2022-08-27 10:29:43

by Liu Shixin

[permalink] [raw]
Subject: [PATCH -next v3 0/5] Delay the initializaton of zswap

In the initialization of zswap, about 18MB memory will be allocated for
zswap_pool. Since not all users use zswap, the memory may be wasted. Save
the memory for these users by delaying the initialization of zswap to
first enablement.

v2->v3: Fix frontswap_ops NULL reported by Nathan and add init for online
swap device in backend register.
v1->v2: Change init_zswap to zswap_init suggested by Andrew.

Liu Shixin (5):
frontswap: skip frontswap_ops init if zswap init failed.
frontswap: invoke ops->init for online swap device in
frontswap_register_ops
mm/zswap: replace zswap_init_{started/failed} with zswap_init_state
mm/zswap: delay the initializaton of zswap until the first enablement
mm/zswap: skip confusing print info

include/linux/swapfile.h | 2 ++
mm/frontswap.c | 50 +++++++++++++++++++++++++-
mm/swapfile.c | 4 +--
mm/zswap.c | 77 ++++++++++++++++++++++++++++++----------
4 files changed, 111 insertions(+), 22 deletions(-)

--
2.25.1


2022-08-27 10:40:11

by Liu Shixin

[permalink] [raw]
Subject: [PATCH -next v3 4/5] mm/zswap: delay the initializaton of zswap until the first enablement

In the initialization of zswap, about 18MB memory will be allocated for
zswap_pool in my machine. Since not all users use zswap, the memory may be
wasted. Save the memory for these users by delaying the initialization of
zswap to first enablement.

Signed-off-by: Liu Shixin <[email protected]>
---
mm/zswap.c | 50 ++++++++++++++++++++++++++++++++++++++++----------
1 file changed, 40 insertions(+), 10 deletions(-)

diff --git a/mm/zswap.c b/mm/zswap.c
index 84e38300f571..4c476c463035 100644
--- a/mm/zswap.c
+++ b/mm/zswap.c
@@ -81,6 +81,8 @@ static bool zswap_pool_reached_full;

#define ZSWAP_PARAM_UNSET ""

+static int zswap_setup(void);
+
/* Enable/disable zswap */
static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
static int zswap_enabled_param_set(const char *,
@@ -220,6 +222,8 @@ static atomic_t zswap_pools_count = ATOMIC_INIT(0);

/* init state */
static int zswap_init_state;
+/* used to ensure the integrity of initialization */
+static DEFINE_MUTEX(zswap_init_lock);

/* init completed, but couldn't create the initial pool */
static bool zswap_has_pool;
@@ -273,13 +277,13 @@ static void zswap_update_total_size(void)
**********************************/
static struct kmem_cache *zswap_entry_cache;

-static int __init zswap_entry_cache_create(void)
+static int zswap_entry_cache_create(void)
{
zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
return zswap_entry_cache == NULL;
}

-static void __init zswap_entry_cache_destroy(void)
+static void zswap_entry_cache_destroy(void)
{
kmem_cache_destroy(zswap_entry_cache);
}
@@ -664,7 +668,7 @@ static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
return NULL;
}

-static __init struct zswap_pool *__zswap_pool_create_fallback(void)
+static struct zswap_pool *__zswap_pool_create_fallback(void)
{
bool has_comp, has_zpool;

@@ -782,11 +786,17 @@ static int __zswap_param_set(const char *val, const struct kernel_param *kp,
if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
return 0;

- /* if this is load-time (pre-init) param setting,
+ /*
+ * if zswap has not been initialized,
* don't create a pool; that's done during init.
*/
- if (zswap_init_state == ZSWAP_UNINIT)
- return param_set_charp(s, kp);
+ mutex_lock(&zswap_init_lock);
+ if (zswap_init_state == ZSWAP_UNINIT) {
+ ret = param_set_charp(s, kp);
+ mutex_unlock(&zswap_init_lock);
+ return ret;
+ }
+ mutex_unlock(&zswap_init_lock);

if (!type) {
if (!zpool_has_pool(s)) {
@@ -876,6 +886,14 @@ static int zswap_zpool_param_set(const char *val,
static int zswap_enabled_param_set(const char *val,
const struct kernel_param *kp)
{
+ if (system_state == SYSTEM_RUNNING) {
+ mutex_lock(&zswap_init_lock);
+ if (zswap_setup()) {
+ mutex_unlock(&zswap_init_lock);
+ return -ENODEV;
+ }
+ mutex_unlock(&zswap_init_lock);
+ }
if (zswap_init_state == ZSWAP_INIT_FAILED) {
pr_err("can't enable, initialization failed\n");
return -ENODEV;
@@ -1432,7 +1450,7 @@ static const struct frontswap_ops zswap_frontswap_ops = {

static struct dentry *zswap_debugfs_root;

-static int __init zswap_debugfs_init(void)
+static int zswap_debugfs_init(void)
{
if (!debugfs_initialized())
return -ENODEV;
@@ -1463,7 +1481,7 @@ static int __init zswap_debugfs_init(void)
return 0;
}
#else
-static int __init zswap_debugfs_init(void)
+static int zswap_debugfs_init(void)
{
return 0;
}
@@ -1472,11 +1490,14 @@ static int __init zswap_debugfs_init(void)
/*********************************
* module init and exit
**********************************/
-static int __init init_zswap(void)
+static int zswap_setup(void)
{
struct zswap_pool *pool;
int ret;

+ if (zswap_init_state != ZSWAP_UNINIT)
+ return 0;
+
if (zswap_entry_cache_create()) {
pr_err("entry cache creation failed\n");
goto cache_fail;
@@ -1534,8 +1555,17 @@ static int __init init_zswap(void)
zswap_enabled = false;
return -ENOMEM;
}
+
+static int __init zswap_init(void)
+{
+ /* skip init if zswap is disabled when system startup */
+ if (!zswap_enabled)
+ return 0;
+ return zswap_setup();
+}
+
/* must be late so crypto has time to come up */
-late_initcall(init_zswap);
+late_initcall(zswap_init);

MODULE_LICENSE("GPL");
MODULE_AUTHOR("Seth Jennings <[email protected]>");
--
2.25.1

2022-08-27 10:41:36

by Liu Shixin

[permalink] [raw]
Subject: [PATCH -next v3 5/5] mm/zswap: skip confusing print info

It's confusing when we disable zswap while zswap is init failed or has no
pool. If no change required, just return directly.

