x86/fpu: Remove the thread::fpu pointer
Subsystems (7)
v3 cover x86/fpu: Remove the thread::fpu pointer
This series is one of the dependencies of the fast-headers work, which aims to reduce header complexity by removing <asm/processor.h> from the <linux/sched.h> dependency chain, which headers are headers are fat enough already even if we do not combine them.
To achieve that decoupling, one of the key steps is to not embedd any C types from <asm/processor.h> into task_struct.
The only architecture that relies on that in a serious fashion is x86, via 'struct thread::fpu', and the series below attempts to resolve it by using a calculated &task->thread.fpu value via the x86_task_fpu() helper.
Code generation: without CONFIG_X86_DEBUG_FPU=y we get a small reduction:
text data bss dec hex filename 26475783 10439082 1740804 38655669 24dd6b5 vmlinux.before 26475601 10435146 1740804 38651551 24dc69f vmlinux.after
With the new debug code - which I think we'll remove soon-ish, there's an expected increase:
text data bss dec hex filename 26476198 10439286 1740804 38656288 24dd920 vmlinux.CONFIG_X86_DEBUG_FPU.before 26477000 10435458 1740804 38653262 24dcd4e vmlinux.CONFIG_X86_DEBUG_FPU.after
The Git tree can be found at:
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git WIP.x86/fpu
Thanks,
Ingo
===============>
Ingo Molnar (3):
x86/fpu: Make task_struct::thread constant size
x86/fpu: Remove the thread::fpu pointer
x86/fpu: Remove init_task FPU state dependencies, add debugging warning
arch/x86/include/asm/fpu/sched.h | 2 +- arch/x86/include/asm/processor.h | 19 ++++++++++--------- arch/x86/kernel/fpu/context.h | 4 ++-- arch/x86/kernel/fpu/core.c | 57 +++++++++++++++++++++++++++++++-------------------------- arch/x86/kernel/fpu/init.c | 23 +++++++++++++---------- arch/x86/kernel/fpu/regset.c | 22 +++++++++++----------- arch/x86/kernel/fpu/signal.c | 18 +++++++++--------- arch/x86/kernel/fpu/xstate.c | 23 ++++++++++------------- arch/x86/kernel/fpu/xstate.h | 6 +++--- arch/x86/kernel/process.c | 6 ++---- arch/x86/kernel/signal.c | 6 +++--- arch/x86/kernel/traps.c | 2 +- arch/x86/math-emu/fpu_aux.c | 2 +- arch/x86/math-emu/fpu_entry.c | 4 ++-- arch/x86/math-emu/fpu_system.h | 2 +- arch/x86/mm/extable.c | 2 +- include/linux/sched.h | 13 +++---------- 17 files changed, 104 insertions(+), 107 deletions(-)
Signature · 1 line
2.43.0
PATCH v3 1/3 x86/fpu: Make task_struct::thread constant size
Turn thread.fpu into a pointer. Since most FPU code internals work by passing around the FPU pointer already, the code generation impact is small.
This allows us to remove the old kludge of task_struct being variable size:
struct task_struct { ...
/*
* New fields for task_struct should be added above here, so that
* they are included in the randomized portion of task_struct.
*/
randomized_struct_fields_end
/* CPU-specific state of this task: */
struct thread_struct thread;
/*
* WARNING: on x86, 'thread_struct' contains a variable-sized
* structure. It *MUST* be at the end of 'task_struct'.
*
* Do not put anything below here!
*/};
... which creates a number of problems, such as requiring thread_struct to be the last member of the struct - not allowing it to be struct-randomized, etc.
But the primary motivation is to allow the decoupling of task_struct from hardware details (<asm/processor.h> in particular), and to eventually allow the per-task infrastructure:
DECLARE_PER_TASK(type, name); ... per_task(current, name) = val;
... which requires task_struct to be a constant size struct.
The fpu_thread_struct_whitelist() quirk to hardened usercopy can be removed, now that the FPU structure is not embedded in the task struct anymore, which reduces text footprint a bit.
--- arch/x86/include/asm/fpu/sched.h | 2 +- arch/x86/include/asm/processor.h | 14 ++++++-------- arch/x86/kernel/fpu/context.h | 4 ++-- arch/x86/kernel/fpu/core.c | 51 ++++++++++++++++++++++++++------------------------- arch/x86/kernel/fpu/init.c | 25 +++++++++++++++---------- arch/x86/kernel/fpu/regset.c | 22 +++++++++++----------- arch/x86/kernel/fpu/signal.c | 18 +++++++++--------- arch/x86/kernel/fpu/xstate.c | 22 +++++++++++----------- arch/x86/kernel/fpu/xstate.h | 6 +++--- arch/x86/kernel/process.c | 6 +++--- arch/x86/kernel/signal.c | 6 +++--- arch/x86/kernel/traps.c | 2 +- arch/x86/math-emu/fpu_aux.c | 2 +- arch/x86/math-emu/fpu_entry.c | 4 ++-- arch/x86/math-emu/fpu_system.h | 2 +- arch/x86/mm/extable.c | 2 +- include/linux/sched.h | 13 +++---------- 17 files changed, 99 insertions(+), 102 deletions(-)
arch/x86/include/asm/fpu/sched.h +1 -1
@@ -41,7 +41,7 @@ static inline void switch_fpu_prepare(struct task_struct *old, int cpu) {if (cpu_feature_enabled (X86_FEATURE_FPU ) && !(old->flags & (PF_KTHREAD |PF_USER_WORKER ))) { -struct fpu *old_fpu = &old->thread.fpu; +struct fpu *old_fpu = old->thread.fpu;save_fpregs_to_fpstate (old_fpu);/*
arch/x86/include/asm/processor.h +6 -8
@@ -504,19 +504,17 @@ struct thread_struct {#endif /* Floating point and extended processor state */ -struct fpu fpu; -/* -* WARNING: 'fpu' is dynamically-sized. It *MUST* be at -* the end. -*/ +struct fpu *fpu; }; -extern void fpu_thread_struct_whitelist (unsigned long *offset,unsigned long *size); - +/* +* X86 doesn't need any embedded-FPU-struct quirks: +*/ static inline void arch_thread_struct_whitelist (unsigned long *offset,unsigned long *size) { -fpu_thread_struct_whitelist (offset, size); + *offset =0 ; + *size =0 ; } static inline void
arch/x86/kernel/fpu/context.h +2 -2
@@ -53,7 +53,7 @@ static inline void fpregs_activate(struct fpu *fpu)/* Internal helper for switch_fpu_return() and signal frame setup */ static inline void fpregs_restore_userregs (void ) { -struct fpu *fpu = ¤t->thread.fpu; +struct fpu *fpu = current->thread.fpu;int cpu =smp_processor_id ();if (WARN_ON_ONCE (current->flags & (PF_KTHREAD |PF_USER_WORKER )))
@@ -67,7 +67,7 @@ static inline void fpregs_restore_userregs(void)* If PKRU is enabled, then the PKRU value is already * correct because it was either set in switch_to() or in * flush_thread(). So it is excluded because it might be -* not up to date in current->thread.fpu.xsave state. +* not up to date in current->thread.fpu->xsave state. * * XFD state is handled in restore_fpregs_from_fpstate(). */
arch/x86/kernel/fpu/core.c +26 -25
141 lines not on this page · raw message
arch/x86/kernel/fpu/init.c +15 -10
@@ -38,7 +38,7 @@ static void fpu__init_cpu_generic(void)/* Flush out any pending x87 state: */ #ifdef CONFIG_MATH_EMULATION if (!boot_cpu_has (X86_FEATURE_FPU )) -fpstate_init_soft (¤t->thread.fpu.fpstate->regs.soft); +fpstate_init_soft (¤t->thread.fpu->fpstate->regs.soft); else#endif asmvolatile ("fninit" );
@@ -71,8 +71,17 @@ static bool __init fpu__probe_without_cpuid(void)return fsw ==0 && (fcw &0x103f ) ==0x003f ; } +static struct fpu x86_init_fpu__read_mostly ; +static void __init fpu__init_system_early_generic (void ) { +int this_cpu =smp_processor_id (); + +fpstate_reset (&x86_init_fpu); + current->thread.fpu = &x86_init_fpu; +per_cpu (fpu_fpregs_owner_ctx, this_cpu) = &x86_init_fpu; + x86_init_fpu.last_cpu = this_cpu; +if (!boot_cpu_has (X86_FEATURE_CPUID ) && !test_bit (X86_FEATURE_FPU , (unsigned long *)cpu_caps_cleared)) {if (fpu__probe_without_cpuid ())
@@ -150,11 +159,13 @@ static void __init fpu__init_task_struct_size(void) {int task_size =sizeof (struct task_struct ); + task_size +=sizeof (struct fpu ); +/* * Subtract off the static size of the register state. * It potentially has a bunch of padding. */ - task_size -=sizeof (current->thread.fpu.__fpstate.regs); + task_size -=sizeof (current->thread.fpu->__fpstate.regs);/* * Add back the dynamically-calculated register state
@@ -164,14 +175,9 @@ static void __init fpu__init_task_struct_size(void)/* * We dynamically size 'struct fpu', so we require that -* it be at the end of 'thread_struct' and that -* 'thread_struct' be at the end of 'task_struct'. If -* you hit a compile error here, check the structure to -* see if something got added to the end. +* 'state' be at the end of 'it: */ CHECK_MEMBER_AT_END_OF(struct fpu, __fpstate); - CHECK_MEMBER_AT_END_OF(struct thread_struct, fpu); - CHECK_MEMBER_AT_END_OF(struct task_struct, thread); arch_task_struct_size = task_size; }
@@ -204,7 +210,7 @@ static void __init fpu__init_system_xstate_size_legacy(void) fpu_kernel_cfg.default_size = size; fpu_user_cfg.max_size = size; fpu_user_cfg.default_size = size; -fpstate_reset (¤t->thread.fpu); +fpstate_reset (current->thread.fpu); }/*
@@ -213,7 +219,6 @@ static void __init fpu__init_system_xstate_size_legacy(void)*/ void __init fpu__init_system (void ) { -fpstate_reset (¤t->thread.fpu);fpu__init_system_early_generic ();/*
arch/x86/kernel/fpu/regset.c +11 -11
@@ -45,7 +45,7 @@ int regset_xregset_fpregs_active(struct task_struct *target, const struct user_r*/ static void sync_fpstate (struct fpu *fpu) { -if (fpu == ¤t->thread.fpu) +if (fpu == current->thread.fpu)fpu_sync_fpstate (fpu); }
@@ -63,7 +63,7 @@ static void fpu_force_restore(struct fpu *fpu)* Only stopped child tasks can be used to modify the FPU * state in the fpstate buffer: */ -WARN_ON_FPU (fpu == ¤t->thread.fpu); +WARN_ON_FPU (fpu == current->thread.fpu);__fpu_invalidate_fpregs_state (fpu); }
@@ -71,7 +71,7 @@ static void fpu_force_restore(struct fpu *fpu)int xfpregs_get (struct task_struct *target,const struct user_regset *regset,struct membuf to) { -struct fpu *fpu = &target->thread.fpu; +struct fpu *fpu = target->thread.fpu;if (!cpu_feature_enabled (X86_FEATURE_FXSR ))return -ENODEV ;
@@ -91,7 +91,7 @@ int xfpregs_set(struct task_struct *target, const struct user_regset *regset, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - struct fpu *fpu = &target->thread.fpu; + struct fpu *fpu = target->thread.fpu;struct fxregs_state newstate;int ret;
@@ -133,7 +133,7 @@ int xstateregs_get(struct task_struct *target, const struct user_regset *regset,if (!cpu_feature_enabled (X86_FEATURE_XSAVE ))return -ENODEV ; -sync_fpstate (&target->thread.fpu); +sync_fpstate (target->thread.fpu);copy_xstate_to_uabi_buf (to, target,XSTATE_COPY_XSAVE );return 0 ;
@@ -143,7 +143,7 @@ int xstateregs_set(struct task_struct *target, const struct user_regset *regset, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - struct fpu *fpu = &target->thread.fpu; + struct fpu *fpu = target->thread.fpu;struct xregs_state *tmpbuf =NULL ;int ret;
@@ -187,7 +187,7 @@ int ssp_active(struct task_struct *target, const struct user_regset *regset)int ssp_get (struct task_struct *target,const struct user_regset *regset,struct membuf to) { -struct fpu *fpu = &target->thread.fpu; +struct fpu *fpu = target->thread.fpu;struct cet_user_state *cetregs;if (!cpu_feature_enabled (X86_FEATURE_USER_SHSTK ))
@@ -213,7 +213,7 @@ int ssp_set(struct task_struct *target, const struct user_regset *regset, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - struct fpu *fpu = &target->thread.fpu; + struct fpu *fpu = target->thread.fpu;struct xregs_state *xsave = &fpu->fpstate->regs.xsave;struct cet_user_state *cetregs;unsigned long user_ssp;
@@ -367,7 +367,7 @@ static void __convert_from_fxsr(struct user_i387_ia32_struct *env,void convert_from_fxsr (struct user_i387_ia32_struct *env,struct task_struct *tsk) { -__convert_from_fxsr (env, tsk, &tsk->thread.fpu.fpstate->regs.fxsave); +__convert_from_fxsr (env, tsk, &tsk->thread.fpu->fpstate->regs.fxsave); } void convert_to_fxsr(struct fxregs_state *fxsave,
@@ -400,7 +400,7 @@ void convert_to_fxsr(struct fxregs_state *fxsave,int fpregs_get (struct task_struct *target,const struct user_regset *regset,struct membuf to) { -struct fpu *fpu = &target->thread.fpu; +struct fpu *fpu = target->thread.fpu;struct user_i387_ia32_struct env;struct fxregs_state fxsave, *fx;
@@ -432,7 +432,7 @@ int fpregs_set(struct task_struct *target, const struct user_regset *regset, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - struct fpu *fpu = &target->thread.fpu; + struct fpu *fpu = target->thread.fpu;struct user_i387_ia32_struct env;int ret;
arch/x86/kernel/fpu/signal.c +9 -9
@@ -38,7 +38,7 @@ static inline bool check_xstate_in_sigframe(struct fxregs_state __user *fxbuf,/* Check for the first magic field and other error scenarios. */ if (fx_sw->magic1 !=FP_XSTATE_MAGIC1 || fx_sw->xstate_size < min_xstate_size || - fx_sw->xstate_size > current->thread.fpu.fpstate->user_size || + fx_sw->xstate_size > current->thread.fpu->fpstate->user_size || fx_sw->xstate_size > fx_sw->extended_size) goto setfx;
@@ -54,7 +54,7 @@ static inline bool check_xstate_in_sigframe(struct fxregs_state __user *fxbuf,if (likely (magic2 ==FP_XSTATE_MAGIC2 ))return true; setfx: -trace_x86_fpu_xstate_check_failed (¤t->thread.fpu); +trace_x86_fpu_xstate_check_failed (current->thread.fpu);/* Set the parameters for fx only state */ fx_sw->magic1 =0 ;
@@ -69,13 +69,13 @@ static inline bool check_xstate_in_sigframe(struct fxregs_state __user *fxbuf,static inline bool save_fsave_header (struct task_struct *tsk,void __user *buf) {if (use_fxsr ()) { -struct xregs_state *xsave = &tsk->thread.fpu.fpstate->regs.xsave; +struct xregs_state *xsave = &tsk->thread.fpu->fpstate->regs.xsave;struct user_i387_ia32_struct env;struct _fpstate_32 __user *fp = buf;fpregs_lock ();if (!test_thread_flag (TIF_NEED_FPU_LOAD )) -fxsave (&tsk->thread.fpu.fpstate->regs.fxsave); +fxsave (&tsk->thread.fpu->fpstate->regs.fxsave);fpregs_unlock ();convert_from_fxsr (&env, tsk);
@@ -188,7 +188,7 @@ static inline int copy_fpregs_to_sigframe(struct xregs_state __user *buf)bool copy_fpstate_to_sigframe (void __user *buf,void __user *buf_fx,int size) {struct task_struct *tsk = current; -struct fpstate *fpstate = tsk->thread.fpu.fpstate; +struct fpstate *fpstate = tsk->thread.fpu->fpstate;bool ia32_fxstate = (buf != buf_fx);int ret;
@@ -276,7 +276,7 @@ static int __restore_fpregs_from_user(void __user *buf, u64 ufeatures,*/ static bool restore_fpregs_from_user (void __user *buf,u64 xrestore,bool fx_only) { -struct fpu *fpu = ¤t->thread.fpu; +struct fpu *fpu = current->thread.fpu;int ret;/* Restore enabled features only. */
@@ -336,7 +336,7 @@ static bool __fpu_restore_sig(void __user *buf, void __user *buf_fx, bool ia32_fxstate) {struct task_struct *tsk = current; -struct fpu *fpu = &tsk->thread.fpu; +struct fpu *fpu = tsk->thread.fpu;struct user_i387_ia32_struct env;bool success, fx_only = false;union fpregs_state *fpregs;
