Limit the number of queued writes per client.
Writes above this threshold are blocked till place
in the transmit queue is available.
The limit is configurable via sysfs and defaults to 50.
The implementation should provide blocking I/O behavior.
Prior to this change one would end up in the hands of OOM.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When mei driver is powering down due to suspend or shutdown
it will iterate over the mei client bus and disconnect
each client device attached in turn.
The power down flow consist of the link rest, which causes all clients
get disconnected at once, hence the individual disconnection
can be omitted and significantly reduce power down flow.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Device disabled state is caught inside the retry loop, so
there is no need to check it once again afterwards.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
AMTHIF has special support in the mei drive, it handles multiplexing
multiple user space connection above single me client connection.
Since there is no additional addressing information there is a strict
requirement on the traffic order on each connection and on the "read
after write" order within the connection. This creates a lot of
complexity mostly because the other client types do not necessarily fall
under the same restriction. After carefully studying the use of the
AMTHIF client, we came to conclusion that the multiplexing is not really
utilized by any application and we may safely remove that support and
significantly simplify the driver.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch fixes 'mei: synchronize irq before initiating a reset'
The patch had introduced a deadlock between irq thread and mei_reset()
as they are both holding the same device lock.
---> device_lock:
mei_reset()
<---- interrupt thread
device_lock
---> synchornize_irq()
wait on interrupt thread == (dead lock)
The fix is to call synchronize_irq
prior to call locked mei_reset function.
Cc: <stable@vger.kernel.org> #4.10+
Fixes: f302bb0de6ac (mei: synchronize irq before initiating a reset)
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mei_dev structure used struct mei_cl_cb type variables as for holding
callbacks list heads. Replace them by the actual struct list_head
as there is no other info that is handled. This slims down
the mei_dev structure and mostly streamline the code.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We need to synchronize irqs before issuing reset to make sure that the
clients communication is concluded and doesn't leak to the reset flow
and confusing the state machine.
This issue is happening during suspend/resume stress testing.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The stall timer worker checks periodically if there is a stalled i/o
transaction. The issue with the current implementation is that the timer
is ticking also when there is no pending i/o transaction.
This patch provides a simple change that prevents rescheduling
of the delayed work when there is no pending i/o.
Cc: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The amthif FW client can appear after the end of client enumeration.
Amthif host client initialization is done now at FW client discovery
time.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since clients can be now added and removed during runtime
we need to run bus rescan whenever me_clients list is modified.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Clean write and write_waiting queues in disconnect.
Requests in those queues are stale and processing will lead to
fat warnings.
In multi thread operations on disconnect and in FW disconnect case -
write/read/event waiters should end wait and return error.
Wake all waiters for disconnecting client to achieve that.
Drop wake all and write queue clean on reset,
as now we waking all waiters and cleaning write queues on disconnect.
No need to do it twice.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Now when we have per client rd_completed list we can remove
the amthif specific amthif_rd_complete_list.
In addition in the function mei_amthif_read do not loop over the
rd_completed list like the original code as the code path is unlocked.
Reviewed-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Instead of integrating the iAMT watchdog in the mei core driver
we will create a watchdog device on the mei client bus and
create a driver for it.
This patch removes the watchdog code from the mei core driver.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In process of client devices removal from the bus there still
might be communication between a driver and the mei device
hence we need to cancel supporting workers only after all
the client devices were removed.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
MEI bus was designed around nfc and was hard to extend.
Instead of the hard coded way of adding the devices on the mei bus
we scan the whole me client list and create a device for each
eligible me client (mei_cl_bus_rescan); currently we support
only clients with single connection and fixed address clients.
NFC radio name detection is run as a fixup routine
The patch replaces handling the device list based on struct me_cl
to device list based on me_cl_devices. The creating a connection
is pushed from the device creation time to device enablement.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Instead of holding the list of host clients (me_cl)
we want to keep the list me client devices (mei_cl_device)
This way we can create host to me client connection only when needed.
Add list head to mei_cl_device and cl_bus_lock
Add bus_added flag to the me client (mei_me_client) to track if
the appropriate mei_cl_device was already created and is_added
flag to mei_cl_device to track if it was already added to the device
list across the bus rescans
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In function mei_nfc_host_exit mei_cl_remove_device cannot be called
under the device mutex as device removing flow invokes the device driver
remove handler that calls in turn to mei_cl_disable_device which
naturally acquires the device mutex.
Also remove mei_cl_bus_remove_devices which has the same issue, but is
never executed as currently the only device on the mei client bus is NFC
and a new device cannot be easily added till the bus revamp is
completed.
This fixes regression caused by commit be9b720a0c ("mei_phy: move all
nfc logic from mei driver to nfc")
Prior to this change the nfc driver remove handler called to no-op
disable function while actual nfc device was disabled directly from the
mei driver.
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Samuel Ortiz <sameo@linux.intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The io callback is clear from write_waitling_list after
we receive interrupt from the hw to ack the write completion.
