From 0643c69caf4d4a138a21529081ab3084963da536 Mon Sep 17 00:00:00 2001
From: Hendrik Greving <hgreving@google.com>
Date: Mon, 29 Jul 2019 11:59:27 -0700
Subject: [PATCH 1/7] i#3756: Fix copying signal magic1 value.
Fixes copying the current state when delivering pending signals by preserving the original
magic1 value of the signal's fpstate.
Re-factor AVX detection in test linux.sigcontext and add support to actually test the AVX
state on supported machines.
Adds new <test>_runavx and <test>_runavx512 variables to the cmake suite similar to
_runcmp, in order to propagate the __AVX__ and __AVX512F__ defines to the template
parsers which are usually automatically defined by the compiler.
Updates a comment w.r.t. alignment of a stack variable we're declaring.
Fixes #3756
---
core/unix/signal_linux_x86.c | 16 +++++++-----
suite/tests/CMakeLists.txt | 12 +++++++++
suite/tests/linux/sigcontext.c | 35 ++++++---------------------
suite/tests/linux/sigcontext.template | 32 ++++++++++++++++--------
4 files changed, 52 insertions(+), 43 deletions(-)
diff --git a/core/unix/signal_linux_x86.c b/core/unix/signal_linux_x86.c
index 035e14489..2ad44ebad 100644
--- a/core/unix/signal_linux_x86.c
+++ b/core/unix/signal_linux_x86.c
@@ -179,7 +179,8 @@ convert_fxsave_to_fpstate(kernel_fpstate_t *fpstate, struct i387_fxsave_struct *
fpstate->status = fxsave->swd;
fpstate->magic = X86_FXSR_MAGIC;
- memcpy(&fpstate->_fxsr_env[0], fxsave, sizeof(struct i387_fxsave_struct));
+ memcpy(&fpstate->_fxsr_env[0], fxsave,
+ sizeof(struct i387_fxsave_struct) - sizeof(fpstate->sw_reserved));
}
#endif /* !X64 */
@@ -258,13 +259,14 @@ save_xmm(dcontext_t *dcontext, sigframe_rt_t *frame)
void
save_fpstate(dcontext_t *dcontext, sigframe_rt_t *frame)
{
- /* FIXME: is there a better way to align this thing?
- * the __attribute__ on the struct above doesn't do it
+ /* The compiler may not be able to properly align stack variables even with
+ * __attribute__((aligned()). We maintain this array to enforce alignment.
*/
- char align[sizeof(union i387_union) + 16];
+ char align[sizeof(union i387_union) + 16] __attribute__((aligned(16)));
union i387_union *temp =
(union i387_union *)((((ptr_uint_t)align) + 16) & ((ptr_uint_t)-16));
sigcontext_t *sc = get_sigcontext_from_rt_frame(frame);
+
LOG(THREAD, LOG_ASYNCH, 3, "save_fpstate\n");
if (sc->fpstate == NULL) {
/* Nothing to do: there was no fpstate to save at the time the kernel
@@ -285,7 +287,8 @@ save_fpstate(dcontext_t *dcontext, sigframe_rt_t *frame)
asm volatile("fxsaveq %0 ; fnclex" : "=m"(temp->fxsave));
/* now convert into kernel_fpstate_t form */
ASSERT(sizeof(kernel_fpstate_t) == sizeof(struct i387_fxsave_struct));
- memcpy(sc->fpstate, &temp->fxsave, sizeof(struct i387_fxsave_struct));
+ memcpy(sc->fpstate, &temp->fxsave,
+ sizeof(struct i387_fxsave_struct) - sizeof(sc->fpstate->sw_reserved));
#else
/* this is "unlazy_fpu" */
asm volatile("fxsave %0 ; fnclex" : "=m"(temp->fxsave));
@@ -299,7 +302,8 @@ save_fpstate(dcontext_t *dcontext, sigframe_rt_t *frame)
asm volatile("fnsave %0 ; fwait" : "=m"(temp->fsave));
/* now convert into kernel_fpstate_t form */
temp->fsave.status = temp->fsave.swd;
- memcpy(sc->fpstate, &temp->fsave, sizeof(struct i387_fsave_struct));
+ memcpy(sc->fpstate, &temp->fsave,
+ sizeof(struct i387_fsave_struct) - sizeof(sc->fpstate->sw_reserved));
}
save_xmm(dcontext, frame);
diff --git a/suite/tests/CMakeLists.txt b/suite/tests/CMakeLists.txt
index d69274bf3..33ad80972 100644
--- a/suite/tests/CMakeLists.txt
+++ b/suite/tests/CMakeLists.txt
@@ -1031,6 +1031,11 @@ function(template2expect outexpect template runops key)
