diff mbox series

[RFC,V2] zram:calculate available memory when zram is used

Message ID 1622910240-4621-1-git-send-email-yongw.pur@gmail.com (mailing list archive)
State New, archived
Headers show
Series [RFC,V2] zram:calculate available memory when zram is used | expand

Commit Message

yong w June 5, 2021, 4:24 p.m. UTC
From: wangyong <yongw.pur@gmail.com>

When zram is used, available+Swap free memory is obviously
bigger than we actually can use, because zram can compress
memory by compression algorithm and zram compressed data
will occupy memory too.

So, we can count the compression ratio of zram in the kernel.
The space will be saved by zram and other swap device are
calculated as follows:
zram[swapfree - swapfree * compress ratio] + swapdev[swapfree]
We can evaluate the available memory of the whole system as:
MemAvailable+zram[swapfree - swapfree * compress ratio]+swapdev[swapfree]

Add an entry to the /proc/meminfo file, returns swap will save space.
Which name is more appropriate is still under consideration.
There are several alternative names: SwapAvailable, SwapSaved,
SwapCompressible

Signed-off-by: wangyong <yongw.pur@gmail.com>
---
 drivers/block/zram/zcomp.h    |  1 +
 drivers/block/zram/zram_drv.c | 19 +++++++++
 drivers/block/zram/zram_drv.h |  1 +
 fs/proc/meminfo.c             |  1 +
 include/linux/swap.h          | 10 +++++
 mm/swapfile.c                 | 95 +++++++++++++++++++++++++++++++++++++++++++
 mm/vmscan.c                   |  1 +
 7 files changed, 128 insertions(+)

Comments

Greg KH June 5, 2021, 5:07 p.m. UTC | #1
On Sat, Jun 05, 2021 at 09:24:00AM -0700, yongw.pur@gmail.com wrote:
> From: wangyong <yongw.pur@gmail.com>
> 
> When zram is used, available+Swap free memory is obviously
> bigger than we actually can use, because zram can compress
> memory by compression algorithm and zram compressed data
> will occupy memory too.
> 
> So, we can count the compression ratio of zram in the kernel.
> The space will be saved by zram and other swap device are
> calculated as follows:
> zram[swapfree - swapfree * compress ratio] + swapdev[swapfree]
> We can evaluate the available memory of the whole system as:
> MemAvailable+zram[swapfree - swapfree * compress ratio]+swapdev[swapfree]
> 
> Add an entry to the /proc/meminfo file, returns swap will save space.
> Which name is more appropriate is still under consideration.
> There are several alternative names: SwapAvailable, SwapSaved,
> SwapCompressible
> 
> Signed-off-by: wangyong <yongw.pur@gmail.com>
> ---
>  drivers/block/zram/zcomp.h    |  1 +
>  drivers/block/zram/zram_drv.c | 19 +++++++++
>  drivers/block/zram/zram_drv.h |  1 +
>  fs/proc/meminfo.c             |  1 +
>  include/linux/swap.h          | 10 +++++
>  mm/swapfile.c                 | 95 +++++++++++++++++++++++++++++++++++++++++++
>  mm/vmscan.c                   |  1 +
>  7 files changed, 128 insertions(+)

You are adding a new sysfs file with no new Documentation/ABI entry.
Also are you sure you are allowed to add a new proc file entry without
breaking existing tools?

thanks,

greg k-h
David Hildenbrand June 6, 2021, 10:45 a.m. UTC | #2
On 05.06.21 19:07, Greg KH wrote:
> On Sat, Jun 05, 2021 at 09:24:00AM -0700, yongw.pur@gmail.com wrote:
>> From: wangyong <yongw.pur@gmail.com>
>>
>> When zram is used, available+Swap free memory is obviously
>> bigger than we actually can use, because zram can compress
>> memory by compression algorithm and zram compressed data
>> will occupy memory too.
>>
>> So, we can count the compression ratio of zram in the kernel.
>> The space will be saved by zram and other swap device are
>> calculated as follows:
>> zram[swapfree - swapfree * compress ratio] + swapdev[swapfree]
>> We can evaluate the available memory of the whole system as:
>> MemAvailable+zram[swapfree - swapfree * compress ratio]+swapdev[swapfree]
>>
>> Add an entry to the /proc/meminfo file, returns swap will save space.
>> Which name is more appropriate is still under consideration.
>> There are several alternative names: SwapAvailable, SwapSaved,
>> SwapCompressible
>>
>> Signed-off-by: wangyong <yongw.pur@gmail.com>
>> ---
>>   drivers/block/zram/zcomp.h    |  1 +
>>   drivers/block/zram/zram_drv.c | 19 +++++++++
>>   drivers/block/zram/zram_drv.h |  1 +
>>   fs/proc/meminfo.c             |  1 +
>>   include/linux/swap.h          | 10 +++++
>>   mm/swapfile.c                 | 95 +++++++++++++++++++++++++++++++++++++++++++
>>   mm/vmscan.c                   |  1 +
>>   7 files changed, 128 insertions(+)
> 
> You are adding a new sysfs file with no new Documentation/ABI entry.

