The driver don't need to explicit enable the power domain, which
can be done by runtime power management, when the power domain tree
defined in device tree.
in this case, the MU initialization can be moved to runtime pm function.
Signed-off-by: Shengjiu Wang <shengjiu.wang@nxp.com>
In order to avoid the name problem going forward with
integration with Qcom, Qcom has their own dsp and hifi
is competitor, so the hifi name should not be used in
our code.
So use the name of dsp instead of hifi to fix this
problem.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
The architecture of dsp framework has been changed, so update dsp
driver to support suspend & resume test of new dsp framework.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
When using memcpy() or fread() function in dsp unit test or wrapper code,
an unhandled alignment fault error will occur randomly, this issue
is caused by the setting of mmap(). After using pgprot_writecombine()
function instead of pgprot_noncached() function, this error will not
occur.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
The architecture of dsp framework has been changed, the role of
dsp driver is transferring messages between dsp framework and user space
application, so change dsp driver to support this function.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
In order to avoid license problem of Cadence header files, these
license files has been wrappered into a library and new interface
has been abstracted to replace the interface of Cadence header
files.
So update the mxc_hifi4.h file to provide new interface for
user space to use.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
add reset command declaration into mxc_hifi4.h file,
this command is used to reset hifi4 codec when seeking
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
Reviewed-by: Daniel Baluta <daniel.baluta@nxp.com>
update the mxc_hifi4.h header file to support multi-codec
decoding or encoding together for hifi4 dsp.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
Reviewed-by: Mihai Serban <mihai.serban@nxp.com>
This patch is used to fix Coverity-1793874, Coverity-1793875,
Coverity-1793876, Coverity-1793877 issue.
The icm_base_info_t is not initialized before using, so
use memset() to initialize it to fix these issues.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
When loading the codec libs in driver, if the destination is
not 4-bytes alignment when doing memset_hifi(), the driver
will print a warning message and the driver may crash in some
cases.
So by changing the memset() function and aligning the virtual address
based on the physical address to fix this issue.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
In order to manage the memory simply, all the memory which is
shared between hifi driver and hifi framework are managed by
hifi framework.
So when the driver wants to get free memory, it can send
"ICM_PI_LIB_MEM_ALLOC" command to hifi framework, then hifi
framework will return the address of available memory to
driver. When the driver wants to release the memory, it can
send "ICM_PI_LIB_MEM_FREE" command to hifi framework, the hifi
framework will mark this memory available.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
In current hifi driver, some resources are shared when multi
codec decodes together. When switching between multi-codec,
the hifi driver and framework need to save and restore the shared
resources,this will waster time and complicate the hifi driver.
So by distributing private resources for each codec to avoid
this problem. When the user space wants to enable a new codec,
it can send "HIFI4_CLIENT_REGISTER" command to hifi driver to apply
an available resource, the driver will send a client id to
user space. When the user space wants to release the resource,
it can send "HIFI4_CLIENT_UNREGISTER" command to hifi driver,
then the driver will mark this resource available.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
For hifi need to enter runtime suspend state in suspend,
then the power of HIFI can be down. In this case content
in internal RAM will be lost, and need to be recovered
in resume.
Move the loading firmware to runtime resume function, and
define ICM_SUSPEND and ICM_RESUME command, with ICM_SUSPEND
the hifi framework will store the data in RAM and with
ICM_RESUME the hifi framework will restore the data to RAM.
Signed-off-by: Shengjiu Wang <shengjiu.wang@nxp.com>
move hifi4 dsp firmware's code and data section to SDRAM space
move hifi4 dsp codec lib's code section to SDRAM space
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
When error occurs in fsl_hifi4_open() function, before this
function exists, "hifi4_priv->hifi4_mutex" should be unlocked.
If not, when the device is opened next time, the kernel will
be hanged.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
Enable pm runtime for hifi4, so the firmware may load many times,
The shdr->sh_addr can't be refined in hifi4_load_firmware, otherwise
it should impact the load operation in next time.
