From: Paolo Bonzini <pbonzini@redhat.com>
To: qemu-devel@nongnu.org
Cc: Richard Henderson <richard.henderson@linaro.org>
Subject: [PULL 37/37] linux-user: i386/signal: support XSAVE/XRSTOR for signal frame fpstate
Date: Tue, 11 Oct 2022 12:27:00 +0200 [thread overview]
Message-ID: <20221011102700.319178-38-pbonzini@redhat.com> (raw)
In-Reply-To: <20221011102700.319178-1-pbonzini@redhat.com>
Add support for saving/restoring extended save states when signals
are delivered. This allows using AVX, MPX or PKRU registers in
signal handlers.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
linux-user/i386/signal.c | 119 +++++++++++++++++++++++++++++------
target/i386/cpu.c | 2 +-
target/i386/cpu.h | 3 +
target/i386/tcg/fpu_helper.c | 64 +++++++++++--------
4 files changed, 142 insertions(+), 46 deletions(-)
diff --git a/linux-user/i386/signal.c b/linux-user/i386/signal.c
index 58b11be116..60fa07d6f9 100644
--- a/linux-user/i386/signal.c
+++ b/linux-user/i386/signal.c
@@ -24,6 +24,10 @@
/* from the Linux kernel - /arch/x86/include/uapi/asm/sigcontext.h */
+#define TARGET_FP_XSTATE_MAGIC1 0x46505853U /* FPXS */
+#define TARGET_FP_XSTATE_MAGIC2 0x46505845U /* FPXE */
+#define TARGET_FP_XSTATE_MAGIC2_SIZE 4
+
struct target_fpreg {
uint16_t significand[4];
uint16_t exponent;
@@ -39,6 +43,15 @@ struct target_xmmreg {
uint32_t element[4];
};
+struct target_fpx_sw_bytes {
+ uint32_t magic1;
+ uint32_t extended_size;
+ uint64_t xfeatures;
+ uint32_t xstate_size;
+ uint32_t reserved[7];
+};
+QEMU_BUILD_BUG_ON(sizeof(struct target_fpx_sw_bytes) != 12*4);
+
struct target_fpstate_fxsave {
/* FXSAVE format */
uint16_t cw;
@@ -51,10 +64,13 @@ struct target_fpstate_fxsave {
uint32_t mxcsr_mask;
uint32_t st_space[32];
uint32_t xmm_space[64];
- uint32_t reserved[24];
+ uint32_t hw_reserved[12];
+ struct target_fpx_sw_bytes sw_reserved;
+ uint8_t xfeatures[];
};
#define TARGET_FXSAVE_SIZE sizeof(struct target_fpstate_fxsave)
QEMU_BUILD_BUG_ON(TARGET_FXSAVE_SIZE != 512);
+QEMU_BUILD_BUG_ON(offsetof(struct target_fpstate_fxsave, sw_reserved) != 464);
struct target_fpstate_32 {
/* Regular FPU environment */
@@ -214,13 +230,39 @@ struct rt_sigframe {
* Set up a signal frame.
*/
-static void fxsave_sigcontext(CPUX86State *env, struct target_fpstate_fxsave *fxsave,
- abi_ulong fxsave_addr)
+static void xsave_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));
+ if (!(env->features[FEAT_1_ECX] & CPUID_EXT_XSAVE)) {
+ /* fxsave_addr must be 16 byte aligned for fxsave */
+ assert(!(fxsave_addr & 0xf));
- cpu_x86_fxsave(env, fxsave_addr);
+ cpu_x86_fxsave(env, fxsave_addr);
+ __put_user(0, &fxsave->sw_reserved.magic1);
+ } else {
+ uint32_t xstate_size = xsave_area_size(env->xcr0, false);
+ uint32_t xfeatures_size = xstate_size - TARGET_FXSAVE_SIZE;
+
+ /*
+ * extended_size is the offset from fpstate_addr to right after the end
+ * of the extended save states. On 32-bit that includes the legacy
+ * FSAVE area.
