To improve kretprobe scalability
Subsystems (5)
RFC PATCH 1/1 To improve kretprobe scalability
Currently list of kretprobe instances are stored in kretprobe object (as used_instances,free_instances) and in kretprobe hash table. We have one global kretprobe lock to serialise the access to these lists. This causes only one kretprobe handler to execute at a time. Hence affects system performance, particularly on SMP systems and when return probe is set on lot of functions (like on all systemcalls).
Solution proposed here gives fine-grain locks that performs better on SMP system compared to present kretprobe implementation.
Solution:
1) Instead of having one global lock to protect kretprobe instances
present in kretprobe object and kretprobe hash table. We will have two locks, one lock for protecting kretprobe hash table and another lock for kretporbe object.
2) We hold lock present in kretprobe object while we modify kretprobe
instance in kretprobe object and we hold per-hash-list lock while modifying kretprobe instances present in that hash list. To prevent deadlock, we never grab a per-hash-list lock while holding a kretprobe lock.
3) We can remove used_instances from struct kretprobe, as we can track used
instances of kretprobe instances using kretprobe hash table.
Time duration for kernel compilation ("make -j 8") on a 8-way ppc64 system with return probes set on all systemcalls looks like this.
Patched-kernel Un-Patched kernel
============================================================
real 10m49.694s real 13m21.276s user 35m6.117s user 40m30.454s sys 2m55.480s sys 3m23.221s
===========================================================
Please let me know your comments on the patch attached here.
--- arch/arm/kernel/kprobes.c | 6 -- arch/ia64/kernel/kprobes.c | 6 -- arch/powerpc/kernel/kprobes.c | 6 -- arch/s390/kernel/kprobes.c | 6 -- arch/sparc64/kernel/kprobes.c | 11 +--- arch/x86/kernel/kprobes.c | 6 -- include/linux/kprobes.h | 7 +- kernel/kprobes.c | 108 +++++++++++++++++++++++++++--------------- 8 files changed, 89 insertions(+), 67 deletions(-)
arch/ia64/kernel/kprobes.c +2 -4
@@ -429,8 +429,7 @@ int __kprobes trampoline_probe_handler(s ((struct fnptr *)kretprobe_trampoline)->ip;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +get_kretprobe_hash_lock (current, &head, &flags);/* * It is possible to have multiple instances associated with a given
@@ -485,7 +484,7 @@ int __kprobes trampoline_probe_handler(skretprobe_assert (ri, orig_ret_address, trampoline_address);reset_current_kprobe (); -spin_unlock_irqrestore (&kretprobe_lock, flags); +put_kretprobe_hash_lock (current, &flags);preempt_enable_no_resched ();hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {
@@ -500,7 +499,6 @@ int __kprobes trampoline_probe_handler(sreturn 1 ; } -/* Called with kretprobe_lock held */ void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri,struct pt_regs *regs) {
arch/powerpc/kernel/kprobes.c +2 -4
@@ -127,7 +127,6 @@ static void __kprobes set_current_kprobe kcb->kprobe_saved_msr = regs->msr; } -/* Called with kretprobe_lock held */ void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri,struct pt_regs *regs) {
@@ -294,8 +293,7 @@ static int __kprobes trampoline_probe_haunsigned long trampoline_address =(unsigned long )&kretprobe_trampoline;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +get_kretprobe_hash_lock (current, &head, &flags);/* * It is possible to have multiple instances associated with a given
@@ -334,7 +332,7 @@ static int __kprobes trampoline_probe_ha regs->nip = orig_ret_address;reset_current_kprobe (); -spin_unlock_irqrestore (&kretprobe_lock, flags); +put_kretprobe_hash_lock (current, &flags);preempt_enable_no_resched ();hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {
arch/s390/kernel/kprobes.c +2 -4
@@ -272,7 +272,6 @@ static void __kprobes set_current_kprobe__ctl_store (kcb->kprobe_saved_ctl,9 ,11 ); } -/* Called with kretprobe_lock held */ void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri,struct pt_regs *regs) {
@@ -379,8 +378,7 @@ static int __kprobes trampoline_probe_haunsigned long trampoline_address = (unsigned long )&kretprobe_trampoline;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +get_kretprobe_hash_lock (current, &head, &flags);/* * It is possible to have multiple instances associated with a given
@@ -419,7 +417,7 @@ static int __kprobes trampoline_probe_ha regs->psw.addr = orig_ret_address |PSW_ADDR_AMODE ;reset_current_kprobe (); -spin_unlock_irqrestore (&kretprobe_lock, flags); +put_kretprobe_hash_lock (current, &flags);preempt_enable_no_resched ();hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {
arch/x86/kernel/kprobes.c +2 -4
@@ -431,7 +431,6 @@ static void __kprobes prepare_singlestep regs->ip = (unsigned long )p->ainsn.insn; } -/* Called with kretprobe_lock held */ void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri,struct pt_regs *regs) {
@@ -682,8 +681,7 @@ static __used __kprobes void *trampolineunsigned long trampoline_address = (unsigned long )&kretprobe_trampoline;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +get_kretprobe_hash_lock (current, &head, &flags);/* fixup registers */ #ifdef CONFIG_X86_64 regs->cs = __KERNEL_CS;
@@ -732,7 +730,7 @@ static __used __kprobes void *trampolinekretprobe_assert (ri, orig_ret_address, trampoline_address); -spin_unlock_irqrestore (&kretprobe_lock, flags); +put_kretprobe_hash_lock (current, &flags);hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {hlist_del (&ri->hlist);
include/linux/kprobes.h +4 -3
@@ -157,11 +157,10 @@ struct kretprobe {int nmissed;size_t data_size;struct hlist_head free_instances; -struct hlist_head used_instances; +spinlock_t lock; };struct kretprobe_instance { -struct hlist_node uflist;/* either on free list or used list */ struct hlist_node hlist;struct kretprobe *rp;kprobe_opcode_t *ret_addr;