Signed-off-by: Liu Shixin <[email protected]>
---
mm/zswap.c | 9 +++++++++
1 file changed, 9 insertions(+)

diff --git a/mm/zswap.c b/mm/zswap.c
index 4c476c463035..ef7463550e49 100644
--- a/mm/zswap.c
+++ b/mm/zswap.c
@@ -886,6 +886,15 @@ static int zswap_zpool_param_set(const char *val,
static int zswap_enabled_param_set(const char *val,
const struct kernel_param *kp)
{
+ bool res;
+
+ if (kstrtobool(val, &res))
+ return -EINVAL;
+
+ /* no change required */
+ if (res == *(bool *)kp->arg)
+ return 0;
+
if (system_state == SYSTEM_RUNNING) {
mutex_lock(&zswap_init_lock);
if (zswap_setup()) {
--
2.25.1

2022-08-27 10:42:19

by Liu Shixin

[permalink] [raw]
Subject: [PATCH -next v3 2/5] frontswap: invoke ops->init for online swap device in frontswap_register_ops

Since we are supported to delay zswap initializaton, we need to invoke
ops->init for the swap device which is already online when register
backend.

This patch is a revert of f328c1d16e4c ("frontswap: simplify frontswap_register_ops")
and 633423a09cb5 ("mm: mark swap_lock and swap_active_head static")

Signed-off-by: Liu Shixin <[email protected]>
---
include/linux/swapfile.h | 2 ++
mm/frontswap.c | 47 ++++++++++++++++++++++++++++++++++++++++
mm/swapfile.c | 4 ++--
3 files changed, 51 insertions(+), 2 deletions(-)

diff --git a/include/linux/swapfile.h b/include/linux/swapfile.h
index 2fbcc9afd814..75fc069594a5 100644
--- a/include/linux/swapfile.h
+++ b/include/linux/swapfile.h
@@ -6,6 +6,8 @@
* these were static in swapfile.c but frontswap.c needs them and we don't
* want to expose them to the dozens of source files that include swap.h
*/
+extern spinlock_t swap_lock;
+extern struct plist_head swap_active_head;
extern struct swap_info_struct *swap_info[];
extern unsigned long generic_max_swapfile_size(void);
/* Maximum swapfile size supported for the arch (not inclusive). */
diff --git a/mm/frontswap.c b/mm/frontswap.c
index 620f95af81dd..449e6f499b88 100644
--- a/mm/frontswap.c
+++ b/mm/frontswap.c
@@ -96,11 +96,58 @@ static inline void inc_frontswap_invalidates(void) { }
*/
int frontswap_register_ops(const struct frontswap_ops *ops)
{
+ DECLARE_BITMAP(a, MAX_SWAPFILES);
+ DECLARE_BITMAP(b, MAX_SWAPFILES);
+ struct swap_info_struct *si;
+ unsigned int i;
+
if (frontswap_ops)
return -EINVAL;

+ bitmap_zero(a, MAX_SWAPFILES);
+ bitmap_zero(b, MAX_SWAPFILES);
+
+ spin_lock(&swap_lock);
+ plist_for_each_entry(si, &swap_active_head, list) {
+ if (!WARN_ON(!si->frontswap_map))
+ __set_bit(si->type, a);
+ }
+ spin_unlock(&swap_lock);
+
+ /* the new ops needs to know the currently active swap devices */
+ for_each_set_bit(i, a, MAX_SWAPFILES) {
+ pr_err("init frontswap_ops\n");
+ ops->init(i);
+ }
+
frontswap_ops = ops;
static_branch_inc(&frontswap_enabled_key);
+
+ spin_lock(&swap_lock);
+ plist_for_each_entry(si, &swap_active_head, list) {
+ if (si->frontswap_map)
+ __set_bit(si->type, b);
+ }
+ spin_unlock(&swap_lock);
+
+ /*
+ * On the very unlikely chance that a swap device was added or
+ * removed between setting the "a" list bits and the ops init
+ * calls, we re-check and do init or invalidate for any changed
+ * bits.
+ */
+ if (unlikely(!bitmap_equal(a, b, MAX_SWAPFILES))) {
+ for (i = 0; i < MAX_SWAPFILES; i++) {
+ if (!test_bit(i, a) && test_bit(i, b)) {
+ pr_err("init frontswap_ops re\n");
+ ops->init(i);
+ } else if (test_bit(i, a) && !test_bit(i, b)) {
+ pr_err("inval frontswap_ops re\n");
+ ops->invalidate_area(i);
+ }
+ }
+ }
+
return 0;
}

diff --git a/mm/swapfile.c b/mm/swapfile.c
index 469d9af86be2..d383b282f269 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -51,7 +51,7 @@ static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
unsigned char);
static void free_swap_count_continuations(struct swap_info_struct *);

-static DEFINE_SPINLOCK(swap_lock);
+DEFINE_SPINLOCK(swap_lock);
static unsigned int nr_swapfiles;
atomic_long_t nr_swap_pages;
/*
@@ -77,7 +77,7 @@ static const char Unused_offset[] = "Unused swap offset entry ";
* all active swap_info_structs
* protected with swap_lock, and ordered by priority.
*/
-static PLIST_HEAD(swap_active_head);
+PLIST_HEAD(swap_active_head);

/*
* all available (active, not full) swap_info_structs
--
2.25.1

2022-08-27 10:44:29

by Liu Shixin

[permalink] [raw]
Subject: [PATCH -next v3 3/5] mm/zswap: replace zswap_init_{started/failed} with zswap_init_state

zswap_init_started indicates that the initialization is started. And
zswap_init_failed indicates that the initialization is failed. As we will
support to init zswap after system startup, it's necessary to add a state
to indicate the initialization is complete and succeed to avoid
concurrency issues. Since we don't care about the difference between
init started with init completion. We only need three states:
uninitialized, initial failed, initial succeed.

Signed-off-by: Liu Shixin <[email protected]>
---
mm/zswap.c | 22 +++++++++++-----------
1 file changed, 11 insertions(+), 11 deletions(-)

diff --git a/mm/zswap.c b/mm/zswap.c
index 2d48fd59cc7a..84e38300f571 100644
--- a/mm/zswap.c
+++ b/mm/zswap.c
@@ -214,11 +214,12 @@ static DEFINE_SPINLOCK(zswap_pools_lock);
/* pool counter to provide unique names to zpool */
static atomic_t zswap_pools_count = ATOMIC_INIT(0);

-/* used by param callback function */
-static bool zswap_init_started;
+#define ZSWAP_UNINIT 0
+#define ZSWAP_INIT_SUCCEED 1
+#define ZSWAP_INIT_FAILED 2

-/* fatal error during init */
-static bool zswap_init_failed;
+/* init state */
+static int zswap_init_state;

/* init completed, but couldn't create the initial pool */
static bool zswap_has_pool;
@@ -772,7 +773,7 @@ static int __zswap_param_set(const char *val, const struct kernel_param *kp,
char *s = strstrip((char *)val);
int ret;

