[lkml]   [2017]   [Mar]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH] PCI: rockchip: don't leak the PCI resource list
On Tue, Mar 21, 2017 at 09:25:16AM +0800, Shawn Lin wrote:
> On 2017/3/21 6:49, Brian Norris wrote:
> >This list is local to the probe() function. We should free it up in both
> >the success case and the error case, but currently we're only freeing it
> >in the error case (see commit f1d722b607d6 ("PCI: rockchip: Fix
> >rockchip_pcie_probe() error path to free resource list")).
> >
> >Caught by kmemleak, when doing repeated bind/unbind tests.
> >
> It doesn't look natural to free it in probe, instead it looks more
> like we should do the cleanup work when calling .remove

Hmm, I did write this one up pretty quickly, so I might have gotten it a
bit wrong. It's not clear there's a *real* problem with this patch,
since the bridge's window list doesn't actually get used much later, but
I agree this isn't correct.

And actually, I think my patch is a red herring; the leak is still
there. I notice that the resource list is actually freed in
pci_release_host_bridge_dev() already anyway.

> I didn't know if there is something that still use this resource
> , for instance, hotplug? But I noticed it from Bjron's statement[1]
> that "The struct resource for each host bridge window must live as long
> as the host bridge itself". So I didn't free it when finishing probe.
> I guess the proper fix is to do it in pci_remove_root_bus or somewhere
> of the cleanup code if I undertand it correctly.

That thread is talking about a different issue; in that driver, the
'struct resource' is on the stack. That's not good. In our case, it's
just the resource list head that's on the stack, which is fine. The
relevant entries get spliced onto a longer-lived list in the end.

But then, we have these to consider for leaks:

(a) the 'resource_entry's allocated in
of_pci_get_host_bridge_resources() (via pci_add_resource() and

(b) the 'resource's allocated in
of_pci_get_host_bridge_resources(), to go with each of the
'resource_entry's from (b)

As noted above, I think (a) is actually taken care of (and so my current
patch is not needed). The problem is only with (b). (I think?)

I'll double check this all tomorrow.


> [1]:
> >Signed-off-by: Brian Norris <>
> >---
> > drivers/pci/host/pcie-rockchip.c | 1 +
> > 1 file changed, 1 insertion(+)
> >
> >diff --git a/drivers/pci/host/pcie-rockchip.c b/drivers/pci/host/pcie-rockchip.c
> >index bd6df7254de4..8087a0698d65 100644
> >--- a/drivers/pci/host/pcie-rockchip.c
> >+++ b/drivers/pci/host/pcie-rockchip.c
> >@@ -1396,6 +1396,7 @@ static int rockchip_pcie_probe(struct platform_device *pdev)
> > goto err_free_res;
> > }
> > rockchip->root_bus = bus;
> >+ pci_free_resource_list(&res);
> >
> > pci_bus_size_bridges(bus);
> > pci_bus_assign_resources(bus);
> >

 \ /
  Last update: 2017-03-21 03:30    [W:0.027 / U:17.508 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site