| Age | Commit message (Collapse) | Author |
|---|
|
This patch handles the following:
1. cleaned the switch statement in dmm_user.c to move the copy_from_user
to individual functions.
2. Seperated the building of Scatter gather list to Map function. The list
is created when the Map function is created and deleted when the unmap
function is called. Previously this list was created during flush function
and deleted during the invalidate function. The previous implementation would
fail if the Users call the invalidate function directly without calling flush
function, as it would require calling flush function prior to calling invalidate
function. The previous implementation also includes the overhead in creating
and deleting the SG list for every flush and invalidate calls.
3. Changed the logic to track mapping object based on MPU VA instead of page
aligned MPU VA. This is because if there are 2 MPU buffers with virtual
addresses within the Page boundary, then there would be 2 Mapping objects with
same virtual address.
4. Delete the DMM pools only when all the handles to IODMM are closed. The
ioctl DMM_IOCDELETEPOOL can be deprecated with this change.
5. Added the SYS ADMIN check for create dmm pools, and set tlb entry ioctls.
This code is currently kept commented out.
6. DMA Flush APIs are still optionally selected using CONFIG_DMM_DMA_API option.
This is not selected by default.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
omap:iodmm-send notification on close only when resources not free
Instead of sending the notification on every IOMMU close, send it
only when there are still mapped resources to avoid the remote
processor accessing these resources.
After compelting the notification, the iommu release function can
safely remove the mapped entries.
Change-Id: Ib162bb4aa4ec0ed287a4b7484f54b0712871804e
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
Previously, when restoring the iommu context, all the
TLB entries marked valid were being reloaded.
Reading dynamically loaded TLBs back from the MMU
registers during a context save operation returned
addresses that were not aligned to the boundaries of
the page sizes indicated in the respective fields.
Attempting to restore these unaligned entries in
the MMU resulted in warnings from the restore routines.
Since only the locked TLB entries need to be loaded
before restarting the processor, we are now restoring
only those. (All the locked entries were marked as
protected/preserved when they were initially loaded.)
We are still investigating the root cause of why the
alignment appears wrong in the TLB entries loaded as
a result of a TLB miss by the table walking logic.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
|
|
iommu_notify_event function is called from interrupt context, and this
function is using internally blocking_notifier_call_chain call that could
sleep.
The fix is changing blocking_notifier to raw_notifier and let the callers
of the iommu_notify_event handle the protection.
iommu_notify_event is used for 2 purposes. One is for MMU FAULT notification
to the device handler where the devh sets the error state to Fault and prevents
the broadcasting of any iommu close notifications. This shouldn't be blocking.
Second, it is used to broadcast the notifications of the Process ID that closed
the iommu handle to remote processor. This could be blocking.
Reported by Hari Nagalla
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Hari Nagalla <hnagalla@ti.com>
|
|
In case omap_hwmod_lookup returns NULL, don't proceed further for this
instance and continue with next device.
Reported by: Rajendra Nayak
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
Use dma calls for cache flushing under a configurable flag.
Use this option until Multimedia integration issues are sorted out.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
The DVFS framework for DSP is dependent on hwmod registeration completed for
DSP in iommu driver. This patch moves the iommu initialization to post-core
init to ensure the registeration is completed prior to DVFS stuff.
Without this fix, the following warning was observed during boot-up.
------------[ cut here ]------------
WARNING: at arch/arm/mach-omap2/pm.c:54 omap4_get_dsp_device+0x34/0x50()
Modules linked in:
Backtrace:
[<c0039f9c>] (dump_backtrace+0x0/0x110) from [<c0380ce8>] (dump_stack+0x18/0x1c)
r7:00000000 r6:c0041f70 r5:c044ba64 r4:00000036
[<c0380cd0>] (dump_stack+0x0/0x1c) from [<c006fde0>] (warn_slowpath_common+0x54/
0x6c)
[<c006fd8c>] (warn_slowpath_common+0x0/0x6c) from [<c006fe1c>] (warn_slowpath_nu
ll+0x24/0x2c)
r9:00000000 r8:00000000 r7:00000001 r6:c000f708 r5:c0508784
r4:c0503024
[<c006fdf8>] (warn_slowpath_null+0x0/0x2c) from [<c0041f70>] (omap4_get_dsp_devi
ce+0x34/0x50)
[<c0041f3c>] (omap4_get_dsp_device+0x0/0x50) from [<c001137c>] (omap4_pm_init_op
p_table+0x144/0x244)
r5:c0508784 r4:c050326c
[<c0011238>] (omap4_pm_init_opp_table+0x0/0x244) from [<c000f79c>] (omap2_common
_pm_init+0x94/0xbc)
r5:c0508784 r4:c0503024
[<c000f708>] (omap2_common_pm_init+0x0/0xbc) from [<c003558c>] (do_one_initcall+
0x64/0x1bc)
r5:c0508784 r4:c002b5bc
[<c0035528>] (do_one_initcall+0x0/0x1bc) from [<c0008644>] (kernel_init+0x168/0x
238)
r7:c050877c r6:c04ca560 r5:c0508784 r4:c002b5bc
[<c00084dc>] (kernel_init+0x0/0x238) from [<c00736f0>] (do_exit+0x0/0x5f0)
r7:00000013 r6:c00736f0 r5:c00084dc r4:00000000
---[ end trace dec6997083161631 ]---
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
|
|
omap:iodmm-dmm code moved to new file and fixes
The patch handles the following
1. Moved the code out of iovmm to new file.
2. Removed the config option related to IOVMM in makefile
3. Moved the DMM pool creation part to iodmm.c
4. Removed the hacks that were made to handle Tiler buffers.
5. Only buffers that are mapped would be permittted for flushing.
6. Removed the ioctl to program PTE entries directly from userspace.
Regular DMM should be used to program the PTE entries. This requires
userspace change
7. Removed unused functions.
8. Provided the support to map Anonymous buffers.
9. Provided the support to map Physical address if passed.
10. Replaced the local spinlock with global mutex lock for critical
critical sections.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
omap:iodmm-resolve compilation issue
This patch resolves the following build issue that was observed
when build with omap3 defcofig.
