[permalink] [raw]

Subject: [RFC v2 7/8] x86, xsave: add kernel support for AMDs Lightweight Profiling (LWP)

This patch extends the xsave structure to support the LWP state. The
xstate feature bit for LWP is added to XCNTXT_NONLAZY, thereby enabling
kernel support for saving/restoring LWP state. The LWP state is also
saved/restored on signal entry/return, just like all other xstates. LWP
state needs to be reset (disabled) when on entry in a signal handler.

Signed-off-by: Hans Rosenfeld <[email protected]>
---
arch/x86/include/asm/msr-index.h | 1 +
arch/x86/include/asm/processor.h | 12 ++++++++++++
arch/x86/include/asm/sigcontext.h | 12 ++++++++++++
arch/x86/include/asm/xsave.h | 3 ++-
arch/x86/kernel/xsave.c | 2 ++
5 files changed, 29 insertions(+), 1 deletions(-)

diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index 823d482..0ba2150 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -130,6 +130,7 @@
#define MSR_AMD64_IBSDCPHYSAD 0xc0011039
#define MSR_AMD64_IBSCTL 0xc001103a
#define MSR_AMD64_IBSBRTARGET 0xc001103b
+#define MSR_AMD64_LWP_CBADDR 0xc0000106

/* Fam 15h MSRs */
#define MSR_F15H_PERF_CTL 0xc0010200
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index 4c25ab4..df2cbd4 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -353,6 +353,17 @@ struct ymmh_struct {
u32 ymmh_space[64];
};

+struct lwp_struct {
+ u64 lwpcb_addr;
+ u32 flags;
+ u32 buf_head_offset;
+ u64 buf_base;
+ u32 buf_size;
+ u32 filters;
+ u64 saved_event_record[4];
+ u32 event_counter[16];
+};
+
struct xsave_hdr_struct {
u64 xstate_bv;
u64 reserved1[2];
@@ -363,6 +374,7 @@ struct xsave_struct {
struct i387_fxsave_struct i387;
struct xsave_hdr_struct xsave_hdr;
struct ymmh_struct ymmh;
+ struct lwp_struct lwp;
/* new processor state extensions will go here */
} __attribute__ ((packed, aligned (64)));

diff --git a/arch/x86/include/asm/sigcontext.h b/arch/x86/include/asm/sigcontext.h
index 04459d2..0a58b82 100644
--- a/arch/x86/include/asm/sigcontext.h
+++ b/arch/x86/include/asm/sigcontext.h
@@ -274,6 +274,17 @@ struct _ymmh_state {
__u32 ymmh_space[64];
};

+struct _lwp_state {
+ __u64 lwpcb_addr;
+ __u32 flags;
+ __u32 buf_head_offset;
+ __u64 buf_base;
+ __u32 buf_size;
+ __u32 filters;
+ __u64 saved_event_record[4];
+ __u32 event_counter[16];
+};
+
/*
* Extended state pointed by the fpstate pointer in the sigcontext.
* In addition to the fpstate, information encoded in the xstate_hdr
@@ -284,6 +295,7 @@ struct _xstate {
struct _fpstate fpstate;
struct _xsave_hdr xstate_hdr;
struct _ymmh_state ymmh;
+ struct _lwp_state lwp;
/* new processor state extensions go here */
};

diff --git a/arch/x86/include/asm/xsave.h b/arch/x86/include/asm/xsave.h
index 4ccee3c..be89f0e 100644
--- a/arch/x86/include/asm/xsave.h
+++ b/arch/x86/include/asm/xsave.h
@@ -9,6 +9,7 @@
#define XSTATE_FP 0x1
#define XSTATE_SSE 0x2
#define XSTATE_YMM 0x4
+#define XSTATE_LWP (1ULL << 62)

#define XSTATE_FPSSE (XSTATE_FP | XSTATE_SSE)

@@ -24,7 +25,7 @@
* These are the features that the OS can handle currently.
*/
#define XCNTXT_LAZY (XSTATE_FP | XSTATE_SSE | XSTATE_YMM)
-#define XCNTXT_NONLAZY 0
+#define XCNTXT_NONLAZY (XSTATE_LWP)

#define XCNTXT_MASK (XCNTXT_LAZY | XCNTXT_NONLAZY)

diff --git a/arch/x86/kernel/xsave.c b/arch/x86/kernel/xsave.c
index 7b08d32..d3dc65e 100644
--- a/arch/x86/kernel/xsave.c
+++ b/arch/x86/kernel/xsave.c
@@ -177,6 +177,8 @@ int save_xstates_sigframe(void __user *buf, unsigned int size)
(struct _fpstate_ia32 __user *) buf) ? -1 : 1;

save_xstates(tsk);
+ if (pcntxt_mask & XSTATE_LWP)
+ wrmsrl(MSR_AMD64_LWP_CBADDR, 0);
if (use_xsaveopt())
sanitize_i387_state(tsk);

--
1.5.6.5

2011-03-23 15:28:38

by Hans Rosenfeld

[permalink] [raw]

Subject: [RFC v2 4/8] x86, xsave: remove unused code

The patches to rework the fpu/xsave handling and signal frame setup have
made a lot of code unused. This patch removes all this now useless stuff.

