[PATCH v6 0/3] PCI: tegra: Allow building as a module

[PATCH v6 0/3] PCI: tegra: Allow building as a module

[Aaron Kling via B4 Relay]

Signed-off-by: Aaron Kling <webgeek1234@gmail.com>
---
Changes in v6:
- Remove unused debugfs cleanup function, as caught by kernel ci
- Link to v5: https://lore.kernel.org/r/20250505-pci-tegra-module-v5-0-827aaac998ba@gmail.com

Changes in v5:
- Copy commit message exactly word for word on patch 1, as required by reviewer
- Delete remove callback in patch 3, per request
- Don't clean up debugfs, per request, which drops patch 4 entirely
- Link to v4: https://lore.kernel.org/r/20250505-pci-tegra-module-v4-0-088b552c4b1a@gmail.com

Changes in v4:
- Updated commit messages for patches 1 and 2, per review
- Link to v3: https://lore.kernel.org/r/20250502-pci-tegra-module-v3-0-556a49732d70@gmail.com

Changes in v3:
- Add patch to drop remove callback, per request
- Link to v2: https://lore.kernel.org/r/20250428-pci-tegra-module-v2-0-c11a4b912446@gmail.com

Changes in v2:
- Add patch to export tegra_cpuidle_pcie_irqs_in_use as required when
  building pci-tegra as a module for arm
- Drop module exit to prevent module unloading, as requested
- Link to v1: https://lore.kernel.org/r/20250420-pci-tegra-module-v1-0-c0a1f831354a@gmail.com

---
Aaron Kling (3):
      irqdomain: Export irq_domain_free_irqs
      cpuidle: tegra: Export tegra_cpuidle_pcie_irqs_in_use
      PCI: tegra: Allow building as a module

 drivers/cpuidle/cpuidle-tegra.c    |  1 +
 drivers/pci/controller/Kconfig     |  2 +-
 drivers/pci/controller/pci-tegra.c | 35 ++++-------------------------------
 kernel/irq/irqdomain.c             |  1 +
 4 files changed, 7 insertions(+), 32 deletions(-)
---
base-commit: 18352e73612d60b81790d2437845276ae499b64a
change-id: 20250313-pci-tegra-module-7cbd1c5e70af

Best regards,
-- 
Aaron Kling <webgeek1234@gmail.com>

[PATCH v6 1/3] irqdomain: Export irq_domain_free_irqs

[Aaron Kling via B4 Relay]

From: Aaron Kling <webgeek1234@gmail.com>

Export irq_domain_free_irqs() to allow PCI/MSI drivers like pci-tegra to
be built as a module.

Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Aaron Kling <webgeek1234@gmail.com>
---
 kernel/irq/irqdomain.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
index ec6d8e72d980f604ded2bfa2143420e0e0095920..36cd79a8a2ce960b07b03c40067343ec8f632452 100644
--- a/kernel/irq/irqdomain.c
+++ b/kernel/irq/irqdomain.c
@@ -1890,6 +1890,7 @@ void irq_domain_free_irqs(unsigned int virq, unsigned int nr_irqs)
 	irq_domain_free_irq_data(virq, nr_irqs);
 	irq_free_descs(virq, nr_irqs);
 }
+EXPORT_SYMBOL_GPL(irq_domain_free_irqs);
 
 static void irq_domain_free_one_irq(struct irq_domain *domain, unsigned int virq)
 {

-- 
2.48.1

[PATCH v6 2/3] cpuidle: tegra: Export tegra_cpuidle_pcie_irqs_in_use

[Aaron Kling via B4 Relay]

From: Aaron Kling <webgeek1234@gmail.com>

Add export for tegra_cpuidle_pcie_irqs_in_use() so that drivers like
pci-tegra can be loaded as a module.

