linux.git/sound/soc/sof/amd/acp.c, branch v6.8

ASoC: SOF: amd: Fix locking in ACP IRQ handler

2024-02-11T16:58:50+00:00

A recent change in acp_irq_thread() was meant to address a potential race
condition while trying to acquire the hardware semaphore responsible for
the synchronization between firmware and host IPC interrupts.

This resulted in an improper use of the IPC spinlock, causing normal
kernel memory allocations (which may sleep) inside atomic contexts:

1707255557.133976 kernel: BUG: sleeping function called from invalid context at include/linux/sched/mm.h:315

...

1707255557.134757 kernel:  sof_ipc3_rx_msg+0x70/0x130 [snd_sof]
1707255557.134793 kernel:  acp_sof_ipc_irq_thread+0x1e0/0x550 [snd_sof_amd_acp]
1707255557.134855 kernel:  acp_irq_thread+0xa3/0x130 [snd_sof_amd_acp]
1707255557.134904 kernel:  ? irq_thread+0xb5/0x1e0
1707255557.134947 kernel:  ? __pfx_irq_thread_fn+0x10/0x10
1707255557.134985 kernel:  irq_thread_fn+0x23/0x60

Moreover, there are attempts to lock a mutex from the same atomic
context:

1707255557.136357 kernel: =============================
1707255557.136393 kernel: [ BUG: Invalid wait context ]
1707255557.136413 kernel: 6.8.0-rc3-next-20240206-audio-next #9 Tainted: G        W
1707255557.136432 kernel: -----------------------------
1707255557.136451 kernel: irq/66-AudioDSP/502 is trying to lock:
1707255557.136470 kernel: ffff965152f26af8 (&sb->s_type->i_mutex_key#2){+.+.}-{3:3}, at: start_creating.part.0+0x5f/0x180

...

1707255557.137429 kernel:  start_creating.part.0+0x5f/0x180
1707255557.137457 kernel:  __debugfs_create_file+0x61/0x210
1707255557.137475 kernel:  snd_sof_debugfs_io_item+0x75/0xc0 [snd_sof]
1707255557.137494 kernel:  sof_ipc3_do_rx_work+0x7cf/0x9f0 [snd_sof]
1707255557.137513 kernel:  sof_ipc3_rx_msg+0xb3/0x130 [snd_sof]
1707255557.137532 kernel:  acp_sof_ipc_irq_thread+0x1e0/0x550 [snd_sof_amd_acp]
1707255557.137551 kernel:  acp_irq_thread+0xa3/0x130 [snd_sof_amd_acp]

Fix the issues by reducing the lock scope in acp_irq_thread(), so that
it guards only the hardware semaphore acquiring attempt.  Additionally,
restore the initial locking in acp_sof_ipc_irq_thread() to synchronize
the handling of immediate replies from DSP core.

Fixes: 802134c8c2c8 ("ASoC: SOF: amd: Refactor spinlock_irq(&sdev->ipc_lock) sequence in irq_handler")
Signed-off-by: Cristian Ciocaltea 
Link: https://lore.kernel.org/r/20240208234315.2182048-1-cristian.ciocaltea@collabora.com
Signed-off-by: Mark Brown

ASoC: SOF: amd: Add acp-psp mailbox interface for iram-dram fence register modification

2023-12-19T14:02:49+00:00

Add acp-psp mailbox communication interface for iram-dram size
modification to notify psp.

Signed-off-by: Venkata Prasad Potturu 
Link: https://msgid.link/r/20231219112416.3334928-5-venkataprasad.potturu@amd.com
Signed-off-by: Mark Brown

ASoC: SOF: amd: Refactor spinlock_irq(&sdev->ipc_lock) sequence in irq_handler

2023-12-19T14:02:45+00:00

Refactor spinlock_irq(&sdev->ipc_lock) sequence in irq_handler
to avoid race conditions for acquiring hw_semaphore.