@@ -201,7 +200,6 @@ static inline int init_test_probes(void) }#endif /* CONFIG_KPROBES_SANITY_TEST */ -extern spinlock_t kretprobe_lock;extern struct mutex kprobe_mutex;extern int arch_prepare_kprobe (struct kprobe *p);extern void arch_arm_kprobe (struct kprobe *p);
@@ -214,6 +212,9 @@ extern void kprobes_inc_nmissed_count(st/* Get the kprobe at this addr (if any) - called with preemption disabled */ struct kprobe *get_kprobe (void *addr); +void get_kretprobe_hash_lock (struct task_struct *tsk, +struct hlist_head **head,unsigned long *flags); +void put_kretprobe_hash_lock (struct task_struct *tsk,unsigned long *flags);struct hlist_head *kretprobe_inst_table_head (struct task_struct *tsk);/* kprobe_running() will just return the current_kprobe on this CPU */
kernel/kprobes.c +9 -9
@@ -62,6 +62,7 @@ addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))#endif +static int kprobes_initialized;static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE ];static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE ];
@@ -69,8 +70,8 @@ static struct hlist_head kretprobe_inst_static bool kprobe_enabled;DEFINE_MUTEX (kprobe_mutex);/* Protects kprobe_table */ -DEFINE_SPINLOCK (kretprobe_lock);/* Protects kretprobe_inst_table */ static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; +static spinlock_t kretprobe_table_locks[KPROBE_TABLE_SIZE ];/* * Normally, functions that we'd want to prohibit kprobes in, are marked
@@ -368,26 +369,41 @@ void __kprobes kprobes_inc_nmissed_countreturn ; } -/* Called with kretprobe_lock held */ void __kprobes recycle_rp_inst (struct kretprobe_instance *ri,struct hlist_head *head) { +struct kretprobe *rp = ri->rp; +/* remove rp inst off the rprobe_inst_table */ hlist_del (&ri->hlist); -if (ri->rp) { -/* remove rp inst off the used list */ -hlist_del (&ri->uflist); -/* put rp inst back onto the free list */ -INIT_HLIST_NODE (&ri->uflist); -hlist_add_head (&ri->uflist, &ri->rp->free_instances); +INIT_HLIST_NODE (&ri->hlist); +if (likely (rp)) { +spin_lock (&rp->lock); +hlist_add_head (&ri->hlist, &rp->free_instances); +spin_unlock (&rp->lock); }else /* Unregistering */ hlist_add_head (&ri->hlist, head); } -struct hlist_head __kprobes *kretprobe_inst_table_head(struct task_struct
*tsk)
+void get_kretprobe_hash_lock(struct task_struct *tsk,
+ struct hlist_head **head, unsigned long *flags)
+{
+ unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
+ spinlock_t *hlist_lock;
+
+ *head = &kretprobe_inst_table[hash];
+ hlist_lock = &kretprobe_table_locks[hash];
+ spin_lock_irqsave(hlist_lock, *flags);
+}
+
+void put_kretprobe_hash_lock(struct task_struct *tsk, unsigned long *flags)
{
- return &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)];
+ unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
+ spinlock_t *hlist_lock;
+
+ hlist_lock = &kretprobe_table_locks[hash];
+ spin_unlock_irqrestore(hlist_lock, *flags);
}
/*
@@ -401,17 +417,21 @@ void __kprobes kprobe_flush_task(struct
struct kretprobe_instance *ri;
struct hlist_head *head, empty_rp;
struct hlist_node *node, *tmp;
- unsigned long flags = 0;
+ unsigned long hash, flags = 0;
- INIT_HLIST_HEAD(&empty_rp);
- spin_lock_irqsave(&kretprobe_lock, flags);
- head = kretprobe_inst_table_head(tk);
+ if (unlikely(!kprobes_initialized))
+ /* Early boot. kretprobe_table_locks not yet initialized. */
+ return;
+
+ hash = hash_ptr(tk, KPROBE_HASH_BITS);
+ head = &kretprobe_inst_table[hash];
+ spin_lock_irqsave(&kretprobe_table_locks[hash], flags);
hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
if (ri->task == tk)
recycle_rp_inst(ri, &empty_rp);
}
- spin_unlock_irqrestore(&kretprobe_lock, flags);
-
+ spin_unlock_irqrestore(&kretprobe_table_locks[hash], flags);
+ INIT_HLIST_HEAD(&empty_rp);
hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
hlist_del(&ri->hlist);
kfree(ri);
@@ -423,24 +443,29 @@ static inline void free_rp_inst(struct k
struct kretprobe_instance *ri;
struct hlist_node *pos, *next;
- hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, uflist) {
- hlist_del(&ri->uflist);
+ hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
+ hlist_del(&ri->hlist);
kfree(ri);
}
}
static void __kprobes cleanup_rp_inst(struct kretprobe *rp) {
- unsigned long flags;
+ unsigned long flags, hash;
struct kretprobe_instance *ri;
struct hlist_node *pos, *next;
+ struct hlist_head *head;
+
/* No race here */
- spin_lock_irqsave(&kretprobe_lock, flags);
- hlist_for_each_entry_safe(ri, pos, next, &rp->used_instances, uflist) {
- ri->rp = NULL;
- hlist_del(&ri->uflist);
+ for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
+ spin_lock_irqsave(&kretprobe_table_locks[hash], flags);
+ head = &kretprobe_inst_table[hash];
+ hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
+ if (ri->rp == rp)
+ ri->rp = NULL;
+ }
+ spin_unlock_irqrestore(&kretprobe_table_locks[hash], flags);
}
- spin_unlock_irqrestore(&kretprobe_lock, flags);
free_rp_inst(rp);
}
@@ -829,32 +854,37 @@ static int __kprobes pre_handler_kretpro
struct pt_regs *regs)
{
struct kretprobe *rp = container_of(p, struct kretprobe, kp);
- unsigned long flags = 0;
+ unsigned long hash, flags = 0;
+ struct kretprobe_instance *ri;
/*TODO: consider to only swap the RA after the last pre_handler fired */
- spin_lock_irqsave(&kretprobe_lock, flags);
+ hash = hash_ptr(current, KPROBE_HASH_BITS);
+ spin_lock_irqsave(&rp->lock, flags);
if (!hlist_empty(&rp->free_instances)) {
- struct kretprobe_instance *ri;
-
ri = hlist_entry(rp->free_instances.first,
- struct kretprobe_instance, uflist);
+ struct kretprobe_instance, hlist);
+ hlist_del(&ri->hlist);
+ spin_unlock_irqrestore(&rp->lock, flags);
+