+ */
+ uint32_t extended_size = TARGET_FPSTATE_FXSAVE_OFFSET
+ + xstate_size + TARGET_FP_XSTATE_MAGIC2_SIZE;
+
+ /* fxsave_addr must be 64 byte aligned for xsave */
+ assert(!(fxsave_addr & 0x3f));
+
+ /* Zero the header, XSAVE *adds* features to an existing save state. */
+ memset(fxsave->xfeatures, 0, 64);
+ cpu_x86_xsave(env, fxsave_addr);
+ __put_user(TARGET_FP_XSTATE_MAGIC1, &fxsave->sw_reserved.magic1);
+ __put_user(extended_size, &fxsave->sw_reserved.extended_size);
+ __put_user(env->xcr0, &fxsave->sw_reserved.xfeatures);
+ __put_user(xstate_size, &fxsave->sw_reserved.xstate_size);
+ __put_user(TARGET_FP_XSTATE_MAGIC2, (uint32_t *) &fxsave->xfeatures[xfeatures_size]);
+ }
}
static void setup_sigcontext(struct target_sigcontext *sc,
@@ -257,8 +299,8 @@ static void setup_sigcontext(struct target_sigcontext *sc,
if (!(env->features[FEAT_1_EDX] & CPUID_FXSR)) {
magic = 0xffff;
} else {
- fxsave_sigcontext(env, &fpstate->fxsave,
- fpstate_addr + TARGET_FPSTATE_FXSAVE_OFFSET);
+ xsave_sigcontext(env, &fpstate->fxsave,
+ fpstate_addr + TARGET_FPSTATE_FXSAVE_OFFSET);
magic = 0;
}
__put_user(magic, &fpstate->magic);
@@ -291,7 +333,7 @@ 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);
+ xsave_sigcontext(env, fpstate, fpstate_addr);
#endif
__put_user(fpstate_addr, &sc->fpstate);
@@ -332,8 +374,12 @@ get_sigframe(struct target_sigaction *ka, CPUX86State *env, size_t fxsave_offset
if (!(env->features[FEAT_1_EDX] & CPUID_FXSR)) {
return (esp - (fxsave_offset + TARGET_FXSAVE_SIZE)) & -8ul;
- } else {
+ } else if (!(env->features[FEAT_1_ECX] & CPUID_EXT_XSAVE)) {
return ((esp - TARGET_FXSAVE_SIZE) & -16ul) - fxsave_offset;
+ } else {
+ size_t xstate_size =
+ xsave_area_size(env->xcr0, false) + TARGET_FP_XSTATE_MAGIC2_SIZE;
+ return ((esp - xstate_size) & -64ul) - fxsave_offset;
}
}
@@ -437,7 +483,11 @@ void setup_rt_frame(int sig, struct target_sigaction *ka,
}
/* Create the ucontext. */
- __put_user(0, &frame->uc.tuc_flags);
+ if (env->features[FEAT_1_ECX] & CPUID_EXT_XSAVE) {
+ __put_user(1, &frame->uc.tuc_flags);
+ } else {
+ __put_user(0, &frame->uc.tuc_flags);
+ }
__put_user(0, &frame->uc.tuc_link);
target_save_altstack(&frame->uc.tuc_stack, env);
setup_sigcontext(&frame->uc.tuc_mcontext, &frame->fpstate, env,
@@ -491,10 +541,37 @@ give_sigsegv:
force_sigsegv(sig);
}
+static int xrstor_sigcontext(CPUX86State *env, struct target_fpstate_fxsave *fxsave,
+ abi_ulong fxsave_addr)
+{
+ if (env->features[FEAT_1_ECX] & CPUID_EXT_XSAVE) {
+ uint32_t extended_size = tswapl(fxsave->sw_reserved.extended_size);
+ uint32_t xstate_size = tswapl(fxsave->sw_reserved.xstate_size);
+ uint32_t xfeatures_size = xstate_size - TARGET_FXSAVE_SIZE;
+
+ /* Linux checks MAGIC2 using xstate_size, not extended_size. */
+ if (tswapl(fxsave->sw_reserved.magic1) == TARGET_FP_XSTATE_MAGIC1 &&
