Saving unsupported state prevents migration when the new host does not
support a XSAVE feature of the original host, even if the feature is not
exposed to the guest.
We've masked host features with guest-visible features before, with
4344ee981e ("KVM: x86: only copy XSAVE state for the supported
features") and dropped it when implementing XSAVES. Do it again.
Fixes: df1daba7d1 ("KVM: x86: support XSAVES usage in the host")
Cc: stable@vger.kernel.org
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
After the rewrite of the runtime PM code, we have only two driver
status: CONNECTED and DISCONNECTED. So it's clearer to use a boolean
flag, and name it easier one, "connected".
Signed-off-by: Takashi Iwai <tiwai@suse.de>
- Add a few more comments to functions.
- Move the initialization of some PCM state variables to open and
prepare callbacks, where these are clearer places.
- Remove superfluous NULL checks.
- Get rid of the bogus drv_status change to CONNECTED at close;
this doesn't make any sense.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
It's a stand-alone small driver code, and we don't have to describe
too much formalized comments in kernel-doc style for local functions
at all.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This is a guess work. Usually the DP audio info frame is just 8-bit
shifted from HDMI AI, so let's try to put CA in DIP frame 2 [24-31].
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Like other drivers, expose the ELD bytes via a control element so that
user-space can parse it.
For the simplicity, the code to register the ctl elements is
refactored using an array. Also, since ELD ctl read copies the bytes
also during disconnection, clear the ELD bytes at hot-unplug, in order
to avoid the leak of the previous bogus ELD.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Currently each register definition contains the own prefix in the
union struct itself; for example, union aud_ch_status_0 has
status_0_regx and status_0_regval fields. These are simply
superfluous, since usually the type of the variable is seen in its
declaration or in its name.
In this patch, we cut off these prefixes. Now all register
definitions have regx and regval fields consistently, instead.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The only remaining field in struct had_stream_data is stream_type that
holds the current stream status. Such information fits better in
struct pcm_stream_info, so move it as a boolean "running" field to be
clearer.
This allows us to get rid or had_stream_data definition and
references.
Also, the superfluous status check get removed in a couple of places
where we can call PCM helpers in anyway.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Although the driver has some PM callbacks, it doesn't do it right:
- the suspend callback doesn't handle to suspend the running PCM,
- the runtime PM ops are missing,
- pm_runtime_get_sync() isn't used at the right place.
This patch covers the above and provides the basic runtime PM
functionality.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The PCM substream is referred not only in the PCM callbacks but also
in the irq handler and in the hotplug/unplug codes. The latter code
paths don't take the PCM lock, thus the PCM may be released
unexpectedly while calling PCM helper functions or accessing
pcm->runtime fields.
This patch implements a simple refcount to assure the PCM substream
accessibility while the other codes are accessing. It needed some
code refactoring in the relevant functions for avoiding the doubly
spinlocks.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The struct pcm_stream_info contains a few unused or useless fields.
str_id is always zero, buffer_ptr is volatile, never read, and sfreq
is nowhere referred. Kill them.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The had_stream_pvt struct assigned to PCM runtime private data tracks
merely the stream running status, and the very same information is
carried by had_stream->stream_type. Kill it.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The state field keeps the connection state and it's basically as same
as drv_status field. Drop this redundancy.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The flag_underrun flag is used to indicate to escalate the XRUN
reporting at the next position inquiry, but there is a much simpler
method to achieve it: just call snd_pcm_stop_xrun().
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The access to chmap can be racy against the hotplug process, where it
recreates the chmap on the fly. For protecting against it, a mutex is
introduced in this patch. It's also used for protecting the change /
reference of eld and state fields, too.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
We don't need to use irqsave/irqrestore versions for each spin lock,
but judge the context properly and use the simpler versions.
Also add some (still simplistic) comments to functions.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Although we dropped the most of the obsoleted *_v1 definitions and
codes, some codes still keep the _v1 or _v2 suffices. Now they are
ripped off.
The only thing to be done carefully here is the definition of control
offsets. The original code defines enum hdmi_ctrl_reg_offset_v1 and
a few new elements just for v2 on its top. After this cleanup, we
remove the old AUD_HDMI_STATUS and AUD_HDMIW_INFOFR definitions and
replace with the v2 values.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The helper function isn't clearer than the plain condition check
"if (drv_status == HDA_DRV_DISCONNECTED)". By expanding this, the
compiler could even catch the possible uninitialized cases, so we
could fix them, too.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The had_get_hwstate() is identical with drv_status==DISCONECTED, which
was already checked before the call. And, returning an error at
resume is simply bad. That is, we should just kill this check.
Also, spewing an error at resume for drv_status!=SUSPENDED is also
annoying, as this is the normal case when the suspend was called
without the monitor connection. Make it debug, too.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
i915 notification is executed in a spinlock, thus it must not sleep;
i.e. we can't use kmalloc with GFP_KERNEL or such.
For making it working properly, move the notification handler in a
work, and handle it gracefully. We have already such a work, and it
was used just at the start. This can be re-used in a more generic
hotplug handling.
Also, the patch adds the proper call of cancel_work_sync() to the
destructor.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
snd_pcm_lib_preallocate_free_for_all() doesn't have to be called from
each driver as it's called in the PCM core.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Again another indirect call... Let's straighten it up.
Also define the had_stream field with a proper type instead of a void
pointer.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The direct access to power.runtime_status is taboo, let's use a helper
macro to avoid the compile error with CONFIG_PM=n.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
dev_xxx() helpers give a tidier output in general.
