Skip to contentSkip to conversation

MSI: Clear the irq_desc's msi pointer on free

PATCH Eric W. Biederman 2 participants 5 messages Mar 22, 2007 – Mar 26, 2007 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 5 lines · Michael Ellerman

Currently we never clear the msi_desc pointer in the irq_desc. This leaves us with a pointer to free'ed memory hanging around. No one seems to have hit this, so presumably other parts of the code are protecting us from ever using the stale pointer .. or we're just lucky, we should still clear it.

Hmm. Maybe. Currently this is done in dynamic_irq_cleanup, at least for everything except sparc64.

Quoted 19 lines

---

 drivers/pci/msi.c |    1 +
 1 file changed, 1 insertion(+)

Index: msi-new/drivers/pci/msi.c
===================================================================

drivers/pci/msi.c +1 -0
@@ -520,6 +520,7 @@ static int msi_free_irq(struct pci_dev* 
 	list_del(&entry->list);
 
 	arch_teardown_msi_irq(irq);
+	set_irq_msi(irq, NULL);
 	kfree(entry);
 
 	if (type == PCI_CAP_ID_MSIX && list_empty(&dev->msi_list))
#2 lore raw
Quoted 10 lines · Eric W. Biederman
Quoted 5 lines · Michael Ellerman

Currently we never clear the msi_desc pointer in the irq_desc. This leaves us with a pointer to free'ed memory hanging around. No one seems to have hit this, so presumably other parts of the code are protecting us from ever using the stale pointer .. or we're just lucky, we should still clear it.

Hmm. Maybe. Currently this is done in dynamic_irq_cleanup, at least for everything except sparc64.

OK, I missed that. I still think we should do it here, otherwise there's a window, however small, where the msi_desc pointer is pointing at freed memory.

cheers

Signature · 8 lines
Michael Ellerman
OzLabs, IBM Australia Development Lab

wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)

We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person
#3 lore raw
Quoted 9 lines · Michael Ellerman
Quoted 8 lines · Eric W. Biederman
Quoted 5 lines · Michael Ellerman

Currently we never clear the msi_desc pointer in the irq_desc. This leaves us with a pointer to free'ed memory hanging around. No one seems to have hit this, so presumably other parts of the code are protecting us from ever using the stale pointer .. or we're just lucky, we should still clear it.

Quoted 6 lines
Quoted 2 lines

Hmm. Maybe. Currently this is done in dynamic_irq_cleanup, at least for everything except sparc64.

OK, I missed that. I still think we should do it here, otherwise there's a window, however small, where the msi_desc pointer is pointing at freed memory.

After following the code through the current cleanup happens before you are proposing, and in fact the irq is return to the set of irq's that can be allocated before you are calling set_irq_msi(irq, NULL).

Therefore you are doing this too late and we need to ensure the architecture code does this in arch_teardown_msi_irq.

Eric

#4 lore raw
Quoted 14 lines · Eric W. Biederman
Quoted 12 lines · Michael Ellerman
Quoted 10 lines · Eric W. Biederman
Quoted 5 lines · Michael Ellerman

Currently we never clear the msi_desc pointer in the irq_desc. This leaves us with a pointer to free'ed memory hanging around. No one seems to have hit this, so presumably other parts of the code are protecting us from ever using the stale pointer .. or we're just lucky, we should still clear it.

Hmm. Maybe. Currently this is done in dynamic_irq_cleanup, at least for everything except sparc64.

Quoted 7 lines
Quoted 3 lines

OK, I missed that. I still think we should do it here, otherwise there's a window, however small, where the msi_desc pointer is pointing at freed memory.

After following the code through the current cleanup happens before you are proposing, and in fact the irq is return to the set of irq's that can be allocated before you are calling set_irq_msi(irq, NULL).

We don't call dynamic_irq_cleanup(), so it never gets done. Perhaps we should be using your dynamic_irq_init/cleanup.

Quoted 2 lines

Therefore you are doing this too late and we need to ensure the architecture code does this in arch_teardown_msi_irq.

As long as the arch teardown routine somehow calls dynamic_irq_cleanup() it should be fine. But I guess it's probably safer to just have all archs do set_msi_irq(irq, NULL) in the teardown.

cheers

Signature · 8 lines
Michael Ellerman
OzLabs, IBM Australia Development Lab

wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)

We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person
#5 lore raw
Quoted 2 lines · Michael Ellerman

We don't call dynamic_irq_cleanup(), so it never gets done. Perhaps we should be using your dynamic_irq_init/cleanup.

It depends. If you are going through another irq controller etc. dynamic_irq_cleanup is probably excessive.

Quoted 6 lines
Quoted 2 lines

Therefore you are doing this too late and we need to ensure the architecture code does this in arch_teardown_msi_irq.

As long as the arch teardown routine somehow calls dynamic_irq_cleanup() it should be fine. But I guess it's probably safer to just have all archs do set_msi_irq(irq, NULL) in the teardown.

Yes. That sounds correct.

Eric