+ extended_size >= TARGET_FPSTATE_FXSAVE_OFFSET + xstate_size + TARGET_FP_XSTATE_MAGIC2_SIZE) {
+ if (!access_ok(env_cpu(env), VERIFY_READ, fxsave_addr,
+ extended_size - TARGET_FPSTATE_FXSAVE_OFFSET)) {
+ return 1;
+ }
+ if (tswapl(*(uint32_t *) &fxsave->xfeatures[xfeatures_size]) == TARGET_FP_XSTATE_MAGIC2) {
+ cpu_x86_xrstor(env, fxsave_addr);
+ return 0;
+ }
+ }
+ /* fall through to fxrstor */
+ }
+
+ cpu_x86_fxrstor(env, fxsave_addr);
+ return 0;
+}
+
static int
restore_sigcontext(CPUX86State *env, struct target_sigcontext *sc)
{
- unsigned int err = 0;
+ int err = 1;
abi_ulong fpstate_addr;
unsigned int tmpflags;
@@ -545,24 +622,28 @@ restore_sigcontext(CPUX86State *env, struct target_sigcontext *sc)
fpstate_addr = tswapl(sc->fpstate);
if (fpstate_addr != 0) {
- if (!access_ok(env_cpu(env), VERIFY_READ, fpstate_addr,
- sizeof(struct target_fpstate))) {
- goto badframe;
+ struct target_fpstate *fpstate;
+ if (!lock_user_struct(VERIFY_READ, fpstate, fpstate_addr,
+ sizeof(struct target_fpstate))) {
+ return err;
}
#ifndef TARGET_X86_64
if (!(env->features[FEAT_1_EDX] & CPUID_FXSR)) {
cpu_x86_frstor(env, fpstate_addr, 1);
+ err = 0;
} else {
- cpu_x86_fxrstor(env, fpstate_addr + TARGET_FPSTATE_FXSAVE_OFFSET);
+ err = xrstor_sigcontext(env, &fpstate->fxsave,
+ fpstate_addr + TARGET_FPSTATE_FXSAVE_OFFSET);
}
#else
- cpu_x86_fxrstor(env, fpstate_addr);
+ err = xrstor_sigcontext(env, fpstate, fpstate_addr);
#endif
+ unlock_user_struct(fpstate, fpstate_addr, 0);
+ } else {
+ err = 0;
}
return err;
-badframe:
- return 1;
}
/* Note: there is no sigreturn on x86_64, there is only rt_sigreturn */
diff --git a/target/i386/cpu.c b/target/i386/cpu.c
index 06884177fa..8a11470507 100644
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -1467,7 +1467,7 @@ ExtSaveArea x86_ext_save_areas[XSAVE_STATE_AREA_COUNT] = {
},
};
-static uint32_t xsave_area_size(uint64_t mask, bool compacted)
+uint32_t xsave_area_size(uint64_t mask, bool compacted)
{
uint64_t ret = x86_ext_save_areas[0].size;
const ExtSaveArea *esa;
diff --git a/target/i386/cpu.h b/target/i386/cpu.h
index d4124973ce..9327353fff 100644
--- a/target/i386/cpu.h
+++ b/target/i386/cpu.h
@@ -2071,6 +2071,8 @@ void cpu_x86_fsave(CPUX86State *s, target_ulong ptr, int data32);
void cpu_x86_frstor(CPUX86State *s, target_ulong ptr, int data32);
void cpu_x86_fxsave(CPUX86State *s, target_ulong ptr);
void cpu_x86_fxrstor(CPUX86State *s, target_ulong ptr);
+void cpu_x86_xsave(CPUX86State *s, target_ulong ptr);
+void cpu_x86_xrstor(CPUX86State *s, target_ulong ptr);
/* cpu.c */
void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,
@@ -2327,6 +2329,7 @@ bool cpu_is_bsp(X86CPU *cpu);
void x86_cpu_xrstor_all_areas(X86CPU *cpu, const void *buf, uint32_t buflen);
void x86_cpu_xsave_all_areas(X86CPU *cpu, void *buf, uint32_t buflen);