- if (zswap_init_failed) {
+ if (zswap_init_state == ZSWAP_INIT_FAILED) {
pr_err("can't set param, initialization failed\n");
return -ENODEV;
}
@@ -784,7 +785,7 @@ static int __zswap_param_set(const char *val, const struct kernel_param *kp,
/* if this is load-time (pre-init) param setting,
* don't create a pool; that's done during init.
*/
- if (!zswap_init_started)
+ if (zswap_init_state == ZSWAP_UNINIT)
return param_set_charp(s, kp);

if (!type) {
@@ -875,11 +876,11 @@ static int zswap_zpool_param_set(const char *val,
static int zswap_enabled_param_set(const char *val,
const struct kernel_param *kp)
{
- if (zswap_init_failed) {
+ if (zswap_init_state == ZSWAP_INIT_FAILED) {
pr_err("can't enable, initialization failed\n");
return -ENODEV;
}
- if (!zswap_has_pool && zswap_init_started) {
+ if (!zswap_has_pool && zswap_init_state == ZSWAP_INIT_SUCCEED) {
pr_err("can't enable, no pool configured\n");
return -ENODEV;
}
@@ -1476,8 +1477,6 @@ static int __init init_zswap(void)
struct zswap_pool *pool;
int ret;

- zswap_init_started = true;
-
if (zswap_entry_cache_create()) {
pr_err("entry cache creation failed\n");
goto cache_fail;
@@ -1517,6 +1516,7 @@ static int __init init_zswap(void)
goto destroy_wq;
if (zswap_debugfs_init())
pr_warn("debugfs initialization failed\n");
+ zswap_init_state = ZSWAP_INIT_SUCCEED;
return 0;

destroy_wq:
@@ -1530,7 +1530,7 @@ static int __init init_zswap(void)
zswap_entry_cache_destroy();
cache_fail:
/* if built-in, we aren't unloaded on failure; don't allow use */
- zswap_init_failed = true;
+ zswap_init_state = ZSWAP_INIT_FAILED;
zswap_enabled = false;
return -ENOMEM;
}
--
2.25.1

2022-08-27 10:47:02

by Liu Shixin

[permalink] [raw]
Subject: [PATCH -next v3 1/5] frontswap: skip frontswap_ops init if zswap init failed.

If zswap initial failed or has not been initial, frontswap_ops will be
NULL. In such situation, swap device would enable failed with following
stack trace:

Unable to handle kernel access to user memory outside uaccess routines at virtual address 0000000000000000
Mem abort info:
ESR = 0x0000000096000004
EC = 0x25: DABT (current EL), IL = 32 bits
SET = 0, FnV = 0
EA = 0, S1PTW = 0
FSC = 0x04: level 0 translation fault
Data abort info:
ISV = 0, ISS = 0x00000004
CM = 0, WnR = 0
user pgtable: 4k pages, 48-bit VAs, pgdp=00000020a4fab000
[0000000000000000] pgd=0000000000000000, p4d=0000000000000000
Internal error: Oops: 96000004 [#1] SMP
Modules linked in: zram fsl_dpaa2_eth pcs_lynx phylink ahci_qoriq crct10dif_ce ghash_ce sbsa_gwdt fsl_mc_dpio nvme lm90 nvme_core at803x xhci_plat_hcd rtc_fsl_ftm_alarm xgmac_mdio ahci_platform i2c_imx ip6_tables ip_tables fuse
Unloaded tainted modules: cppc_cpufreq():1
CPU: 10 PID: 761 Comm: swapon Not tainted 6.0.0-rc2-00454-g22100432cf14 #1
Hardware name: SolidRun Ltd. SolidRun CEX7 Platform, BIOS EDK II Jun 21 2022
pstate: 00400005 (nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : frontswap_init+0x38/0x60
lr : __do_sys_swapon+0x8a8/0x9f4
sp : ffff80000969bcf0
x29: ffff80000969bcf0 x28: ffff37bee0d8fc00 x27: ffff80000a7f5000
x26: fffffcdefb971e80 x25: ffffaba797453b90 x24: 0000000000000064
x23: ffff37c1f209d1a8 x22: ffff37bee880e000 x21: ffffaba797748560
x20: ffff37bee0d8fce4 x19: ffffaba797748488 x18: 0000000000000014
x17: 0000000030ec029a x16: ffffaba795a479b0 x15: 0000000000000000
x14: 0000000000000000 x13: 0000000000000030 x12: 0000000000000001
x11: ffff37c63c0aba18 x10: 0000000000000000 x9 : ffffaba7956b8c88
x8 : ffff80000969bcd0 x7 : 0000000000000000 x6 : 0000000000000000
x5 : 0000000000000001 x4 : 0000000000000000 x3 : ffffaba79730f000
x2 : ffff37bee0d8fc00 x1 : 0000000000000000 x0 : 0000000000000000
Call trace:
frontswap_init+0x38/0x60
__do_sys_swapon+0x8a8/0x9f4
__arm64_sys_swapon+0x28/0x3c
invoke_syscall+0x78/0x100
el0_svc_common.constprop.0+0xd4/0xf4
do_el0_svc+0x38/0x4c
el0_svc+0x34/0x10c
el0t_64_sync_handler+0x11c/0x150
el0t_64_sync+0x190/0x194
Code: d000e283 910003fd f9006c41 f946d461 (f9400021)
---[ end trace 0000000000000000 ]---

Reported-by: Nathan Chancellor <[email protected]>
Signed-off-by: Liu Shixin <[email protected]>
---
mm/frontswap.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/mm/frontswap.c b/mm/frontswap.c
index 1a97610308cb..620f95af81dd 100644
--- a/mm/frontswap.c
+++ b/mm/frontswap.c
@@ -125,7 +125,8 @@ void frontswap_init(unsigned type, unsigned long *map)
* p->frontswap set to something valid to work properly.
*/
frontswap_map_set(sis, map);
- frontswap_ops->init(type);
+ if (frontswap_ops)
+ frontswap_ops->init(type);
}

static bool __frontswap_test(struct swap_info_struct *sis,
--
2.25.1

2022-08-27 22:31:04

by Andrew Morton

[permalink] [raw]
Subject: Re: [PATCH -next v3 0/5] Delay the initializaton of zswap

On Sat, 27 Aug 2022 18:45:55 +0800 Liu Shixin <[email protected]> wrote:

> In the initialization of zswap, about 18MB memory will be allocated for
> zswap_pool. Since not all users use zswap, the memory may be wasted. Save
> the memory for these users by delaying the initialization of zswap to
> first enablement.
>
> v2->v3: Fix frontswap_ops NULL reported by Nathan and add init for online
> swap device in backend register.
> v1->v2: Change init_zswap to zswap_init suggested by Andrew.

Thanks. Konrad, could you please take a look for the frontswap
changes?