Signed-off-by: Hans Rosenfeld <[email protected]>
---
arch/x86/include/asm/i387.h | 155 ++----------------------------
arch/x86/include/asm/xsave.h | 51 ----------
arch/x86/kernel/i387.c | 221 ------------------------------------------
arch/x86/kernel/traps.c | 22 ----
arch/x86/kernel/xsave.c | 163 -------------------------------
5 files changed, 7 insertions(+), 605 deletions(-)

diff --git a/arch/x86/include/asm/i387.h b/arch/x86/include/asm/i387.h
index 30930bf..97867ea 100644
--- a/arch/x86/include/asm/i387.h
+++ b/arch/x86/include/asm/i387.h
@@ -29,8 +29,6 @@
# include <asm/sigcontext32.h>
# include <asm/user32.h>
#else
-# define save_i387_xstate_ia32 save_i387_xstate
-# define restore_i387_xstate_ia32 restore_i387_xstate
# define _fpstate_ia32 _fpstate
# define _xstate_ia32 _xstate
# define sig_xstate_ia32_size sig_xstate_size
@@ -108,75 +106,16 @@ static inline void sanitize_i387_state(struct task_struct *tsk)
}

#ifdef CONFIG_X86_64
-static inline int fxrstor_checking(struct i387_fxsave_struct *fx)
+static inline void fxrstor(struct i387_fxsave_struct *fx)
{
- int err;
-
- /* See comment in fxsave() below. */
-#ifdef CONFIG_AS_FXSAVEQ
- asm volatile("1: fxrstorq %[fx]\n\t"
- "2:\n"
- ".section .fixup,\"ax\"\n"
- "3: movl $-1,%[err]\n"
- " jmp 2b\n"
- ".previous\n"
- _ASM_EXTABLE(1b, 3b)
- : [err] "=r" (err)
- : [fx] "m" (*fx), "0" (0));
-#else
- asm volatile("1: rex64/fxrstor (%[fx])\n\t"
- "2:\n"
- ".section .fixup,\"ax\"\n"
- "3: movl $-1,%[err]\n"
- " jmp 2b\n"
- ".previous\n"
- _ASM_EXTABLE(1b, 3b)
- : [err] "=r" (err)
- : [fx] "R" (fx), "m" (*fx), "0" (0));
-#endif
- return err;
-}
-
-static inline int fxsave_user(struct i387_fxsave_struct __user *fx)
-{
- int err;
-
- /*
- * Clear the bytes not touched by the fxsave and reserved
- * for the SW usage.
- */
- err = __clear_user(&fx->sw_reserved,
- sizeof(struct _fpx_sw_bytes));
- if (unlikely(err))
- return -EFAULT;
-
/* See comment in fxsave() below. */
#ifdef CONFIG_AS_FXSAVEQ
- asm volatile("1: fxsaveq %[fx]\n\t"
- "2:\n"
- ".section .fixup,\"ax\"\n"
- "3: movl $-1,%[err]\n"
- " jmp 2b\n"
- ".previous\n"
- _ASM_EXTABLE(1b, 3b)
- : [err] "=r" (err), [fx] "=m" (*fx)
- : "0" (0));
+ asm volatile("fxrstorq %[fx]\n\t"
+ : : [fx] "m" (*fx));
#else
- asm volatile("1: rex64/fxsave (%[fx])\n\t"
- "2:\n"
- ".section .fixup,\"ax\"\n"
- "3: movl $-1,%[err]\n"
- " jmp 2b\n"
- ".previous\n"
- _ASM_EXTABLE(1b, 3b)
- : [err] "=r" (err), "=m" (*fx)
- : [fx] "R" (fx), "0" (0));
+ asm volatile("rex64/fxrstor (%[fx])\n\t"
+ : : [fx] "R" (fx), "m" (*fx));
#endif
- if (unlikely(err) &&
- __clear_user(fx, sizeof(struct i387_fxsave_struct)))
- err = -EFAULT;
- /* No need to clear here because the caller clears USED_MATH */
- return err;
}

static inline void fpu_fxsave(struct fpu *fpu)
@@ -209,7 +148,7 @@ static inline void fpu_fxsave(struct fpu *fpu)
#else /* CONFIG_X86_32 */

/* perform fxrstor iff the processor has extended states, otherwise frstor */
-static inline int fxrstor_checking(struct i387_fxsave_struct *fx)
+static inline void fxrstor(struct i387_fxsave_struct *fx)
{
/*
* The "nop" is needed to make the instructions the same
@@ -220,8 +159,6 @@ static inline int fxrstor_checking(struct i387_fxsave_struct *fx)
"fxrstor %1",
X86_FEATURE_FXSR,
"m" (*fx));
-
- return 0;
}

static inline void fpu_fxsave(struct fpu *fpu)
@@ -246,7 +183,7 @@ static inline void fpu_fxsave(struct fpu *fpu)
*/
static inline void fpu_restore(struct fpu *fpu)
{
- fxrstor_checking(&fpu->state->fxsave);
+ fxrstor(&fpu->state->fxsave);
}

static inline void fpu_save(struct fpu *fpu)
@@ -278,69 +215,6 @@ static inline void fpu_clean(struct fpu *fpu)
[addr] "m" (safe_address));
}

