From: Byungchul Park <byungchul.park@lge.com>
To: torvalds@linux-foundation.org
Cc: damien.lemoal@opensource.wdc.com, linux-ide@vger.kernel.org,
adilger.kernel@dilger.ca, linux-ext4@vger.kernel.org,
mingo@redhat.com, linux-kernel@vger.kernel.org,
peterz@infradead.org, will@kernel.org, tglx@linutronix.de,
rostedt@goodmis.org, joel@joelfernandes.org, sashal@kernel.org,
daniel.vetter@ffwll.ch, chris@chris-wilson.co.uk,
duyuyang@gmail.com, johannes.berg@intel.com, tj@kernel.org,
tytso@mit.edu, willy@infradead.org, david@fromorbit.com,
amir73il@gmail.com, bfields@fieldses.org,
gregkh@linuxfoundation.org, kernel-team@lge.com,
linux-mm@kvack.org, akpm@linux-foundation.org, mhocko@kernel.org,
minchan@kernel.org, hannes@cmpxchg.org, vdavydov.dev@gmail.com,
sj@kernel.org, jglisse@redhat.com, dennis@kernel.org,
cl@linux.com, penberg@kernel.org, rientjes@google.com,
vbabka@suse.cz, ngupta@vflare.org, linux-block@vger.kernel.org,
paolo.valente@linaro.org, josef@toxicpanda.com,
linux-fsdevel@vger.kernel.org, viro@zeniv.linux.org.uk,
jack@suse.cz, jack@suse.com, jlayton@kernel.org,
dan.j.williams@intel.com, hch@infradead.org, djwong@kernel.org,
dri-devel@lists.freedesktop.org, airlied@linux.ie,
rodrigosiqueiramelo@gmail.com, melissa.srw@gmail.com,
hamohammed.sa@gmail.com, 42.hyeyoo@gmail.com
Subject: [PATCH RFC v6 19/21] dept: Differentiate onstack maps from others of different tasks in class
Date: Wed, 4 May 2022 17:17:47 +0900 [thread overview]
Message-ID: <1651652269-15342-20-git-send-email-byungchul.park@lge.com> (raw)
In-Reply-To: <1651652269-15342-1-git-send-email-byungchul.park@lge.com>
Dept assumes that maps might belong to the same class if the running
code is the same for possibility detection. However, maps on stack would
never belong to a common class between different tasks because each task
has its own instance on stack.
So differentiated onstack maps from others in class, to avoid false
positive alarms.
Signed-off-by: Byungchul Park <byungchul.park@lge.com>
---
include/linux/dept.h | 3 +
kernel/dependency/dept.c | 166 ++++++++++++++++++++++++++++++++++++++---------
kernel/exit.c | 8 ++-
3 files changed, 147 insertions(+), 30 deletions(-)
diff --git a/include/linux/dept.h b/include/linux/dept.h
index 1a3858c..3027121 100644
--- a/include/linux/dept.h
+++ b/include/linux/dept.h
@@ -72,6 +72,7 @@ struct dept_class {
*/
const char *name;
unsigned long key;
+ unsigned long key2;
int sub;
/*
@@ -343,6 +344,7 @@ struct dept_key {
struct dept_map {
const char *name;
struct dept_key *keys;
+ unsigned long key2;
int sub_usr;
/*
@@ -366,6 +368,7 @@ struct dept_map {
{ \
.name = #n, \
.keys = NULL, \
+ .key2 = 0UL, \
.sub_usr = 0, \
.keys_local = { .classes = { 0 } }, \
.wgen = 0U, \
diff --git a/kernel/dependency/dept.c b/kernel/dependency/dept.c
index 6707313..2bc6259 100644
--- a/kernel/dependency/dept.c
+++ b/kernel/dependency/dept.c
@@ -73,6 +73,7 @@
#include <linux/hash.h>
#include <linux/dept.h>
#include <linux/utsname.h>
+#include <linux/sched/task_stack.h>
#include "dept_internal.h"
static int dept_stop;
@@ -523,12 +524,12 @@ static unsigned long key_dep(struct dept_dep *d)
static bool cmp_class(struct dept_class *c1, struct dept_class *c2)
{
- return c1->key == c2->key;
+ return c1->key == c2->key && c1->key2 == c2->key2;
}
static unsigned long key_class(struct dept_class *c)
{
- return c->key;
+ return c->key2 ? mix(c->key, c->key2) : c->key;
}
#define HASH(id, bits) \
@@ -571,14 +572,38 @@ static inline struct dept_dep *lookup_dep(struct dept_class *fc,
return hash_lookup_dep(&onetime_d);
}
-static inline struct dept_class *lookup_class(unsigned long key)
+static inline struct dept_class *lookup_class(unsigned long key,
+ unsigned long key2)
{
- struct dept_class onetime_c = { .key = key };
+ struct dept_class onetime_c = { .key = key, .key2 = key2 };
return hash_lookup_class(&onetime_c);
}
/*
+ * NOTE: Must be called with dept_lock held.
