PCI Hotplug: acpiphp: don't store a pci_dev in acpiphp_func
PATCH 1/1 PCI Hotplug: acpiphp: don't store a pci_dev in acpiphp_func
An oops can occur if a user attempts to use both PCI logical hotplug and the ACPI physical hotplug driver (acpiphp) in this sequence, where $slot/address == $device.
In other words, if acpiphp has claimed a PCI device, and that device is logically removed, then acpiphp may oops when it attempts to access it again.
# echo 1 > /sys/bus/pci/devices/$device/remove # echo 0 > /sys/bus/pci/slots/$slot/power
Unable to handle kernel NULL pointer dereference (address 0000000000000000) Call Trace: [<a000000100016390>] show_stack+0x50/0xa0 [<a000000100016c60>] show_regs+0x820/0x860 [<a00000010003b390>] die+0x190/0x2a0 [<a000000100066a40>] ia64_do_page_fault+0x8e0/0xa40 [<a00000010000c7a0>] ia64_native_leave_kernel+0x0/0x270 [<a0000001003b2660>] pci_remove_bus_device+0x120/0x260 [<a0000002060549f0>] acpiphp_disable_slot+0x410/0x540 [acpiphp] [<a0000002060505c0>] disable_slot+0xc0/0x120 [acpiphp] [<a0000002040d21c0>] power_write_file+0x1e0/0x2a0 [pci_hotplug] [<a0000001003bb820>] pci_slot_attr_store+0x60/0xa0 [<a000000100240f70>] sysfs_write_file+0x230/0x2c0 [<a000000100195750>] vfs_write+0x190/0x2e0 [<a0000001001961a0>] sys_write+0x80/0x100 [<a00000010000c600>] ia64_ret_from_syscall+0x0/0x20 [<a000000000010720>] __kernel_syscall_via_break+0x0/0x20
The root cause of this oops is that the logical remove ("echo 1 > /sys/bus/pci/devices/$device/remove") destroyed the pci_dev. The pci_dev struct itself wasn't deallocated because acpiphp kept a reference, but some of its fields became invalid.
acpiphp doesn't have any real reason to keep a pointer to a pci_dev around. It can always derive it using pci_get_slot().
If a logical remove destroys the pci_dev, acpiphp won't find it and is thus prevented from causing mischief.
--- acpiphp.h | 1 acpiphp_glue.c | 63 +++++++++++++++++++++++---------------------------------- 2 files changed, 26 insertions(+), 38 deletions(-)
---
drivers/pci/hotplug/acpiphp.h +0 -1
@@ -129,7 +129,6 @@ struct acpiphp_func {struct acpiphp_bridge *bridge;/* Ejectable PCI-to-PCI bridge */ struct list_head sibling; -struct pci_dev *pci_dev;struct notifier_block nb;acpi_handle handle;
drivers/pci/hotplug/acpiphp_glue.c +26 -37
132 lines not on this page · raw message
Quoted 15 lines · Alex Chiang
In other words, if acpiphp has claimed a PCI device, and that device is logically removed, then acpiphp may oops when it attempts to access it again.
# echo 1 > /sys/bus/pci/devices/$device/remove # echo 0 > /sys/bus/pci/slots/$slot/powerThe root cause of this oops is that the logical remove ("echo 1 >
/sys/bus/pci/devices/$device/remove") destroyed thepci_dev. Thepci_devstruct itself wasn't deallocated because acpiphp kept a reference, but some of its fields became invalid.acpiphp doesn't have any real reason to keep a pointer to a
pci_devaround. It can always derive it usingpci_get_slot().
Quoted 6 lines
If a logical remove destroys the
pci_dev, acpiphp won't find it and is thus prevented from causing mischief.Reported-by Kenji Kaneshige <[email protected]>Acked-by Bjorn Helgaas <[email protected]>Signed-off-by Alex Chiang <[email protected]>
I think this is the right approach. The more minimal way to fix this would be to check that the pdev was valid before destroying it ... but I approve of deleting more code from acpiphp ;-) You do end up doing slightly more work in the remove case, but this is such an infrequent operation that it really doesn't matter.
