Driver registration fails on SOC imx8mn as its supplier, the clock control module, is probed later than subsys initcall level. This driver uses platform_driver_probe which is not compatible with deferred probing and won't be probed again later if probe function fails due to clock not being available at that time.
This patch replaces the use of platform_driver_probe with platform_driver_register which will allow probing the driver later again when the clock control module will be available.
Fixes: a580b8c5429a ("dmaengine: mxs-dma: add dma support for i.MX23/28") Co-developed-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Dario Binacchi dario.binacchi@amarulasolutions.com Cc: stable@vger.kernel.org
---
Changes in v5: - Update the commit message. - Add the patch "dmaengine: mxs: fix section mismatch" to remove the warning raised by this patch.
Changes in v4: - Restore __init in front of mxs_dma_probe() definition. - Rename the mxs_dma_driver variable to mxs_dma_driver_probe. - Update the commit message. - Use builtin_platform_driver() instead of module_platform_driver().
Changes in v3: - Restore __init in front of mxs_dma_init() definition.
Changes in v2: - Add the tag "Cc: stable@vger.kernel.org" in the sign-off area.
drivers/dma/mxs-dma.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)
diff --git a/drivers/dma/mxs-dma.c b/drivers/dma/mxs-dma.c index 994fc4d2aca4..18f8154b859b 100644 --- a/drivers/dma/mxs-dma.c +++ b/drivers/dma/mxs-dma.c @@ -839,10 +839,6 @@ static struct platform_driver mxs_dma_driver = { .name = "mxs-dma", .of_match_table = mxs_dma_dt_ids, }, + .probe = mxs_dma_probe, }; - -static int __init mxs_dma_module_init(void) -{ - return platform_driver_probe(&mxs_dma_driver, mxs_dma_probe); -} -subsys_initcall(mxs_dma_module_init); +builtin_platform_driver(mxs_dma_driver);
The patch was suggested by the following modpost warning:
WARNING: modpost: vmlinux.o(.data+0xa3900): Section mismatch in reference from the variable mxs_dma_driver to the function .init.text:mxs_dma_probe() The variable mxs_dma_driver references the function __init mxs_dma_probe() If the reference is valid then annotate the variable with __init* or __refdata (see linux/init.h) or name the variable: *_template, *_timer, *_sht, *_ops, *_probe, *_probe_one, *_console
Co-developed-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Dario Binacchi dario.binacchi@amarulasolutions.com Cc: stable@vger.kernel.org ---
(no changes since v1)
drivers/dma/mxs-dma.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/dma/mxs-dma.c b/drivers/dma/mxs-dma.c index 18f8154b859b..a01953e06048 100644 --- a/drivers/dma/mxs-dma.c +++ b/drivers/dma/mxs-dma.c @@ -834,7 +834,7 @@ static int __init mxs_dma_probe(struct platform_device *pdev) return 0; }
-static struct platform_driver mxs_dma_driver = { +static struct platform_driver mxs_dma_driver __initdata = { .driver = { .name = "mxs-dma", .of_match_table = mxs_dma_dt_ids,
On 2022-09-04 15:10, Dario Binacchi wrote:
The patch was suggested by the following modpost warning:
WARNING: modpost: vmlinux.o(.data+0xa3900): Section mismatch in reference from the variable mxs_dma_driver to the function .init.text:mxs_dma_probe() The variable mxs_dma_driver references the function __init mxs_dma_probe() If the reference is valid then annotate the variable with __init* or __refdata (see linux/init.h) or name the variable: *_template, *_timer, *_sht, *_ops, *_probe, *_probe_one, *_console
This is very wrong - even *with* platform_driver_probe(), the driver may remain registered beyond init, so when the driver core walks the list trying to match a driver for some other device later it can access freed data and crash. Which is absolutely no fun to debug...
The correct fix is to remove the __init annotation from the probe routine. If you want to support deferred probe, consider that even your own probe call might potentially be delayed until after initdata is freed.
Thanks, Robin.
Co-developed-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Dario Binacchi dario.binacchi@amarulasolutions.com Cc: stable@vger.kernel.org
(no changes since v1)
drivers/dma/mxs-dma.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/dma/mxs-dma.c b/drivers/dma/mxs-dma.c index 18f8154b859b..a01953e06048 100644 --- a/drivers/dma/mxs-dma.c +++ b/drivers/dma/mxs-dma.c @@ -834,7 +834,7 @@ static int __init mxs_dma_probe(struct platform_device *pdev) return 0; } -static struct platform_driver mxs_dma_driver = { +static struct platform_driver mxs_dma_driver __initdata = { .driver = { .name = "mxs-dma", .of_match_table = mxs_dma_dt_ids,
Hi Dario,
On Sun, Sep 04, 2022 at 04:10:19PM +0200, Dario Binacchi wrote:
Driver registration fails on SOC imx8mn as its supplier, the clock control module, is probed later than subsys initcall level. This driver uses platform_driver_probe which is not compatible with deferred probing and won't be probed again later if probe function fails due to clock not being available at that time.