Signed-off-by: Shengjiu Wang <shengjiu.wang@nxp.com>
add cases to support resetting the hifi4 codec when receiving
HIFI4_RESET_CODEC command from the user space.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
Reviewed-by: Daniel Baluta <daniel.baluta@nxp.com>
Signed-off-by: Dong Aisheng <aisheng.dong@nxp.com>
When abnormal situation occurs and the current process terminates
abnormally, the hifi4 driver can't get the HIFI4_CODEC_CLOSE CMD
from user space to release the multi-codec resource, so the current
resource can't be used again.
Have found that the fsl_hifi4_close() function can be called
implicitly when process terminates abnormally, so add a reference
counter in fsl_hifi4_open() and fsl_hifi4_close() to check this
abnormal situation, when the number is same for opening and closing
hifi4 device, the multi-codec should be reinitialized again and
the hifi4 driver should send ICM_EXT_MSG_ADDR CMD to hifi4 framework
to initialize the multi-codec resources too.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
The previous hifi4 driver and framework code can't support
multi-codec decoding or encoding together, so change the driver
code to support this feature.
Currently, the hifi4 driver and framework can support at most
5 codec working together.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
Reviewed-by: Mihai Serban <mihai.serban@nxp.com>
When loop testing the hifi4 dirver, a random crash issue always
occur when doing memcpy() in decode function.
Have found that memcpy() is not suited to transfer data between
kernel space and user space, so use copy_from_user() and copy_
to_user() function to replace memcpy() to fix this issue.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
Move the load firmware operation from probe function to open,
Then firmware can be loaded from rootfs.
Signed-off-by: Shengjiu Wang <shengjiu.wang@freescale.com>
As the fsl_hifi4.c uses the function from uboot/cmd/elf.c,
so need to add the copyright of elf.c, and change licence to
Dual BSD/GPL.
And mxc_hifi4.h is used by user space, so change license to BSD.
Signed-off-by: Shengjiu Wang <shengjiu.wang@freescale.com>
Reviewed-by: Frank Li <frank.li@nxp.com>
In fsl_hifi4_probe(), the length for dma_alloc_coherent() is
MSG_BUF_SIZE + INPUT_BUF_SIZE + OUTPUT_BUF_SIZE +
FIRMWARE_DATA_BUF_SIZE + SCRATCH_DATA_BUF_SIZE;
However, in fsl_hifi4_remove(), the length for dma_free_coherent()
is MSG_BUF_SIZE + INPUT_BUF_SIZE + OUTPUT_BUF_SIZE +
FIRMWARE_DATA_BUF_SIZE;
By keeping the same length between dma_alloc_coherent() and
dma_free_coherent() to fix this issue.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
When dsp driver can't find the dsp core lib in loading codec
process, the kernel will be crashed. This issue is caused
by unreasonable way of error handling.
By changing the way of error handling to fix this issue.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
When building sound/soc/fsl/fsl_hifi4.c file, a warning
occurs:
warning: cast to pointer from integer of different
size [-Wint-to-pointer-cast]
(struct timestamp_info_t *)pext_msg->dtstamp;
By forced conversing int type to long type to fix this
issue.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
1. add cases to receive error value from hifi4 firmware and
return this error to hifi4 driver's caller.
2. add cases to receive input over indicator variable from
hifi4 dirver's caller and pass this value to hifi4 firmware
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
when transferring struct icm_open_resp_info_t type
between hifi4 framework and hifi4 driver, because this
struct has an element "*dtstamp" which is a pointer,
but for hifi4 firmware, this pointer occupies 4 bytes,
for hifi4 driver, this pointer occupies 8 bytes.
different pointer length will cause issue when reading
this structure's content in hifi4 driver.
By changing the pointer type to unsigned int type to
fix this issue.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
In order to use the hifi4's Cache to cache the firmware's
.rodata, .text, .data, .bss section and hifi4 core lib's
.text section, the firmware's .rodata, .text, .data and
.bss section should be remaped to 0x20700000 - 0x20FFFFFF
address range. This patch is used to parse the firmware
and load each section to corresponding address range.
