]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
zram: use u32 for entry ac_time tracking
authorSergey Senozhatsky <senozhatsky@chromium.org>
Mon, 15 Dec 2025 05:47:11 +0000 (14:47 +0900)
committerAndrew Morton <akpm@linux-foundation.org>
Wed, 21 Jan 2026 03:24:38 +0000 (19:24 -0800)
We can reduce sizeof(zram_table_entry) on 64-bit systems by converting
flags and ac_time to u32.  Entry flags fit into u32, and for ac_time u32
gives us over a century of entry lifespan (approx 136 years) which is
plenty (zram uses system boot time (seconds)).

In struct zram_table_entry we use bytes aliasing, because bit-wait API
(for slot lock) requires a whole unsigned long word.

Link: https://lkml.kernel.org/r/d7c0b48450c70eeb5fd8acd6ecd23593f30dbf1f.1765775954.git.senozhatsky@chromium.org
Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Suggested-by: David Stevens <stevensd@google.com>
Cc: Brian Geffon <bgeffon@google.com>
Cc: Minchan Kim <minchan@google.com>
Cc: Richard Chang <richardycc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
drivers/block/zram/zram_drv.c
drivers/block/zram/zram_drv.h

index 67a9e7c005c3339bcce223251d4dbec6892090b1..65f99ff3e2e54d3ffb6774b3332c431c69a46970 100644 (file)
@@ -81,7 +81,7 @@ static void zram_slot_lock_init(struct zram *zram, u32 index)
  */
 static __must_check bool zram_slot_trylock(struct zram *zram, u32 index)
 {
-       unsigned long *lock = &zram->table[index].flags;
+       unsigned long *lock = &zram->table[index].__lock;
 
        if (!test_and_set_bit_lock(ZRAM_ENTRY_LOCK, lock)) {
                mutex_acquire(slot_dep_map(zram, index), 0, 1, _RET_IP_);
@@ -94,7 +94,7 @@ static __must_check bool zram_slot_trylock(struct zram *zram, u32 index)
 
 static void zram_slot_lock(struct zram *zram, u32 index)
 {
-       unsigned long *lock = &zram->table[index].flags;
+       unsigned long *lock = &zram->table[index].__lock;
 
        mutex_acquire(slot_dep_map(zram, index), 0, 0, _RET_IP_);
        wait_on_bit_lock(lock, ZRAM_ENTRY_LOCK, TASK_UNINTERRUPTIBLE);
@@ -103,7 +103,7 @@ static void zram_slot_lock(struct zram *zram, u32 index)
 
 static void zram_slot_unlock(struct zram *zram, u32 index)
 {
-       unsigned long *lock = &zram->table[index].flags;
+       unsigned long *lock = &zram->table[index].__lock;
 
        mutex_release(slot_dep_map(zram, index), _RET_IP_);
        clear_and_wake_up_bit(ZRAM_ENTRY_LOCK, lock);
@@ -130,34 +130,33 @@ static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
 }
 
 static bool zram_test_flag(struct zram *zram, u32 index,
-                       enum zram_pageflags flag)
+                          enum zram_pageflags flag)
 {
-       return zram->table[index].flags & BIT(flag);
+       return zram->table[index].attr.flags & BIT(flag);
 }
 
 static void zram_set_flag(struct zram *zram, u32 index,
-                       enum zram_pageflags flag)
+                         enum zram_pageflags flag)
 {
-       zram->table[index].flags |= BIT(flag);
+       zram->table[index].attr.flags |= BIT(flag);
 }
 
 static void zram_clear_flag(struct zram *zram, u32 index,
-                       enum zram_pageflags flag)
+                           enum zram_pageflags flag)
 {
-       zram->table[index].flags &= ~BIT(flag);
+       zram->table[index].attr.flags &= ~BIT(flag);
 }
 
 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 {
-       return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
+       return zram->table[index].attr.flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
 }
 
-static void zram_set_obj_size(struct zram *zram,
-                                       u32 index, size_t size)
+static void zram_set_obj_size(struct zram *zram, u32 index, size_t size)
 {
-       unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
+       unsigned long flags = zram->table[index].attr.flags >> ZRAM_FLAG_SHIFT;
 
-       zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
+       zram->table[index].attr.flags = (flags << ZRAM_FLAG_SHIFT) | size;
 }
 
 static inline bool zram_allocated(struct zram *zram, u32 index)
@@ -208,14 +207,14 @@ static inline void zram_set_priority(struct zram *zram, u32 index, u32 prio)
         * Clear previous priority value first, in case if we recompress
         * further an already recompressed page
         */
-       zram->table[index].flags &= ~(ZRAM_COMP_PRIORITY_MASK <<
-                                     ZRAM_COMP_PRIORITY_BIT1);
-       zram->table[index].flags |= (prio << ZRAM_COMP_PRIORITY_BIT1);
+       zram->table[index].attr.flags &= ~(ZRAM_COMP_PRIORITY_MASK <<
+                                          ZRAM_COMP_PRIORITY_BIT1);
+       zram->table[index].attr.flags |= (prio << ZRAM_COMP_PRIORITY_BIT1);
 }
 
