linux-2.6-microblaze.git
3 years agocoresight: tpiu: Allow tpiu to be built as a module
Kim Phillips [Mon, 28 Sep 2020 16:35:00 +0000 (10:35 -0600)]
coresight: tpiu: Allow tpiu to be built as a module

Allow to build coresight-tpiu as a module, for ease of development.

- Kconfig becomes a tristate, to allow =m
- add a tpiu_remove function, for module unload
- add a MODULE_DEVICE_TABLE for autoloading on boot

Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Suzuki K Poulose <Suzuki.Poulose@arm.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Russell King <linux@armlinux.org.uk>
Reviewed-by: Mike Leach <mike.leach@linaro.org>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Kim Phillips <kim.phillips@arm.com>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-13-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: etb: Allow etb to be built as a module
Kim Phillips [Mon, 28 Sep 2020 16:34:59 +0000 (10:34 -0600)]
coresight: etb: Allow etb to be built as a module

Allow to build coresight-etb10 as a module, for ease of development.

- Kconfig becomes a tristate, to allow =m
- add an etb_remove function, for module unload
- add a MODULE_DEVICE_TABLE for autoloading on boot

Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Suzuki K Poulose <Suzuki.Poulose@arm.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Russell King <linux@armlinux.org.uk>
Tested-by: Mike Leach <mike.leach@linaro.org>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Kim Phillips <kim.phillips@arm.com>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-12-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: etm4x: Allow etm4x to be built as a module
Kim Phillips [Mon, 28 Sep 2020 16:34:58 +0000 (10:34 -0600)]
coresight: etm4x: Allow etm4x to be built as a module

Allow to build coresight-etm4x as a module, for ease of development.

- Kconfig becomes a tristate, to allow =m
- append -core to source file name to allow module to
  be called coresight-etm4x by the Makefile
- add an etm4_remove function, for module unload
- add a MODULE_DEVICE_TABLE for autoloading on boot
- delay advertising the per-cpu etmdrvdata
- protect etmdrvdata[] by modifying it on relevant CPU

Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Suzuki K Poulose <Suzuki.Poulose@arm.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Russell King <linux@armlinux.org.uk>
Tested-by: Mike Leach <mike.leach@linaro.org>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Kim Phillips <kim.phillips@arm.com>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-11-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: etm3x: Allow etm3x to be built as a module
Kim Phillips [Mon, 28 Sep 2020 16:34:57 +0000 (10:34 -0600)]
coresight: etm3x: Allow etm3x to be built as a module

Allow to build coresight-etm3x as a module, for ease of development.

- Kconfig becomes a tristate, to allow =m
- append -core to source file name to allow module to
  be called coresight-etm3x by the Makefile
- add an etm_remove function, for module unload
- add a MODULE_DEVICE_TABLE for autoloading on boot
- delay advertising the per-cpu etmdrvdata
- protect etmdrvdata[] by modifying it on relevant CPU

Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Suzuki K Poulose <Suzuki.Poulose@arm.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Russell King <linux@armlinux.org.uk>
Reviewed-by: Mike Leach <mike.leach@linaro.org>
Signed-off-by: Kim Phillips <kim.phillips@arm.com>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-10-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: etm: perf: Fix warning caused by etm_setup_aux failure
Tingwei Zhang [Mon, 28 Sep 2020 16:34:56 +0000 (10:34 -0600)]
coresight: etm: perf: Fix warning caused by etm_setup_aux failure

When coresight_build_path() fails on all the cpus, etm_setup_aux
calls etm_free_aux() to free allocated event_data.
WARN_ON(cpumask_empty(mask) will be triggered since cpu mask is empty.
Check event_data->snk_config is not NULL first to avoid this
warning.

Fixes: f5200aa9831f38 ("coresight: perf: Refactor function free_event_data()")
Reviewed-by: Mike Leach <mike.leach@linaro.org>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-9-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: stm: Allow to build coresight-stm as a module
Tingwei Zhang [Mon, 28 Sep 2020 16:34:55 +0000 (10:34 -0600)]
coresight: stm: Allow to build coresight-stm as a module

- Kconfig becomes a tristate, to allow =m
- add a stm_remove function, for module unload
- add a MODULE_DEVICE_TABLE for autoloading on boot

Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Suzuki K Poulose <Suzuki.Poulose@arm.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Russell King <linux@armlinux.org.uk>
Tested-by: Mike Leach <mike.leach@linaro.org>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Kim Phillips <kim.phillips@arm.com>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-8-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: Add try_get_module() in coresight_grab_device()
Tingwei Zhang [Mon, 28 Sep 2020 16:34:54 +0000 (10:34 -0600)]
coresight: Add try_get_module() in coresight_grab_device()

When coresight device is in an active session, driver module of
that device should not be removed. Use try_get_module() in
coresight_grab_device() to prevent module to be unloaded.
Use get_device()/put_device() to protect device data
in the middle of active session.

Suggested-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Tested-by: Mike Leach <mike.leach@linaro.org>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-7-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: Export global symbols
Mian Yousaf Kaukab [Mon, 28 Sep 2020 16:34:53 +0000 (10:34 -0600)]
coresight: Export global symbols

Export symbols used among coresight modules.

Tested-by: Mike Leach <mike.leach@linaro.org>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Mian Yousaf Kaukab <ykaukab@suse.de>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-6-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: Add coresight prefix to barrier_pkt
Tingwei Zhang [Mon, 28 Sep 2020 16:34:52 +0000 (10:34 -0600)]
coresight: Add coresight prefix to barrier_pkt

Add coresight prefix to make it specific. It will be a export symbol.

Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Mian Yousaf Kaukab <ykaukab@suse.de>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-5-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: Use IS_ENABLED for CONFIGs that may be modules
Kim Phillips [Mon, 28 Sep 2020 16:34:51 +0000 (10:34 -0600)]
coresight: Use IS_ENABLED for CONFIGs that may be modules

Checking for ifdef CONFIG_x fails if CONFIG_x=m.  Use IS_ENABLED
that is true for both built-ins and modules, instead.  Required
when building coresight components as modules.

Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Suzuki K Poulose <Suzuki.Poulose@arm.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Russell King <linux@armlinux.org.uk>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Kim Phillips <kim.phillips@arm.com>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-4-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: cpu_debug: Define MODULE_DEVICE_TABLE
Tingwei Zhang [Mon, 28 Sep 2020 16:34:50 +0000 (10:34 -0600)]
coresight: cpu_debug: Define MODULE_DEVICE_TABLE

Define a MODULE_DEVICE_TABLE for cpu_debug so module can
be auto loaded on boot.

Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-3-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocoresight: cpu_debug: Add module name in Kconfig
Tingwei Zhang [Mon, 28 Sep 2020 16:34:49 +0000 (10:34 -0600)]
coresight: cpu_debug: Add module name in Kconfig

Provide name of cpu_debug module in Kconfig help section.

Tested-by: Mike Leach <mike.leach@linaro.org>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Tingwei Zhang <tingwei@codeaurora.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Link: https://lore.kernel.org/r/20200928163513.70169-2-mathieu.poirier@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonvmem: core: fix possibly memleak when use nvmem_cell_info_to_nvmem_cell()
Vadym Kochan [Wed, 23 Sep 2020 20:44:56 +0000 (23:44 +0300)]
nvmem: core: fix possibly memleak when use nvmem_cell_info_to_nvmem_cell()

Fix missing 'kfree_const(cell->name)' when call to
nvmem_cell_info_to_nvmem_cell() in several places:

     * after nvmem_cell_info_to_nvmem_cell() failed during
       nvmem_add_cells()

     * during nvmem_device_cell_{read,write} when cell->name is
       kstrdup'ed() without calling kfree_const() at the end, but
       really there is no reason to do that 'dup, because the cell
       instance is allocated on the stack for some short period to be
       read/write without exposing it to the caller.

So the new nvmem_cell_info_to_nvmem_cell_nodup() helper is introduced
which is used to convert cell_info -> cell without name duplication as
a lighweight version of nvmem_cell_info_to_nvmem_cell().

Fixes: e2a5402ec7c6 ("nvmem: Add nvmem_device based consumer apis.")
Reviewed-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
Acked-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
Signed-off-by: Vadym Kochan <vadym.kochan@plvision.eu>
Link: https://lore.kernel.org/r/20200923204456.14032-1-vadym.kochan@plvision.eu
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoslimbus: qcom-ngd-ctrl: disable ngd in qmi server down callback
Srinivas Kandagatla [Fri, 25 Sep 2020 09:55:20 +0000 (10:55 +0100)]
slimbus: qcom-ngd-ctrl: disable ngd in qmi server down callback

In QMI new server notification we enable the NGD however during
delete server notification we do not disable the NGD.

This can lead to multiple instances of NGD being enabled, so make
sure that we disable NGD in delete server callback to fix this issue!