This patch replaces the use of platform_driver_probe with platform_driver_register which will allow probing the driver later again when the clock control module will be available.
Fixes: a580b8c5429a ("dmaengine: mxs-dma: add dma support for i.MX23/28") Co-developed-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Dario Binacchi dario.binacchi@amarulasolutions.com Cc: stable@vger.kernel.org
How I see it v3 of this patch is perfectly fine and should be taken instead of this one. I just commented that to v3.
Not sure if Vinod would take v3, or if you should resend v3 as v6 instead. If you do, you can add my Acked-by.
Vinod, please let us know what you prefer.
Sascha
Changes in v5:
- Update the commit message.
- Add the patch "dmaengine: mxs: fix section mismatch" to remove the warning raised by this patch.
Changes in v4:
- Restore __init in front of mxs_dma_probe() definition.
- Rename the mxs_dma_driver variable to mxs_dma_driver_probe.
- Update the commit message.
- Use builtin_platform_driver() instead of module_platform_driver().
Changes in v3:
- Restore __init in front of mxs_dma_init() definition.
Changes in v2:
- Add the tag "Cc: stable@vger.kernel.org" in the sign-off area.
drivers/dma/mxs-dma.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)
diff --git a/drivers/dma/mxs-dma.c b/drivers/dma/mxs-dma.c index 994fc4d2aca4..18f8154b859b 100644 --- a/drivers/dma/mxs-dma.c +++ b/drivers/dma/mxs-dma.c @@ -839,10 +839,6 @@ static struct platform_driver mxs_dma_driver = { .name = "mxs-dma", .of_match_table = mxs_dma_dt_ids, },
- .probe = mxs_dma_probe,
};
-static int __init mxs_dma_module_init(void) -{
- return platform_driver_probe(&mxs_dma_driver, mxs_dma_probe);
-} -subsys_initcall(mxs_dma_module_init);
+builtin_platform_driver(mxs_dma_driver);
2.32.0
Hi Vinoud,
On Tue, Sep 13, 2022 at 6:35 PM Sascha Hauer s.hauer@pengutronix.de wrote:
Hi Dario,
On Sun, Sep 04, 2022 at 04:10:19PM +0200, Dario Binacchi wrote:
Driver registration fails on SOC imx8mn as its supplier, the clock control module, is probed later than subsys initcall level. This driver uses platform_driver_probe which is not compatible with deferred probing and won't be probed again later if probe function fails due to clock not being available at that time.
This patch replaces the use of platform_driver_probe with platform_driver_register which will allow probing the driver later again when the clock control module will be available.
Fixes: a580b8c5429a ("dmaengine: mxs-dma: add dma support for i.MX23/28") Co-developed-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Dario Binacchi dario.binacchi@amarulasolutions.com Cc: stable@vger.kernel.org
How I see it v3 of this patch is perfectly fine and should be taken instead of this one. I just commented that to v3.
Not sure if Vinod would take v3, or if you should resend v3 as v6 instead. If you do, you can add my Acked-by.
Vinod, please let us know what you prefer.
Could you please let me know how to proceed? This patch has been pending for a while and it's a real shame as the change is minimal and fixes a real issue that is still present in the mainline and stable kernels.
Thanks and regards, Dario
Sascha
Changes in v5:
- Update the commit message.
- Add the patch "dmaengine: mxs: fix section mismatch" to remove the warning raised by this patch.
Changes in v4:
- Restore __init in front of mxs_dma_probe() definition.
- Rename the mxs_dma_driver variable to mxs_dma_driver_probe.
- Update the commit message.
- Use builtin_platform_driver() instead of module_platform_driver().
Changes in v3:
- Restore __init in front of mxs_dma_init() definition.
Changes in v2:
- Add the tag "Cc: stable@vger.kernel.org" in the sign-off area.
drivers/dma/mxs-dma.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)
diff --git a/drivers/dma/mxs-dma.c b/drivers/dma/mxs-dma.c index 994fc4d2aca4..18f8154b859b 100644 --- a/drivers/dma/mxs-dma.c +++ b/drivers/dma/mxs-dma.c @@ -839,10 +839,6 @@ static struct platform_driver mxs_dma_driver = { .name = "mxs-dma", .of_match_table = mxs_dma_dt_ids, },
.probe = mxs_dma_probe,
};
-static int __init mxs_dma_module_init(void) -{
return platform_driver_probe(&mxs_dma_driver, mxs_dma_probe);
-} -subsys_initcall(mxs_dma_module_init);
+builtin_platform_driver(mxs_dma_driver);
2.32.0
-- Pengutronix e.K. | | Steuerwalder Str. 21 | http://www.pengutronix.de/ | 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
On 20-09-22, 19:10, Dario Binacchi wrote:
Hi Vinoud,
On Tue, Sep 13, 2022 at 6:35 PM Sascha Hauer s.hauer@pengutronix.de wrote:
Hi Dario,
On Sun, Sep 04, 2022 at 04:10:19PM +0200, Dario Binacchi wrote:
Driver registration fails on SOC imx8mn as its supplier, the clock control module, is probed later than subsys initcall level. This driver uses platform_driver_probe which is not compatible with deferred probing and won't be probed again later if probe function fails due to clock not being available at that time.
