From: Paolo Bonzini <pbonzini@redhat.com>
To: qemu-devel@nongnu.org
Cc: Richard Henderson <richard.henderson@linaro.org>
Subject: [PULL 36/37] linux-user: i386/signal: support FXSAVE fpstate on 32-bit emulation
Date: Tue, 11 Oct 2022 12:26:59 +0200 [thread overview]
Message-ID: <20221011102700.319178-37-pbonzini@redhat.com> (raw)
In-Reply-To: <20221011102700.319178-1-pbonzini@redhat.com>
Linux can use FXSAVE to save/restore XMM registers even on 32-bit
systems. This requires some care in order to keep the FXSAVE area
aligned to 16 bytes; for this reason, get_sigframe is changed to
pass the offset into the FXSAVE area rather than the full frame
size.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
linux-user/i386/signal.c | 129 +++++++++++++++++++++++----------------
1 file changed, 77 insertions(+), 52 deletions(-)
diff --git a/linux-user/i386/signal.c b/linux-user/i386/signal.c
index 76317a3d16..58b11be116 100644
--- a/linux-user/i386/signal.c
+++ b/linux-user/i386/signal.c
@@ -39,29 +39,7 @@ struct target_xmmreg {
uint32_t element[4];
};
-struct target_fpstate_32 {
- /* Regular FPU environment */
- uint32_t cw;
- uint32_t sw;
- uint32_t tag;
- uint32_t ipoff;
- uint32_t cssel;
- uint32_t dataoff;
- uint32_t datasel;
- struct target_fpreg st[8];
- uint16_t status;
- uint16_t magic; /* 0xffff = regular FPU data only */
-
- /* FXSR FPU environment */
- uint32_t _fxsr_env[6]; /* FXSR FPU env is ignored */
- uint32_t mxcsr;
- uint32_t reserved;
- struct target_fpxreg fxsr_st[8]; /* FXSR FPU reg data is ignored */
- struct target_xmmreg xmm[8];
- uint32_t padding[56];
-};
-
-struct target_fpstate_64 {
+struct target_fpstate_fxsave {
/* FXSAVE format */
uint16_t cw;
uint16_t sw;
@@ -75,11 +53,36 @@ struct target_fpstate_64 {
uint32_t xmm_space[64];
uint32_t reserved[24];
};
+#define TARGET_FXSAVE_SIZE sizeof(struct target_fpstate_fxsave)
+QEMU_BUILD_BUG_ON(TARGET_FXSAVE_SIZE != 512);
+
+struct target_fpstate_32 {
+ /* Regular FPU environment */
+ uint32_t cw;
+ uint32_t sw;
+ uint32_t tag;
+ uint32_t ipoff;
+ uint32_t cssel;
+ uint32_t dataoff;
+ uint32_t datasel;
+ struct target_fpreg st[8];
+ uint16_t status;
+ uint16_t magic; /* 0xffff = regular FPU data only */
+ struct target_fpstate_fxsave fxsave;
+};
+
+/*
+ * For simplicity, setup_frame aligns struct target_fpstate_32 to
+ * 16 bytes, so ensure that the FXSAVE area is also aligned.
+ */
+QEMU_BUILD_BUG_ON(offsetof(struct target_fpstate_32, fxsave) & 15);
#ifndef TARGET_X86_64
# define target_fpstate target_fpstate_32
+# define TARGET_FPSTATE_FXSAVE_OFFSET offsetof(struct target_fpstate_32, fxsave)
#else
-# define target_fpstate target_fpstate_64
+# define target_fpstate target_fpstate_fxsave
+# define TARGET_FPSTATE_FXSAVE_OFFSET 0
#endif
struct target_sigcontext_32 {
@@ -172,8 +175,16 @@ struct sigframe {
struct target_fpstate fpstate_unused;
abi_ulong extramask[TARGET_NSIG_WORDS-1];
char retcode[8];
- struct target_fpstate fpstate;
+
+ /*
+ * This field will be 16-byte aligned in memory. Applying QEMU_ALIGNED
+ * to it ensures that the base of the frame has an appropriate alignment
+ * too.
