trace: Add SOF_IPC_TRACE_DMA_FREE IPC command

[ranj063]

This will be used to stop the trace DMA and free
its resources. This change is tagged for ABI 3.20.

kernel PR thesofproject/linux#3197

[ujfalusi]

Should we also put the channel and set d->dc.chan to NULL?
SOF_IPC_TRACE_DMA_PARAMS_EXT will fail if we have a valid channel already.

[ujfalusi]

Unless we also have #4850 along with this patch.

[kv2019i]

#4850 was merged, so this is good right @ujfalusi ?

[ujfalusi]

I would still put the channel for consistency:

if (d->dc.chan) {
	dma_stop(d->dc.chan); /* Should we do an error check? */
	dma_channel_put(d->dc.chan);
	d->dc.chan = NULL;
}

[ranj063]

ok, ill release the channel but error checking would be pointless because trace work is already stopped right?

[kv2019i]

#4850 was merged, so this is good right @ujfalusi ?

[lgirdwood]

@ujfalusi pls approve if good for you.

[lgirdwood]

@ranj063 looks like we have a crash

unknown ipc: unknown command type 0
unknown ipc: debug cmd 0x50000
AddressSanitizer:DEADLYSIGNAL
=================================================================
==22==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000000 (pc 0x0000005870d2 bp 0x7fffb3ce93f0 sp 0x7fffb3ce93b0 T0)
==22==The signal is caused by a READ memory access.
==22==Hint: address points to the zero page.
SCARINESS: 10 (null-deref)
    #0 0x5870d2 in dma_stop /src/sof/src/include/sof/lib/dma.h:316:21
    #1 0x5870d2 in dma_trace_disable /src/sof/src/trace/dma-trace.c:429:2
    #2 0x55f464 in ipc_dma_trace_free /src/sof/src/ipc/ipc3/handler.c:757:2
    #3 0x55f464 in ipc_glb_trace_message /src/sof/src/ipc/ipc3/handler.c:886:3
    #4 0x55f464 in ipc_cmd /src/sof/src/ipc/ipc3/handler.c:1571:9
    #5 0x55dc56 in LLVMFuzzerTestOneInput /src/sof/tools/oss-fuzz/fuzz_ipc.c:28:2
    #6 0x456b03 in fuzzer::Fuzzer::ExecuteCallback(unsigned char const*, unsigned long) cxa_noexception.cpp
    #7 0x4562fd in fuzzer::Fuzzer::RunOne(unsigned char const*, unsigned long, bool, fuzzer::InputInfo*, bool, bool*) cxa_noexception.cpp
    #8 0x457b8b in fuzzer::Fuzzer::MutateAndTestOne() cxa_noexception.cpp
    #9 0x458675 in fuzzer::Fuzzer::Loop(std::__Fuzzer::vector<fuzzer::SizedFile, std::__Fuzzer::allocator<fuzzer::SizedFile> >&) cxa_noexception.cpp
    #10 0x447d68 in fuzzer::FuzzerDriver(int*, char***, int (*)(unsigned char const*, unsigned long)) cxa_noexception.cpp
    #11 0x470bd2 in main /src/llvm-project/compiler-rt/lib/fuzzer/FuzzerMain.cpp:20:10
    #12 0x7fc4ca4a40b2 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x270b2)
    #13 0x41f68d in _start (build-out/fuzz_ipc+0x41f68d)

DEDUP_TOKEN: dma_stop--dma_trace_disable--ipc_dma_trace_free
AddressSanitizer can not provide additional info.
SUMMARY: AddressSanitizer: SEGV /src/sof/src/include/sof/lib/dma.h:316:21 in dma_stop
==22==ABORTING

[ujfalusi]

Why don't we call this as dma_trace_free() to be consistent?

[ranj063]

it balances with trace_dma_enable during ipc_dma_trace_params() no?

[ujfalusi]

I would still put the channel for consistency:

if (d->dc.chan) {
	dma_stop(d->dc.chan); /* Should we do an error check? */
	dma_channel_put(d->dc.chan);
	d->dc.chan = NULL;
}

[ujfalusi]

@ranj063, thank you, looks great!

[lgirdwood]

CI showing a presence detect failure on ADL. @marc-hb I've seen this several time today, looks like we may need to filter the cpufreq errors out ? https://sof-ci.01.org/sofpr/PR4849/build10723/devicetest/?model=ADLP_RVP_SDW&testcase=verify-kernel-boot-log

[marc-hb]

cpufreq - and more. All boot logs on ADK: thesofproject/sof-test#786