If the PCI channel has gone down, don't attempt to poke the hardware.
We need to guard every time cxl_whatever_(read|write) is called. This
is because a call to those functions will dereference an offset into an
mmio register, and the mmio mappings get invalidated in the EEH
teardown.
Check in the read/write functions in the header.
We give them the same semantics as usual PCI operations:
- a write to a channel that is down is ignored.
- a read from a channel that is down returns all fs.
Also, we try to access the MMIO space of a vPHB device as part of the
PCI disable path. Because that's a read that bypasses most of our usual
checks, we handle it explicitly.
As far as user visible warnings go:
- Check link state in file ops, return -EIO if down.
- Be reasonably quiet if there's an error in a teardown path,
or when we already know the hardware is going down.
- Throw a big WARN if someone tries to start a CXL operation
while the card is down. This gives a useful stacktrace for
debugging whatever is doing that.
Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
It was discovered that if a process mmaped their problem state area they
were able to access one page more than expected, potentially allowing
them to access the problem state area of an unrelated process.
This was due to a simple off by one error in the mmap fault handler
introduced in 0712dc7e73 ("cxl: Fix issues
when unmapping contexts"), which is fixed in this patch.
Cc: stable@vger.kernel.org
Fixes: 0712dc7e73 ("cxl: Fix issues when unmapping contexts")
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This patch makes the mmap call fail outright if the requested region is
larger than the problem state area assigned to the context so the error
is reported immediately rather than waiting for an attempt to access an
address out of bounds.
Although we never expect users to map more than the assigned problem
state area and are not aware of anyone doing this (other than for
testing), this does have the potential to break users if someone has
used a larger range regardless. I'm submitting it for consideration, but
if this change is not considered acceptable the previous patch is
sufficient to prevent access out of bounds without breaking anyone.
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This reworks contexts lifetimes a bit to enable the kernel API where we may
want to reuse contexts. Here we will want to start and stop contexts without
freeing them.
Start context does the get pid & ctx so stop context will need to do the puts.
Here we move put pid & ctx to the detach context path which will become part of
the stop context path.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Rework __detach_context() and cxl_context_detach() so we can reuse them in the
kernel API.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Add cookie parameter to afu_release_irqs() so that we can pass in a different
cookie than the context structure. This will be useful for other kernel
drivers that want to call this but get their own cookie back in the interrupt
handler.
Update all existing call sites.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The afu fd release path was identified as a significant bottleneck in
the overall performance of cxl. While an optimal AFU design would
minimise the need to close & reopen the AFU fd, it is not always
practical to avoid.
The bottleneck seems to be down to the call to synchronize_rcu(), which
will block until every other thread is guaranteed to be out of an RCU
critical section. Replace it with call_rcu() to free the context
structures later so we can return to the application sooner.
This reduces the time spent in the fd release path from 13356 usec to
13.3 usec - about a 100x speed up.
Reported-by: Fei K Chen <uchen@cn.ibm.com>
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
An issue was introduced with "cxl: Unmap MMIO regions when detaching a
context" (b123429e6a) where closing a
context normally could also unmap the problem state area of other
contexts currently using the AFU.
It was also discovered that after a context's MMIO space had been
unmapped it would read 0s when accessing it, whereas the expected
behaviour was for the access to fail altogether.
In order to address these issues, this patch does two things:
- Forced mmap unmapping is only done when we are forcefully detaching
all contexts, and not in the normal detach path. Since the normal
context close path is tied to the file release any mmaps must have
already been released so we don't need to worry in that case.
- The mmap path now uses a vm_operations_struct with a fault handler.
The fault handler ensures that the context is in started state,
otherwise it fails the access attempt with a SIGBUS.
Fixes: b123429e6a ("cxl: Unmap MMIO regions when detaching a context")
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
If we need to force detach a context (e.g. due to EEH or simply force
unbinding the driver) we should prevent the userspace contexts from
being able to access the Problem State Area MMIO region further, which
they may have mapped with mmap().
This patch unmaps any mapped MMIO regions when detaching a userspace
context.
Cc: stable@vger.kernel.org
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
We had a known sleep while atomic bug if a CXL device was forcefully
unbound while it was in use. This could occur as a result of EEH, or
manually induced with something like this while the device was in use:
echo 0000:01:00.0 > /sys/bus/pci/drivers/cxl-pci/unbind
The issue was that in this code path we iterated over each context and
forcefully detached it with the contexts_lock spin lock held, however
the detach also needed to take the spu_mutex, and call schedule.
This patch changes the contexts_lock to a mutex so that we are not in
atomic context while doing the detach, thereby avoiding the sleep while
atomic.
Also delete the related TODO comment, which suggested an alternate
solution which turned out to not be workable.
Cc: stable@vger.kernel.org
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This is the core of the cxl driver.
It adds support for using cxl cards in the powernv environment only (ie POWER8
bare metal). It allows access to cxl accelerators by userspace using the
/dev/cxl/afuM.N char devices.
The kernel driver has no knowledge of the function implemented by the
accelerator. It provides services to userspace via the /dev/cxl/afuM.N
devices. When a program opens this device and runs the start work IOCTL, the
accelerator will have coherent access to that processes memory using the same
virtual addresses. That process may mmap the device to access any MMIO space
the accelerator provides. Also, reads on the device will allow interrupts to
be received. These services are further documented in a later patch in
Documentation/powerpc/cxl.txt.
Documentation of the cxl hardware architecture and userspace API is provided in
subsequent patches.
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
