Skip to contentSkip to conversation

intel_pmc: Add telemetry API to read counters

PATCH In review David E. Box 3 participants 28 messages Nov 29, 2023 – Dec 4, 2023

Subsystems (4)
lore
Share & tools
lore

David E. Box

author maintainer
#0 lore raw

v6 cover intel_pmc: Add telemetry API to read counters

On newer Intel silicon, more IP counters are being added in Intel Platform Monitoring Technology (PMT) telemetry spaces hosted in MMIO. There is a need for the intel_pmc_core driver and other drivers to access PMT hosted telemetry in the kernel using an API. This patchset adds driver APIs to allow registering and reading telemetry entries. It makes changes to the intel_pmc_core driver to use these interfaces to access the low power mode counters that are now exclusively available from PMT.

David E. Box (15):
platform/x86/intel/vsec: Fix xa_alloc memory leak platform/x86/intel/vsec: Remove unnecessary return platform/x86/intel/vsec: Move structures to header platform/x86/intel/vsec: remove platform_info from vsec device
structure
platform/x86/intel/vsec: Use cleanup.h platform/x86/intel/vsec: Assign auxdev parent by argument platform/x86/intel/vsec: Add base address field platform/x86/intel/pmt: Add header to struct intel_pmt_entry platform/x86/intel/pmt: telemetry: Export API to read telemetry platform/x86/intel/pmc: Allow pmc_core_ssram_init to fail platform/x86/intel/pmc: Cleanup SSRAM discovery platform/x86/intel/pmc/mtl: Use return value from
pmc_core_ssram_init()
platform/x86/intel/pmc: Find and register PMC telemetry entries platform/x86/intel/pmc: Add debug attribute for Die C6 counter platform/x86/intel/pmc: Show Die C6 counter on Meteor Lake

Gayatri Kammela (1):
platform/x86/intel/vsec: Add intel_vsec_register

Rajvi Jingar (1):
platform/x86/intel/pmc: Display LPM requirements for multiple PMCs

Xi Pardee (3):
platform/x86:intel/pmc: Call pmc_get_low_power_modes from platform
init
platform/x86/intel/pmc: Retrieve LPM information using Intel PMT platform/x86/intel/pmc: Read low power mode requirements for MTL-M and
MTL-P

 drivers/platform/x86/intel/pmc/Kconfig      |   1 +
 drivers/platform/x86/intel/pmc/adl.c        |   2 +
 drivers/platform/x86/intel/pmc/cnp.c        |   2 +
 drivers/platform/x86/intel/pmc/core.c       | 185 +++++++++-----
 drivers/platform/x86/intel/pmc/core.h       |  10 +-
 drivers/platform/x86/intel/pmc/core_ssram.c | 265 +++++++++++++++++---
 drivers/platform/x86/intel/pmc/icl.c        |  10 +-
 drivers/platform/x86/intel/pmc/mtl.c        |  87 ++++++-
 drivers/platform/x86/intel/pmc/spt.c        |  10 +-
 drivers/platform/x86/intel/pmc/tgl.c        |   1 +
 drivers/platform/x86/intel/pmt/class.c      |  43 +++-
 drivers/platform/x86/intel/pmt/class.h      |  30 ++-
 drivers/platform/x86/intel/pmt/crashlog.c   |   2 +-
 drivers/platform/x86/intel/pmt/telemetry.c  | 193 +++++++++++++-
 drivers/platform/x86/intel/pmt/telemetry.h  | 126 ++++++++++
 drivers/platform/x86/intel/vsec.c           | 131 +++++-----
 drivers/platform/x86/intel/vsec.h           |  45 +++-
 17 files changed, 939 insertions(+), 204 deletions(-)
 create mode 100644 drivers/platform/x86/intel/pmt/telemetry.h

base-commit: b85ea95d086471afb4ad062012a4d73cd328fa86

Signature · 1 line
2.34.1
author maintainer
#1 lore raw

PATCH v6 1/20 platform/x86/intel/vsec: Fix xa_alloc memory leak

Commit 936874b77dd0 ("platform/x86/intel/vsec: Add PCI error recovery support to Intel PMT") added an xarray to track the list of vsec devices to be recovered after a PCI error. But it did not provide cleanup for the list leading to a memory leak that was caught by kmemleak. Do xa_alloc() before devm_add_action_or_reset() so that the list may be cleaned up with xa_erase() in the release function.

Fixes 936874b77dd0 ("platform/x86/intel/vsec: Add PCI error recovery support to Intel PMT")

---

V6 - Move xa_alloc() before ida_alloc() to reduce mutex use during error
recovery.
- Fix return value after id_alloc() fail
- Add Fixes tag
- Add more detail to changelog

V5 - New patch

 drivers/platform/x86/intel/vsec.c | 24 ++++++++++++++----------
 drivers/platform/x86/intel/vsec.h |  1 +
 2 files changed, 15 insertions(+), 10 deletions(-)
drivers/platform/x86/intel/vsec.c +14 -10
index c1f9e4471b28..2d568466b4e2 100644
@@ -120,6 +120,8 @@ static void intel_vsec_dev_release(struct device *dev)
 {
 	struct intel_vsec_device *intel_vsec_dev = dev_to_ivdev(dev);
 
+	xa_erase(&auxdev_array, intel_vsec_dev->id);
+
 	mutex_lock(&vsec_ida_lock);
 	ida_free(intel_vsec_dev->ida, intel_vsec_dev->auxdev.id);
 	mutex_unlock(&vsec_ida_lock);
@@ -135,19 +137,27 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
 	struct auxiliary_device *auxdev = &intel_vsec_dev->auxdev;
 	int ret, id;
 
-	mutex_lock(&vsec_ida_lock);
-	ret = ida_alloc(intel_vsec_dev->ida, GFP_KERNEL);
-	mutex_unlock(&vsec_ida_lock);
+	ret = xa_alloc(&auxdev_array, &intel_vsec_dev->id, intel_vsec_dev,
+		       PMT_XA_LIMIT, GFP_KERNEL);
 	if (ret < 0) {
 		kfree(intel_vsec_dev->resource);
 		kfree(intel_vsec_dev);
 		return ret;
 	}
 
+	mutex_lock(&vsec_ida_lock);
+	id = ida_alloc(intel_vsec_dev->ida, GFP_KERNEL);
+	mutex_unlock(&vsec_ida_lock);
+	if (id < 0) {
+		kfree(intel_vsec_dev->resource);
+		kfree(intel_vsec_dev);
+		return id;
+	}
+
 	if (!parent)
 		parent = &pdev->dev;
 
-	auxdev->id = ret;
+	auxdev->id = id;
 	auxdev->name = name;
 	auxdev->dev.parent = parent;
 	auxdev->dev.release = intel_vsec_dev_release;
@@ -169,12 +179,6 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
 	if (ret < 0)
 		return ret;
 
-	/* Add auxdev to list */
-	ret = xa_alloc(&auxdev_array, &id, intel_vsec_dev, PMT_XA_LIMIT,
-		       GFP_KERNEL);
-	if (ret)
-		return ret;
-
 	return 0;
 }
 EXPORT_SYMBOL_NS_GPL(intel_vsec_add_aux, INTEL_VSEC);
drivers/platform/x86/intel/vsec.h +1 -0
index 0fd042c171ba..0a6201b4a0e9 100644
@@ -45,6 +45,7 @@ struct intel_vsec_device {
 	struct ida *ida;
 	struct intel_vsec_platform_info *info;
 	int num_resources;
+	int id; /* xa */
 	void *priv_data;
 	size_t priv_data_size;
 };
Signature · 1 line
2.34.1
maintainer Reviewed-by on v6 1/20
#2 lore raw

Re: [PATCH V6 01/20] platform/x86/intel/vsec: Fix xa_alloc memory leak

Quoted 48 lines · David E. Box

Commit 936874b77dd0 ("platform/x86/intel/vsec: Add PCI error recovery support to Intel PMT") added an xarray to track the list of vsec devices to be recovered after a PCI error. But it did not provide cleanup for the list leading to a memory leak that was caught by kmemleak. Do xa_alloc() before devm_add_action_or_reset() so that the list may be cleaned up with xa_erase() in the release function.

Fixes 936874b77dd0 ("platform/x86/intel/vsec: Add PCI error recovery support to Intel PMT")

---

V6 - Move xa_alloc() before ida_alloc() to reduce mutex use during error
recovery.
- Fix return value after id_alloc() fail
- Add Fixes tag
- Add more detail to changelog

V5 - New patch

 drivers/platform/x86/intel/vsec.c | 24 ++++++++++++++----------
 drivers/platform/x86/intel/vsec.h |  1 +
 2 files changed, 15 insertions(+), 10 deletions(-)
drivers/platform/x86/intel/vsec.c +4 -3
index c1f9e4471b28..2d568466b4e2 100644
@@ -120,6 +120,8 @@ static void intel_vsec_dev_release(struct device *dev)
 {
 	struct intel_vsec_device *intel_vsec_dev = dev_to_ivdev(dev);
 
+	xa_erase(&auxdev_array, intel_vsec_dev->id);
+
 	mutex_lock(&vsec_ida_lock);
 	ida_free(intel_vsec_dev->ida, intel_vsec_dev->auxdev.id);
 	mutex_unlock(&vsec_ida_lock);
@@ -135,19 +137,27 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
 	struct auxiliary_device *auxdev = &intel_vsec_dev->auxdev;
 	int ret, id;
 
-	mutex_lock(&vsec_ida_lock);
-	ret = ida_alloc(intel_vsec_dev->ida, GFP_KERNEL);
-	mutex_unlock(&vsec_ida_lock);
+	ret = xa_alloc(&auxdev_array, &intel_vsec_dev->id, intel_vsec_dev,
+		       PMT_XA_LIMIT, GFP_KERNEL);
 	if (ret < 0) {
 		kfree(intel_vsec_dev->resource);
 		kfree(intel_vsec_dev);

drivers/platform/x86/intel/vsec.c int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent, · line 154

Quoted 10 lines
drivers/platform/x86/intel/vsec.c +7 -0
 		return ret;
 	}
 
+	mutex_lock(&vsec_ida_lock);
+	id = ida_alloc(intel_vsec_dev->ida, GFP_KERNEL);
+	mutex_unlock(&vsec_ida_lock);
+	if (id < 0) {
+		kfree(intel_vsec_dev->resource);
+		kfree(intel_vsec_dev);
+		return id;

Thanks, this looks much better this way around but it is missing xa_alloc() rollback now that the order is reversed.