arch/arm/plat-omap/iodmm.c: In function 'user_to_device_map':
arch/arm/plat-omap/iodmm.c:425: error: dereferencing pointer to incomplete type
arch/arm/plat-omap/iodmm.c: In function 'dmm_user':
arch/arm/plat-omap/iodmm.c:655: error: dereferencing pointer to incomplete type
arch/arm/plat-omap/iodmm.c:744: error: dereferencing pointer to incomplete type
make[1]: *** [arch/arm/plat-omap/iodmm.o] Error 1 m
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
omap:iodmm-allow unconditional cache flush
This patch reverts back Russell King's patch to do the userspace
buffer cache flush. This allows unconditional cache operations and this
patch is considered as a hack patch until the new API is alinged with MM
team
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
omap:iommu: save/restore tlbs entries
Introduce a way to save/restore the tlbs entries.
When saving is call all 32 entries are saved.
When restoring just the valid ones are restored.
A new member in iommu obj was added to save/restore the tlbs
- iommu tlbs context: saved area
struct iotlb_entry *tlbs_e
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
omap:iommu: Change in iommu to save/restore
To save the context the tlbs are being saved using
iommu_save_tlb_entries() then the iommu is disable.
To restore the context the tlbs are being restored using
iommu_restore_tlb_entries() then the iommu is enable.
There is no need to save the registers that is why
ctx was removed from the iommu struct.
omap2_save/restore function were removed.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com
omap:iommu: dump ctx for omap4
Adding the two new registers just for omap4 in the
dumping context function.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
omap:iommu: iommu_debug fix due to api change
New function to get the version of IOMMU was added since now
there is a need to determine if its tesla or ducati.
iommu_debug.c was updated to use this API
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
omap:iommu: set MMU_GP_REG to force a bus error upon MMU fault
MMU_GP_REG[0] determines the behavior of Ducati cores upon
hitting an MMU fault (address is not mapped in L2MMU). If it is not
set, the cores are hung preventing the fault handler to run. If one
introduces a bus error, the fault handler gets to run and prints
state information which is useful for debug.
Signed-off-by: Wajahat Khan <w-khan@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
omap:iodmm-fixes to unmap and general cleanup
Following fixes are handled by this patch.
- Unmapping of buffers
- Pages were allocated on 1K boundary instead of 4K
- Use VM_IO to detect IO buffers (eg: Tiler)
- Page_cache_release calls were missing for unmapping part.
- find_containing_mapping function is extended to check based on
da address as well. This is needed during unmap part
- Cleanup of unused header files in iodmm.c
- Cleanup of unused case to clear PTE entries. PTE entries would get
cleaned during device release
- Added function headers with description
- Uncommented omap_delete_vmm_pool code.
This patch still has a hack code to handle Tiler buffer. Tiler buffer
should be handle dby VM_IO itself and doesn't need to depend on flags.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
omap:iodmm-exclude Tiler buffers from mapped information
Exclude Tiler buffers from getting added to runtime mapped information
since they are already pre-mapped in TLB.
By including them in runtime leads to MMU fault during the process
shutdown.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
Enabled the support to map the buffer that is passed through
mmap.
Also comment out the vmm delete pool function as this is broken.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
This patch adds spinlock protection to event list accesses in order
to avoid list corruption.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
This fixes the following:
1. pgd and pte entries weren't getting flushed out leading to MMU faults.
2. Cache invalidate was setting wrong size parameter to
memory_regain_ownership leading cache invalidate function to fail.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
use platform device instead of omap device for iovmm.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
Disable TWL for ES1.0 by checking the CPU Revision
[Hari] - Amended the patch to move enabling/disabling of TWL
part to iommu_get function since this needs to be executed
only once.
Signed-off-by: Ramesh Gupta G <grgupta@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
Add flush and invalidate suppor for
userspace buffers.
Signed-of-by: Ramesh Gupta G <grgupta@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
Implement notifier callback mechanism for IOMMU events
such as MMU fault and file close.
The Error handler that hooks to this registration would notify
the remote processor whenever a process closed the iommu. This is
to ensure that the Processes on Remote process stop accessing the
buffers mapped by the terminating Process
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Instead of disabling the MMU device, just disable the MMU interrupts
when the MMU fault occurs. This change is required because the user
applications might be releasing the MMU resources after the MMU fault
recovery and this requires MMU to be in enabled state. MMU will be
disabled after all the open handles to IOMMU are closed.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
omap:iommu-ported dspbridge's dmm code to iommu
This patch ports the dspbridge's dmm code to iommu location.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Ramesh Gupta <grgupta@ti.com>
omap:iommu-added cache flushing operation for L2 cache
Signed-off-by: Ramesh Gupta <grgupta@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
omap:iovmm-add interface to userspace DMM
signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
signed-off-by: Ramesh Gupta <grgupta@ti.com>
omap:iommu-event notification to userspace
Implement iommu event notifications to userspace using eventfd
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
IODMM add support for unmap functionality
Adds unmap functionality , adds the map information
of loadable sections to the mapped list.