2022-08-28 20:48:34

by Vitaly Wool

[permalink] [raw]
Subject: Re: [PATCH -next v3 4/5] mm/zswap: delay the initializaton of zswap until the first enablement

On Sat, Aug 27, 2022 at 12:12 PM Liu Shixin <[email protected]> wrote:
>
> In the initialization of zswap, about 18MB memory will be allocated for
> zswap_pool in my machine. Since not all users use zswap, the memory may be
> wasted. Save the memory for these users by delaying the initialization of
> zswap to first enablement.
>
> Signed-off-by: Liu Shixin <[email protected]>

Reviewed-by: Vitaly Wool <[email protected]>

> ---
> mm/zswap.c | 50 ++++++++++++++++++++++++++++++++++++++++----------
> 1 file changed, 40 insertions(+), 10 deletions(-)
>
> diff --git a/mm/zswap.c b/mm/zswap.c
> index 84e38300f571..4c476c463035 100644
> --- a/mm/zswap.c
> +++ b/mm/zswap.c
> @@ -81,6 +81,8 @@ static bool zswap_pool_reached_full;
>
> #define ZSWAP_PARAM_UNSET ""
>
> +static int zswap_setup(void);
> +
> /* Enable/disable zswap */
> static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
> static int zswap_enabled_param_set(const char *,
> @@ -220,6 +222,8 @@ static atomic_t zswap_pools_count = ATOMIC_INIT(0);
>
> /* init state */
> static int zswap_init_state;
> +/* used to ensure the integrity of initialization */
> +static DEFINE_MUTEX(zswap_init_lock);
>
> /* init completed, but couldn't create the initial pool */
> static bool zswap_has_pool;
> @@ -273,13 +277,13 @@ static void zswap_update_total_size(void)
> **********************************/
> static struct kmem_cache *zswap_entry_cache;
>
> -static int __init zswap_entry_cache_create(void)
> +static int zswap_entry_cache_create(void)
> {
> zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
> return zswap_entry_cache == NULL;
> }
>
> -static void __init zswap_entry_cache_destroy(void)
> +static void zswap_entry_cache_destroy(void)
> {
> kmem_cache_destroy(zswap_entry_cache);
> }
> @@ -664,7 +668,7 @@ static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
> return NULL;
> }
>
> -static __init struct zswap_pool *__zswap_pool_create_fallback(void)
> +static struct zswap_pool *__zswap_pool_create_fallback(void)
> {
> bool has_comp, has_zpool;
>
> @@ -782,11 +786,17 @@ static int __zswap_param_set(const char *val, const struct kernel_param *kp,
> if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
> return 0;
>
> - /* if this is load-time (pre-init) param setting,
> + /*
> + * if zswap has not been initialized,
> * don't create a pool; that's done during init.
> */
> - if (zswap_init_state == ZSWAP_UNINIT)
> - return param_set_charp(s, kp);
> + mutex_lock(&zswap_init_lock);
> + if (zswap_init_state == ZSWAP_UNINIT) {
> + ret = param_set_charp(s, kp);
> + mutex_unlock(&zswap_init_lock);
> + return ret;
> + }
> + mutex_unlock(&zswap_init_lock);
>
> if (!type) {
> if (!zpool_has_pool(s)) {
> @@ -876,6 +886,14 @@ static int zswap_zpool_param_set(const char *val,
> static int zswap_enabled_param_set(const char *val,
> const struct kernel_param *kp)
> {
> + if (system_state == SYSTEM_RUNNING) {
> + mutex_lock(&zswap_init_lock);
> + if (zswap_setup()) {
> + mutex_unlock(&zswap_init_lock);
> + return -ENODEV;
> + }
> + mutex_unlock(&zswap_init_lock);
> + }
> if (zswap_init_state == ZSWAP_INIT_FAILED) {
> pr_err("can't enable, initialization failed\n");
> return -ENODEV;
> @@ -1432,7 +1450,7 @@ static const struct frontswap_ops zswap_frontswap_ops = {
>
> static struct dentry *zswap_debugfs_root;
>
> -static int __init zswap_debugfs_init(void)
> +static int zswap_debugfs_init(void)
> {
> if (!debugfs_initialized())
> return -ENODEV;
> @@ -1463,7 +1481,7 @@ static int __init zswap_debugfs_init(void)
> return 0;
> }
> #else
> -static int __init zswap_debugfs_init(void)
> +static int zswap_debugfs_init(void)
> {
> return 0;
> }
> @@ -1472,11 +1490,14 @@ static int __init zswap_debugfs_init(void)
> /*********************************
> * module init and exit
> **********************************/
> -static int __init init_zswap(void)
> +static int zswap_setup(void)
> {
> struct zswap_pool *pool;
> int ret;
>
> + if (zswap_init_state != ZSWAP_UNINIT)
> + return 0;
> +
> if (zswap_entry_cache_create()) {
> pr_err("entry cache creation failed\n");
> goto cache_fail;
> @@ -1534,8 +1555,17 @@ static int __init init_zswap(void)
> zswap_enabled = false;
> return -ENOMEM;
> }
> +
> +static int __init zswap_init(void)
> +{
> + /* skip init if zswap is disabled when system startup */
> + if (!zswap_enabled)
> + return 0;
> + return zswap_setup();
> +}
> +
> /* must be late so crypto has time to come up */
> -late_initcall(init_zswap);
> +late_initcall(zswap_init);
>
> MODULE_LICENSE("GPL");
> MODULE_AUTHOR("Seth Jennings <[email protected]>");
> --
> 2.25.1
>

2022-08-28 20:48:47

by Vitaly Wool

[permalink] [raw]
Subject: Re: [PATCH -next v3 3/5] mm/zswap: replace zswap_init_{started/failed} with zswap_init_state

On Sat, Aug 27, 2022 at 12:12 PM Liu Shixin <[email protected]> wrote:
>
> zswap_init_started indicates that the initialization is started. And
> zswap_init_failed indicates that the initialization is failed. As we will
> support to init zswap after system startup, it's necessary to add a state
> to indicate the initialization is complete and succeed to avoid
> concurrency issues. Since we don't care about the difference between
> init started with init completion. We only need three states:
> uninitialized, initial failed, initial succeed.
>
> Signed-off-by: Liu Shixin <[email protected]>

