PCI: keystone: fix crash with deferred probe

Drop __init annotations for functions called during
probe of keystone PCIe host.

This fixes kernel crashes seen because of init data
reclamation if the driver defer probes because of
phy unavailability.

While at it, drop the useless __refdata annotation
to driver structure. If its really needed, it should
be added back with a comment explaining why.

Signed-off-by: Sekhar Nori <nsekhar@ti.com>
---
 drivers/pci/host/pci-keystone-dw.c | 2 +-
 drivers/pci/host/pci-keystone.c    | 6 +++---
 2 files changed, 4 insertions(+), 4 deletions(-)

Dear Sekhar,

On Fri, 20 Nov 2015 16:47:02 +0530
Sekhar Nori wrote:

> Drop __init annotations for functions called during
> probe of keystone PCIe host.
> 
> This fixes kernel crashes seen because of init data
> reclamation if the driver defer probes because of
> phy unavailability.
> 
> While at it, drop the useless __refdata annotation
> to driver structure. If its really needed, it should
> be added back with a comment explaining why.
> 
> Signed-off-by: Sekhar Nori <nsekhar@ti.com>
> ---
>  drivers/pci/host/pci-keystone-dw.c | 2 +-
>  drivers/pci/host/pci-keystone.c    | 6 +++---
>  2 files changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/pci/host/pci-keystone-dw.c b/drivers/pci/host/pci-keystone-dw.c
> index ed34c9520a02..47ddbcce53ca 100644
> --- a/drivers/pci/host/pci-keystone-dw.c
> +++ b/drivers/pci/host/pci-keystone-dw.c
> @@ -469,7 +469,7 @@ void ks_dw_pcie_initiate_link_train(struct keystone_pcie *ks_pcie)
>   * and call dw_pcie_v3_65_host_init() API to initialize the Keystone
>   * PCI host controller.
>   */
> -int __init ks_dw_pcie_host_init(struct keystone_pcie *ks_pcie,
> +int ks_dw_pcie_host_init(struct keystone_pcie *ks_pcie,
>  				struct device_node *msi_intc_np)

when trying to add berlin pcie support and came up one patch

http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/384575.html

when build testing, I also realized this issue, but keystone relies on
hook_fault_code() which is will be freed after boot, so your patch isn't
completed.

And from the keystone driver implementation, it seems assume the driver won't
be defer probed.

Thanks

>  {
>  	struct pcie_port *pp = &ks_pcie->pp;
> diff --git a/drivers/pci/host/pci-keystone.c b/drivers/pci/host/pci-keystone.c
> index 0aa81bd3de12..593a58834c7d 100644
> --- a/drivers/pci/host/pci-keystone.c
> +++ b/drivers/pci/host/pci-keystone.c
> @@ -291,7 +291,7 @@ static struct pcie_host_ops keystone_pcie_host_ops = {
>  	.scan_bus = ks_dw_pcie_v3_65_scan_bus,
>  };
>  
> -static int __init ks_add_pcie_port(struct keystone_pcie *ks_pcie,
> +static int ks_add_pcie_port(struct keystone_pcie *ks_pcie,
>  			 struct platform_device *pdev)
>  {
>  	struct pcie_port *pp = &ks_pcie->pp;
> @@ -340,7 +340,7 @@ static int __exit ks_pcie_remove(struct platform_device *pdev)
>  	return 0;
>  }
>  
> -static int __init ks_pcie_probe(struct platform_device *pdev)
> +static int ks_pcie_probe(struct platform_device *pdev)
>  {
>  	struct device *dev = &pdev->dev;
>  	struct keystone_pcie *ks_pcie;
> @@ -396,7 +396,7 @@ fail_clk:
>  	return ret;
>  }
>  
> -static struct platform_driver ks_pcie_driver __refdata = {
> +static struct platform_driver ks_pcie_driver = {
>  	.probe  = ks_pcie_probe,
>  	.remove = __exit_p(ks_pcie_remove),
>  	.driver = {

--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Mon, 23 Nov 2015 19:22:09 +0800
Jisheng Zhang <jszhang@marvell.com> wrote:

> Dear Sekhar,
> 
> On Fri, 20 Nov 2015 16:47:02 +0530
> Sekhar Nori wrote:
> 
> > Drop __init annotations for functions called during
> > probe of keystone PCIe host.
> > 
> > This fixes kernel crashes seen because of init data
> > reclamation if the driver defer probes because of
> > phy unavailability.
> > 
> > While at it, drop the useless __refdata annotation
> > to driver structure. If its really needed, it should
> > be added back with a comment explaining why.
> > 
> > Signed-off-by: Sekhar Nori <nsekhar@ti.com>
> > ---
> >  drivers/pci/host/pci-keystone-dw.c | 2 +-
> >  drivers/pci/host/pci-keystone.c    | 6 +++---
> >  2 files changed, 4 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/pci/host/pci-keystone-dw.c b/drivers/pci/host/pci-keystone-dw.c
> > index ed34c9520a02..47ddbcce53ca 100644
> > --- a/drivers/pci/host/pci-keystone-dw.c
> > +++ b/drivers/pci/host/pci-keystone-dw.c
> > @@ -469,7 +469,7 @@ void ks_dw_pcie_initiate_link_train(struct keystone_pcie *ks_pcie)
> >   * and call dw_pcie_v3_65_host_init() API to initialize the Keystone
> >   * PCI host controller.
> >   */
> > -int __init ks_dw_pcie_host_init(struct keystone_pcie *ks_pcie,
> > +int ks_dw_pcie_host_init(struct keystone_pcie *ks_pcie,
> >  				struct device_node *msi_intc_np)  
> 
> when trying to add berlin pcie support and came up one patch
> 
> http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/384575.html
> 
> when build testing, I also realized this issue, but keystone relies on

I mean ks_dw_pcie_host_init() calls hook_fault_code()

> hook_fault_code() which is will be freed after boot, so your patch isn't
> completed.