Once that is fixed,

Signature · 1 line
 i.
#3 lore raw

Re: [PATCH V6 01/20] platform/x86/intel/vsec: Fix xa_alloc memory leak

Hi,

Quoted 64 lines · Ilpo Järvinen
Quoted 48 lines · David E. Box

Commit 936874b77dd0 ("platform/x86/intel/vsec: Add PCI error recovery support to Intel PMT") added an xarray to track the list of vsec devices to be recovered after a PCI error. But it did not provide cleanup for the list leading to a memory leak that was caught by kmemleak. Do xa_alloc() before devm_add_action_or_reset() so that the list may be cleaned up with xa_erase() in the release function.

Fixes 936874b77dd0 ("platform/x86/intel/vsec: Add PCI error recovery support to Intel PMT")

---

V6 - Move xa_alloc() before ida_alloc() to reduce mutex use during error
recovery.
- Fix return value after id_alloc() fail
- Add Fixes tag
- Add more detail to changelog

V5 - New patch

 drivers/platform/x86/intel/vsec.c | 24 ++++++++++++++----------
 drivers/platform/x86/intel/vsec.h |  1 +
 2 files changed, 15 insertions(+), 10 deletions(-)
drivers/platform/x86/intel/vsec.c +4 -3
index c1f9e4471b28..2d568466b4e2 100644
@@ -120,6 +120,8 @@ static void intel_vsec_dev_release(struct device *dev)
 {
 	struct intel_vsec_device *intel_vsec_dev = dev_to_ivdev(dev);
 
+	xa_erase(&auxdev_array, intel_vsec_dev->id);
+
 	mutex_lock(&vsec_ida_lock);
 	ida_free(intel_vsec_dev->ida, intel_vsec_dev->auxdev.id);
 	mutex_unlock(&vsec_ida_lock);
@@ -135,19 +137,27 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
 	struct auxiliary_device *auxdev = &intel_vsec_dev->auxdev;
 	int ret, id;
 
-	mutex_lock(&vsec_ida_lock);
-	ret = ida_alloc(intel_vsec_dev->ida, GFP_KERNEL);
-	mutex_unlock(&vsec_ida_lock);
+	ret = xa_alloc(&auxdev_array, &intel_vsec_dev->id, intel_vsec_dev,
+		       PMT_XA_LIMIT, GFP_KERNEL);
 	if (ret < 0) {
 		kfree(intel_vsec_dev->resource);
 		kfree(intel_vsec_dev);

drivers/platform/x86/intel/vsec.c int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent, · line 154

Quoted 10 lines
drivers/platform/x86/intel/vsec.c +7 -0
 		return ret;
 	}
 
+	mutex_lock(&vsec_ida_lock);
+	id = ida_alloc(intel_vsec_dev->ida, GFP_KERNEL);
+	mutex_unlock(&vsec_ida_lock);
+	if (id < 0) {
+		kfree(intel_vsec_dev->resource);
+		kfree(intel_vsec_dev);
+		return id;

Thanks, this looks much better this way around but it is missing xa_alloc() rollback now that the order is reversed.

Quoted 3 lines

Once that is fixed,

I have fixed this up, adding the missing:

	xa_erase(&auxdev_array, intel_vsec_dev->id);

to this error-exit path while merging this.

While looking into this I did find one other thing which worries me (different issue, needs a separate fix):

intel_vsec_pci_slot_reset() uses

                devm_release_action(&pdev->dev, intel_vsec_remove_aux,
                                    &intel_vsec_dev->auxdev);

and seems to assume that after this intel_vsec_remove_aux() has run for the auxdev-s. *But this is not the case*

devm_release_action() only removes the action from the list of devres resources tied to the parent PCI device.

It does *NOT* call the registered action function, so intel_vsec_remove_aux() is NOT called here.

And then on re-probing the device as is done in intel_vsec_pci_slot_reset() a second set of aux devices with the same parent will be created AFAICT.

So it seems that this also needs an explicit
intel_vsec_remove_aux() call for each auxdev!

###

This makes me wonder if the PCI error handling here and specifically the reset code was ever tested ?

###

Note that simply forcing a reprobe using device_reprobe() will cause all the aux-devices to also get removed through the action on driver-unbind without ever needing the auxdev_array at all!

I guess that you want the removal to happen before the pci_restore_state(pdev) state though, so that simply relying on the removal on driver unbind is not an option ?

Regards,

Hans

author maintainer
#4 lore raw

Re: [PATCH V6 01/20] platform/x86/intel/vsec: Fix xa_alloc memory leak

Hi Hans,

Quoted 75 lines · Hans de Goede

Hi,

Quoted 72 lines · Ilpo Järvinen
Quoted 56 lines · David E. Box

Commit 936874b77dd0 ("platform/x86/intel/vsec: Add PCI error recovery support to Intel PMT") added an xarray to track the list of vsec devices to
be recovered after a PCI error. But it did not provide cleanup for the list
leading to a memory leak that was caught by kmemleak.  Do xa_alloc() before
devm_add_action_or_reset() so that the list may be cleaned up with xa_erase() in the release function.

Fixes 936874b77dd0 ("platform/x86/intel/vsec: Add PCI error recovery

support to Intel PMT")

---

V6 - Move xa_alloc() before ida_alloc() to reduce mutex use during error recovery.
   - Fix return value after id_alloc() fail
   - Add Fixes tag
   - Add more detail to changelog

V5 - New patch

 drivers/platform/x86/intel/vsec.c | 24 ++++++++++++++---------- drivers/platform/x86/intel/vsec.h |  1 +

 2 files changed, 15 insertions(+), 10 deletions(-)
drivers/platform/x86/intel/vsec.c +0 -0
index c1f9e4471b28..2d568466b4e2 100644
@@ -120,6 +120,8 @@ static void intel_vsec_dev_release(struct device *dev)
  {
         struct intel_vsec_device *intel_vsec_dev = dev_to_ivdev(dev);


+       xa_erase(&auxdev_array, intel_vsec_dev->id);
+
        mutex_lock(&vsec_ida_lock); ida_free(intel_vsec_dev->ida, intel_vsec_dev->auxdev.id); mutex_unlock(&vsec_ida_lock);

drivers/platform/x86/intel/vsec.c +0 -0
@@ -135,19 +137,27 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct
 device *parent,
         struct auxiliary_device *auxdev = &intel_vsec_dev->auxdev;
         int ret, id;


-       mutex_lock(&vsec_ida_lock);
-       ret = ida_alloc(intel_vsec_dev->ida, GFP_KERNEL);
-       mutex_unlock(&vsec_ida_lock);
+       ret = xa_alloc(&auxdev_array, &intel_vsec_dev->id, intel_vsec_dev, +                      PMT_XA_LIMIT, GFP_KERNEL);
        if (ret < 0) {
                kfree(intel_vsec_dev->resource); kfree(intel_vsec_dev);
                return ret;
        }

Quoted 8 lines


+       mutex_lock(&vsec_ida_lock);
+       id = ida_alloc(intel_vsec_dev->ida, GFP_KERNEL); +       mutex_unlock(&vsec_ida_lock);
+       if (id < 0) {
+               kfree(intel_vsec_dev->resource); +               kfree(intel_vsec_dev);
+               return id;

Thanks, this looks much better this way around but it is missing xa_alloc() rollback now that the order is reversed.

Once that is fixed,

Quoted 7 lines
Quoted 1 line

I have fixed this up, adding the missing:

        xa_erase(&auxdev_array, intel_vsec_dev->id);

to this error-exit path while merging this.

Thanks. Does that include the rest of the series which was all reviewed by Ilpo?

Quoted 22 lines

While looking into this I did find one other thing which worries me (different issue, needs a separate fix):

intel_vsec_pci_slot_reset() uses

                devm_release_action(&pdev->dev, intel_vsec_remove_aux, &intel_vsec_dev->auxdev);

and seems to assume that after this intel_vsec_remove_aux() has run for the auxdev-s. *But this is not the case*

devm_release_action() only removes the action from the list of devres resources tied to the parent PCI device.

It does *NOT* call the registered action function, so intel_vsec_remove_aux() is NOT called here.

And then on re-probing the device as is done in intel_vsec_pci_slot_reset() a second set of aux devices with the same parent will be created AFAICT.

Quoted 7 lines

So it seems that this also needs an explicit
intel_vsec_remove_aux() call for each auxdev!

###

This makes me wonder if the PCI error handling here and specifically the reset code was ever tested ?

I recall the code was tested using error injection to cause the slot reset to occur and reprobe was confirmed. I'll have to find out the specific test so that we can check it again with the proposed fix and ensure no leaks.

Quoted 7 lines

###

Note that simply forcing a reprobe using device_reprobe() will cause all the aux-devices to also get removed through the action on driver-unbind without ever needing the auxdev_array at all!

Okay. That would be a lot simpler.

Quoted 5 lines

I guess that you want the removal to happen before the pci_restore_state(pdev) state though, so that simply relying on the removal on driver unbind is not an option ?

I'm not sure this is needed given the simplicity of the device. The array was added only to track the list of devices and reprobe the one that was reset. We'll look at changing this to do driver_reprobe() instead. Thanks.

