Re: [PATCH] PCI: Fix dropping valid root bus resources with .end = zero

[Niklas Schnelle]

On Wed, 2022-07-13 at 11:35 +0200, Geert Uytterhoeven wrote:
> On r8a7791/koelsch:
> 
>     kmemleak: 1 new suspected memory leaks (see /sys/kernel/debug/kmemleak)
>     # cat /sys/kernel/debug/kmemleak
>     unreferenced object 0xc3a34e00 (size 64):
>       comm "swapper/0", pid 1, jiffies 4294937460 (age 199.080s)
>       hex dump (first 32 bytes):
> 	b4 5d 81 f0 b4 5d 81 f0 c0 b0 a2 c3 00 00 00 00  .]...]..........
> 	00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
>       backtrace:
> 	[<fe3aa979>] __kmalloc+0xf0/0x140
> 	[<34bd6bc0>] resource_list_create_entry+0x18/0x38
> 	[<767046bc>] pci_add_resource_offset+0x20/0x68
> 	[<b3f3edf2>] devm_of_pci_get_host_bridge_resources.constprop.0+0xb0/0x390
> 
> When coalescing two resources for a contiguous aperture, the first
> resource is enlarged to cover the full contiguous range, while the
> second resource is marked invalid.  This invalidation is done by
> clearing the flags, start, and end members.
> 
> When adding the initial resources to the bus later, invalid resources
> are skipped.  Unfortunately, the check for an invalid resource considers
> only the end member, causing false positives.
> 
> E.g. on r8a7791/koelsch, root bus resource 0 ("bus 00") is skipped, and
> no longer registered with pci_bus_insert_busn_res() (causing the memory
> leak), nor printed:
> 
>      pci-rcar-gen2 ee090000.pci: host bridge /soc/pci@ee090000 ranges:
>      pci-rcar-gen2 ee090000.pci:      MEM 0x00ee080000..0x00ee08ffff -> 0x00ee080000
>      pci-rcar-gen2 ee090000.pci: PCI: revision 11
>      pci-rcar-gen2 ee090000.pci: PCI host bridge to bus 0000:00
>     -pci_bus 0000:00: root bus resource [bus 00]
>      pci_bus 0000:00: root bus resource [mem 0xee080000-0xee08ffff]
> 
> Fix this by only skipping resources where all of the flags, start, and
> end members are zero.
> 
> Fixes: 7c3855c423b17f6c ("PCI: Coalesce host bridge contiguous apertures")
> Signed-off-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> ---
> Is there any side effect of not registering the root bus resource with
> pci_bus_insert_busn_res()?  This is the resource created by
> of_pci_parse_bus_range(), and thus affects any DT platforms using
> "bus-range = <0 0>".
> 
> Perhaps checking for "!res->flags" would be sufficient?
> 
> I assume this still causes memory leaks on systems where resources are
> coalesced, as the second resource of a contiguous aperture is no longer
> referenced? Perhaps instead of clearing the resource, it should be
> removed from the list (and freed? is it actually safe to do that?)?
> 
> Apparently Johannes had identified the bug before, but didn't realize
> the full impact...
> https://lore.kernel.org/r/5331e942ff28bb191d62bb403b03ceb7d750856c.camel@xxxxxxxxxxxxxxxx/
> ---
>  drivers/pci/probe.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index 17a969942d37033a..be628798d279ada0 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -994,7 +994,7 @@ static int pci_register_host_bridge(struct pci_host_bridge *bridge)
>  	resource_list_for_each_entry_safe(window, n, &resources) {
>  		offset = window->offset;
>  		res = window->res;
> -		if (!res->end)
> +		if (!res->flags && !res->start && !res->end)
>  			continue;
>  
>  		list_move_tail(&window->node, &bridge->windows);

Hi Geert, Hi Bjorn, Hi Kai-Heng,

I just stumbled over this issue on s390 with the below kmemleak
splat[0]. On s390 we currently always have a single PCI bus with bus
number 00 per PCI domain so this is triggered whenever there are PCI
devices attached to the system.

Applying the patch from this mail makes the splat go away and the
'pci_bus 0002:00: root bus resource [bus 00]' message reappear. As this
mail is from July I guess it got lost and this was never picked up ;-(

Looking at the commit message and code I'm left with the same questions
as Geert added below the '---' line, so further changes might be needed
for the coalescing case. Either way without this patch the above if
incorrectly skips the root bus resource for us and I think '!res->end'
is just incorrect as a check for invalidity. Geert's proposed change on
the other hand looks correct to me though I do think that the better
solution would be to drop this if and remove the invalidated resource.
The loop even uses '..._for_each_entry_safe' so I would assume removing
it should be okay? This does look a bit like the expansion case in
acpi_pci_root_validate_resources().

For now feel free to add my:

Tested-by: Niklas Schnelle <schnelle@xxxxxxxxxxxxx>

Thanks,
Niklas
 
[0]:
unreferenced object 0x880e20d0 (size 96):
  comm "swapper/0", pid 1, jiffies 4294937765 (age 3029.430s)
  hex dump (first 32 bytes):
    00 00 03 7f ff b1 ba e0 00 00 03 7f ff b1 ba e0  ................
    00 00 00 00 88 1a 18 30 00 00 00 00 00 00 00 00  .......0........
  backtrace:
    [<00000003c0151e6c>] __kmem_cache_alloc_node+0x204/0x438
    [<00000003c00c0a88>] __kmalloc+0x58/0xf0
    [<00000003bfe34a9e>] resource_list_create_entry+0x2e/0x60
    [<00000003c075ea8a>] pci_add_resource+0x32/0x98
    [<00000003bfe22f14>] zpci_bus_device_register+0x124/0x3c0
    [<00000003bfe1baf2>] zpci_create_device+0x142/0x1d0
    [<00000003bfe1fe4c>] clp_scan_pci_devices+0xfc/0x150
    [<00000003c189a468>] pci_base_init+0x148/0x1b0
    [<00000003bfdd4ab0>] do_one_initcall+0x78/0x388
    [<00000003c189081c>] do_initcalls+0x12c/0x150
    [<00000003c1890b1e>] kernel_init_freeable+0x25e/0x2a0
    [<00000003c0bb5136>] kernel_init+0x2e/0x170
    [<00000003bfdd7914>] __ret_from_fork+0x3c/0x58
    [<00000003c0bc79fa>] ret_from_fork+0xa/0x40