Indeed. And in the best case, the change would be accompanied with a man 
page update patch for the "proc" man page.

> Also are you sure you are allowed to add a new proc file entry without
> breaking existing tools?

I think adding new entries is not a problem. Parsers have to deal with 
new entries popping up. For example, we introduced "ShadowCallStack:" in 
2020 via 628d06a48f57 ("scs: Add page accounting for shadow call stack 
allocations").
diff mbox series

Patch

diff --git a/drivers/block/zram/zcomp.h b/drivers/block/zram/zcomp.h
index 40f6420..deb2dbf 100644
--- a/drivers/block/zram/zcomp.h
+++ b/drivers/block/zram/zcomp.h
@@ -40,4 +40,5 @@  int zcomp_decompress(struct zcomp_strm *zstrm,
 		const void *src, unsigned int src_len, void *dst);
 
 bool zcomp_set_max_streams(struct zcomp *comp, int num_strm);
+int get_zram_major(void);
 #endif /* _ZCOMP_H_ */
diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index fcaf275..8f527e0 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c
@@ -59,6 +59,10 @@  static void zram_free_page(struct zram *zram, size_t index);
 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 				u32 index, int offset, struct bio *bio);
 
+int get_zram_major(void)
+{
+	return zram_major;
+}
 
 static int zram_slot_trylock(struct zram *zram, u32 index)
 {
@@ -1040,6 +1044,19 @@  static ssize_t compact_store(struct device *dev,
 	return len;
 }
 
+static ssize_t min_compr_ratio_show(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	struct zram *zram = dev_to_zram(dev);
+	ssize_t ret;
+
+	down_read(&zram->init_lock);
+	ret = scnprintf(buf, PAGE_SIZE, "%d\n", atomic_read(&zram->stats.min_compr_ratio));
+	up_read(&zram->init_lock);
+
+	return ret;
+}
+
 static ssize_t io_stat_show(struct device *dev,
 		struct device_attribute *attr, char *buf)
 {
@@ -1132,6 +1149,7 @@  static ssize_t debug_stat_show(struct device *dev,
 	return ret;
 }
 
+static DEVICE_ATTR_RO(min_compr_ratio);
 static DEVICE_ATTR_RO(io_stat);
 static DEVICE_ATTR_RO(mm_stat);
 #ifdef CONFIG_ZRAM_WRITEBACK
@@ -1859,6 +1877,7 @@  static struct attribute *zram_disk_attrs[] = {
 	&dev_attr_idle.attr,
 	&dev_attr_max_comp_streams.attr,
 	&dev_attr_comp_algorithm.attr,
+	&dev_attr_min_compr_ratio.attr,
 #ifdef CONFIG_ZRAM_WRITEBACK
 	&dev_attr_backing_dev.attr,
 	&dev_attr_writeback.attr,
diff --git a/drivers/block/zram/zram_drv.h b/drivers/block/zram/zram_drv.h
index 80c3b43..5717e06 100644
--- a/drivers/block/zram/zram_drv.h
+++ b/drivers/block/zram/zram_drv.h
@@ -88,6 +88,7 @@  struct zram_stats {
 	atomic64_t bd_reads;		/* no. of reads from backing device */
 	atomic64_t bd_writes;		/* no. of writes from backing device */
 #endif
+	atomic_t min_compr_ratio;
 };
 
 struct zram {
diff --git a/fs/proc/meminfo.c b/fs/proc/meminfo.c
index 6fa761c..34a174b 100644
--- a/fs/proc/meminfo.c
+++ b/fs/proc/meminfo.c
@@ -86,6 +86,7 @@  static int meminfo_proc_show(struct seq_file *m, void *v)
 
 	show_val_kb(m, "SwapTotal:      ", i.totalswap);
 	show_val_kb(m, "SwapFree:       ", i.freeswap);
+	show_val_kb(m, "SwapAvailable:	", count_avail_swaps());
 	show_val_kb(m, "Dirty:          ",
 		    global_node_page_state(NR_FILE_DIRTY));
 	show_val_kb(m, "Writeback:      ",
diff --git a/include/linux/swap.h b/include/linux/swap.h
index bb48893..798f09b 100644
--- a/include/linux/swap.h
+++ b/include/linux/swap.h
@@ -515,6 +515,8 @@  extern int init_swap_address_space(unsigned int type, unsigned long nr_pages);
 extern void exit_swap_address_space(unsigned int type);
 extern struct swap_info_struct *get_swap_device(swp_entry_t entry);
 sector_t swap_page_sector(struct page *page);
+extern void update_zram_zstats(void);
+extern u64 count_avail_swaps(void);
 
 static inline void put_swap_device(struct swap_info_struct *si)
 {
@@ -689,6 +691,14 @@  static inline swp_entry_t get_swap_page(struct page *page)
 	return entry;
 }
 
