On 14/01/2020 3:25 pm, Will Deacon wrote:
On Thu, Dec 19, 2019 at 05:30:31PM +0100, Jean-Philippe Brucker wrote:
Let add_device() clean up after itself. The iommu_bus_init() function
does call remove_device() on error, but other sites (e.g. of_iommu) do
not.
Don't free level-2 stream tables because we'd have to track if we
allocated each of them or if they are used by other endpoints. It's not
worth the hassle since they are managed resources.
Reviewed-by: Eric Auger <eric.auger@xxxxxxxxxx>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx>
Signed-off-by: Jean-Philippe Brucker <jean-philippe@xxxxxxxxxx>
---
drivers/iommu/arm-smmu-v3.c | 28 +++++++++++++++++++++-------
1 file changed, 21 insertions(+), 7 deletions(-)
I think this is alright, with one caveat relating to:
/*
* We _can_ actually withstand dodgy bus code re-calling add_device()
* without an intervening remove_device()/of_xlate() sequence, but
* we're not going to do so quietly...
*/
if (WARN_ON_ONCE(fwspec->iommu_priv)) {
master = fwspec->iommu_priv;
smmu = master->smmu;
} ...
which may be on shakey ground if the subsequent add_device() call can fail
and free stuff that the first one allocated. At least, I don't know what
we're trying to support with this, so it's hard to tell whether or not it
still works as intended after your change.
Hmm, if add_device() ever did fail it should really be expected to
return the device back to an un-added state, so I don't believe that
particular concern should be significant regardless...
How is this supposed to work? I don't recall ever seeing that WARN fire,
so can we just remove this and bail instead? Robin?
However, I am inclined to agree that it's probably better to make it all
moot. Although it indeed should never happen, ISTR at the time there
appeared to be some possible path somewhere by which the notifier may
have been triggered a second time - possibly if some other device failed
or deferred after the first call, triggering the bus code to start all
over again. Since then, though, we've made a lot of changes to how
->add_device usually gets called, plus stuff like the
iommu_device_link() call has snuck in that might not stand up to a
replay anyway, so I don't see any problem with making this condition a
hard failure. It's certainly much easier to reason about.
In fact, there will already be a WARN from iommu_probe_device() now
(because the first call will have set the group), so I don't think we
need any additional diagnostic in the driver any more.
Robin.
Something like below before your changes...
Will
--->8
diff --git a/drivers/iommu/arm-smmu-v3.c b/drivers/iommu/arm-smmu-v3.c
index effe72eb89e7..6ae3df2f3495 100644
--- a/drivers/iommu/arm-smmu-v3.c
+++ b/drivers/iommu/arm-smmu-v3.c
@@ -2534,28 +2534,23 @@ static int arm_smmu_add_device(struct device *dev)
if (!fwspec || fwspec->ops != &arm_smmu_ops)
return -ENODEV;
- /*
- * We _can_ actually withstand dodgy bus code re-calling add_device()
- * without an intervening remove_device()/of_xlate() sequence, but
- * we're not going to do so quietly...
- */
- if (WARN_ON_ONCE(fwspec->iommu_priv)) {
- master = fwspec->iommu_priv;
- smmu = master->smmu;
- } else {
- smmu = arm_smmu_get_by_fwnode(fwspec->iommu_fwnode);
- if (!smmu)
- return -ENODEV;
- master = kzalloc(sizeof(*master), GFP_KERNEL);
- if (!master)
- return -ENOMEM;
- master->dev = dev;
- master->smmu = smmu;
- master->sids = fwspec->ids;
- master->num_sids = fwspec->num_ids;
- fwspec->iommu_priv = master;
- }
+ if (WARN_ON_ONCE(fwspec->iommu_priv))
+ return -EBUSY;
+
+ smmu = arm_smmu_get_by_fwnode(fwspec->iommu_fwnode);
+ if (!smmu)
+ return -ENODEV;
+
+ master = kzalloc(sizeof(*master), GFP_KERNEL);
+ if (!master)
+ return -ENOMEM;
+
+ master->dev = dev;
+ master->smmu = smmu;
+ master->sids = fwspec->ids;
+ master->num_sids = fwspec->num_ids;
+ fwspec->iommu_priv = master;
/* Check the SIDs are in range of the SMMU and our stream table */
for (i = 0; i < master->num_sids; i++) {
