[resend, PATCH net-next v1 1/2] net: thunderbolt: Switch from __maybe_unused to pm_sleep_ptr() etc

[resend, PATCH net-next v1 1/2] net: thunderbolt: Switch from __maybe_unused to pm_sleep_ptr() etc

[Andy Shevchenko]

Letting the compiler remove these functions when the kernel is built
without CONFIG_PM_SLEEP support is simpler and less heavier for builds
than the use of __maybe_unused attributes.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
---
 drivers/net/thunderbolt.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/drivers/net/thunderbolt.c b/drivers/net/thunderbolt.c
index a52ee2bf5575..4dbc6c7f2e10 100644
--- a/drivers/net/thunderbolt.c
+++ b/drivers/net/thunderbolt.c
@@ -1319,7 +1319,7 @@ static void tbnet_shutdown(struct tb_service *svc)
 	tbnet_tear_down(tb_service_get_drvdata(svc), true);
 }
 
-static int __maybe_unused tbnet_suspend(struct device *dev)
+static int tbnet_suspend(struct device *dev)
 {
 	struct tb_service *svc = tb_to_service(dev);
 	struct tbnet *net = tb_service_get_drvdata(svc);
@@ -1334,7 +1334,7 @@ static int __maybe_unused tbnet_suspend(struct device *dev)
 	return 0;
 }
 
-static int __maybe_unused tbnet_resume(struct device *dev)
+static int tbnet_resume(struct device *dev)
 {
 	struct tb_service *svc = tb_to_service(dev);
 	struct tbnet *net = tb_service_get_drvdata(svc);
@@ -1350,9 +1350,7 @@ static int __maybe_unused tbnet_resume(struct device *dev)
 	return 0;
 }
 
-static const struct dev_pm_ops tbnet_pm_ops = {
-	SET_SYSTEM_SLEEP_PM_OPS(tbnet_suspend, tbnet_resume)
-};
+static DEFINE_SIMPLE_DEV_PM_OPS(tbnet_pm_ops, tbnet_suspend, tbnet_resume);
 
 static const struct tb_service_id tbnet_ids[] = {
 	{ TB_SERVICE("network", 1) },
@@ -1364,7 +1362,7 @@ static struct tb_service_driver tbnet_driver = {
 	.driver = {
 		.owner = THIS_MODULE,
 		.name = "thunderbolt-net",
-		.pm = &tbnet_pm_ops,
+		.pm = pm_sleep_ptr(&tbnet_pm_ops),
 	},
 	.probe = tbnet_probe,
 	.remove = tbnet_remove,
-- 
2.35.1

[resend, PATCH net-next v1 2/2] net: thunderbolt: Use separate header data type for the Rx

[Andy Shevchenko]

The same data type structure is used for bitwise operations and
regular ones. It makes sparse unhappy, for example:

  .../thunderbolt.c:718:23: warning: cast to restricted __le32

  .../thunderbolt.c:953:23: warning: incorrect type in initializer (different base types)
  .../thunderbolt.c:953:23:    expected restricted __wsum [usertype] wsum
  .../thunderbolt.c:953:23:    got restricted __be32 [usertype]

Split the header to bitwise one and specific for Rx to make sparse
happy. Assure the layout by involving static_assert() against size
and offsets of the member of the structures.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
---
 drivers/net/thunderbolt.c | 22 +++++++++++++++++++++-
 1 file changed, 21 insertions(+), 1 deletion(-)

diff --git a/drivers/net/thunderbolt.c b/drivers/net/thunderbolt.c
index 4dbc6c7f2e10..f7b3d0d4646c 100644
--- a/drivers/net/thunderbolt.c
+++ b/drivers/net/thunderbolt.c
@@ -58,12 +58,32 @@
  * supported then @frame_id is filled, otherwise it stays %0.
  */
 struct thunderbolt_ip_frame_header {
+	__le32 frame_size;
+	__le16 frame_index;
+	__le16 frame_id;
+	__le32 frame_count;
+};
+
+/* Same as &struct thunderbolt_ip_frame_header for Rx */
+struct thunderbolt_ip_frame_rx_hdr {
 	u32 frame_size;
 	u16 frame_index;
 	u16 frame_id;
 	u32 frame_count;
 };
 