# so currently this is unused.
set(rundefs "${rundefs} -DCYGWIN")
endif ()
+ if (DEFINED ${key}_runavx512)
+ set(rundefs "${rundefs} -D__AVX__ -D__AVX512F__")
+ elseif (DEFINED ${key}_runavx)
+ set(rundefs "${rundefs} -D__AVX__")
+ endif ()
string(STRIP "${rundefs}" rundefs)
@@ -3319,6 +3324,13 @@ if (UNIX)
# FIXME i#58: port test to MacOS
if (NOT ARM AND NOT AARCH64 AND NOT APPLE)
tobuild(linux.sigcontext linux/sigcontext.c)
+ if (proc_supports_avx512)
+ append_property_string(TARGET linux.sigcontext COMPILE_FLAGS "${CFLAGS_AVX512}")
+ set(linux.sigcontext_runavx512 1)
+ elseif (proc_supports_avx)
+ append_property_string(TARGET linux.sigcontext COMPILE_FLAGS "${CFLAGS_AVX}")
+ set(linux.sigcontext_runavx 1)
+ endif ()
endif ()
if (NOT APPLE)
tobuild(linux.thread linux/thread.c)
diff --git a/suite/tests/linux/sigcontext.c b/suite/tests/linux/sigcontext.c
index e768d5e5a..dc1aa8bb3 100644
--- a/suite/tests/linux/sigcontext.c
+++ b/suite/tests/linux/sigcontext.c
@@ -97,9 +97,8 @@ signal_handler(int sig, siginfo_t *siginfo, ucontext_t *ucxt)
kernel_xstate_t *xstate = (kernel_xstate_t *)ucxt->uc_mcontext.fpregs;
if (xstate->fpstate.sw_reserved.magic1 == FP_XSTATE_MAGIC1) {
assert(xstate->fpstate.sw_reserved.xstate_size >= sizeof(*xstate));
- /* we can't print b/c not all processors have avx */
for (i = 0; i < NUM_SIMD_REGS; i++) {
-#if VERBOSE
+#ifdef __AVX__
print("ymmh[%d] = 0x%x 0x%x 0x%x 0x%x\n", i,
xstate->ymmh.ymmh_space[i * 4],
xstate->ymmh.ymmh_space[i * 4 + 1],
@@ -117,28 +116,6 @@ signal_handler(int sig, siginfo_t *siginfo, ucontext_t *ucxt)
}
}
-/* looks for "avx" flag in /proc/cpuinfo */
-static int
-determine_avx(void)
-{
- FILE *cpuinfo;
- char line[1024];
- ulong cpu_mhz = 1, cpu_khz = 0;
-
- cpuinfo = fopen("/proc/cpuinfo", "r");
- assert(cpuinfo != NULL);
-
- while (!feof(cpuinfo)) {
- if (fgets(line, sizeof(line), cpuinfo) == NULL)
- break;
- if (strstr(line, "flags") == line && strstr(line, "avx") != NULL) {
- return 1;
- }
- }
- fclose(cpuinfo);
- return 0;
-}
-
int
main(int argc, char *argv[])
{
@@ -177,12 +154,15 @@ main(int argc, char *argv[])
/* we assume xmm* won't be overwritten by this library call before the signal */
kill(getpid(), SIGUSR1);
- if (determine_avx()) {
+#ifdef __AVX__
+ {
/* put known values in ymm regs */
int buf[INTS_PER_YMM * NUM_SIMD_REGS];
char *ptr = (char *)buf;
int i, j;
intercept_signal(SIGUSR2, signal_handler, false);
+ print("Sending SIGUSR2\n");
