We see spurious interrupts during modem standby if ERDB is enabled.
Making changes to not enable ERDB interrupts.
Change-Id: I5a51101b54f67e7c8e56d46c56bddd739c8852a5
Signed-off-by: Subramanian Ananthanarayanan <skananth@codeaurora.org>
Signed-off-by: Gauri Joshi <gaurjosh@codeaurora.org>
During mhi_dev_restore_mmio() we call mhi_dev_mmio_mask_interrupts,
restore the backup registers and we enable only CTRL,CHDB interrupts.
This causes the CMDB do be in disabled state, so after first D3 cold
transition we don't get CMDB interrupt until it fires along with one of
the already enabled interrupts. This causes delay in processing CMDB
interrupts that host might have queued at a very early stage.
Change is to enable CMDB interrupts post mmio restore along with the
other interrupts.
Change-Id: I75acbeda3f6e7c50669be4db9190b8ccd5b1dd1e
Signed-off-by: Subramanian Ananthanarayanan <skananth@codeaurora.org>
Signed-off-by: Gauri Joshi <gaurjosh@codeaurora.org>
mhi_dev_backup_mmio() and mhi_dev_restore_mmio() does backup/restore
of INTR registers of MHI device, causing it to restore interrupts
that are not valid during PCIE PM_RST_DEAST event. The fix is to
skip the INTR register space.
Change-Id: I2a8cde174cf4c6acd7acce13ad0195ed6a9a6f0b
Signed-off-by: Subramanian Ananthanarayanan <skananth@codeaurora.org>
Signed-off-by: Gauri Joshi <gaurjosh@codeaurora.org>
In hibernate scenario, due to recent changes on host side the
BHI register value is updated before M0 is issued. This causes a
mismatch with the reset value 0xfffffff expectations. As a result
device ends up sending an MSI to indicate ready status.
This is not expected from host side, and causes race conditions
where it leads the host to process MHI initialization and MHI reset
concurrently in two thread contexts. This manifest into two different
signatures primarily:
a) M0 is acknowledged in Thread 1 context, host keep waiting for device to
Acknowledged M0 sent as part of Thread 2. Device ignores the 2nd request
as its already in M0.
b) MMIO event ring is initialized with a primary value x, because of
Thread 2 doing a re-installation the event ring base is updated by the
host to another value y. The device which is only aware of the primary
value send completion event with RP value corresponding to x. This is
treated as a junk RP and ends up resulting in a host crash with
BSOD 0xdeaddead.
Change-Id: Iae0ee996ad086ff5e6430c198f29e56d994c1ef7
Signed-off-by: Subramanian Ananthanarayanan <skananth@codeaurora.org>
Signed-off-by: Gauri Joshi <gaurjosh@codeaurora.org>
num_reqs feild as part of chan_attr is not updated properly, it is set
to true. So there is only 1 reqs allocated for this channel. This causes
the channel to run of reqs easily causing ADB to stall. Updated num_reqs
to 50.
Change-Id: I1d4ec246c5879dbf1e9ca38ceeec7ffd6b8feccb
Signed-off-by: Subramanian Ananthanarayanan <skananth@codeaurora.org>
Signed-off-by: Gauri Joshi <gaurjosh@codeaurora.org>
ring_cache is NULL as channel has not yet been started, read operation
accesses ring_cache before checking if channel is in stopped state.
Adding a check in UCI layer to perform read/write transfers only
if the channel is in valid state.
Change-Id: Ic8e6ae19badf43a7f64f92349e9c2427750a98b6
Signed-off-by: Subramanian Ananthanarayanan <skananth@codeaurora.org>
Signed-off-by: Gauri Joshi <gaurjosh@codeaurora.org>
Invalid memory access can happen if a client closes its channels
with transfers pending because the mhi transfer request structure
gets freed in UCI when channel is closed, but MHI driver accesses the
request structure in the transfer completion callback.
To avoid the invalid memory access, following changes are done:
* Allocate mhi transfer request structures for a client in the
first UCI client open call
* Do not free mhi transfer request structs when client calls release,
to allow for subsequent re-use
* Maintain an in-use list to track active transfers
* If a client exits with transfers pending, move the request structs
corresponding to pending transfers to free list and mark them as
stale.
* Stale transfers, if completed later, are ignored.