-static inline void fpu_save_init(struct fpu *fpu)
-{
- if (use_xsave()) {
- struct xsave_struct *xstate = &fpu->state->xsave;
-
- fpu_xsave(xstate, -1);
-
- /*
- * xsave header may indicate the init state of the FP.
- */
- if (!(xstate->xsave_hdr.xstate_bv & XSTATE_FP))
- return;
- } else if (use_fxsr()) {
- fpu_fxsave(fpu);
- } else {
- asm volatile("fsave %[fx]; fwait"
- : [fx] "=m" (fpu->state->fsave));
- return;
- }
-
- if (unlikely(fpu->state->fxsave.swd & X87_FSW_ES))
- asm volatile("fnclex");
-
- /* AMD K7/K8 CPUs don't save/restore FDP/FIP/FOP unless an exception
- is pending. Clear the x87 state here by setting it to fixed
- values. safe_address is a random variable that should be in L1 */
- alternative_input(
- ASM_NOP8 ASM_NOP2,
- "emms\n\t" /* clear stack tags */
- "fildl %P[addr]", /* set F?P to defined value */
- X86_FEATURE_FXSAVE_LEAK,
- [addr] "m" (safe_address));
-}
-
-static inline void __save_init_fpu(struct task_struct *tsk)
-{
- fpu_save_init(&tsk->thread.fpu);
- task_thread_info(tsk)->status &= ~TS_USEDFPU;
-}
-
-static inline int fpu_restore_checking(struct fpu *fpu)
-{
- if (use_xsave())
- return xrstor_checking(&fpu->state->xsave, -1);
- else
- return fxrstor_checking(&fpu->state->fxsave);
-}
-
-/*
- * Signal frame handlers...
- */
-extern int save_i387_xstate(void __user *buf);
-extern int restore_i387_xstate(void __user *buf);
-
-static inline void __unlazy_fpu(struct task_struct *tsk)
-{
- if (task_thread_info(tsk)->status & TS_USEDFPU) {
- __save_init_fpu(tsk);
- stts();
- } else
- tsk->fpu_counter = 0;
-}
-
static inline void __clear_fpu(struct task_struct *tsk)
{
if (task_thread_info(tsk)->status & TS_USEDFPU) {
@@ -409,21 +283,6 @@ static inline void irq_ts_restore(int TS_state)
/*
* These disable preemption on their own and are safe
*/
-static inline void save_init_fpu(struct task_struct *tsk)
-{
- preempt_disable();
- __save_init_fpu(tsk);
- stts();
- preempt_enable();
-}
-
-static inline void unlazy_fpu(struct task_struct *tsk)
-{
- preempt_disable();
- __unlazy_fpu(tsk);
- preempt_enable();
-}
-
static inline void clear_fpu(struct task_struct *tsk)
{
preempt_disable();
diff --git a/arch/x86/include/asm/xsave.h b/arch/x86/include/asm/xsave.h
index 200c56d..742da4a 100644
--- a/arch/x86/include/asm/xsave.h
+++ b/arch/x86/include/asm/xsave.h
@@ -51,26 +51,6 @@ extern int check_for_xstate(struct i387_fxsave_struct __user *buf,
unsigned int size,
struct _fpx_sw_bytes *sw);

-static inline int xrstor_checking(struct xsave_struct *fx, u64 mask)
-{
- int err;
- u32 lmask = mask;
- u32 hmask = mask >> 32;
-
- asm volatile("1: .byte " REX_PREFIX "0x0f,0xae,0x2f\n\t"
- "2:\n"
- ".section .fixup,\"ax\"\n"
- "3: movl $-1,%[err]\n"
- " jmp 2b\n"
- ".previous\n"
- _ASM_EXTABLE(1b, 3b)
- : [err] "=r" (err)
- : "D" (fx), "m" (*fx), "a" (lmask), "d" (hmask), "0" (0)
- : "memory");
-
- return err;
-}
-
static inline void xrstor_state(struct xsave_struct *fx, u64 mask)
{
u32 lmask = mask;
@@ -81,37 +61,6 @@ static inline void xrstor_state(struct xsave_struct *fx, u64 mask)
: "memory");
}

-static inline int xsave_checking(struct xsave_struct __user *buf)
-{
- int err;
-
- /*
- * Clear the xsave header first, so that reserved fields are
- * initialized to zero.
- */
- err = __clear_user(&buf->xsave_hdr,
- sizeof(struct xsave_hdr_struct));
- if (unlikely(err))
- return -EFAULT;
-
- asm volatile("1: .byte " REX_PREFIX "0x0f,0xae,0x27\n"
- "2:\n"
- ".section .fixup,\"ax\"\n"
- "3: movl $-1,%[err]\n"
- " jmp 2b\n"
- ".previous\n"
- _ASM_EXTABLE(1b,3b)
- : [err] "=r" (err)
- : "D" (buf), "a" (-1), "d" (-1), "0" (0)
- : "memory");
-
- if (unlikely(err) && __clear_user(buf, xstate_size))
- err = -EFAULT;
-
- /* No need to clear here because the caller clears USED_MATH */
- return err;
-}
-
static inline void xsave_state(struct xsave_struct *fx, u64 mask)
{
u32 lmask = mask;
diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c
index 5cec7c2..d2d2b69 100644
--- a/arch/x86/kernel/i387.c
+++ b/arch/x86/kernel/i387.c
@@ -490,227 +490,6 @@ int fpregs_set(struct task_struct *target, const struct user_regset *regset,
}