+
/* put known values in xmm regs (we assume processor has xmm) */
for (i = 0; i < NUM_SIMD_REGS; i++) {
for (j = 0; j < INTS_PER_YMM; j++)
@@ -197,7 +177,7 @@ main(int argc, char *argv[])
__asm("mov %0, %%" XAX "; vmovdqu 0xa0(%%" XAX "),%%ymm5" : : "m"(ptr) : "%" XAX);
__asm("mov %0, %%" XAX "; vmovdqu 0xc0(%%" XAX "),%%ymm6" : : "m"(ptr) : "%" XAX);
__asm("mov %0, %%" XAX "; vmovdqu 0xe0(%%" XAX "),%%ymm7" : : "m"(ptr) : "%" XAX);
-#ifdef X64
+# ifdef X64
__asm("mov %0, %%" XAX "; vmovdqu 0x100(%%" XAX "),%%ymm8"
:
: "m"(ptr)
@@ -230,10 +210,11 @@ main(int argc, char *argv[])
:
: "m"(ptr)
: "%" XAX);
-#endif
+# endif
/* now make sure they show up in signal context */
kill(getpid(), SIGUSR2);
}
+#endif
print("All done\n");
return 0;
diff --git a/suite/tests/linux/sigcontext.template b/suite/tests/linux/sigcontext.template
index 62d4bb308..54f60f0f8 100644
--- a/suite/tests/linux/sigcontext.template
+++ b/suite/tests/linux/sigcontext.template
@@ -1,5 +1,4 @@
Sending SIGUSR1
-#ifdef X64
xmm[0] = 0xdeadbeef 0xdeadbeef 0xdeadbeef 0xdeadbeef
xmm[1] = 0xbd5b7dde 0xbd5b7dde 0xbd5b7dde 0xbd5b7dde
xmm[2] = 0x7ab6fbbc 0x7ab6fbbc 0x7ab6fbbc 0x7ab6fbbc
@@ -8,6 +7,7 @@ xmm[4] = 0xeadbeef0 0xeadbeef0 0xeadbeef0 0xeadbeef0
xmm[5] = 0xd5b7dde0 0xd5b7dde0 0xd5b7dde0 0xd5b7dde0
xmm[6] = 0xab6fbbc0 0xab6fbbc0 0xab6fbbc0 0xab6fbbc0
xmm[7] = 0x56df7780 0x56df7780 0x56df7780 0x56df7780
+#ifdef X64
xmm[8] = 0xadbeef00 0xadbeef00 0xadbeef00 0xadbeef00
xmm[9] = 0x5b7dde00 0x5b7dde00 0x5b7dde00 0x5b7dde00
xmm[10] = 0xb6fbbc00 0xb6fbbc00 0xb6fbbc00 0xb6fbbc00
@@ -16,14 +16,26 @@ xmm[12] = 0xdbeef000 0xdbeef000 0xdbeef000 0xdbeef000
xmm[13] = 0xb7dde000 0xb7dde000 0xb7dde000 0xb7dde000
xmm[14] = 0x6fbbc000 0x6fbbc000 0x6fbbc000 0x6fbbc000
xmm[15] = 0xdf778000 0xdf778000 0xdf778000 0xdf778000
-#else
-xmm[0] = 0xdeadbeef 0xdeadbeef 0xdeadbeef 0xdeadbeef
-xmm[1] = 0xbd5b7dde 0xbd5b7dde 0xbd5b7dde 0xbd5b7dde
-xmm[2] = 0x7ab6fbbc 0x7ab6fbbc 0x7ab6fbbc 0x7ab6fbbc
-xmm[3] = 0xf56df778 0xf56df778 0xf56df778 0xf56df778
-xmm[4] = 0xeadbeef0 0xeadbeef0 0xeadbeef0 0xeadbeef0
-xmm[5] = 0xd5b7dde0 0xd5b7dde0 0xd5b7dde0 0xd5b7dde0
-xmm[6] = 0xab6fbbc0 0xab6fbbc0 0xab6fbbc0 0xab6fbbc0
-xmm[7] = 0x56df7780 0x56df7780 0x56df7780 0x56df7780
+#endif
+#ifdef __AVX__
+Sending SIGUSR2
+ymmh[0] = 0xdeadbeef 0xdeadbeef 0xdeadbeef 0xdeadbeef
+ymmh[1] = 0xbd5b7dde 0xbd5b7dde 0xbd5b7dde 0xbd5b7dde
+ymmh[2] = 0x7ab6fbbc 0x7ab6fbbc 0x7ab6fbbc 0x7ab6fbbc