David

#5 lore raw

Re: [PATCH V6 01/20] platform/x86/intel/vsec: Fix xa_alloc memory leak

Hi,

Quoted 80 lines · David E. Box

Hi Hans,

Quoted 77 lines · Hans de Goede

Hi,

Quoted 73 lines · Ilpo Järvinen
Quoted 56 lines · David E. Box

Commit 936874b77dd0 ("platform/x86/intel/vsec: Add PCI error recovery support to Intel PMT") added an xarray to track the list of vsec devices to
be recovered after a PCI error. But it did not provide cleanup for the list
leading to a memory leak that was caught by kmemleak.  Do xa_alloc() before
devm_add_action_or_reset() so that the list may be cleaned up with xa_erase() in the release function.

Fixes 936874b77dd0 ("platform/x86/intel/vsec: Add PCI error recovery

support to Intel PMT")

---

V6 - Move xa_alloc() before ida_alloc() to reduce mutex use during error recovery.
   - Fix return value after id_alloc() fail
   - Add Fixes tag
   - Add more detail to changelog

V5 - New patch

 drivers/platform/x86/intel/vsec.c | 24 ++++++++++++++---------- drivers/platform/x86/intel/vsec.h |  1 +

 2 files changed, 15 insertions(+), 10 deletions(-)
drivers/platform/x86/intel/vsec.c +0 -0
index c1f9e4471b28..2d568466b4e2 100644
@@ -120,6 +120,8 @@ static void intel_vsec_dev_release(struct device *dev)
  {
         struct intel_vsec_device *intel_vsec_dev = dev_to_ivdev(dev);


+       xa_erase(&auxdev_array, intel_vsec_dev->id);
+
        mutex_lock(&vsec_ida_lock); ida_free(intel_vsec_dev->ida, intel_vsec_dev->auxdev.id); mutex_unlock(&vsec_ida_lock);

drivers/platform/x86/intel/vsec.c +0 -0
@@ -135,19 +137,27 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct
 device *parent,
         struct auxiliary_device *auxdev = &intel_vsec_dev->auxdev;
         int ret, id;


-       mutex_lock(&vsec_ida_lock);
-       ret = ida_alloc(intel_vsec_dev->ida, GFP_KERNEL);
-       mutex_unlock(&vsec_ida_lock);
+       ret = xa_alloc(&auxdev_array, &intel_vsec_dev->id, intel_vsec_dev, +                      PMT_XA_LIMIT, GFP_KERNEL);
        if (ret < 0) {
                kfree(intel_vsec_dev->resource); kfree(intel_vsec_dev);
                return ret;
        }

Quoted 8 lines


+       mutex_lock(&vsec_ida_lock);
+       id = ida_alloc(intel_vsec_dev->ida, GFP_KERNEL); +       mutex_unlock(&vsec_ida_lock);
+       if (id < 0) {
+               kfree(intel_vsec_dev->resource); +               kfree(intel_vsec_dev);
+               return id;

Thanks, this looks much better this way around but it is missing xa_alloc() rollback now that the order is reversed.

Once that is fixed,

Quoted 7 lines
Quoted 5 lines

I have fixed this up, adding the missing:

        xa_erase(&auxdev_array, intel_vsec_dev->id);

to this error-exit path while merging this.

Thanks. Does that include the rest of the series which was all reviewed by Ilpo?

Yes the entire series has been merged into the pdx86/review-hans branch now.

Regards,

Hans

author maintainer
#6 lore raw

PATCH v6 2/20 platform/x86/intel/vsec: Remove unnecessary return

In reply to David E. Box

In intel_vsec_add_aux(), just return from the last call to devm_add_action_or_reset() instead of checking its return value.

---
V6 - New patch

 drivers/platform/x86/intel/vsec.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)
drivers/platform/x86/intel/vsec.c +1 -5
index 2d568466b4e2..340562ae2041 100644
@@ -174,12 +174,8 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
 		return ret;
 	}
 
-	ret = devm_add_action_or_reset(parent, intel_vsec_remove_aux,
+	return devm_add_action_or_reset(parent, intel_vsec_remove_aux,
 				       auxdev);
-	if (ret < 0)
-		return ret;
-
-	return 0;
 }
 EXPORT_SYMBOL_NS_GPL(intel_vsec_add_aux, INTEL_VSEC);
 
Signature · 1 line
2.34.1
maintainer Reviewed-by on v6 2/20
#7 lore raw

Re: [PATCH V6 02/20] platform/x86/intel/vsec: Remove unnecessary return

Quoted 18 lines · David E. Box

In intel_vsec_add_aux(), just return from the last call to devm_add_action_or_reset() instead of checking its return value.

---
V6 - New patch

 drivers/platform/x86/intel/vsec.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)
drivers/platform/x86/intel/vsec.c +0 -0
index 2d568466b4e2..340562ae2041 100644
@@ -174,12 +174,8 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
 		return ret;
 	}

drivers/platform/x86/intel/vsec.c int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent, · line 180

Quoted 10 lines
drivers/platform/x86/intel/vsec.c +1 -5
 
-	ret = devm_add_action_or_reset(parent, intel_vsec_remove_aux,
+	return devm_add_action_or_reset(parent, intel_vsec_remove_aux,
 				       auxdev);
-	if (ret < 0)
-		return ret;
-
-	return 0;
 }
 EXPORT_SYMBOL_NS_GPL(intel_vsec_add_aux, INTEL_VSEC);
Signature · 1 line
 i.
author maintainer
#8 lore raw

PATCH v6 3/20 platform/x86/intel/vsec: Move structures to header

In reply to David E. Box

In preparation for exporting an API to register Intel Vendor Specific Extended Capabilities (VSEC) from other drivers, move needed structures to the header file.

---
V6 - No change

V5 - No change

V4 - No change

V3 - No change

V2 - New patch splitting previous PATCH 1

 drivers/platform/x86/intel/vsec.c | 35 ------------------------------
 drivers/platform/x86/intel/vsec.h | 36 +++++++++++++++++++++++++++++++
 2 files changed, 36 insertions(+), 35 deletions(-)
drivers/platform/x86/intel/vsec.c +0 -35
index 340562ae2041..c83e2c549fc7 100644
@@ -24,13 +24,6 @@
 
 #include "vsec.h"
 
-/* Intel DVSEC offsets */
-#define INTEL_DVSEC_ENTRIES		0xA
-#define INTEL_DVSEC_SIZE		0xB
-#define INTEL_DVSEC_TABLE		0xC
-#define INTEL_DVSEC_TABLE_BAR(x)	((x) & GENMASK(2, 0))
-#define INTEL_DVSEC_TABLE_OFFSET(x)	((x) & GENMASK(31, 3))
-#define TABLE_OFFSET_SHIFT		3
 #define PMT_XA_START			0
 #define PMT_XA_MAX			INT_MAX
 #define PMT_XA_LIMIT			XA_LIMIT(PMT_XA_START, PMT_XA_MAX)
@@ -39,34 +32,6 @@ static DEFINE_IDA(intel_vsec_ida);
 static DEFINE_IDA(intel_vsec_sdsi_ida);
 static DEFINE_XARRAY_ALLOC(auxdev_array);
 
-/**
- * struct intel_vsec_header - Common fields of Intel VSEC and DVSEC registers.
- * @rev:         Revision ID of the VSEC/DVSEC register space
- * @length:      Length of the VSEC/DVSEC register space
- * @id:          ID of the feature
- * @num_entries: Number of instances of the feature
- * @entry_size:  Size of the discovery table for each feature
- * @tbir:        BAR containing the discovery tables
- * @offset:      BAR offset of start of the first discovery table
- */
-struct intel_vsec_header {
-	u8	rev;
-	u16	length;
-	u16	id;
-	u8	num_entries;
-	u8	entry_size;
-	u8	tbir;
-	u32	offset;
-};
-
-enum intel_vsec_id {
-	VSEC_ID_TELEMETRY	= 2,
-	VSEC_ID_WATCHER		= 3,
-	VSEC_ID_CRASHLOG	= 4,
-	VSEC_ID_SDSI		= 65,
-	VSEC_ID_TPMI		= 66,
-};
-
 static const char *intel_vsec_name(enum intel_vsec_id id)
 {
 	switch (id) {
drivers/platform/x86/intel/vsec.h +36 -0
index 0a6201b4a0e9..c242c07ea69c 100644
@@ -11,9 +11,45 @@
 #define VSEC_CAP_SDSI		BIT(3)
 #define VSEC_CAP_TPMI		BIT(4)
 
+/* Intel DVSEC offsets */
+#define INTEL_DVSEC_ENTRIES		0xA
+#define INTEL_DVSEC_SIZE		0xB
+#define INTEL_DVSEC_TABLE		0xC
+#define INTEL_DVSEC_TABLE_BAR(x)	((x) & GENMASK(2, 0))
+#define INTEL_DVSEC_TABLE_OFFSET(x)	((x) & GENMASK(31, 3))
+#define TABLE_OFFSET_SHIFT		3
+
 struct pci_dev;
 struct resource;
 
+enum intel_vsec_id {
+	VSEC_ID_TELEMETRY	= 2,
+	VSEC_ID_WATCHER		= 3,
+	VSEC_ID_CRASHLOG	= 4,
+	VSEC_ID_SDSI		= 65,
+	VSEC_ID_TPMI		= 66,
+};
+
+/**
+ * struct intel_vsec_header - Common fields of Intel VSEC and DVSEC registers.
+ * @rev:         Revision ID of the VSEC/DVSEC register space
+ * @length:      Length of the VSEC/DVSEC register space
+ * @id:          ID of the feature
+ * @num_entries: Number of instances of the feature
+ * @entry_size:  Size of the discovery table for each feature
+ * @tbir:        BAR containing the discovery tables
+ * @offset:      BAR offset of start of the first discovery table
+ */
+struct intel_vsec_header {
+	u8	rev;
+	u16	length;
+	u16	id;
+	u8	num_entries;
+	u8	entry_size;
+	u8	tbir;
+	u32	offset;
+};
+
 enum intel_vsec_quirks {
 	/* Watcher feature not supported */
 	VSEC_QUIRK_NO_WATCHER	= BIT(0),
Signature · 1 line
2.34.1
author maintainer
#9 lore raw

PATCH v6 4/20 platform/x86/intel/vsec: remove platform_info from vsec device structure

In reply to David E. Box

In preparation for exporting an API to register Intel Vendor Specific Extended Capabilities (VSEC) from other drivers, remove the pointer to platform_info from intel_vsec_device. This prevents a potential page fault when auxiliary drivers probe and attempt to dereference this pointer to access the needed quirks field. Instead, just add the quirks to intel_vsec_device.

