Use the dev_iommu_fwspec_get() accessor, instead of directly
accessing the device structure for the IOMMU fwspec, as well
as remove an unnecessary forward declaration.
Change-Id: Id09202bd468bddbf8bdf194513fda34b59756fba
Signed-off-by: Isaac J. Manjarres <[email protected]>
When the ARM SMMU module is unloaded, the exit routine does not
fully clean-up after the initialization routine, as it leaves
the TBU driver registered. Thus, unregister the TBU driver
when unloading the SMMU module.
Change-Id: I536f756dab7eebeffe0dd907629feed6657a2532
Signed-off-by: Isaac J. Manjarres <[email protected]>
This is a snapshot of the crypto drivers as of msm-4.19
commit 52d84be6ae6e151ba ("crypto: msm: qcrypto: Fix null
pointer dereference error in qcrypto driver")
with changes to comply with msm-5.4 format requirement.
Change-Id: Ic566a15e41511d229a9e476c03962753da583946
Signed-off-by: Zhen Kong <[email protected]>
The remove() callback for the ARM SMMU driver should ensure
the SMMU devices that are being registered are no longer
visible to the IOMMU framework, so remove the devices
by invoking iommu_device_unregister().
Change-Id: I87dd7b4619e5b26b7eed64b04e3716f3b9c417c5
Signed-off-by: Isaac J. Manjarres <[email protected]>
ipa_disable_apps_wan_cons_deaggr() has different declarations when
CONFIG_IPA or CONFIG_IPA3 are not set. Thus fix the declaration to
align with CONFIG_IPA or CONFIG_IPA3 enabled one.
Change-Id: Ic09cac2484c9cfbcd0833583c05c6fc19a3c2aed
Signed-off-by: Jia Ding <[email protected]>
This reverts commit 75afa532e7.
Too many things rely on cmd-db to be built in to allow it to
be built as a module.
Change-Id: I317ba89a7d3c9f88832e73834b65bcccac79bbe1
Signed-off-by: Lina Iyer <[email protected]>
The context bank 0 offset and size of each context bank register space
was hardcoded. The context bank 0 offset changed in A660. We now compute
these values from GPU_GFX_SMMU_IDR1 register instead of using hardcoded
values.
Change-Id: If9125dfe823f92e9c61ff1635ae07a13ec6f000b
Signed-off-by: Raghu Ananya Arabolu <[email protected]>
There are only two IOMMU contexts that we care about and there is no
need to keep an array and all the associated detritus that comes with it.
Change-Id: Ic0dedbad658c9e8ef58e676b0294fcc142b544c7
Signed-off-by: Jordan Crouse <[email protected]>
Currently the few global secure memory objects are allocated in a side
band fashion. Add support to kgsl_allocate_secure_global() to allow
secure objects.
Change-Id: Ic0dedbadc03fb462a8b8bb82a2141f1d5ddd9762
Signed-off-by: Jordan Crouse <[email protected]>
Currently all global buffers are mapped when pagetables are created which
means that all global buffers need to be in place before the default
pagetable is created otherwise they will not be available in that context.
Add a new MMU function to map each global buffer. If the default pagetable
has been created then the new buffer can be mapped immediately, otherwise
all the buffers on the list will be mapped when the pagetable finally comes
along. Per-process pagetables will map everything from the list at create
time as they do now.
As a caveat this function only applies to the default pagetable. We will
not go back and retroactively map buffers into existing per-process
pagetables. Previously if somebody mapped a global buffer late it
wouldn't be available in the default pagetable either so it was easier to
debug. Now that this function will always successfully map buffers in the
default pagetable "late" buffers that got missed in a per-process table
will be harder to debug. In recognition of this the new map function will
trigger a warning if a buffer is mapped after the first per-process
pagetable is created. It won't stop breakage from happening but at least
it will be more easily debuggable.
Change-Id: Ic0dedbad2ed20769ea96bfdac3b707a30c6f9737
Signed-off-by: Jordan Crouse <[email protected]>
The ability to vote IPA clock from user space is now required. Given
this, voting for IPA clock can now be done via an ioctl. These changes
implement the ioctl.
Change-Id: Id7e02678e96da98f2a17f35e7f6736fba39ca990
CRs-Fixed: 2592003
Signed-off-by: Perry Randise <[email protected]>
The ability to vote IPA clock from user space is now required. Given
this, voting for IPA clock can now be done via an ioctl. This header
file supports the ioctl and can be independently propagated.
Change-Id: I742a7950f9fc53f17cb6f210f2b7f06ade842011
CRs-Fixed: 2592003
Signed-off-by: Perry Randise <[email protected]>
possibility of OOB access on device->response_pkt in
__response_handler. for e.x if msg queue contains 1000
messages and all 1000 were read and queue is empty.
So __get_q_size api will return zero and _iface_msgq_read
will go in an infinite loop, even if packet_count
== max_packets.
Change-Id: I8e64ff70d5bcefb52d66bfea121f3febf4728384
Signed-off-by: suchawla <[email protected]>
Add usb streaming support to collect trace data externally.
Add support to enable USB mode for coresight devices.
Set memory as the default mode on ETR.
