[ Upstream commit 3cf74230c139f208b7fb313ae0054386eee31a81 ]
If qi_submit_sync() is invoked with 0 invalidation descriptors (for
instance, for DMA draining purposes), we can run into a bug where a
submitting thread fails to detect the completion of invalidation_wait.
Subsequently, this led to a soft lockup. Currently, there is no impact
by this bug on the existing users because no callers are submitting
invalidations with 0 descriptors. This fix will enable future users
(such as DMA drain) calling qi_submit_sync() with 0 count.
Suppose thread T1 invokes qi_submit_sync() with non-zero descriptors, while
concurrently, thread T2 calls qi_submit_sync() with zero descriptors. Both
threads then enter a while loop, waiting for their respective descriptors
to complete. T1 detects its completion (i.e., T1's invalidation_wait status
changes to QI_DONE by HW) and proceeds to call reclaim_free_desc() to
reclaim all descriptors, potentially including adjacent ones of other
threads that are also marked as QI_DONE.
During this time, while T2 is waiting to acquire the qi->q_lock, the IOMMU
hardware may complete the invalidation for T2, setting its status to
QI_DONE. However, if T1's execution of reclaim_free_desc() frees T2's
invalidation_wait descriptor and changes its status to QI_FREE, T2 will
not observe the QI_DONE status for its invalidation_wait and will
indefinitely remain stuck.
This soft lockup does not occur when only non-zero descriptors are
submitted.In such cases, invalidation descriptors are interspersed among
wait descriptors with the status QI_IN_USE, acting as barriers. These
barriers prevent the reclaim code from mistakenly freeing descriptors
belonging to other submitters.
Considered the following example timeline:
T1 T2
========================================
ID1
WD1
while(WD1!=QI_DONE)
unlock
lock
WD1=QI_DONE* WD2
while(WD2!=QI_DONE)
unlock
lock
WD1==QI_DONE?
ID1=QI_DONE WD2=DONE*
reclaim()
ID1=FREE
WD1=FREE
WD2=FREE
unlock
soft lockup! T2 never sees QI_DONE in WD2
Where:
ID = invalidation descriptor
WD = wait descriptor
* Written by hardware
The root of the problem is that the descriptor status QI_DONE flag is used
for two conflicting purposes:
1. signal a descriptor is ready for reclaim (to be freed)
2. signal by the hardware that a wait descriptor is complete
The solution (in this patch) is state separation by using QI_FREE flag
for #1.
Once a thread's invalidation descriptors are complete, their status would
be set to QI_FREE. The reclaim_free_desc() function would then only
free descriptors marked as QI_FREE instead of those marked as
QI_DONE. This change ensures that T2 (from the previous example) will
correctly observe the completion of its invalidation_wait (marked as
QI_DONE).
Signed-off-by: Sanjay K Kumar <sanjay.k.kumar@intel.com>
Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/20240728210059.1964602-1-jacob.jun.pan@linux.intel.com
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit b5e86a95541cea737394a1da967df4cd4d8f7182 ]
Queued invalidation wait descriptor status is volatile in that IOMMU
hardware writes the data upon completion.
Use READ_ONCE() to prevent compiler optimizations which ensures memory
reads every time. As a side effect, READ_ONCE() also enforces strict
types and may add an extra instruction. But it should not have negative
performance impact since we use cpu_relax anyway and the extra time(by
adding an instruction) may allow IOMMU HW request cacheline ownership
easier.
e.g. gcc 12.3
BEFORE:
81 38 ad de 00 00 cmpl $0x2,(%rax)
AFTER (with READ_ONCE())
772f: 8b 00 mov (%rax),%eax
7731: 3d ad de 00 00 cmp $0x2,%eax
//status data is 32 bit
Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Yi Liu <yi.l.liu@intel.com>
Link: https://lore.kernel.org/r/20240607173817.3914600-1-jacob.jun.pan@linux.intel.com
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20240702130839.108139-2-baolu.lu@linux.intel.com
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