Change-Id: I170e06321de9c67b0a1d107b775d75b2ed97febb
Signed-off-by: Siva Kumar Akkireddi <sivaa@codeaurora.org>
Signed-off-by: Gauri Joshi <gaurjosh@codeaurora.org>
Add sched_sync_hint_enable not to enable/disable sync.
Change-Id: Id6bdff3d4b9ea1051d7b55ca4cfd1d802c92fa64
Signed-off-by: Biao Long <blong@codeaurora.org>
WLAN host driver has a new memory requirement for regulatory structure
allocation of order 5. Increase 128 kB slot by one.
Change-Id: If95291c96df0b7588b54f8e19b287d846896b479
Signed-off-by: Mahesh Kumar Kalikot Veetil <mkalikot@codeaurora.org>
Migrate isolate of a pageblock is being checked only at the pageblock
boundary pages thus the isolation of freepages should happen on that
pageblock which otherwise can still cause the isolation of freepages on
an isolated pageblock.
Consider the below case:
Pa Pb
check the pageblock
isolation
zone_lock
Isolate the SWAP_CLUSTER_MAX pages
Zone_unlock
zone_lock
Set Migrate isolate on the
same page block that Pa is
operating on.
zone_unlock
zone_lock
Continue the isolate on the same
page block that was set Migrate
isolate by Pb which is an error.
Change-Id: I1504b031f434fb65ebb11e1472b73e9d34e388b8
Fixes: ff56a97530 ("mem-offline: disallow page migration to isolated page blocks")
Signed-off-by: Charan Teja Reddy <charante@codeaurora.org>
Currently, qcom_icc_bcm_init always assumes the bcm is not initialized.
There is a possibility that same bcm can belong to multiple providers
and re-initing the bcm adds the same bcm multiple times to the nodes
which are under the control of the bcm.
Hence add condition to check for bcm inited.
Change-Id: I3961963a6ef1a7b76d7ab708f648146fe1a30c99
Signed-off-by: Veera Vegivada <vvegivad@codeaurora.org>
Define a vdd_levels enum and vdd_corner[] mapping table which can
be used by clock controller drivers to configure the vdd_class
associated with their clocks.
Change-Id: I7b1ad5f51c414694bc9d5101afc411b1d63bad42
Signed-off-by: Veera Vegivada <vvegivad@codeaurora.org>
To avoid any suspend during probe, we increment the device
usage counter before calling ufshcd_async_scan and decrement
before returning from ufshcd_async_scan function.
For the scenarios where ufs device initialisation fails
or if the ufs device is not present, we are decrementing
device usage counter before calling ufshcd_hba_exit.
Hence there is a possiblity that after we disable the clocks
as part of ufshcd_hba_exit call, runtime suspend may get
triggered. As part of runtime suspend, we are accessing
the register causing noc error due to unclocked access.
To fix this, we are removing pm_runtime_put_sync
from ufshcd_add_lus() and adding it in common place
at the end of ufshcd_async_scan so that we hold auto
suspend until async scan completes in both success and
failure case.
Call Stack:
panic+0x17c
panic+0x0
arm64_serror_panic+0x74
do_serror+0x15c
el1_error+0xbc
el1_irq+0xb0
crypto_qti_resume+0x68
ufshcd_crypto_qti_resume+0x18
ufshcd_crypto_resume+0x3c
ufshcd_suspend+0x5b0
ufshcd_runtime_suspend+0x44
ufshcd_pltfrm_runtime_suspend+0x14
pm_generic_runtime_suspend+0x44
__rpm_callback+0x98
rpm_suspend+0x2a4
rpm_idle+0x158
schedule+0x8c
schedule_preempt_disabled+0x24
__mutex_lock+0x358
__mutex_lock_slowpath+0x14
mutex_lock+0x34
clk_unprepare+0x58
__ufshcd_setup_clocks+0x140
ufshcd_hba_exit+0x170
ufshcd_async_scan$d36fba0b664edbd124b5d17d005db4c5+0x118
async_run_entry_fn$8847b5d726ee295ba72a029a76bec2f5+0x60
process_one_work+0x2cc
worker_thread$f31e2447a3fdcb60f4b193f95acd647c+0x278
kthread$bf1537fa370ac32bde5f056194ddafc6+0x16c
ret_from_fork+0x10.
Change-Id: I059de2b0eba642bf1a35c27e8e12d31c6f30a836
Signed-off-by: Nitin Rawat <nitirawa@codeaurora.org>
