[Skiboot] [PATCH 02/16] core/pci: Add missing lock in set_power_timer

[Andrew Donnellan]

On 9/9/19 2:31 pm, Frederic Barrat wrote:
> set_power_timer() was not using any lock, though it alters the slot
> state and devices found under it. So lock the PHB under which the slot
> is found to avoid concurrent operations.
> 
> Signed-off-by: Frederic Barrat <fbarrat at linux.ibm.com>

This seems sensible

Reviewed-by: Andrew Donnellan <ajd at linux.ibm.com>

> ---
>   core/pci-opal.c | 3 +++
>   1 file changed, 3 insertions(+)
> 
> diff --git a/core/pci-opal.c b/core/pci-opal.c
> index d5209600..175810d0 100644
> --- a/core/pci-opal.c
> +++ b/core/pci-opal.c
> @@ -674,7 +674,9 @@ static void set_power_timer(struct timer *t __unused, void *data,
>   	struct pci_device *pd = slot->pd;
>   	struct dt_node *dn = pd->dn;
>   	uint8_t link;
> +	struct phb *phb = slot->phb;
>   
> +	phb_lock(phb);
>   	switch (slot->state) {
>   	case PCI_SLOT_STATE_SPOWER_START:
>   		if (slot->retries-- == 0) {
> @@ -720,6 +722,7 @@ static void set_power_timer(struct timer *t __unused, void *data,
>   		prlog(PR_ERR, "PCI SLOT %016llx: Unexpected state 0x%08x\n",
>   		      slot->id, slot->state);
>   	}
> +	phb_unlock(phb);
>   }
>   
>   static int64_t opal_pci_set_power_state(uint64_t async_token,
> 

-- 
Andrew Donnellan              OzLabs, ADL Canberra
ajd at linux.ibm.com             IBM Australia Limited