch applied.
> Signed-off-by: Nickolai Zeldovich <nickolai@csail.mit.edu>
> ---
> .../selftests/riscv/sigreturn/.gitignore | 1 +
> .../selftests/riscv/sigreturn/Makefile | 5 +-
> .../selftests/riscv/sigreturn/fp_restore.c | 146 ++++++++++++++++++
> 3 files changed, 151 insertions(+), 1 deletion(-)
> create mode 100644 tools/testing/selftests/riscv/sigreturn/fp_restore.c
>
> diff --git a/tools/testing/selftests/riscv/sigreturn/.gitignore b/tools/testing/selftests/riscv/sigreturn/.gitignore
> index 35002b8ae780..8ed282ecb618 100644
> --- a/tools/testing/selftests/riscv/sigreturn/.gitignore
> +++ b/tools/testing/selftests/riscv/sigreturn/.gitignore
> @@ -1 +1,2 @@
> +fp_restore
> sigreturn
> diff --git a/tools/testing/selftests/riscv/sigreturn/Makefile b/tools/testing/selftests/riscv/sigreturn/Makefile
> index 8c77508641f3..ac7ae7a7c2ae 100644
> --- a/tools/testing/selftests/riscv/sigreturn/Makefile
> +++ b/tools/testing/selftests/riscv/sigreturn/Makefile
> @@ -4,9 +4,12 @@
>
> CFLAGS += -I$(top_srcdir)/tools/include
>
> -TEST_GEN_PROGS := sigreturn
> +TEST_GEN_PROGS := sigreturn fp_restore
>
> include ../../lib.mk
>
> $(OUTPUT)/sigreturn: sigreturn.c
> $(CC) -static -o $@ $(CFLAGS) $(LDFLAGS) $^
> +
> +$(OUTPUT)/fp_restore: fp_restore.c
> + $(CC) -static -pthread -o $@ $(CFLAGS) $(LDFLAGS) $^
> diff --git a/tools/testing/selftests/riscv/sigreturn/fp_restore.c b/tools/testing/selftests/riscv/sigreturn/fp_restore.c
> new file mode 100644
> index 000000000000..1a56684c5f28
> --- /dev/null
> +++ b/tools/testing/selftests/riscv/sigreturn/fp_restore.c
> @@ -0,0 +1,146 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Check that rt_sigreturn() restores the FP registers from the signal
> + * frame when the copy of the frame sleeps on a page fault while the
> + * handler's FP state is dirty. The frame is placed on a userfaultfd
> + * page that the handler drops, so the copy faults after f0-f15.
> + */
> +#include <fcntl.h>
> +#include <pthread.h>
> +#include <signal.h>
> +#include <stdint.h>
> +#include <string.h>
> +#include <sys/auxv.h>
> +#include <sys/ioctl.h>
> +#include <sys/mman.h>
> +#include <sys/syscall.h>
> +#include <ucontext.h>
> +#include <unistd.h>
> +#include <asm/hwcap.h>
> +#include <linux/userfaultfd.h>
> +#include "kselftest_harness.h"
> +
> +#define PAGE 4096UL
> +#define STACK_SIZE 65536UL
> +/* the page boundary falls this far into the FP image, after f0-f15 */
> +#define SPLIT 128
> +
> +#define F0_15 "0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15"
> +/* caller-saved only, so the handler's epilogue does not reload them */
> +#define HANDLER_REGS "0, 1, 2, 3, 4, 5, 6, 7, 10, 11, 12, 13, 14, 15"
> +
> +static uint64_t pattern[16], handler_pattern[16];
> +static void *saved;
> +static int placed;
> +
> +static void handler(int sig, siginfo_t *si, void *context)
> +{
> + ucontext_t *uc = context;
> + unsigned long split = (unsigned long)&uc->uc_mcontext.__fpregs + SPLIT;
> +
> + placed = !(split & (PAGE - 1));
> + if (placed) {
> + /* rt_sigreturn() will fault on this page and wait for serve() */
> + memcpy(saved, (void *)split, PAGE);
> + if (madvise((void *)split, PAGE, MADV_DONTNEED))
> + _exit(1);
> + }
> + asm volatile(".irp i, " HANDLER_REGS "\n"
> + "fld f\\i, \\i * 8(%[h])\n"
> + ".endr\n"
> + : : [h] "r" (handler_pattern)
> + : "memory", "ft0", "ft1", "ft2", "ft3", "ft4", "ft5",
> + "ft6", "ft7", "fa0", "fa1", "fa2", "fa3", "fa4", "fa5");
> +}
> +
> +static void *serve(void *arg)
> +{
> + struct uffdio_copy copy = { .src = (unsigned long)saved, .len = PAGE };
> + struct uffd_msg msg;
> + int fd = (long)arg;
> +
> + while (read(fd, &msg, sizeof(msg)) == sizeof(msg)) {