Signed-off-by: Ramesh Gupta G <grgupta@ti.com>
|
|
This patch adapts iommu to hwmod implementation for OMAP4.
Work needs to be done for OMAP3
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
This patch removes the clk_enable/disable functions for
iommu functions, module auto idle should take care
of this.
Signed-off-by: Ramesh Gupta G<grgupta@ti.com>
|
|
omap: iommu-update irq mask to be specific about twl and tlb
Revise the IRQ mask definitions to handle the MMU faults related
to TWL fault as well as TLB miss fault.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Hiroshi Doyu <Hiroshi.DOYU@nokia.com>
omap: iommu-add functionality to get TLB miss interrupt
In order to enable TLB miss interrupt, the TWL should be
disabled. This patch provides the functionality to get the
MMU fault interrupt for a TLB miss in the cases where the
users are working with the locked TLB entries and with TWL
disabled.
New interface is added to select twl and to enable TLB miss
interrupt.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Ramesh Gupta <grgupta@ti.com>
Signed-off-by: Hiroshi Doyu <Hiroshi.DOYU@nokia.com>
|
|
wdt notification event was not being sent to the deh because
if there are resources in use by IPU the assumption was that
it is in a good state (not hanging) and the timer was being reloaded
without catching the issue. This fix is avoiding that scenario
adding an event PM_ALIVE, sent by ipu, that indicates there are
resources in use but ipu wont hibernate, and the timer needs
to be reloaded in kernel for hib_time seconds again.
Also resources clean up is now done when the wd issue is catched.
Change-Id: I34304763ce50ac51595a9a6b7c9305d95174f8f8
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
|
|
Modified to configure timer 3 that is used to trigger hibernation
and it is also used as watch dog timer. Since both ducati cores
can go to retention together and if one core is not responding
both need to be reset, we only use 1 timer
[Hari K] The free irq while freeing
checks the dev_id as the token with the dev_id that it was registered
earlier. To avoid the race condition as when the remote proc frees the
dmtimer handle save the handle in ipu_pm to use it as a token when passing
to free_irq
A cleaner solution should be investigated for watchdog timer
Change-Id: I15d3c0182d84861819cc36abf1f78cbc1674375f
Signed-off-by: Shahid Akhtar <sakhtar@ti.com>
|
|
This patch adds build related files
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Ramesh Gupta G <grgupta@ti.com>
|
|
Enable gptimer 4 and 9 to be used as ipu clk. Gptimers will
be requested when calling rproc_start() and released them
when calling rproc_stop(), also when hibernating ipu the
gptimers will be stop/start to ensure proper pm functionality.
The gptimer numbers are configurable, currently 4 will be used
by ipu_c0 and 9 by ipu_c1.
Change-Id: I9424991e34e786ab2f57a1dc625ae801290fd7b6
Reported-by: Paul Hunt <hunt@ti.com>
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
|
|
The current code is freeing dmtimer in proc stop only if the
proc state is in RUNNING. But the proc stop could be called
when the proc state is HIBERNATION.
The fix is to free dmtimer without any check
Change-Id: Ib9efbb38b8973f688013f3a8f4a94cae1469e390
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
Moved the debug print statement in omap_rproc_ioctl so that rproc will
be valid when derefenced in the log statement
Change-Id: Ib13c98b953d5254d5227a08631cea99a88c29cb1
Signed-off-by: Axel Haslam <axelhaslam@ti.com>
|
|
The dmtimer module now supports the saving and restoration of timer
registers which would otherwise be lost when going to OSWR or OFF.
This patch adapts remoteproc to use omap_dm_timer_stop( ) and
omap_dm_timer_start( ) which handle the save and restore respectively,
instead of omap_dm_timer_disable( ) and omap_dm_timer_enable( ).
Change-Id: Id13a3365a4ca8dca55178fbba90b11c480a6e2a4
Signed-off-by: Paul Hunt <hunt@ti.com>
|
|
This patch provides the inital series of commits for
remote proc on omap
omap:remoteproc-add remoteproc support
Add remote proc support
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Singed-off-by: Angela Stegmaier <angelabaker@ti.com>
omap:remoteproc-provide multi-omap support
This patch resolves the build issues when trying to build with
multi-omap configuration.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
omap:remoteproc-add mechanism to enable bios timer
BIOS requires the GPTimer for handling the OS activities.
This is to configuer the GPTimers that are required for
the remote processors.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
syslink: rproc - add support for START and STOP events
Now apps can wait for remote proc START and STOP events
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
rproc: fix missing common proc args to match userspace
Add the common proc args structure to store the
status of the api in order to match userspace
side.
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
rproc: add protection to rproc_eventdf_ntfy
The list of fd handles most be protected while traversing
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
omap:remoteproc: disable gpt3&4 as bios timer source
Each core sys/app were requesting gpt3/4, respectively,
as bios timer source clock. This patch is disabling the use
of those timers as bios timer source since now the internal
M3 timers are set for this purpose. Anyway the mechanism to
set any timer as bios timer source was kept.