/*
- * Signal frame handlers.
- */
-
-static inline int save_i387_fsave(struct _fpstate_ia32 __user *buf)
-{
- struct task_struct *tsk = current;
- struct i387_fsave_struct *fp = &tsk->thread.fpu.state->fsave;
-
- fp->status = fp->swd;
- if (__copy_to_user(buf, fp, sizeof(struct i387_fsave_struct)))
- return -1;
- return 1;
-}
-
-static int save_i387_fxsave(struct _fpstate_ia32 __user *buf)
-{
- struct task_struct *tsk = current;
- struct i387_fxsave_struct *fx = &tsk->thread.fpu.state->fxsave;
- struct user_i387_ia32_struct env;
- int err = 0;
-
- convert_from_fxsr(&env, tsk);
- if (__copy_to_user(buf, &env, sizeof(env)))
- return -1;
-
- err |= __put_user(fx->swd, &buf->status);
- err |= __put_user(X86_FXSR_MAGIC, &buf->magic);
- if (err)
- return -1;
-
- if (__copy_to_user(&buf->_fxsr_env[0], fx, xstate_size))
- return -1;
- return 1;
-}
-
-static int save_i387_xsave(void __user *buf)
-{
- struct task_struct *tsk = current;
- struct _fpstate_ia32 __user *fx = buf;
- int err = 0;
-
-
- sanitize_i387_state(tsk);
-
- /*
- * For legacy compatible, we always set FP/SSE bits in the bit
- * vector while saving the state to the user context.
- * This will enable us capturing any changes(during sigreturn) to
- * the FP/SSE bits by the legacy applications which don't touch
- * xstate_bv in the xsave header.
- *
- * xsave aware applications can change the xstate_bv in the xsave
- * header as well as change any contents in the memory layout.
- * xrestore as part of sigreturn will capture all the changes.
- */
- tsk->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FPSSE;
-
- if (save_i387_fxsave(fx) < 0)
- return -1;
-
- err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved_ia32,
- sizeof(struct _fpx_sw_bytes));
- err |= __put_user(FP_XSTATE_MAGIC2,
- (__u32 __user *) (buf + sig_xstate_ia32_size
- - FP_XSTATE_MAGIC2_SIZE));
- if (err)
- return -1;
-
- return 1;
-}
-
-int save_i387_xstate_ia32(void __user *buf)
-{
- struct _fpstate_ia32 __user *fp = (struct _fpstate_ia32 __user *) buf;
- struct task_struct *tsk = current;
-
- if (!used_math())
- return 0;
-
- if (!access_ok(VERIFY_WRITE, buf, sig_xstate_ia32_size))
- return -EACCES;
- /*
- * This will cause a "finit" to be triggered by the next
- * attempted FPU operation by the 'current' process.
- */
- clear_used_math();
-
- if (!HAVE_HWFP) {
- return fpregs_soft_get(current, NULL,
- 0, sizeof(struct user_i387_ia32_struct),
- NULL, fp) ? -1 : 1;
- }
-
- preempt_disable();
- save_xstates(tsk);
- preempt_enable();
-
- if (cpu_has_xsave)
- return save_i387_xsave(fp);
- if (cpu_has_fxsr)
- return save_i387_fxsave(fp);
- else
- return save_i387_fsave(fp);
-}
-
-static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf)
-{
- struct task_struct *tsk = current;
-
- return __copy_from_user(&tsk->thread.fpu.state->fsave, buf,
- sizeof(struct i387_fsave_struct));
-}
-
-static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf,
- unsigned int size)
-{
- struct task_struct *tsk = current;
- struct user_i387_ia32_struct env;
- int err;
-
- err = __copy_from_user(&tsk->thread.fpu.state->fxsave, &buf->_fxsr_env[0],
- size);
- /* mxcsr reserved bits must be masked to zero for security reasons */
- tsk->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask;
- if (err || __copy_from_user(&env, buf, sizeof(env)))
- return 1;
- convert_to_fxsr(tsk, &env);
-
- return 0;
-}
-
-static int restore_i387_xsave(void __user *buf)
-{
- struct _fpx_sw_bytes fx_sw_user;
- struct _fpstate_ia32 __user *fx_user =
- ((struct _fpstate_ia32 __user *) buf);
- struct i387_fxsave_struct __user *fx =
- (struct i387_fxsave_struct __user *) &fx_user->_fxsr_env[0];
- struct xsave_hdr_struct *xsave_hdr =
- &current->thread.fpu.state->xsave.xsave_hdr;
- u64 mask;
- int err;
-
- if (check_for_xstate(fx, sig_xstate_ia32_size -
- offsetof(struct _fpstate_ia32, _fxsr_env),
- &fx_sw_user))
- goto fx_only;
-
- mask = fx_sw_user.xstate_bv;
-
- err = restore_i387_fxsave(buf, fx_sw_user.xstate_size);
-
- xsave_hdr->xstate_bv &= pcntxt_mask;
- /*
- * These bits must be zero.
- */
- xsave_hdr->reserved1[0] = xsave_hdr->reserved1[1] = 0;
-
- /*
- * Init the state that is not present in the memory layout
- * and enabled by the OS.
- */
- mask = ~(pcntxt_mask & ~mask);
- xsave_hdr->xstate_bv &= mask;
-
- return err;
-fx_only:
- /*
- * Couldn't find the extended state information in the memory
- * layout. Restore the FP/SSE and init the other extended state
- * enabled by the OS.
- */
- xsave_hdr->xstate_bv = XSTATE_FPSSE;
- return restore_i387_fxsave(buf, sizeof(struct i387_fxsave_struct));
-}
-
-int restore_i387_xstate_ia32(void __user *buf)
-{
- int err;
- struct task_struct *tsk = current;
- struct _fpstate_ia32 __user *fp = (struct _fpstate_ia32 __user *) buf;
-
- if (HAVE_HWFP)
- clear_fpu(tsk);
-
- if (!buf) {
- if (used_math()) {
- clear_fpu(tsk);
- clear_used_math();
- }
-
- return 0;
- } else
- if (!access_ok(VERIFY_READ, buf, sig_xstate_ia32_size))
- return -EACCES;
-
- if (!used_math()) {
- err = init_fpu(tsk);
- if (err)
- return err;
- }
-
- if (HAVE_HWFP) {
- if (cpu_has_xsave)
- err = restore_i387_xsave(buf);
- else if (cpu_has_fxsr)
- err = restore_i387_fxsave(fp, sizeof(struct
- i387_fxsave_struct));
- else
- err = restore_i387_fsave(fp);
- } else {
- err = fpregs_soft_set(current, NULL,
- 0, sizeof(struct user_i387_ia32_struct),
- NULL, fp) != 0;
- }
- set_used_math();
-
- return err;
-}
-
-/*
* FPU state for core dumps.
* This is only used for a.out dumps now.
* It is declared generically using elf_fpregset_t (which is
diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index 072c30e..872fc78 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -720,28 +720,6 @@ asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
}