ri->rp = rp;
ri->task = current;
if (rp->entry_handler && rp->entry_handler(ri, regs)) {
- spin_unlock_irqrestore(&kretprobe_lock, flags);
+ spin_unlock_irqrestore(&rp->lock, flags);
return 0;
}
arch_prepare_kretprobe(ri, regs);
/* XXX(hch): why is there no hlist_move_head? */
- hlist_del(&ri->uflist);
- hlist_add_head(&ri->uflist, &ri->rp->used_instances);
- hlist_add_head(&ri->hlist, kretprobe_inst_table_head(ri->task));
- } else
+ INIT_HLIST_NODE(&ri->hlist);
+ spin_lock_irqsave(&kretprobe_table_locks[hash], flags);
+ hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
+ spin_unlock_irqrestore(&kretprobe_table_locks[hash], flags);
+ } else {
rp->nmissed++;
- spin_unlock_irqrestore(&kretprobe_lock, flags);
+ spin_unlock_irqrestore(&rp->lock, flags);
+ }
return 0;
}
@@ -890,7 +920,7 @@ static int __kprobes __register_kretprob
rp->maxactive = NR_CPUS;
#endif
}
- INIT_HLIST_HEAD(&rp->used_instances);
+ spin_lock_init(&rp->lock);
INIT_HLIST_HEAD(&rp->free_instances);
for (i = 0; i < rp->maxactive; i++) {
inst = kmalloc(sizeof(struct kretprobe_instance) +
@@ -899,8 +929,8 @@ static int __kprobes __register_kretprob
free_rp_inst(rp);
return -ENOMEM;
}
- INIT_HLIST_NODE(&inst->uflist);
- hlist_add_head(&inst->uflist, &rp->free_instances);
+ INIT_HLIST_NODE(&inst->hlist);
+ hlist_add_head(&inst->hlist, &rp->free_instances);
}
rp->nmissed = 0;
@@ -1006,6 +1036,7 @@ static int __init init_kprobes(void)
for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
INIT_HLIST_HEAD(&kprobe_table[i]); INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
+ spin_lock_init(&kretprobe_table_locks[i]);
}
/*
@@ -1047,6 +1078,7 @@ static int __init init_kprobes(void)
err = arch_init_kprobes();
if (!err)
err = register_die_notifier(&kprobe_exceptions_nb);
+ kprobes_initialized = (err == 0);
if (!err)
init_test_probes();
arch/sparc64/kernel/kprobes.c +5 -6
@@ -478,9 +478,9 @@ int __kprobes longjmp_break_handler(strureturn 0 ; } -/* Called with kretprobe_lock held. The value stored in the return -* address register is actually 2 instructions before where the -* callee will return to. Sequences usually look something like this +/* The value stored in the return address register is actually 2 +* instructions before where the callee will return to. +* Sequences usually look something like this * * call some_function <--- return register points here * nop <--- call delay slot
@@ -512,8 +512,7 @@ int __kprobes trampoline_probe_handler(sunsigned long trampoline_address =(unsigned long )&kretprobe_trampoline;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +get_kretprobe_hash_lock (current, &head, &flags);/* * It is possible to have multiple instances associated with a given
@@ -553,7 +552,7 @@ int __kprobes trampoline_probe_handler(s regs->tnpc = orig_ret_address +4 ;reset_current_kprobe (); -spin_unlock_irqrestore (&kretprobe_lock, flags); +put_kretprobe_hash_lock (current, &flags);preempt_enable_no_resched ();hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {
arch/arm/kernel/kprobes.c +2 -4
@@ -296,8 +296,7 @@ static __used __kprobes void *trampolineunsigned long trampoline_address = (unsigned long )&kretprobe_trampoline;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +get_kretprobe_hash_lock (current, &head, &flags);/* * It is possible to have multiple instances associated with a given
@@ -337,7 +336,7 @@ static __used __kprobes void *trampoline }kretprobe_assert (ri, orig_ret_address, trampoline_address); -spin_unlock_irqrestore (&kretprobe_lock, flags); +put_kretprobe_hash_lock (current, &flags);hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {hlist_del (&ri->hlist);
@@ -347,7 +346,6 @@ static __used __kprobes void *trampolinereturn (void *)orig_ret_address; } -/* Called with kretprobe_lock held. */ void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri,struct pt_regs *regs) {
Quoted 17 lines · On Wed, 21 May 2008 06:32:17 +0530 Srinivasa D S <[email protected]> wrote:
Currently list of kretprobe instances are stored in kretprobe object (as used_instances,free_instances) and in kretprobe hash table. We have one global kretprobe lock to serialise the access to these lists. This causes only one kretprobe handler to execute at a time. Hence affects system performance, particularly on SMP systems and when return probe is set on lot of functions (like on all systemcalls).
Solution proposed here gives fine-grain locks that performs better on SMP system compared to present kretprobe implementation.
Solution:
1) Instead of having one global lock to protect kretprobe instances
present in kretprobe object and kretprobe hash table. We will have two locks, one lock for protecting kretprobe hash table and another lock for kretporbe object.
Quoted 7 lines
2) We hold lock present in kretprobe object while we modify kretprobe
instance in kretprobe object and we hold per-hash-list lock while modifying kretprobe instances present in that hash list. To prevent deadlock, we never grab a per-hash-list lock while holding a kretprobe lock.3) We can remove
used_instancesfrom struct kretprobe, as we can track used
instances of kretprobe instances using kretprobe hash table.
Are you sure that all the code which was previously globally serialised is safe to run concurrently?
Quoted 9 lines
Time duration for kernel compilation ("make -j 8") on a 8-way ppc64 system with return probes set on all systemcalls looks like this.
Patched-kernel Un-Patched kernel
============================================================real 10m49.694s real 13m21.276s user 35m6.117s user 40m30.454s sys 2m55.480s sys 3m23.221s===========================================================
That's a large speedup.
Do you have available the figures for the same workload when no probes are set at all?
