Skip to contentSkip to conversation

perf/x86/uncore: update support for client uncore IMC PMU

PATCH Superseded Stephane Eranian 3 participants 5 messages Sep 15, 2014 – Sep 16, 2014

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf/x86/uncore: update support for client uncore IMC PMU

This patch restructures the memory controller (IMC) uncore PMU support for client SNB/IVB/HSW processors. The main change is that it can now cope with more than one PCI device ID per processor model. There are many flavors of memory controllers for each processor. They have different PCI device ID, yet they behave the same w.r.t. the memory controller PMU that we are interested in.

The patch now supports two distinct memory controllers for IVB
processors: one for mobile, one for desktop.

---
 arch/x86/kernel/cpu/perf_event_intel_uncore_snb.c |   68 +++++++++++++++------
 include/linux/pci_ids.h                           |    1 +
 2 files changed, 51 insertions(+), 18 deletions(-)
arch/x86/kernel/cpu/perf_event_intel_uncore_snb.c +50 -18
index e0e934c..70d0887 100644
@@ -460,6 +460,10 @@ static const struct pci_device_id ivb_uncore_pci_ids[] = {
 		PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IVB_IMC),
 		.driver_data = UNCORE_PCI_DEV_DATA(SNB_PCI_UNCORE_IMC, 0),
 	},
+	{ /* IMC */
+		PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IVB_E3_IMC),
+		.driver_data = UNCORE_PCI_DEV_DATA(SNB_PCI_UNCORE_IMC, 0),
+	},
 	{ /* end: all zeroes */ },
 };
 
@@ -486,34 +490,62 @@ static struct pci_driver hsw_uncore_pci_driver = {
 	.id_table	= hsw_uncore_pci_ids,
 };
 
-int snb_uncore_pci_init(void)
+struct imc_uncore_pci_dev {
+	__u32 pci_id;
+	struct pci_driver *driver;
+};
+#define IMC_DEV(a, d) \
+	{ .pci_id = PCI_DEVICE_ID_INTEL_##a, .driver = (d) }
+
+static const struct imc_uncore_pci_dev desktop_imc_pci_ids[] = {
+	IMC_DEV(SNB_IMC, &snb_uncore_pci_driver),
+	IMC_DEV(IVB_IMC, &ivb_uncore_pci_driver),    /* 3rd Gen Core processor */
+	IMC_DEV(IVB_E3_IMC, &ivb_uncore_pci_driver), /* Xeon E3-1200 v2/3rd Gen Core processor */
+	IMC_DEV(HSW_IMC, &hsw_uncore_pci_driver),    /* 4th Gen Core Processor */
+	{  /* end marker */ }
+};
+
+
+#define for_each_imc_pci_id(x, t) \
+	for (x = (t); (x)->pci_id; x++)
+
+static struct pci_driver *imc_uncore_find_dev(void)
 {
-	int ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_SNB_IMC);
-	if (ret)
-		return ret;
-	uncore_pci_uncores = snb_pci_uncores;
-	uncore_pci_driver = &snb_uncore_pci_driver;
-	return 0;
+	const struct imc_uncore_pci_dev *p;
+	int ret;
+
+	for_each_imc_pci_id (p, desktop_imc_pci_ids) {
+		ret = snb_pci2phy_map_init(p->pci_id);
+		if (ret == 0)
+			return p->driver;
+	}
+	return NULL;
 }
 
-int ivb_uncore_pci_init(void)
+static int imc_uncore_pci_init(void)
 {
-	int ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_IVB_IMC);
-	if (ret)
-		return ret;
+	struct pci_driver *imc_drv = imc_uncore_find_dev();
+	if (!imc_drv)
+		return -ENODEV;
+
 	uncore_pci_uncores = snb_pci_uncores;
-	uncore_pci_driver = &ivb_uncore_pci_driver;
+	uncore_pci_driver = imc_drv;
+
 	return 0;
 }
 
+int snb_uncore_pci_init(void)
+{
+	return imc_uncore_pci_init();
+}
+
+int ivb_uncore_pci_init(void)
+{
+	return imc_uncore_pci_init();
+}
 int hsw_uncore_pci_init(void)
 {
-	int ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_HSW_IMC);
-	if (ret)
-		return ret;
-	uncore_pci_uncores = snb_pci_uncores;
-	uncore_pci_driver = &hsw_uncore_pci_driver;
-	return 0;
+	return imc_uncore_pci_init();
 }
 