Signed-off-by: Aaron Kling <webgeek1234@gmail.com>
---
 drivers/cpuidle/cpuidle-tegra.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/cpuidle/cpuidle-tegra.c b/drivers/cpuidle/cpuidle-tegra.c
index b203a93deac5f378572be90e22c73e7417adb99e..aca907a62bb5de4ee4c71c1900eacedd4b90bc0a 100644
--- a/drivers/cpuidle/cpuidle-tegra.c
+++ b/drivers/cpuidle/cpuidle-tegra.c
@@ -336,6 +336,7 @@ void tegra_cpuidle_pcie_irqs_in_use(void)
 	pr_info("disabling CC6 state, since PCIe IRQs are in use\n");
 	tegra_cpuidle_disable_state(TEGRA_CC6);
 }
+EXPORT_SYMBOL_GPL(tegra_cpuidle_pcie_irqs_in_use);
 
 static void tegra_cpuidle_setup_tegra114_c7_state(void)
 {

-- 
2.48.1

[PATCH v6 3/3] PCI: tegra: Allow building as a module

[Aaron Kling via B4 Relay]

From: Aaron Kling <webgeek1234@gmail.com>

This changes the module macro back to builtin, which does not define an
exit function. This will prevent the module from being unloaded. There
are concerns with modules not cleaning up IRQs on unload, thus this
needs specifically disallowed. The remove callback is also dropped as it
is unused.

Signed-off-by: Aaron Kling <webgeek1234@gmail.com>
---
 drivers/pci/controller/Kconfig     |  2 +-
 drivers/pci/controller/pci-tegra.c | 35 ++++-------------------------------
 2 files changed, 5 insertions(+), 32 deletions(-)

diff --git a/drivers/pci/controller/Kconfig b/drivers/pci/controller/Kconfig
index 9800b768105402d6dd1ba4b134c2ec23da6e4201..a9164dd2eccaead5ae9348c24a5ad75fcb40f507 100644
--- a/drivers/pci/controller/Kconfig
+++ b/drivers/pci/controller/Kconfig
@@ -224,7 +224,7 @@ config PCI_HYPERV_INTERFACE
 	  driver.
 
 config PCI_TEGRA
-	bool "NVIDIA Tegra PCIe controller"
+	tristate "NVIDIA Tegra PCIe controller"
 	depends on ARCH_TEGRA || COMPILE_TEST
 	depends on PCI_MSI
 	help
diff --git a/drivers/pci/controller/pci-tegra.c b/drivers/pci/controller/pci-tegra.c
index b3cdbc5927de3742161310610dc5dcb836f5dd69..b902c696f63310f8f70651976d0fa2d9337134a2 100644
--- a/drivers/pci/controller/pci-tegra.c
+++ b/drivers/pci/controller/pci-tegra.c
@@ -2595,12 +2595,6 @@ static const struct seq_operations tegra_pcie_ports_sops = {
 
 DEFINE_SEQ_ATTRIBUTE(tegra_pcie_ports);
 
-static void tegra_pcie_debugfs_exit(struct tegra_pcie *pcie)
-{
-	debugfs_remove_recursive(pcie->debugfs);
-	pcie->debugfs = NULL;
-}
-
 static void tegra_pcie_debugfs_init(struct tegra_pcie *pcie)
 {
 	pcie->debugfs = debugfs_create_dir("pcie", NULL);
@@ -2674,29 +2668,6 @@ static int tegra_pcie_probe(struct platform_device *pdev)
 	return err;
 }
 
-static void tegra_pcie_remove(struct platform_device *pdev)
-{
-	struct tegra_pcie *pcie = platform_get_drvdata(pdev);
-	struct pci_host_bridge *host = pci_host_bridge_from_priv(pcie);
-	struct tegra_pcie_port *port, *tmp;
-
-	if (IS_ENABLED(CONFIG_DEBUG_FS))
-		tegra_pcie_debugfs_exit(pcie);
-
-	pci_stop_root_bus(host->bus);
-	pci_remove_root_bus(host->bus);
-	pm_runtime_put_sync(pcie->dev);
-	pm_runtime_disable(pcie->dev);
-
-	if (IS_ENABLED(CONFIG_PCI_MSI))
-		tegra_pcie_msi_teardown(pcie);
-
-	tegra_pcie_put_resources(pcie);
-
-	list_for_each_entry_safe(port, tmp, &pcie->ports, list)
-		tegra_pcie_port_free(port);
-}
-
 static int tegra_pcie_pm_suspend(struct device *dev)
 {
 	struct tegra_pcie *pcie = dev_get_drvdata(dev);
@@ -2800,6 +2771,8 @@ static struct platform_driver tegra_pcie_driver = {
 		.pm = &tegra_pcie_pm_ops,
 	},
 	.probe = tegra_pcie_probe,
-	.remove = tegra_pcie_remove,
 };
-module_platform_driver(tegra_pcie_driver);
+builtin_platform_driver(tegra_pcie_driver);
+MODULE_AUTHOR("Thierry Reding <treding@nvidia.com>");
+MODULE_DESCRIPTION("NVIDIA PCI host controller driver");
+MODULE_LICENSE("GPL");