Signature · 4 lines
Matthew Wilcox Intel Open Source Technology Centre "Bill, look, we understand that you're interested in selling us this operating system, but compare it to ours. We can't possibly take such a retrograde step."
Quoted 193 lines · Alex Chiang
An oops can occur if a user attempts to use both PCI logical hotplug and the ACPI physical hotplug driver (acpiphp) in this sequence, where $slot/address == $device.
In other words, if acpiphp has claimed a PCI device, and that device is logically removed, then acpiphp may oops when it attempts to access it again.
# echo 1 > /sys/bus/pci/devices/$device/remove # echo 0 > /sys/bus/pci/slots/$slot/powerUnable to handle kernel NULL pointer dereference (address 0000000000000000) Call Trace: [<a000000100016390>] show_stack+0x50/0xa0 [<a000000100016c60>] show_regs+0x820/0x860 [<a00000010003b390>] die+0x190/0x2a0 [<a000000100066a40>] ia64_do_page_fault+0x8e0/0xa40 [<a00000010000c7a0>] ia64_native_leave_kernel+0x0/0x270 [<a0000001003b2660>] pci_remove_bus_device+0x120/0x260 [<a0000002060549f0>] acpiphp_disable_slot+0x410/0x540 [acpiphp] [<a0000002060505c0>] disable_slot+0xc0/0x120 [acpiphp] [<a0000002040d21c0>] power_write_file+0x1e0/0x2a0 [pci_hotplug] [<a0000001003bb820>] pci_slot_attr_store+0x60/0xa0 [<a000000100240f70>] sysfs_write_file+0x230/0x2c0 [<a000000100195750>] vfs_write+0x190/0x2e0 [<a0000001001961a0>] sys_write+0x80/0x100 [<a00000010000c600>] ia64_ret_from_syscall+0x0/0x20 [<a000000000010720>] __kernel_syscall_via_break+0x0/0x20The root cause of this oops is that the logical remove ("echo 1 >
/sys/bus/pci/devices/$device/remove") destroyed thepci_dev. Thepci_devstruct itself wasn't deallocated because acpiphp kept a reference, but some of its fields became invalid.acpiphp doesn't have any real reason to keep a pointer to a
pci_devaround. It can always derive it usingpci_get_slot().If a logical remove destroys the
pci_dev, acpiphp won't find it and is thus prevented from causing mischief.Reported-by Kenji Kaneshige <[email protected]>Acked-by Bjorn Helgaas <[email protected]>Signed-off-by Alex Chiang <[email protected]>--- acpiphp.h | 1 acpiphp_glue.c | 63 +++++++++++++++++++++++---------------------------------- 2 files changed, 26 insertions(+), 38 deletions(-)---
drivers/pci/hotplug/acpiphp.h +0
-1@@ -129,7 +129,6 @@ struct acpiphp_func {struct acpiphp_bridge *bridge;/* Ejectable PCI-to-PCI bridge */ struct list_head sibling; -struct pci_dev *pci_dev;struct notifier_block nb;acpi_handle handle;drivers/pci/hotplug/acpiphp_glue.c +24
-35@@ -32,9 +32,6 @@/* * Lifetime rules for pci_dev: -* - The one in acpiphp_func has its refcount elevated by pci_get_slot() -* when the driver is loaded or when an insertion event occurs. It loses -* a refcount when its ejected or the driver unloads. * - The one in acpiphp_bridge has its refcount elevated by pci_get_slot() * when the bridge is scanned and it loses a refcount when the bridge * is removed. @@ -130,6 +127,7 @@ register_slot(acpi_handle handle, u32 lvl, void *context, void **rv)unsigned long long adr, sun;int device, function, retval;struct pci_bus *pbus = bridge->pci_bus; +struct pci_dev *pdev;if (!acpi_pci_check_ejectable (pbus, handle) && !is_dock_device (handle))return AE_OK ;@@ -213,10 +211,10 @@ register_slot(acpi_handle handle, u32 lvl, void *context, void **rv) newfunc->slot = slot;list_add_tail (&newfunc->sibling, &slot->funcs); -/* associate corresponding pci_dev */ - newfunc->pci_dev =pci_get_slot (pbus,PCI_DEVFN (device, function)); -if (newfunc->pci_dev) { + pdev =pci_get_slot (pbus,PCI_DEVFN (device, function)); +if (pdev) { slot->flags |= (SLOT_ENABLED |SLOT_POWEREDON ); +pci_dev_put (pdev); }if (is_dock_device (handle)) {@@ -617,7 +615,6 @@ static void cleanup_bridge(struct acpiphp_bridge *bridge)if (ACPI_FAILURE (status))err ("failed to remove notify handler\n" ); } -pci_dev_put (func->pci_dev);list_del (list);kfree (func); }@@ -1101,22 +1098,24 @@ static int __ref enable_device(struct acpiphp_slot *slot)pci_enable_bridges (bus);pci_bus_add_devices (bus); -/* associate pci_dev to our representation */ list_for_each (l, &slot->funcs) { func = list_entry(l, struct acpiphp_func, sibling); - func->pci_dev =pci_get_slot (bus,PCI_DEVFN (slot->device, - func->function)); -if (!func->pci_dev) + dev =pci_get_slot (bus,PCI_DEVFN (slot->device, + func->function)); +if (!dev)continue ; -if (func->pci_dev->hdr_type !=PCI_HEADER_TYPE_BRIDGE && - func->pci_dev->hdr_type !=PCI_HEADER_TYPE_CARDBUS ) +if (dev->hdr_type !=PCI_HEADER_TYPE_BRIDGE && + dev->hdr_type !=PCI_HEADER_TYPE_CARDBUS ) { +pci_dev_put (dev);continue ; + } status =find_p2p_bridge (func->handle, (u32 )1 , bus,NULL );if (ACPI_FAILURE (status))warn ("find_p2p_bridge failed (error code = 0x%x)\n" , status); +pci_dev_put (dev); } slot->flags |=SLOT_ENABLED ;@@ -1142,17 +1141,14 @@ static void disable_bridges(struct pci_bus *bus)*/ static int disable_device (struct acpiphp_slot *slot) { -int retval =0 ;struct acpiphp_func *func; -struct list_head *l; +struct pci_dev *pdev;/* is this slot already disabled? */ if (!(slot->flags &SLOT_ENABLED )) goto err_exit; - list_for_each (l, &slot->funcs) { - func = list_entry(l, struct acpiphp_func, sibling); - +list_for_each_entry (func, &slot->funcs, sibling) {if (func->bridge) {/* cleanup p2p bridges under this P2P bridge */ cleanup_p2p_bridge(func->bridge->handle,@@ -1160,35 +1156,28 @@ static int disable_device(struct acpiphp_slot *slot) func->bridge =NULL ; } -if (func->pci_dev) { -pci_stop_bus_device (func->pci_dev); -if (func->pci_dev->subordinate) { -disable_bridges (func->pci_dev->subordinate); -pci_disable_device (func->pci_dev); + pdev =pci_get_slot (slot->bridge->pci_bus, +PCI_DEVFN (slot->device, func->function)); +if (pdev) { +pci_stop_bus_device (pdev); +if (pdev->subordinate) { +disable_bridges (pdev->subordinate); +pci_disable_device (pdev); } +pci_remove_bus_device (pdev); +pci_dev_put (pdev); } } - list_for_each (l, &slot->funcs) { - func = list_entry(l, struct acpiphp_func, sibling); - +list_for_each_entry (func, &slot->funcs, sibling) {acpiphp_unconfigure_ioapics (func->handle);acpiphp_bus_trim (func->handle); -/* try to remove anyway. -* acpiphp_bus_add might have been failed */ - -if (!func->pci_dev) -continue ; - -pci_remove_bus_device (func->pci_dev); -pci_dev_put (func->pci_dev); - func->pci_dev =NULL ;
drivers/pci/hotplug/acpiphp_glue.c static int disable_device(struct acpiphp_slot *slot) · line 1182
Quoted 10 lines
drivers/pci/hotplug/acpiphp_glue.c +2
-2} slot->flags &= (~SLOT_ENABLED ); - err_exit: -return retval; +err_exit: +return 0 ; }
The patch looks good to me. I confirmed it fixes the kernel oops problem on my test environment.