Quoted 11 lines
kernel/kprobes.c +1
-0@@ -62,6 +62,7 @@ addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))#endif +static int kprobes_initialized;static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE ];static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE ];
kernel/kprobes.c static struct hlist_head kretprobe_inst_ · line 74
Quoted 6 lines
kernel/kprobes.c +1
-1@@ -69,8 +70,8 @@ static struct hlist_head kretprobe_inst_static bool kprobe_enabled;DEFINE_MUTEX (kprobe_mutex);/* Protects kprobe_table */ -DEFINE_SPINLOCK (kretprobe_lock);/* Protects kretprobe_inst_table */ static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; +static spinlock_t kretprobe_table_locks[KPROBE_TABLE_SIZE ];
An array of 64 spinlocks. These will all be crammed into two or four cachelines. If there is really any concurrency in here, the contention will be large.
I suggest that you experiment with one-lock-per-cacheline here and see whether it is worth doing that permanently.
Something along the lines of
static struct {
spinlock_t lock __cacheline_aligned;
} kretprobe_locks[KPROBE_TABLE_SIZE];static spinlock_t *kretprobe_lock_ptr(...)
{
return kretprobe_locks + hash(...);
}Also, none of this new code is needed on uniprocessor builds. Did we just add some bloat?
Quoted 12 lines
kernel/kprobes.c +10
-1-struct hlist_head __kprobes *kretprobe_inst_table_head (struct task_struct *tsk) +void get_kretprobe_hash_lock (struct task_struct *tsk, +struct hlist_head **head,unsigned long *flags) +{ +unsigned long hash =hash_ptr (tsk,KPROBE_HASH_BITS ); +spinlock_t *hlist_lock; + + *head = &kretprobe_inst_table[hash]; + hlist_lock = &kretprobe_table_locks[hash]; +spin_lock_irqsave (hlist_lock, *flags); +}
kernel/kprobes.c static struct hlist_head kretprobe_inst_
Quoted 10 lines
kernel/kprobes.c +7
-1+ +void put_kretprobe_hash_lock (struct task_struct *tsk,unsigned long *flags) { -return &kretprobe_inst_table[hash_ptr (tsk,KPROBE_HASH_BITS )]; +unsigned long hash =hash_ptr (tsk,KPROBE_HASH_BITS ); +spinlock_t *hlist_lock; + + hlist_lock = &kretprobe_table_locks[hash]; +spin_unlock_irqrestore (hlist_lock, *flags); }
I think these functions are misnamed. Generally functions whose names are of the form get_foo() and put_foo() are used for acquiring and releasing reference counts. These functions don't do that.
kretprobe_hash_lock() and kretprobe_hash_unlock() would be better?
Quoted 8 lines
/*
kernel/kprobes.c +1
-1@@ -401,17 +417,21 @@ void __kprobes kprobe_flush_task(structstruct kretprobe_instance *ri;struct hlist_head *head, empty_rp;struct hlist_node *node, *tmp; -unsigned long flags =0 ; +unsigned long hash, flags =0 ;
kernel/kprobes.c void __kprobes kprobe_flush_task(struct · line 428
Quoted 10 lines
kernel/kprobes.c +7
-3-INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (tk); +if (unlikely (!kprobes_initialized)) +/* Early boot. kretprobe_table_locks not yet initialized. */ +return ; + + hash =hash_ptr (tk,KPROBE_HASH_BITS ); + head = &kretprobe_inst_table[hash]; +spin_lock_irqsave (&kretprobe_table_locks[hash], flags);
that's an open-coded copy of get_kretprobe_hash_lock()?
kernel/kprobes.c void __kprobes kprobe_flush_task(struct · line 433
Quoted 7 lines
kernel/kprobes.c +1
-2hlist_for_each_entry_safe (ri, node, tmp, head, hlist) {if (ri->task == tk)recycle_rp_inst (ri, &empty_rp); } -spin_unlock_irqrestore (&kretprobe_lock, flags); - +spin_unlock_irqrestore (&kretprobe_table_locks[hash], flags);
and of put_kretprobe_hash_lock(). Perhaps a separate function which just obtains the spinlock_t* is needed.
Quoted 16 lines
kernel/kprobes.c +3
-2+INIT_HLIST_HEAD (&empty_rp);hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {hlist_del (&ri->hlist);kfree (ri);@@ -423,24 +443,29 @@ static inline void free_rp_inst(struct kstruct kretprobe_instance *ri;struct hlist_node *pos, *next; -hlist_for_each_entry_safe (ri, pos, next, &rp->free_instances, uflist) { -hlist_del (&ri->uflist); +hlist_for_each_entry_safe (ri, pos, next, &rp->free_instances, hlist) { +hlist_del (&ri->hlist);kfree (ri); } }
Quoted 8 lines · Andrew Morton
On Wed, 21 May 2008 06:32:17 +0530
Quoted 2 lines · Srinivasa D S
Currently list of kretprobe instances are stored in kretprobe object (as used_instances,free_instances) and in kretprobe hash table.
Are you sure that all the code which was previously globally serialised is safe to run concurrently?
Yes, Iam sure. I have tested this patch on all architecture. I have written the systemtap script which probes all systemcalls and its return values. This works fine on patched kernel.
Quoted 11 lines
Quoted 10 lines
Time duration for kernel compilation ("make -j 8") on a 8-way ppc64 system with return probes set on all systemcalls looks like this.
Patched-kernel Un-Patched kernel
============================================================real 10m49.694s real 13m21.276s user 35m6.117s user 40m30.454s sys 2m55.480s sys 3m23.221s===========================================================
Quoted 4 lines
That's a large speedup.
Do you have available the figures for the same workload when no probes are set at all?
As I was concerned with the performance of kretprobes, I don't have figures on the workload when probes are set. But, currently I have implemented your cacheline aligned approach and captured this data. Please see below(in the patch).
Quoted 18 lines
kernel/kprobes.c
Quoted 8 lines
kernel/kprobes.c +1
-1DEFINE_MUTEX (kprobe_mutex);/* Protects kprobe_table */ -DEFINE_SPINLOCK (kretprobe_lock);/* Protects kretprobe_inst_table */ static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; +static spinlock_t kretprobe_table_locks[KPROBE_TABLE_SIZE ];An array of 64 spinlocks. These will all be crammed into two or four cachelines. If there is really any concurrency in here, the contention will be large.
