Re: [PATCH] PCI: Fix crash during pci_dev hot-unplug on pseries KVM guest

[Lizhi Hou <lizhi.hou@amd.com>]

On 7/11/24 11:48, Amit Machhiwal wrote:
> On 2024/07/10 09:48 PM, Lizhi Hou wrote:
>> On 7/5/24 12:20, Bjorn Helgaas wrote:
>>> [+cc Lukas, FYI]
>>>
>>> On Wed, Jul 03, 2024 at 07:46:34PM +0530, Amit Machhiwal wrote:
>>>> With CONFIG_PCI_DYNAMIC_OF_NODES [1], a hot-plug and hot-unplug sequence
>>>> of a PCI device attached to a PCI-bridge causes following kernel Oops on
>>>> a pseries KVM guest:
>>>>
>>>>    RTAS: event: 2, Type: Hotplug Event (229), Severity: 1
>>>>    Kernel attempted to read user page (10ec00000048) - exploit attempt? (uid: 0)
>>>>    BUG: Unable to handle kernel data access on read at 0x10ec00000048
>>>>    Faulting instruction address: 0xc0000000012d8728
>>>>    Oops: Kernel access of bad area, sig: 11 [#1]
>>>>    LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA pSeries
>>>> <snip>
>>>>    NIP [c0000000012d8728] __of_changeset_entry_invert+0x10/0x1ac
>>>>    LR [c0000000012da7f0] __of_changeset_revert_entries+0x98/0x180
>>>>    Call Trace:
>>>>    [c00000000bcc3970] [c0000000012daa60] of_changeset_revert+0x58/0xd8
>>>>    [c00000000bcc39c0] [c000000000d0ed78] of_pci_remove_node+0x74/0xb0
>>>>    [c00000000bcc39f0] [c000000000cdcfe0] pci_stop_bus_device+0xf4/0x138
>>>>    [c00000000bcc3a30] [c000000000cdd140] pci_stop_and_remove_bus_device_locked+0x34/0x64
>>>>    [c00000000bcc3a60] [c000000000cf3780] remove_store+0xf0/0x108
>>>>    [c00000000bcc3ab0] [c000000000e89e04] dev_attr_store+0x34/0x78
>>>>    [c00000000bcc3ad0] [c0000000007f8dd4] sysfs_kf_write+0x70/0xa4
>>>>    [c00000000bcc3af0] [c0000000007f7248] kernfs_fop_write_iter+0x1d0/0x2e0
>>>>    [c00000000bcc3b40] [c0000000006c9b08] vfs_write+0x27c/0x558
>>>>    [c00000000bcc3bf0] [c0000000006ca168] ksys_write+0x90/0x170
>>>>    [c00000000bcc3c40] [c000000000033248] system_call_exception+0xf8/0x290
>>>>    [c00000000bcc3e50] [c00000000000d05c] system_call_vectored_common+0x15c/0x2ec
>>>> <snip>
>>>>
>>>> A git bisect pointed this regression to be introduced via [1] that added
>>>> a mechanism to create device tree nodes for parent PCI bridges when a
>>>> PCI device is hot-plugged.
>>>>
>>>> The Oops is caused when `pci_stop_dev()` tries to remove a non-existing
>>>> device-tree node associated with the pci_dev that was earlier
>>>> hot-plugged and was attached under a pci-bridge. The PCI dev header
>>>> `dev->hdr_type` being 0, results a conditional check done with
>>>> `pci_is_bridge()` into false. Consequently, a call to
>>>> `of_pci_make_dev_node()` to create a device node is never made. When at
>>>> a later point in time, in the device node removal path, a memcpy is
>>>> attempted in `__of_changeset_entry_invert()`; since the device node was
>>>> never created, results in an Oops due to kernel read access to a bad
>>>> address.
>>>>
>>>> To fix this issue the patch updates `pci_stop_dev()` to ensure that a
>>>> call to `of_pci_remove_node()` is only made for pci-bridge devices.
>>>>
>>>> [1] commit 407d1a51921e ("PCI: Create device tree node for bridge")
>>>>
>>>> Fixes: 407d1a51921e ("PCI: Create device tree node for bridge")
>>>> Reported-by: Kowshik Jois B S <kowsjois@linux.ibm.com>
>>>> Tested-by: Kowshik Jois B S <kowsjois@linux.ibm.com>
>>>> Signed-off-by: Amit Machhiwal <amachhiw@linux.ibm.com>
>>> Thanks for the patch and testing!  Would like a reviewed-by from
>>> Lizhi.
>> of_pci_make_dev_node() will create of nodes for some endpoint devices (not a
>> bridge) as well. And actually this is the main purpose.
>>
>> Maybe the patch as below would resolve the Oops?
> Thanks for the patch, Lizhi! I tried out this patch and don't see the issue with
> the same. The hot-plug and hot-unplug of PCI device seem to work fine as
> expected.

