Re: [PATCH v2] PCI: Fix crash during pci_dev hot-unplug on pseries KVM guest

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Amit,


I try to follow the option which add a OF flag. If Rob is ok with this, I would suggest to use it instead of V1 patch

diff --git a/drivers/of/dynamic.c b/drivers/of/dynamic.c
index dda6092e6d3a..a401ed0463d9 100644
--- a/drivers/of/dynamic.c
+++ b/drivers/of/dynamic.c
@@ -382,6 +382,11 @@ void of_node_release(struct kobject *kobj)
                               __func__, node);
        }

+       if (of_node_check_flag(node, OF_CREATED_WITH_CSET)) {
+               of_changeset_revert(node->data);
+               of_changeset_destroy(node->data);
+       }
+
        if (node->child)
                pr_err("ERROR: %s() unexpected children for %pOF/%s\n",
                        __func__, node->parent, node->full_name);
@@ -507,6 +512,7 @@ struct device_node *of_changeset_create_node(struct of_changeset *ocs,
        np = __of_node_dup(NULL, full_name);
        if (!np)
                return NULL;
+       of_node_set_flag(np, OF_CREATED_WITH_CSET);
        np->parent = parent;

        ret = of_changeset_attach_node(ocs, np);
diff --git a/drivers/of/unittest.c b/drivers/of/unittest.c
index 445ad13dab98..191ab771d9e8 100644
--- a/drivers/of/unittest.c
+++ b/drivers/of/unittest.c
@@ -895,8 +895,6 @@ static void __init of_unittest_changeset(void)
                 "'%pOF' not added\n", n22);
        of_node_put(np);

-       unittest(!of_changeset_revert(&chgset), "revert failed\n");
-
unittest(!of_find_node_by_path("/testcase-data/changeset/n2/n21"),
                 "'%pOF' still present after revert\n", n21);

@@ -908,8 +906,6 @@ static void __init of_unittest_changeset(void)
        if (!ret)
                unittest(strcmp(propstr, "hello") == 0, "original value not in updated property after revert");

-       of_changeset_destroy(&chgset);
-
        of_node_put(n1);
        of_node_put(n2);
        of_node_put(n21);
diff --git a/drivers/pci/of.c b/drivers/pci/of.c
index 51e3dd0ea5ab..836aaba752bb 100644
--- a/drivers/pci/of.c
+++ b/drivers/pci/of.c
@@ -613,12 +613,10 @@ 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 || !of_node_check_flag(np, OF_CREATED_WITH_CSET))
                return;
        pdev->dev.of_node = NULL;

-       of_changeset_revert(np->data);
-       of_changeset_destroy(np->data);
        of_node_put(np);
 }

diff --git a/include/linux/of.h b/include/linux/of.h
index a0bedd038a05..a46317f6626e 100644
--- a/include/linux/of.h
+++ b/include/linux/of.h
@@ -153,6 +153,7 @@ extern struct device_node *of_stdout;
 #define OF_POPULATED_BUS       4 /* platform bus created for children */
 #define OF_OVERLAY             5 /* allocated for an overlay */
 #define OF_OVERLAY_FREE_CSET   6 /* in overlay cset being freed */
+#define OF_CREATED_WITH_CSET    7 /* created by of_changeset_create_node */

 #define OF_BAD_ADDR    ((u64)-1)

Thanks,

Lizhi

On 7/25/24 10:45, Amit Machhiwal wrote:
Hi Lizhi, Rob,

Sorry for responding late. I got busy with some other things.

On 2024/07/23 02:08 PM, Lizhi Hou wrote:
On 7/23/24 12:54, Rob Herring wrote:
On Tue, Jul 23, 2024 at 12:21 PM Lizhi Hou <lizhi.hou@xxxxxxx> wrote:
On 7/23/24 09:21, Rob Herring wrote:
On Mon, Jul 15, 2024 at 01:52:30PM -0700, Lizhi Hou wrote:
On 7/15/24 11:55, Rob Herring wrote:
On Mon, Jul 15, 2024 at 2:08 AM Amit Machhiwal <amachhiw@xxxxxxxxxxxxx> 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 `of_changeset_create_node()` to
allocate a new node only when the device node doesn't exist and init it
in case it does already. Also, introduce `of_pci_free_node()` to be
called to only revert and destroy the changeset device node that was
created via a call to `of_changeset_create_node()`.

