* [PATCH 1/2] platform/x86/intel/pmc: Do not decode the summary rows in ltr_show
@ 2026-11-21 9:06 Paul Menzel
2026-11-21 9:06 ` [PATCH 2/2] platform/x86/intel/pmc: Fix Tiger Lake LTR ignore maximum to 22 Paul Menzel
0 siblings, 1 reply; 2+ messages in thread
From: Paul Menzel @ 2026-11-21 9:06 UTC (permalink / raw)
To: Rajneesh Bhardwaj, Xi Pardee, Hans de Goede, Ilpo Järvinen,
Andy Shevchenko
Cc: Paul Menzel, Rajneesh Bhardwaj, platform-driver-x86, linux-kernel
On an Intel Tiger Lake laptop Dell XPS 13 9310, the first read of
`/sys/kernel/debug/pmc_core/ltr_show` after loading the driver logs:
intel_pmc_core: Invalid LTR scale factor 7 (only 0-5 are valid per PCIe spec)
The value comes from the last row, AGGREGATED_SYSTEM (CNP_PMC_LTR_CUR_ASLT,
0x1b54). In 120 reads over 30 seconds, it read 0x7fffdff, 0x7fffc1f or
0x7fffc23 (117 times), and 0x40201ff or 0x402001f (3 times); another read
gave 0x7ffffff. Decoded as an LTR message, the snoop half of the first
group has the reserved bits 14:13 set and scale 7, and the second group is
CURRENT_PLATFORM (0x40201) shifted left by eight with the low byte filled.
So these two PMC registers do not seem to hold the PCIe LTR payload the
per-IP rows carry, and decoding them only produces the warning (the decoded
value is 0 anyway).
All ltr_show maps end with CURRENT_PLATFORM and AGGREGATED_SYSTEM, which
pmc_core_send_ltr_ignore() already relies on. Print only the raw value of
these two rows, as for rows without a requirement bit set, and keep the
warning for the per-IP rows.
With the change, 200 reads with AGGREGATED_SYSTEM bit 15 set in 199 of them
log nothing, and the CNV row is still decoded (0x89fe89fe, 522240 ns).
Fixes: 8122e7cd806d ("platform/x86: intel_pmc_core: Decode Snoop / Non Snoop LTR")
Signed-off-by: Paul Menzel <pmenzel@molgen.mpg.de>
---
Notes:
Claude Code (claude-opus-5-5) helped analyse the register values, write
the patch, and test it on the Dell XPS 13 9310.
drivers/platform/x86/intel/pmc/core.c | 17 +++++++++++++++--
1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/drivers/platform/x86/intel/pmc/core.c b/drivers/platform/x86/intel/pmc/core.c
index 71be313bdf0e1..e3f7d6bc8ce47 100644
--- a/drivers/platform/x86/intel/pmc/core.c
+++ b/drivers/platform/x86/intel/pmc/core.c
@@ -631,6 +631,18 @@ static u32 convert_ltr_scale(u32 val)
return 1U << (5 * val);
}
+/*
+ * Every ltr_show map ends with CURRENT_PLATFORM and AGGREGATED_SYSTEM. Unlike
+ * the per-IP rows, these PMC registers do not hold an LTR message payload as
+ * defined by PCIe; on Tiger Lake, for example, AGGREGATED_SYSTEM has the
+ * reserved bits 14:13 set and reads 0x7fffdff. Do not decode them.
