From: Jonathan Matthew <jonathan@d14n.org>
Subject: ksmn(4) on 19h/1x devices
To: tech@openbsd.org
Cc: brynet@openbsd.org
Date: Thu, 21 May 2026 21:04:10 +1000

We've had some of these boxes for a while but I only noticed today that
ksmn doesn't support them.  The interesting part here is that the check
in ksmn_ccd_attach() is a bit too aggressive, it ensures there are fewer
than 12 potential ccd sensors, which happens to be how many these systems
can have.

With this cpu:
cpu0: AMD EPYC 9354P 32-Core Processor, 3250.01 MHz, 19-11-01, patch 0a101154

I now get some readings:
hw.sensors.ksmn0.temp0=96.12 degC (Tctl)
hw.sensors.ksmn0.temp1=40.00 degC (Tccd0)
hw.sensors.ksmn0.temp2=38.75 degC (Tccd1)
hw.sensors.ksmn0.temp3=39.62 degC (Tccd2)
hw.sensors.ksmn0.temp4=39.75 degC (Tccd3)
hw.sensors.ksmn0.temp5=39.00 degC (Tccd4)
hw.sensors.ksmn0.temp6=37.62 degC (Tccd5)
hw.sensors.ksmn0.temp7=39.25 degC (Tccd6)
hw.sensors.ksmn0.temp8=39.50 degC (Tccd7)

(plus 3 more copies because it has 4 pci roots)

ok?

Index: ksmn.c
===================================================================
RCS file: /cvs/src/sys/dev/pci/ksmn.c,v
diff -u -p -r1.10 ksmn.c
--- ksmn.c	7 Aug 2024 17:39:00 -0000	1.10
+++ ksmn.c	21 May 2026 06:22:17 -0000
@@ -105,6 +105,7 @@ static const struct pci_matchid ksmn_dev
 	{ PCI_VENDOR_AMD, PCI_PRODUCT_AMD_17_1X_RC },
 	{ PCI_VENDOR_AMD, PCI_PRODUCT_AMD_17_3X_RC },
 	{ PCI_VENDOR_AMD, PCI_PRODUCT_AMD_17_6X_RC },
+	{ PCI_VENDOR_AMD, PCI_PRODUCT_AMD_19_1X_RC },
 	{ PCI_VENDOR_AMD, PCI_PRODUCT_AMD_19_4X_RC },
 	{ PCI_VENDOR_AMD, PCI_PRODUCT_AMD_19_6X_RC },
 	{ PCI_VENDOR_AMD, PCI_PRODUCT_AMD_19_7X_RC },
@@ -215,7 +216,7 @@ ksmn_ccd_attach(struct ksmn_softc *sc, i
 	uint32_t reg;
 	int i;
 
-	KASSERT(nccd > 0 && nccd < nitems(sc->sc_ccd_sensor));
+	KASSERT(nccd > 0 && nccd <= nitems(sc->sc_ccd_sensor));
 
 	for (i = 0; i < nccd; i++) {
 		reg = ksmn_read_reg(sc, SMU_17H_CCD_THM(sc->sc_ccd_offset, i));