Cool! Thanks for trying it. Will you re-spin the patch or you would like 
me to create a patch?

Lizhi

>
> ~ Amit
>
>> diff --git a/drivers/of/dynamic.c b/drivers/of/dynamic.c
>> index dda6092e6d3a..3c693b091ecf 100644
>> --- a/drivers/of/dynamic.c
>> +++ b/drivers/of/dynamic.c
>> @@ -492,21 +492,29 @@ struct device_node *__of_node_dup(const struct
>> device_node *np,
>>    * a given changeset.
>>    *
>>    * @ocs: Pointer to changeset
>> + * @np: Pointer to device node. If it is not null, init it directly instead
>> of
>> + *      allocate a new node.
>>    * @parent: Pointer to parent device node
>>    * @full_name: Node full name
>>    *
>>    * Return: Pointer to the created device node or NULL in case of an error.
>>    */
>>   struct device_node *of_changeset_create_node(struct of_changeset *ocs,
>> +                                            struct device_node *np,
>>                                               struct device_node *parent,
>>                                               const char *full_name)
>>   {
>> -       struct device_node *np;
>>          int ret;
>>
>> -       np = __of_node_dup(NULL, full_name);
>> -       if (!np)
>> -               return NULL;
>> +       if (!np) {
>> +               np = __of_node_dup(NULL, full_name);
>> +               if (!np)
>> +                       return NULL;
>> +       } else {
>> +               of_node_set_flag(np, OF_DYNAMIC);
>> +               of_node_set_flag(np, OF_DETACHED);
>> +       }
>> +
>>          np->parent = parent;
>>
>>          ret = of_changeset_attach_node(ocs, np);
>> diff --git a/drivers/of/unittest.c b/drivers/of/unittest.c
>> index 445ad13dab98..087de26852cc 100644
>> --- a/drivers/of/unittest.c
>> +++ b/drivers/of/unittest.c
>> @@ -871,7 +871,7 @@ static void __init of_unittest_changeset(void)
>>          unittest(!of_changeset_add_property(&chgset, parent, ppadd), "fail
>> add prop prop-add\n");
>>          unittest(!of_changeset_update_property(&chgset, parent, ppupdate),
>> "fail update prop\n");
>>          unittest(!of_changeset_remove_property(&chgset, parent, ppremove),
>> "fail remove prop\n");
>> -       n22 = of_changeset_create_node(&chgset, n2, "n22");
>> +       n22 = of_changeset_create_node(&chgset, NULL, n2, "n22");
>>          unittest(n22, "fail create n22\n");
>>          unittest(!of_changeset_add_prop_string(&chgset, n22, "prop-str",
>> "abcd"),
>>                   "fail add prop prop-str");
>> diff --git a/drivers/pci/of.c b/drivers/pci/of.c
>> index 51e3dd0ea5ab..92c079b2e570 100644
>> --- a/drivers/pci/of.c
>> +++ b/drivers/pci/of.c
>> @@ -608,18 +608,28 @@ int devm_of_pci_bridge_init(struct device *dev, struct
>> pci_host_bridge *bridge)
>>
>>   #ifdef CONFIG_PCI_DYNAMIC_OF_NODES
>>
>> +void of_pci_free_node(struct device_node *np)
>> +{
>> +       struct of_changeset *cset;
>> +
>> +       cset = (struct of_changeset *)(np + 1);
>> +
>> +       np->data = NULL;
>> +       of_changeset_revert(cset);
>> +       of_changeset_destroy(cset);
>> +       of_node_put(np);
>> +}
>> +
>>   void of_pci_remove_node(struct pci_dev *pdev)
>>   {
>>          struct device_node *np;
>>