/*
- * __math_state_restore assumes that cr0.TS is already clear and the
- * fpu state is all ready for use. Used during context switch.
- */
-void __math_state_restore(void)
-{
- struct thread_info *thread = current_thread_info();
- struct task_struct *tsk = thread->task;
-
- /*
- * Paranoid restore. send a SIGSEGV if we fail to restore the state.
- */
- if (unlikely(fpu_restore_checking(&tsk->thread.fpu))) {
- stts();
- force_sig(SIGSEGV, tsk);
- return;
- }
-
- thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
- tsk->fpu_counter++;
-}
-
-/*
* 'math_state_restore()' saves the current math information in the
* old math state array, and gets the new ones from the current task
*
diff --git a/arch/x86/kernel/xsave.c b/arch/x86/kernel/xsave.c
index 5d07a88..f2714ea 100644
--- a/arch/x86/kernel/xsave.c
+++ b/arch/x86/kernel/xsave.c
@@ -249,169 +249,6 @@ int save_xstates_sigframe(void __user *buf, unsigned int size)
return 1;
}

-#ifdef CONFIG_X86_64
-int save_i387_xstate(void __user *buf)
-{
- struct task_struct *tsk = current;
- int err = 0;
-
- if (!access_ok(VERIFY_WRITE, buf, sig_xstate_size))
- return -EACCES;
-
- BUG_ON(sig_xstate_size < xstate_size);
-
- if ((unsigned long)buf % 64)
- printk("save_i387_xstate: bad fpstate %p\n", buf);
-
- if (!used_math())
- return 0;
-
- if (task_thread_info(tsk)->status & TS_USEDFPU) {
- if (use_xsave())
- err = xsave_checking(buf);
- else
- err = fxsave_user(buf);
-
- if (err)
- return err;
- task_thread_info(tsk)->status &= ~TS_USEDFPU;
- stts();
- } else {
- sanitize_i387_state(tsk);
- if (__copy_to_user(buf, &tsk->thread.fpu.state->fxsave,
- xstate_size))
- return -1;
- }
-
- clear_used_math(); /* trigger finit */
-
- if (use_xsave()) {
- struct _fpstate __user *fx = buf;
- struct _xstate __user *x = buf;
- u64 xstate_bv;
-
- err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved,
- sizeof(struct _fpx_sw_bytes));
-
- err |= __put_user(FP_XSTATE_MAGIC2,
- (__u32 __user *) (buf + sig_xstate_size
- - FP_XSTATE_MAGIC2_SIZE));
-
- /*
- * Read the xstate_bv which we copied (directly from the cpu or
- * from the state in task struct) to the user buffers and
- * set the FP/SSE bits.
- */
- err |= __get_user(xstate_bv, &x->xstate_hdr.xstate_bv);
-
- /*
- * For legacy compatible, we always set FP/SSE bits in the bit
- * vector while saving the state to the user context. This will
- * enable us capturing any changes(during sigreturn) to
- * the FP/SSE bits by the legacy applications which don't touch
- * xstate_bv in the xsave header.
- *
- * xsave aware apps can change the xstate_bv in the xsave
- * header as well as change any contents in the memory layout.
- * xrestore as part of sigreturn will capture all the changes.
- */
- xstate_bv |= XSTATE_FPSSE;
-
- err |= __put_user(xstate_bv, &x->xstate_hdr.xstate_bv);
-
- if (err)
- return err;
- }
-
- return 1;
-}
-
-/*
- * Restore the extended state if present. Otherwise, restore the FP/SSE
- * state.
- */
-static int restore_user_xstate(void __user *buf)
-{
- struct _fpx_sw_bytes fx_sw_user;
- u64 mask;
- int err;
-
- if (((unsigned long)buf % 64) ||
- check_for_xstate(buf, sig_xstate_size, &fx_sw_user))
- goto fx_only;
-
- mask = fx_sw_user.xstate_bv;
-
- /*
- * restore the state passed by the user.
- */
- err = xrstor_checking((__force struct xsave_struct *)buf, mask);
- if (err)
- return err;
-
- /*
- * init the state skipped by the user.
- */
- mask = pcntxt_mask & ~mask;
- if (unlikely(mask))
- xrstor_state(init_xstate_buf, mask);
-
- return 0;
-
-fx_only:
- /*
- * couldn't find the extended state information in the
- * memory layout. Restore just the FP/SSE and init all
- * the other extended state.
- */
- xrstor_state(init_xstate_buf, pcntxt_mask & ~XSTATE_FPSSE);
- return fxrstor_checking((__force struct i387_fxsave_struct *)buf);
-}
-
-/*
- * This restores directly out of user space. Exceptions are handled.
- */
-int restore_i387_xstate(void __user *buf)
-{
- struct task_struct *tsk = current;
- int err = 0;
-
- if (!buf) {
- if (used_math())
- goto clear;
- return 0;
- } else
- if (!access_ok(VERIFY_READ, buf, sig_xstate_size))
- return -EACCES;
-
- if (!used_math()) {
- err = init_fpu(tsk);
- if (err)
- return err;
- }
-
- if (!(task_thread_info(current)->status & TS_USEDFPU)) {
- clts();
- task_thread_info(current)->status |= TS_USEDFPU;
- }
- if (use_xsave())
- err = restore_user_xstate(buf);
- else
- err = fxrstor_checking((__force struct i387_fxsave_struct *)
- buf);
- if (unlikely(err)) {
- /*
- * Encountered an error while doing the restore from the
- * user buffer, clear the fpu state.
- */
-clear:
- clear_fpu(tsk);
- clear_used_math();
- }
- return err;
-}
-#endif
-
int restore_xstates_sigframe(void __user *buf, unsigned int size)
{
#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
--
1.5.6.5