+ */
+static void obtain_classes_from_hlist(struct hlist_head *to,
+ bool (*cmp)(struct dept_class *c, void *data),
+ void *data)
+{
+ struct dept_class *c;
+ struct hlist_node *n;
+ int i;
+
+ for (i = 0; i < ARRAY_SIZE(table_class); i++) {
+ struct hlist_head *h = table_class + i;
+
+ hlist_for_each_entry_safe(c, n, h, hash_node) {
+ if (cmp(c, data)) {
+ hlist_del_rcu(&c->hash_node);
+ hlist_add_head_rcu(&c->hash_node, to);
+ }
+ }
+ }
+}
+
+/*
* Report
* =====================================================================
* DEPT prints useful information to help debuging on detection of
@@ -1899,6 +1924,7 @@ void dept_map_init(struct dept_map *m, struct dept_key *k, int sub,
const char *n)
{
unsigned long flags;
+ bool onstack;
if (unlikely(READ_ONCE(dept_stop) || in_nmi()))
return;
@@ -1908,6 +1934,16 @@ void dept_map_init(struct dept_map *m, struct dept_key *k, int sub,
return;
}
+ onstack = object_is_on_stack(m);
+
+ /*
+ * Require an explicit key for onstack maps.
+ */
+ if (onstack && !k) {
+ m->nocheck = true;
+ return;
+ }
+
/*
* Allow recursive entrance.
*/
@@ -1917,6 +1953,7 @@ void dept_map_init(struct dept_map *m, struct dept_key *k, int sub,
m->sub_usr = sub;
m->keys = k;
+ m->key2 = onstack ? (unsigned long)current : 0UL;
m->name = n;
m->wgen = 0U;
m->nocheck = false;
@@ -2031,7 +2068,7 @@ static inline int map_sub(struct dept_map *m, int e)
static struct dept_class *check_new_class(struct dept_key *local,
struct dept_key *k, int sub,
- const char *n)
+ unsigned long k2, const char *n)
{
struct dept_class *c = NULL;
@@ -2047,14 +2084,14 @@ static struct dept_class *check_new_class(struct dept_key *local,
if (c)
return c;
- c = lookup_class((unsigned long)k->subkeys + sub);
+ c = lookup_class((unsigned long)k->subkeys + sub, k2);
if (c)
goto caching;
if (unlikely(!dept_lock()))
return NULL;
- c = lookup_class((unsigned long)k->subkeys + sub);
+ c = lookup_class((unsigned long)k->subkeys + sub, k2);
if (unlikely(c))
goto unlock;
@@ -2065,6 +2102,7 @@ static struct dept_class *check_new_class(struct dept_key *local,
c->name = n;
c->sub = sub;
c->key = (unsigned long)(k->subkeys + sub);
+ c->key2 = k2;
hash_add_class(c);
list_add(&c->all_node, &dept_classes);
unlock:
@@ -2099,8 +2137,8 @@ static void __dept_wait(struct dept_map *m, unsigned long w_f,
struct dept_key *k;
k = m->keys ?: &m->keys_local;
- c = check_new_class(&m->keys_local, k,
- map_sub(m, e), m->name);
+ c = check_new_class(&m->keys_local, k, map_sub(m, e),
+ m->key2, m->name);
if (!c)
continue;
@@ -2298,7 +2336,8 @@ void dept_ecxt_enter(struct dept_map *m, unsigned long e_f, unsigned long ip,
DEPT_WARN_ON(1UL << e != e_f);
k = m->keys ?: &m->keys_local;
- c = check_new_class(&m->keys_local, k, map_sub(m, e), m->name);
+ c = check_new_class(&m->keys_local, k, map_sub(m, e),
+ m->key2, m->name);
if (c && add_ecxt((void *)m, c, ip, c_fn, e_fn, ne))
goto exit;
@@ -2376,7 +2415,8 @@ void dept_event(struct dept_map *m, unsigned long e_f, unsigned long ip,
