To support dynamic addition and removal of
me clients we add reference counter.
Update kdoc with locking requirements.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add uuid, me_addr addressing also for flow control credits.
The only exception in cases for single buffer clients for which
the host address in flow credits response is always 0
To in order to deal with add/remove race between fw and driver clients
addressing we need to use [uuid, me_addr] tuple to address the clients
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The pointer to client in the callback structure (cb->cl)
can't be NULL with current locking.
We can drop check and warnings as in some cases this just
uselessly complicates the code flow.
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>
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>
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>
Both responses have same flow only the client status
update is different. We introduce handler mei_hbm_cl_res()
that handles both responses
Also we use per client wait queue (cl->wait) rather then
global dev->wait_recvd_msg
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Rename mei_cl_irq_close to mei_cl_irq_disconnect
and MEI_FOP_CLOSE to MEI_FOP_DISCONNECT
Remove unused MEI_FOP_OPEN
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When handling dynamic clients there might be a race
scenario in which two me clients with the same me
address would be linked in the me clients list,
therefore we need to search by both uuid and me address.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To support dynamic addition/remove we add wrappers
for removal of me clients
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>
For support of dynamic addition and removal of me clients
it is more convenient to use a list instead of static array
as is use now.
As the first step of the transition to the new data structure
we change the lookup function so it returns me client address
instead of an index.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Replace open coded loop with an existing service function:
mei_io_list_flush
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>
If connect request is queued (e.g. device in pg) set client state
to initializing, thus avoid preliminary exit in wait if current
state is disconnected.
This is regression from:
commit e4d8270e60
Author: Alexander Usyskin <alexander.usyskin@intel.com>
mei: set connecting state just upon connection request is sent to the fw
CC: stable@vger.kernel.org # 3.15+
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>
wait_event_timeout can return 0 or the remaining jiffies
so return -ETIME if disconnected state not reached.
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>
Link must be reset in case the fw doesn't
respond to client disconnect request.
We did charge the timer only in irq path
from mei_cl_irq_close and not in mei_cl_disconnect
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>
On connection timeout we leave the connecting client in
connecting state. Since a new connection is stalled till
previous connection is completed in this case no new connection
is possible till the user space does release the file handle.
Therefore on timeout we move the client to disconnected state.
Cc: stable@vger.kernel.org # 3.15+
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>
Adding power gating introduced new waiting state for client also during
connection attempt, a connection request can be queued for later either due
device is power gated or due to other on going connection.
We setting client connection state before start of full connect procedure so
in both cased the client state will be MEI_FILE_CONNECTING
which create interlock between the two connection attempts, both
detecting that another connection is in progress.
The interlock is resolved by moving client to connecting state
only upon connection request transmission, so the first
cb in queue can be processed.
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>
Follow-up for bits missed in
commit 7ca96aa278
mei: make return values consistent across 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>
Take rpm token on operation start to initiate rpm resume if needed.
Mark last busy time, release token and advice rpm framework
to try to autosuspend on operation end.
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>
Write callbacks are released on the write completed path but
when file handler is closed before the writes are
completed those are left dangling on write and write_waiting queues.
We add mei_io_list_free function to perform this task
Also move static functions to client.c form client.h
Cc: stable <stable@vger.kernel.org> # 3.11+
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. Propagate ENOTTY to user space if the client is not present
in the system
2. Use ETIME consistently on timeouts
3. Return EIO on write failures
4. Return ENODEV on recoverable device failures such as resets
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 txe use doorbell and not circular buffer
we have to cheat in write slot counting, txe always consume all the
slots upon write. In order for it to work we need to track
slots using mei_hbuf_empty_slots() instead of tracking it in mei layer
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
A client has to acquire host buffer
before writing, we add lock like wrapper
to replace the code snippet
if (dev->hbuf_is_ready)
dev->hbuf_is_ready = false;
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We can use simply list_for_each_entry if there is no
entry removal
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We already have a helper to find me client by id, let's
use it in all relevant places.
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>
Drop not-very-useful check and with this
fix read on index that can be after array end.
Cleanup search function as byproduct.
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>
1. Return -ENOTTY on client connect if the requested client was not found
on the enumeration list or the client was internally disabled, in the later
case FW will return NOT_FOUND.
2. Return -EBUSY if the client cannot be connected because of resource
contention
3. Change response status enum to have MEI_CL_ prefix
4. Add function to translate response status to a string
for more readable logging
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 operation actually only support connection
and not a generic ioctl
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Clear write callbacks sitting in write_waiting list on reset.
Otherwise these callbacks are left dangling and cause memory leak.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Cc: stable <stable@vger.kernel.org>
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>
Fix syntax errors in comments and debug strings
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>
Set hbm header bit 30 for internal commands
This mark commands that are generated by
the device driver
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
list_del_init appears twice in row in mei_cl_unlink, drop one.
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>
1. Perform simple checks first and only then attempt to allocate cl structure.
2. Remove open_handler_count test, this is already checked in mei_cl_link function
3. return -EMFILE instead of -ENOENT as expected by user space
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
we can call list_del_init regardless the client is
linked or not it is always properly initialized
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
this eventually allows as use a single write queue
both for control and data messages and removing possible
race
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
ENODEV will cause application to try to reconnect since
it assumes that device went through the reset
write errors are not always fatal it can happen due to
resource contention
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
define cl_dbg and cl_err macros that add me and host id
prefix for debug and error log messages so we can track
for the client context of the flow.
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>
mei_cl_link is called both from mei_open and also from
in-kernel drivers so we need to protect open_handle_count
from overflow
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Tomas Winkler