Reviewed-by: Vitaly Wool <[email protected]>

> ---
> mm/zswap.c | 22 +++++++++++-----------
> 1 file changed, 11 insertions(+), 11 deletions(-)
>
> diff --git a/mm/zswap.c b/mm/zswap.c
> index 2d48fd59cc7a..84e38300f571 100644
> --- a/mm/zswap.c
> +++ b/mm/zswap.c
> @@ -214,11 +214,12 @@ static DEFINE_SPINLOCK(zswap_pools_lock);
> /* pool counter to provide unique names to zpool */
> static atomic_t zswap_pools_count = ATOMIC_INIT(0);
>
> -/* used by param callback function */
> -static bool zswap_init_started;
> +#define ZSWAP_UNINIT 0
> +#define ZSWAP_INIT_SUCCEED 1
> +#define ZSWAP_INIT_FAILED 2
>
> -/* fatal error during init */
> -static bool zswap_init_failed;
> +/* init state */
> +static int zswap_init_state;
>
> /* init completed, but couldn't create the initial pool */
> static bool zswap_has_pool;
> @@ -772,7 +773,7 @@ static int __zswap_param_set(const char *val, const struct kernel_param *kp,
> char *s = strstrip((char *)val);
> int ret;
>
> - if (zswap_init_failed) {
> + if (zswap_init_state == ZSWAP_INIT_FAILED) {
> pr_err("can't set param, initialization failed\n");
> return -ENODEV;
> }
> @@ -784,7 +785,7 @@ static int __zswap_param_set(const char *val, const struct kernel_param *kp,
> /* if this is load-time (pre-init) param setting,
> * don't create a pool; that's done during init.
> */
> - if (!zswap_init_started)
> + if (zswap_init_state == ZSWAP_UNINIT)
> return param_set_charp(s, kp);
>
> if (!type) {
> @@ -875,11 +876,11 @@ static int zswap_zpool_param_set(const char *val,
> static int zswap_enabled_param_set(const char *val,
> const struct kernel_param *kp)
> {
> - if (zswap_init_failed) {
> + if (zswap_init_state == ZSWAP_INIT_FAILED) {
> pr_err("can't enable, initialization failed\n");
> return -ENODEV;
> }
> - if (!zswap_has_pool && zswap_init_started) {
> + if (!zswap_has_pool && zswap_init_state == ZSWAP_INIT_SUCCEED) {
> pr_err("can't enable, no pool configured\n");
> return -ENODEV;
> }
> @@ -1476,8 +1477,6 @@ static int __init init_zswap(void)
> struct zswap_pool *pool;
> int ret;
>
> - zswap_init_started = true;
> -
> if (zswap_entry_cache_create()) {
> pr_err("entry cache creation failed\n");
> goto cache_fail;
> @@ -1517,6 +1516,7 @@ static int __init init_zswap(void)
> goto destroy_wq;
> if (zswap_debugfs_init())
> pr_warn("debugfs initialization failed\n");
> + zswap_init_state = ZSWAP_INIT_SUCCEED;
> return 0;
>
> destroy_wq:
> @@ -1530,7 +1530,7 @@ static int __init init_zswap(void)
> zswap_entry_cache_destroy();
> cache_fail:
> /* if built-in, we aren't unloaded on failure; don't allow use */
> - zswap_init_failed = true;
> + zswap_init_state = ZSWAP_INIT_FAILED;
> zswap_enabled = false;
> return -ENOMEM;
> }
> --
> 2.25.1
>

2022-08-28 20:58:29

by Vitaly Wool

[permalink] [raw]
Subject: Re: [PATCH -next v3 1/5] frontswap: skip frontswap_ops init if zswap init failed.

On Sat, Aug 27, 2022 at 12:12 PM Liu Shixin <[email protected]> wrote:
>
> If zswap initial failed or has not been initial, frontswap_ops will be
> NULL. In such situation, swap device would enable failed with following
> stack trace:
>
> Unable to handle kernel access to user memory outside uaccess routines at virtual address 0000000000000000
> Mem abort info:
> ESR = 0x0000000096000004
> EC = 0x25: DABT (current EL), IL = 32 bits
> SET = 0, FnV = 0
> EA = 0, S1PTW = 0
> FSC = 0x04: level 0 translation fault
> Data abort info:
> ISV = 0, ISS = 0x00000004
> CM = 0, WnR = 0
> user pgtable: 4k pages, 48-bit VAs, pgdp=00000020a4fab000
> [0000000000000000] pgd=0000000000000000, p4d=0000000000000000
> Internal error: Oops: 96000004 [#1] SMP
> Modules linked in: zram fsl_dpaa2_eth pcs_lynx phylink ahci_qoriq crct10dif_ce ghash_ce sbsa_gwdt fsl_mc_dpio nvme lm90 nvme_core at803x xhci_plat_hcd rtc_fsl_ftm_alarm xgmac_mdio ahci_platform i2c_imx ip6_tables ip_tables fuse
> Unloaded tainted modules: cppc_cpufreq():1
> CPU: 10 PID: 761 Comm: swapon Not tainted 6.0.0-rc2-00454-g22100432cf14 #1
> Hardware name: SolidRun Ltd. SolidRun CEX7 Platform, BIOS EDK II Jun 21 2022
> pstate: 00400005 (nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> pc : frontswap_init+0x38/0x60
> lr : __do_sys_swapon+0x8a8/0x9f4
> sp : ffff80000969bcf0
> x29: ffff80000969bcf0 x28: ffff37bee0d8fc00 x27: ffff80000a7f5000
> x26: fffffcdefb971e80 x25: ffffaba797453b90 x24: 0000000000000064
> x23: ffff37c1f209d1a8 x22: ffff37bee880e000 x21: ffffaba797748560
> x20: ffff37bee0d8fce4 x19: ffffaba797748488 x18: 0000000000000014
> x17: 0000000030ec029a x16: ffffaba795a479b0 x15: 0000000000000000
> x14: 0000000000000000 x13: 0000000000000030 x12: 0000000000000001
> x11: ffff37c63c0aba18 x10: 0000000000000000 x9 : ffffaba7956b8c88
> x8 : ffff80000969bcd0 x7 : 0000000000000000 x6 : 0000000000000000
> x5 : 0000000000000001 x4 : 0000000000000000 x3 : ffffaba79730f000
> x2 : ffff37bee0d8fc00 x1 : 0000000000000000 x0 : 0000000000000000
> Call trace:
> frontswap_init+0x38/0x60
> __do_sys_swapon+0x8a8/0x9f4
> __arm64_sys_swapon+0x28/0x3c
> invoke_syscall+0x78/0x100
> el0_svc_common.constprop.0+0xd4/0xf4
> do_el0_svc+0x38/0x4c
> el0_svc+0x34/0x10c
> el0t_64_sync_handler+0x11c/0x150
> el0t_64_sync+0x190/0x194
> Code: d000e283 910003fd f9006c41 f946d461 (f9400021)
> ---[ end trace 0000000000000000 ]---
>

Well, this issue you are seeing is in fact introduced by the following patch:

author Christoph Hellwig <[email protected]> 2022-01-21 22:15:10 -0800
frontswap: remove support for multiple ops

So I would rather see that one reverted and fixed.

Thanks,
Vitaly

> Reported-by: Nathan Chancellor <[email protected]>
> Signed-off-by: Liu Shixin <[email protected]>
> ---
> mm/frontswap.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/mm/frontswap.c b/mm/frontswap.c
> index 1a97610308cb..620f95af81dd 100644
> --- a/mm/frontswap.c
> +++ b/mm/frontswap.c
> @@ -125,7 +125,8 @@ void frontswap_init(unsigned type, unsigned long *map)
> * p->frontswap set to something valid to work properly.
> */
> frontswap_map_set(sis, map);
> - frontswap_ops->init(type);
> + if (frontswap_ops)
> + frontswap_ops->init(type);
> }
>
> static bool __frontswap_test(struct swap_info_struct *sis,
> --
> 2.25.1
>

2022-08-28 20:58:29

by Vitaly Wool

[permalink] [raw]
Subject: Re: [PATCH -next v3 2/5] frontswap: invoke ops->init for online swap device in frontswap_register_ops

On Sat, Aug 27, 2022 at 12:12 PM Liu Shixin <[email protected]> wrote:
>
> Since we are supported to delay zswap initializaton, we need to invoke
> ops->init for the swap device which is already online when register
> backend.
>
> This patch is a revert of f328c1d16e4c ("frontswap: simplify frontswap_register_ops")
> and 633423a09cb5 ("mm: mark swap_lock and swap_active_head static")
>
> Signed-off-by: Liu Shixin <[email protected]>

Sorry, is this the revert of 2 patches at the same time? I would
rather not do it like that.