-- 
2.48.1

Re: [PATCH v6 3/3] PCI: tegra: Allow building as a module

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 854 bytes --]

On Wed, May 07, 2025 at 10:25:54PM -0500, Aaron Kling via B4 Relay wrote:
> From: Aaron Kling <webgeek1234@gmail.com>
> 
> This changes the module macro back to builtin, which does not define an
> exit function. This will prevent the module from being unloaded. There
> are concerns with modules not cleaning up IRQs on unload, thus this
> needs specifically disallowed. The remove callback is also dropped as it
> is unused.

What exactly are these concerns? I haven't done this lately, but I'm
pretty sure that unbinding the PCI controller is something that I
extensively tested back when this code was introduced. PCI is designed
to be hot-pluggable, so there shouldn't be a need to prevent unloading
of the controller.

Rather than just forcing this to be always there, can we not fix any
issues and keep this unloadable?

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH v6 3/3] PCI: tegra: Allow building as a module

[Aaron Kling]

On Thu, May 8, 2025 at 3:40 AM Thierry Reding <thierry.reding@gmail.com> wrote:
>
> On Wed, May 07, 2025 at 10:25:54PM -0500, Aaron Kling via B4 Relay wrote:
> > From: Aaron Kling <webgeek1234@gmail.com>
> >
> > This changes the module macro back to builtin, which does not define an
> > exit function. This will prevent the module from being unloaded. There
> > are concerns with modules not cleaning up IRQs on unload, thus this
> > needs specifically disallowed. The remove callback is also dropped as it
> > is unused.
>
> What exactly are these concerns? I haven't done this lately, but I'm
> pretty sure that unbinding the PCI controller is something that I
> extensively tested back when this code was introduced. PCI is designed
> to be hot-pluggable, so there shouldn't be a need to prevent unloading
> of the controller.
>
> Rather than just forcing this to be always there, can we not fix any
> issues and keep this unloadable?

For the short version, see this part of the conversation on v1 [0].
For the long version, read comments on all revisions. Basically, I
originally submitted this as unloadable, but got told that due to
generic concerns that affect all pci drivers, including ones already
modules and unloadable, making this one a module would be blocked if
it was unloadable. Which leads us to this revision of the series.

Sincerely,
Aaron

[0] https://lore.kernel.org/all/4u4h27w77sdjvy43b3yonidhfjuvljylms3qxqfaqwyw3v32qo@kzgrrenxr6yz/

Re: [PATCH v6 3/3] PCI: tegra: Allow building as a module

[Aaron Kling]

On Thu, May 8, 2025 at 6:26 AM Aaron Kling <webgeek1234@gmail.com> wrote:
>
> On Thu, May 8, 2025 at 3:40 AM Thierry Reding <thierry.reding@gmail.com> wrote:
> >
> > On Wed, May 07, 2025 at 10:25:54PM -0500, Aaron Kling via B4 Relay wrote:
> > > From: Aaron Kling <webgeek1234@gmail.com>
> > >
> > > This changes the module macro back to builtin, which does not define an
> > > exit function. This will prevent the module from being unloaded. There
> > > are concerns with modules not cleaning up IRQs on unload, thus this
> > > needs specifically disallowed. The remove callback is also dropped as it
> > > is unused.
> >
> > What exactly are these concerns? I haven't done this lately, but I'm
> > pretty sure that unbinding the PCI controller is something that I
> > extensively tested back when this code was introduced. PCI is designed
> > to be hot-pluggable, so there shouldn't be a need to prevent unloading
> > of the controller.
> >
> > Rather than just forcing this to be always there, can we not fix any
> > issues and keep this unloadable?
>
> For the short version, see this part of the conversation on v1 [0].
> For the long version, read comments on all revisions. Basically, I
> originally submitted this as unloadable, but got told that due to
> generic concerns that affect all pci drivers, including ones already
> modules and unloadable, making this one a module would be blocked if
> it was unloadable. Which leads us to this revision of the series.
>
> Sincerely,
> Aaron
>
> [0] https://lore.kernel.org/all/4u4h27w77sdjvy43b3yonidhfjuvljylms3qxqfaqwyw3v32qo@kzgrrenxr6yz/

Is there any more comments on this, Manivannan or Thierry? I'd like to
get some form of this series submitted.