I suggest that you experiment with one-lock-per-cacheline here and see whether it is worth doing that permanently.
Something along the lines of
Quoted 10 lines
static struct { spinlock_t lock __cacheline_aligned; } kretprobe_locks[KPROBE_TABLE_SIZE];static
spinlock_t*kretprobe_lock_ptr(...){ return kretprobe_locks + hash(...); }
Thanks for the suggestion. I have implemented this approach and by looking at the statistics(attached in the patch), there is slight improvement in kretprobe performance.
Quoted 3 lines
Also, none of this new code is needed on uniprocessor builds. Did we just add some bloat?
No Andrew, We just replaced old kretprobe_lock code with the new one. Since kretprobe_lock was also used in arch dependent code, we have modified arch dependent kprobes.c file.
But code added to kernel.[ch] is very less, 51 lines of code gets added to kprobes.c(with the new patch, attached below) and this has caused speedup in performance.
Quoted 14 lines
kernel/kprobes.c
Quoted 9 lines
+void
put_kretprobe_hash_lock(struct task_struct *tsk, unsigned long *flags){kernel/kprobes.c +5
-1-return &kretprobe_inst_table[hash_ptr (tsk,KPROBE_HASH_BITS )]; +unsigned long hash =hash_ptr (tsk,KPROBE_HASH_BITS ); +spinlock_t *hlist_lock; + + hlist_lock = &kretprobe_table_locks[hash]; +spin_unlock_irqrestore (hlist_lock, *flags); }I think these functions are misnamed. Generally functions whose names are of the form
get_foo()andput_foo()are used for acquiring and releasing reference counts. These functions don't do that.
Quoted 1 line
kretprobe_hash_lock()andkretprobe_hash_unlock()would be better?
Sounds good to me, Done.
Quoted 13 lines
kernel/kprobes.c void __kprobes kprobe_flush_task(struct · line 421
Quoted 10 lines
/*
kernel/kprobes.c +1
-2@@ -401,17 +417,21 @@ void __kprobes kprobe_flush_task(structstruct kretprobe_instance *ri;struct hlist_head *head, empty_rp;struct hlist_node *node, *tmp; } -spin_unlock_irqrestore (&kretprobe_lock, flags); - +spin_unlock_irqrestore (&kretprobe_table_locks[hash], flags);and of
put_kretprobe_hash_lock(). Perhaps a separate function which just obtains the spinlock_t* is needed.
Added to the code.
Quoted 4 lines
kernel/kprobes.c void __kprobes kprobe_flush_task(struct · line 424
Quoted 4 lines
kernel/kprobes.c +1
-0+INIT_HLIST_HEAD (&empty_rp);hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {hlist_del (&ri->hlist);
Resending the patch, Again
Currently list of kretprobe instances are stored in kretprobe object (as used_instances,free_instances) and in kretprobe hash table. We have one global kretprobe lock to serialise the access to these lists. This causes only one kretprobe handler to execute at a time. Hence affects system performance, particularly on SMP systems and when return probe is set on lot of functions (like on all systemcalls).
Solution proposed here gives fine-grain locks that performs better on SMP system compared to present kretprobe implementation.
Solution:
1) Instead of having one global lock to protect kretprobe instances
present
in kretprobe object and kretprobe hash table. We will have two locks, one lock for protecting kretprobe hash table and another lock for kretporbe object.
2) We hold lock present in kretprobe object while we modify kretprobe
instance in
kretprobe object and we hold per-hash-list lock while modifying kretprobe instances
present in that hash list. To prevent deadlock, we never grab a per-hash-list lock while holding a kretprobe lock.
3) We can remove used_instances from struct kretprobe, as we can track used
instances of kretprobe instances using kretprobe hash table.
Time duration for kernel compilation ("make -j 8") on a 8-way ppc64 system with return probes set on all systemcalls looks like this.
cacheline non-cacheline Un-patched kernel
aligned patch aligned patch
===============================================================================
real 9m46.784s 9m54.412s 10m2.450s user 40m5.715s 40m7.142s 40m4.273s sys 2m57.754s 2m58.583s 3m17.430s
===========================================================
Time duration for kernel compilation ("make -j 8) on the same system, when kernel is not probed.
=========================
real 9m26.389s user 40m8.775s sys 2m7.283s
=========================
Please let me know your comments on the patch attached here.
