host_bridge::reset_slot() is supposed to reset the PCI root port/slot. Once
that happens, the config space content would be lost. This was reported by
Niklas on the dw-rockchip based platform where the MPS setting of the root
port was lost after the host_bridge::reset_slot() callback. Hence, save the
config space before calling the host_bridge::reset_slot() callback and
restore it afterwards.
While at it, make sure that the callback is only called for root ports by
checking if the bridge is behind the root bus.
Fixes: d5c1e1c25b37 ("PCI/ERR: Add support for resetting the slots in a platform specific way")
Reported-by: Niklas Cassel <cassel@kernel.org>
Closes: https://lore.kernel.org/linux-pci/aC9OrPAfpzB_A4K2@ryzen
Suggested-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
---
drivers/pci/pci.c | 11 ++++++++++-
1 file changed, 10 insertions(+), 1 deletion(-)
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index 4d396bbab4a8..6d6e9ce2bbcc 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -4985,10 +4985,19 @@ void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
int ret;
- if (host->reset_slot) {
+ if (pci_is_root_bus(dev->bus) && host->reset_slot) {
+ /*
+ * Save the config space of the root port before doing the
+ * reset, since the state could be lost. The device state
+ * should've been saved by the caller.
+ */
+ pci_save_state(dev);
ret = host->reset_slot(host, dev);
if (ret)
pci_err(dev, "failed to reset slot: %d\n", ret);
+ else
+ /* Now restore it on success */
+ pci_restore_state(dev);
return;
}
--
2.43.0
On Sun, May 25, 2025 at 12:23:03AM +0530, Manivannan Sadhasivam wrote:
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -4985,10 +4985,19 @@ void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
> struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
> int ret;
>
> - if (host->reset_slot) {
> + if (pci_is_root_bus(dev->bus) && host->reset_slot) {
> + /*
> + * Save the config space of the root port before doing the
> + * reset, since the state could be lost. The device state
> + * should've been saved by the caller.
> + */
> + pci_save_state(dev);
> ret = host->reset_slot(host, dev);
Nit: Capitalize terms as the PCIe Base Spec does, i.e. "Root Port".
"The device state" is ambiguous as the Root Port is a device itself
and even referred to by the "dev" variable. I think what you mean
is "The Endpoint state".
Thanks,
Lukas
On Sun, May 25, 2025 at 09:22:03AM +0200, Lukas Wunner wrote:
> On Sun, May 25, 2025 at 12:23:03AM +0530, Manivannan Sadhasivam wrote:
> > --- a/drivers/pci/pci.c
> > +++ b/drivers/pci/pci.c
> > @@ -4985,10 +4985,19 @@ void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
> > struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
> > int ret;
> >
> > - if (host->reset_slot) {
> > + if (pci_is_root_bus(dev->bus) && host->reset_slot) {
> > + /*
> > + * Save the config space of the root port before doing the
> > + * reset, since the state could be lost. The device state
> > + * should've been saved by the caller.
> > + */
> > + pci_save_state(dev);
> > ret = host->reset_slot(host, dev);
>
> Nit: Capitalize terms as the PCIe Base Spec does, i.e. "Root Port".
>
Ack.
> "The device state" is ambiguous as the Root Port is a device itself
> and even referred to by the "dev" variable. I think what you mean
> is "The Endpoint state".
>
Yes! Will fix them while applying, thanks!
- Mani
--
மணிவண்ணன் சதாசிவம்
On Sun, May 25, 2025 at 01:28:18PM +0530, Manivannan Sadhasivam wrote:
> On Sun, May 25, 2025 at 09:22:03AM +0200, Lukas Wunner wrote:
> > "The device state" is ambiguous as the Root Port is a device itself
> > and even referred to by the "dev" variable. I think what you mean
> > is "The Endpoint state".
> >
>
> Yes! Will fix them while applying, thanks!
ICYMI, current controller/dw-rockchip branch still uses
"The device state", not "The Endpoint state" in commit
56eecfc8f46f ("PCI/ERR: Add support for resetting the
Root Ports in a platform specific way").
Otherwise LGTM.
Thanks,
Lukas
On Mon, May 26, 2025 at 08:54:37AM +0200, Lukas Wunner wrote:
> On Sun, May 25, 2025 at 01:28:18PM +0530, Manivannan Sadhasivam wrote:
> > On Sun, May 25, 2025 at 09:22:03AM +0200, Lukas Wunner wrote:
> > > "The device state" is ambiguous as the Root Port is a device itself
> > > and even referred to by the "dev" variable. I think what you mean
> > > is "The Endpoint state".
