FSP0 is a boot frequency and is not supported to be switched at
runtime.  Overwirte all FSP0 requests to be FSP1.

Bug: 133340008
Change-Id: I0e66d241ff5ca99fe9a0afcc2e1000512fb2517b
Signed-off-by: Cheng Gu <gucheng@google.com>

Provide a DDR memory location pointer via an extra argument
to the SBL FSP switch routine. This memory location is used before
and after the switch to test the validity of DDR read/write
operations.

Change-Id: I79997749a903484138ad192469198c4484467ec4
Signed-off-by: tsrytkon <teemu.s.rytkonen@intel.com>
(cherry picked from commit efdff6a1)

The refresh worker thread toggles auto low pwr on and off
periodically.

In rare cases a hang was observed manifesting as an AXI
interrupt while the memory system is in its deepest
sleep state - see SCU registers DBG_STATUS and IRQ status.

Based on register dumps, analysis pointed to a hang in
mnh_ddr_disable_lp upon sending the exit lp command.
Following up on this theory, the entry/exit from LP mode
was scrutinized and this patch is the result of minimizing
register accesses with respect said LP mode.

Disable/enable lp method as per cadence recommendation

Bug: 111703181
Change-Id: Iddd4733cbd8a119ec26bbef7e5c381269d727fc3
Signed-off-by: Anthony Loeppert <anthony.loeppert@intel.com>
(cherry picked from commit 765eccf3)

Bug: 111703181
Change-Id: I14d9380d7ffbe1b896d9bbf2c39e3ea6233e5f0f
Signed-off-by: tsrytkon <teemu.s.rytkonen@intel.com>
(cherry picked from commit aa8bfe35)

Bug: 111703181
Change-Id: I6cb86a3534778a478b325828370e419f085ec19f
Signed-off-by: Anthony Loeppert <anthony.loeppert@intel.com>
(cherry picked from commit 904c083f)

Need to use disable_irq_nosync vs disable_irq
to avoid a deadlock scenario.

Bug: 111642220
Change-Id: Ia15c3a44341cff2b1341897f91ab5a7baa6b0c5b
Signed-off-by: Cheng Gu <gucheng@google.com>
(cherry picked from commit 7def6e6b)

Force refresh worker execution upon resuming to
make sure worker thread doesn't get starved
with back to back ddr frequency switching.

Bug: 111703181
Change-Id: Icfe67a229ca23c072c39928d26271467211ecb30
Signed-off-by: Anthony Loeppert <anthony.loeppert@intel.com>
(cherry picked from commit c299c29e)

-Add an example sbl fsp sysfs to the mnh clk, which just invokes
get/set methods of SBL via SMC.

Bug: 111703181
Change-Id: I2925aa709c7e4c11bf4033df470a2d7b87ea331f
Signed-off-by: tsrytkon <teemu.s.rytkonen@intel.com>
(cherry picked from commit 8e6d1291)

If temperature of the two DRAM dies don't match, the driver will
use a refresh rate suitable for the higher temperature die.  This
is not an error, so reduce the log level.

Change-Id: Id62f3cfdbad85cdcbef234d41ef534c22c8f04ca
Signed-off-by: Cheng Gu <gucheng@google.com>
(cherry picked from commit a22fda77)

Force disable automatic AXI gating feature at kernel level.  This
is a preventive measure to improve stability.  This also aligns
behavior with B1C1, where this feature is not used.

Bug: 119834879
Change-Id: I51c1f14216a45ff16d59e8c9530aaf354ea047ac
Signed-off-by: Cheng Gu <gucheng@google.com>

FSP0 is a boot frequency and is not supported to be switched at
runtime.  Overwirte all FSP0 requests to be FSP1.

Bug: 133340008
Change-Id: I58e00143e26095ae55f35923de3fdcd06161bbea
Signed-off-by: Cheng Gu <gucheng@google.com>

The two defconfig files mnh_prod_defconfig and mnh_qemu_defconfig don't
have CONFIG_MODULES enabled.  Make respective changes in build.config.

Without this change, we are seeing the following error message when
running "make modules_install"

The present kernel configuration has modules disabled.
Type 'make config' and enable loadable module support.
Then build a kernel with module support enabled.

make[2]: *** [modules_install] Error 1
make[1]: *** [sub-make] Error 2
make: *** [__sub-make] Error 2

Change-Id: I72654f320169795209597795cb3ef4525c791911
Signed-off-by: Daniel Mentz <danielmentz@google.com>

With multi MSI enabled, the retry mechanism when sending wrap marker
should be removed.