While we're at it, remove many useless debug prints (e.g. the ones at
each function entry), replace some too verbose errors with debugs, and
use WARN_ON() for some serious errors.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
As the very last step, we fold intel_hdmi_audio_if.c into the main
file, intel_hdmi_audio.c. This is merely a cleanup, and no functional
change.
By this move, we can mark all functions and variables as static, which
allows the compiler more optimizations.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This is the final stage for a big clean-up series. Here we flatten
the two layers into one. Formerly, the implementation was split to
HDMI "shell" that talks with the platform device, and HDMI audio part
that communicates via caps and other event handlers. All these would
be good if there were multiple instantiations or if there were data
protection. But neither are true in our case. That said, it'll be
easier to have a flat driver structure in the end.
In this patch, the former struct hdmi_lpe_audio_ctx is forged into the
existing struct snd_intelhad. The latter has already a few members
that are basically the copy from the former. Only a few new members
for the lowlevel I/O are added by this change.
Then, the had_get_caps() and had_set_caps() are simply replaced with
the direct calls to copy the data in the struct fields. Also, the
had_event_handler() calls are replaced with the direct call for each
event as well.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The card registration should be done at the last stage of the probe
procedure. Otherwise user-space may access to the device before the
whole initialization is done.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Just a minor optimization; check the presence of platform_data earlier
in the probe function before actually starting allocation, which makes
the error path easier.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Since the config base offset is now set per pipe id, we don't have to
check Cherry Trail PCI IDs any longer. Currently it's used only for
debug prints. Let's drop it.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Instead of allocating snd_intelhad struct, use the card's private_data
and embed it. It simplifies the code a lot.
While we're at it, embed had_stream into snd_intelhad struct instead
of individually allocating, and rename had_pvt_data to a bit more
specific name, had_stream_data.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The previous patch introduced a separate inline asm version of the
krcrctab declaration template for use with 64-bit architectures, which
cannot refer to ELF symbols using 32-bit quantities.
This declaration should be equivalent to the C one for 32-bit
architectures, but just in case - unify them in a separate patch, which
can simply be dropped if it turns out to break anything.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The modversion symbol CRCs are emitted as ELF symbols, which allows us
to easily populate the kcrctab sections by relying on the linker to
associate each kcrctab slot with the correct value.
This has a couple of downsides:
- Given that the CRCs are treated as memory addresses, we waste 4 bytes
for each CRC on 64 bit architectures,
- On architectures that support runtime relocation, a R_<arch>_RELATIVE
relocation entry is emitted for each CRC value, which identifies it
as a quantity that requires fixing up based on the actual runtime
load offset of the kernel. This results in corrupted CRCs unless we
explicitly undo the fixup (and this is currently being handled in the
core module code)
- Such runtime relocation entries take up 24 bytes of __init space
each, resulting in a x8 overhead in [uncompressed] kernel size for
CRCs.
Switching to explicit 32 bit values on 64 bit architectures fixes most
of these issues, given that 32 bit values are not treated as quantities
that require fixing up based on the actual runtime load offset. Note
that on some ELF64 architectures [such as PPC64], these 32-bit values
are still emitted as [absolute] runtime relocatable quantities, even if
the value resolves to a build time constant. Since relative relocations
are always resolved at build time, this patch enables MODULE_REL_CRCS on
powerpc when CONFIG_RELOCATABLE=y, which turns the absolute CRC
references into relative references into .rodata where the actual CRC
value is stored.
So redefine all CRC fields and variables as u32, and redefine the
__CRC_SYMBOL() macro for 64 bit builds to emit the CRC reference using
inline assembler (which is necessary since 64-bit C code cannot use
32-bit types to hold memory addresses, even if they are ultimately
resolved using values that do not exceed 0xffffffff). To avoid
potential problems with legacy 32-bit architectures using legacy
toolchains, the equivalent C definition of the kcrctab entry is retained
for 32-bit architectures.
Note that this mostly reverts commit d4703aefdb ("module: handle ppc64
relocating kcrctabs when CONFIG_RELOCATABLE=y")
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This add the kbuild infrastructure that will allow architectures to emit
vmlinux symbol CRCs as 32-bit offsets to another location in the kernel
where the actual value is stored. This works around problems with CRCs
being mistaken for relocatable symbols on kernels that self relocate at
runtime (i.e., powerpc with CONFIG_RELOCATABLE=y)
For the kbuild side of things, this comes down to the following:
- introducing a Kconfig symbol MODULE_REL_CRCS
- adding a -R switch to genksyms to instruct it to emit the CRC symbols
as references into the .rodata section
- making modpost distinguish such references from absolute CRC symbols
by the section index (SHN_ABS)
- making kallsyms disregard non-absolute symbols with a __crc_ prefix
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Similarly like the previous patch, drop the global variable to keep
the ELD copy. It can be embedded in hdmi_lpe_audio_ctx as well.
And this makes easier to code, it's just a memcpy(), after all.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Now it's the turn to drop the global hlpe_state variable. It can be
gracefully embedded in hdmi_lpe_audio_ctx struct.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Instead of referring to the global hlpe_pdev variable, pass the
platform device object to each function properly. Accessing to the
global object is really ugly and error-prone.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Registering the irq handler at the too early place may cause a system
stall because the irq handler may be triggered before the other
initializations. Move the irq handler registration to the later point.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Currently the driver calls the event callback stored in its ctx
pointer, but it's obviously inefficient. Replace it with the direct
calls.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
had_mutex is (supposedly) used to protect the concurrent calls of
hdmi_audio_probe(). But we may have only one device at most, so it's
utterly useless. Drop it.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