 static inline u32 zram_get_priority(struct zram *zram, u32 index)
 {
-       u32 prio = zram->table[index].flags >> ZRAM_COMP_PRIORITY_BIT1;
+       u32 prio = zram->table[index].attr.flags >> ZRAM_COMP_PRIORITY_BIT1;
 
        return prio & ZRAM_COMP_PRIORITY_MASK;
 }
@@ -225,7 +224,7 @@ static void zram_accessed(struct zram *zram, u32 index)
        zram_clear_flag(zram, index, ZRAM_IDLE);
        zram_clear_flag(zram, index, ZRAM_PP_SLOT);
 #ifdef CONFIG_ZRAM_TRACK_ENTRY_ACTIME
-       zram->table[index].ac_time = ktime_get_boottime();
+       zram->table[index].attr.ac_time = (u32)ktime_get_boottime_seconds();
 #endif
 }
 
@@ -447,7 +446,7 @@ static void mark_idle(struct zram *zram, ktime_t cutoff)
 
 #ifdef CONFIG_ZRAM_TRACK_ENTRY_ACTIME
                is_idle = !cutoff ||
-                       ktime_after(cutoff, zram->table[index].ac_time);
+                       ktime_after(cutoff, zram->table[index].attr.ac_time);
 #endif
                if (is_idle)
                        zram_set_flag(zram, index, ZRAM_IDLE);
@@ -461,18 +460,19 @@ static ssize_t idle_store(struct device *dev, struct device_attribute *attr,
                          const char *buf, size_t len)
 {
        struct zram *zram = dev_to_zram(dev);
-       ktime_t cutoff_time = 0;
+       ktime_t cutoff = 0;
 
        if (!sysfs_streq(buf, "all")) {
                /*
                 * If it did not parse as 'all' try to treat it as an integer
                 * when we have memory tracking enabled.
                 */
-               u64 age_sec;
+               u32 age_sec;
 
-               if (IS_ENABLED(CONFIG_ZRAM_TRACK_ENTRY_ACTIME) && !kstrtoull(buf, 0, &age_sec))
-                       cutoff_time = ktime_sub(ktime_get_boottime(),
-                                       ns_to_ktime(age_sec * NSEC_PER_SEC));
+               if (IS_ENABLED(CONFIG_ZRAM_TRACK_ENTRY_ACTIME) &&
+                   !kstrtouint(buf, 0, &age_sec))
+                       cutoff = ktime_sub((u32)ktime_get_boottime_seconds(),
+                                          age_sec);
                else
                        return -EINVAL;
        }
@@ -482,10 +482,10 @@ static ssize_t idle_store(struct device *dev, struct device_attribute *attr,
                return -EINVAL;
 
        /*
-        * A cutoff_time of 0 marks everything as idle, this is the
+        * A cutoff of 0 marks everything as idle, this is the
         * "all" behavior.
         */
-       mark_idle(zram, cutoff_time);
+       mark_idle(zram, cutoff);
        return len;
 }
 
@@ -1588,7 +1588,7 @@ static ssize_t read_block_state(struct file *file, char __user *buf,
                if (!zram_allocated(zram, index))
                        goto next;
 
-               ts = ktime_to_timespec64(zram->table[index].ac_time);
+               ts = ktime_to_timespec64(zram->table[index].attr.ac_time);
                copied = snprintf(kbuf + written, count,
                        "%12zd %12lld.%06lu %c%c%c%c%c%c\n",
                        index, (s64)ts.tv_sec,
@@ -2013,7 +2013,7 @@ static void zram_slot_free(struct zram *zram, u32 index)
        unsigned long handle;
 
 #ifdef CONFIG_ZRAM_TRACK_ENTRY_ACTIME
-       zram->table[index].ac_time = 0;
+       zram->table[index].attr.ac_time = 0;
 #endif
 
        zram_clear_flag(zram, index, ZRAM_IDLE);
@@ -3286,7 +3286,7 @@ static int __init zram_init(void)
        struct zram_table_entry zram_te;
        int ret;
 
-       BUILD_BUG_ON(__NR_ZRAM_PAGEFLAGS > sizeof(zram_te.flags) * 8);
+       BUILD_BUG_ON(__NR_ZRAM_PAGEFLAGS > sizeof(zram_te.attr.flags) * 8);
 
        ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
                                      zcomp_cpu_up_prepare, zcomp_cpu_dead);
index 72fdf66c78ab577c8b2c50e7564bd86cac0a6093..48d6861c664701ff59a78b46169a672aa06679ed 100644 (file)
@@ -65,10 +65,15 @@ enum zram_pageflags {
  */
 struct zram_table_entry {
        unsigned long handle;
-       unsigned long flags;
+       union {
+               unsigned long __lock;
+               struct attr {
+                       u32 flags;
 #ifdef CONFIG_ZRAM_TRACK_ENTRY_ACTIME
-       ktime_t ac_time;
+                       u32 ac_time;
 #endif
+               } attr;
+       };
        struct lockdep_map dep_map;
 };