@@ -456,7 +456,7 @@ static inline unsigned int xstate_sigframe_size(struct fpstate *fpstate)*/ bool fpu__restore_sig (void __user *buf,int ia32_frame) { -struct fpu *fpu = ¤t->thread.fpu; +struct fpu *fpu = current->thread.fpu;void __user *buf_fx = buf;bool ia32_fxstate = false;bool success = false;
@@ -503,7 +503,7 @@ unsigned long fpu__alloc_mathframe(unsigned long sp, int ia32_frame, unsigned long *buf_fx, unsigned long *size) { - unsigned long frame_size = xstate_sigframe_size(current->thread.fpu.fpstate); + unsigned long frame_size = xstate_sigframe_size(current->thread.fpu->fpstate); *buf_fx = sp =round_down (sp - frame_size,64 );if (ia32_frame &&use_fxsr ()) {
arch/x86/kernel/fpu/xstate.c +11 -11
@@ -735,7 +735,7 @@ static void __init fpu__init_disable_system_xstate(unsigned int legacy_size)*/ init_fpstate.xfd =0 ; -fpstate_reset (¤t->thread.fpu); +fpstate_reset (current->thread.fpu); }/*
@@ -845,7 +845,7 @@ void __init fpu__init_system_xstate(unsigned int legacy_size) goto out_disable;/* Reset the state for the current task */ -fpstate_reset (¤t->thread.fpu); +fpstate_reset (current->thread.fpu);/* * Update info used for ptrace frames; use standard-format size and no
@@ -919,7 +919,7 @@ void fpu__resume_cpu(void) }if (fpu_state_size_dynamic ()) -wrmsrl (MSR_IA32_XFD , current->thread.fpu.fpstate->xfd); +wrmsrl (MSR_IA32_XFD , current->thread.fpu->fpstate->xfd); }/*
@@ -1188,8 +1188,8 @@ void __copy_xstate_to_uabi_buf(struct membuf to, struct fpstate *fpstate,void copy_xstate_to_uabi_buf (struct membuf to,struct task_struct *tsk,enum xstate_copy_mode copy_mode) { -__copy_xstate_to_uabi_buf (to, tsk->thread.fpu.fpstate, - tsk->thread.fpu.fpstate->user_xfeatures, +__copy_xstate_to_uabi_buf (to, tsk->thread.fpu->fpstate, + tsk->thread.fpu->fpstate->user_xfeatures, tsk->thread.pkru, copy_mode); }
@@ -1329,7 +1329,7 @@ int copy_uabi_from_kernel_to_xstate(struct fpstate *fpstate, const void *kbuf, uint copy_sigframe_from_user_to_xstate (struct task_struct *tsk,const void __user *ubuf) { -return copy_uabi_to_xstate (tsk->thread.fpu.fpstate,NULL , ubuf, &tsk->thread.pkru); +return copy_uabi_to_xstate (tsk->thread.fpu->fpstate,NULL , ubuf, &tsk->thread.pkru); }static bool validate_independent_components (u64 mask)
@@ -1423,7 +1423,7 @@ static bool xstate_op_valid(struct fpstate *fpstate, u64 mask, bool rstor)* The XFD MSR does not match fpstate->xfd. That's invalid when * the passed in fpstate is current's fpstate. */ -if (fpstate->xfd == current->thread.fpu.fpstate->xfd) +if (fpstate->xfd == current->thread.fpu->fpstate->xfd)return false;/*
@@ -1500,7 +1500,7 @@ void fpstate_free(struct fpu *fpu)static int fpstate_realloc (u64 xfeatures,unsigned int ksize,unsigned int usize,struct fpu_guest *guest_fpu) { -struct fpu *fpu = ¤t->thread.fpu; +struct fpu *fpu = current->thread.fpu;struct fpstate *curfps, *newfps =NULL ;unsigned int fpsize;bool in_use;
@@ -1593,7 +1593,7 @@ static int __xstate_request_perm(u64 permitted, u64 requested, bool guest)* AVX512. */ bool compacted =cpu_feature_enabled (X86_FEATURE_XCOMPACTED ); -struct fpu *fpu = ¤t->group_leader->thread.fpu; +struct fpu *fpu = current->group_leader->thread.fpu;struct fpu_state_perm *perm;unsigned int ksize, usize;u64 mask;
@@ -1696,7 +1696,7 @@ int __xfd_enable_feature(u64 xfd_err, struct fpu_guest *guest_fpu)return -EPERM ; } - fpu = ¤t->group_leader->thread.fpu; + fpu = current->group_leader->thread.fpu; perm = guest_fpu ? &fpu->guest_perm : &fpu->perm; ksize = perm->__state_size; usize = perm->__user_state_size;
@@ -1801,7 +1801,7 @@ long fpu_xstate_prctl(int option, unsigned long arg2)*/ static void avx512_status (struct seq_file *m,struct task_struct *task) { -unsigned long timestamp =READ_ONCE (task->thread.fpu.avx512_timestamp); +unsigned long timestamp =READ_ONCE (task->thread.fpu->avx512_timestamp);long delta;if (!timestamp) {
arch/x86/kernel/fpu/xstate.h +3 -3
@@ -22,7 +22,7 @@ static inline void xstate_init_xcomp_bv(struct xregs_state *xsave, u64 mask)static inline u64 xstate_get_group_perm (bool guest) { -struct fpu *fpu = ¤t->group_leader->thread.fpu; +struct fpu *fpu = current->group_leader->thread.fpu;struct fpu_state_perm *perm;/* Pairs with WRITE_ONCE() in xstate_request_perm() */
@@ -265,7 +265,7 @@ static inline int xsave_to_user_sigframe(struct xregs_state __user *buf)* internally, e.g. PKRU. That's user space ABI and also required * to allow the signal handler to modify PKRU. */ -struct fpstate *fpstate = current->thread.fpu.fpstate; +struct fpstate *fpstate = current->thread.fpu->fpstate;u64 mask = fpstate->user_xfeatures;u32 lmask;u32 hmask;
@@ -296,7 +296,7 @@ static inline int xrstor_from_user_sigframe(struct xregs_state __user *buf, u64u32 hmask = mask >>32 ;int err; -xfd_validate_state (current->thread.fpu.fpstate, mask, true); +xfd_validate_state (current->thread.fpu->fpstate, mask, true);stac ();XSTATE_OP (XRSTOR , xstate, lmask, hmask, err);
arch/x86/kernel/process.c +3 -3
@@ -97,7 +97,7 @@ int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) dst->thread.vm86 =NULL ;#endif /* Drop the copied pointer to current's fpstate */ - dst->thread.fpu.fpstate =NULL ; + dst->thread.fpu =NULL ;return 0 ; }
@@ -106,7 +106,7 @@ int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)void arch_release_task_struct (struct task_struct *tsk) {if (fpu_state_size_dynamic ()) -fpstate_free (&tsk->thread.fpu); +fpstate_free (tsk->thread.fpu); }#endif
@@ -116,7 +116,7 @@ void arch_release_task_struct(struct task_struct *tsk)void exit_thread (struct task_struct *tsk) {struct thread_struct *t = &tsk->thread; -struct fpu *fpu = &t->fpu; +struct fpu *fpu = t->fpu;if (test_thread_flag (TIF_IO_BITMAP ))io_bitmap_exit (tsk);
arch/x86/kernel/signal.c +3 -3
@@ -228,7 +228,7 @@ static void handle_signal(struct ksignal *ksig, struct pt_regs *regs) { bool stepping, failed; -struct fpu *fpu = ¤t->thread.fpu; +struct fpu *fpu = current->thread.fpu;if (v8086_mode (regs))save_v86_state ((struct kernel_vm86_regs *) regs,VM86_SIGNAL );
@@ -396,14 +396,14 @@ bool sigaltstack_size_valid(size_t ss_size)if (!fpu_state_size_dynamic () && !strict_sigaltstack_size)return true; - fsize += current->group_leader->thread.fpu.perm.__user_state_size; + fsize += current->group_leader->thread.fpu->perm.__user_state_size;if (likely (ss_size > fsize))return true;if (strict_sigaltstack_size)return ss_size > fsize; - mask = current->group_leader->thread.fpu.perm.__state_perm; + mask = current->group_leader->thread.fpu->perm.__state_perm;if (mask &XFEATURE_MASK_USER_DYNAMIC )return ss_size > fsize;
arch/x86/kernel/traps.c +1 -1
@@ -1148,7 +1148,7 @@ DEFINE_IDTENTRY_RAW(exc_debug)static void math_error (struct pt_regs *regs,int trapnr) {struct task_struct *task = current; -struct fpu *fpu = &task->thread.fpu; +struct fpu *fpu = task->thread.fpu;int si_code;char *str = (trapnr ==X86_TRAP_MF ) ?"fpu exception" :"simd exception" ;
arch/x86/math-emu/fpu_aux.c +1 -1
@@ -53,7 +53,7 @@ void fpstate_init_soft(struct swregs_state *soft)void finit (void ) { -fpstate_init_soft (¤t->thread.fpu.fpstate->regs.soft); +fpstate_init_soft (¤t->thread.fpu->fpstate->regs.soft); }/*
arch/x86/math-emu/fpu_entry.c +2 -2
@@ -641,7 +641,7 @@ int fpregs_soft_set(struct task_struct *target, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - struct swregs_state *s387 = &target->thread.fpu.fpstate->regs.soft; + struct swregs_state *s387 = &target->thread.fpu->fpstate->regs.soft;void *space = s387->st_space;int ret;int offset, other, i, tags, regnr, tag, newtop;
@@ -692,7 +692,7 @@ int fpregs_soft_get(struct task_struct *target, const struct user_regset *regset, struct membuf to) { -struct swregs_state *s387 = &target->thread.fpu.fpstate->regs.soft; +struct swregs_state *s387 = &target->thread.fpu->fpstate->regs.soft;const void *space = s387->st_space;int offset = (S387 ->ftop &7 ) *10 , other =80 - offset;
arch/x86/math-emu/fpu_system.h +1 -1
@@ -73,7 +73,7 @@ static inline bool seg_writable(struct desc_struct *d)return (d->type &SEG_TYPE_EXECUTE_MASK ) ==SEG_TYPE_WRITABLE ; } -#define I387 (¤t->thread.fpu.fpstate->regs) +#define I387 (¤t->thread.fpu->fpstate->regs)#define FPU_info (I387->soft.info)#define FPU_CS (*(unsigned short *) &(FPU_info->regs->cs))
arch/x86/mm/extable.c +1 -1
@@ -111,7 +111,7 @@ static bool ex_handler_sgx(const struct exception_table_entry *fixup,/* * Handler for when we fail to restore a task's FPU state. We should never get -* here because the FPU state of a task using the FPU (task->thread.fpu.state) +* here because the FPU state of a task using the FPU (task->thread.fpu->state) * should always be valid. However, past bugs have allowed userspace to set * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn(). * These caused XRSTOR to fail when switching to the task, leaking the FPU
include/linux/sched.h +3 -10
@@ -1554,21 +1554,14 @@ struct task_struct {struct user_event_mm *user_event_mm;#endif +/* CPU-specific state of this task: */ +struct thread_struct thread; +/* * New fields for task_struct should be added above here, so that * they are included in the randomized portion of task_struct. */ randomized_struct_fields_end - -/* CPU-specific state of this task: */ - struct thread_struct thread; - -/* -* WARNING: on x86, 'thread_struct' contains a variable-sized -* structure. It *MUST* be at the end of 'task_struct'. -* -* Do not put anything below here! -*/ };#define TASK_REPORT_IDLE (TASK_REPORT + 1)
Signature · 1 line
2.43.0
Re: [PATCH 1/3] x86/fpu: Make task_struct::thread constant size
Quoted 5 lines · Ingo Molnar
/*
* Handler for when we fail to restore a task's FPU state. We should never get
- * here because the FPU state of a task using the FPU (task->thread.fpu.state)
+ * here because the FPU state of a task using the FPU (task->thread.fpu->state)
Just a nitpick:
fpu::fpstate now points to the active FPU in-memory storage.
Thanks,
Chang
PATCH 1/1 x86/fpu: Fix stale comment in ex_handler_fprestore()
Quoted 8 lines · Chang S. Bae
Quoted 4 lines · Ingo Molnar
/*
* Handler for when we fail to restore a task's FPU state. We should never get
- * here because the FPU state of a task using the FPU (task->thread.fpu.state)
+ * here because the FPU state of a task using the FPU (task->thread.fpu->state)Just a nitpick:
fpu::fpstatenow points to the active FPU in-memory storage.
Yeah, that is a stale comment, but this was just a 'sed' job in essence, and I'd like to keep that particular patch semi-automated.
Note that later on that comment gets further mangled into:
/*
* Handler for when we fail to restore a task's FPU state. We should never get
* here because the FPU state of a task using the FPU (x86_task_fpu(task)->state)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
* should always be valid. However, past bugs have allowed userspace to set
* reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn().
* These caused XRSTOR to fail when switching to the task, leaking the FPU
* registers of the task previously executing on the CPU. Mitigate this class
* of vulnerability by restoring from the initial state (essentially, zeroing
* out all the FPU registers) if we can't restore from the task's FPU state. */
... which didn't improve clarity either. :-)
How about the patch below?
Thanks,
Ingo
=========================>
From: Ingo Molnar <[email protected]>
Date: Thu, 6 Jun 2024 11:27:57 +0200
Subject: [PATCH] x86/fpu: Fix stale comment in ex_handler_fprestore()
The state -> fpstate rename of the fpu::fpstate field didn't get propagated to the comment describing ex_handler_fprestore(), fix it.
--- arch/x86/mm/extable.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/mm/extable.c +1 -1
@@ -111,7 +111,7 @@ static bool ex_handler_sgx(const struct exception_table_entry *fixup,/* * Handler for when we fail to restore a task's FPU state. We should never get -* here because the FPU state of a task using the FPU (x86_task_fpu(task)->state) +* here because the FPU state of a task using the FPU (struct fpu::fpstate) * should always be valid. However, past bugs have allowed userspace to set * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn(). * These caused XRSTOR to fail when switching to the task, leaking the FPU
Re: [PATCH] x86/fpu: Fix stale comment in ex_handler_fprestore()
Quoted 11 lines · Ingo Molnar
From: Ingo Molnar <[email protected]>
Date: Thu, 6 Jun 2024 11:27:57 +0200
Subject: [PATCH] x86/fpu: Fix stale comment inex_handler_fprestore()The state -> fpstate rename of the
fpu::fpstatefield didn't get propagated to the comment describingex_handler_fprestore(), fix it.Reported-by Chang S. Bae <[email protected]>Signed-off-by Ingo Molnar <[email protected]>
Yeah, looks good to me. Feel free to add:
Reviewed-by Chang S. Bae <[email protected]>
Thanks,
Chang
commit 81106b7e0b13 can break asm_int80_emulation on x86_64
In reply to Ingo MolnarSince linux-next-20240730 and in versions next-202408{05,06} the error below can appear when trying to start a game using wine. The error does not always appear but after some failed attempts at bisecting this I found a way to reliably trigger this by first creating high system load (by compiling a kernel with make -j 16) and then trying to start the game until the error occurs. If the error does not occur during the time it take to compile the kernel (about ~6.5min) I declared the commit as good. Usually the bad commit did not take more than three attempts of starting wine to trigger the error.
With this I bisected the problem to commit 81106b7e0b13.
To revert this commit in next-20240806 the following commits have to be reverted:
(HEAD -> wine_usercopy_bug) Revert "x86/fpu: Make task_struct::thread constant size" Revert "x86/fpu: Remove the thread::fpu pointer"
Revert "x86/fpu: Remove init_task FPU state dependencies, add debugging warning for PF_KTHREAD tasks" Revert "x86/CPU/AMD: Always inline amd_clear_divider()"
After reverting these commit the next-20240806 seems to be free from the error. The machine used is an Ryzen 5800H (Msi Alpha15 Laptop) running a fully updated (as of 20240807) debian sid.
Bert Karwatzki
Error message:
[T34926] usercopy: Kernel memory overwrite attempt detected to SLUB object 'task_struct' (offset 3072, size 160)!
[T34926] ------------[ cut here ]------------
[T34926] kernel BUG at mm/usercopy.c:102!
