Messages in this thread | | | Date | Mon, 23 Oct 2023 11:26:54 +0100 | From | Will Deacon <> | Subject | Re: [PATCH 2/3] perf/arm-cmn: Rework DTC counters (again) |
| |
On Fri, Oct 20, 2023 at 06:51:26PM +0100, Robin Murphy wrote: > The bitmap-based scheme for tracking DTC counter usage turns out to be a > complete dead-end for its imagined purpose, since by the time we have to > keep track of a per-DTC counter index anyway, we already have enough > information to make the bitmap itself redundant. Revert the remains of > it back to almost the original scheme, but now expanded to track per-DTC > indices, in preparation for making use of them in anger. > > Note that since cycle count events always use a dedicated counter on a > single DTC, we reuse the field to encode their DTC index directly. > > Signed-off-by: Robin Murphy <robin.murphy@arm.com> > --- > drivers/perf/arm-cmn.c | 126 +++++++++++++++++++++-------------------- > 1 file changed, 64 insertions(+), 62 deletions(-) > > diff --git a/drivers/perf/arm-cmn.c b/drivers/perf/arm-cmn.c > index f1ac8d0cdb3b..675f1638013e 100644 > --- a/drivers/perf/arm-cmn.c > +++ b/drivers/perf/arm-cmn.c > @@ -281,16 +281,13 @@ struct arm_cmn_node { > u16 id, logid; > enum cmn_node_type type; > > - int dtm; > - union { > - /* DN/HN-F/CXHA */ > - struct { > - u8 val : 4; > - u8 count : 4; > - } occupid[SEL_MAX]; > - /* XP */ > - u8 dtc; > - }; > + u8 dtm; > + s8 dtc; > + /* DN/HN-F/CXHA */ > + struct { > + u8 val : 4; > + u8 count : 4; > + } occupid[SEL_MAX]; > union { > u8 event[4]; > __le32 event_sel; > @@ -540,12 +537,12 @@ static int arm_cmn_map_show(struct seq_file *s, void *data) > > seq_puts(s, "\n |"); > for (x = 0; x < cmn->mesh_x; x++) { > - u8 dtc = cmn->xps[xp_base + x].dtc; > + s8 dtc = cmn->xps[xp_base + x].dtc; > > - if (dtc & (dtc - 1)) > + if (dtc < 0) > seq_puts(s, " DTC ?? |"); > else > - seq_printf(s, " DTC %ld |", __ffs(dtc)); > + seq_printf(s, " DTC %d |", dtc); > } > seq_puts(s, "\n |"); > for (x = 0; x < cmn->mesh_x; x++) > @@ -589,8 +586,7 @@ static void arm_cmn_debugfs_init(struct arm_cmn *cmn, int id) {} > struct arm_cmn_hw_event { > struct arm_cmn_node *dn; > u64 dtm_idx[4]; > - unsigned int dtc_idx; > - u8 dtcs_used; > + s8 dtc_idx[CMN_MAX_DTCS]; > u8 num_dns; > u8 dtm_offset; > bool wide_sel; > @@ -600,6 +596,10 @@ struct arm_cmn_hw_event { > #define for_each_hw_dn(hw, dn, i) \ > for (i = 0, dn = hw->dn; i < hw->num_dns; i++, dn++) > > +/* @i is the DTC number, @idx is the counter index on that DTC */ > +#define for_each_hw_dtc_idx(hw, i, idx) \ > + for (int i = 0, idx; i < CMN_MAX_DTCS; i++) if ((idx = hw->dtc_idx[i]) >= 0)
This macro is pretty hideous ;) The kbuild robot complained as well, but given that it's internal to the driver and it does make the callsites quite a bit simpler, I'm inclined to stick with it for now. At least, I couldn't come up with something else which was just as succinct.
Will
| |