BTW, I think the ACPI PCI hotplug driver ('acpiphp') still has some problems because of the lack of mechanism to clean up the slot when its parent bridge is removed by logical hotplug mechanism. The '/sys/bus/pci/slots/<SLOT#>/' directory by acpiphp still remains even after its parent bridge is removed, and user can still read from or write to files under the directory. At this point, acpiphp handles the slot with referring old data structures (ex. struct pci_bus) that are already removed from the PCI core, but not freed yet. I think this behavior would cause some problems like below. The Standard Hot-Plug Controller driver ('shpchp') and the PCI Express Hot-Plug driver ('pciehp') don't have this kind of problem because they clean up the slot if its parent bridge is removed. On the other hand, the ACPI pci slot detection driver ('pci_slot') also have the similar problem.
o There are PCI slots that can be handled by several hotplug drivers (ex. acpiphp or pciehp). The drivers/pci/slot.c provide a mechanism to prevent the PCI slot to be handled by multiple hotplug drivers at the same time. But it would no longer work if the parent bridge of slots handled by acpiphp is removed and added again by logical hotplug mechanism. Please see below (note that PCI hotplug slots on my environment can be handled by both acpiphp and pciehp).
# /sbin/modprobe acpiphp # ls /sys/bus/pci/slots/ 1 2 3 4 # echo 1 > /sys/bus/pci/devices/0000\:2f\:04.0/remove # Remove parent bridge of slot '1'. # ls /sys/bus/pci/slots/ 1 2 3 4 # Slot '1' still remains. # echo 1 > /sys/bus/pci/rescan # Add parent bridge of slot '1'. # ls /sys/bus/pci/slots/ 1 2 3 4 # /sbin/modprobe pciehp # Load pciehp. # ls /sys/bus/pci/slots/ # Slot '1' is managed acpiphp and 1 1-1 2 3 4 # '1-1' is managed pciehp # cat /sys/bus/pci/slots/1/address 0000:40:00 # cat /sys/bus/pci/slots/1-1/address # Slot '1' and '1-1' correspond to 0000:40:00 # the same physical slot.
o I think something wrong would happen by the following steps because some PCI core API (pci_scan_slot(), pci_bus_add_devices()) and so on) will be called with old data structures (ex. struct pci_bus) that are already removed from PCI core.
(1) Remove parent bridge of the slot
(2) echo 0 > /sys/bus/pci/slots/<SLOT#>/power
(3) echo 1 > /sys/bys/pci/slots/<SLOT#>/power
Thanks,
Kenji Kaneshige
Quoted 219 lines · On Tue, 26 May 2009 08:55:22 +0900 Kenji Kaneshige…
Quoted 204 lines · Alex Chiang
An oops can occur if a user attempts to use both PCI logical hotplug and the ACPI physical hotplug driver (acpiphp) in this sequence, where $slot/address == $device.
In other words, if acpiphp has claimed a PCI device, and that device is logically removed, then acpiphp may oops when it attempts to access it again.