[T34926] Oops: invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
[T34926] CPU: 12 UID: 1000 PID: 34926 Comm: start.exe Not tainted 6.11.0-rc2-next-20240806-master-dirty #178
[T34926] Hardware name: Micro-Star International Co., Ltd. Alpha 15 B5EEK/MS-158L, BIOS E158LAMS.107 11/10/2021
[T34926] RIP: 0010:usercopy_abort+0x73/0x75
[T34926] Code: e0 f2 06 86 eb 0e 48 c7 c2 81 6b 08 86 48 c7 c7 e9 f2 06 86 56 48 89 fe 48 c7 c7 d8 47 0c 86 51 48 89 c1 41 52 e8 6d 70 ff ff <0f> 0b 48 89 d9 49 89 e8 44 89 e2 31 f6 48 81 e9 00 00 20 85 48 c7 [T34926] RSP: 0000:ffffa58a47e73cd0 EFLAGS: 00010246
[T34926] RAX: 0000000000000068 RBX: ffff8e1eec1a4a00 RCX: 0000000000000000
[T34926] RDX: 0000000000000000 RSI: ffff8e2cae9177c0 RDI: ffff8e2cae9177c0
[T34926] RBP: 00000000000000a0 R08: 0000000000000000 R09: ffffa58a47e73b78
[T34926] R10: ffffffff8627fe88 R11: 0000000000000003 R12: 0000000000000000
[T34926] R13: ffff8e1eec1a4aa0 R14: ffff8e1eec1a4a00 R15: 0000000000000001
[T34926] FS: 000000003ffe2000(006b) GS:ffff8e2cae900000(0063) knlGS:00000000f7ed20c0
[T34926] CS: 0010 DS: 002b ES: 002b CR0: 0000000080050033
[T34926] CR2: 000000003ffed324 CR3: 00000002a58da000 CR4: 0000000000750ef0
[T34926] PKRU: 55555554
[T34926] Call Trace:
[T34926] <TASK> [T34926] ? __die+0x51/0x92 [T34926] ? die+0x29/0x50 [T34926] ? do_trap+0x105/0x110 [T34926] ? do_error_trap+0x60/0x80 [T34926] ? usercopy_abort+0x73/0x75 [T34926] ? exc_invalid_op+0x4d/0x70 [T34926] ? usercopy_abort+0x73/0x75 [T34926] ? asm_exc_invalid_op+0x1a/0x20 [T34926] ? usercopy_abort+0x73/0x75 [T34926] ? __check_heap_object+0x7d/0xa0 [T34926] ? __check_object_size+0x1f9/0x210 [T34926] ? copy_from_buffer+0x40/0x60 [T34926] ? copy_uabi_to_xstate+0xe1/0x1e0 [T34926] ? __fpu_restore_sig+0x403/0x480 [T34926] ? fpu__restore_sig+0x4c/0x90 [T34926] ? ia32_restore_sigcontext+0x129/0x170 [T34926] ? __do_compat_sys_rt_sigreturn+0x68/0xc0 [T34926] ? do_int80_emulation+0x88/0x140 [T34926] ? asm_int80_emulation+0x1a/0x20 [T34926] </TASK>
[T34926] Modules linked in: ccm snd_seq_dummy snd_hrtimer snd_seq_midi snd_seq_midi_event snd_rawmidi snd_seq snd_seq_device rfcomm bnep nls_ascii nls_cp437 vfat fat snd_ctl_led snd_hda_codec_realtek snd_hda_codec_generic snd_hda_scodec_component snd_hda_codec_hdmi btusb btrtl btintel snd_hda_intel btbcm btmtk snd_intel_dspcfg snd_acp3x_pdm_dma snd_soc_dmic snd_acp3x_rn uvcvideo bluetooth snd_hda_codec videobuf2_vmalloc snd_soc_core uvc snd_hwdep videobuf2_memops videobuf2_v4l2 snd_hda_core snd_pcm_oss videodev snd_mixer_oss snd_pcm snd_rn_pci_acp3x videobuf2_common snd_acp_config msi_wmi snd_soc_acpi ecdh_generic amd_atl ecc mc sparse_keymap edac_mce_amd wmi_bmof snd_timer snd ccp snd_pci_acp3x k10temp soundcore battery ac joydev button hid_sensor_accel_3d hid_sensor_prox hid_sensor_als hid_sensor_magn_3d hid_sensor_gyro_3d hid_sensor_trigger industrialio_triggered_buffer kfifo_buf industrialio amd_pmc hid_sensor_iio_common evdev hid_multitouch serio_raw mt7921e mt7921_common mt792x_lib mt76_connac_lib mt76 [T34926] mac80211 libarc4 cfg80211 rfkill msr fuse nvme_fabrics efi_pstore configfs efivarfs autofs4 ext4 crc32c_generic mbcache jbd2 usbhid amdgpu i2c_algo_bit drm_ttm_helper xhci_pci ttm drm_exec drm_suballoc_helper xhci_hcd amdxcp drm_buddy hid_sensor_hub usbcore mfd_core nvme gpu_sched i2c_piix4 hid_generic crc32c_intel psmouse drm_display_helper amd_sfh usb_common i2c_smbus nvme_core crc16 r8169 i2c_hid_acpi i2c_hid hid i2c_designware_platform i2c_designware_core
[T34926] ---[ end trace 0000000000000000 ]---
Re: commit 81106b7e0b13 can break asm_int80_emulation on x86_64
I did some experimentation on the bug with the help of the following patch:
arch/x86/kernel/fpu/xstate.c +7 -0
@@ -1212,6 +1212,7 @@ static int copy_from_buffer(void *dst, unsigned int offset, unsigned int size,if (kbuf) {memcpy (dst, kbuf + offset, size); }else { +printk (KERN_INFO "%s: calling copy_from_user with to = %px from = %px, n = 0x%x\n" , __func__, dst, ubuf + offset, size);if (copy_from_user (dst, ubuf + offset, size))return -EFAULT ; }
@@ -1257,6 +1258,8 @@ static int copy_uabi_to_xstate(struct fpstate *fpstate, const void *kbuf,int i; offset = offsetof(struct xregs_state , header); +printk (KERN_INFO "%s %d: calling copy_from buffer with offset = 0x%x, size = 0x%lx\n" , + __func__, __LINE__, offset,sizeof (hdr));if (copy_from_buffer (&hdr, offset,sizeof (hdr), kbuf, ubuf))return -EFAULT ;
@@ -1269,6 +1272,8 @@ static int copy_uabi_to_xstate(struct fpstate *fpstate, const void *kbuf,u32 mxcsr[2 ]; offset = offsetof(struct fxregs_state , mxcsr); +printk (KERN_INFO "%s %d: calling copy_from buffer with offset = 0x%x, size = 0x%lx\n" , + __func__, __LINE__, offset,sizeof (mxcsr));if (copy_from_buffer (mxcsr, offset,sizeof (mxcsr), kbuf, ubuf))return -EFAULT ;
@@ -1292,6 +1297,8 @@ static int copy_uabi_to_xstate(struct fpstate *fpstate, const void *kbuf, offset = xstate_offsets[i]; size = xstate_sizes[i]; +printk (KERN_INFO "%s %d: calling copy_from buffer %d with offset = 0x%x, size = 0x%x, dst = %px, kbuf = %px, ubuf = %px\n" , + __func__, __LINE__, i, offset, size, dst, kbuf, ubuf);if (copy_from_buffer (dst, offset, size, kbuf, ubuf))return -EFAULT ; }
drivers/net/wireless/mediatek/mt76/mt7921/main.c +1 -1
@@ -1183,7 +1183,7 @@ static void mt7921_ipv6_addr_change(struct ieee80211_hw *hw, struct inet6_dev *idev) {struct mt792x_vif *mvif = (struct mt792x_vif *)vif->drv_priv; -struct mt792x_dev *dev = mvif->phy->dev; +struct mt792x_dev *dev =mt792x_hw_dev (hw);struct inet6_ifaddr *ifa;struct in6_addr ns_addrs[IEEE80211_BSS_ARP_ADDR_LIST_LEN ];struct sk_buff *skb;
mm/slub.c +4 -0
@@ -5636,6 +5636,10 @@ void __check_heap_object(const void *ptr, unsigned long n, n <= s->useroffset - offset + s->usersize) return; +printk (KERN_INFO "%s: ptr = %px slab = %px s = %px\n" , __func__, ptr, slab, s); +printk (KERN_INFO "%s: offset = 0x%x s->useroffset = 0x%x\n" , __func__, offset, s->useroffset); +printk (KERN_INFO "%s: offset - s->useroffset = 0x%x s->usersize = 0x%x\n" , __func__, offset - s->useroffset, s->usersize); +printk (KERN_INFO "%s: n = 0x%lx s->useroffset - offset + s->usersize = 0x%x\n" , __func__, n, s->useroffset - offset + s->usersize); usercopy_abort("SLUB object", s->name, to_user, offset, n); }#endif /* CONFIG_HARDENED_USERCOPY */
which gives the following output (before the usual backtrace) :
[ 223.785491] [ T46217] copy_uabi_to_xstate 1261: calling copy_from buffer with offset = 0x200, size = 0x40 [ 223.785501] [ T46217] copy_from_buffer: calling copy_from_user with to = ffffa85f5387fd58 from = 000000003ffef840, n = 0x40 [ 223.785506] [ T46217] copy_uabi_to_xstate 1275: calling copy_from buffer with offset = 0x18, size = 0x8 [ 223.785509] [ T46217] copy_from_buffer: calling copy_from_user with to = ffffa85f5387fd50 from = 000000003ffef658, n = 0x8 [ 223.785512] [ T46217] copy_uabi_to_xstate 1300: calling copy_from buffer 0 with offset = 0x0, size = 0xa0, dst = ffff8c819c239b80, kbuf = 0000000000000000, ubuf = 000000003ffef640 [ 223.785516] [ T46217] copy_from_buffer: calling copy_from_user with to = ffff8c819c239b80 from = 000000003ffef640, n = 0xa0 [ 223.785520] [ T46217] __check_heap_object: ptr = ffff8c819c239b80 slab = ffffd5368c708e00 s = ffff8c7f800d1400 [ 223.785523] [ T46217] __check_heap_object: offset = 0xc00 s->useroffset = 0x0 [ 223.785525] [ T46217] __check_heap_object: offset - s->useroffset = 0xc00 s->usersize = 0x0 FIXME? [ 223.785528] [ T46217] __check_heap_object: n = 0xa0 s->useroffset - offset + s->usersize = 0xfffff400 [ 223.785530] [ T46217] usercopy: Kernel memory overwrite attempt detected to SLUB object 'task_struct' (offset 3072, size 160)! [ 223.785545] [ T46217] ------------[ cut here ]------------ [ 223.785547] [ T46217] kernel BUG at mm/usercopy.c:102!
So the problem seems to be that the kmem_cache object *s has usersize 0. This should be impossible in theory as kmem_cache_create_usercopy() should print a warning in case of (!usersize && useroffset).
Bert Karwatzki
Re: commit 81106b7e0b13 can break asm_int80_emulation on x86_64
Quoted 4 lines · Bert Karwatzki
So the problem seems to be that the
kmem_cacheobject *s has usersize 0. This should be impossible in theory askmem_cache_create_usercopy()should print a warning in case of (!usersize && useroffset).
Following along from your original report:
usercopy: Kernel memory overwrite attempt detected to SLUB object
'task_struct' (offset 3072, size 160)!
IOW, the user copy code is unhappy about copying data from user mode into the task_struct allocation.
Anyway, on x86-64 with that commit we now just do
arch_thread_struct_whitelist(unsigned long *offset, unsigned long *size)
{
*offset = 0;
*size = 0;
}and that seems to be the bug.
It's still allocated together with that 'struct task_struct', even if it's no longer inside the 'struct thread_struct'.
Ingo? I think it needs to be something like
*offset = sizeof(struct task_struct)
- offsetof(struct task_struct, thread)
+ offsetof(struct fpu, __fpstate.regs);
*size = fpu_kernel_cfg.default_size;
and the commit message of
The fpu_thread_struct_whitelist() quirk to hardened usercopy can be removed, now that the FPU structure is not embedded in the task struct anymore, which reduces text footprint a bit.
is clearly not true. No, it's not embedded in 'struct task_struct', but it *is* still allocated together with it in the same slab if I read the code correctly (ie this part
static void __init fpu__init_task_struct_size(void)
{
int task_size = sizeof(struct task_struct); task_size += sizeof(struct fpu);
..
arch_task_struct_size = task_size;is because it's still all one single slab allocation.
Linus
Re: commit 81106b7e0b13 can break asm_int80_emulation on x86_64
Am Freitag, dem 09.08.2024 um 11:17 -0700 schrieb Linus Torvalds:
Quoted 48 lines · Bert Karwatzki
Quoted 4 lines
So the problem seems to be that the
kmem_cacheobject *s has usersize 0. This should be impossible in theory askmem_cache_create_usercopy()should print a warning in case of (!usersize && useroffset).Following along from your original report:
usercopy: Kernel memory overwrite attempt detected to SLUB object
'task_struct' (offset 3072, size 160)!IOW, the user copy code is unhappy about copying data from user mode into the
task_structallocation.Anyway, on x86-64 with that commit we now just do
arch_thread_struct_whitelist(unsigned long *offset, unsigned long *size) { *offset = 0; *size = 0; }and that seems to be the bug.
It's still allocated together with that 'struct
task_struct', even if it's no longer inside the 'structthread_struct'.Ingo? I think it needs to be something like
*offset =
sizeof(struct task_struct)
-offsetof(struct task_struct, thread)
+offsetof(struct fpu, __fpstate.regs);
*size = fpu_kernel_cfg.default_size;and the commit message of
The
fpu_thread_struct_whitelist()quirk to hardened usercopy can be removed, now that the FPU structure is not embedded in the task struct anymore, which reduces text footprint a bit.is clearly not true. No, it's not embedded in 'struct
task_struct', but it *is* still allocated together with it in the same slab if I read the code correctly (ie this partstatic void __init fpu__init_task_struct_size(void) { int task_size = sizeof(struct task_struct);
Quoted 8 lines
task_size += sizeof(struct fpu); .. arch_task_struct_size = task_size;is because it's still all one single slab allocation.
Linus
Yes, this seems to be the problem. As the old code (from linux-5.16-rc1 to linux-6.11-rc2) used this
void fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size)
{
*offset = offsetof(struct thread_struct, fpu.__fpstate.regs);
*size = fpu_kernel_cfg.default_size;
}I tried this as a potential solution:
arch/x86/include/asm/processor.h +3 -2
@@ -516,8 +516,9 @@ extern struct fpu *x86_task_fpu(struct task_struct *task);static inline void arch_thread_struct_whitelist (unsigned long *offset,unsigned long *size) { - *offset =0 ; - *size =0 ; + *offset =sizeof (struct thread_struct ) + + offsetof(struct fpu, __fpstate.regs); + *size = fpu_kernel_cfg.default_size; } static inline void
and it seems to solve the problem (debug output from my previous printk patch):
[ T57380] copy_uabi_to_xstate 1261: calling copy_from buffer with offset = 0x200, size = 0x40
[ T57380] copy_from_buffer: calling copy_from_user with to = ffffadf495127d58 from = 000000003ffef840, n = 0x40
[ T57380] copy_uabi_to_xstate 1275: calling copy_from buffer with offset = 0x18, size = 0x8
[ T57380] copy_from_buffer: calling copy_from_user with to = ffffadf495127d50 from = 000000003ffef658, n = 0x8
[ T57380] copy_uabi_to_xstate 1300: calling copy_from buffer 0 with offset = 0x0, size = 0xa0, dst = ffff90d285ec7880, kbuf = 0000000000000000, ubuf = 000000003ffef640
[ T57380] copy_from_buffer: calling copy_from_user with to = ffff90d285ec7880 from = 000000003ffef640, n = 0xa0
Here the bug would happen in linux-next-20240806 which seems to be avoided by the patch.
[ T57380] copy_uabi_to_xstate 1300: calling copy_from buffer 1 with offset = 0xa0, size = 0x100, dst = ffff90d285ec7920, kbuf = 0000000000000000, ubuf = 000000003ffef640
[ T57380] copy_from_buffer: calling copy_from_user with to = ffff90d285ec7920 from = 000000003ffef6e0, n = 0x100
Bert Karwatzki
Re: commit 81106b7e0b13 can break asm_int80_emulation on x86_64
Am Samstag, dem 10.08.2024 um 01:04 +0200 schrieb Bert Karwatzki:
Quoted 105 lines
Am Freitag, dem 09.08.2024 um 11:17 -0700 schrieb Linus Torvalds:
Quoted 48 lines · Bert Karwatzki
Quoted 4 lines
So the problem seems to be that the
kmem_cacheobject *s has usersize 0. This should be impossible in theory askmem_cache_create_usercopy()should print a warning in case of (!usersize && useroffset).Following along from your original report:
usercopy: Kernel memory overwrite attempt detected to SLUB object
'task_struct' (offset 3072, size 160)!IOW, the user copy code is unhappy about copying data from user mode into the
task_structallocation.Anyway, on x86-64 with that commit we now just do
arch_thread_struct_whitelist(unsigned long *offset, unsigned long *size) { *offset = 0; *size = 0; }and that seems to be the bug.
It's still allocated together with that 'struct
task_struct', even if it's no longer inside the 'structthread_struct'.Ingo? I think it needs to be something like
*offset =
sizeof(struct task_struct)
-offsetof(struct task_struct, thread)
+offsetof(struct fpu, __fpstate.regs);
*size = fpu_kernel_cfg.default_size;and the commit message of
The
fpu_thread_struct_whitelist()quirk to hardened usercopy can be removed, now that the FPU structure is not embedded in the task struct anymore, which reduces text footprint a bit.is clearly not true. No, it's not embedded in 'struct
task_struct', but it *is* still allocated together with it in the same slab if I read the code correctly (ie this partstatic void __init fpu__init_task_struct_size(void) { int task_size = sizeof(struct task_struct);Quoted 8 lines
task_size += sizeof(struct fpu); .. arch_task_struct_size = task_size;is because it's still all one single slab allocation.
Linus
Yes, this seems to be the problem. As the old code (from linux-5.16-rc1 to linux-6.11-rc2) used this
void
fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size){ *offset = offsetof(struct thread_struct, fpu.__fpstate.regs); *size = fpu_kernel_cfg.default_size; }I tried this as a potential solution:
arch/x86/include/asm/processor.h +3
-2@@ -516,8 +516,9 @@ extern struct fpu *x86_task_fpu(struct task_struct *task);static inline void arch_thread_struct_whitelist (unsigned long *offset,unsigned long *size) { - *offset =0 ; - *size =0 ; + *offset =sizeof (struct thread_struct ) + + offsetof(struct fpu, __fpstate.regs); + *size = fpu_kernel_cfg.default_size; } static inline voidand it seems to solve the problem (debug output from my previous printk patch):
[ T57380]
copy_uabi_to_xstate1261: callingcopy_frombuffer with offset =0x200, size =0x40
[ T57380]copy_from_buffer: callingcopy_from_userwith to =ffffadf495127d58from =000000003ffef840, n =0x40
[ T57380]copy_uabi_to_xstate1275: callingcopy_frombuffer with offset =0x18, size =0x8
[ T57380]copy_from_buffer: callingcopy_from_userwith to =ffffadf495127d50from =000000003ffef658, n =0x8
[ T57380]copy_uabi_to_xstate1300: callingcopy_frombuffer 0 with offset =0x0, size =0xa0, dst =ffff90d285ec7880, kbuf = 0000000000000000, ubuf =000000003ffef640
[ T57380]copy_from_buffer: callingcopy_from_userwith to =ffff90d285ec7880from =000000003ffef640, n =0xa0Here the bug would happen in linux-next-20240806 which seems to be avoided by the patch.
Quoted 8 lines
[ T57380]
copy_uabi_to_xstate1300: callingcopy_frombuffer 1 with offset =0xa0, size =0x100, dst =ffff90d285ec7920, kbuf = 0000000000000000, ubuf =000000003ffef640
[ T57380]copy_from_buffer: callingcopy_from_userwith to =ffff90d285ec7920from =000000003ffef6e0, n =0x100Bert Karwatzki
After taking a closer look at the code I realized that my solution only works if task_struct and thread_struct are not randomized while your solution
*offset = sizeof(struct task_struct)
- offsetof(struct task_struct, thread)
+ offsetof(struct fpu, __fpstate.regs);
*size = fpu_kernel_cfg.default_size;
works in those cases, too. Here's a way to introdce this fix using fpu_thread_struct_whitelist in arch/x86/kernel/fpu/init.c where we can use sizeof(task_struct) without including additional headers:
arch/x86/include/asm/processor.h +0 -0
@@ -510,14 +510,12 @@ extern struct fpu *x86_task_fpu(struct task_struct *task);# define x86_task_fpu (task) ((struct fpu *)((void *)(task) +
sizeof(*(task))))
#endif
-/*
- * X86 doesn't need any embedded-FPU-struct quirks:
- */
+extern void fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size);
+
static inline void
arch_thread_struct_whitelist(unsigned long *offset, unsigned long *size) {
- *offset = 0; - *size = 0; + fpu_thread_struct_whitelist(offset, size);
}
static inline void
arch/x86/kernel/fpu/init.c +12 -0
@@ -179,6 +179,18 @@ static void __init fpu__init_task_struct_size(void) arch_task_struct_size = task_size; } +/* +* Whitelist the FPU register state embedded into task_struct for hardened +* usercopy. +*/ +void fpu_thread_struct_whitelist (unsigned long *offset,unsigned long *size) +{ + *offset =sizeof (struct task_struct ) + - offsetof(struct task_struct , thread) + + offsetof(struct fpu, __fpstate.regs); + *size = fpu_kernel_cfg.default_size; +} +/* * Set up the user and kernel xstate sizes based on the legacy FPU context
size.