Change-Id: I3bd2bd92c676ff79a3ae5a045a54e7d1e867543d
Signed-off-by: Mulu He <[email protected]>
Signed-off-by: Tingwei Zhang <[email protected]>
Add support for PMIC7 ADC_TM peripheral in ADC_TM common driver files.
Add main ADC_TM driver file for PMIC7. As register offsets and SW flows
have changed significantly for ADC_TM due to ADC architecture change,
this new driver is required for ADC_TM on PMIC7.
Change-Id: I9c92ab0a93a1bf487cb9fa0e9640b3c4eaf2b21a
Signed-off-by: Jishnu Prakash <[email protected]>
ADC_TM peripheral is used by client to set thresholds on channels
supported by VADC and send notification once set thresholds are
crossed. The driver registers thermistors used for thermal
mitigation with of_thermal to allow thermal driver to set
thresholds and receive notification through thermal sysfs.
This is a snapshot of the driver taken from msm-4.19 as of commit
f94ba33e3b90 ("Merge "usb: dwc3-msm: Fix GDSC collapse in host mode
bus suspend").
Change-Id: I9211552317955c503b7287876e11cd161c37f706
Signed-off-by: Jishnu Prakash <[email protected]>
KW raise the issue of "array 'dbg_buff' of size 4096
may use index value(s) 4096." The fix is to increase the
dbg_buff size to avoid it.
Change-Id: Ibb02e7fa73d9c0fb8924c90c66052255acc8275f
Signed-off-by: Bojun Pan <[email protected]>
On uC debug stats, whenever uC provides the invalid protocol id,
we have to do the iounmap to prevent the leak.
Change-Id: I1360a04c5013657b5c9a051ee11d1a6cb5556915
Signed-off-by: Bojun Pan <[email protected]>
Enable all of the configuration options related to memory
hotplug for the Lahaina target.
Change-Id: I1f39c5ee7ecdca18746f09fb2ee58b4c2de63602
Signed-off-by: Isaac J. Manjarres <[email protected]>
After a memory block is offlined, the page-table entries should be removed
in order to avoid any CPU speculative accesses happening on this region.
Clear off the entries at the leaf levels (PUD or PMD) on a block mapped
memory region. This only zeroes the entries and does not free the page
table memory. During online, we create mappings again for these regions.
Change-Id: I71091c33589645567c60457ba569dd393ddcab59
Signed-off-by: Sudarshan Rajagopalan <[email protected]>
[[email protected]: resolve merge conflicts]
Signed-off-by: Isaac J. Manjarres <[email protected]>
Currently we use a percentage property value to determine the amount of
offlinable memory in the system.
However a percentage property doesn't support accurately configuring the
offlinable region sizes on targets with different amounts of DDR.
For example it is not possible to configure the system such that 4GB
targets won't have any offlinable region, 6GB targets will have a 1GB
offlinable region and 8GB targets having a 3GB offlinable region without
creating a separate DT files for each target.
Many customers want to be able to support different DDR sizes without
creating separate DT files for each target.
Replace the mem-percent property with an offline-sizes property which
contains an array of offlinable memory region sizes to apply to targets
based on their DDR size.
Change-Id: I93210beec153ca0179c0087ce98afd4dfc22ea02
Signed-off-by: Liam Mark <[email protected]>
Signed-off-by: Isaac J. Manjarres <[email protected]>
Update the physical end address of available RAM in calculation
used for randomizing the memory base address base.
Change-Id: Id32260d5c15ec4363ab62755238a10cc4dfaabe8
Signed-off-by: Sudarshan Rajagopalan <[email protected]>
Signed-off-by: Isaac J. Manjarres <[email protected]>
The system RAM region would not be ideally contiguous from
start to end of DRAM. Detect any offset/holes present in the
DDR region and update the memory limit being set, so that the
DRAM end address is aligned to the memory block size. This is
a requirement for mem-offline framework to work properly.
Change-Id: I97cb0cce70e7c414ae2d200a4a3f714d145c0e64
Signed-off-by: Sudarshan Rajagopalan <[email protected]>
Signed-off-by: Isaac J. Manjarres <[email protected]>
Add a new sysfs node called allocated_bytes to each memory block which
reports the amount of memory currently allocated from that block.
Userspace can then use the allocated_byes node to better evaluate the
potential cost of offlining an individual block since all the allocated
memory in the block will need to be migrated when the block is offlined.
Change-Id: I520de2dd56e4562a65c48e38edb1236a8cbce5fd
Signed-off-by: Liam Mark <[email protected]>
Signed-off-by: Swathi Sridhar <[email protected]>
[[email protected]: resolve merge conflicts]
Signed-off-by: Isaac J. Manjarres <[email protected]>
remove_pagetable() should only be called when CONFIG_MEMORY_HOTPLUG
is enabled to clean up the page table entries created when memory
was hot-added.
Change-Id: I38cf42673a7637dac62d13a3083b0963c8a435b3
Signed-off-by: Isaac J. Manjarres <[email protected]>
Before killing a process first try to see if there are any offlined
memory blocks, if there are try to online one of them.
Change-Id: Ie97c8b69f0ba173c202a891a38a5c914869ddaae
Signed-off-by: Liam Mark <[email protected]>
[[email protected]: Update walk_memory_range to walk_memory_blocks]
Signed-off-by: Isaac J. Manjarres <[email protected]>