2011-03-23 15:28:05

by Hans Rosenfeld

[permalink] [raw]

Subject: [RFC v2 0/8] x86, xsave: rework of extended state handling, LWP support

Changes since last patch set:
* fixed race in signal return path that could lead to xstate corruptions
* avoid LWP state inconsitency after signal return by disabling LWP
when entering a signal handler

This patch set is a general cleanup and rework of the code related to
handling of FPU and other extended states.

All handling of extended states, including the FPU state, is now handled
by xsave/xrstor wrappers that fall back to fxsave/fxrstor, or even
fsave/frstor, if hardware support for those features is lacking.

Non-lazy xstates, which cannot be restored lazily, can now be easily
supported with almost no processing overhead. This makes adding basic
support for AMDs LWP almost trivial.

Since non-lazy xstates are inherently incompatible with lazy allocation
of the xstate area, the complete removal of lazy allocation to further
reduce code complexity should be considered. Since SSE-optimized library
functions are widely used today, most processes will have an xstate area
anyway, so the memory overhead wouldn't be big enough to be much of an
issue.

Hans Rosenfeld (8):
x86, xsave: cleanup fpu/xsave support
x86, xsave: rework fpu/xsave support
x86, xsave: cleanup fpu/xsave signal frame setup
x86, xsave: remove unused code
x86, xsave: more cleanups
x86, xsave: add support for non-lazy xstates
x86, xsave: add kernel support for AMDs Lightweight Profiling (LWP)
x86, xsave: remove lazy allocation of xstate area

arch/x86/ia32/ia32_signal.c | 4 +-
arch/x86/include/asm/i387.h | 243 ++++++++--------------------
arch/x86/include/asm/msr-index.h | 1 +
arch/x86/include/asm/processor.h | 12 ++
arch/x86/include/asm/sigcontext.h | 12 ++
arch/x86/include/asm/thread_info.h | 4 +-
arch/x86/include/asm/xsave.h | 100 +++----------
arch/x86/kernel/i387.c | 310 ++++--------------------------------
arch/x86/kernel/process_32.c | 29 ++---
arch/x86/kernel/process_64.c | 28 +---
arch/x86/kernel/signal.c | 4 +-
arch/x86/kernel/traps.c | 47 +-----
arch/x86/kernel/xsave.c | 311 +++++++++++++++++++++++-------------
arch/x86/kvm/vmx.c | 2 +-
arch/x86/kvm/x86.c | 11 +-
arch/x86/math-emu/fpu_entry.c | 8 +-
drivers/lguest/x86/core.c | 2 +-
17 files changed, 380 insertions(+), 748 deletions(-)

2011-03-23 15:28:57

by Hans Rosenfeld

[permalink] [raw]

Subject: [RFC v2 8/8] x86, xsave: remove lazy allocation of xstate area

This patch completely removes lazy allocation of the xstate area. All
tasks will always have an xstate area preallocated, just like they
already do when non-lazy features are present. The size of the xsave
area ranges from 112 to 960 bytes, depending on the xstates present and
enabled. Since it is common to use SSE etc. for optimization, the actual
overhead is expected to negligible.

This removes some of the special-case handling of non-lazy xstates. It
also greatly simplifies init_fpu() by removing the allocation code, the
check for presence of the xstate area or init_fpu() return value.