*
Bert Karwatzki
PATCH v3 2/3 x86/fpu: Remove the thread::fpu pointer
In reply to Ingo MolnarAs suggested by Oleg, remove the thread::fpu pointer, as we can calculate it via x86_task_fpu() at compile-time.
This improves code generation a bit:
kepler:~/tip> size vmlinux.before vmlinux.after
text data bss dec hex filename 26475405 10435342 1740804 38651551 24dc69f vmlinux.before 26475339 10959630 1216516 38651485 24dc65d vmlinux.after
==================>
arch/x86/include/asm/fpu/sched.h | 2 +- arch/x86/include/asm/processor.h | 5 ++--- arch/x86/include/asm/vm86.h | 2 +- arch/x86/kernel/fpu/context.h | 2 +- arch/x86/kernel/fpu/core.c | 30 ++++++++++++++---------------- arch/x86/kernel/fpu/init.c | 7 +++---- arch/x86/kernel/fpu/regset.c | 22 +++++++++++----------- arch/x86/kernel/fpu/signal.c | 18 +++++++++--------- arch/x86/kernel/fpu/xstate.c | 22 +++++++++++----------- arch/x86/kernel/fpu/xstate.h | 6 +++--- arch/x86/kernel/process.c | 6 ++---- arch/x86/kernel/signal.c | 6 +++--- arch/x86/kernel/traps.c | 2 +- arch/x86/kernel/vmlinux.lds.S | 4 ++++ arch/x86/math-emu/fpu_aux.c | 2 +- arch/x86/math-emu/fpu_entry.c | 4 ++-- arch/x86/math-emu/fpu_system.h | 2 +- arch/x86/mm/extable.c | 2 +- 18 files changed, 71 insertions(+), 73 deletions(-)
--- arch/x86/include/asm/fpu/sched.h | 2 +- arch/x86/include/asm/processor.h | 5 ++--- arch/x86/kernel/fpu/context.h | 2 +- arch/x86/kernel/fpu/core.c | 30 ++++++++++++++---------------- arch/x86/kernel/fpu/init.c | 7 +++---- arch/x86/kernel/fpu/regset.c | 22 +++++++++++----------- arch/x86/kernel/fpu/signal.c | 18 +++++++++--------- arch/x86/kernel/fpu/xstate.c | 22 +++++++++++----------- arch/x86/kernel/fpu/xstate.h | 6 +++--- arch/x86/kernel/process.c | 6 ++---- arch/x86/kernel/signal.c | 6 +++--- arch/x86/kernel/traps.c | 2 +- arch/x86/kernel/vmlinux.lds.S | 4 ++++ arch/x86/math-emu/fpu_aux.c | 2 +- arch/x86/math-emu/fpu_entry.c | 4 ++-- arch/x86/math-emu/fpu_system.h | 2 +- arch/x86/mm/extable.c | 2 +- 17 files changed, 70 insertions(+), 72 deletions(-)
arch/x86/include/asm/fpu/sched.h +1 -1
@@ -41,7 +41,7 @@ static inline void switch_fpu_prepare(struct task_struct *old, int cpu) {if (cpu_feature_enabled (X86_FEATURE_FPU ) && !(old->flags & (PF_KTHREAD |PF_USER_WORKER ))) { -struct fpu *old_fpu = old->thread.fpu; +struct fpu *old_fpu =x86_task_fpu (old);save_fpregs_to_fpstate (old_fpu);/*
arch/x86/include/asm/processor.h +2 -3
@@ -502,11 +502,10 @@ struct thread_struct {struct thread_shstk shstk;#endif - -/* Floating point and extended processor state */ -struct fpu *fpu; }; +#define x86_task_fpu (task) ((struct fpu *)((void *)task + sizeof(*task))) +/* * X86 doesn't need any embedded-FPU-struct quirks: */
arch/x86/kernel/fpu/context.h +1 -1
@@ -53,7 +53,7 @@ static inline void fpregs_activate(struct fpu *fpu)/* Internal helper for switch_fpu_return() and signal frame setup */ static inline void fpregs_restore_userregs (void ) { -struct fpu *fpu = current->thread.fpu; +struct fpu *fpu =x86_task_fpu (current);int cpu =smp_processor_id ();if (WARN_ON_ONCE (current->flags & (PF_KTHREAD |PF_USER_WORKER )))
arch/x86/kernel/fpu/core.c +14 -16
@@ -204,7 +204,7 @@ static void fpu_init_guest_permissions(struct fpu_guest *gfpu)return ;spin_lock_irq (¤t->sighand->siglock); - fpuperm = ¤t->group_leader->thread.fpu->guest_perm; + fpuperm = &x86_task_fpu (current->group_leader)->guest_perm; perm = fpuperm->__state_perm;/* First fpstate allocation locks down permissions. */
@@ -316,7 +316,7 @@ EXPORT_SYMBOL_GPL(fpu_update_guest_xfd);*/ void fpu_sync_guest_vmexit_xfd_state (void ) { -struct fpstate *fps = current->thread.fpu->fpstate; +struct fpstate *fps =x86_task_fpu (current)->fpstate;lockdep_assert_irqs_disabled ();if (fpu_state_size_dynamic ()) {
@@ -330,7 +330,7 @@ EXPORT_SYMBOL_GPL(fpu_sync_guest_vmexit_xfd_state);int fpu_swap_kvm_fpstate (struct fpu_guest *guest_fpu,bool enter_guest) {struct fpstate *guest_fps = guest_fpu->fpstate; -struct fpu *fpu = current->thread.fpu; +struct fpu *fpu =x86_task_fpu (current);struct fpstate *cur_fps = fpu->fpstate;fpregs_lock ();
@@ -430,7 +430,7 @@ void kernel_fpu_begin_mask(unsigned int kfpu_mask)if (!(current->flags & (PF_KTHREAD |PF_USER_WORKER )) && !test_thread_flag (TIF_NEED_FPU_LOAD )) {set_thread_flag (TIF_NEED_FPU_LOAD ); -save_fpregs_to_fpstate (current->thread.fpu); +save_fpregs_to_fpstate (x86_task_fpu (current)); }__cpu_invalidate_fpregs_state ();
@@ -458,7 +458,7 @@ EXPORT_SYMBOL_GPL(kernel_fpu_end);*/ void fpu_sync_fpstate (struct fpu *fpu) { -WARN_ON_FPU (fpu != current->thread.fpu); +WARN_ON_FPU (fpu !=x86_task_fpu (current));fpregs_lock ();trace_x86_fpu_before_save (fpu);
@@ -543,7 +543,7 @@ void fpstate_reset(struct fpu *fpu)static inline void fpu_inherit_perms (struct fpu *dst_fpu) {if (fpu_state_size_dynamic ()) { -struct fpu *src_fpu = current->group_leader->thread.fpu; +struct fpu *src_fpu =x86_task_fpu (current->group_leader);spin_lock_irq (¤t->sighand->siglock);/* Fork also inherits the permissions of the parent */
@@ -563,7 +563,7 @@ static int update_fpu_shstk(struct task_struct *dst, unsigned long ssp)if (!ssp)return 0 ; - xstate =get_xsave_addr (&dst->thread.fpu->fpstate->regs.xsave, + xstate =get_xsave_addr (&x86_task_fpu (dst)->fpstate->regs.xsave,XFEATURE_CET_USER );/*
@@ -591,13 +591,11 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,* This is safe because task_struct size is a multiple of cacheline size. */ struct fpu *dst_fpu = (void *)dst +sizeof (*dst); -struct fpu *src_fpu = current->thread.fpu; +struct fpu *src_fpu =x86_task_fpu (current);BUILD_BUG_ON (sizeof (*dst) %SMP_CACHE_BYTES !=0 );BUG_ON (!src_fpu); - dst->thread.fpu = dst_fpu; -/* The new task's FPU state cannot be valid in the hardware. */ dst_fpu->last_cpu =-1 ;
@@ -678,7 +676,7 @@ void fpu__drop(struct fpu *fpu) {preempt_disable (); -if (fpu == current->thread.fpu) { +if (fpu ==x86_task_fpu (current)) {/* Ignore delayed exceptions from user space */ asmvolatile ("1: fwait\n" "2:\n"
@@ -712,7 +710,7 @@ static inline void restore_fpregs_from_init_fpstate(u64 features_mask)*/ static void fpu_reset_fpregs (void ) { -struct fpu *fpu = current->thread.fpu; +struct fpu *fpu =x86_task_fpu (current);fpregs_lock ();__fpu_invalidate_fpregs_state (fpu);
@@ -741,7 +739,7 @@ static void fpu_reset_fpregs(void)*/ void fpu__clear_user_states (struct fpu *fpu) { -WARN_ON_FPU (fpu != current->thread.fpu); +WARN_ON_FPU (fpu !=x86_task_fpu (current));fpregs_lock ();if (!cpu_feature_enabled (X86_FEATURE_FPU )) {
@@ -774,7 +772,7 @@ void fpu__clear_user_states(struct fpu *fpu)void fpu_flush_thread (void ) { -fpstate_reset (current->thread.fpu); +fpstate_reset (x86_task_fpu (current));fpu_reset_fpregs (); }/*
@@ -815,7 +813,7 @@ void fpregs_lock_and_load(void)*/ void fpregs_assert_state_consistent (void ) { -struct fpu *fpu = current->thread.fpu; +struct fpu *fpu =x86_task_fpu (current);if (test_thread_flag (TIF_NEED_FPU_LOAD ))return ;
@@ -827,7 +825,7 @@ EXPORT_SYMBOL_GPL(fpregs_assert_state_consistent);void fpregs_mark_activate (void ) { -struct fpu *fpu = current->thread.fpu; +struct fpu *fpu =x86_task_fpu (current);fpregs_activate (fpu); fpu->last_cpu =smp_processor_id ();
arch/x86/kernel/fpu/init.c +3 -4
@@ -38,7 +38,7 @@ static void fpu__init_cpu_generic(void)/* Flush out any pending x87 state: */ #ifdef CONFIG_MATH_EMULATION if (!boot_cpu_has (X86_FEATURE_FPU )) -fpstate_init_soft (¤t->thread.fpu->fpstate->regs.soft); +fpstate_init_soft (&x86_task_fpu (current)->fpstate->regs.soft); else#endif asmvolatile ("fninit" );
@@ -78,7 +78,6 @@ static void __init fpu__init_system_early_generic(void)int this_cpu =smp_processor_id ();fpstate_reset (&x86_init_fpu); - current->thread.fpu = &x86_init_fpu;per_cpu (fpu_fpregs_owner_ctx, this_cpu) = &x86_init_fpu; x86_init_fpu.last_cpu = this_cpu;
@@ -165,7 +164,7 @@ static void __init fpu__init_task_struct_size(void)* Subtract off the static size of the register state. * It potentially has a bunch of padding. */ - task_size -=sizeof (current->thread.fpu->__fpstate.regs); + task_size -=sizeof (x86_task_fpu (current)->__fpstate.regs);/* * Add back the dynamically-calculated register state
@@ -210,7 +209,7 @@ static void __init fpu__init_system_xstate_size_legacy(void) fpu_kernel_cfg.default_size = size; fpu_user_cfg.max_size = size; fpu_user_cfg.default_size = size; -fpstate_reset (current->thread.fpu); +fpstate_reset (x86_task_fpu (current)); }/*
arch/x86/kernel/fpu/regset.c +11 -11
@@ -45,7 +45,7 @@ int regset_xregset_fpregs_active(struct task_struct *target, const struct user_r*/ static void sync_fpstate (struct fpu *fpu) { -if (fpu == current->thread.fpu) +if (fpu ==x86_task_fpu (current))fpu_sync_fpstate (fpu); }
@@ -63,7 +63,7 @@ static void fpu_force_restore(struct fpu *fpu)* Only stopped child tasks can be used to modify the FPU * state in the fpstate buffer: */ -WARN_ON_FPU (fpu == current->thread.fpu); +WARN_ON_FPU (fpu ==x86_task_fpu (current));__fpu_invalidate_fpregs_state (fpu); }
@@ -71,7 +71,7 @@ static void fpu_force_restore(struct fpu *fpu)int xfpregs_get (struct task_struct *target,const struct user_regset *regset,struct membuf to) { -struct fpu *fpu = target->thread.fpu; +struct fpu *fpu =x86_task_fpu (target);if (!cpu_feature_enabled (X86_FEATURE_FXSR ))return -ENODEV ;
@@ -91,7 +91,7 @@ int xfpregs_set(struct task_struct *target, const struct user_regset *regset, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - struct fpu *fpu = target->thread.fpu; + struct fpu *fpu = x86_task_fpu(target);struct fxregs_state newstate;int ret;
@@ -133,7 +133,7 @@ int xstateregs_get(struct task_struct *target, const struct user_regset *regset,if (!cpu_feature_enabled (X86_FEATURE_XSAVE ))return -ENODEV ; -sync_fpstate (target->thread.fpu); +sync_fpstate (x86_task_fpu (target));copy_xstate_to_uabi_buf (to, target,XSTATE_COPY_XSAVE );return 0 ;
@@ -143,7 +143,7 @@ int xstateregs_set(struct task_struct *target, const struct user_regset *regset, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - struct fpu *fpu = target->thread.fpu; + struct fpu *fpu = x86_task_fpu(target);struct xregs_state *tmpbuf =NULL ;int ret;
@@ -187,7 +187,7 @@ int ssp_active(struct task_struct *target, const struct user_regset *regset)int ssp_get (struct task_struct *target,const struct user_regset *regset,struct membuf to) { -struct fpu *fpu = target->thread.fpu; +struct fpu *fpu =x86_task_fpu (target);struct cet_user_state *cetregs;if (!cpu_feature_enabled (X86_FEATURE_USER_SHSTK ))
@@ -213,7 +213,7 @@ int ssp_set(struct task_struct *target, const struct user_regset *regset, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - struct fpu *fpu = target->thread.fpu; + struct fpu *fpu = x86_task_fpu(target);struct xregs_state *xsave = &fpu->fpstate->regs.xsave;struct cet_user_state *cetregs;unsigned long user_ssp;
@@ -367,7 +367,7 @@ static void __convert_from_fxsr(struct user_i387_ia32_struct *env,void convert_from_fxsr (struct user_i387_ia32_struct *env,struct task_struct *tsk) { -__convert_from_fxsr (env, tsk, &tsk->thread.fpu->fpstate->regs.fxsave); +__convert_from_fxsr (env, tsk, &x86_task_fpu (tsk)->fpstate->regs.fxsave); } void convert_to_fxsr(struct fxregs_state *fxsave,
@@ -400,7 +400,7 @@ void convert_to_fxsr(struct fxregs_state *fxsave,int fpregs_get (struct task_struct *target,const struct user_regset *regset,struct membuf to) { -struct fpu *fpu = target->thread.fpu; +struct fpu *fpu =x86_task_fpu (target);struct user_i387_ia32_struct env;struct fxregs_state fxsave, *fx;
@@ -432,7 +432,7 @@ int fpregs_set(struct task_struct *target, const struct user_regset *regset, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - struct fpu *fpu = target->thread.fpu; + struct fpu *fpu = x86_task_fpu(target);struct user_i387_ia32_struct env;int ret;
arch/x86/kernel/fpu/signal.c +9 -9
@@ -38,7 +38,7 @@ static inline bool check_xstate_in_sigframe(struct fxregs_state __user *fxbuf,/* Check for the first magic field and other error scenarios. */ if (fx_sw->magic1 !=FP_XSTATE_MAGIC1 || fx_sw->xstate_size < min_xstate_size || - fx_sw->xstate_size > current->thread.fpu->fpstate->user_size || + fx_sw->xstate_size >x86_task_fpu (current)->fpstate->user_size || fx_sw->xstate_size > fx_sw->extended_size) goto setfx;
@@ -54,7 +54,7 @@ static inline bool check_xstate_in_sigframe(struct fxregs_state __user *fxbuf,if (likely (magic2 ==FP_XSTATE_MAGIC2 ))return true; setfx: -trace_x86_fpu_xstate_check_failed (current->thread.fpu); +trace_x86_fpu_xstate_check_failed (x86_task_fpu (current));/* Set the parameters for fx only state */ fx_sw->magic1 =0 ;
@@ -69,13 +69,13 @@ static inline bool check_xstate_in_sigframe(struct fxregs_state __user *fxbuf,static inline bool save_fsave_header (struct task_struct *tsk,void __user *buf) {if (use_fxsr ()) { -struct xregs_state *xsave = &tsk->thread.fpu->fpstate->regs.xsave; +struct xregs_state *xsave = &x86_task_fpu (tsk)->fpstate->regs.xsave;struct user_i387_ia32_struct env;struct _fpstate_32 __user *fp = buf;fpregs_lock ();if (!test_thread_flag (TIF_NEED_FPU_LOAD )) -fxsave (&tsk->thread.fpu->fpstate->regs.fxsave); +fxsave (&x86_task_fpu (tsk)->fpstate->regs.fxsave);fpregs_unlock ();convert_from_fxsr (&env, tsk);
@@ -188,7 +188,7 @@ static inline int copy_fpregs_to_sigframe(struct xregs_state __user *buf)bool copy_fpstate_to_sigframe (void __user *buf,void __user *buf_fx,int size) {struct task_struct *tsk = current; -struct fpstate *fpstate = tsk->thread.fpu->fpstate; +struct fpstate *fpstate =x86_task_fpu (tsk)->fpstate;bool ia32_fxstate = (buf != buf_fx);int ret;
@@ -276,7 +276,7 @@ static int __restore_fpregs_from_user(void __user *buf, u64 ufeatures,*/ static bool restore_fpregs_from_user (void __user *buf,u64 xrestore,bool fx_only) { -struct fpu *fpu = current->thread.fpu; +struct fpu *fpu =x86_task_fpu (current);int ret;/* Restore enabled features only. */
@@ -336,7 +336,7 @@ static bool __fpu_restore_sig(void __user *buf, void __user *buf_fx, bool ia32_fxstate) {struct task_struct *tsk = current; -struct fpu *fpu = tsk->thread.fpu; +struct fpu *fpu =x86_task_fpu (tsk);struct user_i387_ia32_struct env;bool success, fx_only = false;union fpregs_state *fpregs;
