| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
The inode can be different in a container, for example, a docker and host
both open the same uacce parent device, which uses the same uacce struct
but different inode, so uacce->inode is not enough.
What's worse, when docker stops, the inode will be destroyed as well,
causing use-after-free in uacce_remove.
So use q->mapping to replace uacce->inode->i_mapping.
Signed-off-by: Weili Qian <[email protected]>
Signed-off-by: Zhangfei Gao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Greg Kroah-Hartman <[email protected]>
|
|
UACCE adds the hardware error isolation feature. To improve service
reliability, some uacce devices that frequently encounter hardware
errors are isolated. Therefore, this feature is added.
Users can configure the hardware error threshold by 'isolate_strategy'
sysfs node. The user space can get the device isolated state by 'isolate'
sysfs node. If the number of device errors exceeds the configured error
threshold, the device will be isolated. It means the uacce device is
unavailable.
Signed-off-by: Kai Ye <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Greg Kroah-Hartman <[email protected]>
|
|
The uacce driver must deal with a possible removal of the parent device
or parent driver module rmmod at any time.
Although uacce_remove(), called on device removal and on driver unbind,
prevents future use of the uacce fops by removing the cdev, fops that
were called before that point may still be running.
Serialize uacce_fops_open() and uacce_remove() with uacce->mutex.
Serialize other fops against uacce_remove() with q->mutex.
Since we need to protect uacce_fops_poll() which gets called on the fast
path, replace uacce->queues_lock with q->mutex to improve scalability.
The other fops are only used during setup.
uacce_queue_is_valid(), checked under q->mutex or uacce->mutex, denotes
whether uacce_remove() has disabled all queues. If that is the case,
don't go any further since the parent device is being removed and
uacce->ops should not be called anymore.
Reported-by: Yang Shen <[email protected]>
Signed-off-by: Zhangfei Gao <[email protected]>
Signed-off-by: Jean-Philippe Brucker <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Greg Kroah-Hartman <[email protected]>
|
|
PASID is defined as a few different types in iommu including "int",
"u32", and "unsigned int". To be consistent and to match with uapi
definitions, define PASID and its variations (e.g. max PASID) as "u32".
"u32" is also shorter and a little more explicit than "unsigned int".
No PASID type change in uapi although it defines PASID as __u64 in
some places.
Suggested-by: Thomas Gleixner <[email protected]>
Signed-off-by: Fenghua Yu <[email protected]>
Signed-off-by: Borislav Petkov <[email protected]>
Reviewed-by: Tony Luck <[email protected]>
Reviewed-by: Lu Baolu <[email protected]>
Acked-by: Felix Kuehling <[email protected]>
Acked-by: Joerg Roedel <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
|
|
The mm_exit() op will be removed from the SVA API. When a process dies
and its mm goes away, the IOMMU driver won't notify device drivers
anymore. Drivers should expect to handle a lot more aborted DMA. On the
upside, it does greatly simplify the queue management.
The uacce_mm struct, that tracks all queues bound to an mm, was only
used by the mm_exit() callback. Remove it.
Signed-off-by: Jean-Philippe Brucker <[email protected]>
Acked-by: Jacob Pan <[email protected]>
Acked-by: Lu Baolu <[email protected]>
Acked-by: Zhangfei Gao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Joerg Roedel <[email protected]>
|
|
When uacce parent device module is removed, user app may
still keep the mmaped area, which can be accessed unsafely.
When rmmod, Parent device driver will call uacce_remove,
which unmap all remaining mapping from user space for safety.
VM_FAULT_SIGBUS is also reported to user space accordingly.
Suggested-by: Dave Jiang <[email protected]>
Signed-off-by: Zhangfei Gao <[email protected]>
Signed-off-by: Herbert Xu <[email protected]>
|
|
Uacce (Unified/User-space-access-intended Accelerator Framework) targets to
provide Shared Virtual Addressing (SVA) between accelerators and processes.
So accelerator can access any data structure of the main cpu.
This differs from the data sharing between cpu and io device, which share
only data content rather than address.
Since unified address, hardware and user space of process can share the
same virtual address in the communication.
Uacce create a chrdev for every registration, the queue is allocated to
the process when the chrdev is opened. Then the process can access the
hardware resource by interact with the queue file. By mmap the queue
file space to user space, the process can directly put requests to the
hardware without syscall to the kernel space.
The IOMMU core only tracks mm<->device bonds at the moment, because it
only needs to handle IOTLB invalidation and PASID table entries. However
uacce needs a finer granularity since multiple queues from the same
device can be bound to an mm. When the mm exits, all bound queues must
be stopped so that the IOMMU can safely clear the PASID table entry and
reallocate the PASID.
An intermediate struct uacce_mm links uacce devices and queues.
Note that an mm may be bound to multiple devices but an uacce_mm
structure only ever belongs to a single device, because we don't need
anything more complex (if multiple devices are bound to one mm, then
we'll create one uacce_mm for each bond).
uacce_device --+-- uacce_mm --+-- uacce_queue
| '-- uacce_queue
|
'-- uacce_mm --+-- uacce_queue
+-- uacce_queue
'-- uacce_queue
Reviewed-by: Greg Kroah-Hartman <[email protected]>
Reviewed-by: Jonathan Cameron <[email protected]>
Signed-off-by: Kenneth Lee <[email protected]>
Signed-off-by: Zaibo Xu <[email protected]>
Signed-off-by: Zhou Wang <[email protected]>
Signed-off-by: Jean-Philippe Brucker <[email protected]>
Signed-off-by: Zhangfei Gao <[email protected]>
Signed-off-by: Herbert Xu <[email protected]>
|