Also increases the timeout for waiting for wrap_done signal to 1 sec.

Bug: 68210429
Change-Id: I93599f5464cbcf6d7748da6959f06d664e2cd025
Signed-off-by: Cheng Gu <gucheng@google.com>

Change-Id: I60a855d9237bfcfc1fa2ebad45f2c938ba6c6d2e
Signed-off-by: Trevor Bunker <trevorbunker@google.com>

Bug: 68210429
Change-Id: I7f1cd6cb17e225761c808565fb8ee3945244acb3
Signed-off-by: Trevor Bunker <trevorbunker@google.com>

Bug: 68210429
Change-Id: I82814ff1d74401e4a158a2fc81c7bdb4d04fcb47
Signed-off-by: Trevor Bunker <trevorbunker@google.com>

If the IRQ entry in the wait object has already been triggered then do
not add the event again.  The event will be left in the IRQ object's
pending queue and will be reported in a later wait for interrupt.

Also do not add the IRQ event if the wait object has met its exit
criteria and the waiting thread has been signaled.

Bug: 68724906
Change-Id: I7b40f0b59e0fddbdd224cdfa1da62a22d7cbdf38
Signed-off-by: Adam Hampson <ahampson@google.com>

The Runtime uses a try, fail, try another method for allocating IPU
resources.  Access errors do not indicate a problem, just a resource
in  use.  This changes the log level for the access messages to
warning.

Change-Id: Idbfefdef8f26217b5459988c0e37f64782a96f78
Signed-off-by: Adam Hampson <ahampson@google.com>

The previous implementation tried to reuse an existing stream
enable interface which created a race condition with regard to the
free_running and frame_count fields in the stream data structure.
This change simplifies the code and explicitly configures the stream
for free running mode and for disable on error.

Bug: 68045211

Change-Id: I9bd98c29f991de2f8a35d6d1ab7166b4668d1f58
Signed-off-by: Adam Hampson <ahampson@google.com>

On a MIPI SOF, check to make sure there are sufficient buffers for
the current frame and the next.  If there is not a sufficient numbers
of buffers for the next frame then disable the stream.

If a DMA buffer that has a MIPI source is enqueued and the
corresponding MIPI stream is not enabled then enable it after the
DMA transfer has been started.

Bug: 68045211
Change-Id: Id60140c0ba0847ada0245c1bd4cc11202924bc5c
Signed-off-by: Adam Hampson <ahampson@google.com>

Only permitted on exclusive sessions.

Allocations and interrupt bindings are preserved, but active and pending
DMAs are cancelled, MIPI streams are disabled, STPs are reset, and all
setup is cleared.

Bug: 67108372
Change-Id: I94d11b806b8dd160edcda6298ebf268311e11845
Signed-off-by: Sean Howarth <showarth@google.com>

Change-Id: I5ca5c0efc4a3a0a7db157d7b1a80e2b51a846b11
Signed-off-by: Adam Hampson <ahampson@google.com>

1.  Make stop completion non-interruptible.
The IPU driver was using the interruptible version of the wait for
completion call but was not handling signal interrupt properly.  The
stop operation should take at most 500us so the delay to signal
processing should a signal come in is not onerous.

2.  Report cancel on both active transfers
Previously the second transfer enqueued in hardware was reported as
canceled as part of the reset operation.  This makes it difficult to
detect the case where a reset is occurring while there is an active
transfer.  This change explicitly reports the transfer as canceled
before issuing the DMA channel reset.

Change-Id: I0ae8db6083f68a24a8df7b707bc8d8de2ed6fe7b
Signed-off-by: Adam Hampson <ahampson@google.com>

If a MIPI overflow occurs because there is no DMA buffer to receive
the data then there will be no DMA EOF at the end of the frame.  In
this scenario the end of frame MIPI tasks need to be handled in the
overflow interrupt handler.

This change also improves MIPI overflow kernel logging.

Bug: 65757237
Bug: 67041967
Change-Id: I3fc9ea0c58183d22078e6823447d0f865ff3e64d
Signed-off-by: Adam Hampson <ahampson@google.com>

The absence of this kernel config option was preventing tests with
multiple launches from succeeding.

Change-Id: I1522e432805d0857295e2b7607be837c0a66fa47
Signed-off-by: Vince Harron <vharron@google.com>

mnh_qemu_defconfig was created by taking options from mnh_prod_defconfig
(and a few from mnh_defconfig) and mixing in some VM support.