DEPT_WARN_ON(1UL << e != e_f);
k = m->keys ?: &m->keys_local;
- c = check_new_class(&m->keys_local, k, map_sub(m, e), m->name);
+ c = check_new_class(&m->keys_local, k, map_sub(m, e),
+ m->key2, m->name);
if (c && add_ecxt((void *)m, c, 0UL, NULL, e_fn, 0)) {
do_event((void *)m, c, READ_ONCE(m->wgen), ip);
@@ -2427,7 +2467,8 @@ void dept_ecxt_exit(struct dept_map *m, unsigned long e_f,
DEPT_WARN_ON(1UL << e != e_f);
k = m->keys ?: &m->keys_local;
- c = check_new_class(&m->keys_local, k, map_sub(m, e), m->name);
+ c = check_new_class(&m->keys_local, k, map_sub(m, e),
+ m->key2, m->name);
if (c && pop_ecxt((void *)m, c))
goto exit;
@@ -2504,7 +2545,7 @@ void dept_wait_split_map(struct dept_map_each *me,
flags = dept_enter();
k = mc->keys ?: &mc->keys_local;
- c = check_new_class(&mc->keys_local, k, 0, mc->name);
+ c = check_new_class(&mc->keys_local, k, 0, 0UL, mc->name);
if (c)
add_wait(c, ip, w_fn, ne);
@@ -2568,7 +2609,7 @@ void dept_event_split_map(struct dept_map_each *me,
flags = dept_enter();
k = mc->keys ?: &mc->keys_local;
- c = check_new_class(&mc->keys_local, k, 0, mc->name);
+ c = check_new_class(&mc->keys_local, k, 0, 0UL, mc->name);
if (c && add_ecxt((void *)me, c, 0UL, NULL, e_fn, 0)) {
do_event((void *)me, c, READ_ONCE(me->wgen), ip);
@@ -2584,12 +2625,64 @@ void dept_event_split_map(struct dept_map_each *me,
}
EXPORT_SYMBOL_GPL(dept_event_split_map);
+static bool cmp_class_key2(struct dept_class *c, void *k2)
+{
+ return c->key2 == (unsigned long)k2;
+}
+
+static void per_task_key_destroy(void)
+{
+ struct dept_class *c;
+ struct hlist_node *n;
+ HLIST_HEAD(h);
+
+ /*
+ * per_task_key_destroy() should not fail.
+ *
+ * FIXME: Should be fixed if per_task_key_destroy() causes
+ * deadlock with dept_lock().
+ */
+ while (unlikely(!dept_lock()))
+ cpu_relax();
+
+ obtain_classes_from_hlist(&h, cmp_class_key2, current);
+
+ hlist_for_each_entry_safe(c, n, &h, hash_node) {
+ hash_del_class(c);
+ disconnect_class(c);
+ list_del(&c->all_node);
+ inval_class(c);
+
+ /*
+ * Actual deletion will happen on the rcu callback
+ * that has been added in disconnect_class().
+ */
+ del_class(c);
+ }
+
+ dept_unlock();
+}
+
void dept_task_exit(struct task_struct *t)
{
- struct dept_task *dt = &t->dept_task;
+ struct dept_task *dt = dept_task();
+ unsigned long flags;
int i;
- raw_local_irq_disable();
+ if (unlikely(READ_ONCE(dept_stop) || in_nmi()))
+ return;
+
+ if (dt->recursive) {
+ DEPT_STOP("Entered task_exit() while Dept is working.\n");
+ return;
+ }
+
+ if (t != current) {
+ DEPT_STOP("Never expect task_exit() done by others.\n");
+ return;
+ }
+
+ flags = dept_enter();
if (dt->stack)
put_stack(dt->stack);
@@ -2601,9 +2694,17 @@ void dept_task_exit(struct task_struct *t)
if (dt->wait_hist[i].wait)
put_wait(dt->wait_hist[i].wait);
+ per_task_key_destroy();
+
dept_off();
+ dept_exit(flags);
- raw_local_irq_enable();
+ /*
+ * Wait until even lockless hash_lookup_class() for the class
+ * returns NULL.