Signed-off-by: Hans Rosenfeld <[email protected]>
---
arch/x86/include/asm/i387.h | 12 +++-------
arch/x86/kernel/i387.c | 46 +++++++++++-----------------------------
arch/x86/kernel/traps.c | 16 +------------
arch/x86/kernel/xsave.c | 21 ++----------------
arch/x86/kvm/x86.c | 4 +-
arch/x86/math-emu/fpu_entry.c | 8 +-----
6 files changed, 26 insertions(+), 81 deletions(-)

diff --git a/arch/x86/include/asm/i387.h b/arch/x86/include/asm/i387.h
index 22ad24c..0448f45 100644
--- a/arch/x86/include/asm/i387.h
+++ b/arch/x86/include/asm/i387.h
@@ -40,7 +40,7 @@
extern unsigned int sig_xstate_size;
extern void fpu_init(void);
extern void mxcsr_feature_mask_init(void);
-extern int init_fpu(struct task_struct *child);
+extern void init_fpu(struct task_struct *child);
extern asmlinkage void math_state_restore(void);
extern int dump_fpu(struct pt_regs *, struct user_i387_struct *);

@@ -332,13 +332,9 @@ extern void fpu_finit(struct fpu *fpu);

static inline void fpu_clear(struct fpu *fpu)
{
- if (pcntxt_mask & XCNTXT_NONLAZY) {
- memset(fpu->state, 0, xstate_size);
- fpu_finit(fpu);
- set_used_math();
- } else {
- fpu_free(fpu);
- }
+ memset(fpu->state, 0, xstate_size);
+ fpu_finit(fpu);
+ set_used_math();
}

#endif /* __ASSEMBLY__ */
diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c
index 88fefba..32b3c8d 100644
--- a/arch/x86/kernel/i387.c
+++ b/arch/x86/kernel/i387.c
@@ -42,6 +42,8 @@ void __cpuinit mxcsr_feature_mask_init(void)

static void __cpuinit init_thread_xstate(void)
{
+ static union thread_xstate x;
+
/*
* Note that xstate_size might be overwriten later during
* xsave_init().
@@ -62,6 +64,9 @@ static void __cpuinit init_thread_xstate(void)
xstate_size = sizeof(struct i387_fxsave_struct);
else
xstate_size = sizeof(struct i387_fsave_struct);
+
+ init_task.thread.fpu.state = &x;
+ fpu_finit(&init_task.thread.fpu);
}

/*
@@ -127,30 +132,20 @@ EXPORT_SYMBOL_GPL(fpu_finit);
* value at reset if we support XMM instructions and then
* remeber the current task has used the FPU.
*/
-int init_fpu(struct task_struct *tsk)
+void init_fpu(struct task_struct *tsk)
{
- int ret;
-
if (tsk_used_math(tsk)) {
if (HAVE_HWFP && tsk == current) {
preempt_disable();
save_xstates(tsk);
preempt_enable();
}
- return 0;
+ return;
}

- /*
- * Memory allocation at the first usage of the FPU and other state.
- */
- ret = fpu_alloc(&tsk->thread.fpu);
- if (ret)
- return ret;
-
fpu_finit(&tsk->thread.fpu);

set_stopped_child_used_math(tsk);
- return 0;
}
EXPORT_SYMBOL_GPL(init_fpu);

@@ -173,14 +168,10 @@ int xfpregs_get(struct task_struct *target, const struct user_regset *regset,
unsigned int pos, unsigned int count,
void *kbuf, void __user *ubuf)
{
- int ret;
-
if (!cpu_has_fxsr)
return -ENODEV;

- ret = init_fpu(target);
- if (ret)
- return ret;
+ init_fpu(target);

if (use_xsaveopt())
sanitize_i387_state(target);
@@ -198,9 +189,7 @@ int xfpregs_set(struct task_struct *target, const struct user_regset *regset,
if (!cpu_has_fxsr)
return -ENODEV;

- ret = init_fpu(target);
- if (ret)
- return ret;
+ init_fpu(target);

if (use_xsaveopt())
sanitize_i387_state(target);
@@ -232,9 +221,7 @@ int xstateregs_get(struct task_struct *target, const struct user_regset *regset,
if (!cpu_has_xsave)
return -ENODEV;

- ret = init_fpu(target);
- if (ret)
- return ret;
+ init_fpu(target);

/*
* Copy the 48bytes defined by the software first into the xstate
@@ -262,9 +249,7 @@ int xstateregs_set(struct task_struct *target, const struct user_regset *regset,
if (!cpu_has_xsave)
return -ENODEV;

- ret = init_fpu(target);
- if (ret)
- return ret;
+ init_fpu(target);

ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
&target->thread.fpu.state->xsave, 0, -1);
@@ -427,11 +412,8 @@ int fpregs_get(struct task_struct *target, const struct user_regset *regset,
void *kbuf, void __user *ubuf)
{
struct user_i387_ia32_struct env;
- int ret;

- ret = init_fpu(target);
- if (ret)
- return ret;
+ init_fpu(target);

if (!HAVE_HWFP)
return fpregs_soft_get(target, regset, pos, count, kbuf, ubuf);
@@ -462,9 +444,7 @@ int fpregs_set(struct task_struct *target, const struct user_regset *regset,
struct user_i387_ia32_struct env;
int ret;