# echo 1 > /sys/bus/pci/devices/$device/remove # echo 0 > /sys/bus/pci/slots/$slot/powerUnable to handle kernel NULL pointer dereference (address 0000000000000000) Call Trace: [<a000000100016390>] show_stack+0x50/0xa0 [<a000000100016c60>] show_regs+0x820/0x860 [<a00000010003b390>] die+0x190/0x2a0 [<a000000100066a40>] ia64_do_page_fault+0x8e0/0xa40 [<a00000010000c7a0>] ia64_native_leave_kernel+0x0/0x270 [<a0000001003b2660>] pci_remove_bus_device+0x120/0x260 [<a0000002060549f0>] acpiphp_disable_slot+0x410/0x540 [acpiphp] [<a0000002060505c0>] disable_slot+0xc0/0x120 [acpiphp] [<a0000002040d21c0>] power_write_file+0x1e0/0x2a0 [pci_hotplug] [<a0000001003bb820>] pci_slot_attr_store+0x60/0xa0 [<a000000100240f70>] sysfs_write_file+0x230/0x2c0 [<a000000100195750>] vfs_write+0x190/0x2e0 [<a0000001001961a0>] sys_write+0x80/0x100 [<a00000010000c600>] ia64_ret_from_syscall+0x0/0x20 [<a000000000010720>] __kernel_syscall_via_break+0x0/0x20The root cause of this oops is that the logical remove ("echo 1 >
/sys/bus/pci/devices/$device/remove") destroyed thepci_dev. Thepci_devstruct itself wasn't deallocated because acpiphp kept a reference, but some of its fields became invalid.acpiphp doesn't have any real reason to keep a pointer to a
pci_devaround. It can always derive it usingpci_get_slot().If a logical remove destroys the
pci_dev, acpiphp won't find it and is thus prevented from causing mischief.Reported-by Kenji Kaneshige <[email protected]>Acked-by Bjorn Helgaas <[email protected]>Signed-off-by Alex Chiang <[email protected]>--- acpiphp.h | 1 acpiphp_glue.c | 63 +++++++++++++++++++++++---------------------------------- 2 files changed, 26 insertions(+), 38 deletions(-)---
(quoted diff) +0
-0
b/drivers/pci/hotplug/acpiphp.hindex 4fc168b..e68d5f2 100644drivers/pci/hotplug/acpiphp.h +0
-0@@ -129,7 +129,6 @@ struct acpiphp_func {struct acpiphp_bridge *bridge;/* Ejectable PCI-to-PCI bridge */
drivers/pci/hotplug/acpiphp.h +0
-1struct list_head sibling; -struct pci_dev *pci_dev;struct notifier_block nb;acpi_handle handle;(quoted diff) +0
-0
b/drivers/pci/hotplug/acpiphp_glue.cindex a33794d..3a6064b 100644drivers/pci/hotplug/acpiphp_glue.c +0
-2@@ -32,9 +32,6 @@/* * Lifetime rules for pci_dev: -* - The one in acpiphp_func has its refcount elevated by pci_get_slot() -* when the driver is loaded or when an insertion event occurs. It loses
- * a refcount when its ejected or the driver unloads.
* - The one inacpiphp_bridgehas its refcount elevated bypci_get_slot()
* when the bridge is scanned and it loses a refcount when the
bridge
* is removed.drivers/pci/hotplug/acpiphp_glue.c +4
-3@@ -130,6 +127,7 @@ register_slot(acpi_handle handle, u32 lvl, void *context, void **rv) unsigned long long adr, sun;int device, function, retval;struct pci_bus *pbus = bridge->pci_bus; +struct pci_dev *pdev;if (!acpi_pci_check_ejectable (pbus, handle) && !is_dock_device (handle))return AE_OK ;@@ -213,10 +211,10 @@ register_slot(acpi_handle handle, u32 lvl, void *context, void **rv) newfunc->slot = slot;list_add_tail (&newfunc->sibling, &slot->funcs); -/* associate corresponding pci_dev */ - newfunc->pci_dev =pci_get_slot (pbus,PCI_DEVFN (device, function)); -if (newfunc->pci_dev) { + pdev =pci_get_slot (pbus,PCI_DEVFN (device, function)); +if (pdev) { slot->flags |= (SLOT_ENABLED |SLOT_POWEREDON ); +pci_dev_put (pdev); }drivers/pci/hotplug/acpiphp_glue.c +3