Fixes: 917809e2280b ("slimbus: ngd: Add qcom SLIMBus NGD driver")
Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
Link: https://lore.kernel.org/r/20200925095520.27316-4-srinivas.kandagatla@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoslimbus: core: do not enter to clock pause mode in core
Srinivas Kandagatla [Fri, 25 Sep 2020 09:55:19 +0000 (10:55 +0100)]
slimbus: core: do not enter to clock pause mode in core

Let the controller logic decide when to enter into clock pause mode!
Entering in to pause mode during unregistration does not really make
sense as the controller is totally going down at that point in time.

Fixes: 4b14e62ad3c9e ("slimbus: Add support for 'clock-pause' feature")
Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
Link: https://lore.kernel.org/r/20200925095520.27316-3-srinivas.kandagatla@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoslimbus: core: check get_addr before removing laddr ida
Srinivas Kandagatla [Fri, 25 Sep 2020 09:55:18 +0000 (10:55 +0100)]
slimbus: core: check get_addr before removing laddr ida

logical address can be either assigned by the SLIMBus controller or the core.
Core uses IDA in cases where get_addr callback is not provided by the
controller.
Core already has this check while allocating IDR, however during absence
reporting this is not checked. This patch fixes this issue.

Fixes: 46a2bb5a7f7e ("slimbus: core: Add slim controllers support")
Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
Link: https://lore.kernel.org/r/20200925095520.27316-2-srinivas.kandagatla@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoMerge tag 'misc-habanalabs-next-2020-09-25' of git://people.freedesktop.org/~gabbayo...
Greg Kroah-Hartman [Fri, 25 Sep 2020 12:36:48 +0000 (14:36 +0200)]
Merge tag 'misc-habanalabs-next-2020-09-25' of git://people.freedesktop.org/~gabbayo/linux into char-misc-next

Oded writes:

This tag contains the following changes for kernel 5.10-rc1:

- release the kernel context object after we reset the device. This is
  needed to prevent a race where the firmware still has some in-flight
  transcations that require the kernel context (and its memory mappings) to
  be alive.

- replace constant numbers with defines in QMAN initialization of GAUDI

- correct an error message text and add a few debug messages to help debug
  issues that happen during context open and close.

* tag 'misc-habanalabs-next-2020-09-25' of git://people.freedesktop.org/~gabbayo/linux:
  habanalabs/gaudi: configure QMAN LDMA registers properly
  habanalabs: add notice of device not idle
  habanalabs: add debug messages for opening/closing context
  habanalabs: release kernel context after hw_fini
  habanalabs: correct an error message

3 years agohabanalabs/gaudi: configure QMAN LDMA registers properly
Ofir Bitton [Thu, 24 Sep 2020 05:22:58 +0000 (08:22 +0300)]
habanalabs/gaudi: configure QMAN LDMA registers properly

LDMA registers are configured with a fixed value.
We add new define set which gives the configuration
a proper meaning.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: add notice of device not idle
Oded Gabbay [Wed, 23 Sep 2020 20:56:18 +0000 (23:56 +0300)]
habanalabs: add notice of device not idle

The device should be idle after a context is closed. If not, print a
notice.

Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: add debug messages for opening/closing context
Oded Gabbay [Wed, 23 Sep 2020 20:51:03 +0000 (23:51 +0300)]
habanalabs: add debug messages for opening/closing context

During debugging of error we sometimes need to know whether the error
happened when a user context was open. Add debug prints when opening and
closing user contexts.

Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: release kernel context after hw_fini
Oded Gabbay [Wed, 23 Sep 2020 12:06:52 +0000 (15:06 +0300)]
habanalabs: release kernel context after hw_fini

Some engines use resources that belong to the kernel context (e.g. MMU
mappings). In case the halt-engines doesn't work properly due to H/W
restriction, we need to make sure the kernel context lives on until after
the hw_fini. The hw_fini resets the ASIC after that no engine is alive and
we can safely close the kernel context.

Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: correct an error message
Oded Gabbay [Wed, 23 Sep 2020 11:07:32 +0000 (14:07 +0300)]
habanalabs: correct an error message

We don't try to allocate huge pages here so remove the huge word.

Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agoMerge tag 'icc-5.10-rc1' of https://git.linaro.org/people/georgi.djakov/linux into...
Greg Kroah-Hartman [Thu, 24 Sep 2020 13:15:11 +0000 (15:15 +0200)]
Merge tag 'icc-5.10-rc1' of https://git.linaro.org/people/georgi.djakov/linux into char-misc-next

Georgi writes:

interconnect changes for 5.10

Here are the interconnect changes for the 5.10-rc1 merge window
consisting of core changes, new drivers and cleanups.

Core changes:
 - New bulk API helpers for managing multiple interconnect paths.
 - New xlate_extended() interface for parsing additional data from DT.
 - Support for sync_state().

Driver changes:
 - New drivers for SM8150 and SM8250 platforms.
 - New drivers for the Qualcomm OSM and EPSS hardware blocks.
 - Per-BCM scaling factor support.
 - Misc cleanups.

Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
* tag 'icc-5.10-rc1' of https://git.linaro.org/people/georgi.djakov/linux: (28 commits)
  interconnect: imx: simplify the return expression of imx_icc_unregister
  interconnect: imx: Simplify with dev_err_probe()
  interconnect: core: Simplify with dev_err_probe()
  interconnect: qcom: Use icc_sync_state
  interconnect: Add sync state support
  interconnect: Add get_bw() callback
  interconnect: qcom: osm-l3: Mark more structures const
  interconnect: qcom: Add EPSS L3 support on SM8250
  dt-bindings: interconnect: Add EPSS L3 DT binding on SM8250
  interconnect: qcom: Lay the groundwork for adding EPSS support
  interconnect: qcom: Add OSM L3 support on SM8150
  dt-bindings: interconnect: Add OSM L3 DT binding on SM8150
  interconnect: qcom: sc7180: Replace xlate with xlate_extended
  interconnect: qcom: sdm845: Replace xlate with xlate_extended
  interconnect: qcom: Implement xlate_extended() to parse tags
  dt-bindings: interconnect: Document the support of optional path tag
  interconnect: Introduce xlate_extended() callback
  interconnect: qcom: Add support for per-BCM scaling factors
  interconnect: qcom: Only wait for completion in AMC/WAKE by default
  interconnect: qcom: Support bcm-voter-specific TCS wait behavior
  ...

3 years agomisc: hisi_hikey_usb: delete a stray tab
Dan Carpenter [Fri, 18 Sep 2020 14:34:05 +0000 (17:34 +0300)]
misc: hisi_hikey_usb: delete a stray tab

This return statement is indented one tab too far.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Link: https://lore.kernel.org/r/20200918143405.GF909725@mwanda
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agomisc: mic: scif: Fix error handling path
Souptick Joarder [Sun, 20 Sep 2020 02:51:35 +0000 (08:21 +0530)]
misc: mic: scif: Fix error handling path

Inside __scif_pin_pages(), when map_flags != SCIF_MAP_KERNEL it
will call pin_user_pages_fast() to map nr_pages. However,
pin_user_pages_fast() might fail with a return value -ERRNO.

The return value is stored in pinned_pages->nr_pages. which in
turn is passed to unpin_user_pages(), which expects
pinned_pages->nr_pages >=0, else disaster.

Fix this by assigning pinned_pages->nr_pages to 0 if
pin_user_pages_fast() returns -ERRNO.

Fixes: ba612aa8b487 ("misc: mic: SCIF memory registration and unregistration")
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Link: https://lore.kernel.org/r/1600570295-29546-1-git-send-email-jrdr.linux@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agomisc: pvpanic: Use devm_platform_ioremap_resource()
Wang ShaoBo [Fri, 18 Sep 2020 08:36:34 +0000 (16:36 +0800)]
misc: pvpanic: Use devm_platform_ioremap_resource()

Make use of devm_platform_ioremap_resource() provided by driver
core platform instead of duplicated analogue.

Acked-by: Arnd Bergmann <arn@arndb.de>
Signed-off-by: Wang ShaoBo <bobo.shaobowang@huawei.com>
Link: https://lore.kernel.org/r/20200918083634.33124-1-bobo.shaobowang@huawei.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoMerge tag 'misc-habanalabs-next-2020-09-22' of git://people.freedesktop.org/~gabbayo...
Greg Kroah-Hartman [Tue, 22 Sep 2020 16:38:09 +0000 (18:38 +0200)]
Merge tag 'misc-habanalabs-next-2020-09-22' of git://people.freedesktop.org/~gabbayo/linux into char-misc-next

Oded writes:

This tag contains the following changes for kernel 5.10-rc1:

- Stop using the DRM's dma-fence module and instead use kernel completions.
- Support PCIe AER
- Use dma_mmap_coherent for memory allocated using dma_alloc_coherent
- Use smallest possible alignment when allocating virtual addresses in our
  MMU driver.
- Refactor MMU driver code to be device-oriented
- Allow user to check CS status without any sleep
- Add an option to map a Command Buffer to the Device's MMU
- Expose sync manager resource allocation to user through INFO IOCTL
- Convert code to use standard BIT(), GENMASK() and FIELD_PREP()
- Many small fixes (casting, better error messages, remove unused
  defines, h/w configuration fixes, etc.)