- ret = init_fpu(target);
- if (ret)
- return ret;
+ init_fpu(target);

if (use_xsaveopt())
sanitize_i387_state(target);
diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index 872fc78..c8fbd04 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -734,20 +734,8 @@ asmlinkage void math_state_restore(void)
struct thread_info *thread = current_thread_info();
struct task_struct *tsk = thread->task;

- if (!tsk_used_math(tsk)) {
- local_irq_enable();
- /*
- * does a slab alloc which can sleep
- */
- if (init_fpu(tsk)) {
- /*
- * ran out of memory!
- */
- do_group_exit(SIGKILL);
- return;
- }
- local_irq_disable();
- }
+ if (!tsk_used_math(tsk))
+ init_fpu(tsk);

restore_xstates(tsk, XCNTXT_LAZY);
}
diff --git a/arch/x86/kernel/xsave.c b/arch/x86/kernel/xsave.c
index d3dc65e..81f54e9 100644
--- a/arch/x86/kernel/xsave.c
+++ b/arch/x86/kernel/xsave.c
@@ -264,7 +264,6 @@ int restore_xstates_sigframe(void __user *buf, unsigned int size)
struct _fpstate_ia32 __user *fp = buf;
struct xsave_struct *xsave;
u64 xstate_mask = pcntxt_mask & XCNTXT_NONLAZY;
- int err;

if (!buf) {
if (used_math()) {
@@ -277,11 +276,8 @@ int restore_xstates_sigframe(void __user *buf, unsigned int size)
if (!access_ok(VERIFY_READ, buf, size))
return -EACCES;

- if (!used_math()) {
- err = init_fpu(tsk);
- if (err)
- return err;
- }
+ if (!used_math())
+ init_fpu(tsk);

if (!HAVE_HWFP) {
set_used_math();
@@ -481,13 +477,8 @@ static void __init xstate_enable_boot_cpu(void)
"cntxt size 0x%x\n",
pcntxt_mask, xstate_size);

- if (pcntxt_mask & XCNTXT_NONLAZY) {
- static union thread_xstate x;
-
+ if (pcntxt_mask & XCNTXT_NONLAZY)
task_thread_info(&init_task)->xstate_mask |= XCNTXT_NONLAZY;
- init_task.thread.fpu.state = &x;
- fpu_finit(&init_task.thread.fpu);
- }
}

/*
@@ -530,9 +521,6 @@ void save_xstates(struct task_struct *tsk)
{
struct thread_info *ti = task_thread_info(tsk);

- if (!fpu_allocated(&tsk->thread.fpu))
- return;
-
xsave(&tsk->thread.fpu.state->xsave, ti->xstate_mask);

if (!(ti->xstate_mask & XCNTXT_LAZY))
@@ -566,9 +554,6 @@ void restore_xstates(struct task_struct *tsk, u64 mask)
{
struct thread_info *ti = task_thread_info(tsk);

- if (!fpu_allocated(&tsk->thread.fpu))
- return;
-
xrstor(&tsk->thread.fpu.state->xsave, mask);

ti->xstate_mask |= mask;
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 10aeb04..bd71b12 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -5377,8 +5377,8 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
int r;
sigset_t sigsaved;

- if (!tsk_used_math(current) && init_fpu(current))
- return -ENOMEM;
+ if (!tsk_used_math(current))
+ init_fpu(current);

if (vcpu->sigset_active)
sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
diff --git a/arch/x86/math-emu/fpu_entry.c b/arch/x86/math-emu/fpu_entry.c
index 7718541..472e2b9 100644
--- a/arch/x86/math-emu/fpu_entry.c
+++ b/arch/x86/math-emu/fpu_entry.c
@@ -147,12 +147,8 @@ void math_emulate(struct math_emu_info *info)
unsigned long code_limit = 0; /* Initialized to stop compiler warnings */
struct desc_struct code_descriptor;

- if (!used_math()) {
- if (init_fpu(current)) {
- do_group_exit(SIGKILL);
- return;
- }
- }
+ if (!used_math())
+ init_fpu(current);

#ifdef RE_ENTRANT_CHECKING
if (emulating) {
--
1.5.6.5

2011-03-23 15:28:59

by Hans Rosenfeld

[permalink] [raw]

[permalink] [raw]

Subject: Re: [RFC v3 0/8] x86, xsave: rework of extended state handling, LWP support

On Tue, May 17, 2011 at 01:30:20PM +0200, Ingo Molnar wrote:
>
> * Hans Rosenfeld <[email protected]> wrote:
>
> > Hi,
> >
> > On Thu, Apr 07, 2011 at 03:23:05AM -0400, Ingo Molnar wrote:
> > >
> > > FYI, the bits in tip:x86/xsave crash on boot on an AMD X2 testbox:
> >
> > > Full crashlog and kernel config attached. I've excluded x86/save from
> > > tip:master for now.
> >
> > this issue has been fixed a few weeks ago.
> >
> > Are there any plans to include x86/xsave into tip:master again?
>
> Regarding the LWP bits, that branch was indeed excluded because of that crash,
> while re-checking the branch today i noticed at least one serious design error
> in it, which makes me reconsider the whole thing:

Independend of all the concerns and useful comments regarding LWP ...
it seems to me that patches 1-7 are still of use (cleaning up the
save/restore code making it more maintainable).

Wouldn't it make sense to add patches 1-7 to a branch that is tested
with linux-next to find potential regressions?

Andreas