+ */
+static bool pmc_core_ltr_is_summary(const struct pmc_bit_map *map,
+ unsigned int index)
+{
+ return !map[index + 1].name || !map[index + 2].name;
+}
+
static int pmc_core_ltr_show(struct seq_file *s, void *unused)
{
struct pmc_dev *pmcdev = s->private;
@@ -654,6 +666,7 @@ static int pmc_core_ltr_show(struct seq_file *s, void *unused)
map = pmc->map->ltr_show_sts;
for (index = 0; map[index].name; index++) {
bool ltr_ign_data;
+ bool summary = pmc_core_ltr_is_summary(map, index);
if (index > pmc->map->ltr_ignore_max)
ltr_ign_data = false;
@@ -666,12 +679,12 @@ static int pmc_core_ltr_show(struct seq_file *s, void *unused)
snoop_ltr = ltr_raw_data & ~MTPMC_MASK;
nonsnoop_ltr = (ltr_raw_data >> 0x10) & ~MTPMC_MASK;
- if (FIELD_GET(LTR_REQ_NONSNOOP, ltr_raw_data)) {
+ if (!summary && FIELD_GET(LTR_REQ_NONSNOOP, ltr_raw_data)) {
scale = FIELD_GET(LTR_DECODED_SCALE, nonsnoop_ltr);
val = FIELD_GET(LTR_DECODED_VAL, nonsnoop_ltr);
decoded_non_snoop_ltr = val * convert_ltr_scale(scale);
}
- if (FIELD_GET(LTR_REQ_SNOOP, ltr_raw_data)) {
+ if (!summary && FIELD_GET(LTR_REQ_SNOOP, ltr_raw_data)) {
scale = FIELD_GET(LTR_DECODED_SCALE, snoop_ltr);
val = FIELD_GET(LTR_DECODED_VAL, snoop_ltr);
decoded_snoop_ltr = val * convert_ltr_scale(scale);
--
2.54.0
^ permalink raw reply [flat|nested] 2+ messages in thread
* [PATCH 2/2] platform/x86/intel/pmc: Fix Tiger Lake LTR ignore maximum to 22
2026-11-21 9:06 [PATCH 1/2] platform/x86/intel/pmc: Do not decode the summary rows in ltr_show Paul Menzel
@ 2026-11-21 9:06 ` Paul Menzel
0 siblings, 0 replies; 2+ messages in thread
From: Paul Menzel @ 2026-11-21 9:06 UTC (permalink / raw)
To: Rajneesh Bhardwaj, Xi Pardee, Hans de Goede, Ilpo Järvinen,
David E. Box, Gayatri Kammela
Cc: Paul Menzel, platform-driver-x86, linux-kernel
ltr_ignore_max is the highest index in ltr_show_sts that has a bit in the
LTR_IGNORE register: pmc_core_send_ltr_ignore() accepts indices up to and
including it, pmc_core_ltr_ignore_all() sets GENMASK(ltr_ignore_max, 0),
and pmc_core_ltr_show() reads the LTR_IGNORE column for those rows. Cannon
Lake (19, EMMC), Ice Lake (20, WIGIG), Alder Lake and Meteor Lake set it to
the index of their last IP.
Tiger Lake shares cnp_ltr_show_map[], where THC1 is at index 22 and
CURRENT_PLATFORM at index 23, but sets TGL_NUM_IP_IGN_ALLOWED to 23, the
number of IPs. So debugfs accepts index 23, which is CURRENT_PLATFORM, a
row that “cannot be used for LTR_IGNORE”, and the hardware drops the bit.
On a Dell XPS 13 9310:
$ echo 23 | sudo tee /sys/kernel/debug/pmc_core/ltr_ignore
23
$ sudo grep CURRENT_PLATFORM /sys/kernel/debug/pmc_core/ltr_show
23 PMC0:CURRENT_PLATFORM LTR: RAW: 0x40201 […] LTR_IGNORE: 0
whereas index 22 (THC1) reads back LTR_IGNORE: 1.
Set TGL_NUM_IP_IGN_ALLOWED to 22. Now the write of 23 fails with -EINVAL,
like 24 before, and 22 still works.
Fixes: 43ef6c226a60 ("platform/x86: intel_pmc_core: Add LTR registers for Tiger Lake")
Signed-off-by: Paul Menzel <pmenzel@molgen.mpg.de>
---
Notes:
Claude Code (claude-opus-5-5) helped find the off-by-one, write the
patch, and test it on the Dell XPS 13 9310.
drivers/platform/x86/intel/pmc/core.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/platform/x86/intel/pmc/core.h b/drivers/platform/x86/intel/pmc/core.h
index e55164e5807fd..3edad5a1f72cd 100644
--- a/drivers/platform/x86/intel/pmc/core.h
+++ b/drivers/platform/x86/intel/pmc/core.h
@@ -214,7 +214,7 @@ enum ppfear_regs {
#define TGL_PMC_SLP_S0_RES_COUNTER_STEP 0x7A
#define TGL_PMC_LTR_THC0 0x1C04
#define TGL_PMC_LTR_THC1 0x1C08
-#define TGL_NUM_IP_IGN_ALLOWED 23
+#define TGL_NUM_IP_IGN_ALLOWED 22
#define TGL_PMC_LPM_RES_COUNTER_STEP_X2 61 /* 30.5us * 2 */
#define ADL_PMC_LTR_SPF 0x1C00
--
2.54.0
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2026-11-21 9:07 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-21 9:06 [PATCH 1/2] platform/x86/intel/pmc: Do not decode the summary rows in ltr_show Paul Menzel
2026-11-21 9:06 ` [PATCH 2/2] platform/x86/intel/pmc: Fix Tiger Lake LTR ignore maximum to 22 Paul Menzel
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®