If a gptimer is set as bios timer source changes in Ducati side
are needed.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
omap:remoteproc: enable gpt3&4 for ipu hibernation
gptimer 3/4 are set to be use at the hibernation timers for IPU,
both clocks are enabled and configured in rproc_start and disable
in rproc_stop.
When mpu is hibernating IPU the iommu and mbox context will be
saved and then rproc_sleep will be called to disable each core and
also the associated gptimer needed to allowed retention. Whenever a
message is sent to IPU the ipu_pm_restore will check and wake up
ipu, if needed, restoring the iommu and mbox context and calling
rproc_wake that will enable ipu and the associated gptimer again.
When disable, the hibernating process will only happen if the
system suspend/resume path is sent, IPU self hibernation wont be
executed. Also gpt3/4 won't be requested at all.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
omap: remoteproc - fix sparse warnings in remote proc about static
This patch fixes these warnings:
arch/arm/plat-omap/remoteproc.c:45:6:
warning: symbol 'rproc_eventfd_ntfy' was not declared. Should it be static?
arch/arm/plat-omap/remoteproc.c:180:5:
warning: symbol 'rproc_reg_user_event' was not declared. Should it be static?
arch/arm/plat-omap/remoteproc.c:227:5:
warning: symbol 'rproc_unreg_user_event' was not declared. Should it be static?
Change-Id: Ie822bbf97dc9dbf4b76bff8e115b3e9063037710
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
omap: remoteproc: CONFIG flag for self hibernation
A new CONFIG flag is introduce to control IPU self hibernation
the name of the new flag is CONFIG_SYSLINK_IPU_SELF_HIBERNATION
when disable the timers used by IPU to hibernate and
the flag in sh mem to allow hibernation of IPU are not set and IPU
will hibernate only when sending the system suspend.
On the other hand if the flag is set after N seconds of inactivity
and if there is no resource requested by IPU (using ipu_pm) IPU will
hibernate.
Change-Id: I3f14257930a861c8784d7159a821228111b3ba37
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
|
|
The hwspinlock hwmod is implemented only for omap4 and not
supported for omap3 and omap2.
This patch enable hwspinlocks only for omap4.
Boot warning on omap3:
[ 0.000000] ------------[ cut here ]------------
[ 0.000000] WARNING: at arch/arm/mach-omap2/hwspinlocks.c:61 hwspinlocks_init+0x24/0xbc()
[ 0.000000] Modules linked in:
[ 0.000000] [<c00470ec>] (unwind_backtrace+0x0/0xe4) from [<c0078978>] (warn_slowpath_common+0x4c/0x64)
[ 0.000000] [<c0078978>] (warn_slowpath_common+0x4c/0x64) from [<c00789a8>] (warn_slowpath_null+0x18/0x1c)
[ 0.000000] [<c00789a8>] (warn_slowpath_null+0x18/0x1c) from [<c0017040>] (hwspinlocks_init+0x24/0xbc)
[ 0.000000] [<c0017040>] (hwspinlocks_init+0x24/0xbc) from [<c0041340>] (do_one_initcall+0x58/0x1b4)
[ 0.000000] [<c0041340>] (do_one_initcall+0x58/0x1b4) from [<c0008574>] (kernel_init+0x98/0x150)
[ 0.000000] [<c0008574>] (kernel_init+0x98/0x150) from [<c0042970>] (kernel_thread_exit+0x0/0x8)
[ 0.000000] ---[ end trace 1b75b31a2719ed1c ]--
Signed-off-by: G, Manjunath Kondaiah <manjugk@ti.com>
|
|
Ensure that the hwspinlock driver is registered prior to
I2C driver registration since I2C is dependent on hwspinlock.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
Add pm_runtime_enable to hwspinlock_probe
change pm_runtime_get to pm_runtime_get_sync
Signed-off-by: Balaji T K <balajitk@ti.com>
|
|
Created driver for OMAP hardware spinlock. This driver supports:
- Reserved spinlocks for internal use
- Dynamic allocation of unreserved locks
- Lock, unlock, and trylock functions, with or without disabling irqs/preempt
- Registered as a platform device driver
The device initialization uses hwmod to configure the devices. One device will be created for each IP. It will pass spinlock register offset info to the driver. The device initialization file is:
arch/arm/mach-omap2/hwspinlocks.c
The driver takes in register offset info passed in device initialization. It uses hwmod to obtain the base address of the hardware spinlock module. Then it reads info from the registers. The function hwspinlock_probe() initializes the array of spinlock structures, each containing a spinlock register address calculated from the base address and lock offsets. The device driver file is:
arch/arm/plat-omap/hwspinlock.c
Here's an API summary:
int hwspinlock_lock(struct hwspinlock *);
Attempt to lock a hardware spinlock. If it is busy, the function will
keep trying until it succeeds. This is a blocking function.
int hwspinlock_trylock(struct hwspinlock *);
Attempt to lock a hardware spinlock. If it is busy, the function will
return BUSY. If it succeeds in locking, the function will return
ACQUIRED. This is a non-blocking function.
int hwspinlock_unlock(struct hwspinlock *);
Unlock a hardware spinlock.
struct hwspinlock *hwspinlock_request(void);
Provides for "dynamic allocation" of a hardware spinlock. It returns
the handle to the next available (unallocated) spinlock. If no more
locks are available, it returns NULL.
struct hwspinlock *hwspinlock_request_specific(unsigned int);
Provides for "static allocation" of a specific hardware spinlock. This
allows the system to use a specific spinlock, identified by an ID. If
the ID is invalid or if the desired lock is already allocated, this
will return NULL. Otherwise it returns a spinlock handle.
int hwspinlock_free(struct hwspinlock *);
Frees an allocated hardware spinlock (either reserved or unreserved).