--- arch/arm/kernel/kprobes.c | 6 - arch/ia64/kernel/kprobes.c | 6 - arch/powerpc/kernel/kprobes.c | 6 - arch/s390/kernel/kprobes.c | 6 - arch/sparc64/kernel/kprobes.c | 11 +-- arch/x86/kernel/kprobes.c | 6 - include/linux/kprobes.h | 7 +- kernel/kprobes.c | 127 +++++++++++++++++++++++++++++------------- 8 files changed, 108 insertions(+), 67 deletions(-)
arch/ia64/kernel/kprobes.c +2 -4
@@ -429,8 +429,7 @@ int __kprobes trampoline_probe_handler(s ((struct fnptr *)kretprobe_trampoline)->ip;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +kretprobe_hash_lock (current, &head, &flags);/* * It is possible to have multiple instances associated with a given
@@ -485,7 +484,7 @@ int __kprobes trampoline_probe_handler(skretprobe_assert (ri, orig_ret_address, trampoline_address);reset_current_kprobe (); -spin_unlock_irqrestore (&kretprobe_lock, flags); +kretprobe_hash_unlock (current, &flags);preempt_enable_no_resched ();hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {
@@ -500,7 +499,6 @@ int __kprobes trampoline_probe_handler(sreturn 1 ; } -/* Called with kretprobe_lock held */ void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri,struct pt_regs *regs) {
arch/powerpc/kernel/kprobes.c +2 -4
@@ -127,7 +127,6 @@ static void __kprobes set_current_kprobe kcb->kprobe_saved_msr = regs->msr; } -/* Called with kretprobe_lock held */ void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri,struct pt_regs *regs) {
@@ -294,8 +293,7 @@ static int __kprobes trampoline_probe_haunsigned long trampoline_address =(unsigned long )&kretprobe_trampoline;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +kretprobe_hash_lock (current, &head, &flags);/* * It is possible to have multiple instances associated with a given
@@ -334,7 +332,7 @@ static int __kprobes trampoline_probe_ha regs->nip = orig_ret_address;reset_current_kprobe (); -spin_unlock_irqrestore (&kretprobe_lock, flags); +kretprobe_hash_unlock (current, &flags);preempt_enable_no_resched ();hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {
arch/s390/kernel/kprobes.c +2 -4
@@ -272,7 +272,6 @@ static void __kprobes set_current_kprobe__ctl_store (kcb->kprobe_saved_ctl,9 ,11 ); } -/* Called with kretprobe_lock held */ void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri,struct pt_regs *regs) {
@@ -379,8 +378,7 @@ static int __kprobes trampoline_probe_haunsigned long trampoline_address = (unsigned long )&kretprobe_trampoline;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +kretprobe_hash_lock (current, &head, &flags);/* * It is possible to have multiple instances associated with a given
@@ -419,7 +417,7 @@ static int __kprobes trampoline_probe_ha regs->psw.addr = orig_ret_address |PSW_ADDR_AMODE ;reset_current_kprobe (); -spin_unlock_irqrestore (&kretprobe_lock, flags); +kretprobe_hash_unlock (current, &flags);preempt_enable_no_resched ();hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {
arch/x86/kernel/kprobes.c +2 -4
@@ -431,7 +431,6 @@ static void __kprobes prepare_singlestep regs->ip = (unsigned long )p->ainsn.insn; } -/* Called with kretprobe_lock held */ void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri,struct pt_regs *regs) {
@@ -682,8 +681,7 @@ static __used __kprobes void *trampolineunsigned long trampoline_address = (unsigned long )&kretprobe_trampoline;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +kretprobe_hash_lock (current, &head, &flags);/* fixup registers */ #ifdef CONFIG_X86_64 regs->cs = __KERNEL_CS;
@@ -732,7 +730,7 @@ static __used __kprobes void *trampolinekretprobe_assert (ri, orig_ret_address, trampoline_address); -spin_unlock_irqrestore (&kretprobe_lock, flags); +kretprobe_hash_unlock (current, &flags);hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {hlist_del (&ri->hlist);
include/linux/kprobes.h +4 -3
@@ -157,11 +157,10 @@ struct kretprobe {int nmissed;size_t data_size;struct hlist_head free_instances; -struct hlist_head used_instances; +spinlock_t lock; };struct kretprobe_instance { -struct hlist_node uflist;/* either on free list or used list */ struct hlist_node hlist;struct kretprobe *rp;kprobe_opcode_t *ret_addr;
@@ -201,7 +200,6 @@ static inline int init_test_probes(void) }#endif /* CONFIG_KPROBES_SANITY_TEST */ -extern spinlock_t kretprobe_lock;extern struct mutex kprobe_mutex;extern int arch_prepare_kprobe (struct kprobe *p);extern void arch_arm_kprobe (struct kprobe *p);
@@ -214,6 +212,9 @@ extern void kprobes_inc_nmissed_count(st/* Get the kprobe at this addr (if any) - called with preemption disabled */ struct kprobe *get_kprobe (void *addr); +void kretprobe_hash_lock (struct task_struct *tsk, +struct hlist_head **head,unsigned long *flags); +void kretprobe_hash_unlock (struct task_struct *tsk,unsigned long *flags);struct hlist_head *kretprobe_inst_table_head (struct task_struct *tsk);/* kprobe_running() will just return the current_kprobe on this CPU */
kernel/kprobes.c +16 -9
@@ -62,6 +62,7 @@ addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))#endif +static int kprobes_initialized;static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE ];static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE ];
@@ -69,8 +70,15 @@ static struct hlist_head kretprobe_inst_static bool kprobe_enabled;DEFINE_MUTEX (kprobe_mutex);/* Protects kprobe_table */ -DEFINE_SPINLOCK (kretprobe_lock);/* Protects kretprobe_inst_table */ static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; +static struct { +spinlock_t lock____cacheline_aligned ; +} kretprobe_table_locks[KPROBE_TABLE_SIZE ]; + +static spinlock_t *kretprobe_table_lock_ptr (unsigned long hash) +{ +return &(kretprobe_table_locks[hash].lock); +}/* * Normally, functions that we'd want to prohibit kprobes in, are marked
@@ -368,26 +376,53 @@ void __kprobes kprobes_inc_nmissed_countreturn ; } -/* Called with kretprobe_lock held */ void __kprobes recycle_rp_inst (struct kretprobe_instance *ri,struct hlist_head *head) { +struct kretprobe *rp = ri->rp; +/* remove rp inst off the rprobe_inst_table */ hlist_del (&ri->hlist); -if (ri->rp) { -/* remove rp inst off the used list */ -hlist_del (&ri->uflist); -/* put rp inst back onto the free list */ -INIT_HLIST_NODE (&ri->uflist); -hlist_add_head (&ri->uflist, &ri->rp->free_instances); +INIT_HLIST_NODE (&ri->hlist); +if (likely (rp)) { +spin_lock (&rp->lock); +hlist_add_head (&ri->hlist, &rp->free_instances); +spin_unlock (&rp->lock); }else /* Unregistering */ hlist_add_head (&ri->hlist, head); } -struct hlist_head __kprobes *kretprobe_inst_table_head(struct task_struct
