Instead of passing a pointer to the interrupt descriptor, embedded in
a cascading interrupt object and then using container_of() to get
back to the cascading interrupt, pass a pointer to it directly to
functions.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
When registering a cascading interrupt controller as a child of
another interrupt controller, no new child descriptor has to be
allocated, instead the embedded into the cascading interrupt
controller object descriptor has to be used.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Currently on cAVS platforms the two kinds of interrupts are treated
differently: the native DSP interrupts and "level" interrupts,
multiplexed by cascading interrupt controllers. For the latter
bitfields are used with both the parent DSP interrupt and the child
multiplexed interrupt numbers encoded in them. This is inflexible and
doesn't scale when more cascading interrupt controllers are added.
This patch switches over from using those bitfields to dynamically
mapping cascaded interrupts into a single plain interrupt number
space. Specifically this patch does the following:
- For interrupt conttrollers it adds an .irq_base field, which is the
base for child interrupts, multiplexed by this controller.
- An interrupt mapping function interrupt_get_irq() is added, which
uses the name of the cascading interrupt controller and an
interrupt number on that controller to create a logical interrupt
number.
- All uses of SOF_IRQ() and SOF_ID_IRQ() and related bitfield-
manipulation macros are removed.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
When multiple interrupt handlers share the same interrupt, only
those, that enabled the interrupt on a specific CPU, have to be
called on that CPU.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
The SPI driver only supports the DMA mode, its IRQ line isn't used
and the IRQ number in Sue Creek data is wrong. Remove the field
completely.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Use the platform provided IDC IRQ number only once and save the
number in a private field. This relies on the fact, that the
interrupt number is the same on all cores.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Use the platform provided scheduler IRQ number only once and save the
number in a private field.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Where available, use the IRQ number of the parent instead of
extracting it from the child's interrupt, using SOF_IRQ_NUMBER().
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
DSP interrupts have their own interrupt number in parent interrupt
bits, therefore using SOF_IRQ_NUMBER() for them is redundant.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
The only two uses of SOF_IRQ_CPU() are currently in irq_mask() and
irq_unmask() on cAVS. However, those functions are always called
after interrupt_get_parent(), which makes sure, that the CPU, encoded
into the IRQ is the same as the current one. Therefore we can also
just use the current CPU in irq_mask() and irq_unmask().
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Most interrupts in SOF are handled completely locally: they are
registered, enabled, disabled, and handled on one and the same core.
There is only one case, where interrupts have to be unmasked on a
different core: in idc_enable_interrupts(). To make this possible
without using the CPU field in the IRQ bitfield we add a "cpu"
argument to interrupt_mask() and interrupt_unmask() functions.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Add interrupt controller .mask() and .unmask() operations. Use them
instead of platform_interrupt_mask() and platform_interrupt_unmask().
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Currently cascading interrupt controller descriptors are registered
per core. This works well with Xtensa's "levelN" IRQ controllers
because their child interrupt lines can be masked per core. However
generic interrupt controllers do not have such a capability, which
makes registering them per core meaningless and error-prone. Switch
over to global IRQ descriptors, allocating them from non-cacheable
memory to avoid having to manually synchronise them at run-time.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Instead of checking one bit at a time, use ffs() to only handle set
bits in the interrupt status. Also fix the unhandled interrupt
detection logic: complain only if no handler has been found for a
given interrupt.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Add a list of interrupt controllers, that will be allocated
at run-time from non-cacheable memory, and functions to perform such
a registration and to search through the list.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Until now cavs/interrupt.c used to call interrupt_get_parent() where
it had to verify whether an interrupt was a DSP interrupt or a
cascaded one. That function is rather heavy - it takes a spin-lock,
scans a list of all cascading interrupt controllers and checks their
interrupt numbers. Whereas that check can be performed much easier
and faster by simply checking the "level" field of the IRQ token.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
platform_interrupt_mask() and platform_interrupt_unmask() have a
"mask" parameter, that is actually never used. It is only used on
baytrail, but the only use of those functions on that platform is
from the ssp.c driver, where mask is fixed to 1. Besides the meaning
of that parameter was unclear: it would be logical to assume, that it
is a bitmask that has to be applied to a masking register, but often
those functions were called with mask=0, which makes no sense. This
patch removes that parameter.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Currently the SOF API in addition to interrupt_clear() and
interrupt_set() functions also has platform_interrupt_clear() and
platform_interrupt_set(). Those functions are partially overlapping
and their choice is confusing. Unify them to only use the former
variant - without the platform_ prefix.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
platform_interrupt_clear() is a dummy function on cAVS platforms, no
need to call it in cAVS-specific drivers.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Platform interrupt handling code should call arch_interrupt_clear()
directly, no need to use the interrupt_clear() wrapper.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Enabling interrupts before allocating context for them can at least
in theory lead to such an interrupt being triggered and then
attempting to access the context. Fix the order for task interrupts.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Sharing interrupts is currently implemented by inserting an interrupt
ID field into the interrupt bitmask. We want to get rid of bitmasked
interrupt numbers, therefore that method has to be replaced. This
patch uses an approach, similar to the Linux kernel: we pass the
interrupt handler argument to the interrupt_unregister() function to
be compared to the argument, used when registering the interrupt
handler. This means, that users must use unique arguments when
registering interrupts, but this is anyway the case in most cases. We
also add a check for this to interrupt_register().