+ymmh[3] = 0xf56df778 0xf56df778 0xf56df778 0xf56df778
+ymmh[4] = 0xeadbeef0 0xeadbeef0 0xeadbeef0 0xeadbeef0
+ymmh[5] = 0xd5b7dde0 0xd5b7dde0 0xd5b7dde0 0xd5b7dde0
+ymmh[6] = 0xab6fbbc0 0xab6fbbc0 0xab6fbbc0 0xab6fbbc0
+ymmh[7] = 0x56df7780 0x56df7780 0x56df7780 0x56df7780
+#ifdef X64
+ymmh[8] = 0xadbeef00 0xadbeef00 0xadbeef00 0xadbeef00
+ymmh[9] = 0x5b7dde00 0x5b7dde00 0x5b7dde00 0x5b7dde00
+ymmh[10] = 0xb6fbbc00 0xb6fbbc00 0xb6fbbc00 0xb6fbbc00
+ymmh[11] = 0x6df77800 0x6df77800 0x6df77800 0x6df77800
+ymmh[12] = 0xdbeef000 0xdbeef000 0xdbeef000 0xdbeef000
+ymmh[13] = 0xb7dde000 0xb7dde000 0xb7dde000 0xb7dde000
+ymmh[14] = 0x6fbbc000 0x6fbbc000 0x6fbbc000 0x6fbbc000
+ymmh[15] = 0xdf778000 0xdf778000 0xdf778000 0xdf778000
+#endif
#endif
All done
--
GitLab
From 5787df2d86907ee7ac7ae4dac6ab4c8a8cbfe7d7 Mon Sep 17 00:00:00 2001
From: Hendrik Greving <hgreving@google.com>
Date: Mon, 29 Jul 2019 14:55:08 -0700
Subject: [PATCH 2/7] *** Temporary commit, due to only reproducible on Travis
***
---
suite/runsuite_wrapper.pl | 2 +-
suite/tests/linux/sigcontext.c | 3 +++
2 files changed, 4 insertions(+), 1 deletion(-)
diff --git a/suite/runsuite_wrapper.pl b/suite/runsuite_wrapper.pl
index a8684ff4c..55fd2c0a8 100755
--- a/suite/runsuite_wrapper.pl
+++ b/suite/runsuite_wrapper.pl
@@ -86,7 +86,7 @@ if ($child) {
$mydir = `/usr/bin/cygpath -wi \"$mydir\"`;
chomp $mydir;
}
- system("ctest --output-on-failure ${verbose} -S \"${mydir}/runsuite.cmake${args}\" 2>&1");
+ system("ctest -R sigcontext --output-on-failure ${verbose} -S \"${mydir}/runsuite.cmake${args}\" 2>&1");
exit 0;
}
diff --git a/suite/tests/linux/sigcontext.c b/suite/tests/linux/sigcontext.c
index dc1aa8bb3..5d1a44022 100644
--- a/suite/tests/linux/sigcontext.c
+++ b/suite/tests/linux/sigcontext.c
@@ -97,6 +97,9 @@ signal_handler(int sig, siginfo_t *siginfo, ucontext_t *ucxt)
kernel_xstate_t *xstate = (kernel_xstate_t *)ucxt->uc_mcontext.fpregs;
if (xstate->fpstate.sw_reserved.magic1 == FP_XSTATE_MAGIC1) {
assert(xstate->fpstate.sw_reserved.xstate_size >= sizeof(*xstate));
+ print("was here: xstate->fpstate.sw_reserved.xstate_size:%d\n",
+ xstate->fpstate.sw_reserved.xstate_size);
+ print("was here: sizeof(*xstate):%d\n", sizeof(*xstate));
for (i = 0; i < NUM_SIMD_REGS; i++) {
#ifdef __AVX__
print("ymmh[%d] = 0x%x 0x%x 0x%x 0x%x\n", i,
--
GitLab
From b29b20721a50d47238488e4199a7195bb8319da2 Mon Sep 17 00:00:00 2001