>>          np = pci_device_to_OF_node(pdev);
>> -       if (!np || !of_node_check_flag(np, OF_DYNAMIC))
>> +       if (!np || np->data != of_pci_free_node)
>>                  return;
>>          pdev->dev.of_node = NULL;
>>
>> -       of_changeset_revert(np->data);
>> -       of_changeset_destroy(np->data);
>> -       of_node_put(np);
>> +       of_pci_free_node(np);
>>   }
>>
>>   void of_pci_make_dev_node(struct pci_dev *pdev)
>> @@ -655,14 +665,18 @@ void of_pci_make_dev_node(struct pci_dev *pdev)
>>          if (!name)
>>                  return;
>>
>> -       cset = kmalloc(sizeof(*cset), GFP_KERNEL);
>> -       if (!cset)
>> +       np = kzalloc(sizeof(*np) + sizeof(*cset), GFP_KERNEL);
>> +       if (!np)
>>                  goto out_free_name;
>> +       np->full_name = name;
>> +       of_node_init(np);
>> +
>> +       cset = (struct of_changeset *)(np + 1);
>>          of_changeset_init(cset);
>>
>> -       np = of_changeset_create_node(cset, ppnode, name);
>> +       np = of_changeset_create_node(cset, np, ppnode, NULL);
>>          if (!np)
>> -               goto out_destroy_cset;
>> +               goto out_free_node;
>>
>>          ret = of_pci_add_properties(pdev, cset, np);
>>          if (ret)
>> @@ -672,9 +686,8 @@ void of_pci_make_dev_node(struct pci_dev *pdev)
>>          if (ret)
>>                  goto out_free_node;
>>
>> -       np->data = cset;
>> +       np->data = of_pci_free_node;
>>          pdev->dev.of_node = np;
>> -       kfree(name);
>>
>>          return;
>>
>> diff --git a/include/linux/of.h b/include/linux/of.h
>> index a0bedd038a05..f774459d0d84 100644
>> --- a/include/linux/of.h
>> +++ b/include/linux/of.h
>> @@ -1631,6 +1631,7 @@ static inline int of_changeset_update_property(struct
>> of_changeset *ocs,
>>   }
>>
>>   struct device_node *of_changeset_create_node(struct of_changeset *ocs,
>> +                                            struct device_node *np,
>>                                               struct device_node *parent,
>>                                               const char *full_name);
>>   int of_changeset_add_prop_string(struct of_changeset *ocs,
>>
>> Thanks,
>>
>> Lizhi
>>
>>>> ---
>>>>    drivers/pci/remove.c | 3 ++-
>>>>    1 file changed, 2 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/pci/remove.c b/drivers/pci/remove.c
>>>> index d749ea8250d6..4e51c64af416 100644
>>>> --- a/drivers/pci/remove.c
>>>> +++ b/drivers/pci/remove.c
>>>> @@ -22,7 +22,8 @@ static void pci_stop_dev(struct pci_dev *dev)
>>>>    		device_release_driver(&dev->dev);
>>>>    		pci_proc_detach_device(dev);
>>>>    		pci_remove_sysfs_dev_files(dev);
>>>> -		of_pci_remove_node(dev);
>>>> +		if (pci_is_bridge(dev))
>>>> +			of_pci_remove_node(dev);
>>> IIUC, this basically undoes the work that was done by
>>> of_pci_make_dev_node().
>>>
>>> The call of of_pci_make_dev_node() from pci_bus_add_device() was added
>>> by 407d1a51921e and is conditional on pci_is_bridge(), so it makes
>>> sense to me that the remove needs a similar condition.
>>>
>>>>    		pci_dev_assign_added(dev, false);
>>>>    	}
>>>>
>>>> base-commit: e9d22f7a6655941fc8b2b942ed354ec780936b3e
>>>> -- 
>>>> 2.45.2
>>>>