Thanks,
Vitaly

> ---
> include/linux/swapfile.h | 2 ++
> mm/frontswap.c | 47 ++++++++++++++++++++++++++++++++++++++++
> mm/swapfile.c | 4 ++--
> 3 files changed, 51 insertions(+), 2 deletions(-)
>
> diff --git a/include/linux/swapfile.h b/include/linux/swapfile.h
> index 2fbcc9afd814..75fc069594a5 100644
> --- a/include/linux/swapfile.h
> +++ b/include/linux/swapfile.h
> @@ -6,6 +6,8 @@
> * these were static in swapfile.c but frontswap.c needs them and we don't
> * want to expose them to the dozens of source files that include swap.h
> */
> +extern spinlock_t swap_lock;
> +extern struct plist_head swap_active_head;
> extern struct swap_info_struct *swap_info[];
> extern unsigned long generic_max_swapfile_size(void);
> /* Maximum swapfile size supported for the arch (not inclusive). */
> diff --git a/mm/frontswap.c b/mm/frontswap.c
> index 620f95af81dd..449e6f499b88 100644
> --- a/mm/frontswap.c
> +++ b/mm/frontswap.c
> @@ -96,11 +96,58 @@ static inline void inc_frontswap_invalidates(void) { }
> */
> int frontswap_register_ops(const struct frontswap_ops *ops)
> {
> + DECLARE_BITMAP(a, MAX_SWAPFILES);
> + DECLARE_BITMAP(b, MAX_SWAPFILES);
> + struct swap_info_struct *si;
> + unsigned int i;
> +
> if (frontswap_ops)
> return -EINVAL;
>
> + bitmap_zero(a, MAX_SWAPFILES);
> + bitmap_zero(b, MAX_SWAPFILES);
> +
> + spin_lock(&swap_lock);
> + plist_for_each_entry(si, &swap_active_head, list) {
> + if (!WARN_ON(!si->frontswap_map))
> + __set_bit(si->type, a);
> + }
> + spin_unlock(&swap_lock);
> +
> + /* the new ops needs to know the currently active swap devices */
> + for_each_set_bit(i, a, MAX_SWAPFILES) {
> + pr_err("init frontswap_ops\n");
> + ops->init(i);
> + }
> +
> frontswap_ops = ops;
> static_branch_inc(&frontswap_enabled_key);
> +
> + spin_lock(&swap_lock);
> + plist_for_each_entry(si, &swap_active_head, list) {
> + if (si->frontswap_map)
> + __set_bit(si->type, b);
> + }
> + spin_unlock(&swap_lock);
> +
> + /*
> + * On the very unlikely chance that a swap device was added or
> + * removed between setting the "a" list bits and the ops init
> + * calls, we re-check and do init or invalidate for any changed
> + * bits.
> + */
> + if (unlikely(!bitmap_equal(a, b, MAX_SWAPFILES))) {
> + for (i = 0; i < MAX_SWAPFILES; i++) {
> + if (!test_bit(i, a) && test_bit(i, b)) {
> + pr_err("init frontswap_ops re\n");
> + ops->init(i);
> + } else if (test_bit(i, a) && !test_bit(i, b)) {
> + pr_err("inval frontswap_ops re\n");
> + ops->invalidate_area(i);
> + }
> + }
> + }
> +
> return 0;
> }
>
> diff --git a/mm/swapfile.c b/mm/swapfile.c
> index 469d9af86be2..d383b282f269 100644
> --- a/mm/swapfile.c
> +++ b/mm/swapfile.c
> @@ -51,7 +51,7 @@ static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
> unsigned char);
> static void free_swap_count_continuations(struct swap_info_struct *);
>
> -static DEFINE_SPINLOCK(swap_lock);
> +DEFINE_SPINLOCK(swap_lock);
> static unsigned int nr_swapfiles;
> atomic_long_t nr_swap_pages;
> /*
> @@ -77,7 +77,7 @@ static const char Unused_offset[] = "Unused swap offset entry ";
> * all active swap_info_structs
> * protected with swap_lock, and ordered by priority.
> */
> -static PLIST_HEAD(swap_active_head);
> +PLIST_HEAD(swap_active_head);
>
> /*
> * all available (active, not full) swap_info_structs
> --
> 2.25.1
>

2022-08-29 02:14:41

by Liu Shixin

[permalink] [raw]
Subject: Re: [PATCH -next v3 2/5] frontswap: invoke ops->init for online swap device in frontswap_register_ops