 /* end of Sandy Bridge uncore support */
include/linux/pci_ids.h +1 -0
index 6ed0bb7..3102b7e 100644
@@ -2536,6 +2536,7 @@
 #define PCI_DEVICE_ID_INTEL_EESSC	0x0008
 #define PCI_DEVICE_ID_INTEL_SNB_IMC	0x0100
 #define PCI_DEVICE_ID_INTEL_IVB_IMC	0x0154
+#define PCI_DEVICE_ID_INTEL_IVB_E3_IMC	0x0150
 #define PCI_DEVICE_ID_INTEL_HSW_IMC	0x0c00
 #define PCI_DEVICE_ID_INTEL_PXHD_0	0x0320
 #define PCI_DEVICE_ID_INTEL_PXHD_1	0x0321
Signature · 1 line
1.7.10.4
#1 lore raw
Quoted 11 lines · Stephane Eranian

This patch restructures the memory controller (IMC) uncore PMU support for client SNB/IVB/HSW processors. The main change is that it can now cope with more than one PCI device ID per processor model. There are many flavors of memory controllers for each processor. They have different PCI device ID, yet they behave the same w.r.t. the memory controller PMU that we are interested in.

The patch now supports two distinct memory controllers for IVB
processors: one for mobile, one for desktop.

I haven't investigated yet, but on my Lenovo ULT Haswell laptop the IMC PCI-ID doesn't exist, so the driver doesn't initialize.

So either it's the BIOS disabling it, or it's a different ID there too.

BTW any reason you didn't enable the CBOX/ARB PMUs originally for HSW? At least ARB events are useful for some things.

-Andi

#2 lore raw
Quoted 11 lines · Andi Kleen
Quoted 9 lines · Stephane Eranian

This patch restructures the memory controller (IMC) uncore PMU support for client SNB/IVB/HSW processors. The main change is that it can now cope with more than one PCI device ID per processor model. There are many flavors of memory controllers for each processor. They have different PCI device ID, yet they behave the same w.r.t. the memory controller PMU that we are interested in.

Quoted 6 lines
Quoted 2 lines

The patch now supports two distinct memory controllers for IVB
processors: one for mobile, one for desktop.

I haven't investigated yet, but on my Lenovo ULT Haswell laptop the IMC PCI-ID doesn't exist, so the driver doesn't initialize.

This updates makes it easier to add new PCI-IDs. I don't have all of them. I don't have all the HW combinations.

Quoted 4 lines

So either it's the BIOS disabling it, or it's a different ID there too.

I bet it is a different ID.
Now, this patch still has a PCI-ID table per-processor and not a global table. I debated this for a while. But I don't know if a PCI-ID can be used for something
else on another platform.

Quoted 4 lines

BTW any reason you didn't enable the CBOX/ARB PMUs originally for HSW? At least ARB events are useful for some things.

I have never looked at the ARB events. If they work like on SNB/IVB, then they should be easy to enable.

Quoted 2 lines

-Andi

#3 lore raw
In reply to Andi Kleen
Quoted 11 lines · Andi Kleen
Quoted 9 lines · Stephane Eranian

This patch restructures the memory controller (IMC) uncore PMU support for client SNB/IVB/HSW processors. The main change is that it can now cope with more than one PCI device ID per processor model. There are many flavors of memory controllers for each processor. They have different PCI device ID, yet they behave the same w.r.t. the memory controller PMU that we are interested in.

Quoted 6 lines
Quoted 2 lines

The patch now supports two distinct memory controllers for IVB
processors: one for mobile, one for desktop.

I haven't investigated yet, but on my Lenovo ULT Haswell laptop the IMC PCI-ID doesn't exist, so the driver doesn't initialize.

Send me the PCI-ID for this memory controller and I will add it to V2 of the patch.
Thanks.

#4 lore raw
In reply to Stephane Eranian
Quoted 9 lines · Stephane Eranian

This patch restructures the memory controller (IMC) uncore PMU support for client SNB/IVB/HSW processors. The main change is that it can now cope with more than one PCI device ID per processor model. There are many flavors of memory controllers for each processor. They have different PCI device ID, yet they behave the same w.r.t. the memory controller PMU that we are interested in.

Quoted 8 lines

The patch now supports two distinct memory controllers for IVB
processors: one for mobile, one for desktop.

---
 arch/x86/kernel/cpu/perf_event_intel_uncore_snb.c |   68 +++++++++++++++------
 include/linux/pci_ids.h                           |    1 +
 2 files changed, 51 insertions(+), 18 deletions(-)

Please run patches through scripts/checkpatch.pl before submission, in case it points out legit problems. Checkpatch emits 6 warnings for your patch, the first four are bogus, the last two are worth addressing:

WARNING: space prohibited between function name and open parenthesis '('
#77: FILE: arch/x86/kernel/cpu/perf_event_intel_uncore_snb.c:517:
+ for_each_imc_pci_id (p, desktop_imc_pci_ids) {

WARNING: Missing a blank line after declarations
#92: FILE: arch/x86/kernel/cpu/perf_event_intel_uncore_snb.c:528:
+ struct pci_driver *imc_drv = imc_uncore_find_dev();
+ if (!imc_drv)

Thanks,

Ingo