On Mon, 2016-08-29 at 11:16 -0700, Bart Van Assche wrote:

On 08/14/2016 10:21 AM, James Bottomley wrote:

quoted

diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
index d3e852a..222771d 100644
--- a/drivers/scsi/sd.c
+++ b/drivers/scsi/sd.c

@@ -3000,7 +3000,13 @@ static void sd_probe_async(void *data,

async_cookie_t cookie)
 	}
 
 	blk_pm_runtime_init(sdp->request_queue, dev);
-	device_add_disk(dev, gd);
+	/*
+	 * previously the parent of the gendisk was the scsi
device.  It
+	 * was moved to fix lifetime rules, so now we install a
symlink
+	 * to the new location of the block class directory
+	 */
+	device_add_disk(&sdkp->dev, gd);
+	WARN_ON(sysfs_add_link_to_group(&dev->kobj, "block", &sdkp
->dev.kobj, "block"));
 	if (sdkp->capacity)
 		sd_dif_config_host(sdkp);
 

@@ -3142,6 +3148,7 @@ static int sd_remove(struct device *dev)
 
 	async_synchronize_full_domain(&scsi_sd_pm_domain);
 	async_synchronize_full_domain(&scsi_sd_probe_domain);
+	sysfs_remove_link(&dev->kobj, "block");
 	device_del(&sdkp->dev);
 	del_gendisk(sdkp->disk);
 	sd_shutdown(dev);

Hello James,

Sorry that it took so long before I could test this patch and
the previous patch that was posted in this e-mail thread. But I
did so earlier this morning. What I see is that the following
warning message is reported frequently:

WARNING: CPU: 1 PID: 136 at drivers/scsi/sd.c:3009
sd_probe_async+0x1ce/0x1e0

That's because the link is created too early, I think.  Let me dig into
this; I managed to hose my big device machine, so I'll need time to
resurrect it.

James