On 2022/8/29 4:47, Vitaly Wool wrote:
> On Sat, Aug 27, 2022 at 12:12 PM Liu Shixin <[email protected]> wrote:
>> Since we are supported to delay zswap initializaton, we need to invoke
>> ops->init for the swap device which is already online when register
>> backend.
>>
>> This patch is a revert of f328c1d16e4c ("frontswap: simplify frontswap_register_ops")
>> and 633423a09cb5 ("mm: mark swap_lock and swap_active_head static")
>>
>> Signed-off-by: Liu Shixin <[email protected]>
> Sorry, is this the revert of 2 patches at the same time? I would
> rather not do it like that.
>
> Thanks,
> Vitaly
Thanks for your advice, I'll split it to two patches and resend it.
>> ---
>> include/linux/swapfile.h | 2 ++
>> mm/frontswap.c | 47 ++++++++++++++++++++++++++++++++++++++++
>> mm/swapfile.c | 4 ++--
>> 3 files changed, 51 insertions(+), 2 deletions(-)
>>
>> diff --git a/include/linux/swapfile.h b/include/linux/swapfile.h
>> index 2fbcc9afd814..75fc069594a5 100644
>> --- a/include/linux/swapfile.h
>> +++ b/include/linux/swapfile.h
>> @@ -6,6 +6,8 @@
>> * these were static in swapfile.c but frontswap.c needs them and we don't
>> * want to expose them to the dozens of source files that include swap.h
>> */
>> +extern spinlock_t swap_lock;
>> +extern struct plist_head swap_active_head;
>> extern struct swap_info_struct *swap_info[];
>> extern unsigned long generic_max_swapfile_size(void);
>> /* Maximum swapfile size supported for the arch (not inclusive). */
>> diff --git a/mm/frontswap.c b/mm/frontswap.c
>> index 620f95af81dd..449e6f499b88 100644
>> --- a/mm/frontswap.c
>> +++ b/mm/frontswap.c
>> @@ -96,11 +96,58 @@ static inline void inc_frontswap_invalidates(void) { }
>> */
>> int frontswap_register_ops(const struct frontswap_ops *ops)
>> {
>> + DECLARE_BITMAP(a, MAX_SWAPFILES);
>> + DECLARE_BITMAP(b, MAX_SWAPFILES);
>> + struct swap_info_struct *si;
>> + unsigned int i;
>> +
>> if (frontswap_ops)
>> return -EINVAL;
>>
>> + bitmap_zero(a, MAX_SWAPFILES);
>> + bitmap_zero(b, MAX_SWAPFILES);
>> +
>> + spin_lock(&swap_lock);
>> + plist_for_each_entry(si, &swap_active_head, list) {
>> + if (!WARN_ON(!si->frontswap_map))
>> + __set_bit(si->type, a);
>> + }
>> + spin_unlock(&swap_lock);
>> +
>> + /* the new ops needs to know the currently active swap devices */
>> + for_each_set_bit(i, a, MAX_SWAPFILES) {
>> + pr_err("init frontswap_ops\n");
>> + ops->init(i);
>> + }
>> +
>> frontswap_ops = ops;
>> static_branch_inc(&frontswap_enabled_key);
>> +
>> + spin_lock(&swap_lock);
>> + plist_for_each_entry(si, &swap_active_head, list) {
>> + if (si->frontswap_map)
>> + __set_bit(si->type, b);
>> + }
>> + spin_unlock(&swap_lock);
>> +
>> + /*
>> + * On the very unlikely chance that a swap device was added or
>> + * removed between setting the "a" list bits and the ops init
>> + * calls, we re-check and do init or invalidate for any changed
>> + * bits.
>> + */
>> + if (unlikely(!bitmap_equal(a, b, MAX_SWAPFILES))) {
>> + for (i = 0; i < MAX_SWAPFILES; i++) {
>> + if (!test_bit(i, a) && test_bit(i, b)) {
>> + pr_err("init frontswap_ops re\n");
>> + ops->init(i);
>> + } else if (test_bit(i, a) && !test_bit(i, b)) {
>> + pr_err("inval frontswap_ops re\n");
>> + ops->invalidate_area(i);
>> + }
>> + }
>> + }
>> +
>> return 0;
>> }
>>
>> diff --git a/mm/swapfile.c b/mm/swapfile.c
>> index 469d9af86be2..d383b282f269 100644
>> --- a/mm/swapfile.c
>> +++ b/mm/swapfile.c
>> @@ -51,7 +51,7 @@ static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
>> unsigned char);
>> static void free_swap_count_continuations(struct swap_info_struct *);
>>
>> -static DEFINE_SPINLOCK(swap_lock);
>> +DEFINE_SPINLOCK(swap_lock);
>> static unsigned int nr_swapfiles;
>> atomic_long_t nr_swap_pages;
>> /*
>> @@ -77,7 +77,7 @@ static const char Unused_offset[] = "Unused swap offset entry ";
>> * all active swap_info_structs
>> * protected with swap_lock, and ordered by priority.
>> */
>> -static PLIST_HEAD(swap_active_head);
>> +PLIST_HEAD(swap_active_head);
>>
>> /*
>> * all available (active, not full) swap_info_structs
>> --
>> 2.25.1
>>
> .
>

2022-08-29 02:33:46

by Liu Shixin

[permalink] [raw]
Subject: Re: [PATCH -next v3 1/5] frontswap: skip frontswap_ops init if zswap init failed.

On 2022/8/29 4:44, Vitaly Wool wrote:
> On Sat, Aug 27, 2022 at 12:12 PM Liu Shixin <[email protected]> wrote:
>> If zswap initial failed or has not been initial, frontswap_ops will be
>> NULL. In such situation, swap device would enable failed with following
>> stack trace:
>>
>> Unable to handle kernel access to user memory outside uaccess routines at virtual address 0000000000000000
>> Mem abort info:
>> ESR = 0x0000000096000004
>> EC = 0x25: DABT (current EL), IL = 32 bits
>> SET = 0, FnV = 0
>> EA = 0, S1PTW = 0
>> FSC = 0x04: level 0 translation fault
>> Data abort info:
>> ISV = 0, ISS = 0x00000004
>> CM = 0, WnR = 0
>> user pgtable: 4k pages, 48-bit VAs, pgdp=00000020a4fab000
>> [0000000000000000] pgd=0000000000000000, p4d=0000000000000000
>> Internal error: Oops: 96000004 [#1] SMP
>> Modules linked in: zram fsl_dpaa2_eth pcs_lynx phylink ahci_qoriq crct10dif_ce ghash_ce sbsa_gwdt fsl_mc_dpio nvme lm90 nvme_core at803x xhci_plat_hcd rtc_fsl_ftm_alarm xgmac_mdio ahci_platform i2c_imx ip6_tables ip_tables fuse
>> Unloaded tainted modules: cppc_cpufreq():1
>> CPU: 10 PID: 761 Comm: swapon Not tainted 6.0.0-rc2-00454-g22100432cf14 #1
>> Hardware name: SolidRun Ltd. SolidRun CEX7 Platform, BIOS EDK II Jun 21 2022
>> pstate: 00400005 (nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
>> pc : frontswap_init+0x38/0x60
>> lr : __do_sys_swapon+0x8a8/0x9f4
>> sp : ffff80000969bcf0
>> x29: ffff80000969bcf0 x28: ffff37bee0d8fc00 x27: ffff80000a7f5000
>> x26: fffffcdefb971e80 x25: ffffaba797453b90 x24: 0000000000000064
>> x23: ffff37c1f209d1a8 x22: ffff37bee880e000 x21: ffffaba797748560
>> x20: ffff37bee0d8fce4 x19: ffffaba797748488 x18: 0000000000000014
>> x17: 0000000030ec029a x16: ffffaba795a479b0 x15: 0000000000000000
>> x14: 0000000000000000 x13: 0000000000000030 x12: 0000000000000001
>> x11: ffff37c63c0aba18 x10: 0000000000000000 x9 : ffffaba7956b8c88
>> x8 : ffff80000969bcd0 x7 : 0000000000000000 x6 : 0000000000000000
>> x5 : 0000000000000001 x4 : 0000000000000000 x3 : ffffaba79730f000
>> x2 : ffff37bee0d8fc00 x1 : 0000000000000000 x0 : 0000000000000000
>> Call trace:
>> frontswap_init+0x38/0x60
>> __do_sys_swapon+0x8a8/0x9f4
>> __arm64_sys_swapon+0x28/0x3c
>> invoke_syscall+0x78/0x100
>> el0_svc_common.constprop.0+0xd4/0xf4
>> do_el0_svc+0x38/0x4c
>> el0_svc+0x34/0x10c
>> el0t_64_sync_handler+0x11c/0x150
>> el0t_64_sync+0x190/0x194
>> Code: d000e283 910003fd f9006c41 f946d461 (f9400021)
>> ---[ end trace 0000000000000000 ]---
>>
> Well, this issue you are seeing is in fact introduced by the following patch:
>
> author Christoph Hellwig <[email protected]> 2022-01-21 22:15:10 -0800
> frontswap: remove support for multiple ops
>
> So I would rather see that one reverted and fixed.
>
> Thanks,
> Vitaly
It is surely introduced by the previous patch ,but is it need to revert that patch? Do we have
any plans to add new backend in the future?