This patch replaces the use of platform_driver_probe with platform_driver_register which will allow probing the driver later again when the clock control module will be available.
Fixes: a580b8c5429a ("dmaengine: mxs-dma: add dma support for i.MX23/28") Co-developed-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Dario Binacchi dario.binacchi@amarulasolutions.com Cc: stable@vger.kernel.org
How I see it v3 of this patch is perfectly fine and should be taken instead of this one. I just commented that to v3.
Not sure if Vinod would take v3, or if you should resend v3 as v6 instead. If you do, you can add my Acked-by.
Vinod, please let us know what you prefer.
Could you please let me know how to proceed? This patch has been pending for a while and it's a real shame as the change is minimal and fixes a real issue that is still present in the mainline and stable kernels.
Ooops, Somehow this seems to have really slipped. Sorry I owe you an apology for this
I am still not sure of this patch yet, lets get it right and merged quickly. I will send my review later today
On Wed, Sep 21, 2022 at 08:53:23AM +0530, Vinod Koul wrote:
On 20-09-22, 19:10, Dario Binacchi wrote:
Hi Vinoud,
On Tue, Sep 13, 2022 at 6:35 PM Sascha Hauer s.hauer@pengutronix.de wrote:
Hi Dario,
On Sun, Sep 04, 2022 at 04:10:19PM +0200, Dario Binacchi wrote:
Driver registration fails on SOC imx8mn as its supplier, the clock control module, is probed later than subsys initcall level. This driver uses platform_driver_probe which is not compatible with deferred probing and won't be probed again later if probe function fails due to clock not being available at that time.
This patch replaces the use of platform_driver_probe with platform_driver_register which will allow probing the driver later again when the clock control module will be available.
Fixes: a580b8c5429a ("dmaengine: mxs-dma: add dma support for i.MX23/28") Co-developed-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Michael Trimarchi michael@amarulasolutions.com Signed-off-by: Dario Binacchi dario.binacchi@amarulasolutions.com Cc: stable@vger.kernel.org
How I see it v3 of this patch is perfectly fine and should be taken instead of this one. I just commented that to v3.
Not sure if Vinod would take v3, or if you should resend v3 as v6 instead. If you do, you can add my Acked-by.
Vinod, please let us know what you prefer.
Could you please let me know how to proceed? This patch has been pending for a while and it's a real shame as the change is minimal and fixes a real issue that is still present in the mainline and stable kernels.
Ooops, Somehow this seems to have really slipped. Sorry I owe you an apology for this
I am still not sure of this patch yet, lets get it right and merged quickly. I will send my review later today
I just realized that unlike what I said v3 of this patch is still wrong as it leaves the __init annotation on mxs_dma_init() which is called from (non __init) mxs_dma_probe(). v3 probably doesn't give a section mismatch warning because mxs_dma_init() is inlined.
Really v2 is the one we should take which is at:
https://lore.kernel.org/linux-arm-kernel/20220523132247.1429321-1-dario.bina...
Sascha
On 21-09-22, 12:39, Sascha Hauer wrote:
On Wed, Sep 21, 2022 at 08:53:23AM +0530, Vinod Koul wrote:
On 20-09-22, 19:10, Dario Binacchi wrote:
How I see it v3 of this patch is perfectly fine and should be taken instead of this one. I just commented that to v3.
Not sure if Vinod would take v3, or if you should resend v3 as v6 instead. If you do, you can add my Acked-by.
Vinod, please let us know what you prefer.
Could you please let me know how to proceed? This patch has been pending for a while and it's a real shame as the change is minimal and fixes a real issue that is still present in the mainline and stable kernels.
Ooops, Somehow this seems to have really slipped. Sorry I owe you an apology for this
I am still not sure of this patch yet, lets get it right and merged quickly. I will send my review later today
I just realized that unlike what I said v3 of this patch is still wrong as it leaves the __init annotation on mxs_dma_init() which is called from (non __init) mxs_dma_probe(). v3 probably doesn't give a section mismatch warning because mxs_dma_init() is inlined.
Really v2 is the one we should take which is at:
hmmm, looking at the old revs, that does look sane. My question was why __init change is there, it needs to be documented and if there are two different reasons, add that
I agree rev 2 is the right things to do and changelog needs to add why we dropped __init (i dont think this should be a different patchset as that leads to warnings ...
https://lore.kernel.org/linux-arm-kernel/20220523132247.1429321-1-dario.bina...
linux-stable-mirror@lists.linaro.org