> > >
> >
> > Yes! Will fix them while applying, thanks!
>
> ICYMI, current controller/dw-rockchip branch still uses
> "The device state", not "The Endpoint state" in commit
> 56eecfc8f46f ("PCI/ERR: Add support for resetting the
> Root Ports in a platform specific way").
>
Ah, missed this comment. Incorported now!
- Mani
--
மணிவண்ணன் சதாசிவம்
On Sun, 2025-05-25 at 00:23 +0530, Manivannan Sadhasivam wrote:
> host_bridge::reset_slot() is supposed to reset the PCI root
> port/slot. Once
> that happens, the config space content would be lost. This was
> reported by
> Niklas on the dw-rockchip based platform where the MPS setting of the
> root
> port was lost after the host_bridge::reset_slot() callback. Hence,
> save the
> config space before calling the host_bridge::reset_slot() callback
> and
> restore it afterwards.
>
> While at it, make sure that the callback is only called for root
> ports by
> checking if the bridge is behind the root bus.
>
> Fixes: d5c1e1c25b37 ("PCI/ERR: Add support for resetting the slots in
> a platform specific way")
> Reported-by: Niklas Cassel <cassel@kernel.org>
> Closes: https://lore.kernel.org/linux-pci/aC9OrPAfpzB_A4K2@ryzen
> Suggested-by: Lukas Wunner <lukas@wunner.de>
> Signed-off-by: Manivannan Sadhasivam
> <manivannan.sadhasivam@linaro.org>
> ---
> drivers/pci/pci.c | 11 ++++++++++-
> 1 file changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index 4d396bbab4a8..6d6e9ce2bbcc 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -4985,10 +4985,19 @@ void __weak
> pcibios_reset_secondary_bus(struct pci_dev *dev)
> struct pci_host_bridge *host = pci_find_host_bridge(dev-
> >bus);
> int ret;
>
> - if (host->reset_slot) {
> + if (pci_is_root_bus(dev->bus) && host->reset_slot) {
> + /*
> + * Save the config space of the root port before
> doing the
> + * reset, since the state could be lost. The device
> state
> + * should've been saved by the caller.
> + */
> + pci_save_state(dev);
> ret = host->reset_slot(host, dev);
> if (ret)
> pci_err(dev, "failed to reset slot: %d\n",
> ret);
> + else
> + /* Now restore it on success */
> + pci_restore_state(dev);
>
> return;
> }
Reviewed-by: Wilfred Mallawa <wilfred.mallawa@wdc.com>
On Sun, May 25, 2025 at 12:23:03AM +0530, Manivannan Sadhasivam wrote:
> host_bridge::reset_slot() is supposed to reset the PCI root port/slot. Once
> that happens, the config space content would be lost. This was reported by
> Niklas on the dw-rockchip based platform where the MPS setting of the root
> port was lost after the host_bridge::reset_slot() callback. Hence, save the
> config space before calling the host_bridge::reset_slot() callback and
> restore it afterwards.
>
> While at it, make sure that the callback is only called for root ports by
> checking if the bridge is behind the root bus.
>
> Fixes: d5c1e1c25b37 ("PCI/ERR: Add support for resetting the slots in a platform specific way")
> Reported-by: Niklas Cassel <cassel@kernel.org>
> Closes: https://lore.kernel.org/linux-pci/aC9OrPAfpzB_A4K2@ryzen
> Suggested-by: Lukas Wunner <lukas@wunner.de>
> Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
> ---
> drivers/pci/pci.c | 11 ++++++++++-
> 1 file changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index 4d396bbab4a8..6d6e9ce2bbcc 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -4985,10 +4985,19 @@ void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
> struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
> int ret;
>
> - if (host->reset_slot) {
> + if (pci_is_root_bus(dev->bus) && host->reset_slot) {
> + /*
> + * Save the config space of the root port before doing the
> + * reset, since the state could be lost. The device state
> + * should've been saved by the caller.
> + */
> + pci_save_state(dev);
> ret = host->reset_slot(host, dev);
> if (ret)
> pci_err(dev, "failed to reset slot: %d\n", ret);
> + else
> + /* Now restore it on success */
> + pci_restore_state(dev);
>
> return;
> }
> --
> 2.43.0
>
Looks good to me:
Reviewed-by: Niklas Cassel <cassel@kernel.org>
© 2016 - 2025 Red Hat, Inc.