Re: [PATCH 6/9] perf: Generic pci uncore device support

["Yan, Zheng" <zheng.z.yan@intel.com>]

On 05/04/2012 05:46 AM, Peter Zijlstra wrote:
> On Wed, 2012-05-02 at 10:07 +0800, Yan, Zheng wrote:
>> +static void __devexit uncore_pci_remove(struct pci_dev *pdev)
>> +{
>> +       struct intel_uncore_box *box = pci_get_drvdata(pdev);
>> +       int phyid = pcibus_to_phyid[pdev->bus->number];
>> +
>> +       if (WARN_ON_ONCE(phyid != box->phy_id))
>> +               return;
>> +
>> +       box->pci_dev = NULL;
>> +       if (--box->refcnt == 0) {
> 
> This appears completely unserialized (as is all the refcnt stuff in
> patch 4 it seems), now I figure the only way to actually have this pci
> device go away is by hotplug, which is serialized. Still looks very odd.
> 
yes, box->refcnt should always be 1 here. will remove the 'if (--box->refcnt == 0)'

>> +               spin_lock(&uncore_pci_lock);
>> +               hlist_del_rcu(&box->hlist);
>> +               spin_unlock(&uncore_pci_lock);
>> +               kfree_rcu(box, rcu_head);
>> +       }
>> +}