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Currently each interrupt descriptor has an array of list heads, that
can be used to cascade further interrupts to it. But most interrupts
are simple decide interrupts, that cannot have children. This patch
introduces a separate type for interrupt controllers, that can indeed
have children.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
to align with machine driver
The physical link iDisp1 ID for HDMI should be changed to 3,4,5
to align with machine driver. ID 2 is defined for DMIC16k.
Signed-off-by: Zhang Keqiao <keqiao.zhang@linux.intel.com>
This change is to align sof_ipc_stream_params with its
kernel equivalent. The change is needed in both kernel
and FW because FW needs to use host_period_bytes which
where previously used to inform host about period IRQ.
Therefore now we introduce new member for it period_irq
and let host_period_bytes keep the proper value.
Signed-off-by: Marcin Rajwa <marcin.rajwa@linux.intel.com>
When a block is used, we need reset the searching, to get the real
continuous blocks as required, here fix it.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
This change allows us to free components and pipelines
in any order. Added setting component pointer to NULL
after comp_free() invocation. In pipeline_free() there is
a incovation of pipeline_comp_free() only if component
pointer is not NULL.
Signed-off-by: Bartosz Kokoszko <bartoszx.kokoszko@linux.intel.com>
These traces shouldn't be duplicated, they should be in
more public header because they are used outside of FW.
Signed-off-by: Janusz Jankowski <janusz.jankowski@linux.intel.com>
Reverts usage of __packed macro in ipc, kernel and user
headers. These headers are not only used by firmware,
so they shouldn't have any dependency on it.
Signed-off-by: Tomasz Lauda <tomasz.lauda@linux.intel.com>
Changes cmake script to use CONFIG_BOOT_LOADER instead
of specific platform configs when choosing whether to
build bootloader.
Signed-off-by: Tomasz Lauda <tomasz.lauda@linux.intel.com>
Removes static assert for HEAP_BUF_ALIGNMENT, because it
doesn't make sense anymore to keep it. The value of this
define has been changed to PLATFORM_DCACHE_ALIGN, so it
will always be valid.
Signed-off-by: Tomasz Lauda <tomasz.lauda@linux.intel.com>
All cache related defines should be kept inside
dedicated cache header. Also changes sizeof(uint32_t)
to sizeof(void *), so the header could be included
from assembly files.
Signed-off-by: Tomasz Lauda <tomasz.lauda@linux.intel.com>
Removes PLATFORM_TIMER_START_OFFSET define from
platforms, because it's deprecated and not used
anymore.
Signed-off-by: Tomasz Lauda <tomasz.lauda@linux.intel.com>
This commit does not change FW behaviour since
both COMP_STATUS_STATE_ALREADY_SET and PPL_STATUS_PATH_STOP
defines are equal to 1. It just makes code more clear.
Signed-off-by: Bartosz Kokoszko <bartoszx.kokoszko@linux.intel.com>
Moves task structure and defines to the sof/task.h,
where they should be from the beginning. sof/schedule.h is
just for scheduler related things.
Signed-off-by: Tomasz Lauda <tomasz.lauda@linux.intel.com>
Guennadi Liakhovetski