Skip to contentSkip to conversation

PCI: fix recursive device locking

PATCH RESEND Superseded Keith Busch 3 participants 4 messages Jul 9, 2024 – Jul 11, 2024 PCI

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH RESEND 1/1 PCI: fix recursive device locking

From: Keith Busch <[email protected]>

If one of the bus' devices has subordinates, the recursive call locks itself, so no need to lock the device before the recursion or it will surely deadlock.

Fixes dbc5b5c0d268f87 ("PCI: Add missing bridge lock to pci_bus_lock()"

---
Sending to the correct mailing list this time, and fixed up two typos in the commit message.

 drivers/pci/pci.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/pci/pci.c +4 -2
index df550953fa260..5ab13bf5a3caa 100644
@@ -5488,9 +5488,10 @@ static void pci_bus_lock(struct pci_bus *bus)
 
 	pci_dev_lock(bus->self);
 	list_for_each_entry(dev, &bus->devices, bus_list) {
-		pci_dev_lock(dev);
 		if (dev->subordinate)
 			pci_bus_lock(dev->subordinate);
+		else
+			pci_dev_lock(dev);
 	}
 }
 
@@ -5502,7 +5503,8 @@ static void pci_bus_unlock(struct pci_bus *bus)
 	list_for_each_entry(dev, &bus->devices, bus_list) {
 		if (dev->subordinate)
 			pci_bus_unlock(dev->subordinate);
-		pci_dev_unlock(dev);
+		else
+			pci_dev_unlock(dev);
 	}
 	pci_dev_unlock(bus->self);
 }
Signature · 1 line
2.43.0
#1 lore raw
Quoted 13 lines · Keith Busch
drivers/pci/pci.c +2 -1
@@ -5488,9 +5488,10 @@ static void pci_bus_lock(struct pci_bus *bus)
 
 	pci_dev_lock(bus->self);
 	list_for_each_entry(dev, &bus->devices, bus_list) {
-		pci_dev_lock(dev);
 		if (dev->subordinate)
 			pci_bus_lock(dev->subordinate);
+		else
+			pci_dev_lock(dev);
 	}
 }
 

drivers/pci/pci.c static void pci_bus_unlock(struct pci_bus *bus) · line 5510

Quoted 9 lines
drivers/pci/pci.c +2 -1
@@ -5502,7 +5503,8 @@ static void pci_bus_unlock(struct pci_bus *bus)
 	list_for_each_entry(dev, &bus->devices, bus_list) {
 		if (dev->subordinate)
 			pci_bus_unlock(dev->subordinate);
-		pci_dev_unlock(dev);
+		else
+			pci_dev_unlock(dev);
 	}
 	pci_dev_unlock(bus->self);
 }

I realized pci_slot_lock() has the same problem. I wasn't able to test that path from not having a pcie topology with a subordinate on the slot device, but it follows the same pattern. Same thing with pci_bus_trylock() for that matter, so I will make a new version.

#2 lore raw
Quoted 13 lines · Keith Busch
Quoted 12 lines · Keith Busch
drivers/pci/pci.c +2 -1
@@ -5488,9 +5488,10 @@ static void pci_bus_lock(struct pci_bus *bus)
 
 	pci_dev_lock(bus->self);
 	list_for_each_entry(dev, &bus->devices, bus_list) {
-		pci_dev_lock(dev);
 		if (dev->subordinate)
 			pci_bus_lock(dev->subordinate);
+		else
+			pci_dev_lock(dev);
 	}
 }
 
Quoted 15 lines

drivers/pci/pci.c static void pci_bus_unlock(struct pci_bus *bus) · line 5510

Quoted 10 lines
drivers/pci/pci.c +2 -1
@@ -5502,7 +5503,8 @@ static void pci_bus_unlock(struct pci_bus *bus)
 	list_for_each_entry(dev, &bus->devices, bus_list) {
 		if (dev->subordinate)
 			pci_bus_unlock(dev->subordinate);
-		pci_dev_unlock(dev);
+		else
+			pci_dev_unlock(dev);
 	}
 	pci_dev_unlock(bus->self);
 }

I realized pci_slot_lock() has the same problem. I wasn't able to test that path from not having a pcie topology with a subordinate on the slot device, but it follows the same pattern. Same thing with pci_bus_trylock() for that matter, so I will make a new version.

I can take another look at that one as well, but feel free to carry my Reviewed-by tag on that one, and just trust that I'll scream if I notice something late.

#3 lore raw
Quoted 8 lines · Dan Williams
Quoted 4 lines

I realized pci_slot_lock() has the same problem. I wasn't able to test that path from not having a pcie topology with a subordinate on the slot device, but it follows the same pattern. Same thing with pci_bus_trylock() for that matter, so I will make a new version.

I can take another look at that one as well, but feel free to carry my Reviewed-by tag on that one, and just trust that I'll scream if I notice something late.

Disregard, I mistook this as comment about v2 rather than v1. v2 is good to go.