Re: [PATCH] pci: shpchp: set the bridge busmaster if MSI are enabled

Re: [PATCH] pci: shpchp: set the bridge busmaster if MSI are enabled

[Marcel Apfelbaum]

On 19/07/2017 16:36, Marcel Apfelbaum wrote:
> On 18/07/2017 17:12, Aleksandr Bezzubikov wrote:
>> An MSI-based SHPC built in PCI bridges can configure hotplugged devices
>> only if they notify the bridge with MSI.
>> But they can't trigger interrupt without the bridge being busmaster,
>> that's why it should be enabled.
> 
> Hi Aleksandr,
> 
> Hot-plugging an empty bridge does require making it bus-master,
> otherwise we end up with an unusable pci brigde.
> 
>>
>> Signed-off-by: Aleksandr Bezzubikov <zuban32s@gmail.com>
>> ---
>>   drivers/pci/hotplug/shpchp_hpc.c | 2 ++
>>   1 file changed, 2 insertions(+)
>>
>> diff --git a/drivers/pci/hotplug/shpchp_hpc.c 
>> b/drivers/pci/hotplug/shpchp_hpc.c
>> index de0ea47..e5824c7 100644
>> --- a/drivers/pci/hotplug/shpchp_hpc.c
>> +++ b/drivers/pci/hotplug/shpchp_hpc.c
>> @@ -1062,6 +1062,8 @@ int shpc_init(struct controller *ctrl, struct 
>> pci_dev *pdev)
>>           if (rc) {
>>               ctrl_info(ctrl, "Can't get msi for the hotplug 
>> controller\n");
>>               ctrl_info(ctrl, "Use INTx for the hotplug controller\n");
>> +        } else {
>> +            pci_set_master(pdev);
>>           }
>>           rc = request_irq(ctrl->pci_dev->irq, shpc_isr, IRQF_SHARED,
>>
> 
> 
> I am not really familiar with the shpc code,
> but the change looks OK to me.
> 
> Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
> 

Hi,

Since is a fix:
Cc: stable@vger.kernel.org

Thanks,
Marcel

> Thanks,
> Marcel

Re: pci: shpchp: set the bridge busmaster if MSI are enabled

[Michael S. Tsirkin]

On Tue, Jul 18, 2017 at 05:12:25PM +0300, Aleksandr Bezzubikov wrote:
> An MSI-based SHPC built in PCI bridges can configure hotplugged devices
> only if they notify the bridge with MSI.
> But they can't trigger interrupt without the bridge being busmaster,
> that's why it should be enabled.
> 
> Signed-off-by: Aleksandr Bezzubikov <zuban32s@gmail.com>
> Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>


Given there's pci_enable_msi above, not enabling bus master
seems like a very strange thing to do.

Acked-by: Michael S. Tsirkin <mst@redhat.com>

This also looks like a good candidate for stable.

> ---
>  drivers/pci/hotplug/shpchp_hpc.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/pci/hotplug/shpchp_hpc.c b/drivers/pci/hotplug/shpchp_hpc.c
> index de0ea47..e5824c7 100644
> --- a/drivers/pci/hotplug/shpchp_hpc.c
> +++ b/drivers/pci/hotplug/shpchp_hpc.c
> @@ -1062,6 +1062,8 @@ int shpc_init(struct controller *ctrl, struct pci_dev *pdev)
>  		if (rc) {
>  			ctrl_info(ctrl, "Can't get msi for the hotplug controller\n");
>  			ctrl_info(ctrl, "Use INTx for the hotplug controller\n");
> +		} else {
> +			pci_set_master(pdev);
>  		}
>  
>  		rc = request_irq(ctrl->pci_dev->irq, shpc_isr, IRQF_SHARED,