@@ -456,7 +456,7 @@ static inline unsigned int xstate_sigframe_size(struct fpstate *fpstate)*/ bool fpu__restore_sig (void __user *buf,int ia32_frame) { -struct fpu *fpu = current->thread.fpu; +struct fpu *fpu =x86_task_fpu (current);void __user *buf_fx = buf;bool ia32_fxstate = false;bool success = false;
@@ -503,7 +503,7 @@ unsigned long fpu__alloc_mathframe(unsigned long sp, int ia32_frame, unsigned long *buf_fx, unsigned long *size) { - unsigned long frame_size = xstate_sigframe_size(current->thread.fpu->fpstate); + unsigned long frame_size = xstate_sigframe_size(x86_task_fpu(current)->fpstate); *buf_fx = sp =round_down (sp - frame_size,64 );if (ia32_frame &&use_fxsr ()) {
arch/x86/kernel/fpu/xstate.c +11 -11
@@ -735,7 +735,7 @@ static void __init fpu__init_disable_system_xstate(unsigned int legacy_size)*/ init_fpstate.xfd =0 ; -fpstate_reset (current->thread.fpu); +fpstate_reset (x86_task_fpu (current)); }/*
@@ -845,7 +845,7 @@ void __init fpu__init_system_xstate(unsigned int legacy_size) goto out_disable;/* Reset the state for the current task */ -fpstate_reset (current->thread.fpu); +fpstate_reset (x86_task_fpu (current));/* * Update info used for ptrace frames; use standard-format size and no
@@ -919,7 +919,7 @@ void fpu__resume_cpu(void) }if (fpu_state_size_dynamic ()) -wrmsrl (MSR_IA32_XFD , current->thread.fpu->fpstate->xfd); +wrmsrl (MSR_IA32_XFD ,x86_task_fpu (current)->fpstate->xfd); }/*
@@ -1188,8 +1188,8 @@ void __copy_xstate_to_uabi_buf(struct membuf to, struct fpstate *fpstate,void copy_xstate_to_uabi_buf (struct membuf to,struct task_struct *tsk,enum xstate_copy_mode copy_mode) { -__copy_xstate_to_uabi_buf (to, tsk->thread.fpu->fpstate, - tsk->thread.fpu->fpstate->user_xfeatures, +__copy_xstate_to_uabi_buf (to,x86_task_fpu (tsk)->fpstate, +x86_task_fpu (tsk)->fpstate->user_xfeatures, tsk->thread.pkru, copy_mode); }
@@ -1329,7 +1329,7 @@ int copy_uabi_from_kernel_to_xstate(struct fpstate *fpstate, const void *kbuf, uint copy_sigframe_from_user_to_xstate (struct task_struct *tsk,const void __user *ubuf) { -return copy_uabi_to_xstate (tsk->thread.fpu->fpstate,NULL , ubuf, &tsk->thread.pkru); +return copy_uabi_to_xstate (x86_task_fpu (tsk)->fpstate,NULL , ubuf, &tsk->thread.pkru); }static bool validate_independent_components (u64 mask)
@@ -1423,7 +1423,7 @@ static bool xstate_op_valid(struct fpstate *fpstate, u64 mask, bool rstor)* The XFD MSR does not match fpstate->xfd. That's invalid when * the passed in fpstate is current's fpstate. */ -if (fpstate->xfd == current->thread.fpu->fpstate->xfd) +if (fpstate->xfd ==x86_task_fpu (current)->fpstate->xfd)return false;/*
@@ -1500,7 +1500,7 @@ void fpstate_free(struct fpu *fpu)static int fpstate_realloc (u64 xfeatures,unsigned int ksize,unsigned int usize,struct fpu_guest *guest_fpu) { -struct fpu *fpu = current->thread.fpu; +struct fpu *fpu =x86_task_fpu (current);struct fpstate *curfps, *newfps =NULL ;unsigned int fpsize;bool in_use;
@@ -1593,7 +1593,7 @@ static int __xstate_request_perm(u64 permitted, u64 requested, bool guest)* AVX512. */ bool compacted =cpu_feature_enabled (X86_FEATURE_XCOMPACTED ); -struct fpu *fpu = current->group_leader->thread.fpu; +struct fpu *fpu =x86_task_fpu (current->group_leader);struct fpu_state_perm *perm;unsigned int ksize, usize;u64 mask;
@@ -1696,7 +1696,7 @@ int __xfd_enable_feature(u64 xfd_err, struct fpu_guest *guest_fpu)return -EPERM ; } - fpu = current->group_leader->thread.fpu; + fpu =x86_task_fpu (current->group_leader); perm = guest_fpu ? &fpu->guest_perm : &fpu->perm; ksize = perm->__state_size; usize = perm->__user_state_size;
@@ -1801,7 +1801,7 @@ long fpu_xstate_prctl(int option, unsigned long arg2)*/ static void avx512_status (struct seq_file *m,struct task_struct *task) { -unsigned long timestamp =READ_ONCE (task->thread.fpu->avx512_timestamp); +unsigned long timestamp =READ_ONCE (x86_task_fpu (task)->avx512_timestamp);long delta;if (!timestamp) {
arch/x86/kernel/fpu/xstate.h +3 -3
@@ -22,7 +22,7 @@ static inline void xstate_init_xcomp_bv(struct xregs_state *xsave, u64 mask)static inline u64 xstate_get_group_perm (bool guest) { -struct fpu *fpu = current->group_leader->thread.fpu; +struct fpu *fpu =x86_task_fpu (current->group_leader);struct fpu_state_perm *perm;/* Pairs with WRITE_ONCE() in xstate_request_perm() */
@@ -265,7 +265,7 @@ static inline int xsave_to_user_sigframe(struct xregs_state __user *buf)* internally, e.g. PKRU. That's user space ABI and also required * to allow the signal handler to modify PKRU. */ -struct fpstate *fpstate = current->thread.fpu->fpstate; +struct fpstate *fpstate =x86_task_fpu (current)->fpstate;u64 mask = fpstate->user_xfeatures;u32 lmask;u32 hmask;
@@ -296,7 +296,7 @@ static inline int xrstor_from_user_sigframe(struct xregs_state __user *buf, u64u32 hmask = mask >>32 ;int err; -xfd_validate_state (current->thread.fpu->fpstate, mask, true); +xfd_validate_state (x86_task_fpu (current)->fpstate, mask, true);stac ();XSTATE_OP (XRSTOR , xstate, lmask, hmask, err);
arch/x86/kernel/process.c +2 -4
@@ -96,8 +96,6 @@ int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)#ifdef CONFIG_VM86 dst->thread.vm86 =NULL ;#endif -/* Drop the copied pointer to current's fpstate */ - dst->thread.fpu =NULL ;return 0 ; }
@@ -106,7 +104,7 @@ int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)void arch_release_task_struct (struct task_struct *tsk) {if (fpu_state_size_dynamic ()) -fpstate_free (tsk->thread.fpu); +fpstate_free (x86_task_fpu (tsk)); }#endif
@@ -116,7 +114,7 @@ void arch_release_task_struct(struct task_struct *tsk)void exit_thread (struct task_struct *tsk) {struct thread_struct *t = &tsk->thread; -struct fpu *fpu = t->fpu; +struct fpu *fpu =x86_task_fpu (tsk);if (test_thread_flag (TIF_IO_BITMAP ))io_bitmap_exit (tsk);
arch/x86/kernel/signal.c +3 -3
@@ -228,7 +228,7 @@ static void handle_signal(struct ksignal *ksig, struct pt_regs *regs) { bool stepping, failed; -struct fpu *fpu = current->thread.fpu; +struct fpu *fpu =x86_task_fpu (current);if (v8086_mode (regs))save_v86_state ((struct kernel_vm86_regs *) regs,VM86_SIGNAL );
@@ -396,14 +396,14 @@ bool sigaltstack_size_valid(size_t ss_size)if (!fpu_state_size_dynamic () && !strict_sigaltstack_size)return true; - fsize += current->group_leader->thread.fpu->perm.__user_state_size; + fsize +=x86_task_fpu (current->group_leader)->perm.__user_state_size;if (likely (ss_size > fsize))return true;if (strict_sigaltstack_size)return ss_size > fsize; - mask = current->group_leader->thread.fpu->perm.__state_perm; + mask =x86_task_fpu (current->group_leader)->perm.__state_perm;if (mask &XFEATURE_MASK_USER_DYNAMIC )return ss_size > fsize;
arch/x86/kernel/traps.c +1 -1
@@ -1148,7 +1148,7 @@ DEFINE_IDTENTRY_RAW(exc_debug)static void math_error (struct pt_regs *regs,int trapnr) {struct task_struct *task = current; -struct fpu *fpu = task->thread.fpu; +struct fpu *fpu =x86_task_fpu (task);int si_code;char *str = (trapnr ==X86_TRAP_MF ) ?"fpu exception" :"simd exception" ;
arch/x86/kernel/vmlinux.lds.S +4 -0
@@ -170,6 +170,10 @@ SECTIONS /* equivalent to task_pt_regs(&init_task) */ __top_init_kernel_stack = __end_init_stack - TOP_OF_KERNEL_STACK_PADDING - PTREGS_SIZE; + __x86_init_fpu_begin = .; + . = __x86_init_fpu_begin + 128*PAGE_SIZE; + __x86_init_fpu_end = .; + #ifdef CONFIG_X86_32 /* 32 bit has nosave before _edata */ NOSAVE_DATA
arch/x86/math-emu/fpu_aux.c +1 -1
@@ -53,7 +53,7 @@ void fpstate_init_soft(struct swregs_state *soft)void finit (void ) { -fpstate_init_soft (¤t->thread.fpu->fpstate->regs.soft); +fpstate_init_soft (&x86_task_fpu (current)->fpstate->regs.soft); }/*
arch/x86/math-emu/fpu_entry.c +2 -2
@@ -641,7 +641,7 @@ int fpregs_soft_set(struct task_struct *target, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - struct swregs_state *s387 = &target->thread.fpu->fpstate->regs.soft; + struct swregs_state *s387 = &x86_task_fpu(target)->fpstate->regs.soft;void *space = s387->st_space;int ret;int offset, other, i, tags, regnr, tag, newtop;
@@ -692,7 +692,7 @@ int fpregs_soft_get(struct task_struct *target, const struct user_regset *regset, struct membuf to) { -struct swregs_state *s387 = &target->thread.fpu->fpstate->regs.soft; +struct swregs_state *s387 = &x86_task_fpu (target)->fpstate->regs.soft;const void *space = s387->st_space;int offset = (S387 ->ftop &7 ) *10 , other =80 - offset;
arch/x86/math-emu/fpu_system.h +1 -1
@@ -73,7 +73,7 @@ static inline bool seg_writable(struct desc_struct *d)return (d->type &SEG_TYPE_EXECUTE_MASK ) ==SEG_TYPE_WRITABLE ; } -#define I387 (¤t->thread.fpu->fpstate->regs) +#define I387 (&x86_task_fpu(current)->fpstate->regs)#define FPU_info (I387->soft.info)#define FPU_CS (*(unsigned short *) &(FPU_info->regs->cs))
arch/x86/mm/extable.c +1 -1
@@ -111,7 +111,7 @@ static bool ex_handler_sgx(const struct exception_table_entry *fixup,/* * Handler for when we fail to restore a task's FPU state. We should never get -* here because the FPU state of a task using the FPU (task->thread.fpu->state) +* here because the FPU state of a task using the FPU (x86_task_fpu(task)->state) * should always be valid. However, past bugs have allowed userspace to set * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn(). * These caused XRSTOR to fail when switching to the task, leaking the FPU
Signature · 1 line
2.43.0
arch/x86/kernel/fpu/core.c int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal, · line 594
Quoted 7 lines · Ingo Molnar
arch/x86/kernel/fpu/core.c +1
-1@@ -591,13 +591,11 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,* This is safe because task_struct size is a multiple of cacheline size. */ struct fpu *dst_fpu = (void *)dst +sizeof (*dst); -struct fpu *src_fpu = current->thread.fpu; +struct fpu *src_fpu =x86_task_fpu (current);
I think this patch can also change
struct fpu *dst_fpu = (void *)dst + sizeof(*dst);
above to use x86_task_fpu(dst).
Oleg.
Quoted 11 lines · Oleg Nesterov
arch/x86/kernel/fpu/core.c int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal, · line 594
Quoted 7 lines · Ingo Molnar
arch/x86/kernel/fpu/core.c +1
-1@@ -591,13 +591,11 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,* This is safe because task_struct size is a multiple of cacheline size. */ struct fpu *dst_fpu = (void *)dst +sizeof (*dst); -struct fpu *src_fpu = current->thread.fpu; +struct fpu *src_fpu =x86_task_fpu (current);I think this patch can also change
Quoted 3 lines
struct fpu *dst_fpu = (void *)dst + sizeof(*dst);above to use
x86_task_fpu(dst).
Yeah, so I'd prefer to keep it open coded, because of the comment and the debug check makes a lot more sense if the pointer calculation is visible:
/*
* We allocate the new FPU structure right after the end of the task struct.
* task allocation size already took this into account.
*
* This is safe because task_struct size is a multiple of cacheline size.
*/
struct fpu *dst_fpu = (void *)dst + sizeof(*dst);
BUILD_BUG_ON(sizeof(*dst) % SMP_CACHE_BYTES != 0);But yes, you are right, this is technically an open-coded x86_task_fpu().
Thanks,
Ingo
Quoted 14 lines · Ingo Molnar
Quoted 11 lines · Oleg Nesterov
arch/x86/kernel/fpu/core.c int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal, · line 594
Quoted 7 lines · Ingo Molnar
arch/x86/kernel/fpu/core.c +1
-1@@ -591,13 +591,11 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,* This is safe because task_struct size is a multiple of cacheline size. */ struct fpu *dst_fpu = (void *)dst +sizeof (*dst); -struct fpu *src_fpu = current->thread.fpu; +struct fpu *src_fpu =x86_task_fpu (current);I think this patch can also change
Quoted 6 lines
Quoted 3 lines
struct fpu *dst_fpu = (void *)dst + sizeof(*dst);above to use
x86_task_fpu(dst).Yeah, so I'd prefer to keep it open coded, because of the comment and the debug check makes a lot more sense if the pointer calculation is visible:
On a second thought I changed it to your suggested variant:
struct fpu *src_fpu = x86_task_fpu(current);
struct fpu *dst_fpu = x86_task_fpu(dst);because you are right, it's in fact easier to read this way.
Thanks,
Ingo
Quoted 24 lines · Ingo Molnar
Quoted 14 lines · Ingo Molnar
Quoted 11 lines · Oleg Nesterov
arch/x86/kernel/fpu/core.c int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal, · line 594
Quoted 7 lines · Ingo Molnar
arch/x86/kernel/fpu/core.c +1
-1@@ -591,13 +591,11 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,* This is safe because task_struct size is a multiple of cacheline size. */ struct fpu *dst_fpu = (void *)dst +sizeof (*dst); -struct fpu *src_fpu = current->thread.fpu; +struct fpu *src_fpu =x86_task_fpu (current);I think this patch can also change
Quoted 6 lines
Quoted 3 lines
struct fpu *dst_fpu = (void *)dst + sizeof(*dst);above to use
x86_task_fpu(dst).Yeah, so I'd prefer to keep it open coded, because of the comment and the debug check makes a lot more sense if the pointer calculation is visible:
Quoted 6 lines
On a second thought I changed it to your suggested variant:
struct fpu *src_fpu = x86_task_fpu(current); struct fpu *dst_fpu = x86_task_fpu(dst);because you are right, it's in fact easier to read this way.
On a third thought, while more readable, this doesn't work in practice with the current scheme, because x86_task_fpu() gets called on kthreads in fpu_clone(), which trips up the new debugging code.
We could resolve it by special-casing PF_KTHREAD here too, but that weakens the whole readability argument. I'll leave it as-is for now.
Thanks,
Ingo
Quoted 4 lines · Ingo Molnar
On a third thought, while more readable, this doesn't work in practice with the current scheme, because
x86_task_fpu()gets called on kthreads infpu_clone(), which trips up the new debugging code.
Yes, yes, I even mentioned this in reply to 3/3.
Quoted 2 lines
We could resolve it by special-casing
PF_KTHREADhere too, but that weakens the whole readability argument. I'll leave it as-is for now.
Agreed.
Note that PF_KTHREAD | PF_USER_WORKER is already a special case in fpu_clone(), see the "if (minimal)" check. We can probably do more changes later, for example I don't even understand why do we need to initialize dst_fpu->fpstate->regs, switch_fpu_return() is only called by arch_exit_to_user_mode_prepare().
Oleg.
Quoted 2 lines · Ingo Molnar
As suggested by Oleg, remove the
thread::fpupointer, as we can calculate it viax86_task_fpu()at compile-time.