Signed-off-by: Simon Que <sque@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
SYSLINK: ipc - move check for in recovery state per ipc submodule
With this approach, now each submodule can allow which ioctl
can go through when the IPC is in recovery state. It is done
by exporting a new function in ipc.
Change-Id: I3402f2f83cd3dddbdd6a0d8caf94bf061a5b981c
Signed-off-by: Fernando Guzman Lugo <fernando.lugo@ti.com>
SYSLINK: ipc - allow some ioctl go through in recovery state
This api can go through even in recovery state:
messageq_unblock
messageq_delete
messageq_close
messageq_destroy
Change-Id: If738b58c21f1d18c3ef1ba019bb598290096d32b
Signed-off-by: Fernando Guzman Lugo <fernando.lugo@ti.com>
SYSLINK: IPC- provide ipc start and stop notification
Enable the notification of IPC start and stop, which devh can
use to register/unregister for SYS-ERRORS.
This also includes the changes to export ipc_recovery_schedule
function. This function should be called by error handlers such
as sys error and watchdog error.
Change-Id: I436c37b9c4b74d00c771e88c480ea0e96a0268fd
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Wajahat Khan <w-khan@ti.com>
syslink: ipc: fix compilation warnings
Fixing the following compilation warning when
including ipc.h and notify_ducatidriver.h in another
module:
warning: function declaration isn't a prototype
Change-Id: Idbb0b5289a3e8406fbe0b86e70d63d55d7e3060f
Reported-by: Ricardo Neri <ricardo.neri@ti.com>
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
|
|
If the IPU subsystem has hibernated, there is no need to wake it up
and forward suspend notifications. Also, resume notifications will
not be forwarded to the IPU SS if it has hibernated. This is because
self-hibernation only occurs when no resources are in use. In this
case the SysLink PM context save/restore should be all that is needed.
Change-Id: I892e58b569d922038db8fb8604d1432087235f75
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Signed-off-by: Axel Haslam <axelhaslam@ti.com>
|
|
Also has compiler warning fix in AUXCLK request function
syslink: ipu_pm: refactor and cleanup functions
Including:
- ipu_pm module code was a reviewed and reworked to remove
redundant code, simplify and enhance the implementation.
- Change ipu_notifications to receive proc_id.
- Add IOMMU_FAULT notifier callback function.
- Function to release all resources when IOMMU_FAULT event
is received or when detaching, this is called in a
workqueue to avoid calling context problems in the case of
the FAULT event and called directly in the case of detaching.
- Add the possibility to select the SRC_CLK when requesting
a gptimer.
- Add pr_debug traces for debugging purpose:
Previosly all the errors when requesting/releasing
resources were just reported to IPU using notify_event
in the payload. Now kernel is printing errors/warnings/
debug messages and also reporting the error back to IPU.
This is needed because IPU could be dead and not able to
see/print the error ipu_pm is reporting.
Change-Id: I22614bafc7cb7e3070d5bf565be6138fca5a809b
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
syslink: ipu_pm: fix suspend/resume when no IPU image loaded
ipu_pm_drv_suspend and _resume were trying to notify ipu for
suspend/resume without cheking if an image was loaded leading
to an error in the suspend path.
Checking the handle to each proc before sending the notification
is avoiding this.
Suspend/resume is working now with/without an image loaded in IPU.
Change-Id: I55aca1674b1cdffba9fb3247e7da605b4082a521
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
SYSLINK: IPU-PM: clean up warning in AUXCLK request function
clean up clk_disable( ) warning in AUXCLK request function:
ipu_pm_get_aux_clk( ).
Change-Id: I6bcc41e220f80f4e248bef65b3ed0fc379451d98
Signed-off-by: Paul Hunt <hunt@ti.com>
SYSLINK: IPU-PM: restore IOMMU during ipu pm detach and mmu close
The IPU PM during the Ducati hibernation is shutting down the
IOMMU. If the User process that is using IOMMU is killed when the
Ducati is in hibernation state, the board hangs due to access to IOMMU during
resource cleanup.
The issue was observed when the Syslink daemon is killed with Ducati in
hibernation.
Change-Id: Ia5bd05fe382488c33ac858657f05aee9e22a48ee
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
- Call ipu_pm_save_ctx() to truly simulate system suspend, this
is just for testing purpose and is intented to validate IPU
suspend/resume cb's and proper restore of IPU.
Change-Id: I63a957b5b2980e4c4f1f0443fd5ce541b2b3b281
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
|
|
SYSLINK: IPC-replace printks with pr_xxx
Replace
printk(KERN_ERR) with pr_err
printk(KERN_INFO) with pr_info
printk(KERN_WARNING) with pr_info
Change-Id: Ic434b3ca34f131528c71f8f3fc6b5b51252db92c
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
SYSLINK: notify-replace printks with pr_xxx
Replace
printk(KERN_ERR) with pr_err
printk(KERN_INFO) with pr_info
printk(KERN_WARNING) with pr_info
Change-Id: Iad946fec668a2fa190ea4c54067f59378cb6f147
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
SYSLINK: notify-ducati-replace printks with pr_xxx
Replace
printk(KERN_ERR) with pr_err
printk(KERN_INFO) with pr_info
printk(KERN_WARNING) with pr_info
Change-Id: Ic65782b709447033912a7b3fe76daa3548b1ad4c
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
SYSLINK: PROCMGR-replace printks with pr_xxx
Replace
printk(KERN_ERR) with pr_err
printk(KERN_INFO) with pr_info
printk(KERN_WARNING) with pr_info
Change-Id: Iee8c0a38dbb32dc33b42d4ba757a746f791a3047
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
|
|
Added a new command in messageq to unblock. A new variable
is also added to the messageq object to unblock the messageq.
