Re: [PATCH for kernel 3.19] Avoid that scsi_device_put() triggers a kernel warning

[James Bottomley <jbottomley@parallels.com>]

On Mon, 2015-01-05 at 12:03 +0100, Bart Van Assche wrote:
> Avoid that the following warning is reported when a SCSI LLD kernel
> module is unloaded:
> 
> WARNING: CPU: 5 PID: 228 at kernel/module.c:954 module_put+0x207/0x220()
> Call Trace:
>  [<ffffffff814d1fcf>] dump_stack+0x4c/0x65
>  [<ffffffff81053ada>] warn_slowpath_common+0x8a/0xc0
>  [<ffffffff81053bca>] warn_slowpath_null+0x1a/0x20
>  [<ffffffff810d0507>] module_put+0x207/0x220
>  [<ffffffffa000bea8>] scsi_device_put+0x48/0x50 [scsi_mod]
>  [<ffffffffa03676d2>] scsi_disk_put+0x32/0x50 [sd_mod]
>  [<ffffffffa0368d4c>] sd_shutdown+0x8c/0x150 [sd_mod]
>  [<ffffffffa0368e79>] sd_remove+0x69/0xc0 [sd_mod]
>  [<ffffffff813457ef>] __device_release_driver+0x7f/0xf0
>  [<ffffffff81345885>] device_release_driver+0x25/0x40
>  [<ffffffff81345134>] bus_remove_device+0x124/0x1b0
>  [<ffffffff8134189e>] device_del+0x13e/0x250
>  [<ffffffffa001cdcd>] __scsi_remove_device+0xcd/0xe0 [scsi_mod]
>  [<ffffffffa001b39f>] scsi_forget_host+0x6f/0x80 [scsi_mod]
>  [<ffffffffa000d5f6>] scsi_remove_host+0x86/0x140 [scsi_mod]
>  [<ffffffffa07d5c0b>] srp_remove_work+0x9b/0x210 [ib_srp]
>  [<ffffffff8106fd28>] process_one_work+0x1d8/0x780
>  [<ffffffff810703eb>] worker_thread+0x11b/0x4a0
>  [<ffffffff81075a6f>] kthread+0xef/0x110
>  [<ffffffff814dad6c>] ret_from_fork+0x7c/0xb0
> 
> See also patch "module: Remove stop_machine from module unloading"
> (Masami Hiramatsu; commit e513cc1c07e2; kernel v3.19-rc1).
> 
> Signed-off-by: Bart Van Assche <bvanassche@acm.org>
> Cc: Christoph Hellwig <hch@lst.de>
> Cc: Hannes Reinecke <hare@suse.de>
> ---
>  drivers/scsi/hosts.c     |  1 +
>  drivers/scsi/scsi.c      | 13 ++++---------
>  include/scsi/scsi_host.h |  3 +++
>  3 files changed, 8 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/scsi/hosts.c b/drivers/scsi/hosts.c
> index 8bb173e..e9155d0 100644
> --- a/drivers/scsi/hosts.c
> +++ b/drivers/scsi/hosts.c
> @@ -380,6 +380,7 @@ struct Scsi_Host *scsi_host_alloc(struct scsi_host_template *sht, int privsize)
>  
>  	shost->host_lock = &shost->default_lock;
>  	spin_lock_init(shost->host_lock);
> +	atomic_set(&shost->module_refcnt, 0);
>  	shost->shost_state = SHOST_CREATED;
>  	INIT_LIST_HEAD(&shost->__devices);
>  	INIT_LIST_HEAD(&shost->__targets);
> diff --git a/drivers/scsi/scsi.c b/drivers/scsi/scsi.c
> index e028854..ed325d1 100644
> --- a/drivers/scsi/scsi.c
> +++ b/drivers/scsi/scsi.c
> @@ -988,7 +988,8 @@ int scsi_device_get(struct scsi_device *sdev)
>  		return -ENXIO;
>  	/* We can fail this if we're doing SCSI operations
>  	 * from module exit (like cache flush) */
> -	try_module_get(sdev->host->hostt->module);
> +	if (try_module_get(sdev->host->hostt->module))
> +		atomic_inc(&sdev->host->module_refcnt);
>  
>  	return 0;
>  }
> @@ -1004,14 +1005,8 @@ EXPORT_SYMBOL(scsi_device_get);
>   */
>  void scsi_device_put(struct scsi_device *sdev)
>  {
> -#ifdef CONFIG_MODULE_UNLOAD
> -	struct module *module = sdev->host->hostt->module;
> -
> -	/* The module refcount will be zero if scsi_device_get()
> -	 * was called from a module removal routine */
> -	if (module && module_refcount(module) != 0)
> -		module_put(module);
> -#endif
> +	if (atomic_dec_if_positive(&sdev->host->module_refcnt) >= 0)
> +		module_put(sdev->host->hostt->module);
>  	put_device(&sdev->sdev_gendev);
>  }
>  EXPORT_SYMBOL(scsi_device_put);
> diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
> index 019e668..b8e6f01 100644
> --- a/include/scsi/scsi_host.h
> +++ b/include/scsi/scsi_host.h
> @@ -566,6 +566,9 @@ struct Scsi_Host {
>  	struct scsi_host_template *hostt;
>  	struct scsi_transport_template *transportt;
>  
> +	/* Number of LLD kernel module references held by the SCSI core. */
> +	atomic_t		 module_refcnt;
> +
>  	/*
>  	 * Area to keep a shared tag map (if needed, will be
>  	 * NULL if not).

This is a bit over coded for a fix.  We don't need to duplicate the
module reference, merely take into account that in the remove code
module_refcnt() can now return -1, which we're not expecting.

Masami, this looks like a nasty unintended consequence of 

commit e513cc1c07e2ab93a4514eec9833e031df3e30bb
Author: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Date:   Mon Nov 10 09:30:29 2014 +1030

    module: Remove stop_machine from module unloading

Are you sure we all should have to alter our module refcounting in
removal routines like this?  I think the better fix is having
module_refcount() never return a negative number:

unsigned long module_refcount(struct module *mod)
{
	unsigned long ret = (unsigned long)atomic_read(&mod->refcnt);

	if (ret == 0)
		return ret
	return ret - MODULE_REF_BASE;
}

James

---

diff --git a/drivers/scsi/scsi.c b/drivers/scsi/scsi.c
index e028854..90dcb2f 100644
--- a/drivers/scsi/scsi.c
+++ b/drivers/scsi/scsi.c
@@ -1009,7 +1009,7 @@ void scsi_device_put(struct scsi_device *sdev)
 
 	/* The module refcount will be zero if scsi_device_get()
 	 * was called from a module removal routine */
-	if (module && module_refcount(module) != 0)
+	if (module && module_refcount(module) > 0)
 		module_put(module);
 #endif
 	put_device(&sdev->sdev_gendev);