On 4/10/20 3:18 AM, Brent Lu wrote:
After sending an ipc command to DSP, the host waits for the reply message which will be read from SST_IPCD register in sst_byt_irq_thread() to complete the transaction. Sometimes the value read from SST_IPCD register is still the reply message for previous command instead of the waiting command so ipc command timeout happens.
In an experiment we read the same SST_IPCD register again when the defect happens and found the value of second read is different from previous one and is the correct reply message. It suggests the DSP is okay but the way we read the register may be the cause.
Currently the driver is using memcpy_fromio() to read the value of 64-bit registers. This function is based on __builtin_memcpy() call and depends on the implementation of compiler. Since this issue happens right after the toolchain switched from clang 10 to clang 11, we replace the register read with two readl() function calls to avoid all optimization from compiler's library.
I have mixed feelings about this.
One one hand, this looks simple enough.
But on the other hand we have other users of memcpy_fromio(), including SOF drivers, so what are the odds we have the same problems in other places? Wouldn't it be safer to either change this function so that it's behavior is not ambiguous or compiler-dependent, or fix the compiler?
Signed-off-by: Brent Lu brent.lu@intel.com
sound/soc/intel/common/sst-dsp.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/sound/soc/intel/common/sst-dsp.c b/sound/soc/intel/common/sst-dsp.c index ec66be2..12af7aa 100644 --- a/sound/soc/intel/common/sst-dsp.c +++ b/sound/soc/intel/common/sst-dsp.c @@ -42,7 +42,7 @@ u64 sst_shim32_read64(void __iomem *addr, u32 offset) { u64 val;
- memcpy_fromio(&val, addr + offset, sizeof(val));
- sst_memcpy_fromio_32(NULL, &val, addr + offset, sizeof(val)); return val; } EXPORT_SYMBOL_GPL(sst_shim32_read64);