The purpose of unblock is to unblock the messageq_get() when
it is pending for a new message with MessageQ_FOREVER timeout. The
messageq_unblock is intended for use before deletion of messageq
and it unblocks messageq_get() with MessageQ_FOREVER timeout value
for termination.
Signed-off-by: Shahid Akhtar <sakhtar@ti.com>
|
|
ipu_dev header to mach-omap2 tree
Header file needed fot mach-omap2/ipu_dev.c file
Signed-off-by: Paul Hunt <hunt@ti.com>
omap:ipu_pm: add ipu_utility to mach-omap2 tree
IVAHD Sequencer WFI Boot Code for IPU PM Driver
Signed-off-by: Paul Hunt <hunt@ti.com>
omap:ipu_pm: add ipu_drv to mach-omap2 tree
Add needed api for ipu_pm module
Signed-off-by: Paul Hunt <hunt@ti.com>
omap:ipu_pm: add ipu_dev to mach-omap2 tree.
Platform Device handling for IPU PM Driver
Includes:
- ipu_pm_module_start
- ipu_pm_module_stop
- platform data definition for the following devices:
+ fdif
+ ipu
+ ipu_c0
+ ipu_c1
+ iss
+ iva
+ iva_seq0
+ iva_seq1
+ L3
+ mpu
-ipussdev_init to init all devices.
Signed-off-by: Paul Hunt <hunt@ti.com>
syslink:ipu_pm: add/remove ipu_pm_start/stop
Add the calls to api that will enable:
- iva_hd
- iva_seq0
- iva_seq1
- fdif
- iss
modules using the hwmod implementation.
Hibernation disabled.
Gptimer3 used for hibernation was disabled.
Setting CM_MPU_M3_CLKSTCTRL.CLKTRCTRL[1:0] = HW_AUTO
Retention is disable by default.
Signed-off-by: Paul Hunt <hunt@ti.com>
syslink:ipu_pm: fix req/rel aux_clk/iss/iva/idle
The current implementation was not meeting all
the requiriments needed to properly enable and
use aux_clk.
Anyway this is a hack way to do it, the proper
implementation is still pending.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Temporary path using cmm_write_reg to enable CAM_PHY
Signed-off-by: Paul Hunt <hunt@ti.com>
Ivahd requesting was not fully working since
sl2 was not being requested and that is needed
for ivahd to be fully functional.
In ipu_dev sl2 is initialized
In ipu_pm the flow to request iva should be:
- Requesting (ivaseq1 call is requesting sl2)
ivahd>ivaseq0>(ivaseq1>sl2)
- Releasing (iva call is first releasing sl2)
ivaseq0>ivaseq1>(sl2>ivahd)
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Setting IdleAllowed flag that enables the idle mode
in Ducati. WFI can be called in Ducati flag
whenever this flag is set
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Fix a compile error related to iva/iss when trying
to build with ES1.0 configuration. In ES1.0, the
PM is not supported.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
syslink:ipu_pm: DFVS support for freq
Fix bug in error message string referencing
Signed-off-by: Paul Hunt <hunt@ti.com>
Add platform data for DSP resource and fix
attributes for MPU and L3 interconnect.
Signed-off-by: Paul Hunt <hunt@ti.com>
Provide internal APIs for performance and
latency constraint frameworks to use to assert
these constraints into the system power management.
Signed-off-by: Paul Hunt <hunt@ti.com>
MPU and CORE freq/lat cstrs
Add special handling for MPU and CORE frequency and
latency constraints.
To specify a constraint to MPU or CORE the api needs
to be called passing the following to the api:
- IPUPM_SELF
- IPUPM_MPU
- IPUPM_CORE
Signed-off-by: Paul Hunt <hunt@ti.com>
DVFS support in ipu_pm
Calling the dvfs apis per resource.
Included:
- ipu[perf|lat]
- iss[perf|lat]
- ivahd[perf|lat]
- L3 bus[lat]
- mpu[perf|lat]
Pending:
- fdif
- dsp
Note: Latency calls are working but hasnt been tested.
Perf/rate calls are working.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
(cherry picked from commit 3c1cdb45bb67d11eda14420a7b8eaacff0c24173)
SYSLINK:IPU-PM-move iommu handles to attach and detach
Move getting iommu handle to attach and detach from setup/destroy.
Having the iommu get in setup is causing the mmu fault recovery fail
since the iommu object gets incremented before the fault application
closes the iommu handle which leads to failure in shutting down the
iommu device completely.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
syslink:ipu_pm: add L3 bw and fdif cstr
ipu_pm api to set the bandwidth constraint for L3 bus
New api is:
ipu_pm_module_set_bandwidth(rsrc,
target_rsrc,
bw);
bw is received in KiB/s and converted into Hz
Add support to set performance and latency cstrs to
fdif.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
syslink:ipu_pm: send pid_death just to appm3
ipu_pm_notification is broadcasting all events to both
sys/app processors whenever an event is sent using that
api.