* tag 'misc-habanalabs-next-2020-09-22' of git://people.freedesktop.org/~gabbayo/linux: (46 commits)
  habanalabs: update scratchpad register map
  habanalabs: add indication of security-enabled F/W
  habanalabs/gaudi: fix DMA completions max outstanding to 15
  habanalabs/gaudi: remove axi drain support
  habanalabs: update firmware interface file
  habanalabs: Add an option to map CB to device MMU
  habanalabs: Save context in a command buffer object
  habanalabs: no need for DMA_SHARED_BUFFER
  habanalabs: allow to wait on CS without sleep
  habanalabs/gaudi: increase timeout for boot fit load
  habanalabs: add debugfs support for MMU with 6 HOPs
  habanalabs: add num_hops to hl_mmu_properties
  habanalabs: refactor MMU as device-oriented
  habanalabs: rename mmu.c to mmu_v1.c
  habanalabs: use smallest possible alignment for virtual addresses
  habanalabs: check flag before reset because of f/w event
  habanalabs: increase PQ COMP_OFFSET by one nibble
  habanalabs: Fix alignment issue in cpucp_info structure
  habanalabs: remove unused define
  habanalabs: remove unused ASIC function pointer
  ...

3 years agohabanalabs: update scratchpad register map
Oded Gabbay [Thu, 17 Sep 2020 14:57:03 +0000 (17:57 +0300)]
habanalabs: update scratchpad register map

Our firmware use some scratchpad registers in the device for different
roles. Update the file to the latest version of the firmware code.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: add indication of security-enabled F/W
Oded Gabbay [Sun, 13 Sep 2020 12:51:28 +0000 (15:51 +0300)]
habanalabs: add indication of security-enabled F/W

Future F/W versions will have enhanced security measures and the driver
won't be able to do certain configurations that it always did and those
configurations will be done by the firmware.

We use the firmware's preboot version to determine whether security
measures are enabled or not. Because we need this very early in our code,
the read of the preboot version is moved to the earliest possible place,
right after the device's PCI initialization.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs/gaudi: fix DMA completions max outstanding to 15
Oded Gabbay [Mon, 14 Sep 2020 06:26:54 +0000 (09:26 +0300)]
habanalabs/gaudi: fix DMA completions max outstanding to 15

This is a workaround for H/W bug H3-2116, where if there are more than 16
outstanding completions in the DMA transpose engine, there can be a
deadlock in the engine.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs/gaudi: remove axi drain support
Oded Gabbay [Sun, 13 Sep 2020 12:35:33 +0000 (15:35 +0300)]
habanalabs/gaudi: remove axi drain support

AXI drain is broken in GAUDI so remove support for enabling it.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: update firmware interface file
Oded Gabbay [Thu, 10 Sep 2020 13:37:59 +0000 (16:37 +0300)]
habanalabs: update firmware interface file

Add new packet to fetch PLL information from firmware. This will be needed
in the future when the driver won't be able to access the PLL registers
directly

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: Add an option to map CB to device MMU
Tomer Tayar [Thu, 9 Jul 2020 13:17:48 +0000 (16:17 +0300)]
habanalabs: Add an option to map CB to device MMU

There are cases in which the device should access the host memory of a
CB through the device MMU, and thus this memory should be mapped.
The patch adds a flag to the CB IOCTL, in which a user can ask the
driver to perform the mapping when creating a CB.
The mapping is allowed only if a dedicated VA range was allocated for
the specific ASIC.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: Save context in a command buffer object
Tomer Tayar [Mon, 7 Sep 2020 14:36:41 +0000 (17:36 +0300)]
habanalabs: Save context in a command buffer object

Future changes require using a context while handling a command buffer,
and thus need to save the context in the command buffer object.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: no need for DMA_SHARED_BUFFER
Oded Gabbay [Wed, 9 Sep 2020 07:39:00 +0000 (10:39 +0300)]
habanalabs: no need for DMA_SHARED_BUFFER

Now that the driver no longer uses dma_buf, we can remove the select of
DMA_SHARED_BUFFER from kconfig.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: allow to wait on CS without sleep
Oded Gabbay [Mon, 7 Sep 2020 15:08:51 +0000 (18:08 +0300)]
habanalabs: allow to wait on CS without sleep

The user sometimes wants to check if a CS has completed to clean resources.
In that case, the user doesn't want to sleep but just to check if the CS
has finished and continue with his code.

Add a new definition to the API of the wait on CS. The new definition says
that if the timeout is 0, the driver won't sleep at all but return
immediately after checking if the CS has finished.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs/gaudi: increase timeout for boot fit load
Oded Gabbay [Tue, 8 Sep 2020 12:17:03 +0000 (15:17 +0300)]
habanalabs/gaudi: increase timeout for boot fit load

The firmware running in the boot stage takes more time to execute due to
increased security mechanisms. Therefore, we need to increase the timeout
we wait for the boot fit to finish loading.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: add debugfs support for MMU with 6 HOPs
Moti Haimovski [Thu, 13 Aug 2020 13:54:50 +0000 (16:54 +0300)]
habanalabs: add debugfs support for MMU with 6 HOPs

This commit modify the existing debugfs code to support future devices that
have a 6 HOPs MMU implementation instead of 5 HOPs implementation.

Signed-off-by: Moti Haimovski <mhaimovski@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: add num_hops to hl_mmu_properties
Moti Haimovski [Sun, 23 Aug 2020 10:23:13 +0000 (13:23 +0300)]
habanalabs: add num_hops to hl_mmu_properties

This commit adds the number of HOPs supported by the device to the
device MMU properties.

Signed-off-by: Moti Haimovski <mhaimovski@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: refactor MMU as device-oriented
Moti Haimovski [Wed, 12 Aug 2020 10:33:44 +0000 (13:33 +0300)]
habanalabs: refactor MMU as device-oriented

As preparation to MMU v2, rework MMU to be device oriented
instantiated according to the device in hand.

Signed-off-by: Moti Haimovski <mhaimovski@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: rename mmu.c to mmu_v1.c
Moti Haimovski [Wed, 12 Aug 2020 08:40:08 +0000 (11:40 +0300)]
habanalabs: rename mmu.c to mmu_v1.c

In the future we will have MMU v2 code, so we need to prepare the
driver for it. The first step is to rename the current MMU file to
mmu_v1.c.

Signed-off-by: Moti Haimovski <mhaimovski@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: use smallest possible alignment for virtual addresses
Omer Shpigelman [Sun, 28 Jun 2020 18:15:53 +0000 (21:15 +0300)]
habanalabs: use smallest possible alignment for virtual addresses

Change the acquiring of a device virtual address for mapping by using the
smallest possible alignment, rather than the biggest, depending on the
page size used by the user for allocating the memory. This will lower the
virtual space memory consumption.

Signed-off-by: Omer Shpigelman <oshpigelman@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: check flag before reset because of f/w event
Oded Gabbay [Fri, 4 Sep 2020 18:39:14 +0000 (21:39 +0300)]
habanalabs: check flag before reset because of f/w event

For consistency with GAUDI code, add check of the relevant flag in the
device structure before resetting the GOYA device in case of firmware
event.

Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: increase PQ COMP_OFFSET by one nibble
Oded Gabbay [Fri, 4 Sep 2020 18:36:46 +0000 (21:36 +0300)]
habanalabs: increase PQ COMP_OFFSET by one nibble

For future ASICs, we increase this field by one nibble. This field was not
used by the current ASICs so this change doesn't break anything.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: Fix alignment issue in cpucp_info structure
Ofir Bitton [Fri, 4 Sep 2020 18:33:53 +0000 (21:33 +0300)]
habanalabs: Fix alignment issue in cpucp_info structure

Because the device CPU compiler aligns structures to 8 bytes,
struct cpucp_info has an alignment issue as some parts
in the structure are not aligned to 8 bytes.
It is preferred that we explicitly insert placeholders inside
the structure to avoid confusion

in order to validate this scenario, we printed both pointers:

__u8 cpucp_version[VERSION_MAX_LEN]; (0xffff899c67ed4cbc)
__le64 dram_size;                    (0xffff899c67ed4d40)

we see difference of 132 bytes although the first array
is only 128 bytes long, Meaning compiler added a 4 byte padding.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: remove unused define
Oded Gabbay [Fri, 4 Sep 2020 17:22:35 +0000 (20:22 +0300)]
habanalabs: remove unused define

Cleanup the code.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: remove unused ASIC function pointer
Oded Gabbay [Fri, 4 Sep 2020 17:21:39 +0000 (20:21 +0300)]
habanalabs: remove unused ASIC function pointer

Old function pointer that was left when the call to this function pointer
was removed.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: rename ArmCP to CPU-CP
Oded Gabbay [Fri, 4 Sep 2020 17:18:16 +0000 (20:18 +0300)]
habanalabs: rename ArmCP to CPU-CP

There were a couple of comments where the name ArmCP was still used. Rename
it to CPU-CP.

In addition, rename ArmCP or ARM in log messages to "device CPU".

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: count dropped CS because max CS in-flight
Oded Gabbay [Tue, 1 Sep 2020 08:22:05 +0000 (11:22 +0300)]
habanalabs: count dropped CS because max CS in-flight

There is a case where the user reaches the maximum number of CS in-flight.
In that case, the driver rejects the new CS of the user with EAGAIN. Count
that event so the user can query the driver later to see if it happened.

Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: make use of dma_mmap_coherent
Hillf Danton [Sat, 22 Aug 2020 23:32:42 +0000 (07:32 +0800)]
habanalabs: make use of dma_mmap_coherent

Add dma_mmap_coherent() for goya and gaudi to match their use of
dma_alloc_coherent(), see the Link tag for why.

Link: https://lore.kernel.org/lkml/20200609091727.GA23814@lst.de/
Cc: Christoph Hellwig <hch@lst.de>
Cc: Zhang Li <li.zhang@bitmain.com>
Cc: Ding Z Nan <oshack@hotmail.com>
Signed-off-by: Hillf Danton <hdanton@sina.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: clear vm_pgoff before doing the mmap
Oded Gabbay [Sat, 29 Aug 2020 08:55:15 +0000 (11:55 +0300)]
habanalabs: clear vm_pgoff before doing the mmap

The driver use vm_pgoff to hold the CB idr handle. Before we actually call
the mapping function, we need to clear the handle so there won't be any
garbage left in vm_pgoff.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: restructure hl_mmap
Oded Gabbay [Sat, 29 Aug 2020 08:51:39 +0000 (11:51 +0300)]
habanalabs: restructure hl_mmap

Arrange the hl_mmap code to be more structured and expandable for the
future. Add better defines that describe our usage of the vm_pgoff.

Note that I shamelessly took the code and defines from the amdkfd driver
(my previous driver).

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: cast to u64 before shift > 31 bits
Oded Gabbay [Sat, 29 Aug 2020 08:24:03 +0000 (11:24 +0300)]
habanalabs: cast to u64 before shift > 31 bits

When shifting a boolean variable by more than 31 bits and putting the
result into a u64 variable, we need to cast the boolean into unsigned 64
bits to prevent possible overflow.

Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: replace armcp with the generic cpucp
Oded Gabbay [Sat, 15 Aug 2020 13:28:10 +0000 (16:28 +0300)]
habanalabs: replace armcp with the generic cpucp

ArmCP mandates that the device CPU is always an ARM processor, which might
be wrong in the future.

Most of this change is an internal renaming of variables, functions and
defines but there are two entries in sysfs which have armcp in their
names. Add identical cpucp entries but don't remove yet the armcp entries.
Those will be deprecated next year. Add the documentation about it in sysfs
documentation.

Signed-off-by: Moti Haimovski <mhaimovski@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: update GAUDI hardware specs
Oded Gabbay [Fri, 21 Aug 2020 06:59:18 +0000 (09:59 +0300)]
habanalabs: update GAUDI hardware specs

Add define for the 2 MME slave engines.

Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: add support for getting device total energy
farah kassabri [Sun, 9 Aug 2020 13:25:53 +0000 (16:25 +0300)]
habanalabs: add support for getting device total energy

Add driver implementation for reading the total energy consumption
from the device ARM FW.

Signed-off-by: farah kassabri <fkassabri@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: Include linux/bitfield.h only in habanalabs.h
Tomer Tayar [Tue, 18 Aug 2020 12:06:56 +0000 (15:06 +0300)]
habanalabs: Include linux/bitfield.h only in habanalabs.h

Include linux/bitfield.h only in habanalabs.h, instead of in each and
every file that needs it, as habanalabs.h is already included by all.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: extend busy engines mask to 64 bits
farah kassabri [Wed, 12 Aug 2020 14:20:13 +0000 (17:20 +0300)]
habanalabs: extend busy engines mask to 64 bits

change busy engines bitmask to 64 bits in order to represent
more engines, needed for future ASIC support.

Signed-off-by: farah kassabri <fkassabri@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: use 1U when shifting bits
Oded Gabbay [Wed, 12 Aug 2020 08:32:27 +0000 (11:32 +0300)]
habanalabs: use 1U when shifting bits

Eliminate following warning:
warning: Shifting signed 32-bit value by 31 bits is undefined behavior

Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: check TPC vector pipe is empty
Oded Gabbay [Wed, 12 Aug 2020 08:28:13 +0000 (11:28 +0300)]
habanalabs: check TPC vector pipe is empty

The driver waits for the TPC vector pipe to be empty before checking if the
TPC kernel has finished executing, but the code doesn't validate that the
pipe was indeed empty, it just wait for it without checking the return
value.

Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: remove redundant assignment to variable
Oded Gabbay [Wed, 12 Aug 2020 08:24:05 +0000 (11:24 +0300)]
habanalabs: remove redundant assignment to variable

new_dma_pkt->ctl is assigned a value and then is reassigned a new value
without the first value ever being used.

Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: use FIELD_PREP() instead of <<
Oded Gabbay [Wed, 12 Aug 2020 08:21:01 +0000 (11:21 +0300)]
habanalabs: use FIELD_PREP() instead of <<

Use the standard FIELD_PREP() macro instead of << operator to perform
bitmask operations. This ensures type check safety and eliminate compiler
warnings.

Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: use standard BIT() and GENMASK()
Oded Gabbay [Wed, 12 Aug 2020 07:46:33 +0000 (10:46 +0300)]
habanalabs: use standard BIT() and GENMASK()

Use the standard macros to define bitmasks.

Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: eliminate redundant else condition
Oded Gabbay [Wed, 12 Aug 2020 07:19:28 +0000 (10:19 +0300)]
habanalabs: eliminate redundant else condition

If both parts of if-else are goto statements, we can remove the else and
put the else goto statement after the if statement.

Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: cast int to u32 before printing it with %u
Oded Gabbay [Wed, 12 Aug 2020 07:15:27 +0000 (10:15 +0300)]
habanalabs: cast int to u32 before printing it with %u

%u is used for unsigned so we need to cast the int variable to u32 to avoid
compiler warning.

Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: change CB's ID to be 64 bits
Oded Gabbay [Wed, 12 Aug 2020 07:11:20 +0000 (10:11 +0300)]
habanalabs: change CB's ID to be 64 bits

Although the possible values for CB's ID are only 32 bits, there are a few
places in the code where this field is shifted and passed into a function
which expects 64 bits.

Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: print the queue id in case of an error
Dotan Barak [Thu, 6 Aug 2020 06:20:49 +0000 (09:20 +0300)]
habanalabs: print the queue id in case of an error

If there is a failure during the testing of a queue,
to ease up debugging - print the queue id.

Signed-off-by: Dotan Barak <dbarak@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: remove security from ARB_MST_QUIET register
farah kassabri [Tue, 4 Aug 2020 06:41:32 +0000 (09:41 +0300)]
habanalabs: remove security from ARB_MST_QUIET register

Allow user application to write to this register in order
to be able to configure the quiet period of the QMAN between grants.

Signed-off-by: farah kassabri <fkassabri@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: PCIe Advanced Error Reporting support
Ofir Bitton [Wed, 20 May 2020 13:35:08 +0000 (16:35 +0300)]
habanalabs: PCIe Advanced Error Reporting support

driver will now get notified upon any PCI error occurred and
will respond according to the severity of the error.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: expose sync manager resources allocation in INFO IOCTL
Ofir Bitton [Sun, 19 Jul 2020 08:08:09 +0000 (11:08 +0300)]
habanalabs: expose sync manager resources allocation in INFO IOCTL

Although the driver defines the first user-available sync manager object
and monitor in habanalabs.h, we would like to also expose this information
via the INFO IOCTL so the runtime can get this information dynamically.
This is because in future ASICs we won't need to define it statically.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: add information about PCIe controller
Ofir Bitton [Tue, 21 Jul 2020 07:49:51 +0000 (10:49 +0300)]
habanalabs: add information about PCIe controller

Update firmware header with new API for getting pcie info
such as tx/rx throughput and replay counter.
These counters are needed by customers for monitor and maintenance
of multiple devices.
Add new opcodes to the INFO ioctl to retrieve these counters.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: Replace dma-fence mechanism with completions
Ofir Bitton [Wed, 15 Jul 2020 05:52:39 +0000 (08:52 +0300)]
habanalabs: Replace dma-fence mechanism with completions

habanalabs driver uses dma-fence mechanism for synchronization.
dma-fence mechanism was designed solely for GPUs, hence we purpose
a simpler mechanism based on completions to replace current
dma-fence objects.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
3 years agohabanalabs: increase length of ASIC name
Oded Gabbay [Sun, 2 Aug 2020 08:49:09 +0000 (11:49 +0300)]
habanalabs: increase length of ASIC name

