On Tue, Jan 23, 2024 at 10:54:16AM -0800, Chris Lew wrote: > > > On 1/17/2024 7:52 PM, Bjorn Andersson wrote: > > @@ -235,6 +238,8 @@ int qmp_send(struct qmp *qmp, const char *fmt, ...) > > mutex_lock(&qmp->tx_lock); > > + trace_aoss_send(buf); > > + > > /* The message RAM only implements 32-bit accesses */ > > __iowrite32_copy(qmp->msgram + qmp->offset + sizeof(u32), > > buf, sizeof(buf) / sizeof(u32)); > > @@ -256,6 +261,8 @@ int qmp_send(struct qmp *qmp, const char *fmt, ...) > > ret = 0; > > } > > + trace_aoss_send_done(buf, ret); > > + > > As a side note, another place where we've traced before is on the receiving > irq to get the full timing of how long AOSS takes to process a command. I > think we've discussed this in the past and decided that we can use kprobes > if that need occurs. > Right, these two tracepoints would not provide an accurate measurement for the that time. But as we discussed, we can use this in combination with other events to acquire such measurements. > > mutex_unlock(&qmp->tx_lock); > > return ret; > > diff --git a/drivers/soc/qcom/trace-aoss.h b/drivers/soc/qcom/trace-aoss.h > > new file mode 100644 > > index 000000000000..48cd3f0f4cb8 > > --- /dev/null > > +++ b/drivers/soc/qcom/trace-aoss.h > > @@ -0,0 +1,48 @@ > > +/* SPDX-License-Identifier: GPL-2.0 */ > > +/* > > + * Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved. > > + */ > > + > > +#undef TRACE_SYSTEM > > +#define TRACE_SYSTEM qcom_aoss > > + > > +#if !defined(_TRACE_RPMH_H) || defined(TRACE_HEADER_MULTI_READ) > > +#define _TRACE_RPMH_H > > Any Reason for this to be _TRACE_RPMH_H instead of _TRACE_AOSS_H? > That's what copy-paste gave me... Thanks for spotting that! Regards, Bjorn