This patch also set csr_gpr_control to 0x515A2080 to
remap the hifi4's address range in SCFW.
In addtion, add cases to support hifi4 framework's
performance test.
Signed-off-by: Weiguang Kong <weiguang.kong@nxp.com>
The function of driver is to communicate with hifi firmware.
The mu13 is dedicated for hifi communication, driver allocate
a share memory for message transfer between driver and firmware.
The calling sequence is that LOAD_CODEC,INIT_CODEC,CODEC_OPEN,
DECODE_ONE_FRAME, CODEC_CLOSE.
Signed-off-by: Shengjiu Wang <shengjiu.wang@freescale.com>
[ Aisheng: clean up for a new base ]
Signed-off-by: Dong Aisheng <aisheng.dong@nxp.com>
Pull cramfs fix from Al Viro:
"Regression fix, fallen through the cracks"
* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
cramfs: fix usage on non-MTD device
When both CONFIG_CRAMFS_MTD and CONFIG_CRAMFS_BLOCKDEV are enabled, if
we fail to mount on MTD, we don't try on block device.
Note: this relies upon cramfs_mtd_fill_super() leaving no side
effects on fc state in case of failure; in general, failing
get_tree_...() does *not* mean "fine to try again"; e.g. parsed
options might've been consumed by fill_super callback and freed
on failure.
Fixes: 74f78fc5ef ("vfs: Convert cramfs to use the new mount API")
Signed-off-by: Maxime Bizon <mbizon@freebox.fr>
Signed-off-by: Nicolas Pitre <nico@fluxnic.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Pull last minute virtio bugfixes from Michael Tsirkin:
"Minor bugfixes all over the place"
* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
virtio_balloon: fix shrinker count
virtio_balloon: fix shrinker scan number of pages
virtio_console: allocate inbufs in add_port() only if it is needed
virtio_ring: fix return code on DMA mapping fails
Pull input fix from Dmitry Torokhov:
"Just a single revert as RMI mode should not have been enabled for this
model [yet?]"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
Revert "Input: synaptics - enable RMI mode for X1 Extreme 2nd Generation"
This reverts commit 68b9c5066e39af41d3448abfc887c77ce22dd64d.
Ugh, I really dropped the ball on this one :\. So as it turns out RMI4
works perfectly fine on the X1 Extreme Gen 2 except for one thing I
didn't notice because I usually use the trackpoint: clicking with the
touchpad. Somehow this is broken, in fact we don't even seem to indicate
BTN_LEFT as a valid event type for the RMI4 touchpad. And, I don't even
see any RMI4 events coming from the touchpad when I press down on it.
This only seems to work for PS/2 mode.
Since that means we have a regression, and PS/2 mode seems to work fine
for the time being - revert this for now. We'll have to do a more
thorough investigation on this.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://lore.kernel.org/r/20191119234534.10725-1-lyude@redhat.com
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Pull networking fixes from David Miller:
1) Validate tunnel options length in act_tunnel_key, from Xin Long.
2) Fix DMA sync bug in gve driver, from Adi Suresh.
3) TSO kills performance on some r8169 chips due to HW issues, disable
by default in that case, from Corinna Vinschen.
4) Fix clock disable mismatch in fec driver, from Chubong Yuan.
5) Fix interrupt status bits define in hns3 driver, from Huazhong Tan.
6) Fix workqueue deadlocks in qeth driver, from Julian Wiedmann.
7) Don't napi_disable() twice in r8152 driver, from Hayes Wang.
8) Fix SKB extension memory leak, from Florian Westphal.
* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (54 commits)
r8152: avoid to call napi_disable twice
MAINTAINERS: Add myself as maintainer of virtio-vsock
udp: drop skb extensions before marking skb stateless
net: rtnetlink: prevent underflows in do_setvfinfo()
can: m_can_platform: remove unnecessary m_can_class_resume() call
can: m_can_platform: set net_device structure as driver data
hv_netvsc: Fix send_table offset in case of a host bug
hv_netvsc: Fix offset usage in netvsc_send_table()
net-ipv6: IPV6_TRANSPARENT - check NET_RAW prior to NET_ADMIN
sfc: Only cancel the PPS workqueue if it exists
nfc: port100: handle command failure cleanly
net-sysfs: fix netdev_queue_add_kobject() breakage
r8152: Re-order napi_disable in rtl8152_close
net: qca_spi: Move reset_count to struct qcaspi
net: qca_spi: fix receive buffer size check
net/ibmvnic: Ignore H_FUNCTION return from H_EOI to tolerate XIVE mode
Revert "net/ibmvnic: Fix EOI when running in XIVE mode"
net/mlxfw: Verify FSM error code translation doesn't exceed array size
net/mlx5: Update the list of the PCI supported devices
net/mlx5: Fix auto group size calculation
...
By default s_maxbytes is set to MAX_NON_LFS, which limits the usable
file size to 2GB, enforced by the vfs.
Commit b9b1f8d593 ("AFS: write support fixes") added support for the
64-bit fetch and store server operations, but did not change this value.
As a result, attempts to write past the 2G mark result in EFBIG errors:
$ dd if=/dev/zero of=foo bs=1M count=1 seek=2048
dd: error writing 'foo': File too large
Set s_maxbytes to MAX_LFS_FILESIZE.
Fixes: b9b1f8d593 ("AFS: write support fixes")
Signed-off-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Servers sending callback breaks to the YFS_CM_SERVICE service may
send up to YFSCBMAX (1024) fids in a single RPC. Anything over
AFSCBMAX (50) will cause the assert in afs_break_callbacks to trigger.
Remove the assert, as the count has already been checked against
the appropriate max values in afs_deliver_cb_callback and
afs_deliver_yfs_cb_callback.
Fixes: 35dbfba311 ("afs: Implement the YFS cache manager service")
Signed-off-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Call napi_disable() twice would cause dead lock. There are three situations
may result in the issue.
1. rtl8152_pre_reset() and set_carrier() are run at the same time.
2. Call rtl8152_set_tunable() after rtl8152_close().
3. Call rtl8152_set_ringparam() after rtl8152_close().
For #1, use the same solution as commit 8481141246 ("r8152: Re-order
napi_disable in rtl8152_close"). For #2 and #3, add checking the flag
of IFF_UP and using napi_disable/napi_enable during mutex.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Merge misc fixes from Andrew Morton:
"Three fixes"
* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
mm/ksm.c: don't WARN if page is still mapped in remove_stable_node()
mm/memory_hotplug: don't access uninitialized memmaps in shrink_zone_span()
Revert "fs: ocfs2: fix possible null-pointer dereferences in ocfs2_xa_prepare_entry()"
Marc Kleine-Budde says:
====================
pull-request: can 2019-11-22
this is a pull request of 2 patches for net/master, if possible for the
current release cycle. Otherwise these patches should hit v5.4 via the
stable tree.
Both patches of this pull request target the m_can driver. Pankaj Sharma
fixes the fallout in the m_can_platform part, which appeared with the
introduction of the m_can platform framework.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Since I'm actively working on vsock and virtio/vhost transports,
Stefan suggested to help him to maintain it.
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Once udp stack has set the UDP_SKB_IS_STATELESS flag, later skb free
assumes all skb head state has been dropped already.
This will leak the extension memory in case the skb has extensions other
than the ipsec secpath, e.g. bridge nf data.
To fix this, set the UDP_SKB_IS_STATELESS flag only if we don't have
extensions or if the extension space can be free'd.
Fixes: 895b5c9f20 ("netfilter: drop bridge nf reset from nf_reset")
Cc: Paolo Abeni <pabeni@redhat.com>
Reported-by: Byron Stanoszek <gandalf@winds.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