+ */
+ might_sleep();
+ synchronize_rcu();
}
void dept_task_init(struct task_struct *t)
@@ -2611,10 +2712,18 @@ void dept_task_init(struct task_struct *t)
memset(&t->dept_task, 0x0, sizeof(struct dept_task));
}
+static bool cmp_class_key1(struct dept_class *c, void *k1)
+{
+ return c->key == (unsigned long)k1;
+}
+
void dept_key_init(struct dept_key *k)
{
struct dept_task *dt = dept_task();
unsigned long flags;
+ struct dept_class *c;
+ struct hlist_node *n;
+ HLIST_HEAD(h);
int sub;
if (unlikely(READ_ONCE(dept_stop) || in_nmi()))
@@ -2636,13 +2745,11 @@ void dept_key_init(struct dept_key *k)
while (unlikely(!dept_lock()))
cpu_relax();
- for (sub = 0; sub < DEPT_MAX_SUBCLASSES; sub++) {
- struct dept_class *c;
-
- c = lookup_class((unsigned long)k->subkeys + sub);
- if (!c)
- continue;
+ for (sub = 0; sub < DEPT_MAX_SUBCLASSES; sub++)
+ obtain_classes_from_hlist(&h, cmp_class_key1,
+ k->subkeys + sub);
+ hlist_for_each_entry_safe(c, n, &h, hash_node) {
DEPT_STOP("The class(%s/%d) has not been removed.\n",
c->name, sub);
break;
@@ -2657,6 +2764,9 @@ void dept_key_destroy(struct dept_key *k)
{
struct dept_task *dt = dept_task();
unsigned long flags;
+ struct dept_class *c;
+ struct hlist_node *n;
+ HLIST_HEAD(h);
int sub;
if (unlikely(READ_ONCE(dept_stop) || in_nmi()))
@@ -2678,13 +2788,11 @@ void dept_key_destroy(struct dept_key *k)
while (unlikely(!dept_lock()))
cpu_relax();
- for (sub = 0; sub < DEPT_MAX_SUBCLASSES; sub++) {
- struct dept_class *c;
-
- c = lookup_class((unsigned long)k->subkeys + sub);
- if (!c)
- continue;
+ for (sub = 0; sub < DEPT_MAX_SUBCLASSES; sub++)
+ obtain_classes_from_hlist(&h, cmp_class_key1,
+ k->subkeys + sub);
+ hlist_for_each_entry_safe(c, n, &h, hash_node) {
hash_del_class(c);
disconnect_class(c);
list_del(&c->all_node);
diff --git a/kernel/exit.c b/kernel/exit.c
index bac41ee..d381fd4 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -738,6 +738,13 @@ void __noreturn do_exit(long code)
struct task_struct *tsk = current;
int group_dead;
+ /*
+ * dept_task_exit() requires might_sleep() because it needs to
+ * wait on the grace period after cleaning the objects that have
+ * been coupled with the current task_struct.
+ */
+ dept_task_exit(tsk);
+
WARN_ON(tsk->plug);
kcov_task_exit(tsk);
@@ -844,7 +851,6 @@ void __noreturn do_exit(long code)
exit_tasks_rcu_finish();
lockdep_free_task(tsk);
- dept_task_exit(tsk);
do_task_dead();
}
--
1.9.1
next prev parent reply other threads:[~2022-05-04 8:19 UTC|newest]
Thread overview: 52+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-05-04 8:17 [PATCH RFC v6 00/21] DEPT(Dependency Tracker) Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 01/21] llist: Move llist_{head,node} definition to types.h Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 02/21] dept: Implement Dept(Dependency Tracker) Byungchul Park
2022-05-21 3:24 ` Hyeonggon Yoo
2022-05-04 8:17 ` [PATCH RFC v6 03/21] dept: Apply Dept to spinlock Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 04/21] dept: Apply Dept to mutex families Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 05/21] dept: Apply Dept to rwlock Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 06/21] dept: Apply Dept to wait_for_completion()/complete() Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 07/21] dept: Apply Dept to seqlock Byungchul Park
2022-05-21 5:25 ` Hyeonggon Yoo