I'm seeing boot failures in a TDX VM that bisects to this commit in tip (807333522953). The host is a pile of out-of-tree KVM patches, but the nature of the change makes me wonder if it's not TDX related. The failure looks like the below on the first bad commit. Some of the later commits had a failure with more of an FPU associated stack trace. It also only shows when I have lock debugging on:
#
# Lock Debugging (spinlocks, mutexes, etc...)
#CONFIG_LOCK_DEBUGGING_SUPPORT=yCONFIG_PROVE_LOCKING=y
# CONFIG_PROVE_RAW_LOCK_NESTING is not set
# CONFIG_LOCK_STAT is not setCONFIG_DEBUG_RT_MUTEXES=yCONFIG_DEBUG_SPINLOCK=yCONFIG_DEBUG_MUTEXES=yCONFIG_DEBUG_WW_MUTEX_SLOWPATH=yCONFIG_DEBUG_RWSEMS=yCONFIG_DEBUG_LOCK_ALLOC=yCONFIG_LOCKDEP=yCONFIG_LOCKDEP_BITS=15CONFIG_LOCKDEP_CHAINS_BITS=16CONFIG_LOCKDEP_STACK_TRACE_BITS=19 CONFIG_LOCKDEP_STACK_TRACE_HASH_BITS=14 CONFIG_LOCKDEP_CIRCULAR_QUEUE_BITS=12
# CONFIG_DEBUG_LOCKDEP is not setCONFIG_DEBUG_ATOMIC_SLEEP=y
# CONFIG_DEBUG_LOCKING_API_SELFTESTS is not set
# CONFIG_LOCK_TORTURE_TEST is not set
# CONFIG_WW_MUTEX_SELFTEST is not set
# CONFIG_SCF_TORTURE_TEST is not set
# CONFIG_CSD_LOCK_WAIT_DEBUG is not set
# end of Lock Debugging (spinlocks, mutexes, etc...)
If it's not obvious, I can investigate some more tomorrow on a more normal VM configuration.
[ 8.830714] ------------[ cut here ]------------ [ 8.830714] DEBUG_LOCKS_WARN_ON(1) [ 8.830714] WARNING: CPU: 0 PID: 0 at kernel/locking/lockdep.c:232 __lock_acquire+0xa5c/0x2120 [ 8.830714] CPU: 0 PID: 0 Comm: swapper/0 Tainted: G S 6.10.0-rc3-00004-g807333522953 #117 [ 8.830714] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS unknown 2/2/2022 [ 8.830714] RIP: 0010:__lock_acquire+0xa5c/0x2120 [ 8.830714] Code: b8 85 c0 0f 84 39 fd ff ff 8b 05 13 4d 92 01 85 c0 0f 85 2b fd ff ff 48 c7 c6 c5 93 3e 82 48 c7 c7 66 aa 39 82 e8 24 34 f8 ff <0f> 0b 31 c0 44 8b 5d b8 e9 60 f7 ff ff 88 55 b0 44 89 5d b8 e8 4b [ 8.830714] RSP: 0000:ffffffff82603b28 EFLAGS: 00010086 [ 8.830714] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000 [ 8.830714] RDX: 0000000000000004 RSI: 00000000ffffffea RDI: 00000000ffffffff [ 8.830714] RBP: ffffffff82603ba8 R08: ff1100046fffdfe8 R09: 0000000000000003 [ 8.830714] R10: ff11000437ffe000 R11: ff11000464000d78 R12: ffffffff826b57a8 [ 8.830714] R13: ffffffff826b4e00 R14: ffffffff826b57a8 R15: 51eb851eb8f20808 [ 8.830714] FS: 0000000000000000(0000) GS:ff11000427a00000(0000) knlGS:0000000000000000 [ 8.830714] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 8.830714] CR2: ff1100047ffff000 CR3: 0000000002eda001 CR4: 0000000000771ef0 [ 8.830714] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 8.830714] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400 [ 8.830714] PKRU: 55555554 [ 8.830714] Call Trace: [ 8.830714] <TASK> [ 8.830714] ? show_regs+0x60/0x70 [ 8.830714] ? __warn+0x84/0x180 [ 8.830714] ? __lock_acquire+0xa5c/0x2120 [ 8.830714] ? report_bug+0x1f3/0x200 [ 8.830714] ? handle_bug+0x40/0x70 [ 8.830714] ? exc_invalid_op+0x19/0x70 [ 8.830714] ? asm_exc_invalid_op+0x1b/0x20 [ 8.830714] ? __lock_acquire+0xa5c/0x2120 [ 8.830714] ? __lock_acquire+0xa5c/0x2120 [ 8.830714] ? common_startup_64+0x13e/0x148 [ 8.830714] lock_acquire+0xc8/0x2e0 [ 8.830714] ? copy_process+0x107/0x2b80 [ 8.830714] ? __this_cpu_preempt_check+0x13/0x20 [ 8.830714] ? _raw_spin_lock_irq+0x4b/0x50 [ 8.830714] _raw_spin_lock_irq+0x37/0x50 [ 8.830714] ? copy_process+0x107/0x2b80 [ 8.830714] copy_process+0x107/0x2b80 [ 8.830714] ? find_held_lock+0x31/0x90 [ 8.830714] ? lock_release+0x130/0x290 [ 8.830714] ? _raw_spin_unlock_irqrestore+0x2c/0x60 [ 8.830714] ? find_held_lock+0x31/0x90 [ 8.830714] kernel_clone+0x97/0x3b0 [ 8.830714] ? __mutex_unlock_slowpath+0x3c/0x2a0 [ 8.830714] user_mode_thread+0x59/0x70 [ 8.830714] ? rest_init+0x190/0x190 [ 8.830714] rest_init+0x1e/0x190 [ 8.830714] start_kernel+0x672/0x790 [ 8.830714] x86_64_start_reservations+0x18/0x30 [ 8.830714] x86_64_start_kernel+0xd0/0xe0 [ 8.830714] common_startup_64+0x13e/0x148 [ 8.830714] </TASK> [ 8.830714] irq event stamp: 27428 [ 8.830714] hardirqs last enabled at (27427): [<ffffffff81dc236c>] _raw_spin_unlock_irqrestore+0x2c/0x60 [ 8.830714] hardirqs last disabled at (27428): [<ffffffff81dc208b>] _raw_spin_lock_irq+0x4b/0x50 [ 8.830714] softirqs last enabled at (27360): [<ffffffff8116d93c>] cgroup_idr_alloc.constprop.0+0x5c/0x100 [ 8.830714] softirqs last disabled at (27358): [<ffffffff8116d916>] cgroup_idr_alloc.constprop.0+0x36/0x100 [ 8.830714] ---[ end trace 0000000000000000 ]--- [ 8.830714] BUG: kernel NULL pointer dereference, address: 00000000000000c4 [ 8.830714] #PF: supervisor read access in kernel mode [ 8.830714] #PF: error_code(0x0000) - not-present page [ 8.830714] PGD 0 [ 8.830714] Oops: Oops: 0000 [#1] PREEMPT SMP [ 8.830714] CPU: 0 PID: 0 Comm: swapper/0 Tainted: G S W 6.10.0-rc3-00004-g807333522953 #117 [ 8.830714] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS unknown 2/2/2022 [ 8.830714] RIP: 0010:__lock_acquire+0x1c9/0x2120 [ 8.830714] Code: 45 28 41 89 44 24 24 4c 89 f8 25 ff 1f 00 00 48 0f a3 05 2a 03 dc 01 0f 83 aa 05 00 00 48 69 c0 c8 00 00 00 48 05 a0 bc eb 82 <0f> b6 90 c4 00 00 00 41 0f b7 44 24 20 66 25 ff 1f 0f b7 c0 48 0f [ 8.830714] RSP: 0000:ffffffff82603b28 EFLAGS: 00010046 [ 8.830714] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000 [ 8.830714] RDX: 0000000000000004 RSI: 00000000ffffffea RDI: 00000000ffffffff [ 8.830714] RBP: ffffffff82603ba8 R08: ff1100046fffdfe8 R09: 0000000000000003 [ 8.830714] R10: ff11000437ffe000 R11: 0000000000000001 R12: ffffffff826b57a8 [ 8.830714] R13: ffffffff826b4e00 R14: ffffffff826b57a8 R15: 51eb851eb8f20808 [ 8.830714] FS: 0000000000000000(0000) GS:ff11000427a00000(0000) knlGS:0000000000000000 [ 8.830714] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 8.830714] CR2: 00000000000000c4 CR3: 0000000002eda001 CR4: 0000000000771ef0 [ 8.830714] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 8.830714] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400 [ 8.830714] PKRU: 55555554 [ 8.830714] Call Trace: [ 8.830714] <TASK> [ 8.830714] ? show_regs+0x60/0x70 [ 8.830714] ? __die+0x20/0x60 [ 8.830714] ? page_fault_oops+0x15a/0x480 [ 8.830714] ? __lock_acquire+0xa5c/0x2120 [ 8.830714] ? exc_page_fault+0x437/0x910 [ 8.830714] ? asm_exc_page_fault+0x27/0x30 [ 8.830714] ? __lock_acquire+0x1c9/0x2120 [ 8.830714] ? __lock_acquire+0xa5c/0x2120 [ 8.830714] ? common_startup_64+0x13e/0x148 [ 8.830714] lock_acquire+0xc8/0x2e0 [ 8.830714] ? copy_process+0x107/0x2b80 [ 8.830714] ? __this_cpu_preempt_check+0x13/0x20 [ 8.830714] ? _raw_spin_lock_irq+0x4b/0x50 [ 8.830714] _raw_spin_lock_irq+0x37/0x50 [ 8.830714] ? copy_process+0x107/0x2b80 [ 8.830714] copy_process+0x107/0x2b80 [ 8.830714] ? find_held_lock+0x31/0x90 [ 8.830714] ? lock_release+0x130/0x290 [ 8.830714] ? _raw_spin_unlock_irqrestore+0x2c/0x60 [ 8.830714] ? find_held_lock+0x31/0x90 [ 8.830714] kernel_clone+0x97/0x3b0 [ 8.830714] ? __mutex_unlock_slowpath+0x3c/0x2a0 [ 8.830714] user_mode_thread+0x59/0x70 [ 8.830714] ? rest_init+0x190/0x190 [ 8.830714] rest_init+0x1e/0x190 [ 8.830714] start_kernel+0x672/0x790 [ 8.830714] x86_64_start_reservations+0x18/0x30 [ 8.830714] x86_64_start_kernel+0xd0/0xe0 [ 8.830714] common_startup_64+0x13e/0x148 [ 8.830714] </TASK> [ 8.830714] CR2: 00000000000000c4 [ 8.830714] ---[ end trace 0000000000000000 ]--- [ 8.830714] RIP: 0010:__lock_acquire+0x1c9/0x2120 [ 8.830714] Code: 45 28 41 89 44 24 24 4c 89 f8 25 ff 1f 00 00 48 0f a3 05 2a 03 dc 01 0f 83 aa 05 00 00 48 69 c0 c8 00 00 00 48 05 a0 bc eb 82 <0f> b6 90 c4 00 00 00 41 0f b7 44 24 20 66 25 ff 1f 0f b7 c0 48 0f [ 8.830714] RSP: 0000:ffffffff82603b28 EFLAGS: 00010046 [ 8.830714] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000 [ 8.830714] RDX: 0000000000000004 RSI: 00000000ffffffea RDI: 00000000ffffffff [ 8.830714] RBP: ffffffff82603ba8 R08: ff1100046fffdfe8 R09: 0000000000000003 [ 8.830714] R10: ff11000437ffe000 R11: 0000000000000001 R12: ffffffff826b57a8 [ 8.830714] R13: ffffffff826b4e00 R14: ffffffff826b57a8 R15: 51eb851eb8f20808 [ 8.830714] FS: 0000000000000000(0000) GS:ff11000427a00000(0000) knlGS:0000000000000000 [ 8.830714] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 8.830714] CR2: 00000000000000c4 CR3: 0000000002eda001 CR4: 0000000000771ef0 [ 8.830714] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 8.830714] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400 [ 8.830714] PKRU: 55555554 [ 8.830714] Kernel panic - not syncing: Attempted to kill the idle task! [ 8.830714] ---[ end Kernel panic - not syncing: Attempted to kill the idle
task! ]---
Quoted 3 lines · Rick Edgecombe
If it's not obvious, I can investigate some more tomorrow on a more normal VM configuration.
I see the same issue on a normal VM. CONFIG_DEBUG_LOCK_ALLOC seems to be the critical config.
PATCH v3 3/3 x86/fpu: Remove init_task FPU state dependencies, add debugging warning
In reply to Ingo Molnarinit_task's FPU state initialization was a bit of a hack:
__x86_init_fpu_begin = .; . = __x86_init_fpu_begin + 128*PAGE_SIZE; __x86_init_fpu_end = .;
But the init task isn't supposed to be using the FPU in any case, so remove the hack and add in some debug warnings.
--- arch/x86/include/asm/processor.h | 6 +++++- arch/x86/kernel/fpu/core.c | 12 +++++++++--- arch/x86/kernel/fpu/init.c | 5 ++--- arch/x86/kernel/fpu/xstate.c | 3 --- arch/x86/kernel/vmlinux.lds.S | 4 ---- 5 files changed, 16 insertions(+), 14 deletions(-)
arch/x86/include/asm/processor.h +5 -1
@@ -504,7 +504,11 @@ struct thread_struct {#endif }; -#define x86_task_fpu (task) ((struct fpu *)((void *)task + sizeof(*task))) +#ifdef CONFIG_X86_DEBUG_FPU +extern struct fpu *x86_task_fpu (struct task_struct *task); +#else +# define x86_task_fpu (task) ((struct fpu *)((void *)task + sizeof(*task))) +#endif /* * X86 doesn't need any embedded-FPU-struct quirks:
arch/x86/kernel/fpu/core.c +9 -3
@@ -51,6 +51,15 @@ static DEFINE_PER_CPU(bool, in_kernel_fpu);*/ DEFINE_PER_CPU(struct fpu *, fpu_fpregs_owner_ctx); +#ifdef CONFIG_X86_DEBUG_FPU +struct fpu *x86_task_fpu (struct task_struct *task) +{ +WARN_ON_ONCE (task == &init_task); + +return (void *)task +sizeof (*task); +} +#endif +/* * Can we use the FPU in kernel mode with the * whole "kernel_fpu_begin/end()" sequence?
@@ -591,10 +600,8 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,* This is safe because task_struct size is a multiple of cacheline size. */ struct fpu *dst_fpu = (void *)dst +sizeof (*dst); -struct fpu *src_fpu =x86_task_fpu (current);BUILD_BUG_ON (sizeof (*dst) %SMP_CACHE_BYTES !=0 ); -BUG_ON (!src_fpu);/* The new task's FPU state cannot be valid in the hardware. */ dst_fpu->last_cpu =-1 ;
@@ -657,7 +664,6 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,if (update_fpu_shstk (dst, ssp))return 1 ; -trace_x86_fpu_copy_src (src_fpu);trace_x86_fpu_copy_dst (dst_fpu);return 0 ;
arch/x86/kernel/fpu/init.c +2 -3
@@ -38,7 +38,7 @@ static void fpu__init_cpu_generic(void)/* Flush out any pending x87 state: */ #ifdef CONFIG_MATH_EMULATION if (!boot_cpu_has (X86_FEATURE_FPU )) -fpstate_init_soft (&x86_task_fpu (current)->fpstate->regs.soft); + ; else#endif asmvolatile ("fninit" );
@@ -164,7 +164,7 @@ static void __init fpu__init_task_struct_size(void)* Subtract off the static size of the register state. * It potentially has a bunch of padding. */ - task_size -=sizeof (x86_task_fpu (current)->__fpstate.regs); + task_size -=sizeof (union fpregs_state );/* * Add back the dynamically-calculated register state
@@ -209,7 +209,6 @@ static void __init fpu__init_system_xstate_size_legacy(void) fpu_kernel_cfg.default_size = size; fpu_user_cfg.max_size = size; fpu_user_cfg.default_size = size; -fpstate_reset (x86_task_fpu (current)); }/*
arch/x86/kernel/fpu/xstate.c +0 -3
@@ -844,9 +844,6 @@ void __init fpu__init_system_xstate(unsigned int legacy_size)if (err) goto out_disable; -/* Reset the state for the current task */ -fpstate_reset (x86_task_fpu (current)); -/* * Update info used for ptrace frames; use standard-format size and no * supervisor xstates:
arch/x86/kernel/vmlinux.lds.S +0 -4
@@ -170,10 +170,6 @@ SECTIONS /* equivalent to task_pt_regs(&init_task) */ __top_init_kernel_stack = __end_init_stack - TOP_OF_KERNEL_STACK_PADDING - PTREGS_SIZE; - __x86_init_fpu_begin = .; - . = __x86_init_fpu_begin + 128*PAGE_SIZE; - __x86_init_fpu_end = .; - #ifdef CONFIG_X86_32 /* 32 bit has nosave before _edata */ NOSAVE_DATA
Signature · 1 line
2.43.0
Re: [PATCH 3/3] x86/fpu: Remove init_task FPU state dependencies, add debugging warning
Quoted 2 lines · Ingo Molnar
But the init task isn't supposed to be using the FPU in any case,
Afaics, the same is true for any PF_KTHREAD/USER_WORKER thread?
arch/x86/kernel/fpu/core.c static DEFINE_PER_CPU(bool, in_kernel_fpu); · line 60
Quoted 7 lines
arch/x86/kernel/fpu/core.c +7
-0+#ifdef CONFIG_X86_DEBUG_FPU +struct fpu *x86_task_fpu (struct task_struct *task) +{ +WARN_ON_ONCE (task == &init_task); + +return (void *)task +sizeof (*task); +}
So perhaps WARN_ON_ONCE(task->flags & PF_KTHREAD) makes more sense?
Although in this case fpu_clone() can't use x86_task_fpu(dst) as I tried to suggest in reply to 2/3.