[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@xxxxxxxxxxxxx>
Signed-off-by: Lizhi Hou <lizhi.hou@xxxxxxx>
Signed-off-by: Amit Machhiwal <amachhiw@xxxxxxxxxxxxx>
---
Changes since v1:
        * Included Lizhi's suggested changes on V1
        * Fixed below two warnings from Lizhi's changes and rearranged the cleanup
          part a bit in `of_pci_make_dev_node`
            drivers/pci/of.c:611:6: warning: no previous prototype for ‘of_pci_free_node’ [-Wmissing-prototypes]
              611 | void of_pci_free_node(struct device_node *np)
                  |      ^~~~~~~~~~~~~~~~
            drivers/pci/of.c: In function ‘of_pci_make_dev_node’:
            drivers/pci/of.c:696:1: warning: label ‘out_destroy_cset’ defined but not used [-Wunused-label]
              696 | out_destroy_cset:
                  | ^~~~~~~~~~~~~~~~
        * V1: https://lore.kernel.org/all/20240703141634.2974589-1-amachhiw@xxxxxxxxxxxxx/

     drivers/of/dynamic.c  | 16 ++++++++++++----
     drivers/of/unittest.c |  2 +-
     drivers/pci/bus.c     |  3 +--
     drivers/pci/of.c      | 39 ++++++++++++++++++++++++++-------------
     drivers/pci/pci.h     |  2 ++
     include/linux/of.h    |  1 +
     6 files changed, 43 insertions(+), 20 deletions(-)

diff --git a/drivers/of/dynamic.c b/drivers/of/dynamic.c
index dda6092e6d3a..9bba5e82a384 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 null, allocate a new node. If not, init an
+ *     existing one.
      * @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);
Are we going to rename the function to
of_changeset_create_or_maybe_modify_node()? No. The functions here are
very clear in that they allocate new objects and don't reuse what's
passed in.
Ok. How about keeping of_changeset_create_node unchanged.

Instead, call kzalloc(), of_node_init() and of_changeset_attach_node()

in of_pci_make_dev_node() directly.

A similar example is dlpar_parse_cc_node().


Does this sound better?
No, because really that code should be re-written using of_changeset
API.

My suggestion is add a data pointer to struct of_changeset and then set
that to something to know the data ptr is a changeset and is your
changeset.
I do not fully understand the point. I think the issue is that we do not
know if a given of_node is created by of_pci_make_dev_node(), correct?
Yes.

of_node->data can point to anything. And we do not know if it points a
cset or not.
Right. But instead of checking "of_node->data == of_pci_free_node",
you would just be checking "*(of_node->data) == of_pci_free_node"
if of_node->data is a char* pointer, it would be panic. So I used
of_node->data == of_pci_free_node.

(omitting a NULL check and cast for simplicity). I suppose in theory
that could have a false match, but that could happen in this patch
already.
I think if any other kernel code  put of_pci_free_node to of_node->data, it
can be fixed over there.

Do you mean to add a flag (e.g. OF_PCI_DYNAMIC) to
indicate of_node->data points to cset?
That would be another option, but OF_PCI_DYNAMIC would not be a good
name because that would be a flag bit for every single caller needing
similar functionality. Name it just what it indicates: of_node->data
points to cset

If we have that flag, then possibly the DT core can handle more
clean-up itself like calling detach and freeing the changeset.
Ideally, the flags should be internal to the DT code.
Sure. If you prefer this option, I will propose another fix.

The crash in question is a critical issue that we would want to have a fix for
soon. And while this is still being figured out, is it okay to go with the fix I
proposed in the V1 of this patch?

Thanks,
Amit

Thanks,

Lizhi

Rob




[Index of Archives]     [KVM Development]     [KVM ARM]     [KVM ia64]     [Linux Virtualization]     [Linux USB Devel]     [Linux Video]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [Big List of Linux Books]

  Powered by Linux