*tsk)
+void kretprobe_hash_lock(struct task_struct *tsk,
+ struct hlist_head **head, unsigned long *flags)
+{
+ unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
+ spinlock_t *hlist_lock;
+
+ *head = &kretprobe_inst_table[hash];
+ hlist_lock = kretprobe_table_lock_ptr(hash);
+ spin_lock_irqsave(hlist_lock, *flags);
+}
+
+void kretprobe_table_lock(unsigned long hash, unsigned long *flags)
{
- return &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)];
+ spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
+ spin_lock_irqsave(hlist_lock, *flags);
+}
+
+void kretprobe_hash_unlock(struct task_struct *tsk, unsigned long *flags) +{
+ unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
+ spinlock_t *hlist_lock;
+
+ hlist_lock = kretprobe_table_lock_ptr(hash);
+ spin_unlock_irqrestore(hlist_lock, *flags);
+}
+
+void kretprobe_table_unlock(unsigned long hash, unsigned long *flags) +{
+ spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
+ spin_unlock_irqrestore(hlist_lock, *flags);
}
/*
@@ -401,17 +436,21 @@ void __kprobes kprobe_flush_task(struct
struct kretprobe_instance *ri;
struct hlist_head *head, empty_rp;
struct hlist_node *node, *tmp;
- unsigned long flags = 0;
+ unsigned long hash, flags = 0;
- INIT_HLIST_HEAD(&empty_rp);
- spin_lock_irqsave(&kretprobe_lock, flags);
- head = kretprobe_inst_table_head(tk);
+ if (unlikely(!kprobes_initialized))
+ /* Early boot. kretprobe_table_locks not yet initialized. */
+ return;
+
+ hash = hash_ptr(tk, KPROBE_HASH_BITS);
+ head = &kretprobe_inst_table[hash];
+ kretprobe_table_lock(hash, &flags);
hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
if (ri->task == tk)
recycle_rp_inst(ri, &empty_rp);
}
- spin_unlock_irqrestore(&kretprobe_lock, flags);
-
+ kretprobe_table_unlock(hash, &flags);
+ INIT_HLIST_HEAD(&empty_rp);
hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
hlist_del(&ri->hlist);
kfree(ri);
@@ -423,24 +462,29 @@ static inline void free_rp_inst(struct k
struct kretprobe_instance *ri;
struct hlist_node *pos, *next;
- hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, uflist) {
- hlist_del(&ri->uflist);
+ hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
+ hlist_del(&ri->hlist);
kfree(ri);
}
}
static void __kprobes cleanup_rp_inst(struct kretprobe *rp) {
- unsigned long flags;
+ unsigned long flags, hash;
struct kretprobe_instance *ri;
struct hlist_node *pos, *next;
+ struct hlist_head *head;
+
/* No race here */
- spin_lock_irqsave(&kretprobe_lock, flags);
- hlist_for_each_entry_safe(ri, pos, next, &rp->used_instances, uflist) {
- ri->rp = NULL;
- hlist_del(&ri->uflist);
+ for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
+ kretprobe_table_lock(hash, &flags);
+ head = &kretprobe_inst_table[hash];
+ hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
+ if (ri->rp == rp)
+ ri->rp = NULL;
+ }
+ kretprobe_table_unlock(hash, &flags);
}
- spin_unlock_irqrestore(&kretprobe_lock, flags);
free_rp_inst(rp);
}
@@ -829,32 +873,37 @@ static int __kprobes pre_handler_kretpro
struct pt_regs *regs)
{
struct kretprobe *rp = container_of(p, struct kretprobe, kp);
- unsigned long flags = 0;
+ unsigned long hash, flags = 0;
+ struct kretprobe_instance *ri;
/*TODO: consider to only swap the RA after the last pre_handler fired */
- spin_lock_irqsave(&kretprobe_lock, flags);
+ hash = hash_ptr(current, KPROBE_HASH_BITS);
+ spin_lock_irqsave(&rp->lock, flags);
if (!hlist_empty(&rp->free_instances)) {
- struct kretprobe_instance *ri;
-
ri = hlist_entry(rp->free_instances.first,
- struct kretprobe_instance, uflist);
+ struct kretprobe_instance, hlist);
+ hlist_del(&ri->hlist);
+ spin_unlock_irqrestore(&rp->lock, flags);
+
ri->rp = rp;
ri->task = current;
if (rp->entry_handler && rp->entry_handler(ri, regs)) {
- spin_unlock_irqrestore(&kretprobe_lock, flags);
+ spin_unlock_irqrestore(&rp->lock, flags);
return 0;
}
arch_prepare_kretprobe(ri, regs);
/* XXX(hch): why is there no hlist_move_head? */
- hlist_del(&ri->uflist);
- hlist_add_head(&ri->uflist, &ri->rp->used_instances);
- hlist_add_head(&ri->hlist, kretprobe_inst_table_head(ri->task));
- } else
+ INIT_HLIST_NODE(&ri->hlist);
+ kretprobe_table_lock(hash, &flags);
+ hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
+ kretprobe_table_unlock(hash, &flags);
+ } else {
rp->nmissed++;
- spin_unlock_irqrestore(&kretprobe_lock, flags);
+ spin_unlock_irqrestore(&rp->lock, flags);
+ }
return 0;
}
@@ -890,7 +939,7 @@ static int __kprobes __register_kretprob
rp->maxactive = NR_CPUS;
#endif
}
- INIT_HLIST_HEAD(&rp->used_instances);
+ spin_lock_init(&rp->lock);
INIT_HLIST_HEAD(&rp->free_instances);
for (i = 0; i < rp->maxactive; i++) {
inst = kmalloc(sizeof(struct kretprobe_instance) +
@@ -899,8 +948,8 @@ static int __kprobes __register_kretprob
free_rp_inst(rp);
return -ENOMEM;
}
- INIT_HLIST_NODE(&inst->uflist);
- hlist_add_head(&inst->uflist, &rp->free_instances);
+ INIT_HLIST_NODE(&inst->hlist);
+ hlist_add_head(&inst->hlist, &rp->free_instances);
}
rp->nmissed = 0;
@@ -1006,6 +1055,7 @@ static int __init init_kprobes(void)
for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
INIT_HLIST_HEAD(&kprobe_table[i]); INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
+ spin_lock_init(&(kretprobe_table_locks[i].lock));
}
/*
@@ -1047,6 +1097,7 @@ static int __init init_kprobes(void)
err = arch_init_kprobes();
if (!err)
err = register_die_notifier(&kprobe_exceptions_nb);
+ kprobes_initialized = (err == 0);
if (!err)
init_test_probes();
arch/sparc64/kernel/kprobes.c +5 -6
@@ -478,9 +478,9 @@ int __kprobes longjmp_break_handler(strureturn 0 ; } -/* Called with kretprobe_lock held. The value stored in the return -* address register is actually 2 instructions before where the -* callee will return to. Sequences usually look something like this +/* The value stored in the return address register is actually 2 +* instructions before where the callee will return to. +* Sequences usually look something like this * * call some_function <--- return register points here * nop <--- call delay slot