This is offtopic right now, but I am wondering if it makes any sense to make arch_task_struct_size depend on PF_KTHREAD... I mean, create another task_struct_kthread_cachep used by alloc_task_struct_node() if PF_KTHREAD.
Oleg.
Re: [PATCH 3/3] x86/fpu: Remove init_task FPU state dependencies, add debugging warning
Quoted 6 lines · Oleg Nesterov
Quoted 2 lines · Ingo Molnar
But the init task isn't supposed to be using the FPU in any case,
Afaics, the same is true for any PF_KTHREAD/USER_WORKER thread?
I don't think so. We have various users of kernel_fpu_begin()/end() that are very much about things like crypto and RAID xor memory copies etc that will be used by kernel worker threads.
In fact, as far as I know, we'll use the FPU in the init_task too thanks to irq_fpu_usable(). Look up the nft_pipapo_avx2_lookup() function.
Maybe other patches removed that, I didn't check the context, but the "init_task doesn't use FPU" doesn't actually sound true to me.
Linus
Re: [PATCH 3/3] x86/fpu: Remove init_task FPU state dependencies, add debugging warning
Quoted 5 lines · Linus Torvalds
Quoted 3 lines · Oleg Nesterov
Quoted 1 line · Ingo Molnar
Quoted 7 lines
Quoted 3 lines
Quoted 1 line
But the init task isn't supposed to be using the FPU in any case,
Afaics, the same is true for any PF_KTHREAD/USER_WORKER thread?
I don't think so. We have various users of kernel_fpu_begin()/end() that are very much about things like crypto and RAID xor memory copies etc that will be used by kernel worker threads.
Yes, but kernel_fpu_begin() never does save_fpregs_to_fpstate() if current->flags & PF_KTHREAD ?
Oleg.
Re: [PATCH 3/3] x86/fpu: Remove init_task FPU state dependencies, add debugging warning
Quoted 3 lines · Oleg Nesterov
Yes, but
kernel_fpu_begin()never doessave_fpregs_to_fpstate()ifcurrent->flags&PF_KTHREAD?
Ahh, and init_thread does have PF_KTHREAD.
Ok, looks fine to me, except I think the commit message should be cleared up.
The whole sentence about
"But the init task isn't supposed to be using the FPU in any case ..."
is just simply not true.
It should be more along the lines of "kernel threads don't need an FPU save area, because their FPU use is not preemptible or reentrant and they don't return to user space".
Linus
PATCH v4 3/3 x86/fpu: Remove init_task FPU state dependencies, add debugging warning for PF_KTHREAD tasks
Quoted 11 lines · Linus Torvalds
Quoted 3 lines · Oleg Nesterov
Yes, but
kernel_fpu_begin()never doessave_fpregs_to_fpstate()ifcurrent->flags&PF_KTHREAD?Ahh, and
init_threaddoes havePF_KTHREAD.Ok, looks fine to me, except I think the commit message should be cleared up.
The whole sentence about
Quoted 7 lines
"But the init task isn't supposed to be using the FPU in any case ..."
is just simply not true.
It should be more along the lines of "kernel threads don't need an FPU save area, because their FPU use is not preemptible or reentrant and they don't return to user space".
Indeed - I fixed & clarified the changelog accordingly - see the new patch further below.
I changed the debug check to test for PF_KTHREAD, and to return NULL:
+#ifdef CONFIG_X86_DEBUG_FPU
+struct fpu *x86_task_fpu(struct task_struct *task) +{
+ if (WARN_ON_ONCE(task->flags & PF_KTHREAD))
+ return NULL;
+
+ return (void *)task + sizeof(*task);
+}
+#endif
... and the NULL we return will likely crash & exit any kthreads attempting to use the FPU context area - which I think is preferable to returning invalid memory that may then be corrupted.
Hopefully this remains a hypothethical concern. :-)
Alternatively, this may be one of the very few cases where a BUG_ON() might be justified? This condition is not recoverable in any sane fashion IMO.
Thanks,
Ingo
============================>
From: Ingo Molnar <[email protected]>
Date: Wed, 5 Jun 2024 10:35:57 +0200
Subject: [PATCH] x86/fpu: Remove init_task FPU state dependencies, add debugging warning for PF_KTHREAD tasks
init_task's FPU state initialization was a bit of a hack:
__x86_init_fpu_begin = .; . = __x86_init_fpu_begin + 128*PAGE_SIZE; __x86_init_fpu_end = .;
But the init task isn't supposed to be using the FPU context in any case, so remove the hack and add in some debug warnings.
As Linus noted in the discussion, the init task (and other PF_KTHREAD tasks) *can* use the FPU via kernel_fpu_begin()/_end(), but they don't need the context area because their FPU use is not preemptible or reentrant, and they don't return to user-space.
--- arch/x86/include/asm/processor.h | 6 +++++- arch/x86/kernel/fpu/core.c | 13 ++++++++++--- arch/x86/kernel/fpu/init.c | 5 ++--- arch/x86/kernel/fpu/xstate.c | 3 --- arch/x86/kernel/vmlinux.lds.S | 4 ---- 5 files changed, 17 insertions(+), 14 deletions(-)
arch/x86/include/asm/processor.h +5 -1
@@ -504,7 +504,11 @@ struct thread_struct {#endif }; -#define x86_task_fpu (task) ((struct fpu *)((void *)task + sizeof(*task))) +#ifdef CONFIG_X86_DEBUG_FPU +extern struct fpu *x86_task_fpu (struct task_struct *task); +#else +# define x86_task_fpu (task) ((struct fpu *)((void *)task + sizeof(*task))) +#endif /* * X86 doesn't need any embedded-FPU-struct quirks:
arch/x86/kernel/fpu/core.c +10 -3
@@ -51,6 +51,16 @@ static DEFINE_PER_CPU(bool, in_kernel_fpu);*/ DEFINE_PER_CPU(struct fpu *, fpu_fpregs_owner_ctx); +#ifdef CONFIG_X86_DEBUG_FPU +struct fpu *x86_task_fpu (struct task_struct *task) +{ +if (WARN_ON_ONCE (task->flags &PF_KTHREAD )) +return NULL ; + +return (void *)task +sizeof (*task); +} +#endif +/* * Can we use the FPU in kernel mode with the * whole "kernel_fpu_begin/end()" sequence?
@@ -591,10 +601,8 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,* This is safe because task_struct size is a multiple of cacheline size. */ struct fpu *dst_fpu = (void *)dst +sizeof (*dst); -struct fpu *src_fpu =x86_task_fpu (current);BUILD_BUG_ON (sizeof (*dst) %SMP_CACHE_BYTES !=0 ); -BUG_ON (!src_fpu);/* The new task's FPU state cannot be valid in the hardware. */ dst_fpu->last_cpu =-1 ;
@@ -657,7 +665,6 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,if (update_fpu_shstk (dst, ssp))return 1 ; -trace_x86_fpu_copy_src (src_fpu);trace_x86_fpu_copy_dst (dst_fpu);return 0 ;
arch/x86/kernel/fpu/init.c +2 -3
@@ -38,7 +38,7 @@ static void fpu__init_cpu_generic(void)/* Flush out any pending x87 state: */ #ifdef CONFIG_MATH_EMULATION if (!boot_cpu_has (X86_FEATURE_FPU )) -fpstate_init_soft (&x86_task_fpu (current)->fpstate->regs.soft); + ; else#endif asmvolatile ("fninit" );
@@ -164,7 +164,7 @@ static void __init fpu__init_task_struct_size(void)* Subtract off the static size of the register state. * It potentially has a bunch of padding. */ - task_size -=sizeof (x86_task_fpu (current)->__fpstate.regs); + task_size -=sizeof (union fpregs_state );/* * Add back the dynamically-calculated register state
@@ -209,7 +209,6 @@ static void __init fpu__init_system_xstate_size_legacy(void) fpu_kernel_cfg.default_size = size; fpu_user_cfg.max_size = size; fpu_user_cfg.default_size = size; -fpstate_reset (x86_task_fpu (current)); }/*
arch/x86/kernel/fpu/xstate.c +0 -3
@@ -844,9 +844,6 @@ void __init fpu__init_system_xstate(unsigned int legacy_size)if (err) goto out_disable; -/* Reset the state for the current task */ -fpstate_reset (x86_task_fpu (current)); -/* * Update info used for ptrace frames; use standard-format size and no * supervisor xstates:
arch/x86/kernel/vmlinux.lds.S +0 -4
@@ -170,10 +170,6 @@ SECTIONS /* equivalent to task_pt_regs(&init_task) */ __top_init_kernel_stack = __end_init_stack - TOP_OF_KERNEL_STACK_PADDING - PTREGS_SIZE; - __x86_init_fpu_begin = .; - . = __x86_init_fpu_begin + 128*PAGE_SIZE; - __x86_init_fpu_end = .; - #ifdef CONFIG_X86_32 /* 32 bit has nosave before _edata */ NOSAVE_DATA
[PATCH 4/3] x86/fpu: Push 'fpu' pointer calculation into the fpu__drop() call
This encapsulates the fpu__drop() functionality better, and it will also enable other changes that want to check a task for PF_KTHREAD before calling x86_task_fpu().
--- arch/x86/include/asm/fpu/sched.h | 2 +- arch/x86/kernel/fpu/core.c | 4 +++- arch/x86/kernel/process.c | 3 +-- 3 files changed, 5 insertions(+), 4 deletions(-)
arch/x86/include/asm/fpu/sched.h +1 -1
@@ -10,7 +10,7 @@#include <asm/trace/fpu.h> extern void save_fpregs_to_fpstate (struct fpu *fpu); -extern void fpu__drop (struct fpu *fpu); +extern void fpu__drop (struct task_struct *tsk);extern int fpu_clone (struct task_struct *dst,unsigned long clone_flags,bool minimal,unsigned long shstk_addr);extern void fpu_flush_thread (void );
arch/x86/kernel/fpu/core.c +3 -1
@@ -679,8 +679,10 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,* a state-restore is coming: either an explicit one, * or a reschedule. */ -void fpu__drop (struct fpu *fpu) +void fpu__drop (struct task_struct *tsk) { +struct fpu *fpu =x86_task_fpu (tsk); +preempt_disable ();if (fpu ==x86_task_fpu (current)) {
arch/x86/kernel/process.c +1 -2
@@ -114,7 +114,6 @@ void arch_release_task_struct(struct task_struct *tsk)void exit_thread (struct task_struct *tsk) {struct thread_struct *t = &tsk->thread; -struct fpu *fpu =x86_task_fpu (tsk);if (test_thread_flag (TIF_IO_BITMAP ))io_bitmap_exit (tsk);
@@ -122,7 +121,7 @@ void exit_thread(struct task_struct *tsk)free_vm86 (t);shstk_free (tsk); -fpu__drop (fpu); +fpu__drop (tsk); }static int set_new_tls (struct task_struct *p,unsigned long tls)
[PATCH 5/3] x86/fpu: Make sure x86_task_fpu() doesn't get called for PF_KTHREAD tasks during exit
In reply to Ingo Molnarfpu__drop() calls x86_task_fpu() unconditionally, while the FPU context area will not be present if it's the init task, and should not be in use when it's some other type of kthread.
Return early for PF_KTHREAD tasks. The debug warning in x86_task_fpu() will catch any kthreads attempting to use the FPU save area.
--- arch/x86/kernel/fpu/core.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)
arch/x86/kernel/fpu/core.c +7 -1
@@ -681,7 +681,13 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,*/ void fpu__drop (struct task_struct *tsk) { -struct fpu *fpu =x86_task_fpu (tsk); +struct fpu *fpu; + +/* PF_KTHREAD tasks do not use the FPU context area: */ +if (tsk->flags &PF_KTHREAD ) +return ; + + fpu =x86_task_fpu (tsk);preempt_disable ();
Re: [PATCH 3/3, v4] x86/fpu: Remove init_task FPU state dependencies, add debugging warning for PF_KTHREAD tasks
In reply to Ingo MolnarQuoted 12 lines · Ingo Molnar
I changed the debug check to test for
PF_KTHREAD, and to return NULL:arch/x86/kernel/fpu/core.c +9
-0+#ifdef CONFIG_X86_DEBUG_FPU +struct fpu *x86_task_fpu (struct task_struct *task) +{ +if (WARN_ON_ONCE (task->flags &PF_KTHREAD )) +return NULL ; + +return (void *)task +sizeof (*task); +} +#endif
Quoted 8 lines
... and the NULL we return will likely crash & exit any kthreads attempting to use the FPU context area - which I think is preferable to returning invalid memory that may then be corrupted.
Hopefully this remains a hypothethical concern. :-)
Alternatively, this may be one of the very few cases where a
BUG_ON()might be justified? This condition is not recoverable in any sane fashion IMO.
And promptly this triggered in live testing, because while kthreads do not use the FPU context area, the current fpu__drop() code does call x86_task_cpu() even for kthreads...
See the two new 4/3 and 5/3 patches in this thread I've sent that clean this up:
[PATCH 4/3] x86/fpu: Push 'fpu' pointer calculation into the fpu__drop() call [PATCH 5/3] x86/fpu: Make sure x86_task_fpu() doesn't get called for PF_KTHREAD tasks during exit
I'll also reorder the patches to apply these fpu__drop() changes before adding the debug warning.
Thanks,
Ingo
Re: [PATCH 3/3, v4] x86/fpu: Remove init_task FPU state dependencies, add debugging warning for PF_KTHREAD tasks
In reply to Ingo Molnararch/x86/kernel/fpu/core.c static DEFINE_PER_CPU(bool, in_kernel_fpu); · line 62
Quoted 12 lines · Ingo Molnar
I changed the debug check to test for
PF_KTHREAD, and to return NULL:arch/x86/kernel/fpu/core.c +9
-0+#ifdef CONFIG_X86_DEBUG_FPU +struct fpu *x86_task_fpu (struct task_struct *task) +{ +if (WARN_ON_ONCE (task->flags &PF_KTHREAD )) +return NULL ; + +return (void *)task +sizeof (*task); +} +#endif
How many users enable CONFIG_X86_DEBUG_FPU? Perhaps it makes sense to check PF_KTHREAD unconditionally for the start, them add if (IS_ENABLED(X86_DEBUG_FPU)). But I won't insist.
For the record, I think we can later change this code to check
task->flags & (PF_KTHREAD | PF_USER_WORKER)
but I guess this needs some (simple) changes in the ptrace/coredump paths.
Oleg.
Re: [PATCH 3/3, v4] x86/fpu: Remove init_task FPU state dependencies, add debugging warning for PF_KTHREAD tasks
Quoted 4 lines · Oleg Nesterov
Quoted 3 lines · Ingo Molnar
I changed the debug check to test for
PF_KTHREAD, and to return NULL:
Quoted 11 lines
arch/x86/kernel/fpu/core.c static DEFINE_PER_CPU(bool, in_kernel_fpu); · line 62
Quoted 9 lines
arch/x86/kernel/fpu/core.c +9
-0+#ifdef CONFIG_X86_DEBUG_FPU +struct fpu *x86_task_fpu (struct task_struct *task) +{ +if (WARN_ON_ONCE (task->flags &PF_KTHREAD )) +return NULL ; + +return (void *)task +sizeof (*task); +} +#endif How many users enable
CONFIG_X86_DEBUG_FPU?
Ubuntu does:
kepler:~/tip> grep X86_DEBUG_FPU /boot/config-6.* /boot/config-6.8.0-35-generic:CONFIG_X86_DEBUG_FPU=y
Fedora doesn't:
kepler:~/s/tmp> grep X86_DEBUG_FPU ./lib/modules/6.9.0-64.fc41.x86_64/config # CONFIG_X86_DEBUG_FPU is not set
So it's a bit of a hit and miss, but at least one major distribution does, which is all that we need really.
Quoted 2 lines
[...] Perhaps it makes sense to check
PF_KTHREADunconditionally for the start, them add if (IS_ENABLED(X86_DEBUG_FPU)). But I won't insist.
I think Ubuntu ought to add enough debug coverage - and this check isn't exactly cheap, given how it replaces a simple build time structure offset with a full-blown function call ...
Quoted 6 lines
For the record, I think we can later change this code to check
task->flags & (PF_KTHREAD | PF_USER_WORKER)but I guess this needs some (simple) changes in the ptrace/coredump paths.
Sounds useful, would you be interested in cooking up a series on top of tip:master (or tip:WIP.x86/fpu), if you are interested and have the time?
I'll send out one last iteration of this series today, otherwise the changes seem close to final for an upstream merge via the tip:x86/fpu tree.
Thanks,
Ingo
Re: [PATCH 3/3] x86/fpu: Remove init_task FPU state dependencies, add debugging warning
In reply to Ingo MolnarQuoted 133 lines · Ingo Molnar
init_task's FPU state initialization was a bit of a hack:
__x86_init_fpu_begin = .; . = __x86_init_fpu_begin + 128*PAGE_SIZE; __x86_init_fpu_end = .;But the init task isn't supposed to be using the FPU in any case, so remove the hack and add in some debug warnings.