-6if (is_dock_device (handle)) {@@ -617,7 +615,6 @@ static void cleanup_bridge(struct acpiphp_bridge *bridge) if (ACPI_FAILURE(status)) err("failed to remove notify handler\n"); } - pci_dev_put(func->pci_dev); list_del(list); kfree(func); }@@ -1101,22 +1098,24 @@ static int __ref enable_device(struct acpiphp_slot *slot) pci_enable_bridges(bus);pci_bus_add_devices (bus); -/* associate pci_dev to our representation */ list_for_each (l, &slot->funcs) { func = list_entry(l, struct acpiphp_func, sibling); - func->pci_dev =pci_get_slot (bus,PCI_DEVFN (slot->device, - func->function)); -if (!func->pci_dev) + dev =pci_get_slot (bus,PCI_DEVFN (slot->device, + func->function)); +if (!dev)continue ; - if (func->pci_dev->hdr_type !=
PCI_HEADER_TYPE_BRIDGE&&drivers/pci/hotplug/acpiphp_glue.c +0
-1- func->pci_dev->hdr_type !=
PCI_HEADER_TYPE_CARDBUS)drivers/pci/hotplug/acpiphp_glue.c +4
-0+if (dev->hdr_type !=PCI_HEADER_TYPE_BRIDGE && + dev->hdr_type !=PCI_HEADER_TYPE_CARDBUS ) { +pci_dev_put (dev);continue ; + } status = find_p2p_bridge(func->handle, (u32)1,bus, NULL); if (
ACPI_FAILURE(status))
warn("find_p2p_bridge failed (error code =
0x%x)\n", status);drivers/pci/hotplug/acpiphp_glue.c +13
-21+pci_dev_put (dev); } slot->flags |=SLOT_ENABLED ;@@ -1142,17 +1141,14 @@ static void disable_bridges(struct pci_bus*bus) */ static int disable_device (struct acpiphp_slot *slot) { -int retval =0 ;struct acpiphp_func *func; -struct list_head *l; +struct pci_dev *pdev;/* is this slot already disabled? */ if (!(slot->flags &SLOT_ENABLED )) goto err_exit; - list_for_each (l, &slot->funcs) { - func = list_entry(l, struct acpiphp_func, sibling); - +list_for_each_entry (func, &slot->funcs, sibling) {if (func->bridge) {/* cleanup p2p bridges under this P2P bridge */ cleanup_p2p_bridge(func->bridge->handle,@@ -1160,35 +1156,28 @@ static int disable_device(struct acpiphp_slot *slot) func->bridge = NULL; } -if (func->pci_dev) { -pci_stop_bus_device (func->pci_dev); -if (func->pci_dev->subordinate) { -disable_bridges (func->pci_dev->subordinate); -pci_disable_device (func->pci_dev); + pdev =pci_get_slot (slot->bridge->pci_bus, +PCI_DEVFN (slot->device, func->function)); +if (pdev) { +pci_stop_bus_device (pdev); +if (pdev->subordinate) { +disable_bridges (pdev->subordinate); +pci_disable_device (pdev); } +pci_remove_bus_device (pdev); +pci_dev_put (pdev); } } - list_for_each (l, &slot->funcs) { - func = list_entry(l, struct acpiphp_func, sibling); - +list_for_each_entry (func, &slot->funcs, sibling) {acpiphp_unconfigure_ioapics (func->handle);acpiphp_bus_trim (func->handle); -/* try to remove anyway. -* acpiphp_bus_add might have been failed */ - -if (!func->pci_dev) -continue ; - -pci_remove_bus_device (func->pci_dev); -pci_dev_put (func->pci_dev);drivers/pci/hotplug/acpiphp_glue.c static int disable_device(struct · line 1183
Quoted 10 lines
drivers/pci/hotplug/acpiphp_glue.c +2
-3- func->pci_dev =NULL ; } slot->flags &= (~SLOT_ENABLED ); - err_exit: -return retval; +err_exit: +return 0 ; }The patch looks good to me. I confirmed it fixes the kernel oops problem on my test environment.
Quoted 2 lines
Reviewed-by Kenji Kaneshige <[email protected]>Tested-by Kenji Kaneshige <[email protected]>
Thanks guys, I just pushed this to my for-linus tree. Assuming Linus hasn't released yet, I'll send this over to him tomorrow.
Thanks,
Jesse