Other trivial fixes.

Change-Id: I20742c715dd7466275c8093dfdb701e6a09b42c8
Signed-off-by: Vince Harron <vharron@google.com>

Fixes this warning:
drivers/pcie_ep/mnh_pcie_ep.c:135:14: warning: 'mnh_scu_base' defined
but not used [-Wunused-variable]
 static void *mnh_scu_base;

Bug: 65563892
Test: build.config.qemu from go/pag/911754
Change-Id: I3767a41599e086719a834a11740c119711ca8e71
Signed-off-by: Cheng Gu <gucheng@google.com>

In a situation where race condition causes WRAP_DONE MSI to be
missing, a sending message will be blocked forever.

This patch adds a timeout on waiting for WRAP_DONE MSI.  It will
retry multiple times if it happens to hit the race condition that
WRAP_DONE MSI is overridden.

Backport of msm-google:
commit 84aa856f1518 ("misc: easelcomm: Set timeout on waiting for wrap done")

Bug: 63140997
Bug: 65678253
Change-Id: I35ef6fca3f304abcbaddf54714cd8fa619d8f837
Signed-off-by: Cheng Gu <gucheng@google.com>

Easelcomm single-block DMA handling is broken at the moment.  It's not
needed anymore, turn it off.

Bug: 65762838
Test: easelcomm_test
Change-Id: Ic7b2e8a158b059228f60188f78508c49777e44b7
Signed-off-by: Todd Poynor <toddpoynor@google.com>

Note this is an iterative drop on top of the previous driver drops.

Kernel Commit:
commit f2e9b4a03289 ("misc: paintbox: fixed regs debugfs entry")

Change-Id: Ic811c00efb40376c1464f7a159b2018722ba0fa6
Signed-off-by: Sean Howarth <showarth@google.com>

When wait_for_completion_interruptible_timeout() returns, the caller
checks if ret is set.  This variable should be cleared before
calling that function.

This patch fixes false positive where a valid reply returns -ETIMEDOUT
back to userspace.

Bug: 65553136
Change-Id: If508a454a8cf3e8a4db2359117a848853043194b
Signed-off-by: Cheng Gu <gucheng@google.com>

Add CONFIG_MNH_PCIE_BOOT_TRACE=y to defconfig if the kernel is
needed to log boot trace by writing to PCIe SCU registers

Bug: 65563892
Change-Id: Ibbbd4c94e7878e7115ddde88b18c0f6253df7233
Signed-off-by: Cheng Gu <gucheng@google.com>

Enable /proc/sysrq-trigger to test OOM killer with production
kernel.

Change-Id: I08d8db941979c15a2d95642615925cd88bdc5ff3
Signed-off-by: Cheng Gu <gucheng@google.com>

Note this is an iterative drop on top of the previous driver drops.

Kernel Commit:
commit 7e54b9a119c4 ("misc: paintbox: add ioctl to wait for mipi quiescence")

Bug: 64945572
Change-Id: I7a4b8ca146dad74895c2f233229e9da5b21e077d
Signed-off-by: Adam Hampson <ahampson@google.com>

This helps identify the size intended to be transferred and debug
possible issues when transferring a userspace buffer.

Bug: 65177181
Change-Id: Id9150318299cbe2696aa031f29a57b49132c5d47
Signed-off-by: Cheng Gu <gucheng@google.com>

Bug: 37647527
Change-Id: I1d89125f31f47e0de0b9fb956c6f7440abcb93d3
Signed-off-by: Trevor Bunker <trevorbunker@google.com>

There is an aritrary size limit per message.  When user attempts
sending a large message, the driver returns -EINVAL as error.

This patch prints an error log in kernel to help debugging.

Backport of msm kernel:
commit 853cfe1e82e7 ("misc: easelcomm: Print error when msg is too big")

Bug: 64402906
Change-Id: Ic237a3fb7903627ddd639781116b931699e70c64
Signed-off-by: Cheng Gu <gucheng@google.com>

Returned value was not accounting for zero-terminator, resulting in
short, unterminated scatter-gather lists being sent over to the
Easel side. Updated sg list calculation used to decide whether
to use single or multi-block transfers

Bug: 63809896
Change-Id: I98b74b31996346b60dbe9b39840b79a968180319
Signed-off-by: Shac Ron <sron@google.com>

Bug: 64936837
Change-Id: I4b2346afd3f801bb906232f808402a412072cac1
Signed-off-by: Trevor Bunker <trevorbunker@google.com>