2022-05-24 6:00 ` Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 08/21] dept: Apply Dept to rwsem Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 09/21] dept: Add proc knobs to show stats and dependency graph Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 10/21] dept: Introduce split map concept and new APIs for them Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 11/21] dept: Apply Dept to wait/event of PG_{locked,writeback} Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 12/21] dept: Apply SDT to swait Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 13/21] dept: Apply SDT to wait(waitqueue) Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 14/21] locking/lockdep, cpu/hotplus: Use a weaker annotation in AP thread Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 15/21] dept: Distinguish each syscall context from another Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 16/21] dept: Distinguish each work " Byungchul Park
2022-05-04 11:23 ` Sergey Shtylyov
2022-05-04 8:17 ` [PATCH RFC v6 17/21] dept: Disable Dept within the wait_bit layer by default Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 18/21] dept: Disable Dept on struct crypto_larval's completion for now Byungchul Park
2022-05-04 8:17 ` Byungchul Park [this message]
2022-05-04 8:17 ` [PATCH RFC v6 20/21] dept: Do not add dependencies between events within scheduler and sleeps Byungchul Park
2022-05-04 8:17 ` [PATCH RFC v6 21/21] dept: Unstage wait when tagging a normal sleep wait Byungchul Park
2022-05-04 18:17 ` [PATCH RFC v6 00/21] DEPT(Dependency Tracker) Linus Torvalds
2022-05-06 0:11 ` Byungchul Park
2022-05-07 7:20 ` Hyeonggon Yoo
2022-05-09 0:16 ` Byungchul Park
2022-05-09 20:47 ` Steven Rostedt
2022-05-09 23:38 ` Byungchul Park
2022-05-10 14:12 ` Steven Rostedt
2022-05-10 23:26 ` Byungchul Park
2022-05-10 11:18 ` Hyeonggon Yoo
2022-05-10 23:39 ` Byungchul Park
2022-05-11 10:04 ` Hyeonggon Yoo
2022-05-19 10:11 ` Catalin Marinas
2022-05-23 2:43 ` Byungchul Park
2022-05-09 1:22 ` Byungchul Park
2022-05-09 21:05 ` Theodore Ts'o
2022-05-09 22:28 ` Theodore Ts'o
2022-05-10 0:32 ` Byungchul Park
2022-05-10 1:32 ` Theodore Ts'o
2022-05-10 5:37 ` Byungchul Park
2022-05-11 1:16 ` Byungchul Park
2022-05-12 5:25 ` [REPORT] syscall reboot + umh + firmware fallback Byungchul Park
2022-05-12 9:15 ` Tejun Heo
2022-05-12 11:18 ` Byungchul Park
2022-05-12 13:56 ` Theodore Ts'o
2022-05-23 1:10 ` Byungchul Park
2022-05-12 16:41 ` Tejun Heo
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1651652269-15342-20-git-send-email-byungchul.park@lge.com \
--to=byungchul.park@lge.com \
--cc=42.hyeyoo@gmail.com \
--cc=adilger.kernel@dilger.ca \
--cc=airlied@linux.ie \
--cc=akpm@linux-foundation.org \
--cc=amir73il@gmail.com \
--cc=bfields@fieldses.org \
--cc=chris@chris-wilson.co.uk \
--cc=cl@linux.com \
--cc=damien.lemoal@opensource.wdc.com \
--cc=dan.j.williams@intel.com \
--cc=daniel.vetter@ffwll.ch \
--cc=david@fromorbit.com \
--cc=dennis@kernel.org \
--cc=djwong@kernel.org \
--cc=dri-devel@lists.freedesktop.org \
--cc=duyuyang@gmail.com \
--cc=gregkh@linuxfoundation.org \
--cc=hamohammed.sa@gmail.com \
--cc=hannes@cmpxchg.org \
--cc=hch@infradead.org \
--cc=jack@suse.com \
--cc=jack@suse.cz \
--cc=jglisse@redhat.com \
--cc=jlayton@kernel.org \
--cc=joel@joelfernandes.org \
--cc=johannes.berg@intel.com \
--cc=josef@toxicpanda.com \
--cc=kernel-team@lge.com \
--cc=linux-block@vger.kernel.org \
--cc=linux-ext4@vger.kernel.org \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-ide@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=melissa.srw@gmail.com \
--cc=mhocko@kernel.org \
--cc=minchan@kernel.org \
--cc=mingo@redhat.com \
--cc=ngupta@vflare.org \
--cc=paolo.valente@linaro.org \
--cc=penberg@kernel.org \
--cc=peterz@infradead.org \
--cc=rientjes@google.com \
--cc=rodrigosiqueiramelo@gmail.com \
--cc=rostedt@goodmis.org \
--cc=sashal@kernel.org \
--cc=sj@kernel.org \
--cc=tglx@linutronix.de \
--cc=tj@kernel.org \
--cc=torvalds@linux-foundation.org \
--cc=tytso@mit.edu \
--cc=vbabka@suse.cz \
--cc=vdavydov.dev@gmail.com \
--cc=viro@zeniv.linux.org.uk \
--cc=will@kernel.org \
--cc=willy@infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).