From: Hendrik Greving <hgreving@google.com>
Date: Mon, 29 Jul 2019 14:58:15 -0700
Subject: [PATCH 3/7] *** Temporary commit, due to only reproducible on Travis
***
---
suite/tests/linux/sigcontext.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/suite/tests/linux/sigcontext.c b/suite/tests/linux/sigcontext.c
index 5d1a44022..7a64e1ccd 100644
--- a/suite/tests/linux/sigcontext.c
+++ b/suite/tests/linux/sigcontext.c
@@ -96,10 +96,10 @@ signal_handler(int sig, siginfo_t *siginfo, ucontext_t *ucxt)
*/
kernel_xstate_t *xstate = (kernel_xstate_t *)ucxt->uc_mcontext.fpregs;
if (xstate->fpstate.sw_reserved.magic1 == FP_XSTATE_MAGIC1) {
- assert(xstate->fpstate.sw_reserved.xstate_size >= sizeof(*xstate));
print("was here: xstate->fpstate.sw_reserved.xstate_size:%d\n",
xstate->fpstate.sw_reserved.xstate_size);
print("was here: sizeof(*xstate):%d\n", sizeof(*xstate));
+ assert(xstate->fpstate.sw_reserved.xstate_size >= sizeof(*xstate));
for (i = 0; i < NUM_SIMD_REGS; i++) {
#ifdef __AVX__
print("ymmh[%d] = 0x%x 0x%x 0x%x 0x%x\n", i,
--
GitLab
From 9142cf723f23335aefb13210a1073f1f4fba16da Mon Sep 17 00:00:00 2001
From: Hendrik Greving <hgreving@google.com>
Date: Mon, 29 Jul 2019 15:05:36 -0700
Subject: [PATCH 4/7] *** Temporary commit, due to only reproducible on Travis
***
---
suite/tests/linux/sigcontext.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/suite/tests/linux/sigcontext.c b/suite/tests/linux/sigcontext.c
index 7a64e1ccd..9dd64f53e 100644
--- a/suite/tests/linux/sigcontext.c
+++ b/suite/tests/linux/sigcontext.c
@@ -98,8 +98,10 @@ signal_handler(int sig, siginfo_t *siginfo, ucontext_t *ucxt)
if (xstate->fpstate.sw_reserved.magic1 == FP_XSTATE_MAGIC1) {
print("was here: xstate->fpstate.sw_reserved.xstate_size:%d\n",
xstate->fpstate.sw_reserved.xstate_size);
+ print("was here: xstate->fpstate.sw_reserved.extended_size:%d\n",
+ xstate->fpstate.sw_reserved.extended_size);
print("was here: sizeof(*xstate):%d\n", sizeof(*xstate));
- assert(xstate->fpstate.sw_reserved.xstate_size >= sizeof(*xstate));
+ assert(xstate->fpstate.sw_reserved.extended_size >= sizeof(*xstate));
for (i = 0; i < NUM_SIMD_REGS; i++) {
#ifdef __AVX__
print("ymmh[%d] = 0x%x 0x%x 0x%x 0x%x\n", i,
--
GitLab
From c92e3e0c227d6a13959fc1065a3ae50cd1c69823 Mon Sep 17 00:00:00 2001
From: Hendrik Greving <hgreving@google.com>
Date: Mon, 29 Jul 2019 15:13:30 -0700
Subject: [PATCH 5/7] Fix latent bug, test assertion should be extended_size.