+static_assert(sizeof(struct thunderbolt_ip_frame_header) ==
+	      sizeof(struct thunderbolt_ip_frame_rx_hdr));
+
+#define TBIP_FRAME_HDR_MATCH(x)							 \
+	static_assert(offsetof(struct thunderbolt_ip_frame_header, frame_##x) == \
+		      offsetof(struct thunderbolt_ip_frame_rx_hdr, frame_##x))
+TBIP_FRAME_HDR_MATCH(size);
+TBIP_FRAME_HDR_MATCH(index);
+TBIP_FRAME_HDR_MATCH(id);
+TBIP_FRAME_HDR_MATCH(count);
+#undef TBIP_FRAME_HDR_MATCH
+
 enum thunderbolt_ip_frame_pdf {
 	TBIP_PDF_FRAME_START = 1,
 	TBIP_PDF_FRAME_END,
@@ -193,7 +213,7 @@ struct tbnet {
 	struct delayed_work login_work;
 	struct work_struct connected_work;
 	struct work_struct disconnect_work;
-	struct thunderbolt_ip_frame_header rx_hdr;
+	struct thunderbolt_ip_frame_rx_hdr rx_hdr;
 	struct tbnet_ring rx_ring;
 	atomic_t frame_id;
 	struct tbnet_ring tx_ring;
-- 
2.35.1

Re: [resend, PATCH net-next v1 1/2] net: thunderbolt: Switch from __maybe_unused to pm_sleep_ptr() etc

[Mika Westerberg]

On Tue, Nov 29, 2022 at 06:13:58PM +0200, Andy Shevchenko wrote:
> Letting the compiler remove these functions when the kernel is built
> without CONFIG_PM_SLEEP support is simpler and less heavier for builds
> than the use of __maybe_unused attributes.
> 
> Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>

Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>

Re: [resend, PATCH net-next v1 2/2] net: thunderbolt: Use separate header data type for the Rx

[Mika Westerberg]

On Tue, Nov 29, 2022 at 06:13:59PM +0200, Andy Shevchenko wrote:
> The same data type structure is used for bitwise operations and
> regular ones. It makes sparse unhappy, for example:
> 
>   .../thunderbolt.c:718:23: warning: cast to restricted __le32
> 
>   .../thunderbolt.c:953:23: warning: incorrect type in initializer (different base types)
>   .../thunderbolt.c:953:23:    expected restricted __wsum [usertype] wsum
>   .../thunderbolt.c:953:23:    got restricted __be32 [usertype]
> 
> Split the header to bitwise one and specific for Rx to make sparse
> happy. Assure the layout by involving static_assert() against size
> and offsets of the member of the structures.
> 
> Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> ---
>  drivers/net/thunderbolt.c | 22 +++++++++++++++++++++-
>  1 file changed, 21 insertions(+), 1 deletion(-)

I would much rather keep the humans reading this happy than add 20+
lines just to silence a tool. Unless this of course is some kind of a
real bug.

Re: [resend, PATCH net-next v1 2/2] net: thunderbolt: Use separate header data type for the Rx

[Andy Shevchenko]

On Wed, Nov 30, 2022 at 09:46:16AM +0200, Mika Westerberg wrote:
> On Tue, Nov 29, 2022 at 06:13:59PM +0200, Andy Shevchenko wrote:
> > The same data type structure is used for bitwise operations and
> > regular ones. It makes sparse unhappy, for example:
> > 
> >   .../thunderbolt.c:718:23: warning: cast to restricted __le32
> > 
> >   .../thunderbolt.c:953:23: warning: incorrect type in initializer (different base types)
> >   .../thunderbolt.c:953:23:    expected restricted __wsum [usertype] wsum
> >   .../thunderbolt.c:953:23:    got restricted __be32 [usertype]
> > 
> > Split the header to bitwise one and specific for Rx to make sparse
> > happy. Assure the layout by involving static_assert() against size
> > and offsets of the member of the structures.