@@ -512,8 +512,7 @@ int __kprobes trampoline_probe_handler(sunsigned long trampoline_address =(unsigned long )&kretprobe_trampoline;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +kretprobe_hash_lock (current, &head, &flags);/* * It is possible to have multiple instances associated with a given
@@ -553,7 +552,7 @@ int __kprobes trampoline_probe_handler(s regs->tnpc = orig_ret_address +4 ;reset_current_kprobe (); -spin_unlock_irqrestore (&kretprobe_lock, flags); +kretprobe_hash_unlock (current, &flags);preempt_enable_no_resched ();hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {
arch/arm/kernel/kprobes.c +2 -4
@@ -296,8 +296,7 @@ static __used __kprobes void *trampolineunsigned long trampoline_address = (unsigned long )&kretprobe_trampoline;INIT_HLIST_HEAD (&empty_rp); -spin_lock_irqsave (&kretprobe_lock, flags); - head =kretprobe_inst_table_head (current); +kretprobe_hash_lock (current, &head, &flags);/* * It is possible to have multiple instances associated with a given
@@ -337,7 +336,7 @@ static __used __kprobes void *trampoline }kretprobe_assert (ri, orig_ret_address, trampoline_address); -spin_unlock_irqrestore (&kretprobe_lock, flags); +kretprobe_hash_unlock (current, &flags);hlist_for_each_entry_safe (ri, node, tmp, &empty_rp, hlist) {hlist_del (&ri->hlist);
@@ -347,7 +346,6 @@ static __used __kprobes void *trampolinereturn (void *)orig_ret_address; } -/* Called with kretprobe_lock held. */ void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri,struct pt_regs *regs) {
Ananth N Mavinakayanahalli
Acked-byQuoted 4 lines · Srinivasa D S
Quoted 3 lines · Andrew Morton
On Wed, 21 May 2008 06:32:17 +0530
Srinivasa D S <[email protected]> wrote:
...
Quoted 46 lines
Resending the patch, Again
Currently list of kretprobe instances are stored in kretprobe object (as used_instances,free_instances) and in kretprobe hash table. We have one global kretprobe lock to serialise the access to these lists. This causes only one kretprobe handler to execute at a time. Hence affects system performance, particularly on SMP systems and when return probe is set on lot of functions (like on all systemcalls).
Solution proposed here gives fine-grain locks that performs better on SMP system compared to present kretprobe implementation.
Solution:
1) Instead of having one global lock to protect kretprobe instances
present
in kretprobe object and kretprobe hash table. We will have two locks, one lock for protecting kretprobe hash table and another lock for kretporbe object.2) We hold lock present in kretprobe object while we modify kretprobe
instance in
kretprobe object and we hold per-hash-list lock while modifying kretprobe instances
present in that hash list. To prevent deadlock, we never grab a per-hash-list lock while holding a kretprobe lock.3) We can remove
used_instancesfrom struct kretprobe, as we can track used
instances of kretprobe instances using kretprobe hash table.Time duration for kernel compilation ("make -j 8") on a 8-way ppc64 system with return probes set on all systemcalls looks like this.
cacheline non-cacheline Un-patched kernel
aligned patch aligned patch
===============================================================================real 9m46.784s 9m54.412s 10m2.450s user 40m5.715s 40m7.142s 40m4.273s sys 2m57.754s 2m58.583s 3m17.430s===========================================================
Time duration for kernel compilation ("make -j 8) on the same system, when kernel is not probed.
=========================real 9m26.389s user 40m8.775s sys 2m7.283s
Quoted 6 lines
=========================
Please let me know your comments on the patch attached here.
Signed-off-by Srinivasa DS <[email protected]>Signed-off-by Jim Keniston <[email protected]>
Quoted 5 lines · Srinivasa D S
Solution:
1) Instead of having one global lock to protect kretprobe instances
present in kretprobe object and kretprobe hash table. We will have two locks, one lock for protecting kretprobe hash table and another lock for kretporbe object.
Is it possible to get rid of the kretprobe hash table itself and lose the kretprobe_lock? It seems like it is just doing a pid-to-instance mapping. These return instances could be queued in the "current" task_struct in a LIFO manner. Mutation to this per-task list can be done with local irqs off...
--
Regards,
Abhishek Sagar
Quoted 7 lines · Abhishek Sagar
Quoted 5 lines · Srinivasa D S
Solution:
1) Instead of having one global lock to protect kretprobe instances
present in kretprobe object and kretprobe hash table. We will have two locks, one lock for protecting kretprobe hash table and another lock for kretporbe object.
Quoted 6 lines
Is it possible to get rid of the kretprobe hash table itself and lose the
kretprobe_lock? It seems like it is just doing a pid-to-instance mapping. These return instances could be queued in the "current"task_structin a LIFO manner. Mutation to this per-task list can be done with local irqs off...
There were ideas of storing kretprobe instances in task_struct to get rid of locking, but that would require extending task_struct and catching each task exit, destroying its kretprobe instances. This makes code more invasive. But in this implementation (global hash table, hashed by task), we
lock only the current task's hash bucket and hence we have fairly low contention.
Thanks
Srinivasa DS
Quoted 2 lines · Srinivasa DS
There were ideas of storing kretprobe instances in
task_structto get rid
of locking, but that would require extendingtask_struct
Wouldn't chaining of return instances in task_struct only increase its size by sizeof(struct list_head) bytes?
Quoted 1 line
and catching each task exit, destroying its kretprobe instances.
Which is kind of stil done by (...or at least we have a precendent of this issue's awareness) kprobe_flush_task().
Quoted 1 line
This makes code more invasive.
Ok.
Quoted 3 lines
But in this implementation (global hash table, hashed by task), we lock only the current task's hash bucket and hence we have fairly low
contention.
I may be underestimating the complexity of having returns instances associated with current task_struct, but anything else seems counter intuitive. There might be more possibilites to exploit the fact that functions instances are per-task.
A step in the right direction nevertheless :-)
Quoted 2 lines
Thanks
Srinivasa DS
--
Regards,
Abhishek Sagar