Signed-off-by Ingo Molnar <[email protected]>Cc Andy Lutomirski <[email protected]>Cc Borislav Petkov <[email protected]>Cc Fenghua Yu <[email protected]>Cc H. Peter Anvin <[email protected]>Cc Linus Torvalds <[email protected]>Cc Oleg Nesterov <[email protected]>Cc Dave Hansen <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Uros Bizjak <[email protected]>--- arch/x86/include/asm/processor.h | 6 +++++- arch/x86/kernel/fpu/core.c | 12 +++++++++--- arch/x86/kernel/fpu/init.c | 5 ++--- arch/x86/kernel/fpu/xstate.c | 3 --- arch/x86/kernel/vmlinux.lds.S | 4 ---- 5 files changed, 16 insertions(+), 14 deletions(-)arch/x86/include/asm/processor.h +5
-1@@ -504,7 +504,11 @@ struct thread_struct {#endif }; -#define x86_task_fpu (task) ((struct fpu *)((void *)task + sizeof(*task))) +#ifdef CONFIG_X86_DEBUG_FPU +extern struct fpu *x86_task_fpu (struct task_struct *task); +#else +# define x86_task_fpu (task) ((struct fpu *)((void *)task + sizeof(*task))) +#endif /* * X86 doesn't need any embedded-FPU-struct quirks: arch/x86/kernel/fpu/core.c +9
-3@@ -51,6 +51,15 @@ static DEFINE_PER_CPU(bool, in_kernel_fpu);*/ DEFINE_PER_CPU(struct fpu *, fpu_fpregs_owner_ctx); +#ifdef CONFIG_X86_DEBUG_FPU +struct fpu *x86_task_fpu (struct task_struct *task) +{ +WARN_ON_ONCE (task == &init_task); + +return (void *)task +sizeof (*task); +} +#endif +/* * Can we use the FPU in kernel mode with the * whole "kernel_fpu_begin/end()" sequence? @@ -591,10 +600,8 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,* This is safe because task_struct size is a multiple of cacheline size. */ struct fpu *dst_fpu = (void *)dst +sizeof (*dst); -struct fpu *src_fpu =x86_task_fpu (current);BUILD_BUG_ON (sizeof (*dst) %SMP_CACHE_BYTES !=0 ); -BUG_ON (!src_fpu);/* The new task's FPU state cannot be valid in the hardware. */ dst_fpu->last_cpu =-1 ;@@ -657,7 +664,6 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal,if (update_fpu_shstk (dst, ssp))return 1 ; -trace_x86_fpu_copy_src (src_fpu);trace_x86_fpu_copy_dst (dst_fpu);return 0 ;arch/x86/kernel/fpu/init.c +2
-3@@ -38,7 +38,7 @@ static void fpu__init_cpu_generic(void)/* Flush out any pending x87 state: */ #ifdef CONFIG_MATH_EMULATION if (!boot_cpu_has (X86_FEATURE_FPU )) -fpstate_init_soft (&x86_task_fpu (current)->fpstate->regs.soft); + ; else#endif asmvolatile ("fninit" );@@ -164,7 +164,7 @@ static void __init fpu__init_task_struct_size(void)* Subtract off the static size of the register state. * It potentially has a bunch of padding. */ - task_size -=sizeof (x86_task_fpu (current)->__fpstate.regs); + task_size -=sizeof (union fpregs_state );/* * Add back the dynamically-calculated register state @@ -209,7 +209,6 @@ static void __init fpu__init_system_xstate_size_legacy(void) fpu_kernel_cfg.default_size = size; fpu_user_cfg.max_size = size; fpu_user_cfg.default_size = size; -fpstate_reset (x86_task_fpu (current)); }/* arch/x86/kernel/fpu/xstate.c +0
-3@@ -844,9 +844,6 @@ void __init fpu__init_system_xstate(unsigned int legacy_size)if (err) goto out_disable; -/* Reset the state for the current task */ -fpstate_reset (x86_task_fpu (current)); -/* * Update info used for ptrace frames; use standard-format size and no * supervisor xstates: arch/x86/kernel/vmlinux.lds.S +0
-0
arch/x86/kernel/vmlinux.lds.S SECTIONS · line 175
Quoted 10 lines
arch/x86/kernel/vmlinux.lds.S +0
-4@@ -170,10 +170,6 @@ SECTIONS /* equivalent to task_pt_regs(&init_task) */ __top_init_kernel_stack = __end_init_stack - TOP_OF_KERNEL_STACK_PADDING - PTREGS_SIZE; - __x86_init_fpu_begin = .; - . = __x86_init_fpu_begin + 128*PAGE_SIZE; - __x86_init_fpu_end = .; - #ifdef CONFIG_X86_32 /* 32 bit has nosave before _edata */ NOSAVE_DATA
Hi,
we've hit x86/fpu related WARNING and NULL pointer issue during KVM/QEMU VM booting with latest linux-next kernel, bisect results show it's related to this commit, would you take a look?
detailed description in https://bugzilla.kernel.org/show_bug.cgi?id=218980
Re: [PATCH 3/3] x86/fpu: Remove init_task FPU state dependencies, add debugging warning
Quoted 144 lines · Hongyu
Quoted 137 lines · Ingo Molnar
init_task's FPU state initialization was a bit of a hack:
__x86_init_fpu_begin= .;
. =__x86_init_fpu_begin+ 128*PAGE_SIZE;
__x86_init_fpu_end= .;But the init task isn't supposed to be using the FPU in any case, so remove the hack and add in some debug warnings.
Signed-off-by Ingo Molnar <[email protected]>Cc Andy Lutomirski <[email protected]>Cc Borislav Petkov <[email protected]>Cc Fenghua Yu <[email protected]>Cc H. Peter Anvin <[email protected]>Cc Linus Torvalds <[email protected]>Cc Oleg Nesterov <[email protected]>Cc Dave Hansen <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Uros Bizjak <[email protected]>---
arch/x86/include/asm/processor.h| 6 +++++- arch/x86/kernel/fpu/core.c | 12 +++++++++--- arch/x86/kernel/fpu/init.c | 5 ++--- arch/x86/kernel/fpu/xstate.c | 3 --- arch/x86/kernel/vmlinux.lds.S | 4 ----5 files changed, 16 insertions(+), 14 deletions(-)arch/x86/include/asm/processor.h +5
-1@@ -504,7 +504,11 @@ struct thread_struct { #endif }; -#define x86_task_fpu (task) ((struct fpu *)((void *)task + sizeof(*task))) +#ifdef CONFIG_X86_DEBUG_FPU +extern struct fpu *x86_task_fpu (struct task_struct *task); +#else +# define x86_task_fpu (task) ((struct fpu *)((void *)task + sizeof(*task))) +#endif /** X86 doesn't need any embedded-FPU-struct quirks: arch/x86/kernel/fpu/core.c +9
-3@@ -51,6 +51,15 @@ static DEFINE_PER_CPU(bool, in_kernel_fpu);*/ DEFINE_PER_CPU(struct fpu *, fpu_fpregs_owner_ctx); +#ifdef CONFIG_X86_DEBUG_FPU +struct fpu *x86_task_fpu (struct task_struct *task) +{ + WARN_ON_ONCE(task == &init_task); + + return (void *)task + sizeof(*task); +} +#endif + /** Can we use the FPU in kernel mode with the * whole "kernel_fpu_begin/end()" sequence? @@ -591,10 +600,8 @@ int fpu_clone(struct task_struct *dst, unsignedlong clone_flags, bool minimal, * This is safe because task_struct size is a multiple of cacheline size. */ struct fpu *dst_fpu = (void *)dst + sizeof(*dst); - struct fpu *src_fpu = x86_task_fpu(current); BUILD_BUG_ON(sizeof(*dst) % SMP_CACHE_BYTES != 0); - BUG_ON(!src_fpu); /* The new task's FPU state cannot be valid in the hardware. */ dst_fpu->last_cpu = -1;@@ -657,7 +664,6 @@ int fpu_clone(struct task_struct *dst, unsigned long clone_flags, bool minimal, if (update_fpu_shstk(dst, ssp)) return 1; - trace_x86_fpu_copy_src(src_fpu); trace_x86_fpu_copy_dst(dst_fpu); return 0;arch/x86/kernel/fpu/init.c +1
-1@@ -38,7 +38,7 @@ static void fpu__init_cpu_generic(void) /* Flush out any pending x87 state: */ #ifdef CONFIG_MATH_EMULATION if (!boot_cpu_has(X86_FEATURE_FPU)) - fpstate_init_soft(&x86_task_fpu(current)->fpstate->regs.soft); + ;else
#endif
asm volatile ("fninit");arch/x86/kernel/fpu/init.c +1
-2@@ -164,7 +164,7 @@ static void __init fpu__init_task_struct_size(void)* Subtract off the static size of the register state. * It potentially has a bunch of padding. */ - task_size -= sizeof(x86_task_fpu(current)->__fpstate.regs); + task_size -= sizeof(union fpregs_state); /** Add back the dynamically-calculated register state @@ -209,7 +209,6 @@ static void __initfpu__init_system_xstate_size_legacy (void ) fpu_kernel_cfg.default_size = size; fpu_user_cfg.max_size = size; fpu_user_cfg.default_size = size; - fpstate_reset(x86_task_fpu(current)); } /*arch/x86/kernel/fpu/xstate.c +0
-3@@ -844,9 +844,6 @@ void __init fpu__init_system_xstate(unsigned int legacy_size) if (err) goto out_disable; - /* Reset the state for the current task */ - fpstate_reset(x86_task_fpu(current)); - /** Update info used for ptrace frames; use standard-format size and no * supervisor xstates: arch/x86/kernel/vmlinux.lds.S +0
-4@@ -170,10 +170,6 @@ SECTIONS /* equivalent to task_pt_regs(&init_task) */ __top_init_kernel_stack = __end_init_stack - TOP_OF_KERNEL_STACK_PADDING - PTREGS_SIZE; - __x86_init_fpu_begin = .; - . = __x86_init_fpu_begin + 128*PAGE_SIZE; - __x86_init_fpu_end = .; - #ifdef CONFIG_X86_32 /* 32 bit has nosave before _edata */Quoted 1 line
NOSAVE_DATAHi,
Quoted 4 lines
we've hit x86/fpu related WARNING and NULL pointer issue during KVM/QEMU VM booting with latest linux-next kernel, bisect results show it's related to this commit, would you take a look?
detailed description in https://bugzilla.kernel.org/show_bug.cgi?id=218980
add a quick update:
1. CONFIG_X86_DEBUG_FPU=y was set by auto regression framework
2. disable CONFIG_X86_DEBUG_FPU will bypass above WARNING and NULL pointer issue
it may not make sense for general kernel regression check to enable CONFIG_X86_DEBUG_FPU=y, will revise auto regression framework to keep CONFIG_X86_DEBUG_FPU disabled to bypass it.
in the meanwhile, please let me know if this issue is still valuable to look into.
Quoted 51 lines · Ingo Molnar
This series is one of the dependencies of the fast-headers work, which aims to reduce header complexity by removing <
asm/processor.h> from the <linux/sched.h> dependency chain, which headers are headers are fat enough already even if we do not combine them.To achieve that decoupling, one of the key steps is to not embedd any C types from <
asm/processor.h> intotask_struct.The only architecture that relies on that in a serious fashion is x86, via 'struct
thread::fpu', and the series below attempts to resolve it by using a calculated&task->thread.fpuvalue via thex86_task_fpu()helper.Code generation: without
CONFIG_X86_DEBUG_FPU=ywe get a small reduction:text data bss dec hex filename 26475783 10439082 1740804 38655669 24dd6b5 vmlinux.before 26475601 10435146 1740804 38651551 24dc69f vmlinux.afterWith the new debug code - which I think we'll remove soon-ish, there's an expected increase:
text data bss dec hex filename 26476198 10439286 1740804 38656288 24dd920 vmlinux.CONFIG_X86_DEBUG_FPU.before 26477000 10435458 1740804 38653262 24dcd4e vmlinux.CONFIG_X86_DEBUG_FPU.afterThe Git tree can be found at:
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.gitWIP.x86/fpuThanks,
Ingo
===============>
Ingo Molnar (3):
x86/fpu: Maketask_struct::threadconstant size
x86/fpu: Remove thethread::fpupointer
x86/fpu: Removeinit_taskFPU state dependencies, add debugging warningarch/x86/include/asm/fpu/sched.h | 2 +- arch/x86/include/asm/processor.h | 19 ++++++++++--------- arch/x86/kernel/fpu/context.h | 4 ++-- arch/x86/kernel/fpu/core.c | 57 +++++++++++++++++++++++++++++++-------------------------- arch/x86/kernel/fpu/init.c | 23 +++++++++++++---------- arch/x86/kernel/fpu/regset.c | 22 +++++++++++----------- arch/x86/kernel/fpu/signal.c | 18 +++++++++--------- arch/x86/kernel/fpu/xstate.c | 23 ++++++++++------------- arch/x86/kernel/fpu/xstate.h | 6 +++--- arch/x86/kernel/process.c | 6 ++----
Quoted 8 lines
arch/x86/kernel/signal.c | 6 +++--- arch/x86/kernel/traps.c | 2 +- arch/x86/math-emu/fpu_aux.c | 2 +- arch/x86/math-emu/fpu_entry.c | 4 ++-- arch/x86/math-emu/fpu_system.h | 2 +- arch/x86/mm/extable.c | 2 +- include/linux/sched.h | 13 +++---------- 17 files changed, 104 insertions(+), 107 deletions(-)
This series would be better if you added the x86_task_fpu() helper in an initial patch without any other changes. That would make the actual changes more visible with less code churn.
Brian Gerst
PATCH 1/1 x86/fpu: Introduce the x86_task_fpu() helper method
Quoted 5 lines · Brian Gerst
Quoted 1 line
17 files changed, 104 insertions(+), 107 deletions(-)This series would be better if you added the
x86_task_fpu()helper in an initial patch without any other changes. That would make the actual changes more visible with less code churn.
Makes sense - I've split out the patch below and adjusted the rest of the series. Is this what you had in mind?
Note that I also robustified the macro a bit:
-# define x86_task_fpu(task) ((struct fpu *)((void *)task + sizeof(*task))) +# define x86_task_fpu(task) ((struct fpu *)((void *)(task) + sizeof(*(task))))
Thanks,
Ingo
========================>
From: Ingo Molnar <[email protected]>
Date: Thu, 6 Jun 2024 11:01:14 +0200
Subject: [PATCH] x86/fpu: Introduce the x86_task_fpu() helper method
The per-task FPU context/save area is allocated right next to task_struct() - introduce the x86_task_fpu() helper that calculates this explicitly from the task pointer.
--- arch/x86/include/asm/processor.h | 2 ++ 1 file changed, 2 insertions(+)
arch/x86/include/asm/processor.h +2 -0
@@ -507,6 +507,8 @@ struct thread_struct {struct fpu *fpu; }; +#define x86_task_fpu (task) ((struct fpu *)((void *)(task) + sizeof(*(task)))) +/* * X86 doesn't need any embedded-FPU-struct quirks: */
Re: [PATCH] x86/fpu: Introduce the x86_task_fpu() helper method
Quoted 42 lines · Ingo Molnar
Quoted 5 lines · Brian Gerst
Quoted 1 line
17 files changed, 104 insertions(+), 107 deletions(-)This series would be better if you added the
x86_task_fpu()helper in an initial patch without any other changes. That would make the actual changes more visible with less code churn.Makes sense - I've split out the patch below and adjusted the rest of the series. Is this what you had in mind?
Note that I also robustified the macro a bit:
-# define
x86_task_fpu(task)((struct fpu *)((void *)task +sizeof(*task))) +# definex86_task_fpu(task)((struct fpu *)((void *)(task) + sizeof(*(task))))Thanks,
Ingo
========================>
From: Ingo Molnar <[email protected]>
Date: Thu, 6 Jun 2024 11:01:14 +0200
Subject: [PATCH] x86/fpu: Introduce thex86_task_fpu()helper methodThe per-task FPU context/save area is allocated right next to
task_struct()- introduce thex86_task_fpu()helper that calculates this explicitly from the task pointer.Signed-off-by Ingo Molnar <[email protected]>--- arch/x86/include/asm/processor.h | 2 ++ 1 file changed, 2 insertions(+)arch/x86/include/asm/processor.h +0
-0
arch/x86/include/asm/processor.h struct thread_struct { · line 514
Quoted 8 lines
arch/x86/include/asm/processor.h +2
-0@@ -507,6 +507,8 @@ struct thread_struct {struct fpu *fpu; }; +#define x86_task_fpu (task) ((struct fpu *)((void *)(task) + sizeof(*(task)))) +/* * X86 doesn't need any embedded-FPU-struct quirks: */
Since this should be the first patch in the series, It would be:
#define #define x86_task_fpu(task) (&(task)->thread.fpu)
along with converting the existing accesses to task->thread.fpu in one patch with no other functional changes. Then you could change how the fpu struct is allocated without touching every access site again.
Brian Gerst
Re: [PATCH] x86/fpu: Introduce the x86_task_fpu() helper method
Quoted 54 lines · Brian Gerst
Quoted 42 lines · Ingo Molnar
Quoted 5 lines · Brian Gerst
Quoted 1 line
17 files changed, 104 insertions(+), 107 deletions(-)This series would be better if you added the
x86_task_fpu()helper in an initial patch without any other changes. That would make the actual changes more visible with less code churn.Makes sense - I've split out the patch below and adjusted the rest of the series. Is this what you had in mind?
Note that I also robustified the macro a bit:
-# define
x86_task_fpu(task)((struct fpu *)((void *)task +sizeof(*task))) +# definex86_task_fpu(task)((struct fpu *)((void *)(task) + sizeof(*(task))))Thanks,
Ingo
========================>
From: Ingo Molnar <[email protected]>
Date: Thu, 6 Jun 2024 11:01:14 +0200
Subject: [PATCH] x86/fpu: Introduce thex86_task_fpu()helper methodThe per-task FPU context/save area is allocated right next to
task_struct()- introduce thex86_task_fpu()helper that calculates this explicitly from the task pointer.Signed-off-by Ingo Molnar <[email protected]>--- arch/x86/include/asm/processor.h | 2 ++ 1 file changed, 2 insertions(+)arch/x86/include/asm/processor.h +0
-0arch/x86/include/asm/processor.h struct thread_struct { · line 514
Quoted 8 lines
arch/x86/include/asm/processor.h +2
-0@@ -507,6 +507,8 @@ struct thread_struct {struct fpu *fpu; }; +#define x86_task_fpu (task) ((struct fpu *)((void *)(task) + sizeof(*(task)))) +/* * X86 doesn't need any embedded-FPU-struct quirks: */ Since this should be the first patch in the series, It would be:
Quoted 5 lines
#define #define
x86_task_fpu(task)(&(task)->thread.fpu)along with converting the existing accesses to
task->thread.fpuin one patch with no other functional changes. Then you could change how the fpu struct is allocated without touching every access site again.
Yeah, you are right of course - I've restructured the series accordingly, it's now indeed a lot easier to review internally now, but has the same end result. Will post it later today.
Thanks,
Ingo