---
V6 - No change

V5 - No change

V4 - No change

V3 - No change

V2 - New patch splitting previous PATCH 1

 drivers/platform/x86/intel/pmt/class.c | 2 +-
 drivers/platform/x86/intel/vsec.c      | 2 +-
 drivers/platform/x86/intel/vsec.h      | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)
drivers/platform/x86/intel/pmt/class.c +1 -1
index f32a233470de..2ad91d2fd954 100644
@@ -31,7 +31,7 @@ bool intel_pmt_is_early_client_hw(struct device *dev)
 	 * differences from the server platforms (which use the Out Of Band
 	 * Management Services Module OOBMSM).
 	 */
-	return !!(ivdev->info->quirks & VSEC_QUIRK_EARLY_HW);
+	return !!(ivdev->quirks & VSEC_QUIRK_EARLY_HW);
 }
 EXPORT_SYMBOL_NS_GPL(intel_pmt_is_early_client_hw, INTEL_PMT);
 
drivers/platform/x86/intel/vsec.c +1 -1
index c83e2c549fc7..ba352af65dde 100644
@@ -193,7 +193,7 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 	intel_vsec_dev->pcidev = pdev;
 	intel_vsec_dev->resource = res;
 	intel_vsec_dev->num_resources = header->num_entries;
-	intel_vsec_dev->info = info;
+	intel_vsec_dev->quirks = info->quirks;
 
 	if (header->id == VSEC_ID_SDSI)
 		intel_vsec_dev->ida = &intel_vsec_sdsi_ida;
drivers/platform/x86/intel/vsec.h +1 -1
index c242c07ea69c..8b9fad170503 100644
@@ -79,11 +79,11 @@ struct intel_vsec_device {
 	struct pci_dev *pcidev;
 	struct resource *resource;
 	struct ida *ida;
-	struct intel_vsec_platform_info *info;
 	int num_resources;
 	int id; /* xa */
 	void *priv_data;
 	size_t priv_data_size;
+	unsigned long quirks;
 };
 
 int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
Signature · 1 line
2.34.1
author maintainer
#10 lore raw

PATCH v6 5/20 platform/x86/intel/vsec: Use cleanup.h

In reply to David E. Box

Use cleanup.h helpers to handle cleanup of resources in intel_vsec_add_dev() after failures.

---
V6 - No change

V5 - Assign no_free_ptr(res) to resource member. This fixes __must_check_fn
warning.
- Remove unsed ret variable.

V4 - Do no_free_ptr() before and in call to intel_vsec_add_aux().
- Add resource cleanup in this patch.

V3 - New patch.

 drivers/platform/x86/intel/vsec.c | 18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)
drivers/platform/x86/intel/vsec.c +11 -7
index ba352af65dde..27e8698ebe5e 100644
@@ -15,6 +15,7 @@
 
 #include <linux/auxiliary_bus.h>
 #include <linux/bits.h>
+#include <linux/cleanup.h>
 #include <linux/delay.h>
 #include <linux/kernel.h>
 #include <linux/idr.h>
@@ -147,8 +148,9 @@ EXPORT_SYMBOL_NS_GPL(intel_vsec_add_aux, INTEL_VSEC);
 static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *header,
 			      struct intel_vsec_platform_info *info)
 {
-	struct intel_vsec_device *intel_vsec_dev;
-	struct resource *res, *tmp;
+	struct intel_vsec_device __free(kfree) *intel_vsec_dev = NULL;
+	struct resource __free(kfree) *res = NULL;
+	struct resource *tmp;
 	unsigned long quirks = info->quirks;
 	int i;
 
@@ -170,10 +172,8 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 		return -ENOMEM;
 
 	res = kcalloc(header->num_entries, sizeof(*res), GFP_KERNEL);
-	if (!res) {
-		kfree(intel_vsec_dev);
+	if (!res)
 		return -ENOMEM;
-	}
 
 	if (quirks & VSEC_QUIRK_TABLE_SHIFT)
 		header->offset >>= TABLE_OFFSET_SHIFT;
@@ -191,7 +191,7 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 	}
 
 	intel_vsec_dev->pcidev = pdev;
-	intel_vsec_dev->resource = res;
+	intel_vsec_dev->resource = no_free_ptr(res);
 	intel_vsec_dev->num_resources = header->num_entries;
 	intel_vsec_dev->quirks = info->quirks;
 
@@ -200,7 +200,11 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 	else
 		intel_vsec_dev->ida = &intel_vsec_ida;
 
-	return intel_vsec_add_aux(pdev, NULL, intel_vsec_dev,
+	/*
+	 * Pass the ownership of intel_vsec_dev and resource within it to
+	 * intel_vsec_add_aux()
+	 */
+	return intel_vsec_add_aux(pdev, NULL, no_free_ptr(intel_vsec_dev),
 				  intel_vsec_name(header->id));
 }
 
Signature · 1 line
2.34.1
author maintainer
#11 lore raw

PATCH v6 6/20 platform/x86/intel/vsec: Assign auxdev parent by argument

In reply to David E. Box

Instead of checking for a NULL parent argument in intel_vsec_add_aux() and then assigning it to the probed device, remove this check and just pass the device in the call. Since this function is exported, return -EINVAL if the parent is not specified.

---
V6 - No change

V5 - New patch

 drivers/platform/x86/intel/vsec.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
drivers/platform/x86/intel/vsec.c +4 -4
index 27e8698ebe5e..329ba03a6da0 100644
@@ -103,6 +103,9 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
 	struct auxiliary_device *auxdev = &intel_vsec_dev->auxdev;
 	int ret, id;
 
+	if (!parent)
+		return -EINVAL;
+
 	ret = xa_alloc(&auxdev_array, &intel_vsec_dev->id, intel_vsec_dev,
 		       PMT_XA_LIMIT, GFP_KERNEL);
 	if (ret < 0) {
@@ -120,9 +123,6 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
 		return id;
 	}
 
-	if (!parent)
-		parent = &pdev->dev;
-
 	auxdev->id = id;
 	auxdev->name = name;
 	auxdev->dev.parent = parent;
@@ -204,7 +204,7 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 	 * Pass the ownership of intel_vsec_dev and resource within it to
 	 * intel_vsec_add_aux()
 	 */
-	return intel_vsec_add_aux(pdev, NULL, no_free_ptr(intel_vsec_dev),
+	return intel_vsec_add_aux(pdev, &pdev->dev, no_free_ptr(intel_vsec_dev),
 				  intel_vsec_name(header->id));
 }
 
Signature · 1 line
2.34.1
author maintainer
#12 lore raw

PATCH v6 7/20 platform/x86/intel/vsec: Add intel_vsec_register

In reply to David E. Box

From: Gayatri Kammela <[email protected]>

Add and export intel_vsec_register() to allow the registration of Intel extended capabilities from other drivers. Add check to look for memory conflicts before registering a new capability. Since the vsec provider may not be a PCI device, add a parent field to intel_vsec_platform_info() to allow specifying the parent device for device managed cleanup.

---
V6 - In changelog, explain need for the parent field

V5 - Add parent variable in intel_vsec_add_dev() and set parent to

     the pci device if info->parent is not set.

V4 - Move res cleanup to previous patch

V3 - Replace kfree on request_mem_region fail with use of cleanup.h helper.

V2 - New patch splitting previous PATCH 1

 drivers/platform/x86/intel/vsec.c | 24 +++++++++++++++++++++++-
 drivers/platform/x86/intel/vsec.h |  4 ++++
 2 files changed, 27 insertions(+), 1 deletion(-)
drivers/platform/x86/intel/vsec.c +23 -1
index 329ba03a6da0..2978d4e95bb0 100644
@@ -151,9 +151,15 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 	struct intel_vsec_device __free(kfree) *intel_vsec_dev = NULL;
 	struct resource __free(kfree) *res = NULL;
 	struct resource *tmp;
+	struct device *parent;
 	unsigned long quirks = info->quirks;
 	int i;
 
+	if (info->parent)
+		parent = info->parent;
+	else
+		parent = &pdev->dev;
+
 	if (!intel_vsec_supported(header->id, info->caps))
 		return -EINVAL;
 
@@ -188,6 +194,12 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 			     header->offset + i * (header->entry_size * sizeof(u32));
 		tmp->end = tmp->start + (header->entry_size * sizeof(u32)) - 1;
 		tmp->flags = IORESOURCE_MEM;
+
+		/* Check resource is not in use */
+		if (!request_mem_region(tmp->start, resource_size(tmp), ""))
+			return -EBUSY;
+
+		release_mem_region(tmp->start, resource_size(tmp));
 	}
 
 	intel_vsec_dev->pcidev = pdev;
@@ -204,7 +216,7 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 	 * Pass the ownership of intel_vsec_dev and resource within it to
 	 * intel_vsec_add_aux()
 	 */
-	return intel_vsec_add_aux(pdev, &pdev->dev, no_free_ptr(intel_vsec_dev),
+	return intel_vsec_add_aux(pdev, parent, no_free_ptr(intel_vsec_dev),
 				  intel_vsec_name(header->id));
 }
 