> + copy.dst = msg.arg.pagefault.address & ~(PAGE - 1);
> + copy.copy = 0;
> + ioctl(fd, UFFDIO_COPY, ©);
> + }
> + return NULL;
> +}
> +
> +TEST(fp_restore_frame_fault)
> +{
> + struct sigaction sa = {
> + .sa_sigaction = handler,
> + .sa_flags = SA_SIGINFO | SA_ONSTACK,
> + };
> + struct uffdio_api api = { .api = UFFD_API };
> + struct uffdio_register reg = { .mode = UFFDIO_REGISTER_MODE_MISSING };
> + stack_t ss = { .ss_size = STACK_SIZE };
> + long pid = getpid(), tid = syscall(__NR_gettid);
> + uint64_t out[16];
> + unsigned long fcsr, shift;
> + pthread_t thread;
> + int fd, i;
> +
> + if (!(getauxval(AT_HWCAP) & COMPAT_HWCAP_ISA_D))
> + SKIP(return, "no D extension");
> +
> + fd = syscall(__NR_userfaultfd, O_CLOEXEC);
> + if (fd < 0)
> + SKIP(return, "userfaultfd not available");
> + ASSERT_EQ(0, ioctl(fd, UFFDIO_API, &api));
> +
> + for (i = 0; i < 16; i++) {
> + pattern[i] = 0x4010000000000000ULL + i;
> + handler_pattern[i] = 0xbff0000000000000ULL + i;
> + }
> +
> + saved = mmap(NULL, PAGE, PROT_READ | PROT_WRITE,
> + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
> + ASSERT_NE(MAP_FAILED, saved);
> + ss.ss_sp = mmap(NULL, STACK_SIZE, PROT_READ | PROT_WRITE,
> + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
> + ASSERT_NE(MAP_FAILED, ss.ss_sp);
> + /* populate it, so only the page the handler drops faults */
> + memset(ss.ss_sp, 0, STACK_SIZE);
> + reg.range.start = (unsigned long)ss.ss_sp;
> + reg.range.len = STACK_SIZE;
> + ASSERT_EQ(0, ioctl(fd, UFFDIO_REGISTER, ®));
> + ASSERT_EQ(0, pthread_create(&thread, NULL, serve, (void *)(long)fd));
> + ASSERT_EQ(0, sigaction(SIGUSR1, &sa, NULL));
> +
> + /* lower the frame until the FP image crosses a page at SPLIT */
> + for (shift = 0; shift < PAGE && !placed; shift += 16) {
> + register long a0 asm("a0") = pid;
> + register long a1 asm("a1") = tid;
> + register long a2 asm("a2") = SIGUSR1;
> + register long a7 asm("a7") = __NR_tgkill;
> +
> + ss.ss_size = STACK_SIZE - shift;
> + ASSERT_EQ(0, sigaltstack(&ss, NULL));
> + fcsr = 0x1f;
> + asm volatile(".irp i, " F0_15 "\n"
> + "fld f\\i, \\i * 8(%[in])\n"
> + ".endr\n"
> + "csrw fcsr, %[fcsr]\n"
> + "ecall\n"
> + "csrr %[fcsr], fcsr\n"
> + ".irp i, " F0_15 "\n"
> + "fsd f\\i, \\i * 8(%[out])\n"
> + ".endr\n"
> + : "+r" (a0), [fcsr] "+r" (fcsr)
> + : "r" (a1), "r" (a2), "r" (a7),
> + [in] "r" (pattern), [out] "r" (out)
> + : "memory", "ft0", "ft1", "ft2", "ft3", "ft4", "ft5",
> + "ft6", "ft7", "fs0", "fs1", "fa0", "fa1", "fa2",
> + "fa3", "fa4", "fa5");
> + }
> + ASSERT_TRUE(placed);
> +
> + for (i = 0; i < 16; i++) {
> + EXPECT_EQ(pattern[i], out[i]) TH_LOG("f%d", i);
> + }
> + EXPECT_EQ(0x1fUL, fcsr);
> +}
> +
> +TEST_HARNESS_MAIN
> --
> 2.53.0
>
prev parent reply other threads:[~2026-11-05 17:59 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-10-31 16:18 [PATCH v2 0/2] riscv: signal: fix the FP state restore in rt_sigreturn() Nickolai Zeldovich
2026-10-31 16:18 ` [PATCH v2 1/2] riscv: signal: mark the FP state clean before restoring it from the frame Nickolai Zeldovich
2026-10-31 16:18 ` [PATCH v2 2/2] selftests: riscv: test FP state restore when rt_sigreturn() faults Nickolai Zeldovich
2026-11-05 17:59 ` Andy Chiu [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='auzD6JgYSnW186Bb@Tao-Chiu'\''s-Mac' \
--to=tchiu@tenstorrent.com \
--cc=alex@ghiti.fr \
--cc=andybnac@gmail.com \
--cc=aou@eecs.berkeley.edu \
--cc=debug@rivosinc.com \
--cc=guobin.zhang@intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-kselftest@vger.kernel.org \
--cc=linux-riscv@lists.infradead.org \
--cc=nickolai@csail.mit.edu \
--cc=palmer@dabbelt.com \
--cc=pjw@kernel.org \
--cc=shuah@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®