+uint32_t xsave_area_size(uint64_t mask, bool compacted);
void x86_update_hflags(CPUX86State* env);
static inline bool hyperv_feat_enabled(X86CPU *cpu, int feat)
diff --git a/target/i386/tcg/fpu_helper.c b/target/i386/tcg/fpu_helper.c
index 30bc44fcf8..ad58931751 100644
--- a/target/i386/tcg/fpu_helper.c
+++ b/target/i386/tcg/fpu_helper.c
@@ -2502,18 +2502,6 @@ void helper_frstor(CPUX86State *env, target_ulong ptr, int data32)
do_frstor(env, ptr, data32, GETPC());
}
-#if defined(CONFIG_USER_ONLY)
-void cpu_x86_fsave(CPUX86State *env, target_ulong ptr, int data32)
-{
- do_fsave(env, ptr, data32, 0);
-}
-
-void cpu_x86_frstor(CPUX86State *env, target_ulong ptr, int data32)
-{
- do_frstor(env, ptr, data32, 0);
-}
-#endif
-
#define XO(X) offsetof(X86XSaveArea, X)
static void do_xsave_fpu(CPUX86State *env, target_ulong ptr, uintptr_t ra)
@@ -2787,21 +2775,8 @@ void helper_fxrstor(CPUX86State *env, target_ulong ptr)
do_fxrstor(env, ptr, GETPC());
}
-#if defined(CONFIG_USER_ONLY)
-void cpu_x86_fxsave(CPUX86State *env, target_ulong ptr)
+static void do_xrstor(CPUX86State *env, target_ulong ptr, uint64_t rfbm, uintptr_t ra)
{
- do_fxsave(env, ptr, 0);
-}
-
-void cpu_x86_fxrstor(CPUX86State *env, target_ulong ptr)
-{
- do_fxrstor(env, ptr, 0);
-}
-#endif
-
-void helper_xrstor(CPUX86State *env, target_ulong ptr, uint64_t rfbm)
-{
- uintptr_t ra = GETPC();
uint64_t xstate_bv, xcomp_bv, reserve0;
rfbm &= env->xcr0;
@@ -2894,6 +2869,43 @@ void helper_xrstor(CPUX86State *env, target_ulong ptr, uint64_t rfbm)
#undef XO
+void helper_xrstor(CPUX86State *env, target_ulong ptr, uint64_t rfbm)
+{
+ do_xrstor(env, ptr, rfbm, GETPC());
+}
+
+#if defined(CONFIG_USER_ONLY)
+void cpu_x86_fsave(CPUX86State *env, target_ulong ptr, int data32)
+{
+ do_fsave(env, ptr, data32, 0);
+}
+
+void cpu_x86_frstor(CPUX86State *env, target_ulong ptr, int data32)
+{
+ do_frstor(env, ptr, data32, 0);
+}
+
+void cpu_x86_fxsave(CPUX86State *env, target_ulong ptr)
+{
+ do_fxsave(env, ptr, 0);
+}
+
+void cpu_x86_fxrstor(CPUX86State *env, target_ulong ptr)
+{
+ do_fxrstor(env, ptr, 0);
+}
+
+void cpu_x86_xsave(CPUX86State *env, target_ulong ptr)
+{
+ do_xsave(env, ptr, -1, get_xinuse(env), -1, 0);
+}
+
+void cpu_x86_xrstor(CPUX86State *env, target_ulong ptr)
+{
+ do_xrstor(env, ptr, -1, 0);
+}
+#endif
+
uint64_t helper_xgetbv(CPUX86State *env, uint32_t ecx)
{
/* The OS must have enabled XSAVE. */
--
2.37.3
next prev parent reply other threads:[~2022-10-11 12:09 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 ` [PULL 36/37] linux-user: i386/signal: support FXSAVE fpstate on 32-bit emulation Paolo Bonzini
2022-10-11 10:27 ` Paolo Bonzini [this message]
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=20221011102700.319178-38-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).