Thanks,
>
>> Reported-by: Nathan Chancellor <[email protected]>
>> Signed-off-by: Liu Shixin <[email protected]>
>> ---
>> mm/frontswap.c | 3 ++-
>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/mm/frontswap.c b/mm/frontswap.c
>> index 1a97610308cb..620f95af81dd 100644
>> --- a/mm/frontswap.c
>> +++ b/mm/frontswap.c
>> @@ -125,7 +125,8 @@ void frontswap_init(unsigned type, unsigned long *map)
>> * p->frontswap set to something valid to work properly.
>> */
>> frontswap_map_set(sis, map);
>> - frontswap_ops->init(type);
>> + if (frontswap_ops)
>> + frontswap_ops->init(type);
>> }
>>
>> static bool __frontswap_test(struct swap_info_struct *sis,
>> --
>> 2.25.1
>>
> .
>

2022-08-29 06:28:43

by Vitaly Wool

[permalink] [raw]
Subject: Re: [PATCH -next v3 1/5] frontswap: skip frontswap_ops init if zswap init failed.

On Mon, Aug 29, 2022 at 4:11 AM Liu Shixin <[email protected]> wrote:
>
> On 2022/8/29 4:44, Vitaly Wool wrote:
> > On Sat, Aug 27, 2022 at 12:12 PM Liu Shixin <[email protected]> wrote:
> >> If zswap initial failed or has not been initial, frontswap_ops will be
> >> NULL. In such situation, swap device would enable failed with following
> >> stack trace:
> >>
> >> Unable to handle kernel access to user memory outside uaccess routines at virtual address 0000000000000000
> >> Mem abort info:
> >> ESR = 0x0000000096000004
> >> EC = 0x25: DABT (current EL), IL = 32 bits
> >> SET = 0, FnV = 0
> >> EA = 0, S1PTW = 0
> >> FSC = 0x04: level 0 translation fault
> >> Data abort info:
> >> ISV = 0, ISS = 0x00000004
> >> CM = 0, WnR = 0
> >> user pgtable: 4k pages, 48-bit VAs, pgdp=00000020a4fab000
> >> [0000000000000000] pgd=0000000000000000, p4d=0000000000000000
> >> Internal error: Oops: 96000004 [#1] SMP
> >> Modules linked in: zram fsl_dpaa2_eth pcs_lynx phylink ahci_qoriq crct10dif_ce ghash_ce sbsa_gwdt fsl_mc_dpio nvme lm90 nvme_core at803x xhci_plat_hcd rtc_fsl_ftm_alarm xgmac_mdio ahci_platform i2c_imx ip6_tables ip_tables fuse
> >> Unloaded tainted modules: cppc_cpufreq():1
> >> CPU: 10 PID: 761 Comm: swapon Not tainted 6.0.0-rc2-00454-g22100432cf14 #1
> >> Hardware name: SolidRun Ltd. SolidRun CEX7 Platform, BIOS EDK II Jun 21 2022
> >> pstate: 00400005 (nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> >> pc : frontswap_init+0x38/0x60
> >> lr : __do_sys_swapon+0x8a8/0x9f4
> >> sp : ffff80000969bcf0
> >> x29: ffff80000969bcf0 x28: ffff37bee0d8fc00 x27: ffff80000a7f5000
> >> x26: fffffcdefb971e80 x25: ffffaba797453b90 x24: 0000000000000064
> >> x23: ffff37c1f209d1a8 x22: ffff37bee880e000 x21: ffffaba797748560
> >> x20: ffff37bee0d8fce4 x19: ffffaba797748488 x18: 0000000000000014
> >> x17: 0000000030ec029a x16: ffffaba795a479b0 x15: 0000000000000000
> >> x14: 0000000000000000 x13: 0000000000000030 x12: 0000000000000001
> >> x11: ffff37c63c0aba18 x10: 0000000000000000 x9 : ffffaba7956b8c88
> >> x8 : ffff80000969bcd0 x7 : 0000000000000000 x6 : 0000000000000000
> >> x5 : 0000000000000001 x4 : 0000000000000000 x3 : ffffaba79730f000
> >> x2 : ffff37bee0d8fc00 x1 : 0000000000000000 x0 : 0000000000000000
> >> Call trace:
> >> frontswap_init+0x38/0x60
> >> __do_sys_swapon+0x8a8/0x9f4
> >> __arm64_sys_swapon+0x28/0x3c
> >> invoke_syscall+0x78/0x100
> >> el0_svc_common.constprop.0+0xd4/0xf4
> >> do_el0_svc+0x38/0x4c
> >> el0_svc+0x34/0x10c
> >> el0t_64_sync_handler+0x11c/0x150
> >> el0t_64_sync+0x190/0x194
> >> Code: d000e283 910003fd f9006c41 f946d461 (f9400021)
> >> ---[ end trace 0000000000000000 ]---
> >>
> > Well, this issue you are seeing is in fact introduced by the following patch:
> >
> > author Christoph Hellwig <[email protected]> 2022-01-21 22:15:10 -0800
> > frontswap: remove support for multiple ops
> >
> > So I would rather see that one reverted and fixed.
> >
> > Thanks,
> > Vitaly
> It is surely introduced by the previous patch ,but is it need to revert that patch? Do we have
> any plans to add new backend in the future?

I believe we do. Besides, this patch introduces the bug you have hit,
before this patch frontswap just wouldn't go doing anything on an
empty list.
It's my bad I didn't NAK that patch then, but we have an opportunity
to do it the right way now.

Thanks,
Vitaly

> Thanks,
> >
> >> Reported-by: Nathan Chancellor <[email protected]>
> >> Signed-off-by: Liu Shixin <[email protected]>
> >> ---
> >> mm/frontswap.c | 3 ++-
> >> 1 file changed, 2 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/mm/frontswap.c b/mm/frontswap.c
> >> index 1a97610308cb..620f95af81dd 100644
> >> --- a/mm/frontswap.c
> >> +++ b/mm/frontswap.c
> >> @@ -125,7 +125,8 @@ void frontswap_init(unsigned type, unsigned long *map)
> >> * p->frontswap set to something valid to work properly.
> >> */
> >> frontswap_map_set(sis, map);
> >> - frontswap_ops->init(type);
> >> + if (frontswap_ops)
> >> + frontswap_ops->init(type);
> >> }
> >>
> >> static bool __frontswap_test(struct swap_info_struct *sis,
> >> --
> >> 2.25.1
> >>
> > .
> >
>