For PID_DEATH event the requirement is only to send the
message to APPM3 since is the one that will be executing
the resource manager; also because of that sending the
message to SYSM3 is unnecessary.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
syslink:ipu_pm: allow IPU hibernation
-IPU self hibernation-
After 5 seconds of inactivity ipu will send a notification to
MPU indicating it can be turned off, mpu can proceed to turn off
IPU just if there is no message (notify_send_event) going to IPU,
in that case hibernation will be aborted. IPU is prepared to abort
hibernation when a message arrives to its mbox. When mpu is
hibernating IPU the iommu and mbox context will be saved and then
rproc_sleep will be called to disable each core and also the
associated gptimer, that is needed to allow retention. Whenever a
message is sent to IPU the ipu_pm_restore will check and wake up IPU,
if needed, restoring the iommu and mbox context and calling rproc_wake
that will enable IPU and the associated gptimer again.
-MPU hibernating IPU-
Whenever the system suspend is sent to ipu_drv the suspend_drv will
send a SUSPEND notification to IPU, it will send back an ack to MPU
and then start the hibernation proccess by it self following the same
scheme mentioned above but without waiting the 5 seconds, once
ipu_drv receives the suspend ack it will call ipu_pm_save_ctx() in
order to wait for IPU to be hibernated and in really idle then turn IPU
off. This will work even if self hibernation in IPU is disable.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
IPU PM: adapt aux clk funcs to new framework entities
Signed-off-by: Paul Hunt <hunt@ti.com>
IPU PM: temporary workaround for missing MM AUX_CLK requests
Signed-off-by: Paul Hunt <hunt@ti.com>
syslink: ipu_pm: add support for all the auxclk
Previously ipu_pm was only supporting the request of
AUX_CLK_1/3 now all the available aux_clk 0-5 are
supported and can be requested/released via ipu_pm.
Also the base address of the AUX_CLK was fixed since
it was assuming AUX_CLK_0 as AUX_CLK_1 leading to a
misalignment when requesting.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
syslink: ipu_pm: add a core latency constraint when the ipu is running.
Due to the enabling of C4 and C5 states IPU was being
put in reset without allowing it to properly save the
context leading to an invalid resume operation and an
MMU fault by IPU side.
Putting a latency constraint on IPU_CORE whenever IPU
is up and running is avoiding this scenario.
The constraint is released once the IPU has properly
saved the context and requested again when resuming
to avoid letting the ipu_core going to retention.
Change-Id: Iced64ff0744b2d6b0a0ecfc1952ed26e9e560278
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
omap:syslink: removing gpt4 as available
Gptimers 3 & 4 are reserving for ipu_pm internal use
gpt3 was removed in previous patch from the list of
available gptimers that IPU can request but gpt4 was
still in the list.
This patch removes gpt4 from the list.
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
|
|
This patch adds devh to syslink tree
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
SYSLINK: Devh - Add device event handling and SYSERROR support for OMAP4
SYSLINK: Devh - Add event handling and SYSERROR support
DSP exceptions that are unrecoverable need to be communicated
to the host processor so that it can print debug information,
do resource cleanup and ultimately reload DSP. The notification
mechanism for sending events to the host processor has been
consolidated in a new module named Device Error Handler on the
remote processor.
The notification mechanism relies on a NotifyShm event to know
about error events on the remote cores. The support has been
built into the existing Devh module which registers a call
back for the previously agreed event id to be used for error
events.
Upon receving an error event, the module will notify other
kernel modules which have registered with it. It will also
notify userspace applicatios through an agreed fd event. This
support will be built in a following patch.
For now, the module uses event id 4 to receive notify messages
from remote processor. Enhacements that will follow include using
a reserved event for sending notify messages.
Initialize the Device Error Handler functionality in Devh. Devh
registers a callback for NotifyShm event to be triggered when
an error occurs on the remote core.
Devh module has the support for receiving notifications from remote cores about
irrecoverble errors. Userspace processes need to be notified for such events
as well in order to do cleanup and recovery. Besides, such notifications can
also be used to output valuable debug information about error scenarios.
[Suman A] Corrected err_event_id from 4 to use as
err_event_id = (4 | (NOTIFY_SYSTEMKEY << 16)),
Change-Id: I961d83e62e5ce21b32f5e2c461644e6f2bd8cb3c
Signed-off-by: Wajahat Khan <w-khan@ti.com>
Signed-off-by: Hari Kanigeri <h-kangieri2@ti.com>
SYSLINK: devh - add possibility of register for specific events
New paratemer is added to event register ioctl to allow users
register for a specific event.
Modified to register notifier for IPU_PM watch-dog timer
[Hari K] Added the call to ipc_recover_schedule on sys error and
watchdog timer
Change-Id: I5fcfc28bc424a47f95c32104d2200ad946ef9027
Signed-off-by: Fernando Guzman Lugo <fernando.lugo@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
SYSLINK: Devh - Add userspace events unregistration support
Implement the unregister ioctl which lets an application unregister
an fd it had registered with the module earlier. The call returns
successfully even if no matching fd was present or duplicates were
present.
Also fixed traces because of addition of watch dog event.
