x86, fpu: lazy allocation of FPU area - v5

Only allocate the FPU area when the application actually uses FPU, i.e., in the
first lazy FPU trap. This could save memory for non-fpu using apps.

for example: on my system after boot, there are around 300 processes, with
only 17 using FPU.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

6 files changed, 88 insertions, 22 deletions

diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c
index baf632b221d4..db6839b53195 100644
--- a/arch/x86/kernel/i387.c
+++ b/arch/x86/kernel/i387.c
@@ -8,7 +8,6 @@
 #include <linux/module.h>
 #include <linux/regset.h>
 #include <linux/sched.h>
-#include <linux/bootmem.h>
 
 #include <asm/sigcontext.h>
 #include <asm/processor.h>
@@ -63,7 +62,6 @@ void __init init_thread_xstate(void)
         else
                 xstate_size = sizeof(struct i387_fsave_struct);
 #endif
-        init_task.thread.xstate = alloc_bootmem(xstate_size);
 }
 
 #ifdef CONFIG_X86_64
@@ -93,12 +91,22 @@ void __cpuinit fpu_init(void)
  * value at reset if we support XMM instructions and then
  * remeber the current task has used the FPU.
  */
-void init_fpu(struct task_struct *tsk)
+int init_fpu(struct task_struct *tsk)
 {
         if (tsk_used_math(tsk)) {
                 if (tsk == current)
                         unlazy_fpu(tsk);
-                return;
+                return 0;
+        }
+
+        /*
+         * Memory allocation at the first usage of the FPU and other state.
+         */
+        if (!tsk->thread.xstate) {
+                tsk->thread.xstate = kmem_cache_alloc(task_xstate_cachep,
+                                                      GFP_KERNEL);
+                if (!tsk->thread.xstate)
+                        return -ENOMEM;
         }
 
         if (cpu_has_fxsr) {
@@ -120,6 +128,7 @@ void init_fpu(struct task_struct *tsk)
          * Only the device not available exception or ptrace can call init_fpu.
          */
         set_stopped_child_used_math(tsk);
+        return 0;
 }
 
 int fpregs_active(struct task_struct *target, const struct user_regset *regset)
@@ -136,10 +145,14 @@ int xfpregs_get(struct task_struct *target, const struct user_regset *regset,
                 unsigned int pos, unsigned int count,
                 void *kbuf, void __user *ubuf)
 {
+        int ret;
+
         if (!cpu_has_fxsr)
                 return -ENODEV;
 
-        init_fpu(target);
+        ret = init_fpu(target);
+        if (ret)
+                return ret;
 
         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
                                    &target->thread.xstate->fxsave, 0, -1);
@@ -154,7 +167,10 @@ int xfpregs_set(struct task_struct *target, const struct user_regset *regset,
         if (!cpu_has_fxsr)
                 return -ENODEV;
 
-        init_fpu(target);
+        ret = init_fpu(target);
+        if (ret)
+                return ret;
+
         set_stopped_child_used_math(target);
 
         ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
@@ -312,11 +328,14 @@ int fpregs_get(struct task_struct *target, const struct user_regset *regset,
                void *kbuf, void __user *ubuf)
 {
         struct user_i387_ia32_struct env;
+        int ret;
 
         if (!HAVE_HWFP)
                 return fpregs_soft_get(target, regset, pos, count, kbuf, ubuf);
 
-        init_fpu(target);
+        ret = init_fpu(target);
+        if (ret)
+                return ret;
 
         if (!cpu_has_fxsr) {
                 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
@@ -344,7 +363,10 @@ int fpregs_set(struct task_struct *target, const struct user_regset *regset,
         if (!HAVE_HWFP)
                 return fpregs_soft_set(target, regset, pos, count, kbuf, ubuf);
 
-        init_fpu(target);
+        ret = init_fpu(target);
+        if (ret)
+                return ret;
+
         set_stopped_child_used_math(target);
 
         if (!cpu_has_fxsr) {
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index ead24efbcba0..0e613e7e7b5e 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -5,24 +5,34 @@
 #include <linux/slab.h>
 #include <linux/sched.h>
 
-static struct kmem_cache *task_xstate_cachep;
+struct kmem_cache *task_xstate_cachep;
 
