[PATCH 0/3] mtd: core: protect access to mtd devices while in suspend

[Boris Brezillon]

On Wed, 20 Oct 2021 08:52:50 +0200
Boris Brezillon <boris.brezillon at collabora.com> wrote:

> On Tue, 19 Oct 2021 20:08:00 +0200
> Sean Nyekjaer <sean at geanix.com> wrote:
> 
> > On Fri, Oct 15, 2021 at 08:22:06AM +0200, Miquel Raynal wrote:  
> > > Hi Sean,
> > > 
> > > boris.brezillon at collabora.com wrote on Mon, 11 Oct 2021 16:05:46 +0200:
> > >     
> > > > On Mon, 11 Oct 2021 13:52:50 +0200    
> > 
> > [ ... ]
> >   
> > >     
> > > > > One (small) issue still present. gpmi_nand.c uses the rwsem before it's
> > > > > initialized. Seems cumbersome to have every mtd/nand driver to call
> > > > > init_waitqueue_head() and init_rwsem(). Could we somehow move the call
> > > > > to mtd_set_dev_defaults() before nand_create_bbt()?      
> > > > 
> > > > I have a nasty trick for that one, but I'm not sure Miquel will like it
> > > > (actually, I don't like it either, but it's so simple compared to the
> > > > other options we have that I'm tempted to go for this approach until
> > > > someone has time to invest in a cleaner solution :-)):
> > > > 
> > > > diff --git a/drivers/mtd/nand/raw/nand_base.c b/drivers/mtd/nand/raw/nand_base.c
> > > > index 3d6c6e880520..a9ac2d528a4d 100644
> > > > --- a/drivers/mtd/nand/raw/nand_base.c
> > > > +++ b/drivers/mtd/nand/raw/nand_base.c
> > > > @@ -6222,8 +6222,6 @@ static int nand_scan_tail(struct nand_chip *chip)
> > > >         mtd->_sync = nand_sync;
> > > >         mtd->_lock = nand_lock;
> > > >         mtd->_unlock = nand_unlock;
> > > > -       mtd->_suspend = nand_suspend;
> > > > -       mtd->_resume = nand_resume;
> > > >         mtd->_reboot = nand_shutdown;
> > > >         mtd->_block_isreserved = nand_block_isreserved;
> > > >         mtd->_block_isbad = nand_block_isbad;
> > > > @@ -6269,6 +6267,13 @@ static int nand_scan_tail(struct nand_chip *chip)
> > > >         if (ret)
> > > >                 goto err_free_secure_regions;
> > > >  
> > > > +       /*
> > > > +        * Populate the suspend/resume hooks after the BBT has been scanned to
> > > > +        * avoid using the suspend lock and resume waitqueue which are only
> > > > +        * initialized when mtd_device_register() is called.
> > > > +        */
> > > > +       mtd->_suspend = nand_suspend;
> > > > +       mtd->_resume = nand_resume;
> > > >         return 0;    
> > > 
> > > I'm fine with this as long as it is documented for now.
> > >     
> > 
> > Hi Boris and Miquel,
> > 
> > gpmi-nand.c sets NAND_SKIP_BBTSCAN so we won't get there and populate
> > suspend resume hooks :(
> > Guess there is other drivers that does the same thing...  
> 
> This should fix the issue:
> 
> diff --git a/drivers/mtd/nand/raw/nand_base.c b/drivers/mtd/nand/raw/nand_base.c
> index 3d6c6e880520..c92b17f66994 100644
> --- a/drivers/mtd/nand/raw/nand_base.c
> +++ b/drivers/mtd/nand/raw/nand_base.c
> @@ -6222,8 +6222,6 @@ static int nand_scan_tail(struct nand_chip *chip)
>         mtd->_sync = nand_sync;
>         mtd->_lock = nand_lock;
>         mtd->_unlock = nand_unlock;
> -       mtd->_suspend = nand_suspend;
> -       mtd->_resume = nand_resume;
>         mtd->_reboot = nand_shutdown;
>         mtd->_block_isreserved = nand_block_isreserved;
>         mtd->_block_isbad = nand_block_isbad;
> @@ -6262,13 +6260,21 @@ static int nand_scan_tail(struct nand_chip *chip)
>  
>         /* Check, if we should skip the bad block table scan */
>         if (chip->options & NAND_SKIP_BBTSCAN)
> -               return 0;
> +               goto out;
>  
>         /* Build bad block table */
>         ret = nand_create_bbt(chip);
>         if (ret)
>                 goto err_free_secure_regions;
>  
> +out:
> +       /*
> +        * Populate the suspend/resume hooks after the BBT has been scanned to
> +        * avoid using the suspend lock and resume waitqueue which are only
> +        * initialized when mtd_device_register() is called.
> +        */
> +       mtd->_suspend = nand_suspend;
> +       mtd->_resume = nand_resume;
>         return 0;
>  
>  err_free_secure_regions:

Actually, this version is even cleaner:

diff --git a/drivers/mtd/nand/raw/nand_base.c b/drivers/mtd/nand/raw/nand_base.c
index 3d6c6e880520..98c39b7f6279 100644
--- a/drivers/mtd/nand/raw/nand_base.c
+++ b/drivers/mtd/nand/raw/nand_base.c
@@ -6222,8 +6222,6 @@ static int nand_scan_tail(struct nand_chip *chip)
        mtd->_sync = nand_sync;
        mtd->_lock = nand_lock;
        mtd->_unlock = nand_unlock;
-       mtd->_suspend = nand_suspend;
-       mtd->_resume = nand_resume;
        mtd->_reboot = nand_shutdown;
        mtd->_block_isreserved = nand_block_isreserved;
        mtd->_block_isbad = nand_block_isbad;
@@ -6261,14 +6259,20 @@ static int nand_scan_tail(struct nand_chip *chip)
                goto err_free_interface_config;
 
        /* Check, if we should skip the bad block table scan */
-       if (chip->options & NAND_SKIP_BBTSCAN)
-               return 0;
-
-       /* Build bad block table */
-       ret = nand_create_bbt(chip);
-       if (ret)
-               goto err_free_secure_regions;
+       if (chip->options & NAND_SKIP_BBTSCAN) {
+               /* Build bad block table */
+               ret = nand_create_bbt(chip);
+               if (ret)
+                       goto err_free_secure_regions;
+       }
 
+       /*
+        * Populate the suspend/resume hooks after the BBT has been scanned to
+        * avoid using the suspend lock and resume waitqueue which are only
+        * initialized when mtd_device_register() is called.
+        */
+       mtd->_suspend = nand_suspend;
+       mtd->_resume = nand_resume;
        return 0;
 
 err_free_secure_regions: