Age | Commit message (Collapse) | Author | Files | Lines |
|
On driver unload any pending descriptors are flushed and pending
DMA descriptors are explicitly completed:
idxd_dmaengine_drv_remove() ->
drv_disable_wq() ->
idxd_wq_free_irq() ->
idxd_flush_pending_descs() ->
idxd_dma_complete_txd()
With this done during driver unload any remaining descriptor is
likely stuck and can be dropped. Even so, the descriptor may still
have a callback set that could no longer be accessible. An
example of such a problem is when the dmatest fails and the dmatest
module is unloaded. The failure of dmatest leaves descriptors with
dma_async_tx_descriptor::callback pointing to code that no longer
exist. This causes a page fault as below at the time the IDXD driver
is unloaded when it attempts to run the callback:
BUG: unable to handle page fault for address: ffffffffc0665190
#PF: supervisor instruction fetch in kernel mode
#PF: error_code(0x0010) - not-present page
Fix this by clearing the callback pointers on the transmit
descriptors only when workqueue is disabled.
Fixes: 403a2e236538 ("dmaengine: idxd: change MSIX allocation based on per wq activation")
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Fenghua Yu <fenghua.yu@intel.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/37d06b772aa7f8863ca50f90930ea2fd80b38fc3.1670452419.git.reinette.chatre@intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
On driver unload any pending descriptors are flushed at the
time the interrupt is freed:
idxd_dmaengine_drv_remove() ->
drv_disable_wq() ->
idxd_wq_free_irq() ->
idxd_flush_pending_descs().
If there are any descriptors present that need to be flushed this
flow triggers a "not present" page fault as below:
BUG: unable to handle page fault for address: ff391c97c70c9040
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
The address that triggers the fault is the address of the
descriptor that was freed moments earlier via:
drv_disable_wq()->idxd_wq_free_resources()
Fix the use after free by freeing the descriptors after any possible
usage. This is done after idxd_wq_reset() to ensure that the memory
remains accessible during possible completion writes by the device.
Fixes: 63c14ae6c161 ("dmaengine: idxd: refactor wq driver enable/disable operations")
Suggested-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Fenghua Yu <fenghua.yu@intel.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/6c4657d9cff0a0a00501a7b928297ac966e9ec9d.1670452419.git.reinette.chatre@intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
The workqueue is enabled when the appropriate driver is loaded and
disabled when the driver is removed. When the driver is removed it
assumes that the workqueue was enabled successfully and proceeds to
free allocations made during workqueue enabling.
Failure during workqueue enabling does not prevent the driver from
being loaded. This is because the error path within drv_enable_wq()
returns success unless a second failure is encountered
during the error path. By returning success it is possible to load
the driver even if the workqueue cannot be enabled and
allocations that do not exist are attempted to be freed during
driver remove.
Some examples of problematic flows:
(a)
idxd_dmaengine_drv_probe() -> drv_enable_wq() -> idxd_wq_request_irq():
In above flow, if idxd_wq_request_irq() fails then
idxd_wq_unmap_portal() is called on error exit path, but
drv_enable_wq() returns 0 because idxd_wq_disable() succeeds. The
driver is thus loaded successfully.
idxd_dmaengine_drv_remove()->drv_disable_wq()->idxd_wq_unmap_portal()
Above flow on driver unload triggers the WARN in devm_iounmap() because
the device resource has already been removed during error path of
drv_enable_wq().
(b)
idxd_dmaengine_drv_probe() -> drv_enable_wq() -> idxd_wq_request_irq():
In above flow, if idxd_wq_request_irq() fails then
idxd_wq_init_percpu_ref() is never called to initialize the percpu
counter, yet the driver loads successfully because drv_enable_wq()
returns 0.
idxd_dmaengine_drv_remove()->__idxd_wq_quiesce()->percpu_ref_kill():
Above flow on driver unload triggers a BUG when attempting to drop the
initial ref of the uninitialized percpu ref:
BUG: kernel NULL pointer dereference, address: 0000000000000010
Fix the drv_enable_wq() error path by returning the original error that
indicates failure of workqueue enabling. This ensures that the probe
fails when an error is encountered and the driver remove paths are only
attempted when the workqueue was enabled successfully.
Fixes: 1f2bb40337f0 ("dmaengine: idxd: move wq_enable() to device.c")
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Fenghua Yu <fenghua.yu@intel.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/e8d8116e5efa0fd14fadc5adae6ffd319f0e5ff1.1670452419.git.reinette.chatre@intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Nothing in this file needs anything from linux/msi.h
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Vinod Koul <vkoul@kernel.org>
Cc: dmaengine@vger.kernel.org
Link: https://lore.kernel.org/r/20221113202428.573536003@linutronix.de
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
When IDXD is not configurable, that means its WQ, engine, and group
configurations cannot be changed. But it can be disabled and its state
should be set as disabled regardless it's configurable or not.
Fix this by setting device state IDXD_DEV_DISABLED for read-only device
as well in idxd_device_clear_state().
Fixes: cf4ac3fef338 ("dmaengine: idxd: fix lockdep warning on device driver removal")
Signed-off-by: Fengqian Gao <fengqian.gao@intel.com>
Reviewed-by: Xiaochen Shen <xiaochen.shen@intel.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Fenghua Yu <fenghua.yu@intel.com>
Link: https://lore.kernel.org/r/20220930032835.2290-1-fengqian.gao@intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
>From Intel IAA spec [1], Intel IAA does not support batch processing.
Two batch related default values for IAA are incorrect in current code:
(1) The max batch size of device is set during device initialization,
that indicates batch is supported. It should be always 0 on IAA.
(2) The max batch size of work queue is set to WQ_DEFAULT_MAX_BATCH (32)
as the default value regardless of Intel DSA or IAA device during
work queue setup and cleanup. It should be always 0 on IAA.
Fix the issues by setting the max batch size of device and max batch
size of work queue to 0 on IAA device, that means batch is not
supported.
[1]: https://cdrdv2.intel.com/v1/dl/getContent/721858
Fixes: 23084545dbb0 ("dmaengine: idxd: set max_xfer and max_batch for RO device")
Fixes: 92452a72ebdf ("dmaengine: idxd: set defaults for wq configs")
Fixes: bfe1d56091c1 ("dmaengine: idxd: Init and probe for Intel data accelerators")
Signed-off-by: Xiaochen Shen <xiaochen.shen@intel.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Fenghua Yu <fenghua.yu@intel.com>
Link: https://lore.kernel.org/r/20220930201528.18621-2-xiaochen.shen@intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Add sysfs knob to allow control of the number of batch descriptors that can
be concurrently processed by an engine in the group as a fraction of the
Maximum Work Descriptors in Progress value specfied in ENGCAP register.
This control knob is part of toggle for QoS control.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Co-developed-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Link: https://lore.kernel.org/r/20220917161222.2835172-6-fenghua.yu@intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Add sysfs knob to allow control of the number of work descriptors that can
be concurrently processed by an engine in the group as a fraction of the
Maximum Work Descriptors in Progress value specified in ENGCAP register.
This control knob is part of toggle for QoS control.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Co-developed-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Link: https://lore.kernel.org/r/20220917161222.2835172-5-fenghua.yu@intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
DSA 2.0 add the capability of configuring DMA ops on a per workqueue basis.
This means that certain ops can be disabled by the system administrator for
certain wq. By default, all ops are available. A bitmap is used to store
the ops due to total op size of 256 bits and it is more convenient to use a
range list to specify which bits are enabled.
One of the usage to support this is for VM migration between different
iteration of devices. The newer ops are disabled in order to allow guest to
migrate to a host that only support older ops. Another usage is to
restrict the WQ to certain operations for QoS of performance.
A sysfs of ops_config attribute is added per wq. It is only usable when the
ops_config bit is set under WQ_CAP register. This means that this attribute
will return -EOPNOTSUPP on DSA 1.x devices. The expected input is a range
list for the bits per operation the WQ supports.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Co-developed-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Link: https://lore.kernel.org/r/20220917161222.2835172-4-fenghua.yu@intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Make wq attributes access consistent. Convert ats_dis to wq flag
WQ_FLAG_ATS_DISABLE.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Co-developed-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Link: https://lore.kernel.org/r/20220917161222.2835172-2-fenghua.yu@intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Now that idxd_wq_disable_cleanup() sets the workqueue state to
IDXD_WQ_DISABLED, use a bitmap to track which workqueues have been
enabled. This will then be used to determine which workqueues
should be re-enabled when attempting a software reset to recover
from a device halt state.
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Vinod Koul <vkoul@kernel.org>
Signed-off-by: Jerry Snitselaar <jsnitsel@redhat.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/20220928154856.623545-3-jsnitsel@redhat.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
If we are calling idxd_wq_disable_cleanup(), the workqueue should be
in a disabled state. So set the workqueue state to IDXD_WQ_DISABLED so
that the state reflects that. Currently if there is a device failure,
and a software reset is attempted the workqueues will not be
re-enabled due to idxd_wq_enable() seeing that state as already being
IDXD_WQ_ENABLED.
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Vinod Koul <vkoul@kernel.org>
Signed-off-by: Jerry Snitselaar <jsnitsel@redhat.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/20220928154856.623545-2-jsnitsel@redhat.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Testing shown that when a wq mode is setup to be dedicated and then torn
down and reconfigured to shared, the wq configured end up being dedicated
anyays. The root cause is when idxd_device_wqs_clear_state() gets called
during idxd_driver removal, idxd_wq_disable_cleanup() does not get called
vs when the wq driver is removed first. The check of wq state being
"enabled" causes the cleanup to be bypassed. However, idxd_driver->remove()
releases all wq drivers. So the wqs goes to "disabled" state and will never
be "enabled". By that point, the driver has no idea if the wq was
previously configured or clean. So force call idxd_wq_disable_cleanup() on
all wqs always to make sure everything gets cleaned up.
Reported-by: Tony Zhu <tony.zhu@intel.com>
Tested-by: Tony Zhu <tony.zhu@intel.com>
Fixes: 0dcfe41e9a4c ("dmanegine: idxd: cleanup all device related bits after disabling device")
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Co-developed-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Link: https://lore.kernel.org/r/20220628230056.2527816-1-fenghua.yu@intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/vkoul/dmaengine
Pull dmaengine updates from Vinod Koul:
"Nothing special, this includes a couple of new device support and new
driver support and bunch of driver updates.
New support:
- Tegra gpcdma driver support
- Qualcomm SM8350, Sm8450 and SC7280 device support
- Renesas RZN1 dma and platform support
Updates:
- stm32 device pause/resume support and updates
- DMA memset ops Documentation and usage clarification
- deprecate '#dma-channels' & '#dma-requests' bindings
- driver updates for stm32, ptdma idsx etc"
* tag 'dmaengine-5.19-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/vkoul/dmaengine: (87 commits)
dmaengine: idxd: make idxd_wq_enable() return 0 if wq is already enabled
dmaengine: sun6i: Add support for the D1 variant
dmaengine: sun6i: Add support for 34-bit physical addresses
dmaengine: sun6i: Do not use virt_to_phys
dt-bindings: dma: sun50i-a64: Add compatible for D1
dmaengine: tegra: Remove unused switch case
dmaengine: tegra: Fix uninitialized variable usage
dmaengine: stm32-dma: add device_pause/device_resume support
dmaengine: stm32-dma: rename pm ops before dma pause/resume introduction
dmaengine: stm32-dma: pass DMA_SxSCR value to stm32_dma_handle_chan_done()
dmaengine: stm32-dma: introduce stm32_dma_sg_inc to manage chan->next_sg
dmaengine: stm32-dmamux: avoid reset of dmamux if used by coprocessor
dmaengine: qcom: gpi: Add support for sc7280
dt-bindings: dma: pl330: Add power-domains
dmaengine: stm32-mdma: use dev_dbg on non-busy channel spurious it
dmaengine: stm32-mdma: fix chan initialization in stm32_mdma_irq_handler()
dmaengine: stm32-mdma: remove GISR1 register
dmaengine: ti: deprecate '#dma-channels'
dmaengine: mmp: deprecate '#dma-channels'
dmaengine: pxa: deprecate '#dma-channels' and '#dma-requests'
...
|
|
When calling idxd_wq_enable() and wq is already enabled, code should return 0
and indicate function is successful instead of return error code and fail.
This should also put idxd_wq_enable() in sync with idxd_wq_disable() where
it returns 0 if wq is already disabled.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/165090980906.1378449.1939401700832432886.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Calling synchronize_irq() right before free_irq() is quite useless. On one
hand the IRQ can easily fire again before free_irq() is entered, on the
other hand free_irq() itself calls synchronize_irq() internally (in a race
condition free way), before any state associated with the IRQ is freed.
Signed-off-by: Minghao Chi <chi.minghao@zte.com.cn>
Link: https://lore.kernel.org/r/20220516115412.1651772-1-chi.minghao@zte.com.cn
Acked-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Skip wq irq resources freeing when wq type is not kernel since the driver
skips the irq alloction during wq enable. Add check in wq type check in
idxd_wq_free_irq() to mirror idxd_wq_request_irq().
Fixes: 63c14ae6c161 ("dmaengine: idxd: refactor wq driver enable/disable operations")
Reported-by: Tony Zu <tony.zhu@intel.com>
Tested-by: Tony Zu <tony.zhu@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/165176310726.2112428.7474366910758522079.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
idxd_wq_device_reset_cleanup() already calls idxd_wq_disable_cleanup().
There is no need to call idxd_wq_disable_cleanup() again in
idxd_device_wqs_clear_state(). Remove redudant call from
idxd_wq_device_reset_cleanup().
Fixes: 0dcfe41e9a4c ("dmanegine: idxd: cleanup all device related bits after disabling device")
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/165231365717.986350.2441351765955825964.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Call idxd_wq_free_irq() in the drv_disable_wq() function before
idxd_wq_reset() is called. Otherwise the wq type is reset and the irq does
not get freed.
Fixes: 63c14ae6c161 ("dmaengine: idxd: refactor wq driver enable/disable operations")
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/165231367316.986407.11001767338124941736.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Jacob reported that with lockdep debug turned on, idxd_device_driver
removal causes kernel splat from lock assert warning for
idxd_device_wqs_clear_state(). Make sure
idxd_device_wqs_clear_state() holds the wq lock for each wq when
cleaning the wq state. Move the call outside of the device spinlock.
Reported-by: Jacob Pan <jacob.jun.pan@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/165231364426.986304.9294302800482492780.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
The idxd driver always gated the pasid enabling under a single knob and
this assumption is incorrect. The pasid used for kernel operation can be
independently toggled and has no dependency on the user pasid (and vice
versa). Split the two so they are independent "enabled" flags.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/165231431746.986466.5666862038354800551.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Move the core driver operations from wq driver to the drv_enable_wq() and
drv_disable_wq() functions. The move should reduce the wq driver's
knowledge of the core driver operations and prevent code confusion for
future wq drivers.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/165047301643.3841827.11222723219862233060.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
If the device shows up as read-only configuration, skip the clearing of the
state as the context must be preserved for device re-enable after being
disabled.
Fixes: 0dcfe41e9a4c ("dmanegine: idxd: cleanup all device related bits after disabling device")
Reported-by: Tony Zhu <tony.zhu@intel.com>
Tested-by: Tony Zhu <tony.zhu@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/164971479479.2200566.13980022473526292759.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Load the max_xfer_size and max_batch_size values from the values read from
registers to the shadow variables. This will allow the read-only device to
display the correct values for the sysfs attributes.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/164971507673.2201761.11244446608988838897.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
The driver currently programs the system pasid to the WQ preemptively when
system pasid is enabled. Given that a dwq will reprogram the pasid and
possibly a different pasid, the programming is not necessary. The pasid_en
bit can be set for swq as it does not need pasid programming but
needs the pasid_en bit. Remove system pasid programming on device config
write. Add pasid programming for kernel wq type on wq driver enable. The
char dev driver already reprograms the dwq on ->open() call so there's no
change.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/164935607115.1660372.6734518676950372366.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
There are certain parts of WQ that needs to be cleaned up even after WQ is
disabled during the device disable. Those are the unchangeable parts for a
WQ when the device is still enabled. Move the cleanup outside of WQ state
check. Remove idxd_wq_disable_cleanup() inside idxd_wq_device_reset_cleanup()
since only the unchangeable parts need to be cleared.
Fixes: 0f225705cf65 ("dmaengine: idxd: fix wq settings post wq disable")
Reported-by: Tony Zhu <tony.zhu@intel.com>
Tested-by: Tony Zhu <tony.zhu@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/164919561905.1455025.13542366389944678346.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
When clearing the group configurations, the driver fails to restore the
default setting for DSA 1.x based devices. Add defaults in
idxd_groups_clear_state() for traffic class configuration.
Fixes: ade8a86b512c ("dmaengine: idxd: Set defaults for GRPCFG traffic class")
Reported-by: Binuraj Ravindran <binuraj.ravindran@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/164304123369.824298.6952463420266592087.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
DSA spec v1.2 has changed the term of "bandwidth tokens" to "read buffers"
in order to make the concept clearer. Deprecate bandwidth token
naming in the driver and convert to read buffers in order to match with
the spec and reduce confusion when reading the spec.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163951338932.2988321.6162640806935567317.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
By the spec, wq size and group association is not changeable unless device
is disabled. Exclude clearing the shadow copy on wq disable/reset. This
allows wq type to be changed after disable to be re-enabled.
Move the size and group association to its own cleanup and only call it
during device disable.
Fixes: 0dcfe41e9a4c ("dmanegine: idxd: cleanup all device related bits after disabling device")
Reported-by: Lucas Van <lucas.van@intel.com>
Tested-by: Lucas Van <lucas.van@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163951291732.2987775.13576571320501115257.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Change the driver where WQ interrupt is requested only when wq is being
enabled. This new scheme set things up so that request_threaded_irq() is
only called when a kernel wq type is being enabled. This also sets up for
future interrupt request where different interrupt handler such as wq
occupancy interrupt can be setup instead of the wq completion interrupt.
Not calling request_irq() until the WQ actually needs an irq also prevents
wasting of CPU irq vectors on x86 systems, which is a limited resource.
idxd_flush_pending_descs() is moved to device.c since descriptor flushing
is now part of wq disable rather than shutdown().
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163942149487.2412839.6691222855803875848.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
With irq_entry already being associated with the wq in a 1:1 relationship,
embed the irq_entry in the idxd_wq struct and remove back pointers for
idxe_wq and idxd_device. In the process of this work, clean up the interrupt
handle assignment so that there's no decision to be made during submit
call on where interrupt handle value comes from. Set the interrupt handle
during irq request initialization time.
irq_entry 0 is designated as special and is tied to the device itself.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163942148362.2412839.12055447853311267866.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Add a sysfs knob to allow tuning of retries for the kernel ENQCMDS
descriptor submission. While on host, it is not as likely that ENQCMDS
return busy during normal operations due to the driver controlling the
number of descriptors allocated for submission. However, when the driver is
operating as a guest driver, the chance of retry goes up significantly due
to sharing a wq with multiple VMs. A default value is provided with the
system admin being able to tune the value on a per WQ basis.
Suggested-by: Sanjay Kumar <sanjay.k.kumar@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163820629464.2702134.7577370098568297574.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Add default values for wq size, max_xfer_size and max_batch_size. These
values should provide a general guidance for the wq configuration when
the user does not specify any specific values.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163528473483.3926048.7950067926287180976.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
"Interrupt handle revoked" is an event that happens when the driver is
running on a guest kernel and the VM is migrated to a new machine.
The device will trigger an interrupt that signals to the guest driver
that the interrupt handles need to be replaced.
The misc irq thread function calls a helper function to handle the
event. The function uses the WQ percpu_ref to quiesce the kernel
submissions. It then replaces the interrupt handles by requesting
interrupt handle command for each I/O MSIX vector. Once the handle is
updated, the driver will unblock the submission path to allow new
submissions.
The submitter will attempt to acquire a percpu_ref before submission. When
the request fails, it will wait on the wq_resurrect 'completion'.
The driver does anticipate the possibility of descriptors being submitted
before the WQ percpu_ref is killed. If a descriptor has already been
submitted, it will return with incorrect interrupt handle status. The
descriptor will be re-submitted with the new interrupt handle on the
completion path. For descriptors with incorrect interrupt handles,
completion interrupt won't be triggered.
At the completion of the interrupt handle refresh, the handling function
will call idxd_int_handle_refresh_drain() to issue drain descriptors to
each of the wq with associated interrupt handle. The drain descriptor will have
interrupt request set but without completion record. This will ensure all
descriptors with incorrect interrupt completion handle get drained and
a completion interrupt is triggered for the guest driver to process them.
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Co-Developed-by: Sanjay Kumar <sanjay.k.kumar@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163528420189.3925689.18212568593220415551.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Add a locked version of idxd_quiesce() call so that the quiesce can be
called with a lock in situations where the lock is not held by the caller.
In the driver probe/remove path, the lock is already held, so the raw
version can be called w/o locking.
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163528418980.3925689.5841907054957931211.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Attach int_handle to irq_entry. This removes the separate management of int
handles and reduces the confusion of interating through int handles that is
off by 1 count.
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163528417065.3925689.11505755433684476288.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
According to core-api/dma-api-howto.rst, the address from
dma_alloc_coherent is gauranteed to align to the smallest PAGE_SIZE order.
That supercedes the 64B/32B alignment requirement of the completion record.
Remove alignment adjustment code.
Tested-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163517396063.3484297.7494385225280705372.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Device reset clears the MSIXPERM table and the device registers. Re-program
the MSIXPERM table and re-enable the error interrupts post reset.
Fixes: 745e92a6d816 ("dmaengine: idxd: idxd: move remove() bits for idxd 'struct device' to device.c")
Reported-by: Sanjay Kumar <sanjay.k.kumar@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163054188513.2853562.12077053294595278181.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Remove setting of wq type on guest kernel during configuration load on RO
device config. The user will set the kernel wq type and this setting based
on config is not necessary.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163474724511.2607444.1876715711451990426.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
DSA spec 1.2 has moved the GENCFG register under the GENCAP configuration
support with respect to writability. Add check in driver before writing to
GENCFG register.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163406171896.1303830.11217958011385656998.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
percpu_ref_tryget_live() is safe to call as long as ref is between init and
exit according to the function comment. Move percpu_ref_exit() so it is
called after the dma channel is no longer valid to ensure this holds true.
Fixes: 93a40a6d7428 ("dmaengine: idxd: add percpu_ref to descriptor submission path")
Suggested-by: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163294293832.914350.10326422026738506152.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
The spinlock is not being used in hard interrupt context. There is no need
to disable irq when acquiring the lock. The interrupt thread handler also
is not in bottom half context, therefore we can also remove disabling of
the bh. Convert all dev_lock acquisition to plain spin_lock() calls.
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/162984026772.1939166.11504067782824765879.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
The cmd_lock spinlock is not being used in hard interrupt context. There is
no need to disable irq when acquiring the lock. Convert all cmd_lock
acquisition to plain spin_lock() calls.
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/162984027930.1939209.15758413737332339204.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
DSA spec says WQ priv bit is 0 if the Privileged Mode Enable field of the
PCI Express PASID capability is 0 and pasid is enabled. Make sure that the
WQCFG priv field is set correctly according to usage type. Reject config if
setting up kernel WQ type and no support. Also add the correct priv setup
for a descriptor.
Fixes: 484f910e93b4 ("dmaengine: idxd: fix wq config registers offset programming")
Cc: Ramesh Thomas <ramesh.thomas@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/162939084657.903168.14160019185148244596.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Since submission is sent to limited portal, the actual wq size for shared
wq is set by the threshold rather than the wq size. When the wq type is
shared, set the allocated descriptors to the threshold.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/162827151733.3459223.3829837172226042408.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
The block on fault flag is not cleared when we disable or reset wq. This
causes it to remain set if the user does not clear it on the next
configuration load. Add clear of flag in dxd_wq_disable_cleanup()
routine.
Fixes: da32b28c95a7 ("dmaengine: idxd: cleanup workqueue config after disabling")
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/162803023553.3086015.8158952172068868803.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
Enabling device and wq returns standard errno and that does not provide
enough details to indicate what exactly failed. The hardware command status
is only 8bits. Expand the command status to 32bits and use the upper 16
bits to define software errors to provide more details on the exact
failure. Bit 31 will be used to indicate the error is software set as the
driver is using some of the spec defined hardware error as well.
Cc: Ramesh Thomas <ramesh.thomas@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/162681373579.1968485.5891788397526827892.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
The device submission portal is on a 4k page and any of those 64bit aligned
address on the page can be used for descriptor submission. By rotating the
offset through the 4k range and prevent successive writes to the same MMIO
address, performance improvement is observed through testing.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/162681372446.1968485.10634280461681015569.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
The original architecture of /sys/bus/dsa invented a scheme whereby
a single entry in the list of bus drivers, /sys/bus/drivers/dsa,
handled all device types and internally routed them to different
different drivers. Those internal drivers were invisible to
userspace.
With the idxd driver transitioned to a proper bus device-driver model,
the legacy behavior needs to be preserved due to it being exposed to
user space via sysfs. Create a compat driver to provide the legacy
behavior for /sys/bus/dsa/drivers/dsa. This should satisfy user
tool accel-config v3.2 or ealier where this behavior is expected.
If the distro has a newer accel-config then the legacy mode does
not need to be enabled.
When the compat driver binds the device (i.e. dsa0) to the dsa driver,
it will be bound to the new idxd_drv. The wq device (i.e. wq0.0) will
be bound to either the dmaengine_drv or the user_drv. The dsa_drv
becomes a routing mechansim for the new drivers. It will not support
additional external drivers that are implemented later.
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/162637468705.744545.4399080971745974435.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|
|
The original architecture of /sys/bus/dsa invented a scheme whereby a
single entry in the list of bus drivers, /sys/bus/drivers/dsa, handled
all device types and internally routed them to different drivers.
Those internal drivers were invisible to userspace. Now, as
/sys/bus/dsa wants to grow support for alternate drivers for a given
device, for example vfio-mdev instead of kernel-internal-dmaengine, a
proper bus device-driver model is needed. The first step in that process
is separating the existing omnibus/implicit "dsa" driver into proper
individual drivers registered on /sys/bus/dsa. Establish the
idxd_user_drv driver that controls the enabling and disabling of the
wq and also register and unregister a char device to allow user space
to mmap the descriptor submission portal.
The cdev related bits are moved to the cdev driver probe/remove and out of
the drv_enabe/disable_wq() calls. These bits are exclusive to the cdev
operation and not part of the generic enable/disable of the wq device.
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/162637467578.744545.10203997610072341376.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
|