 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 {
         *dst = *src;
-        dst->thread.xstate = kmem_cache_alloc(task_xstate_cachep, GFP_KERNEL);
-        if (!dst->thread.xstate)
-                return -ENOMEM;
-        WARN_ON((unsigned long)dst->thread.xstate & 15);
-        memcpy(dst->thread.xstate, src->thread.xstate, xstate_size);
+        if (src->thread.xstate) {
+                dst->thread.xstate = kmem_cache_alloc(task_xstate_cachep,
+                                                      GFP_KERNEL);
+                if (!dst->thread.xstate)
+                        return -ENOMEM;
+                WARN_ON((unsigned long)dst->thread.xstate & 15);
+                memcpy(dst->thread.xstate, src->thread.xstate, xstate_size);
+        }
         return 0;
 }
 
-void free_thread_info(struct thread_info *ti)
+void free_thread_xstate(struct task_struct *tsk)
 {
-        kmem_cache_free(task_xstate_cachep, ti->task->thread.xstate);
-        ti->task->thread.xstate = NULL;
+        if (tsk->thread.xstate) {
+                kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
+                tsk->thread.xstate = NULL;
+        }
+}
+
 
+void free_thread_info(struct thread_info *ti)
+{
+        free_thread_xstate(ti->task);
         free_pages((unsigned long)(ti), get_order(THREAD_SIZE));
 }
 
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index 3890a5dd25f9..7adad088e373 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -521,6 +521,10 @@ start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
         regs->cs                = __USER_CS;
         regs->ip                = new_ip;
         regs->sp                = new_sp;
+        /*
+         * Free the old FP and other extended state
+         */
+        free_thread_xstate(current);
 }
 EXPORT_SYMBOL_GPL(start_thread);
 
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index b795e831afd6..891af1a1b48a 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -533,6 +533,10 @@ start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
         regs->ss                = __USER_DS;
         regs->flags             = 0x200;
         set_fs(USER_DS);
+        /*
+         * Free the old FP and other extended state
+         */
+        free_thread_xstate(current);
 }
 EXPORT_SYMBOL_GPL(start_thread);
 
diff --git a/arch/x86/kernel/traps_32.c b/arch/x86/kernel/traps_32.c
index 8d136a73ce8e..471e694d6713 100644
--- a/arch/x86/kernel/traps_32.c
+++ b/arch/x86/kernel/traps_32.c
@@ -1148,9 +1148,22 @@ asmlinkage void math_state_restore(void)
         struct thread_info *thread = current_thread_info();
         struct task_struct *tsk = thread->task;
 
+        if (!tsk_used_math(tsk)) {
+                local_irq_enable();
+                /*
+                 * does a slab alloc which can sleep
+                 */
+                if (init_fpu(tsk)) {
+                        /*
+                         * ran out of memory!
+                         */
+                        do_group_exit(SIGKILL);
+                        return;
+                }
+                local_irq_disable();
+        }
+
         clts();                         /* Allow maths ops (or we recurse) */
-        if (!tsk_used_math(tsk))
-                init_fpu(tsk);
         restore_fpu(tsk);
         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
         tsk->fpu_counter++;
diff --git a/arch/x86/kernel/traps_64.c b/arch/x86/kernel/traps_64.c
index dc0cb497eec3..adff76ea97c4 100644
--- a/arch/x86/kernel/traps_64.c
+++ b/arch/x86/kernel/traps_64.c
@@ -1124,10 +1124,23 @@ asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 asmlinkage void math_state_restore(void)
 {
         struct task_struct *me = current;
-        clts();                 /* Allow maths ops (or we recurse) */
 
-        if (!used_math())
-                init_fpu(me);
+        if (!used_math()) {
+                local_irq_enable();
+                /*
+                 * does a slab alloc which can sleep
+                 */
+                if (init_fpu(me)) {
+                        /*
+                         * ran out of memory!
+                         */
+                        do_group_exit(SIGKILL);
+                        return;
+                }
+                local_irq_disable();
+        }
+
+        clts();                 /* Allow maths ops (or we recurse) */
         restore_fpu_checking(&me->thread.xstate->fxsave);
         task_thread_info(me)->status |= TS_USEDFPU;
         me->fpu_counter++;