@@ -322,6 +334,16 @@ static bool intel_vsec_walk_vsec(struct pci_dev *pdev,
 	return have_devices;
 }
 
+void intel_vsec_register(struct pci_dev *pdev,
+			 struct intel_vsec_platform_info *info)
+{
+	if (!pdev || !info)
+		return;
+
+	intel_vsec_walk_header(pdev, info);
+}
+EXPORT_SYMBOL_NS_GPL(intel_vsec_register, INTEL_VSEC);
+
 static int intel_vsec_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 {
 	struct intel_vsec_platform_info *info;
drivers/platform/x86/intel/vsec.h +4 -0
index 8b9fad170503..bb8b6452df70 100644
@@ -69,6 +69,7 @@ enum intel_vsec_quirks {
 
 /* Platform specific data */
 struct intel_vsec_platform_info {
+	struct device *parent;
 	struct intel_vsec_header **headers;
 	unsigned long caps;
 	unsigned long quirks;
@@ -99,4 +100,7 @@ static inline struct intel_vsec_device *auxdev_to_ivdev(struct auxiliary_device
 {
 	return container_of(auxdev, struct intel_vsec_device, auxdev);
 }
+
+void intel_vsec_register(struct pci_dev *pdev,
+			 struct intel_vsec_platform_info *info);
 #endif
Signature · 1 line
2.34.1
maintainer Reviewed-by on v6 7/20
#13 lore raw

Re: [PATCH V6 07/20] platform/x86/intel/vsec: Add intel_vsec_register

Quoted 9 lines · David E. Box

From: Gayatri Kammela <[email protected]>

Add and export intel_vsec_register() to allow the registration of Intel extended capabilities from other drivers. Add check to look for memory conflicts before registering a new capability. Since the vsec provider may not be a PCI device, add a parent field to intel_vsec_platform_info() to allow specifying the parent device for device managed cleanup.

Quoted 3 lines

---

Signature · 1 line
 i.
author maintainer
#14 lore raw

PATCH v6 8/20 platform/x86/intel/vsec: Add base address field

In reply to David E. Box

Some devices may emulate PCI VSEC capabilities in MMIO. In such cases the BAR is not readable from a config space. Provide a field for drivers to indicate the base address to be used.

---
V6 - No change

V5 - No change

V4 - No change

V3 - No change

V2 - No change

 drivers/platform/x86/intel/pmt/class.c | 14 +++++++++++---
 drivers/platform/x86/intel/vsec.c      | 10 ++++++++--
 drivers/platform/x86/intel/vsec.h      |  2 ++
 3 files changed, 21 insertions(+), 5 deletions(-)
drivers/platform/x86/intel/pmt/class.c +11 -3
index 2ad91d2fd954..32608baaa56c 100644
@@ -160,10 +160,11 @@ static struct class intel_pmt_class = {
 
 static int intel_pmt_populate_entry(struct intel_pmt_entry *entry,
 				    struct intel_pmt_header *header,
-				    struct device *dev,
+				    struct intel_vsec_device *ivdev,
 				    struct resource *disc_res)
 {
-	struct pci_dev *pci_dev = to_pci_dev(dev->parent);
+	struct pci_dev *pci_dev = ivdev->pcidev;
+	struct device *dev = &ivdev->auxdev.dev;
 	u8 bir;
 
 	/*
@@ -215,6 +216,13 @@ static int intel_pmt_populate_entry(struct intel_pmt_entry *entry,
 
 		break;
 	case ACCESS_BARID:
+		/* Use the provided base address if it exists */
+		if (ivdev->base_addr) {
+			entry->base_addr = ivdev->base_addr +
+				   GET_ADDRESS(header->base_offset);
+			break;
+		}
+
 		/*
 		 * If another BAR was specified then the base offset
 		 * represents the offset within that BAR. SO retrieve the
@@ -319,7 +327,7 @@ int intel_pmt_dev_create(struct intel_pmt_entry *entry, struct intel_pmt_namespa
 	if (ret)
 		return ret;
 
-	ret = intel_pmt_populate_entry(entry, &header, dev, disc_res);
+	ret = intel_pmt_populate_entry(entry, &header, intel_vsec_dev, disc_res);
 	if (ret)
 		return ret;
 
drivers/platform/x86/intel/vsec.c +8 -2
index 2978d4e95bb0..648ee842d413 100644
@@ -153,6 +153,7 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 	struct resource *tmp;
 	struct device *parent;
 	unsigned long quirks = info->quirks;
+	u64 base_addr;
 	int i;
 
 	if (info->parent)
@@ -184,14 +185,18 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 	if (quirks & VSEC_QUIRK_TABLE_SHIFT)
 		header->offset >>= TABLE_OFFSET_SHIFT;
 
+	if (info->base_addr)
+		base_addr = info->base_addr;
+	else
+		base_addr = pdev->resource[header->tbir].start;
+
 	/*
 	 * The DVSEC/VSEC contains the starting offset and count for a block of
 	 * discovery tables. Create a resource array of these tables to the
 	 * auxiliary device driver.
 	 */
 	for (i = 0, tmp = res; i < header->num_entries; i++, tmp++) {
-		tmp->start = pdev->resource[header->tbir].start +
-			     header->offset + i * (header->entry_size * sizeof(u32));
+		tmp->start = base_addr + header->offset + i * (header->entry_size * sizeof(u32));
 		tmp->end = tmp->start + (header->entry_size * sizeof(u32)) - 1;
 		tmp->flags = IORESOURCE_MEM;
 
@@ -206,6 +211,7 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
 	intel_vsec_dev->resource = no_free_ptr(res);
 	intel_vsec_dev->num_resources = header->num_entries;
 	intel_vsec_dev->quirks = info->quirks;
+	intel_vsec_dev->base_addr = info->base_addr;
 
 	if (header->id == VSEC_ID_SDSI)
 		intel_vsec_dev->ida = &intel_vsec_sdsi_ida;
drivers/platform/x86/intel/vsec.h +2 -0
index bb8b6452df70..e23e76129691 100644
@@ -73,6 +73,7 @@ struct intel_vsec_platform_info {
 	struct intel_vsec_header **headers;
 	unsigned long caps;
 	unsigned long quirks;
+	u64 base_addr;
 };
 
 struct intel_vsec_device {
@@ -85,6 +86,7 @@ struct intel_vsec_device {
 	void *priv_data;
 	size_t priv_data_size;
 	unsigned long quirks;
+	u64 base_addr;
 };
 
 int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
Signature · 1 line
2.34.1
author maintainer
#15 lore raw

PATCH v6 9/20 platform/x86/intel/pmt: Add header to struct intel_pmt_entry

In reply to David E. Box

The PMT header is passed to several functions. Instead, store the header in struct intel_pmt_entry which is also passed to these functions and shorten the argument list. This simplifies the calls in preparation for later changes. While here also perform a newline cleanup.

---
V6 - No change

V5 - Mention newline cleanup in changelog

V4 - No change

V3 - No change

V2 - No change

 drivers/platform/x86/intel/pmt/class.c     |  8 +++-----
 drivers/platform/x86/intel/pmt/class.h     | 16 ++++++++--------
 drivers/platform/x86/intel/pmt/crashlog.c  |  2 +-
 drivers/platform/x86/intel/pmt/telemetry.c |  2 +-
 4 files changed, 13 insertions(+), 15 deletions(-)
drivers/platform/x86/intel/pmt/class.c +3 -5
index 32608baaa56c..142a24e3727d 100644
@@ -159,12 +159,12 @@ static struct class intel_pmt_class = {
 };
 
 static int intel_pmt_populate_entry(struct intel_pmt_entry *entry,
-				    struct intel_pmt_header *header,
 				    struct intel_vsec_device *ivdev,
 				    struct resource *disc_res)
 {
 	struct pci_dev *pci_dev = ivdev->pcidev;
 	struct device *dev = &ivdev->auxdev.dev;
+	struct intel_pmt_header *header = &entry->header;
 	u8 bir;
 
 	/*
@@ -313,7 +313,6 @@ int intel_pmt_dev_create(struct intel_pmt_entry *entry, struct intel_pmt_namespa
 			 struct intel_vsec_device *intel_vsec_dev, int idx)
 {
 	struct device *dev = &intel_vsec_dev->auxdev.dev;
-	struct intel_pmt_header header;
 	struct resource	*disc_res;
 	int ret;
 
@@ -323,16 +322,15 @@ int intel_pmt_dev_create(struct intel_pmt_entry *entry, struct intel_pmt_namespa
 	if (IS_ERR(entry->disc_table))
 		return PTR_ERR(entry->disc_table);
 
-	ret = ns->pmt_header_decode(entry, &header, dev);
+	ret = ns->pmt_header_decode(entry, dev);
 	if (ret)
 		return ret;
 
-	ret = intel_pmt_populate_entry(entry, &header, intel_vsec_dev, disc_res);
+	ret = intel_pmt_populate_entry(entry, intel_vsec_dev, disc_res);
 	if (ret)
 		return ret;
 
 	return intel_pmt_dev_register(entry, ns, dev);
-
 }
 EXPORT_SYMBOL_NS_GPL(intel_pmt_dev_create, INTEL_PMT);
 
drivers/platform/x86/intel/pmt/class.h +8 -8
index db11d58867ce..e477a19f6700 100644
@@ -18,7 +18,15 @@
 #define GET_BIR(v)		((v) & GENMASK(2, 0))
 #define GET_ADDRESS(v)		((v) & GENMASK(31, 3))
 
+struct intel_pmt_header {
+	u32	base_offset;
+	u32	size;
+	u32	guid;
+	u8	access_type;
+};
+
 struct intel_pmt_entry {
+	struct intel_pmt_header	header;
 	struct bin_attribute	pmt_bin_attr;
 	struct kobject		*kobj;
 	void __iomem		*disc_table;
@@ -29,19 +37,11 @@ struct intel_pmt_entry {
 	int			devid;
 };
 
