On Tue, Jun 22, 2021 at 09:39:40AM +0200, Greg KH wrote:

On Mon, Jun 21, 2021 at 04:30:11PM -0700, Luis Chamberlain wrote:

quoted

CPU 1                            CPU 2

class_unregister(...);

Now the sysfs files are removed and invalidated for all devices
associated with that class.

You're right the disksize_store() would have to happen before.

quoted

idr_for_each(...);
zram_debugfs_destroy();
                                disksize_store(...);

How will this call into the kobject's store function if
class_unregister() has already happened?

The disksize_store() would indeed have to happen before.

quoted

idr_destroy(...);
unregister_blkdev(...);

Ah, it's a block device's store function you are worried about, not the
class one?

In this case its about any files which can change the cpu compression
streams.

quoted

diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index cf8deecc39ef..431b60cd85c1 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c

@@ -44,6 +44,8 @@ static DEFINE_MUTEX(zram_index_mutex);
 static int zram_major;
 static const char *default_compressor = CONFIG_ZRAM_DEF_COMP;
 
+bool zram_up;

static?

Will fix.

quoted

+
 /* Module params (documentation at end) */
 static unsigned int num_devices = 1;
 /*

@@ -1704,6 +1706,7 @@ static void zram_reset_device(struct zram *zram)
 	comp = zram->comp;
 	disksize = zram->disksize;
 	zram->disksize = 0;
+	zram->comp = NULL;

Is this a new change?

It is a sanity change, but indeed, it is separate. I'll let Minchan
decide if he would prefer this to go out as a separate change.

  Luis