+ */
+ struct target_fpstate fpstate QEMU_ALIGNED(8);
};
+#define TARGET_SIGFRAME_FXSAVE_OFFSET ( \
+ offsetof(struct sigframe, fpstate) + TARGET_FPSTATE_FXSAVE_OFFSET)
struct rt_sigframe {
abi_ulong pretcode;
@@ -183,25 +194,35 @@ struct rt_sigframe {
struct target_siginfo info;
struct target_ucontext uc;
char retcode[8];
- struct target_fpstate fpstate;
+ struct target_fpstate fpstate QEMU_ALIGNED(8);
};
-
+#define TARGET_RT_SIGFRAME_FXSAVE_OFFSET ( \
+ offsetof(struct rt_sigframe, fpstate) + TARGET_FPSTATE_FXSAVE_OFFSET)
#else
struct rt_sigframe {
abi_ulong pretcode;
struct target_ucontext uc;
struct target_siginfo info;
- struct target_fpstate fpstate;
+ struct target_fpstate fpstate QEMU_ALIGNED(16);
};
-
+#define TARGET_RT_SIGFRAME_FXSAVE_OFFSET ( \
+ offsetof(struct rt_sigframe, fpstate) + TARGET_FPSTATE_FXSAVE_OFFSET)
#endif
/*
* Set up a signal frame.
*/
-/* XXX: save x87 state */
+static void fxsave_sigcontext(CPUX86State *env, struct target_fpstate_fxsave *fxsave,
+ abi_ulong fxsave_addr)
+{
+ /* fxsave_addr must be 16 byte aligned for fxsave */
+ assert(!(fxsave_addr & 0xf));
+
+ cpu_x86_fxsave(env, fxsave_addr);
+}
+
static void setup_sigcontext(struct target_sigcontext *sc,
struct target_fpstate *fpstate, CPUX86State *env, abi_ulong mask,
abi_ulong fpstate_addr)
@@ -233,13 +254,14 @@ static void setup_sigcontext(struct target_sigcontext *sc,
cpu_x86_fsave(env, fpstate_addr, 1);
fpstate->status = fpstate->sw;
- magic = 0xffff;
+ if (!(env->features[FEAT_1_EDX] & CPUID_FXSR)) {
+ magic = 0xffff;
+ } else {
+ fxsave_sigcontext(env, &fpstate->fxsave,
+ fpstate_addr + TARGET_FPSTATE_FXSAVE_OFFSET);
+ magic = 0;
+ }
__put_user(magic, &fpstate->magic);
- __put_user(fpstate_addr, &sc->fpstate);
-
- /* non-iBCS2 extensions.. */
- __put_user(mask, &sc->oldmask);
- __put_user(env->cr[2], &sc->cr2);
#else
__put_user(env->regs[R_EDI], &sc->rdi);
__put_user(env->regs[R_ESI], &sc->rsi);
@@ -269,15 +291,14 @@ static void setup_sigcontext(struct target_sigcontext *sc,
__put_user((uint16_t)0, &sc->fs);
__put_user(env->segs[R_SS].selector, &sc->ss);
+ fxsave_sigcontext(env, fpstate, fpstate_addr);
+#endif
+
+ __put_user(fpstate_addr, &sc->fpstate);
+
+ /* non-iBCS2 extensions.. */
__put_user(mask, &sc->oldmask);
__put_user(env->cr[2], &sc->cr2);
-
- /* fpstate_addr must be 16 byte aligned for fxsave */
- assert(!(fpstate_addr & 0xf));
-
- cpu_x86_fxsave(env, fpstate_addr);
- __put_user(fpstate_addr, &sc->fpstate);
-#endif
}
/*
@@ -285,7 +306,7 @@ static void setup_sigcontext(struct target_sigcontext *sc,
*/
static inline abi_ulong
-get_sigframe(struct target_sigaction *ka, CPUX86State *env, size_t frame_size)
+get_sigframe(struct target_sigaction *ka, CPUX86State *env, size_t fxsave_offset)