+void update_zram_zstats(void)
+{
+}
+
+u64 count_avail_swaps(void)
+{
+}
+
 #endif /* CONFIG_SWAP */
 
 #ifdef CONFIG_THP_SWAP
diff --git a/mm/swapfile.c b/mm/swapfile.c
index 1e07d1c..fc934845 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -44,6 +44,7 @@ 
 #include <asm/tlbflush.h>
 #include <linux/swapops.h>
 #include <linux/swap_cgroup.h>
+#include "../drivers/block/zram/zram_drv.h"
 
 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
 				 unsigned char);
@@ -3387,6 +3388,100 @@  SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
 	return error;
 }
 
+u64 count_avail_swap(struct swap_info_struct *si)
+{
+	u64 result;
+	struct zram *z;
+	unsigned int free;
+	unsigned int ratio;
+
+	result = 0;
+	if (!si)
+		return 0;
+
+	//zram calculate available mem
+	if (si->flags & SWP_USED && si->swap_map) {
+		if (si->bdev->bd_disk->major == get_zram_major()) {
+			z = (struct zram *)si->bdev->bd_disk->private_data;
+			down_read(&z->init_lock);
+			ratio = atomic_read(&z->stats.min_compr_ratio);
+			free = (si->pages << (PAGE_SHIFT - 10))
+				- (si->inuse_pages << (PAGE_SHIFT - 10));
+			if (!ratio)
+				result += free / 2;
+			else
+				result = free * (100 - 10000 / ratio) / 100;
+			up_read(&z->init_lock);
+		}
+	} else
+		result += (si->pages << (PAGE_SHIFT - 10))
+				- (si->inuse_pages << (PAGE_SHIFT - 10));
+
+	return result;
+}
+
+u64 count_avail_swaps(void)
+{
+	int type;
+	u64 result;
+	struct swap_info_struct *si;
+
+	result = 0;
+	spin_lock(&swap_lock);
+	for (type = 0; type < nr_swapfiles; type++) {
+		si = swap_info[type];
+		spin_lock(&si->lock);
+		result += count_avail_swap(si);
+		spin_unlock(&si->lock);
+	}
+	spin_unlock(&swap_lock);
+
+	return result;
+}
+
+void update_zram_zstat(struct swap_info_struct *si)
+{
+	struct zram *z;
+	struct zram_stats *stat;
+	int ratio;
+	u64 orig_size, compr_data_size;
+
+	if (!si)
+		return;
+
+	//update zram min compress ratio
+	if (si->flags & SWP_USED && si->swap_map) {
+		if (si->bdev->bd_disk->major == get_zram_major()) {
+			z = (struct zram *)si->bdev->bd_disk->private_data;
+			down_read(&z->init_lock);
+			stat = &z->stats;
+			ratio = atomic_read(&stat->min_compr_ratio);
+			orig_size = atomic64_read(&stat->pages_stored) << PAGE_SHIFT;
+			compr_data_size = atomic64_read(&stat->compr_data_size);
+			if (compr_data_size && (!ratio
+					    || ((orig_size * 100) / compr_data_size < ratio)))
+				atomic_set(&stat->min_compr_ratio,
+					   (orig_size * 100) / compr_data_size);
+			up_read(&z->init_lock);
+		}
+	}
+}
+
+void update_zram_zstats(void)
+{
+	int type;
+	struct swap_info_struct *si;
+
+	spin_lock(&swap_lock);
+	for (type = 0; type < nr_swapfiles; type++) {
+		si = swap_info[type];
+		spin_lock(&si->lock);
+		update_zram_zstat(si);
+		spin_unlock(&si->lock);
+	}
+	spin_unlock(&swap_lock);
+}
+
 void si_swapinfo(struct sysinfo *val)
 {
 	unsigned int type;
diff --git a/mm/vmscan.c b/mm/vmscan.c
index eb31452..ffaf59b 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -4159,6 +4159,7 @@  static int kswapd(void *p)
 						alloc_order);
 		reclaim_order = balance_pgdat(pgdat, alloc_order,
 						highest_zoneidx);
+		update_zram_zstats();
 		if (reclaim_order < alloc_order)
 			goto kswapd_try_sleep;
 	}