mbox series

[0/3] hwspinlock: Convert to platform remove callback returning void

Message ID 20230314180020.2865734-1-u.kleine-koenig@pengutronix.de
Headers show
Series hwspinlock: Convert to platform remove callback returning void | expand

Message

Uwe Kleine-König March 14, 2023, 6 p.m. UTC
Hello,

this patch series adapts the platform drivers below drivers/hwspinlock to use
the .remove_new() callback. Compared to the traditional .remove() callback
.remove_new() returns no value. This is a good thing because the driver core
doesn't (and cannot) cope for errors during remove. The only effect of a
non-zero return value in .remove() is that the driver core emits a warning. The
device is removed anyhow and an early return from .remove() usually yields a
resource leak.

By changing the remove callback to return void driver authors cannot reasonably
assume any more that there is some kind of cleanup later.

The omap driver could return -EBUSY. This is first changed to return
zero to drop the duplicated error message. I assume this error path is
dangerous. For sure the platform device binding goes away and so
devm_platform_ioremap_resource is undone. So probably the user of the
hwspinlock that prevented its removal will soon access an unmapped
virtual address resulting in an oops. This is true with and without my
patch. IMHO hwspin_lock_unregister() shouldn't return an error code but
care that all users go away and then return void.

After this change the two platform_drivers can be trivially converted to
.remove_new().

Best regards
Uwe

Uwe Kleine-König (3):
  hwspinlock: omap: Emit only one error message for errors in .remove()
  hwspinlock: omap: Convert to platform remove callback returning void
  hwspinlock: u8500: Convert to platform remove callback returning void

 drivers/hwspinlock/omap_hwspinlock.c | 8 +++-----
 drivers/hwspinlock/u8500_hsem.c      | 6 ++----
 2 files changed, 5 insertions(+), 9 deletions(-)

base-commit: fe15c26ee26efa11741a7b632e9f23b01aca4cc6

Comments

Uwe Kleine-König April 12, 2023, 5:16 p.m. UTC | #1
On Tue, Mar 14, 2023 at 07:00:20PM +0100, Uwe Kleine-König wrote:
> Hello,
> 
> this patch series adapts the platform drivers below drivers/hwspinlock to use
> the .remove_new() callback. Compared to the traditional .remove() callback
> .remove_new() returns no value. This is a good thing because the driver core
> doesn't (and cannot) cope for errors during remove. The only effect of a
> non-zero return value in .remove() is that the driver core emits a warning. The
> device is removed anyhow and an early return from .remove() usually yields a
> resource leak.
> 
> By changing the remove callback to return void driver authors cannot reasonably
> assume any more that there is some kind of cleanup later.
> 
> The omap driver could return -EBUSY. This is first changed to return
> zero to drop the duplicated error message. I assume this error path is
> dangerous. For sure the platform device binding goes away and so
> devm_platform_ioremap_resource is undone. So probably the user of the
> hwspinlock that prevented its removal will soon access an unmapped
> virtual address resulting in an oops. This is true with and without my
> patch. IMHO hwspin_lock_unregister() shouldn't return an error code but
> care that all users go away and then return void.
> 
> After this change the two platform_drivers can be trivially converted to
> .remove_new().

Gentle ping!

Who is supposed to apply this series (or point out a good reason to not
do that)?

Best regards
Uwe
Uwe Kleine-König June 28, 2023, 8:17 a.m. UTC | #2
Hello,

On Wed, May 31, 2023 at 09:12:40AM +0800, Baolin Wang wrote:
> On 5/30/2023 9:56 PM, Uwe Kleine-König wrote:
> > On Wed, Apr 12, 2023 at 07:16:10PM +0200, Uwe Kleine-König wrote:
> > > On Tue, Mar 14, 2023 at 07:00:20PM +0100, Uwe Kleine-König wrote:
> > > > Hello,
> > > > 
> > > > this patch series adapts the platform drivers below drivers/hwspinlock to use
> > > > the .remove_new() callback. Compared to the traditional .remove() callback
> > > > .remove_new() returns no value. This is a good thing because the driver core
> > > > doesn't (and cannot) cope for errors during remove. The only effect of a
> > > > non-zero return value in .remove() is that the driver core emits a warning. The
> > > > device is removed anyhow and an early return from .remove() usually yields a
> > > > resource leak.
> > > > 
> > > > By changing the remove callback to return void driver authors cannot reasonably
> > > > assume any more that there is some kind of cleanup later.
> > > > 
> > > > The omap driver could return -EBUSY. This is first changed to return
> > > > zero to drop the duplicated error message. I assume this error path is
> > > > dangerous. For sure the platform device binding goes away and so
> > > > devm_platform_ioremap_resource is undone. So probably the user of the
> > > > hwspinlock that prevented its removal will soon access an unmapped
> > > > virtual address resulting in an oops. This is true with and without my
> > > > patch. IMHO hwspin_lock_unregister() shouldn't return an error code but
> > > > care that all users go away and then return void.
> > > > 
> > > > After this change the two platform_drivers can be trivially converted to
> > > > .remove_new().
> > > 
> > > Gentle ping!
> > > 
> > > Who is supposed to apply this series (or point out a good reason to not
> > > do that)?
> > 
> > Still no maintainer feedback on my series :-\ Would a resend help?
> 
> For the whole patchset,
> Acked-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> 
> I think Bjorn can help to apply this patchset, Bjorn?

up to now he didn't. I guess it's to late now for v6.5-rc1, but can at
least someone pick it up for the next cycle?

It still applies fine to current Linus's master. With am -3 it also
still applies to current next. If a resend would help, please tell me
what I should choose as base.

Best regards
Uwe
Bjorn Andersson July 15, 2023, 10:13 p.m. UTC | #3
On Tue, 14 Mar 2023 19:00:20 +0100, Uwe Kleine-König wrote:
> this patch series adapts the platform drivers below drivers/hwspinlock to use
> the .remove_new() callback. Compared to the traditional .remove() callback
> .remove_new() returns no value. This is a good thing because the driver core
> doesn't (and cannot) cope for errors during remove. The only effect of a
> non-zero return value in .remove() is that the driver core emits a warning. The
> device is removed anyhow and an early return from .remove() usually yields a
> resource leak.
> 
> [...]

Applied, thanks!

[1/3] hwspinlock: omap: Emit only one error message for errors in .remove()
      commit: 72a3a509f992b6bd182b3380913fe7b4f801075f
[2/3] hwspinlock: omap: Convert to platform remove callback returning void
      commit: 4cf16b6b743e0bbe3128cf97a193ee37110d597b
[3/3] hwspinlock: u8500: Convert to platform remove callback returning void
      commit: 9519793bb6a731a3dd2453ad8515e8866e84c48e

Best regards,