{
unsigned long esp;
@@ -309,11 +330,11 @@ get_sigframe(struct target_sigaction *ka, CPUX86State *env, size_t frame_size)
#endif
}
-#ifndef TARGET_X86_64
- return (esp - frame_size) & -8ul;
-#else
- return ((esp - frame_size) & (~15ul)) - 8;
-#endif
+ if (!(env->features[FEAT_1_EDX] & CPUID_FXSR)) {
+ return (esp - (fxsave_offset + TARGET_FXSAVE_SIZE)) & -8ul;
+ } else {
+ return ((esp - TARGET_FXSAVE_SIZE) & -16ul) - fxsave_offset;
+ }
}
#ifndef TARGET_X86_64
@@ -341,7 +362,7 @@ void setup_frame(int sig, struct target_sigaction *ka,
struct sigframe *frame;
int i;
- frame_addr = get_sigframe(ka, env, sizeof(*frame));
+ frame_addr = get_sigframe(ka, env, TARGET_SIGFRAME_FXSAVE_OFFSET);
trace_user_setup_frame(env, frame_addr);
if (!lock_user_struct(VERIFY_WRITE, frame, frame_addr, 0))
@@ -397,7 +418,7 @@ void setup_rt_frame(int sig, struct target_sigaction *ka,
struct rt_sigframe *frame;
int i;
- frame_addr = get_sigframe(ka, env, sizeof(*frame));
+ frame_addr = get_sigframe(ka, env, TARGET_RT_SIGFRAME_FXSAVE_OFFSET);
trace_user_setup_rt_frame(env, frame_addr);
if (!lock_user_struct(VERIFY_WRITE, frame, frame_addr, 0))
@@ -529,7 +550,11 @@ restore_sigcontext(CPUX86State *env, struct target_sigcontext *sc)
goto badframe;
}
#ifndef TARGET_X86_64
- cpu_x86_frstor(env, fpstate_addr, 1);
+ if (!(env->features[FEAT_1_EDX] & CPUID_FXSR)) {
+ cpu_x86_frstor(env, fpstate_addr, 1);
+ } else {
+ cpu_x86_fxrstor(env, fpstate_addr + TARGET_FPSTATE_FXSAVE_OFFSET);
+ }
#else
cpu_x86_fxrstor(env, fpstate_addr);
#endif
--
2.37.3
next prev parent reply other threads:[~2022-10-11 11:32 UTC|newest]
Thread overview: 39+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-10-11 10:26 [PULL 00/37] SCSI, i386 patches for 2022-10-11 Paolo Bonzini
2022-10-11 10:26 ` [PULL 01/37] scsi-disk: support setting CD-ROM block size via device options Paolo Bonzini
2022-10-11 10:26 ` [PULL 02/37] i386: kvm: extend kvm_{get, put}_vcpu_events to support pending triple fault Paolo Bonzini
2022-10-11 10:26 ` [PULL 03/37] kvm: allow target-specific accelerator properties Paolo Bonzini
2022-10-11 10:26 ` [PULL 04/37] kvm: expose struct KVMState Paolo Bonzini
2022-10-11 10:26 ` [PULL 05/37] i386: add notify VM exit support Paolo Bonzini
2022-10-11 10:26 ` [PULL 06/37] target/i386: Remove pc_start Paolo Bonzini
2022-10-11 10:26 ` [PULL 07/37] target/i386: Return bool from disas_insn Paolo Bonzini
2022-10-11 10:26 ` [PULL 08/37] target/i386: Remove cur_eip argument to gen_exception Paolo Bonzini
2022-10-11 10:26 ` [PULL 09/37] target/i386: Remove cur_eip, next_eip arguments to gen_interrupt Paolo Bonzini
2022-10-11 10:26 ` [PULL 10/37] target/i386: Create gen_update_eip_cur Paolo Bonzini