Future ASIC names are longer than 15 chars so increase the variable length
to 32 chars.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
3 years agoMAINTAINERS: Add entry for the Nitro Enclaves driver
Andra Paraschiv [Mon, 21 Sep 2020 12:17:32 +0000 (15:17 +0300)]
MAINTAINERS: Add entry for the Nitro Enclaves driver

Add entry in the MAINTAINERS file for the Nitro Enclaves files such as
the documentation, the header files, the driver itself and the user
space sample.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Update the location of the documentation, as it has been moved to the
  "virt" directory.

v7 -> v8

* No changes.

v6 -> v7

* No changes.

v5 -> v6

* No changes.

v4 -> v5

* No changes.

v3 -> v4

* No changes.

v2 -> v3

* Update file entries to be in alphabetical order.

v1 -> v2

* No changes.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-19-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add overview documentation
Andra Paraschiv [Mon, 21 Sep 2020 12:17:31 +0000 (15:17 +0300)]
nitro_enclaves: Add overview documentation

Add documentation on the overview of Nitro Enclaves. Include it in the
virtualization specific directory.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Move the Nitro Enclaves documentation to the "virt" directory and add
  an entry for it in the corresponding index file.

v7 -> v8

* Add info about the primary / parent VM CID value.
* Update reference link for huge pages.
* Add reference link for the x86 boot protocol.
* Add license mention and update doc title / chapter formatting.

v6 -> v7

* No changes.

v5 -> v6

* No changes.

v4 -> v5

* No changes.

v3 -> v4

* Update doc type from .txt to .rst.
* Update documentation based on the changes from v4.

v2 -> v3

* No changes.

v1 -> v2

* New in v2.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-18-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add sample for ioctl interface usage
Andra Paraschiv [Mon, 21 Sep 2020 12:17:30 +0000 (15:17 +0300)]
nitro_enclaves: Add sample for ioctl interface usage

Add a user space sample for the usage of the ioctl interface provided by
the Nitro Enclaves driver.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* No changes.

v7 -> v8

* Track NE custom error codes for invalid page size, invalid flags and
  enclave CID.
* Update the heartbeat logic to have a listener fd first, then start the
  enclave and then accept connection to get the heartbeat.
* Update the reference link to the hugetlb documentation.

v6 -> v7

* Track POLLNVAL as poll event in addition to POLLHUP.

v5 -> v6

* Remove "rc" mentioning when printing errno string.
* Remove the ioctl to query API version.
* Include usage info for NUMA-aware hugetlb configuration.
* Update documentation to kernel-doc format.
* Add logic for enclave image loading.

v4 -> v5

* Print enclave vCPU ids when they are created.
* Update logic to map the modified vCPU ioctl call.
* Add check for the path to the enclave image to be less than PATH_MAX.
* Update the ioctl calls error checking logic to match the NE specific
  error codes.

v3 -> v4

* Update usage details to match the updates in v4.
* Update NE ioctl interface usage.

v2 -> v3

* Remove the include directory to use the uapi from the kernel.
* Remove the GPL additional wording as SPDX-License-Identifier is
  already in place.

v1 -> v2

* New in v2.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru Vasile <lexnv@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-17-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add Makefile for the Nitro Enclaves driver
Andra Paraschiv [Mon, 21 Sep 2020 12:17:29 +0000 (15:17 +0300)]
nitro_enclaves: Add Makefile for the Nitro Enclaves driver

Add Makefile for the Nitro Enclaves driver, considering the option set
in the kernel config.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Remove -Wall flags, could use W=1 as an option for this.

v7 -> v8

* No changes.

v6 -> v7

* No changes.

v5 -> v6

* No changes.

v4 -> v5

* No changes.

v3 -> v4

* No changes.

v2 -> v3

* Remove the GPL additional wording as SPDX-License-Identifier is
  already in place.

v1 -> v2

* Update path to Makefile to match the drivers/virt/nitro_enclaves
  directory.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-16-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add Kconfig for the Nitro Enclaves driver
Andra Paraschiv [Mon, 21 Sep 2020 12:17:28 +0000 (15:17 +0300)]
nitro_enclaves: Add Kconfig for the Nitro Enclaves driver

Add kernel config entry for Nitro Enclaves, including dependencies.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* No changes.

v7 -> v8

* No changes.

v6 -> v7

* Remove, for now, the dependency on ARM64 arch. x86 is currently
  supported, with Arm to come afterwards. The NE kernel driver can be
  built for aarch64 arch.

v5 -> v6

* No changes.

v4 -> v5

* Add arch dependency for Arm / x86.

v3 -> v4

* Add PCI and SMP dependencies.

v2 -> v3

* Remove the GPL additional wording as SPDX-License-Identifier is
  already in place.

v1 -> v2

* Update path to Kconfig to match the drivers/virt/nitro_enclaves
  directory.
* Update help in Kconfig.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-15-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add logic for terminating an enclave
Andra Paraschiv [Mon, 21 Sep 2020 12:17:27 +0000 (15:17 +0300)]
nitro_enclaves: Add logic for terminating an enclave

An enclave is associated with an fd that is returned after the enclave
creation logic is completed. This enclave fd is further used to setup
enclave resources. Once the enclave needs to be terminated, the enclave
fd is closed.

Add logic for enclave termination, that is mapped to the enclave fd
release callback. Free the internal enclave info used for bookkeeping.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Use the ne_devs data structure to get the refs for the NE PCI device.

v7 -> v8

* No changes.

v6 -> v7

* Remove the pci_dev_put() call as the NE misc device parent field is
  used now to get the NE PCI device.
* Update the naming and add more comments to make more clear the logic
  of handling full CPU cores and dedicating them to the enclave.

v5 -> v6

* Update documentation to kernel-doc format.
* Use directly put_page() instead of unpin_user_pages(), to match the
  get_user_pages() calls.

v4 -> v5

* Release the reference to the NE PCI device on enclave fd release.
* Adapt the logic to cpumask enclave vCPU ids and CPU cores.
* Remove sanity checks for situations that shouldn't happen, only if
  buggy system or broken logic at all.

v3 -> v4

* Use dev_err instead of custom NE log pattern.

v2 -> v3

* Remove the WARN_ON calls.
* Update static calls sanity checks.
* Update kzfree() calls to kfree().

v1 -> v2

* Add log pattern for NE.
* Remove the BUG_ON calls.
* Update goto labels to match their purpose.
* Add early exit in release() if there was a slot alloc error in the fd
  creation path.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru Vasile <lexnv@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-14-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add logic for starting an enclave
Andra Paraschiv [Mon, 21 Sep 2020 12:17:26 +0000 (15:17 +0300)]
nitro_enclaves: Add logic for starting an enclave

After all the enclave resources are set, the enclave is ready for
beginning to run.

Add ioctl command logic for starting an enclave after all its resources,
memory regions and CPUs, have been set.

The enclave start information includes the local channel addressing -
vsock CID - and the flags associated with the enclave.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Use the ne_devs data structure to get the refs for the NE PCI device.

v7 -> v8

* Add check for invalid enclave CID value e.g. well-known CIDs and
  parent VM CID.
* Add custom error code for incorrect flag in enclave start info and
  invalid enclave CID.

v6 -> v7

* Update the naming and add more comments to make more clear the logic
  of handling full CPU cores and dedicating them to the enclave.

v5 -> v6

* Check for invalid enclave start flags.
* Update documentation to kernel-doc format.

v4 -> v5

* Add early exit on enclave start ioctl function call error.
* Move sanity checks in the enclave start ioctl function, outside of the
  switch-case block.
* Remove log on copy_from_user() / copy_to_user() failure.

v3 -> v4

* Use dev_err instead of custom NE log pattern.
* Update the naming for the ioctl command from metadata to info.
* Check for minimum enclave memory size.

v2 -> v3

* Remove the WARN_ON calls.
* Update static calls sanity checks.

v1 -> v2

* Add log pattern for NE.
* Check if enclave state is init when starting an enclave.
* Remove the BUG_ON calls.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru Vasile <lexnv@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-13-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add logic for setting an enclave memory region
Andra Paraschiv [Mon, 21 Sep 2020 12:17:25 +0000 (15:17 +0300)]
nitro_enclaves: Add logic for setting an enclave memory region

Another resource that is being set for an enclave is memory. User space
memory regions, that need to be backed by contiguous memory regions,
are associated with the enclave.

One solution for allocating / reserving contiguous memory regions, that
is used for integration, is hugetlbfs. The user space process that is
associated with the enclave passes to the driver these memory regions.

The enclave memory regions need to be from the same NUMA node as the
enclave CPUs.

Add ioctl command logic for setting user space memory region for an
enclave.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Use the ne_devs data structure to get the refs for the NE PCI device.

v7 -> v8

* Add early check, while getting user pages, to be multiple of 2 MiB for
  the pages that back the user space memory region.
* Add custom error code for incorrect user space memory region flag.
* Include in a separate function the sanity checks for each page of the
  user space memory region.

v6 -> v7

* Update check for duplicate user space memory regions to cover
  additional possible scenarios.

v5 -> v6

* Check for max number of pages allocated for the internal data
  structure for pages.
* Check for invalid memory region flags.
* Check for aligned physical memory regions.
* Update documentation to kernel-doc format.
* Check for duplicate user space memory regions.
* Use directly put_page() instead of unpin_user_pages(), to match the
  get_user_pages() calls.

v4 -> v5

* Add early exit on set memory region ioctl function call error.
* Remove log on copy_from_user() failure.
* Exit without unpinning the pages on NE PCI dev request failure as
  memory regions from the user space range may have already been added.
* Add check for the memory region user space address to be 2 MiB
  aligned.
* Update logic to not have a hardcoded check for 2 MiB memory regions.

v3 -> v4

* Check enclave memory regions are from the same NUMA node as the
  enclave CPUs.
* Use dev_err instead of custom NE log pattern.
* Update the NE ioctl call to match the decoupling from the KVM API.

v2 -> v3

* Remove the WARN_ON calls.
* Update static calls sanity checks.
* Update kzfree() calls to kfree().

v1 -> v2

* Add log pattern for NE.
* Update goto labels to match their purpose.
* Remove the BUG_ON calls.
* Check if enclave max memory regions is reached when setting an enclave
  memory region.
* Check if enclave state is init when setting an enclave memory region.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru Vasile <lexnv@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-12-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add logic for getting the enclave image load info
Andra Paraschiv [Mon, 21 Sep 2020 12:17:24 +0000 (15:17 +0300)]
nitro_enclaves: Add logic for getting the enclave image load info

Before setting the memory regions for the enclave, the enclave image
needs to be placed in memory. After the memory regions are set, this
memory cannot be used anymore by the VM, being carved out.

Add ioctl command logic to get the offset in enclave memory where to
place the enclave image. Then the user space tooling copies the enclave
image in the memory using the given memory offset.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* No changes.

v7 -> v8

* Add custom error code for incorrect enclave image load info flag.

v6 -> v7

* No changes.

v5 -> v6

* Check for invalid enclave image load flags.

v4 -> v5

* Check for the enclave not being started when invoking this ioctl call.
* Remove log on copy_from_user() / copy_to_user() failure.

v3 -> v4

* Use dev_err instead of custom NE log pattern.
* Set enclave image load offset based on flags.
* Update the naming for the ioctl command from metadata to info.

v2 -> v3

* No changes.

v1 -> v2

* New in v2.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-11-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add logic for setting an enclave vCPU
Andra Paraschiv [Mon, 21 Sep 2020 12:17:23 +0000 (15:17 +0300)]
nitro_enclaves: Add logic for setting an enclave vCPU

An enclave, before being started, has its resources set. One of its
resources is CPU.

A NE CPU pool is set and enclave CPUs are chosen from it. Offline the
CPUs from the NE CPU pool during the pool setup and online them back
during the NE CPU pool teardown. The CPU offline is necessary so that
there would not be more vCPUs than physical CPUs available to the
primary / parent VM. In that case the CPUs would be overcommitted and
would change the initial configuration of the primary / parent VM of
having dedicated vCPUs to physical CPUs.

The enclave CPUs need to be full cores and from the same NUMA node. CPU
0 and its siblings have to remain available to the primary / parent VM.

Add ioctl command logic for setting an enclave vCPU.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Use the ne_devs data structure to get the refs for the NE PCI device.

v7 -> v8

* No changes.

v6 -> v7

* Check for error return value when setting the kernel parameter string.
* Use the NE misc device parent field to get the NE PCI device.
* Update the naming and add more comments to make more clear the logic
  of handling full CPU cores and dedicating them to the enclave.
* Calculate the number of threads per core and not use smp_num_siblings
  that is x86 specific.

v5 -> v6

* Check CPUs are from the same NUMA node before going through CPU
  siblings during the NE CPU pool setup.
* Update documentation to kernel-doc format.

v4 -> v5

* Set empty string in case of invalid NE CPU pool.
* Clear NE CPU pool mask on pool setup failure.
* Setup NE CPU cores out of the NE CPU pool.
* Early exit on NE CPU pool setup if enclave(s) already running.
* Remove sanity checks for situations that shouldn't happen, only if
  buggy system or broken logic at all.
* Add check for maximum vCPU id possible before looking into the CPU
  pool.
* Remove log on copy_from_user() / copy_to_user() failure and on admin
  capability check for setting the NE CPU pool.
* Update the ioctl call to not create a file descriptor for the vCPU.
* Split the CPU pool usage logic in 2 separate functions - one to get a
  CPU from the pool and the other to check the given CPU is available in
  the pool.

v3 -> v4

* Setup the NE CPU pool at runtime via a sysfs file for the kernel
  parameter.
* Check enclave CPUs to be from the same NUMA node.
* Use dev_err instead of custom NE log pattern.
* Update the NE ioctl call to match the decoupling from the KVM API.

v2 -> v3

* Remove the WARN_ON calls.
* Update static calls sanity checks.
* Update kzfree() calls to kfree().
* Remove file ops that do nothing for now - open, ioctl and release.

v1 -> v2

* Add log pattern for NE.
* Update goto labels to match their purpose.
* Remove the BUG_ON calls.
* Check if enclave state is init when setting enclave vCPU.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru Vasile <lexnv@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-10-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add logic for creating an enclave VM
Andra Paraschiv [Mon, 21 Sep 2020 12:17:22 +0000 (15:17 +0300)]
nitro_enclaves: Add logic for creating an enclave VM

Add ioctl command logic for enclave VM creation. It triggers a slot
allocation. The enclave resources will be associated with this slot and
it will be used as an identifier for triggering enclave run.

Return a file descriptor, namely enclave fd. This is further used by the
associated user space enclave process to set enclave resources and
trigger enclave termination.

The poll function is implemented in order to notify the enclave process
when an enclave exits without a specific enclave termination command
trigger e.g. when an enclave crashes.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Use the ne_devs data structure to get the refs for the NE PCI device.

v7 -> v8

* No changes.

v6 -> v7

* Use the NE misc device parent field to get the NE PCI device.
* Update the naming and add more comments to make more clear the logic
  of handling full CPU cores and dedicating them to the enclave.

v5 -> v6

* Update the code base to init the ioctl function in this patch.
* Update documentation to kernel-doc format.

v4 -> v5

* Release the reference to the NE PCI device on create VM error.
* Close enclave fd on copy_to_user() failure; rename fd to enclave fd
  while at it.
* Remove sanity checks for situations that shouldn't happen, only if
  buggy system or broken logic at all.
* Remove log on copy_to_user() failure.

v3 -> v4

* Use dev_err instead of custom NE log pattern.
* Update the NE ioctl call to match the decoupling from the KVM API.
* Add metadata for the NUMA node for the enclave memory and CPUs.

v2 -> v3

* Remove the WARN_ON calls.
* Update static calls sanity checks.
* Update kzfree() calls to kfree().
* Remove file ops that do nothing for now - open.

v1 -> v2

* Add log pattern for NE.
* Update goto labels to match their purpose.
* Remove the BUG_ON calls.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru Vasile <lexnv@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-9-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Init misc device providing the ioctl interface
Andra Paraschiv [Mon, 21 Sep 2020 12:17:21 +0000 (15:17 +0300)]
nitro_enclaves: Init misc device providing the ioctl interface

The Nitro Enclaves driver provides an ioctl interface to the user space
for enclave lifetime management e.g. enclave creation / termination and
setting enclave resources such as memory and CPU.

This ioctl interface is mapped to a Nitro Enclaves misc device.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Use the ne_devs data structure to get the refs for the NE misc device
  in the NE PCI device driver logic.

v7 -> v8

* Add define for the CID of the primary / parent VM.
* Update the NE PCI driver shutdown logic to include misc device
  deregister.

v6 -> v7

* Set the NE PCI device the parent of the NE misc device to be able to
  use it in the ioctl logic.
* Update the naming and add more comments to make more clear the logic
  of handling full CPU cores and dedicating them to the enclave.

v5 -> v6

* Remove the ioctl to query API version.
* Update documentation to kernel-doc format.

v4 -> v5

* Update the size of the NE CPU pool string from 4096 to 512 chars.

v3 -> v4

* Use dev_err instead of custom NE log pattern.
* Remove the NE CPU pool init during kernel module loading, as the CPU
  pool is now setup at runtime, via a sysfs file for the kernel
  parameter.
* Add minimum enclave memory size definition.

v2 -> v3

* Remove the GPL additional wording as SPDX-License-Identifier is
  already in place.
* Remove the WARN_ON calls.
* Remove linux/bug and linux/kvm_host includes that are not needed.
* Remove "ratelimited" from the logs that are not in the ioctl call
  paths.
* Remove file ops that do nothing for now - open and release.

v1 -> v2

* Add log pattern for NE.
* Update goto labels to match their purpose.
* Update ne_cpu_pool data structure to include the global mutex.
* Update NE misc device mode to 0660.
* Check if the CPU siblings are included in the NE CPU pool, as full CPU
  cores are given for the enclave(s).