---
suite/runsuite_wrapper.pl | 2 +-
suite/tests/linux/sigcontext.c | 5 -----
2 files changed, 1 insertion(+), 6 deletions(-)
diff --git a/suite/runsuite_wrapper.pl b/suite/runsuite_wrapper.pl
index 55fd2c0a8..a8684ff4c 100755
--- a/suite/runsuite_wrapper.pl
+++ b/suite/runsuite_wrapper.pl
@@ -86,7 +86,7 @@ if ($child) {
$mydir = `/usr/bin/cygpath -wi \"$mydir\"`;
chomp $mydir;
}
- system("ctest -R sigcontext --output-on-failure ${verbose} -S \"${mydir}/runsuite.cmake${args}\" 2>&1");
+ system("ctest --output-on-failure ${verbose} -S \"${mydir}/runsuite.cmake${args}\" 2>&1");
exit 0;
}
diff --git a/suite/tests/linux/sigcontext.c b/suite/tests/linux/sigcontext.c
index 9dd64f53e..da4214a86 100644
--- a/suite/tests/linux/sigcontext.c
+++ b/suite/tests/linux/sigcontext.c
@@ -96,11 +96,6 @@ signal_handler(int sig, siginfo_t *siginfo, ucontext_t *ucxt)
*/
kernel_xstate_t *xstate = (kernel_xstate_t *)ucxt->uc_mcontext.fpregs;
if (xstate->fpstate.sw_reserved.magic1 == FP_XSTATE_MAGIC1) {
- print("was here: xstate->fpstate.sw_reserved.xstate_size:%d\n",
- xstate->fpstate.sw_reserved.xstate_size);
- print("was here: xstate->fpstate.sw_reserved.extended_size:%d\n",
- xstate->fpstate.sw_reserved.extended_size);
- print("was here: sizeof(*xstate):%d\n", sizeof(*xstate));
assert(xstate->fpstate.sw_reserved.extended_size >= sizeof(*xstate));
for (i = 0; i < NUM_SIMD_REGS; i++) {
#ifdef __AVX__
--
GitLab
From f48b8d7b0cc4ad60df3b3dd2df553a91ac8d6923 Mon Sep 17 00:00:00 2001
From: Hendrik Greving <hgreving@google.com>
Date: Tue, 30 Jul 2019 10:22:57 -0700
Subject: [PATCH 6/7] Use offsetof for i387_fxsave_struct. Use entire
sizeof(i387,fsave_struct).