2022-10-11 10:26 ` [PULL 11/37] target/i386: Create gen_update_eip_next Paolo Bonzini
2022-10-11 10:26 ` [PULL 12/37] target/i386: Introduce DISAS_EOB* Paolo Bonzini
2022-10-11 10:26 ` [PULL 13/37] target/i386: Use DISAS_EOB* in gen_movl_seg_T0 Paolo Bonzini
2022-10-11 10:26 ` [PULL 14/37] target/i386: Use DISAS_EOB_NEXT Paolo Bonzini
2022-10-11 10:26 ` [PULL 15/37] target/i386: USe DISAS_EOB_ONLY Paolo Bonzini
2022-10-11 10:26 ` [PULL 16/37] target/i386: Create cur_insn_len, cur_insn_len_i32 Paolo Bonzini
2022-10-11 10:26 ` [PULL 17/37] target/i386: Remove cur_eip, next_eip arguments to gen_repz* Paolo Bonzini
2022-10-11 10:26 ` [PULL 18/37] target/i386: Introduce DISAS_JUMP Paolo Bonzini
2022-10-11 10:26 ` [PULL 19/37] target/i386: Truncate values for lcall_real to i32 Paolo Bonzini
2022-10-11 10:26 ` [PULL 20/37] target/i386: Create eip_next_* Paolo Bonzini
2022-10-11 10:26 ` [PULL 21/37] target/i386: Use DISAS_TOO_MANY to exit after gen_io_start Paolo Bonzini
2022-10-11 10:26 ` [PULL 22/37] target/i386: Create gen_jmp_rel Paolo Bonzini
2022-10-11 10:26 ` [PULL 23/37] target/i386: Use gen_jmp_rel for loop, repz, jecxz insns Paolo Bonzini
2022-10-11 10:26 ` [PULL 24/37] target/i386: Use gen_jmp_rel for gen_jcc Paolo Bonzini
2022-10-11 10:26 ` [PULL 25/37] target/i386: Use gen_jmp_rel for DISAS_TOO_MANY Paolo Bonzini
2022-10-11 10:26 ` [PULL 26/37] target/i386: Remove MemOp argument to gen_op_j*_ecx Paolo Bonzini
2022-10-11 10:26 ` [PULL 27/37] target/i386: Merge gen_jmp_tb and gen_goto_tb into gen_jmp_rel Paolo Bonzini
2022-10-11 10:26 ` [PULL 28/37] target/i386: Create eip_cur_tl Paolo Bonzini
2022-10-11 10:26 ` [PULL 29/37] target/i386: Add cpu_eip Paolo Bonzini
2022-10-11 10:26 ` [PULL 30/37] target/i386: Inline gen_jmp_im Paolo Bonzini
2022-10-11 10:26 ` [PULL 31/37] target/i386: Enable TARGET_TB_PCREL Paolo Bonzini
2022-10-11 10:26 ` [PULL 32/37] x86: Implement MSR_CORE_THREAD_COUNT MSR Paolo Bonzini
2022-10-11 10:26 ` [PULL 33/37] i386: kvm: Add support for MSR filtering Paolo Bonzini
2022-10-11 10:26 ` [PULL 34/37] KVM: x86: Implement MSR_CORE_THREAD_COUNT MSR Paolo Bonzini
2022-10-11 10:26 ` [PULL 35/37] linux-user: i386/signal: move fpstate at the end of the 32-bit frames Paolo Bonzini
2022-10-11 10:26 ` Paolo Bonzini [this message]
2022-10-11 10:27 ` [PULL 37/37] linux-user: i386/signal: support XSAVE/XRSTOR for signal frame fpstate Paolo Bonzini
2022-10-13 20:29 ` [PULL 00/37] SCSI, i386 patches for 2022-10-11 Stefan Hajnoczi
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=20221011102700.319178-37-pbonzini@redhat.com \
--to=pbonzini@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.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;
as well as URLs for NNTP newsgroup(s).