Re: [PATCH] xhci-mtk: Fix NULL pointer dereference with xhci_irq() for shared_hcd

[Mathias Nyman <mathias.nyman@linux.intel.com>]

On 4.3.2020 5.16, Macpaul Lin wrote:
> On Tue, 2020-02-04 at 17:44 +0800, Mathias Nyman wrote:
>> On 1.2.2020 13.20, Macpaul Lin wrote:
>>> On Fri, 2020-01-31 at 16:50 +0200, Mathias Nyman wrote:
>>>> On 17.1.2020 9.41, Macpaul Lin wrote:
>>>>> According to NULL pointer fix: https://tinyurl.com/uqft5ra
>>>>> xhci: Fix NULL pointer dereference with xhci_irq() for shared_hcd
>>>>> The similar issue has also been found in QC activities in Mediatek.
>>>>>
>>>>> Here quote the description from the referenced patch as follows.
>>>>> "Commit ("f068090426ea xhci: Fix leaking USB3 shared_hcd
>>>>> at xhci removal") sets xhci_shared_hcd to NULL without
>>>>> stopping xhci host. This results into a race condition
>>>>> where shared_hcd (super speed roothub) related interrupts
>>>>> are being handled with xhci_irq happens when the
>>>>> xhci_plat_remove is called and shared_hcd is set to NULL.
>>>>> Fix this by setting the shared_hcd to NULL only after the
>>>>> controller is halted and no interrupts are generated."
>>>>>
>>>>> Signed-off-by: Sriharsha Allenki <sallenki@codeaurora.org>
>>>>> Signed-off-by: Macpaul Lin <macpaul.lin@mediatek.com>
>>>>> ---
>>>>>    drivers/usb/host/xhci-mtk.c | 2 +-
>>>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/usb/host/xhci-mtk.c b/drivers/usb/host/xhci-mtk.c
>>>>> index b18a6baef204..c227c67f5dc5 100644
>>>>> --- a/drivers/usb/host/xhci-mtk.c
>>>>> +++ b/drivers/usb/host/xhci-mtk.c
>>>>> @@ -593,11 +593,11 @@ static int xhci_mtk_remove(struct platform_device *dev)
>>>>>    	struct usb_hcd  *shared_hcd = xhci->shared_hcd;
>>>>>    
>>>>>    	usb_remove_hcd(shared_hcd);
>>>>> -	xhci->shared_hcd = NULL;
>>>>>    	device_init_wakeup(&dev->dev, false);
>>>>>    
>>>>>    	usb_remove_hcd(hcd);
>>>>>    	usb_put_hcd(shared_hcd);
>>>>> +	xhci->shared_hcd = NULL;
>>>>>    	usb_put_hcd(hcd);
>>>>>    	xhci_mtk_sch_exit(mtk);
>>>>>    	xhci_mtk_clks_disable(mtk);
>>>>>
>>>>
>>>> Could you share details of the NULL pointer dereference, (backtrace).
>>>
>>> This bug was found by our QA staff while doing 500 times plug-in and
>>> plug-out devices. The backtrace I have was recorded by QA and I didn't
>>> reproduce this issue on my own environment. However, after applied this
>>> patch the issue seems resolve. Here is the backtrace:
>>>
>>> Exception Class: Kernel (KE)
>>> PC is at [<ffffff8008cccbc0>] xhci_irq+0x728/0x2364
>>> LR is at [<ffffff8008ccc788>] xhci_irq+0x2f0/0x2364
>>>
>>> Current Executing Process:
>>> [iptables, 859][netdagent, 770]
>>>
>>> Backtrace:
>>> [<ffffff80080ead58>] __atomic_notifier_call_chain+0xa8/0x130
>>> [<ffffff80080eb6d4>] notify_die+0x84/0xac
>>> [<ffffff800808e874>] die+0x1d8/0x3b8
>>> [<ffffff80080a89b0>] __do_kernel_fault+0x178/0x188
>>> [<ffffff80080a81b4>] do_page_fault+0x44/0x3b0
>>> [<ffffff80080a811c>] do_translation_fault+0x44/0x98
>>> [<ffffff8008080e08>] do_mem_abort+0x4c/0x128
>>> [<ffffff80080832d0>] el1_da+0x24/0x3c
>>> [<ffffff8008cccbc0>] xhci_irq+0x728/0x2364
>>> [<ffffff8008c98804>] usb_hcd_irq+0x2c/0x44
>>> [<ffffff8008179bb0>] __handle_irq_event_percpu+0x26c/0x4a4
>>> [<ffffff8008179ec8>] handle_irq_event+0x5c/0xd0
>>> [<ffffff800817e3c0>] handle_fasteoi_irq+0x10c/0x1e0
>>> [<ffffff80081787b0>] __handle_domain_irq+0x32c/0x738
>>> [<ffffff800808159c>] gic_handle_irq+0x174/0x1c4
>>> [<ffffff8008083cf8>] el0_irq_naked+0x50/0x5c
>>> [<ffffffffffffffff>] 0xffffffffffffffff
>>>
>>
>> Thanks,
>> Could you help me find out which line of code xhci_irq+0x728 is in your case.
>>
>> As Guenter pointed out there is a risk of turning the NULL pointer dereference
>> into a use after free if we just solve this by setting xhci->shared_hcd = NULL
>> later.
>>
>> If you still have that kernel around, and xhci is compiled in:
>> gdb vmlinux
>> gdb li *(xhci_irq+0x728)
>>
> 
> Sorry that I couldn't get back to you soon. The internal code version
> for this issue was really old and a little bit difficult to rewind to
> that version.
> However, I think the following dump might be correct for the code base.
> 
> (gdb) li *(xhci_irq+0x728)
> 0xffffff8008cc8634 is in xhci_irq (*stripped*
> kernel-4.14/drivers/usb/host/xhci.h:1694).
> 1689     */
> 1690    #define XHCI_MAX_REXIT_TIMEOUT_MS       20
> 1691
> 1692    static inline unsigned int hcd_index(struct usb_hcd *hcd)
> 1693    {
> 1694            if (hcd->speed >= HCD_USB3)
> 1695                    return 0;
> 1696            else
> 1697                    return 1;
> 1698    }
> (gdb)
> 
> Thanks
> Macpaul Lin
> 

Ah, it was a 4.14 kernel.
This should be fixed in 4.20 with patch:
1245374e9b83 xhci: handle port status events for removed USB3 hcd

Port arrays/structures were changed completely in 4.18

Something like the below should work for 4.14:

diff --git a/drivers/usb/host/xhci-ring.c b/drivers/usb/host/xhci-ring.c
index 61fa3007a74a..e7367b9f19c5 100644
--- a/drivers/usb/host/xhci-ring.c
+++ b/drivers/usb/host/xhci-ring.c
@@ -1640,6 +1640,12 @@ static void handle_port_status(struct xhci_hcd *xhci,
 	if ((major_revision == 0x03) != (hcd->speed >= HCD_USB3))
 		hcd = xhci->shared_hcd;
 
+	if (!hcd) {
+		xhci_dbg(xhci, "No hcd found for port %u event\n", port_id);
+		bogus_port_status = true;
+		goto cleanup;
+	}
+
 	if (major_revision == 0) {
 		xhci_warn(xhci, "Event for port %u not in "
 				"Extended Capabilities, ignoring.\n",


-Mathias