> I would much rather keep the humans reading this happy than add 20+
> lines just to silence a tool. Unless this of course is some kind of a
> real bug.

Actually, changing types to bitwise ones reduces the sparse noise
(I will double check this) without reducing readability.
Would it be accepted?

-- 
With Best Regards,
Andy Shevchenko

Re: [resend, PATCH net-next v1 2/2] net: thunderbolt: Use separate header data type for the Rx

[Mika Westerberg]

On Wed, Nov 30, 2022 at 12:51:06PM +0200, Andy Shevchenko wrote:
> On Wed, Nov 30, 2022 at 09:46:16AM +0200, Mika Westerberg wrote:
> > On Tue, Nov 29, 2022 at 06:13:59PM +0200, Andy Shevchenko wrote:
> > > The same data type structure is used for bitwise operations and
> > > regular ones. It makes sparse unhappy, for example:
> > > 
> > >   .../thunderbolt.c:718:23: warning: cast to restricted __le32
> > > 
> > >   .../thunderbolt.c:953:23: warning: incorrect type in initializer (different base types)
> > >   .../thunderbolt.c:953:23:    expected restricted __wsum [usertype] wsum
> > >   .../thunderbolt.c:953:23:    got restricted __be32 [usertype]
> > > 
> > > Split the header to bitwise one and specific for Rx to make sparse
> > > happy. Assure the layout by involving static_assert() against size
> > > and offsets of the member of the structures.
> 
> > I would much rather keep the humans reading this happy than add 20+
> > lines just to silence a tool. Unless this of course is some kind of a
> > real bug.
> 
> Actually, changing types to bitwise ones reduces the sparse noise
> (I will double check this) without reducing readability.
> Would it be accepted?

Sure if it makes it more readable and does not add too many lines :)

Re: [resend, PATCH net-next v1 2/2] net: thunderbolt: Use separate header data type for the Rx

[Andy Shevchenko]

On Wed, Nov 30, 2022 at 01:09:59PM +0200, Mika Westerberg wrote:
> On Wed, Nov 30, 2022 at 12:51:06PM +0200, Andy Shevchenko wrote:
> > On Wed, Nov 30, 2022 at 09:46:16AM +0200, Mika Westerberg wrote:
> > > On Tue, Nov 29, 2022 at 06:13:59PM +0200, Andy Shevchenko wrote:
> > > > The same data type structure is used for bitwise operations and
> > > > regular ones. It makes sparse unhappy, for example:
> > > > 
> > > >   .../thunderbolt.c:718:23: warning: cast to restricted __le32
> > > > 
> > > >   .../thunderbolt.c:953:23: warning: incorrect type in initializer (different base types)
> > > >   .../thunderbolt.c:953:23:    expected restricted __wsum [usertype] wsum
> > > >   .../thunderbolt.c:953:23:    got restricted __be32 [usertype]
> > > > 
> > > > Split the header to bitwise one and specific for Rx to make sparse
> > > > happy. Assure the layout by involving static_assert() against size
> > > > and offsets of the member of the structures.
> > 
> > > I would much rather keep the humans reading this happy than add 20+
> > > lines just to silence a tool. Unless this of course is some kind of a
> > > real bug.
> > 
> > Actually, changing types to bitwise ones reduces the sparse noise
> > (I will double check this) without reducing readability.
> > Would it be accepted?
> 
> Sure if it makes it more readable and does not add too many lines :)

It replaces types u* by __le*, that's it: -4 +4 LoCs.

-- 
With Best Regards,
Andy Shevchenko