-struct intel_pmt_header {
-	u32	base_offset;
-	u32	size;
-	u32	guid;
-	u8	access_type;
-};
-
 struct intel_pmt_namespace {
 	const char *name;
 	struct xarray *xa;
 	const struct attribute_group *attr_grp;
 	int (*pmt_header_decode)(struct intel_pmt_entry *entry,
-				 struct intel_pmt_header *header,
 				 struct device *dev);
 };
 
drivers/platform/x86/intel/pmt/crashlog.c +1 -1
index bbb3d61d09f4..4014c02cafdb 100644
@@ -223,10 +223,10 @@ static const struct attribute_group pmt_crashlog_group = {
 };
 
 static int pmt_crashlog_header_decode(struct intel_pmt_entry *entry,
-				      struct intel_pmt_header *header,
 				      struct device *dev)
 {
 	void __iomem *disc_table = entry->disc_table;
+	struct intel_pmt_header *header = &entry->header;
 	struct crashlog_entry *crashlog;
 
 	if (!pmt_crashlog_supported(entry))
drivers/platform/x86/intel/pmt/telemetry.c +1 -1
index 39cbc87cc28a..f86080e8bebd 100644
@@ -58,10 +58,10 @@ static bool pmt_telem_region_overlaps(struct intel_pmt_entry *entry,
 }
 
 static int pmt_telem_header_decode(struct intel_pmt_entry *entry,
-				   struct intel_pmt_header *header,
 				   struct device *dev)
 {
 	void __iomem *disc_table = entry->disc_table;
+	struct intel_pmt_header *header = &entry->header;
 
 	if (pmt_telem_region_overlaps(entry, dev))
 		return 1;
Signature · 1 line
2.34.1
author maintainer
#16 lore raw

PATCH v6 10/20 platform/x86/intel/pmt: telemetry: Export API to read telemetry

In reply to David E. Box

Export symbols to allow access to Intel PMT Telemetry data on available devices. Provides APIs to search, register, and read telemetry using a kref managed pointer that serves as a handle to a telemetry endpoint. To simplify searching for present devices, have the IDA start at 1 instead of 0 so that 0 can be used to indicate end of search.

---
V6 - No change

V5 - No change

V4 - No change

V3 - Add comment for mutex lock
- Fix multiline comments
- Fix line lengths
- Fix space/tab inconsistences in header doc

V2 - Add explanation of PMT_XA_START change in changelog
- change return and argument type of pmt_telem_get_next_endpoint() to unsigned long
- style fixes

 drivers/platform/x86/intel/pmt/class.c     |  21 ++-
 drivers/platform/x86/intel/pmt/class.h     |  14 ++
 drivers/platform/x86/intel/pmt/telemetry.c | 191 ++++++++++++++++++++-
 drivers/platform/x86/intel/pmt/telemetry.h | 126 ++++++++++++++
 4 files changed, 344 insertions(+), 8 deletions(-)
 create mode 100644 drivers/platform/x86/intel/pmt/telemetry.h
drivers/platform/x86/intel/pmt/class.c +16 -5
index 142a24e3727d..4b53940a64e2 100644
@@ -17,7 +17,7 @@
 #include "../vsec.h"
 #include "class.h"
 
-#define PMT_XA_START		0
+#define PMT_XA_START		1
 #define PMT_XA_MAX		INT_MAX
 #define PMT_XA_LIMIT		XA_LIMIT(PMT_XA_START, PMT_XA_MAX)
 #define GUID_SPR_PUNIT		0x9956f43f
@@ -247,6 +247,7 @@ static int intel_pmt_dev_register(struct intel_pmt_entry *entry,
 				  struct intel_pmt_namespace *ns,
 				  struct device *parent)
 {
+	struct intel_vsec_device *ivdev = dev_to_ivdev(parent);
 	struct resource res = {0};
 	struct device *dev;
 	int ret;
@@ -270,7 +271,7 @@ static int intel_pmt_dev_register(struct intel_pmt_entry *entry,
 	if (ns->attr_grp) {
 		ret = sysfs_create_group(entry->kobj, ns->attr_grp);
 		if (ret)
-			goto fail_sysfs;
+			goto fail_sysfs_create_group;
 	}
 
 	/* if size is 0 assume no data buffer, so no file needed */
@@ -295,13 +296,23 @@ static int intel_pmt_dev_register(struct intel_pmt_entry *entry,
 	entry->pmt_bin_attr.size = entry->size;
 
 	ret = sysfs_create_bin_file(&dev->kobj, &entry->pmt_bin_attr);
-	if (!ret)
-		return 0;
+	if (ret)
+		goto fail_ioremap;
 
+	if (ns->pmt_add_endpoint) {
+		ret = ns->pmt_add_endpoint(entry, ivdev->pcidev);
+		if (ret)
+			goto fail_add_endpoint;
+	}
+
+	return 0;
+
+fail_add_endpoint:
+	sysfs_remove_bin_file(entry->kobj, &entry->pmt_bin_attr);
 fail_ioremap:
 	if (ns->attr_grp)
 		sysfs_remove_group(entry->kobj, ns->attr_grp);
-fail_sysfs:
+fail_sysfs_create_group:
 	device_unregister(dev);
 fail_dev_create:
 	xa_erase(ns->xa, entry->devid);
drivers/platform/x86/intel/pmt/class.h +14 -0
index e477a19f6700..d23c63b73ab7 100644
@@ -9,6 +9,7 @@
 #include <linux/io.h>
 
 #include "../vsec.h"
+#include "telemetry.h"
 
 /* PMT access types */
 #define ACCESS_BARID		2
@@ -18,6 +19,16 @@
 #define GET_BIR(v)		((v) & GENMASK(2, 0))
 #define GET_ADDRESS(v)		((v) & GENMASK(31, 3))
 
+struct pci_dev;
+
+struct telem_endpoint {
+	struct pci_dev		*pcidev;
+	struct telem_header	header;
+	void __iomem		*base;
+	bool			present;
+	struct kref		kref;
+};
+
 struct intel_pmt_header {
 	u32	base_offset;
 	u32	size;
@@ -26,6 +37,7 @@ struct intel_pmt_header {
 };
 
 struct intel_pmt_entry {
+	struct telem_endpoint	*ep;
 	struct intel_pmt_header	header;
 	struct bin_attribute	pmt_bin_attr;
 	struct kobject		*kobj;
@@ -43,6 +55,8 @@ struct intel_pmt_namespace {
 	const struct attribute_group *attr_grp;
 	int (*pmt_header_decode)(struct intel_pmt_entry *entry,
 				 struct device *dev);
+	int (*pmt_add_endpoint)(struct intel_pmt_entry *entry,
+				struct pci_dev *pdev);
 };
 
 bool intel_pmt_is_early_client_hw(struct device *dev);
drivers/platform/x86/intel/pmt/telemetry.c +188 -3

222 lines not on this page · raw message

drivers/platform/x86/intel/pmt/telemetry.h new file +126 -0

126 lines not on this page · raw message

Signature · 1 line
2.34.1
author maintainer
#17 lore raw

PATCH v6 11/20 platform/x86:intel/pmc: Call pmc_get_low_power_modes from platform init

In reply to David E. Box

From: Xi Pardee <[email protected]>

In order to setup a table of low power mode requirements for Meteor Lake, pmc_core_get_low_power_modes() will need to be run from platform init code so that the enabled modes are known, allowing the use of the pmc_for_each_mode helper. Make the function global and call it from the platform init code.

---
V6 - No change

V5 - No change

V4 - No change

V3 - No change

V2 - Added clearer explanation for the change in the changelog

 drivers/platform/x86/intel/pmc/adl.c  |  2 ++
 drivers/platform/x86/intel/pmc/cnp.c  |  2 ++
 drivers/platform/x86/intel/pmc/core.c |  7 +++----
 drivers/platform/x86/intel/pmc/core.h |  1 +
 drivers/platform/x86/intel/pmc/icl.c  | 10 +++++++++-
 drivers/platform/x86/intel/pmc/mtl.c  |  4 +++-
 drivers/platform/x86/intel/pmc/spt.c  | 10 +++++++++-
 drivers/platform/x86/intel/pmc/tgl.c  |  1 +
 8 files changed, 30 insertions(+), 7 deletions(-)
drivers/platform/x86/intel/pmc/adl.c +2 -0
index 5006008e01be..64c492391ede 100644
@@ -319,6 +319,8 @@ int adl_core_init(struct pmc_dev *pmcdev)
 	if (ret)
 		return ret;
 
+	pmc_core_get_low_power_modes(pmcdev);
+
 	/* Due to a hardware limitation, the GBE LTR blocks PC10
 	 * when a cable is attached. Tell the PMC to ignore it.
 	 */
drivers/platform/x86/intel/pmc/cnp.c +2 -0
index 420aaa1d7c76..59298f184d0e 100644
@@ -214,6 +214,8 @@ int cnp_core_init(struct pmc_dev *pmcdev)
 	if (ret)
 		return ret;
 
+	pmc_core_get_low_power_modes(pmcdev);
+
 	/* Due to a hardware limitation, the GBE LTR blocks PC10
 	 * when a cable is attached. Tell the PMC to ignore it.
 	 */
drivers/platform/x86/intel/pmc/core.c +3 -4
index 84c175b9721a..3894119d61b0 100644
@@ -966,9 +966,8 @@ static bool pmc_core_pri_verify(u32 lpm_pri, u8 *mode_order)
 	return true;
 }
 