Sincerely,
Aaron

Re: [PATCH v6 0/3] PCI: tegra: Allow building as a module

[Manivannan Sadhasivam]

On Wed, May 07, 2025 at 10:25:51PM -0500, Aaron Kling via B4 Relay wrote:

You should always add the context of the series in the cover letter.

> Signed-off-by: Aaron Kling <webgeek1234@gmail.com>
> ---
> Changes in v6:
> - Remove unused debugfs cleanup function, as caught by kernel ci
> - Link to v5: https://lore.kernel.org/r/20250505-pci-tegra-module-v5-0-827aaac998ba@gmail.com
> 
> Changes in v5:
> - Copy commit message exactly word for word on patch 1, as required by reviewer
> - Delete remove callback in patch 3, per request
> - Don't clean up debugfs, per request, which drops patch 4 entirely
> - Link to v4: https://lore.kernel.org/r/20250505-pci-tegra-module-v4-0-088b552c4b1a@gmail.com
> 
> Changes in v4:
> - Updated commit messages for patches 1 and 2, per review
> - Link to v3: https://lore.kernel.org/r/20250502-pci-tegra-module-v3-0-556a49732d70@gmail.com
> 
> Changes in v3:
> - Add patch to drop remove callback, per request
> - Link to v2: https://lore.kernel.org/r/20250428-pci-tegra-module-v2-0-c11a4b912446@gmail.com
> 
> Changes in v2:
> - Add patch to export tegra_cpuidle_pcie_irqs_in_use as required when
>   building pci-tegra as a module for arm
> - Drop module exit to prevent module unloading, as requested
> - Link to v1: https://lore.kernel.org/r/20250420-pci-tegra-module-v1-0-c0a1f831354a@gmail.com
> 
> ---
> Aaron Kling (3):
>       irqdomain: Export irq_domain_free_irqs
>       cpuidle: tegra: Export tegra_cpuidle_pcie_irqs_in_use

I need an ACK from the cpuidle maintainers to take these 3 patches through PCI
tree.

- Mani

>       PCI: tegra: Allow building as a module
> 
>  drivers/cpuidle/cpuidle-tegra.c    |  1 +
>  drivers/pci/controller/Kconfig     |  2 +-
>  drivers/pci/controller/pci-tegra.c | 35 ++++-------------------------------
>  kernel/irq/irqdomain.c             |  1 +
>  4 files changed, 7 insertions(+), 32 deletions(-)
> ---
> base-commit: 18352e73612d60b81790d2437845276ae499b64a
> change-id: 20250313-pci-tegra-module-7cbd1c5e70af
> 
> Best regards,
> -- 
> Aaron Kling <webgeek1234@gmail.com>
> 
> 

-- 
மணிவண்ணன் சதாசிவம்

Re: [PATCH v6 0/3] PCI: tegra: Allow building as a module

[Aaron Kling]

On Fri, Jun 13, 2025 at 1:17 AM Manivannan Sadhasivam <mani@kernel.org> wrote:
>
> On Wed, May 07, 2025 at 10:25:51PM -0500, Aaron Kling via B4 Relay wrote:
>
> You should always add the context of the series in the cover letter.
>
> > Signed-off-by: Aaron Kling <webgeek1234@gmail.com>
> > ---
> > Changes in v6:
> > - Remove unused debugfs cleanup function, as caught by kernel ci
> > - Link to v5: https://lore.kernel.org/r/20250505-pci-tegra-module-v5-0-827aaac998ba@gmail.com
> >
> > Changes in v5:
> > - Copy commit message exactly word for word on patch 1, as required by reviewer
> > - Delete remove callback in patch 3, per request
> > - Don't clean up debugfs, per request, which drops patch 4 entirely
> > - Link to v4: https://lore.kernel.org/r/20250505-pci-tegra-module-v4-0-088b552c4b1a@gmail.com
> >
> > Changes in v4:
> > - Updated commit messages for patches 1 and 2, per review
> > - Link to v3: https://lore.kernel.org/r/20250502-pci-tegra-module-v3-0-556a49732d70@gmail.com
> >
> > Changes in v3:
> > - Add patch to drop remove callback, per request
> > - Link to v2: https://lore.kernel.org/r/20250428-pci-tegra-module-v2-0-c11a4b912446@gmail.com
> >
> > Changes in v2:
> > - Add patch to export tegra_cpuidle_pcie_irqs_in_use as required when
> >   building pci-tegra as a module for arm
> > - Drop module exit to prevent module unloading, as requested
> > - Link to v1: https://lore.kernel.org/r/20250420-pci-tegra-module-v1-0-c0a1f831354a@gmail.com
> >
> > ---
> > Aaron Kling (3):
> >       irqdomain: Export irq_domain_free_irqs
> >       cpuidle: tegra: Export tegra_cpuidle_pcie_irqs_in_use
>
> I need an ACK from the cpuidle maintainers to take these 3 patches through PCI
> tree.