We need to wait for this interrupt deliver before we try
to enter low power state
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Replace clunky read state machine with read stack
implemented as per client read list, this is important
mostly for mei drivers with unsolicited reads
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
1. Use rw lock to access the me_clients list
2. Reuse already defined find functions also when
removing particular me client
3. Add wrappers for addition and deletion
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Set the internal device state to to disabled after hardware reset in stop flow.
This will cover cases when driver was not brought to disabled state because of
an error and in stop flow we wish not to retry the reset.
Cc: <stable@vger.kernel.org> #3.10+
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
ME devices prior to PCH8 (Lynx Point) have two FW status registers,
on PCH8 and newer excluding txe there are six FW status registers.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Remove inclusion of linux/pci.h in mei layer
however we need to include the headers that before
got included implicitly from linux/pci.h.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fw status retrieval has pci specific implementation
so we push it back to the hw layer
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For purpose of adding testing HW we would like
to get rid of pci dependency in generic mei code
This patch provides only straight forward changes
FW status and prob quirks need to be handled separately
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Setting of hw_ops and device has should be in
mei_device_init.
We add reference to the parent device and remove
pci dependent cfg
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fix new style warning: Missing a blank line after declarations
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add hbm state, pg enablement and state to devstate file in debugfs
(<debugfs>/mei/devstate)
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We cannot handle user interrupt in context of hbm start
so we only wait for time out which is reasonably short.
1. Add kdoc
2. Rename state to better reflect its function
3. Simplify wait condition and rename
wait_recvd_msg to wait_hbm_start
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To support dynamic addition/remove of clients
it is more convenient to use list instead of
static array
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add mei_cfg structure that holds per device configuration
data and hooks, as the first step we add firmware
status register offsets
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
we never freed the mei_clients array on driver shutdown
only on reset
add mei_hbm_reset function that wraps the hbm cleanup
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fetch FW status registers, as they are important in
in understanding of FW reset reasons
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The driver needs to check whether the write
queue idle before entering power gating
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since the runtime pm and the internal power gating
cannot be in complete sync in regards to I/O
operations, we need to expose the device
hardware internal power gating state to mei layer
2. We add pg_state handler that translate the hw
internal pg state to mei layer
2. We add power gating event variable to keep
power track of power gating transitions
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Reviewed-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add send message functions and receive dispatch stubs
for power gating isolation hbm protocol.
The protocol consist of requests for entering and exiting
the power gating isolation state and their responses.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When stopping the MEI, we should remove and potentially unregister
all bus devices queued on the mei_dev linked list.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Use more standard message writing for
oob data.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In some rare case mei hw reset may take long time to settle.
Instead of blocking resume flow we span another driver reset flow in
separate work context
This allows as to shorten hw reset timeout to something more acceptable
by DPM_WATCHDOG_TIMEOUT
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
give up reseting after 3 unsuccessful tries
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
1. MEI_DEV_RESETTING device state spans only hardware reset flow
while starting dev state is saved into a local variable for further
reference, this let us to reduce big if statements in case we
are trying to avoid nested resets
2. During initializations if the reset ended in MEI_DEV_DISABLED device
state we bail out with -ENODEV
3. Remove redundant interrupts_enabled parameter as this
can be deduced from the starting dev_state
4. mei_reset propagates error code to the caller
5. Add mei_restart function to wrap the pci resume
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When reset is caused by hbm protocol mismatch or timeout
we might end up in an endless reset loop and hbm protocol
will never sync
Cc: <stable@vger.kernel.org>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This fixes a potential deadlock in case of a firmware
initiated reset
mei_reset has a dialog with the interrupt thread hence
it has to be run from an another work item
Most of the mei_resets were called from mei_hbm_dispatch
which is called in interrupt thread context so this
function underwent major revamp. The error code is
propagated to the interrupt thread and if needed
the reset is scheduled from there.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cancel each work properly and remove flash_work_queue.
Quoting documentation:
In most situations flushing the entire workqueue is overkill; you merely
need to know that a particular work item isn't queued and isn't running.
In such cases you should use cancel_delayed_work_sync() or
cancel_work_sync() instead.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Move the unexpected state print to the beginning of mei_reset,
thus printing right state.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Move host_clients_map cleanup from host client init to device init.
This fixes bug where we cleaned up the bitmask that servers as pool for
host client ids while file descriptors are kept open during suspend.
On resume a new connection will be assigned id that is already taken
by opened file descriptor. Consequently read/write will fail due to
mismatched book keeping.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Make open counter to be incremented and decremented
from mei_cl_link and mei_cl_unlik function respectively
Nfc was assuming symmetric linking API and thus open handler
count was never decreased. This patch fixes that.
We need to add separate open hander count for amthif which
is handled out of link/unlink functions and doesn't break
the symmetric API.
Last we do not waste clients slots if amthif or wd are not present
in the device. we don't need to allocates slots ahead
it is all covered by link/unlink before the devices is responding
to user space connection and thus not racing on allocation
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Alexander Usyskin