And will introduce section mismatch warnings

> 
> And from the keystone driver implementation, it seems assume the driver won't
> be defer probed.
> 
> Thanks
> 
> >  {
> >  	struct pcie_port *pp = &ks_pcie->pp;
> > diff --git a/drivers/pci/host/pci-keystone.c b/drivers/pci/host/pci-keystone.c
> > index 0aa81bd3de12..593a58834c7d 100644
> > --- a/drivers/pci/host/pci-keystone.c
> > +++ b/drivers/pci/host/pci-keystone.c
> > @@ -291,7 +291,7 @@ static struct pcie_host_ops keystone_pcie_host_ops = {
> >  	.scan_bus = ks_dw_pcie_v3_65_scan_bus,
> >  };
> >  
> > -static int __init ks_add_pcie_port(struct keystone_pcie *ks_pcie,
> > +static int ks_add_pcie_port(struct keystone_pcie *ks_pcie,
> >  			 struct platform_device *pdev)
> >  {
> >  	struct pcie_port *pp = &ks_pcie->pp;
> > @@ -340,7 +340,7 @@ static int __exit ks_pcie_remove(struct platform_device *pdev)
> >  	return 0;
> >  }
> >  
> > -static int __init ks_pcie_probe(struct platform_device *pdev)
> > +static int ks_pcie_probe(struct platform_device *pdev)
> >  {
> >  	struct device *dev = &pdev->dev;
> >  	struct keystone_pcie *ks_pcie;
> > @@ -396,7 +396,7 @@ fail_clk:
> >  	return ret;
> >  }
> >  
> > -static struct platform_driver ks_pcie_driver __refdata = {
> > +static struct platform_driver ks_pcie_driver = {
> >  	.probe  = ks_pcie_probe,
> >  	.remove = __exit_p(ks_pcie_remove),
> >  	.driver = {  
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Mon, Nov 23, 2015 at 07:22:09PM +0800, Jisheng Zhang wrote:
> when trying to add berlin pcie support and came up one patch
> 
> http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/384575.html
> 
> when build testing, I also realized this issue, but keystone relies on
> hook_fault_code() which is will be freed after boot, so your patch isn't
> completed.

hook_fault_code() is intentionally not callable after boot.  It's unsafe
to try to change the fault handling for a running system.  It's even more
unsafe to try to remove a fault handler which is located in module code.

On Monday 23 November 2015 04:57 PM, Jisheng Zhang wrote:
> On Mon, 23 Nov 2015 19:22:09 +0800
> Jisheng Zhang <jszhang@marvell.com> wrote:
> 
>> Dear Sekhar,
>>
>> On Fri, 20 Nov 2015 16:47:02 +0530
>> Sekhar Nori wrote:
>>
>>> Drop __init annotations for functions called during
>>> probe of keystone PCIe host.
>>>
>>> This fixes kernel crashes seen because of init data
>>> reclamation if the driver defer probes because of
>>> phy unavailability.
>>>
>>> While at it, drop the useless __refdata annotation
>>> to driver structure. If its really needed, it should
>>> be added back with a comment explaining why.
>>>
>>> Signed-off-by: Sekhar Nori <nsekhar@ti.com>
>>> ---
>>>  drivers/pci/host/pci-keystone-dw.c | 2 +-
>>>  drivers/pci/host/pci-keystone.c    | 6 +++---
>>>  2 files changed, 4 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/drivers/pci/host/pci-keystone-dw.c b/drivers/pci/host/pci-keystone-dw.c
>>> index ed34c9520a02..47ddbcce53ca 100644
>>> --- a/drivers/pci/host/pci-keystone-dw.c
>>> +++ b/drivers/pci/host/pci-keystone-dw.c
>>> @@ -469,7 +469,7 @@ void ks_dw_pcie_initiate_link_train(struct keystone_pcie *ks_pcie)
>>>   * and call dw_pcie_v3_65_host_init() API to initialize the Keystone
>>>   * PCI host controller.
>>>   */
>>> -int __init ks_dw_pcie_host_init(struct keystone_pcie *ks_pcie,
>>> +int ks_dw_pcie_host_init(struct keystone_pcie *ks_pcie,
>>>  				struct device_node *msi_intc_np)  
>>
>> when trying to add berlin pcie support and came up one patch
>>
>> http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/384575.html
>>
>> when build testing, I also realized this issue, but keystone relies on
> 
> I mean ks_dw_pcie_host_init() calls hook_fault_code()

Okay, at least in v4.4-rc2, it does not call it. ks_pcie_host_init()
calls it. But your comment is valid. The driver needs more changes to
support probe deferral.

> 
>> hook_fault_code() which is will be freed after boot, so your patch isn't
>> completed.
> 
> And will introduce section mismatch warnings

because the annotation on ks_pcie_host_init() is intact, I did not see
any warnings.

> 
>>
>> And from the keystone driver implementation, it seems assume the driver won't
>> be defer probed.

The reason I worked on this patch is because I am using an out-of-tree
phy driver and that can be built as module and I did see ks_pcie_probe
get deferred. But as you said, this driver is not ready for deferred
probing even with this patch and so this patch should be discarded.

Thanks for your review.

Regards,
Sekhar
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html