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-8-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Handle out-of-band PCI device events
Andra Paraschiv [Mon, 21 Sep 2020 12:17:20 +0000 (15:17 +0300)]
nitro_enclaves: Handle out-of-band PCI device events

In addition to the replies sent by the Nitro Enclaves PCI device in
response to command requests, out-of-band enclave events can happen e.g.
an enclave crashes. In this case, the Nitro Enclaves driver needs to be
aware of the event and notify the corresponding user space process that
abstracts the enclave.

Register an MSI-X interrupt vector to be used for this kind of
out-of-band events. The interrupt notifies that the state of an enclave
changed and the driver logic scans the state of each running enclave to
identify for which this notification is intended.

Create an workqueue to handle the out-of-band events. Notify user space
enclave process that is using a polling mechanism on the enclave fd.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Use the reference to the pdev directly from the ne_pci_dev instead of
  the one from the enclave data structure.

v7 -> v8

* No changes.

v6 -> v7

* No changes.

v5 -> v6

* Update documentation to kernel-doc format.

v4 -> v5

* Remove sanity checks for situations that shouldn't happen, only if
  buggy system or broken logic at all.

v3 -> v4

* Use dev_err instead of custom NE log pattern.
* Return IRQ_NONE when interrupts are not handled.

v2 -> v3

* Remove the WARN_ON calls.
* Update static calls sanity checks.
* Remove "ratelimited" from the logs that are not in the ioctl call
  paths.

v1 -> v2

* Add log pattern for NE.
* Update goto labels to match their purpose.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru-Catalin Vasile <lexnv@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-7-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Handle PCI device command requests
Andra Paraschiv [Mon, 21 Sep 2020 12:17:19 +0000 (15:17 +0300)]
nitro_enclaves: Handle PCI device command requests

The Nitro Enclaves PCI device exposes a MMIO space that this driver
uses to submit command requests and to receive command replies e.g. for
enclave creation / termination or setting enclave resources.

Add logic for handling PCI device command requests based on the given
command type.

Register an MSI-X interrupt vector for command reply notifications to
handle this type of communication events.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* No changes.

v7 -> v8

* Update function signature for submit request and retrive reply
  functions as they only returned 0, no error code.
* Include command type value in the error logs of ne_do_request().

v6 -> v7

* No changes.

v5 -> v6

* Update documentation to kernel-doc format.

v4 -> v5

* Remove sanity checks for situations that shouldn't happen, only if
  buggy system or broken logic at all.

v3 -> v4

* Use dev_err instead of custom NE log pattern.
* Return IRQ_NONE when interrupts are not handled.

v2 -> v3

* Remove the WARN_ON calls.
* Update static calls sanity checks.
* Remove "ratelimited" from the logs that are not in the ioctl call
  paths.

v1 -> v2

* Add log pattern for NE.
* Remove the BUG_ON calls.
* Update goto labels to match their purpose.
* Add fix for kbuild report:
  https://lore.kernel.org/lkml/202004231644.xTmN4Z1z%25lkp@intel.com/

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru-Catalin Vasile <lexnv@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-6-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Init PCI device driver
Andra Paraschiv [Mon, 21 Sep 2020 12:17:18 +0000 (15:17 +0300)]
nitro_enclaves: Init PCI device driver

The Nitro Enclaves PCI device is used by the kernel driver as a means of
communication with the hypervisor on the host where the primary VM and
the enclaves run. It handles requests with regard to enclave lifetime.

Setup the PCI device driver and add support for MSI-X interrupts.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Init the reference to the ne_pci_dev in the ne_devs data structure.

v7 -> v8

* Add NE PCI driver shutdown logic.

v6 -> v7

* No changes.

v5 -> v6

* Update documentation to kernel-doc format.

v4 -> v5

* Remove sanity checks for situations that shouldn't happen, only if
  buggy system or broken logic at all.

v3 -> v4

* Use dev_err instead of custom NE log pattern.
* Update NE PCI driver name to "nitro_enclaves".

v2 -> v3

* Remove the GPL additional wording as SPDX-License-Identifier is
  already in place.
* Remove the WARN_ON calls.
* Remove linux/bug include that is not needed.
* Update static calls sanity checks.
* Remove "ratelimited" from the logs that are not in the ioctl call
  paths.
* Update kzfree() calls to kfree().

v1 -> v2

* Add log pattern for NE.
* Update PCI device setup functions to receive PCI device data structure and
  then get private data from it inside the functions logic.
* Remove the BUG_ON calls.
* Add teardown function for MSI-X setup.
* Update goto labels to match their purpose.
* Implement TODO for NE PCI device disable state check.
* Update function name for NE PCI device probe / remove.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru-Catalin Vasile <lexnv@amazon.com>
Signed-off-by: Alexandru Ciobotaru <alcioa@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-5-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Define enclave info for internal bookkeeping
Andra Paraschiv [Mon, 21 Sep 2020 12:17:17 +0000 (15:17 +0300)]
nitro_enclaves: Define enclave info for internal bookkeeping

The Nitro Enclaves driver keeps an internal info per each enclave.

This is needed to be able to manage enclave resources state, enclave
notifications and have a reference of the PCI device that handles
command requests for enclave lifetime management.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Add data structure to keep references to both Nitro Enclaves misc and
  PCI devices.

v7 -> v8

* No changes.

v6 -> v7

* Update the naming and add more comments to make more clear the logic
  of handling full CPU cores and dedicating them to the enclave.

v5 -> v6

* Update documentation to kernel-doc format.
* Include in the enclave memory region data structure the user space
  address and size for duplicate user space memory regions checks.

v4 -> v5

* Include enclave cores field in the enclave metadata.
* Update the vCPU ids data structure to be a cpumask instead of a list.

v3 -> v4

* Add NUMA node field for an enclave metadata as the enclave memory and
  CPUs need to be from the same NUMA node.

v2 -> v3

* Remove the GPL additional wording as SPDX-License-Identifier is
  already in place.

v1 -> v2

* Add enclave memory regions and vcpus count for enclave bookkeeping.
* Update ne_state comments to reflect NE_START_ENCLAVE ioctl naming
  update.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru-Catalin Vasile <lexnv@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-4-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Define the PCI device interface
Andra Paraschiv [Mon, 21 Sep 2020 12:17:16 +0000 (15:17 +0300)]
nitro_enclaves: Define the PCI device interface

The Nitro Enclaves (NE) driver communicates with a new PCI device, that
is exposed to a virtual machine (VM) and handles commands meant for
handling enclaves lifetime e.g. creation, termination, setting memory
regions. The communication with the PCI device is handled using a MMIO
space and MSI-X interrupts.

This device communicates with the hypervisor on the host, where the VM
that spawned the enclave itself runs, e.g. to launch a VM that is used
for the enclave.

Define the MMIO space of the NE PCI device, the commands that are
provided by this device. Add an internal data structure used as private
data for the PCI device driver and the function for the PCI device
command requests handling.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* Fix indent for the NE PCI device command types enum.

v7 -> v8

* No changes.

v6 -> v7

* Update the documentation to include references to the NE PCI device id
  and MMIO bar.

v5 -> v6

* Update documentation to kernel-doc format.

v4 -> v5

* Add a TODO for including flags in the request to the NE PCI device to
  set a memory region for an enclave. It is not used for now.

v3 -> v4

* Remove the "packed" attribute and include padding in the NE data
  structures.

v2 -> v3

* Remove the GPL additional wording as SPDX-License-Identifier is
  already in place.

v1 -> v2

* Update path naming to drivers/virt/nitro_enclaves.
* Update NE_ENABLE_OFF / NE_ENABLE_ON defines.

Reviewed-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Alexandru-Catalin Vasile <lexnv@amazon.com>
Signed-off-by: Alexandru Ciobotaru <alcioa@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-3-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonitro_enclaves: Add ioctl interface definition
Andra Paraschiv [Mon, 21 Sep 2020 12:17:15 +0000 (15:17 +0300)]
nitro_enclaves: Add ioctl interface definition

The Nitro Enclaves driver handles the enclave lifetime management. This
includes enclave creation, termination and setting up its resources such
as memory and CPU.