---
core/unix/signal_linux_x86.c | 7 +++----
1 file changed, 3 insertions(+), 4 deletions(-)
diff --git a/core/unix/signal_linux_x86.c b/core/unix/signal_linux_x86.c
index 2ad44ebad..7cb6c7131 100644
--- a/core/unix/signal_linux_x86.c
+++ b/core/unix/signal_linux_x86.c
@@ -180,7 +180,7 @@ convert_fxsave_to_fpstate(kernel_fpstate_t *fpstate, struct i387_fxsave_struct *
fpstate->magic = X86_FXSR_MAGIC;
memcpy(&fpstate->_fxsr_env[0], fxsave,
- sizeof(struct i387_fxsave_struct) - sizeof(fpstate->sw_reserved));
+ offsetof(struct i387_fxsave_struct, xmm_space));
}
#endif /* !X64 */
@@ -288,7 +288,7 @@ save_fpstate(dcontext_t *dcontext, sigframe_rt_t *frame)
/* now convert into kernel_fpstate_t form */
ASSERT(sizeof(kernel_fpstate_t) == sizeof(struct i387_fxsave_struct));
memcpy(sc->fpstate, &temp->fxsave,
- sizeof(struct i387_fxsave_struct) - sizeof(sc->fpstate->sw_reserved));
+ offsetof(struct i387_fxsave_struct, xmm_space));
#else
/* this is "unlazy_fpu" */
asm volatile("fxsave %0 ; fnclex" : "=m"(temp->fxsave));
@@ -302,8 +302,7 @@ save_fpstate(dcontext_t *dcontext, sigframe_rt_t *frame)
asm volatile("fnsave %0 ; fwait" : "=m"(temp->fsave));
/* now convert into kernel_fpstate_t form */
temp->fsave.status = temp->fsave.swd;
- memcpy(sc->fpstate, &temp->fsave,
- sizeof(struct i387_fsave_struct) - sizeof(sc->fpstate->sw_reserved));
+ memcpy(sc->fpstate, &temp->fsave, sizeof(struct i387_fsave_struct));
}
save_xmm(dcontext, frame);
--
GitLab
From 28a285c8f064608ca262ff2b3dbb30b3edf5bbd8 Mon Sep 17 00:00:00 2001
From: Hendrik Greving <hgreving@google.com>
Date: Tue, 30 Jul 2019 12:21:41 -0700
Subject: [PATCH 7/7] Pull in updated comments from up-to-date kernel version.
---
core/unix/include/sigcontext.h | 20 +++++++++++++-------
1 file changed, 13 insertions(+), 7 deletions(-)
diff --git a/core/unix/include/sigcontext.h b/core/unix/include/sigcontext.h
index 3945f6e39..4551fab68 100644
--- a/core/unix/include/sigcontext.h
+++ b/core/unix/include/sigcontext.h
@@ -25,16 +25,22 @@
* are used to extended the fpstate pointer in the sigcontext, which now
* includes the extended state information along with fpstate information.
*
- * Presence of FP_XSTATE_MAGIC1 at the beginning of this SW reserved
- * area and FP_XSTATE_MAGIC2 at the end of memory layout
- * (extended_size - FP_XSTATE_MAGIC2_SIZE) indicates the presence of the
- * extended state information in the memory layout pointed by the fpstate
- * pointer in sigcontext.
+ * If sw_reserved.magic1 == FP_XSTATE_MAGIC1 then there's a
+ * sw_reserved.extended_size bytes large extended context area present. (The
+ * last 32-bit word of this extended area (at the
+ * fpstate+extended_size-FP_XSTATE_MAGIC2_SIZE address) is set to
+ * FP_XSTATE_MAGIC2 so that you can sanity check your size calculations.)
+ *
+ * This extended area typically grows with newer CPUs that have larger and
+ * larger XSAVE areas.
*/
typedef struct _kernel_fpx_sw_bytes_t {
__u32 magic1; /* FP_XSTATE_MAGIC1 */
- __u32 extended_size; /* total size of the layout referred by
- * fpstate pointer in the sigcontext.
+ __u32 extended_size; /* Total size of the fpstate area:
+ *
+ * - if magic1 == 0 then it's sizeof(struct _fpstate)
+ * - if magic1 == FP_XSTATE_MAGIC1 then it's sizeof(struct
+ * _xstate) plus extensions (if any)
*/
__u64 xstate_bv;
/* feature bit mask (including fp/sse/extended
--
GitLab