[PATCH v5 03/13] coresight: Refactor etm4_config_timestamp_event()
Leo Yan
leo.yan at arm.com
Thu Nov 20 06:18:21 PST 2025
On Thu, Nov 20, 2025 at 01:52:03PM +0000, James Clark wrote:
[...]
> > > + config->cntr_ctrl[ctridx] = TRCCNTCTLRn_RLDSELF |
> > > + FIELD_PREP(TRCCNTCTLRn_RLDSEL_MASK, ETM_RES_SEL_FALSE) |
> > > + FIELD_PREP(TRCCNTCTLRn_CNTSEL_MASK, ETM_RES_SEL_TRUE);
> > >
> >
> > So if we define generic event generators:-
> >
> > #define ETM4_SEL_RS_PAIR BIT(7)
> > #defiine ETM4_RS_SEL_EVENT_SINGLE(rs_sel_idx) (GENMASK(4:0) & rs_sel_idx)
> > #define ETM4_RS_SEL_EVENT_PAIR(rs_sel_pair_idx) ((GENMASK(3:0) &
> > rs_sel_pair_idx) | ETM4_SEL_RS_PAIR)
> >
> > these are then reuseable for all registers that need the 8 bit event
> > selectors, beyond just this register.
> >
> > Thus we now accurately define the fields in the TRCCNTCTLRn
> >
> > #define TRCCNTCTLRn_RLDEVENT_MASK GENMASK(15, 8)
> >
> > and use
> >
> > FIELD_PREP(TRCCNTCTLRn_RLDEVENT_MASK,
> > ETM4_RS_SEL_EVENT_SINGLE(ETM_RES_SEL_FALSE))
> >
> > etc.
> >
> >
>
> I'm not sure I agree with that, the Arm ARM has CNTEVENT_TYPE as a regular
> bit in the TRCCNTCTLRn register so it should be set like any other. Hiding
> it as a subfield of "EVENT" when it always exists and always does the same
> thing was maybe seen as a bad decision which is why it was updated?
>
> Also IMO, beyond using labels instead of raw numbers, the code should just
> show what's programmed into the register. ETM4_RS_SEL_EVENT_SINGLE() would
> be one more macro to jump through to see what's actually happening.
Maybe define a general macro but with extra checking:
#define TRCCNTCTLRn_RLDEVENT_MASK GENMASK(15, 8)
#define ETM4_RS_SEL_EVENT(paired, sel) ({ \
if (paired) \
assert(!(sel & ~GENMASK(3, 0))); \
else \
assert(!(sel & ~GENMASK(4, 0))); \
FIELD_PREP(TRCCNTCTLRn_RLDEVENT_MASK, \
((paird << 7) | sel)); \
})
Thanks,
Leo
More information about the linux-arm-kernel
mailing list