An enclave runs alongside the VM that spawned it. It is abstracted as a
process running in the VM that launched it. The process interacts with
the NE driver, that exposes an ioctl interface for creating an enclave
and setting up its resources.

Changelog

v9 -> v10

* Update commit message to include the changelog before the SoB tag(s).

v8 -> v9

* No changes.

v7 -> v8

* Add NE custom error codes for user space memory regions not backed by
  pages multiple of 2 MiB, invalid flags and enclave CID.
* Add max flag value for enclave image load info.

v6 -> v7

* Clarify in the ioctls documentation that the return value is -1 and
  errno is set on failure.
* Update the error code value for NE_ERR_INVALID_MEM_REGION_SIZE as it
  gets in user space as value 25 (ENOTTY) instead of 515. Update the
  NE custom error codes values range to not be the same as the ones
  defined in include/linux/errno.h, although these are not propagated
  to user space.

v5 -> v6

* Fix typo in the description about the NE CPU pool.
* Update documentation to kernel-doc format.
* Remove the ioctl to query API version.

v4 -> v5

* Add more details about the ioctl calls usage e.g. error codes, file
  descriptors used.
* Update the ioctl to set an enclave vCPU to not return a file
  descriptor.
* Add specific NE error codes.

v3 -> v4

* Decouple NE ioctl interface from KVM API.
* Add NE API version and the corresponding ioctl call.
* Add enclave / image load flags options.

v2 -> v3

* Remove the GPL additional wording as SPDX-License-Identifier is
  already in place.

v1 -> v2

* Add ioctl for getting enclave image load metadata.
* Update NE_ENCLAVE_START ioctl name to NE_START_ENCLAVE.
* Add entry in Documentation/userspace-api/ioctl/ioctl-number.rst for NE
  ioctls.
* Update NE ioctls definition based on the updated ioctl range for major
  and minor.

Reviewed-by: Alexander Graf <graf@amazon.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Alexandru Vasile <lexnv@amazon.com>
Signed-off-by: Andra Paraschiv <andraprs@amazon.com>
Link: https://lore.kernel.org/r/20200921121732.44291-2-andraprs@amazon.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agointerconnect: imx: simplify the return expression of imx_icc_unregister
Liu Shixin [Mon, 21 Sep 2020 08:24:37 +0000 (16:24 +0800)]
interconnect: imx: simplify the return expression of imx_icc_unregister

Simplify the return expression.

Signed-off-by: Liu Shixin <liushixin2@huawei.com>
Link: https://lore.kernel.org/r/20200921082437.2591461-1-liushixin2@huawei.com
Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
3 years agointerconnect: imx: Simplify with dev_err_probe()
Krzysztof Kozlowski [Wed, 2 Sep 2020 17:24:33 +0000 (19:24 +0200)]
interconnect: imx: Simplify with dev_err_probe()

Common pattern of handling deferred probe can be simplified with
dev_err_probe().  Less code and the error value gets printed.

Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
Link: https://lore.kernel.org/r/20200902172433.1138-2-krzk@kernel.org
Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
3 years agointerconnect: core: Simplify with dev_err_probe()
Krzysztof Kozlowski [Wed, 2 Sep 2020 17:24:32 +0000 (19:24 +0200)]
interconnect: core: Simplify with dev_err_probe()

Common pattern of handling deferred probe can be simplified with
dev_err_probe().  Less code and the error value gets printed.

Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
Link: https://lore.kernel.org/r/20200902172433.1138-1-krzk@kernel.org
Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
3 years agoMerge branch 'icc-syncstate' into icc-next
Georgi Djakov [Fri, 18 Sep 2020 06:12:40 +0000 (09:12 +0300)]
Merge branch 'icc-syncstate' into icc-next

* icc-syncstate:
  interconnect: Add get_bw() callback
  interconnect: Add sync state support
  interconnect: qcom: Use icc_sync_state

Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
3 years agointerconnect: qcom: Use icc_sync_state
Georgi Djakov [Tue, 25 Aug 2020 17:01:52 +0000 (20:01 +0300)]
interconnect: qcom: Use icc_sync_state

Lowering the bandwidth on the bus might have negative consequences if
it's done before all consumers had a chance to cast their vote. Now by
default the framework sets the bandwidth to maximum during boot. We need
to use the icc_sync_state callback to notify the framework when all
consumers are probed and there is no need to keep the bandwidth set to
maximum anymore.

Link: https://lore.kernel.org/r/20200825170152.6434-4-georgi.djakov@linaro.org
Reviewed-by: Saravana Kannan <saravanak@google.com>
Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
3 years agointerconnect: Add sync state support
Georgi Djakov [Tue, 25 Aug 2020 17:01:51 +0000 (20:01 +0300)]
interconnect: Add sync state support

The bootloaders often do some initial configuration of the interconnects
in the system and we want to keep this configuration until all consumers
have probed and expressed their bandwidth needs. This is because we don't
want to change the configuration by starting to disable unused paths until
every user had a chance to request the amount of bandwidth it needs.

To accomplish this we will implement an interconnect specific sync_state
callback which will synchronize (aggregate and set) the current bandwidth
settings when all consumers have been probed.

Link: https://lore.kernel.org/r/20200825170152.6434-3-georgi.djakov@linaro.org
Reviewed-by: Saravana Kannan <saravanak@google.com>
Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
3 years agointerconnect: Add get_bw() callback
Georgi Djakov [Tue, 25 Aug 2020 17:01:50 +0000 (20:01 +0300)]
interconnect: Add get_bw() callback

The interconnect controller hardware may support querying the current
bandwidth settings, so add a callback for providers to implement this
functionality if supported.

Link: https://lore.kernel.org/r/20200825170152.6434-2-georgi.djakov@linaro.org
Reviewed-by: Saravana Kannan <saravanak@google.com>
Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
3 years agouio: free uio id after uio file node is freed
Lang Dai [Mon, 14 Sep 2020 03:26:41 +0000 (11:26 +0800)]
uio: free uio id after uio file node is freed

uio_register_device() do two things.
1) get an uio id from a global pool, e.g. the id is <A>
2) create file nodes like /sys/class/uio/uio<A>

uio_unregister_device() do two things.
1) free the uio id <A> and return it to the global pool
2) free the file node /sys/class/uio/uio<A>

There is a situation is that one worker is calling uio_unregister_device(),
and another worker is calling uio_register_device().
If the two workers are X and Y, they go as below sequence,
1) X free the uio id <AAA>
2) Y get an uio id <AAA>
3) Y create file node /sys/class/uio/uio<AAA>
4) X free the file note /sys/class/uio/uio<AAA>
Then it will failed at the 3rd step and cause the phenomenon we saw as it
is creating a duplicated file node.

Failure reports as follows:
sysfs: cannot create duplicate filename '/class/uio/uio10'
Call Trace:
   sysfs_do_create_link_sd.isra.2+0x9e/0xb0
   sysfs_create_link+0x25/0x40
   device_add+0x2c4/0x640
   __uio_register_device+0x1c5/0x576 [uio]
   adf_uio_init_bundle_dev+0x231/0x280 [intel_qat]
   adf_uio_register+0x1c0/0x340 [intel_qat]
   adf_dev_start+0x202/0x370 [intel_qat]
   adf_dev_start_async+0x40/0xa0 [intel_qat]
   process_one_work+0x14d/0x410
   worker_thread+0x4b/0x460
   kthread+0x105/0x140
 ? process_one_work+0x410/0x410
 ? kthread_bind+0x40/0x40
 ret_from_fork+0x1f/0x40
 Code: 85 c0 48 89 c3 74 12 b9 00 10 00 00 48 89 c2 31 f6 4c 89 ef
 e8 ec c4 ff ff 4c 89 e2 48 89 de 48 c7 c7 e8 b4 ee b4 e8 6a d4 d7
 ff <0f> 0b 48 89 df e8 20 fa f3 ff 5b 41 5c 41 5d 5d c3 66 0f 1f 84
---[ end trace a7531c1ed5269e84 ]---
 c6xxvf b002:00:00.0: Failed to register UIO devices
 c6xxvf b002:00:00.0: Failed to register UIO devices

Signed-off-by: Lang Dai <lang.dai@intel.com>
Link: https://lore.kernel.org/r/1600054002-17722-1-git-send-email-lang.dai@intel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agonvmem: core: fix missing of_node_put() in of_nvmem_device_get()
Vadym Kochan [Thu, 17 Sep 2020 13:44:37 +0000 (14:44 +0100)]
nvmem: core: fix missing of_node_put() in of_nvmem_device_get()

of_parse_phandle() returns device_node with incremented ref count
which needs to be decremented by of_node_put() when device_node
is not used.

Fixes: e2a5402ec7c6 ("nvmem: Add nvmem_device based consumer apis.")
Signed-off-by: Vadym Kochan <vadym.kochan@plvision.eu>
Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
Link: https://lore.kernel.org/r/20200917134437.16637-5-srinivas.kandagatla@linaro.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>