-static void pmc_core_get_low_power_modes(struct platform_device *pdev)
+void pmc_core_get_low_power_modes(struct pmc_dev *pmcdev)
 {
-	struct pmc_dev *pmcdev = platform_get_drvdata(pdev);
 	struct pmc *pmc = pmcdev->pmcs[PMC_IDX_MAIN];
 	u8 pri_order[LPM_MAX_NUM_MODES] = LPM_DEFAULT_PRI;
 	u8 mode_order[LPM_MAX_NUM_MODES];
@@ -1000,7 +999,8 @@ static void pmc_core_get_low_power_modes(struct platform_device *pdev)
 		for (mode = 0; mode < LPM_MAX_NUM_MODES; mode++)
 			pri_order[mode_order[mode]] = mode;
 	else
-		dev_warn(&pdev->dev, "Assuming a default substate order for this platform\n");
+		dev_warn(&pmcdev->pdev->dev,
+			 "Assuming a default substate order for this platform\n");
 
 	/*
 	 * Loop through all modes from lowest to highest priority,
@@ -1250,7 +1250,6 @@ static int pmc_core_probe(struct platform_device *pdev)
 	}
 
 	pmcdev->pmc_xram_read_bit = pmc_core_check_read_lock_bit(primary_pmc);
-	pmc_core_get_low_power_modes(pdev);
 	pmc_core_do_dmi_quirks(primary_pmc);
 
 	pmc_core_dbgfs_register(pmcdev);
drivers/platform/x86/intel/pmc/core.h +1 -0
index 0729f593c6a7..ccf24e0f5e50 100644
@@ -490,6 +490,7 @@ extern int pmc_core_send_ltr_ignore(struct pmc_dev *pmcdev, u32 value);
 
 int pmc_core_resume_common(struct pmc_dev *pmcdev);
 int get_primary_reg_base(struct pmc *pmc);
+extern void pmc_core_get_low_power_modes(struct pmc_dev *pmcdev);
 
 extern void pmc_core_ssram_init(struct pmc_dev *pmcdev);
 
drivers/platform/x86/intel/pmc/icl.c +9 -1
index d08e3174230d..71b0fd6cb7d8 100644
@@ -53,7 +53,15 @@ const struct pmc_reg_map icl_reg_map = {
 int icl_core_init(struct pmc_dev *pmcdev)
 {
 	struct pmc *pmc = pmcdev->pmcs[PMC_IDX_MAIN];
+	int ret;
 
 	pmc->map = &icl_reg_map;
-	return get_primary_reg_base(pmc);
+
+	ret = get_primary_reg_base(pmc);
+	if (ret)
+		return ret;
+
+	pmc_core_get_low_power_modes(pmcdev);
+
+	return ret;
 }
drivers/platform/x86/intel/pmc/mtl.c +3 -1
index 2204bc666980..c3b5f4fe01d1 100644
@@ -985,7 +985,7 @@ static int mtl_resume(struct pmc_dev *pmcdev)
 int mtl_core_init(struct pmc_dev *pmcdev)
 {
 	struct pmc *pmc = pmcdev->pmcs[PMC_IDX_SOC];
-	int ret = 0;
+	int ret;
 
 	mtl_d3_fixup();
 
@@ -1002,6 +1002,8 @@ int mtl_core_init(struct pmc_dev *pmcdev)
 			return ret;
 	}
 
+	pmc_core_get_low_power_modes(pmcdev);
+
 	/* Due to a hardware limitation, the GBE LTR blocks PC10
 	 * when a cable is attached. Tell the PMC to ignore it.
 	 */
drivers/platform/x86/intel/pmc/spt.c +9 -1
index 4b6f5cbda16c..ab993a69e33e 100644
@@ -137,7 +137,15 @@ const struct pmc_reg_map spt_reg_map = {
 int spt_core_init(struct pmc_dev *pmcdev)
 {
 	struct pmc *pmc = pmcdev->pmcs[PMC_IDX_MAIN];
+	int ret;
 
 	pmc->map = &spt_reg_map;
-	return get_primary_reg_base(pmc);
+
+	ret = get_primary_reg_base(pmc);
+	if (ret)
+		return ret;
+
+	pmc_core_get_low_power_modes(pmcdev);
+
+	return ret;
 }
drivers/platform/x86/intel/pmc/tgl.c +1 -0
index 2449940102db..d5f1d2223c5a 100644
@@ -263,6 +263,7 @@ int tgl_core_init(struct pmc_dev *pmcdev)
 	if (ret)
 		return ret;
 
+	pmc_core_get_low_power_modes(pmcdev);
 	pmc_core_get_tgl_lpm_reqs(pmcdev->pdev);
 	/* Due to a hardware limitation, the GBE LTR blocks PC10
 	 * when a cable is attached. Tell the PMC to ignore it.
Signature · 1 line
2.34.1
author maintainer
#18 lore raw

PATCH v6 12/20 platform/x86/intel/pmc: Allow pmc_core_ssram_init to fail

In reply to David E. Box

Currently, if the PMC SSRAM initialization fails, no error is returned and the only indication is that a PMC device has not been created. Instead, allow an error to be returned and handled directly by the caller.

---
V6 - No change

V5 - No change

V4 - Remove return value check in mtl.c. Proper use of the value would
require returning an error status from pmc_core_add_pmc(). But the function that calls it will be removed in the next patch so wait to use it then.

V3 - New patch split from V2 PATCH 9
- Add dev_warn on pmc_core_ssram_init fail

 drivers/platform/x86/intel/pmc/core.h       |  2 +-
 drivers/platform/x86/intel/pmc/core_ssram.c | 21 +++++++++++++--------
 2 files changed, 14 insertions(+), 9 deletions(-)
drivers/platform/x86/intel/pmc/core.h +1 -1
index ccf24e0f5e50..edaa70067e41 100644
@@ -492,7 +492,7 @@ int pmc_core_resume_common(struct pmc_dev *pmcdev);
 int get_primary_reg_base(struct pmc *pmc);
 extern void pmc_core_get_low_power_modes(struct pmc_dev *pmcdev);
 
-extern void pmc_core_ssram_init(struct pmc_dev *pmcdev);
+extern int pmc_core_ssram_init(struct pmc_dev *pmcdev);
 
 int spt_core_init(struct pmc_dev *pmcdev);
 int cnp_core_init(struct pmc_dev *pmcdev);
drivers/platform/x86/intel/pmc/core_ssram.c +13 -8
index 13fa16f0d52e..815950713e25 100644
@@ -35,20 +35,20 @@ static inline u64 get_base(void __iomem *addr, u32 offset)
 	return lo_hi_readq(addr + offset) & GENMASK_ULL(63, 3);
 }
 
-static void
+static int
 pmc_core_pmc_add(struct pmc_dev *pmcdev, u64 pwrm_base,
 		 const struct pmc_reg_map *reg_map, int pmc_index)
 {
 	struct pmc *pmc = pmcdev->pmcs[pmc_index];
 
 	if (!pwrm_base)
-		return;
+		return -ENODEV;
 
 	/* Memory for primary PMC has been allocated in core.c */
 	if (!pmc) {
 		pmc = devm_kzalloc(&pmcdev->pdev->dev, sizeof(*pmc), GFP_KERNEL);
 		if (!pmc)
-			return;
+			return -ENOMEM;
 	}
 
 	pmc->map = reg_map;
@@ -57,10 +57,12 @@ pmc_core_pmc_add(struct pmc_dev *pmcdev, u64 pwrm_base,
 
 	if (!pmc->regbase) {
 		devm_kfree(&pmcdev->pdev->dev, pmc);
-		return;
+		return -ENOMEM;
 	}
 
 	pmcdev->pmcs[pmc_index] = pmc;
+
+	return 0;
 }
 
 static void
@@ -96,7 +98,7 @@ pmc_core_ssram_get_pmc(struct pmc_dev *pmcdev, void __iomem *ssram, u32 offset,
 		iounmap(ssram);
 }
 
-void pmc_core_ssram_init(struct pmc_dev *pmcdev)
+int pmc_core_ssram_init(struct pmc_dev *pmcdev)
 {
 	void __iomem *ssram;
 	struct pci_dev *pcidev;
@@ -105,7 +107,7 @@ void pmc_core_ssram_init(struct pmc_dev *pmcdev)
 
 	pcidev = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(20, 2));
 	if (!pcidev)
-		goto out;
+		return -ENODEV;
 
 	ret = pcim_enable_device(pcidev);
 	if (ret)
@@ -123,11 +125,14 @@ void pmc_core_ssram_init(struct pmc_dev *pmcdev)
 	pmc_core_ssram_get_pmc(pmcdev, ssram, SSRAM_PCH_OFFSET, PMC_IDX_PCH);
 
 	iounmap(ssram);
-out:
-	return;
+
+	return 0;
 
 disable_dev:
+	pmcdev->ssram_pcidev = NULL;
 	pci_disable_device(pcidev);
 release_dev:
 	pci_dev_put(pcidev);
+
+	return ret;
 }
Signature · 1 line
2.34.1
author maintainer
#19 lore raw

PATCH v6 13/20 platform/x86/intel/pmc: Cleanup SSRAM discovery

In reply to David E. Box

Clean up the code handling SSRAM discovery. Handle all resource allocation and cleanup in pmc_core_ssram_get_pmc(). Return the error status from this function but only fail the init if we fail to discover the primary PMC.

---
V6 - Drops patch adding cleanup helpers for iounmap() and ioport_unmap()
to io.h until s390 build issue is resolved. Add pmc_core specific DEFINE_FREE helper for iounmap() for now.

V5 - Use single function to handle SSRAM discovery of all PMCs.