Signed-off-by: Venkata Prasad Potturu 
Link: https://msgid.link/r/20231219112416.3334928-1-venkataprasad.potturu@amd.com
Signed-off-by: Mark Brown

ASoC: SOF: make .remove callback return void

2023-10-12T19:28:12+00:00

We don't use the returned value and return 0 anyways, let's follow the
example of platform drivers and simplify the definitions.

Signed-off-by: Pierre-Louis Bossart 
Reviewed-by: Daniel Baluta 
Reviewed-by: Rander Wang 
Reviewed-by: Péter Ujfalusi 
Reviewed-by: Bard Liao 
Link: https://lore.kernel.org/r/20231012191850.147140-5-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown

ASoC: SOF: amd: clear dsp to host interrupt status

2023-08-23T12:27:19+00:00

DSP_SW_INTR_STAT_OFFSET is a common interrupt register which will be
accessed by both ACP firmware and driver. This register contains register
bits corresponds to host to dsp interrupts and vice versa.

when dsp to host interrupt is reported, only clear dsp to host
interrupt bit in DSP_SW_INTR_STAT_OFFSET.

Fixes: 2e7c6652f9b8 ("ASoC: SOF: amd: Fix for handling spurious interrupts from DSP")

Signed-off-by: Vijendar Mukunda 
Link: https://lore.kernel.org/r/20230823073340.2829821-7-Vijendar.Mukunda@amd.com
Signed-off-by: Mark Brown

ASoC: SOF: amd: add conditional check for acp_clkmux_sel register

2023-08-23T12:27:17+00:00

Few AMD platforms require ACP ACLK as clock source.
Add conditional check for clock mux selection register for
switching between internal clock and ACP ACLK.

Signed-off-by: Vijendar Mukunda 
Link: https://lore.kernel.org/r/20230823073340.2829821-5-Vijendar.Mukunda@amd.com
Signed-off-by: Mark Brown

ASoC: SOF: amd: remove redundant clock mux selection register write

2023-08-23T12:27:16+00:00

ACP clock mux selection register is already programmed during acp init
sequence.
Remove the redundant register write.

Signed-off-by: Vijendar Mukunda 
Link: https://lore.kernel.org/r/20230823073340.2829821-4-Vijendar.Mukunda@amd.com
Signed-off-by: Mark Brown

ASoC: SOF: amd: add module parameter for firmware debug

2023-08-23T12:27:15+00:00

Add module parameter for firmware debug. If firmware debug
flag is enabled, clear the fusion stall bit which is required
for enabling firmware debugging through JTAG.

Signed-off-by: Vijendar Mukunda 
Link: https://lore.kernel.org/r/20230823073340.2829821-3-Vijendar.Mukunda@amd.com
Signed-off-by: Mark Brown

ASoC: SOF: amd: enable ACP external global interrupt

2023-08-23T12:27:14+00:00

Previously ACP SOF firmware used to enable the ACP external
global interrupt register.
This will restrict to report ACP host interrupts only after
firmware loading is successful.
This register needs to be set from host driver to handle
other ACP interrupts(SoundWire Interrupts) before loading
the ACP firmware.

Add field for external interrupt enable register in acp descriptor
structure and enable the external interrupt enable register.

Signed-off-by: Vijendar Mukunda 
Link: https://lore.kernel.org/r/20230823073340.2829821-2-Vijendar.Mukunda@amd.com
Signed-off-by: Mark Brown

ASoC: SOF: amd: remove unused sha dma interrupt code

2023-08-23T12:27:13+00:00

During initial development time for RN platform, when SHA
dma gets completed, SHA DMA engine used to raise the ACP interrupt.
In ACP interrupt handler, SHA DMA interrupt got handled.
Currently SHA DMA compleition is verified by checking
transfer count using read poll time out logic.
Remove unused SHA dma interrupt handling code.

Signed-off-by: Vijendar Mukunda 
Link: https://lore.kernel.org/r/20230823073340.2829821-1-Vijendar.Mukunda@amd.com
Signed-off-by: Mark Brown