Change-Id: Idd075c6a96a5c5438e02cfc1efa19d3e3859c0af
Signed-off-by: Wajahat Khan <w-khan@ti.com>
|
|
Add ipu pm to syslink tree
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Add syslink headers
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Subramaniam C.A <subramaniam.ca@ti.com>
Signed-off-by: Arun M G <arunmg@ti.com>
Signed-off-by: Ramesh Gupta G <grgupta@ti.com>
Signed-off-by: Jayan John <x00jayan@ti.com>
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
Signed-off-by: Arun Radhakrishnan <x0051460@ti.com>
Signed-off-by: Subin Gangadharan <subin.kg@ti.com>
|
|
Add multicore ipc module to syslink.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Subramaniam C.A <subramaniam.ca@ti.com>
Signed-off-by: Arun M G <arunmg@ti.com>
Signed-off-by: Ramesh Gupta G <grgupta@ti.com>
Signed-off-by: Jayan John <x00jayan@ti.com>
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Arun Radhakrishnan <x0051460@ti.com>
Signed-off-by: Subin Gangadharan <subin.kg@ti.com>
|
|
SYSLINK:notify-add notify module to syslink
add notify module to syslink
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Ramesh Gupta G <grgupta@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Subramaniam C.A <subramaniam.ca@ti.com>
SYSLINK:notify-added notify ducati driver to syslink
Add notify ducati driver to syslink
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Subramaniam C.A <subramaniam.ca@ti.com>
Signed-off-by: Ramesh Gupta G <grgupta@ti.com>
SYSLINK: notify - minor fix for error check
copy_to_user returns a non-zero value in case of error.
This patch fixes it and resolved an associated KW defect:
Comparison of unsigned value against 0 is always false
Signed-off-by: Suman Anna <s-anna@ti.com>
omap:notify: call restore_ctx when send_event first time
In notify.c when sending an event using notify_send_event()
the funtion ipu_pm_restore_ctx() is called to enable just
the first time idle and disable hibernation.
Also is changing the ducati clock state transition control
to HW_auto to allowed retention.
The rest of the restore ctx remains under a menuconfig macro.
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
SYSLINK: notify - reset notify registration chart correctly
This patch fixes a minor bug in the notify driver unregistration process.
The notify registration chart is an array and stores all the registered
events in ascending order (dictates the notify event priorities). This
array is compacted whenever an event is unregistered. The compaction is
done shifting all the subsequent events to the left. Currently, the next
element to the element being unregistered is not reset correctly, thereby
leading to duplicate registrations in the array. This still leaves a valid
duplicate event when that event is unregistered. This scenario happens
when a lower event is unregistered while a higher event is also present.
Change-Id: I47a04e4d01a5081099117280be286a51afa7d1cf
Signed-off-by: Suman Anna <s-anna@ti.com>
syslink: notifier - add a mutext per notify_object
Critital section are per notify_object we not need
to block other notify_object of other processors.
Instead create a new mutex per notify_object to
protect critical sections which belong to a
specific processor. This would increase the performance.
Also fix unprotected part while calling the registered
callback fucntions that was causing a kernel panic.
Change-Id: Ie99eef8eb5eca6a75d203df33c2a31981ce50489
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
SYSLINK:IPC - Notify reserve events moved from 4 to 5
Making room for new reserved event for devh. Devh is using
event 4 for sys error notifications.
Change-Id: I545b3da56167d6f539092ab434d33f993129324f
Signed-off-by: Shahid Akhtar <sakhtar@ti.com>
|
|
Add ProcMgr to Syslink
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Ramesh Gupta G <grgupta@ti.com>
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Arun M G <arunmg@ti.com>
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
As pointed by Ohad Ben-Cohen, the variable rq_full flag is a
global variable, so if there are multiple mailbox users
there will be conflics. Now there is a full flag per
mailbox queue.
Version 2:
- Rebase to the latest.
Reported-by: Ohad Ben-Cohen <ohad@wizery.com>
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
|
|
- Fix an issue related with the offsets of the registers
- Each user (Ducati, Tesla ...) should not have its own ctx
memory since the mbox module is one for all of them. A share
ctx is created to store the needed registers when saving.
- save_context function is saving just the irqs per user and
calling the mbox_shutdown.
- restore_context function is calling the mbox_startup and
restoring just the irqs per user. mbox-startup function is
setting all the rest of the register to the original state.
There is no need to save the Messages.
Currently ipu_pm is not taking in to consideration Tesla for
disabling or not mbox.
This patch depends on the following list of patches provided
by Hari Kanigeri:
- omap:change spinlock to mutexlock
- omap:mailbox-add notification support for multiple readers
- SYSLINK:notify-adapt to new mailbox interface
Signed-off-by: Miguel Vadillo <vadillo@ti.com>
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Fernando Guzman Lugo <x0095840@ti.com>
Signed-off-by: Juan Gutierrez <jgutierrez@ti.com>
Signed-off-by: Paul Hunt <hunt@ti.com>
|
|
The ISR is handling only the mailbox instance that was registered last.
So if both mailbox instances are running at the same time, the first mailbox
that registered wouldn't get the mailbox message. The same issue is present
in Transmit Interrupt case too. Only the last registered mailbox is handled.
The fix is to iterate through the list of mailbox that were registered.
Signed-off-by: Hari Kanigeri <h-kanigeri2@ti.com>
Signed-off-by: Ramesh Gupta <grgupta@ti.com>
Signed-off-by: Subramaniam C.A <subramaniam.ca@ti.com>
|
|
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
|