Reminder about this series. I'm normally interacting with the tegra
maintainers, but I don't particularly care which tree it goes through
as long as something that works gets merged. So, can one of the
cpuidle maintainers take a quick look through this, please?

Aaron

Re: [PATCH v6 0/3] PCI: tegra: Allow building as a module

[Aaron Kling]

On Mon, Jun 30, 2025 at 1:31 PM Aaron Kling <webgeek1234@gmail.com> wrote:
>
> On Fri, Jun 13, 2025 at 1:17 AM Manivannan Sadhasivam <mani@kernel.org> wrote:
> >
> > On Wed, May 07, 2025 at 10:25:51PM -0500, Aaron Kling via B4 Relay wrote:
> >
> > You should always add the context of the series in the cover letter.
> >
> > > Signed-off-by: Aaron Kling <webgeek1234@gmail.com>
> > > ---
> > > Changes in v6:
> > > - Remove unused debugfs cleanup function, as caught by kernel ci
> > > - Link to v5: https://lore.kernel.org/r/20250505-pci-tegra-module-v5-0-827aaac998ba@gmail.com
> > >
> > > Changes in v5:
> > > - Copy commit message exactly word for word on patch 1, as required by reviewer
> > > - Delete remove callback in patch 3, per request
> > > - Don't clean up debugfs, per request, which drops patch 4 entirely
> > > - Link to v4: https://lore.kernel.org/r/20250505-pci-tegra-module-v4-0-088b552c4b1a@gmail.com
> > >
> > > Changes in v4:
> > > - Updated commit messages for patches 1 and 2, per review
> > > - Link to v3: https://lore.kernel.org/r/20250502-pci-tegra-module-v3-0-556a49732d70@gmail.com
> > >
> > > Changes in v3:
> > > - Add patch to drop remove callback, per request
> > > - Link to v2: https://lore.kernel.org/r/20250428-pci-tegra-module-v2-0-c11a4b912446@gmail.com
> > >
> > > Changes in v2:
> > > - Add patch to export tegra_cpuidle_pcie_irqs_in_use as required when
> > >   building pci-tegra as a module for arm
> > > - Drop module exit to prevent module unloading, as requested
> > > - Link to v1: https://lore.kernel.org/r/20250420-pci-tegra-module-v1-0-c0a1f831354a@gmail.com
> > >
> > > ---
> > > Aaron Kling (3):
> > >       irqdomain: Export irq_domain_free_irqs
> > >       cpuidle: tegra: Export tegra_cpuidle_pcie_irqs_in_use
> >
> > I need an ACK from the cpuidle maintainers to take these 3 patches through PCI
> > tree.
>
> Reminder about this series. I'm normally interacting with the tegra
> maintainers, but I don't particularly care which tree it goes through
> as long as something that works gets merged. So, can one of the
> cpuidle maintainers take a quick look through this, please?
>
> Aaron

There's been no responses here for a month. If no cpuidle maintainers
wish to respond, then what?

It makes more sense to me to take this through the tegra tree. But
Thierry had concerns about removing functionality that was confirmed
working. Manivannan, if this was to go through the tegra tree, would
you still block this conversion if it didn't drop the unplug/unload
functionality? There's many existing drivers, including one for newer
tegra archs, that are unloadable modules that would be affected by the
IRQ concern. I don't want to cause friction here, but I do want to get
this merged before the next LTS, so I can use it for my projects.

Aaron