V4 - Add checking the return value from pmc_core_sram_init() to mtl.c
- Use iounmap cleanup from io.h

V3 - New patch split from previous PATCH 2
- Update changelog
- Use cleanup.h to cleanup ioremap

V2 - no change

 drivers/platform/x86/intel/pmc/core_ssram.c | 62 +++++++++++----------
 1 file changed, 33 insertions(+), 29 deletions(-)
drivers/platform/x86/intel/pmc/core_ssram.c +33 -29
index 815950713e25..c1b984255571 100644
@@ -8,6 +8,7 @@
  *
  */
 
+#include <linux/cleanup.h>
 #include <linux/pci.h>
 #include <linux/io-64-nonatomic-lo-hi.h>
 
@@ -21,6 +22,8 @@
 #define SSRAM_IOE_OFFSET	0x68
 #define SSRAM_DEVID_OFFSET	0x70
 
+DEFINE_FREE(pmc_core_iounmap, void __iomem *, iounmap(_T));
+
 static const struct pmc_reg_map *pmc_core_find_regmap(struct pmc_info *list, u16 devid)
 {
 	for (; list->map; ++list)
@@ -65,44 +68,49 @@ pmc_core_pmc_add(struct pmc_dev *pmcdev, u64 pwrm_base,
 	return 0;
 }
 
-static void
-pmc_core_ssram_get_pmc(struct pmc_dev *pmcdev, void __iomem *ssram, u32 offset,
-		       int pmc_idx)
+static int
+pmc_core_ssram_get_pmc(struct pmc_dev *pmcdev, int pmc_idx, u32 offset)
 {
-	u64 pwrm_base;
+	struct pci_dev *ssram_pcidev = pmcdev->ssram_pcidev;
+	void __iomem __free(pmc_core_iounmap) *tmp_ssram = NULL;
+	void __iomem __free(pmc_core_iounmap) *ssram = NULL;
+	const struct pmc_reg_map *map;
+	u64 ssram_base, pwrm_base;
 	u16 devid;
 
-	if (pmc_idx != PMC_IDX_SOC) {
-		u64 ssram_base = get_base(ssram, offset);
+	if (!pmcdev->regmap_list)
+		return -ENOENT;
 
-		if (!ssram_base)
-			return;
+	ssram_base = ssram_pcidev->resource[0].start;
+	tmp_ssram = ioremap(ssram_base, SSRAM_HDR_SIZE);
 
+	if (pmc_idx != PMC_IDX_MAIN) {
+		/*
+		 * The secondary PMC BARS (which are behind hidden PCI devices)
+		 * are read from fixed offsets in MMIO of the primary PMC BAR.
+		 */
+		ssram_base = get_base(tmp_ssram, offset);
 		ssram = ioremap(ssram_base, SSRAM_HDR_SIZE);
 		if (!ssram)
-			return;
+			return -ENOMEM;
+
+	} else {
+		ssram = no_free_ptr(tmp_ssram);
 	}
 
 	pwrm_base = get_base(ssram, SSRAM_PWRM_OFFSET);
 	devid = readw(ssram + SSRAM_DEVID_OFFSET);
 
-	if (pmcdev->regmap_list) {
-		const struct pmc_reg_map *map;
+	map = pmc_core_find_regmap(pmcdev->regmap_list, devid);
+	if (!map)
+		return -ENODEV;
 
-		map = pmc_core_find_regmap(pmcdev->regmap_list, devid);
-		if (map)
-			pmc_core_pmc_add(pmcdev, pwrm_base, map, pmc_idx);
-	}
-
-	if (pmc_idx != PMC_IDX_SOC)
-		iounmap(ssram);
+	return pmc_core_pmc_add(pmcdev, pwrm_base, map, PMC_IDX_MAIN);
 }
 
 int pmc_core_ssram_init(struct pmc_dev *pmcdev)
 {
-	void __iomem *ssram;
 	struct pci_dev *pcidev;
-	u64 ssram_base;
 	int ret;
 
 	pcidev = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(20, 2));
@@ -113,18 +121,14 @@ int pmc_core_ssram_init(struct pmc_dev *pmcdev)
 	if (ret)
 		goto release_dev;
 
-	ssram_base = pcidev->resource[0].start;
-	ssram = ioremap(ssram_base, SSRAM_HDR_SIZE);
-	if (!ssram)
-		goto disable_dev;
-
 	pmcdev->ssram_pcidev = pcidev;
 
-	pmc_core_ssram_get_pmc(pmcdev, ssram, 0, PMC_IDX_SOC);
-	pmc_core_ssram_get_pmc(pmcdev, ssram, SSRAM_IOE_OFFSET, PMC_IDX_IOE);
-	pmc_core_ssram_get_pmc(pmcdev, ssram, SSRAM_PCH_OFFSET, PMC_IDX_PCH);
+	ret = pmc_core_ssram_get_pmc(pmcdev, PMC_IDX_MAIN, 0);
+	if (ret)
+		goto disable_dev;
 
-	iounmap(ssram);
+	pmc_core_ssram_get_pmc(pmcdev, PMC_IDX_IOE, SSRAM_IOE_OFFSET);
+	pmc_core_ssram_get_pmc(pmcdev, PMC_IDX_PCH, SSRAM_PCH_OFFSET);
 
 	return 0;
 
Signature · 1 line
2.34.1
author maintainer
#20 lore raw

PATCH v6 14/20 platform/x86/intel/pmc/mtl: Use return value from pmc_core_ssram_init()

In reply to David E. Box

Instead of checking for a NULL regbase, use the return value from pmc_core_ssram_init() to check if PMC discovery was successful. If not, use the legacy enumeration method (which only works for the primary PMC).

---
V6 - No change

V5 - New patch. Split from previous.

 drivers/platform/x86/intel/pmc/mtl.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
drivers/platform/x86/intel/pmc/mtl.c +8 -4
index c3b5f4fe01d1..d1d3d33fb4b8 100644
@@ -990,12 +990,16 @@ int mtl_core_init(struct pmc_dev *pmcdev)
 	mtl_d3_fixup();
 
 	pmcdev->resume = mtl_resume;
-
 	pmcdev->regmap_list = mtl_pmc_info_list;
-	pmc_core_ssram_init(pmcdev);
 
-	/* If regbase not assigned, set map and discover using legacy method */
-	if (!pmc->regbase) {
+	/*
+	 * If ssram init fails use legacy method to at least get the
+	 * primary PMC
+	 */
+	ret = pmc_core_ssram_init(pmcdev);
+	if (ret) {
+		dev_warn(&pmcdev->pdev->dev,
+			 "ssram init failed, %d, using legacy init\n", ret);
 		pmc->map = &mtl_socm_reg_map;
 		ret = get_primary_reg_base(pmc);
 		if (ret)
Signature · 1 line
2.34.1
author maintainer
#21 lore raw

PATCH v6 15/20 platform/x86/intel/pmc: Find and register PMC telemetry entries

In reply to David E. Box

The PMC SSRAM device contains counters that are structured in Intel Platform Monitoring Technology (PMT) telemetry regions. Look for and register these telemetry regions from the driver so that they may be read using the Intel PMT ABI.

---
V6 - No change

V5 - no change

V4 - no change

V3 - no change

V2 - no change

 drivers/platform/x86/intel/pmc/Kconfig      |  1 +
 drivers/platform/x86/intel/pmc/core_ssram.c | 49 +++++++++++++++++++++
 2 files changed, 50 insertions(+)
drivers/platform/x86/intel/pmc/Kconfig +1 -0
index b526597e4deb..d2f651fbec2c 100644
@@ -7,6 +7,7 @@ config INTEL_PMC_CORE
 	tristate "Intel PMC Core driver"
 	depends on PCI
 	depends on ACPI
+	depends on INTEL_PMT_TELEMETRY
 	help
 	  The Intel Platform Controller Hub for Intel Core SoCs provides access
 	  to Power Management Controller registers via various interfaces. This
drivers/platform/x86/intel/pmc/core_ssram.c +49 -0

70 lines not on this page · raw message

Signature · 1 line
2.34.1
author maintainer
#22 lore raw

PATCH v6 16/20 platform/x86/intel/pmc: Display LPM requirements for multiple PMCs

In reply to David E. Box

From: Rajvi Jingar <[email protected]>

Update the substate_requirements attribute to display the requirements for all the PMCs on a package.

---
V6 - Add comment to document reason to check lpm_req_regs for NULL again

V5 - Use conditional instead of if/else to do seq_printf()

V4 - No change

V3 - Add missing submitter signoff

V2 - no change

 drivers/platform/x86/intel/pmc/core.c | 123 ++++++++++++++------------
 1 file changed, 68 insertions(+), 55 deletions(-)
drivers/platform/x86/intel/pmc/core.c +68 -55
index 3894119d61b0..9d3a1b6ef622 100644
@@ -728,7 +728,7 @@ static int pmc_core_substate_l_sts_regs_show(struct seq_file *s, void *unused)
 }
 DEFINE_SHOW_ATTRIBUTE(pmc_core_substate_l_sts_regs);
 
-static void pmc_core_substate_req_header_show(struct seq_file *s)
+static void pmc_core_substate_req_header_show(struct seq_file *s, int pmc_index)
 {
 	struct pmc_dev *pmcdev = s->private;
 	int i, mode;
@@ -743,68 +743,81 @@ static void pmc_core_substate_req_header_show(struct seq_file *s)
 static int pmc_core_substate_req_regs_show(struct seq_file *s, void *unused)
 {
 	struct pmc_dev *pmcdev = s->private;
-	struct pmc *pmc = pmcdev->pmcs[PMC_IDX_MAIN];
-	const struct pmc_bit_map **maps = pmc->map->lpm_sts;
-	const struct pmc_bit_map *map;
-	const int num_maps = pmc->map->lpm_num_maps;
-	u32 sts_offset = pmc->map->lpm_status_offset;
-	u32 *lpm_req_regs = pmc->lpm_req_regs;
-	int mp;
+	u32 sts_offset;
+	u32 *lpm_req_regs;
+	int num_maps, mp, pmc_index;
 
-	/* Display the header */
-	pmc_core_substate_req_header_show(s);
+	for (pmc_index = 0; pmc_index < ARRAY_SIZE(pmcdev->pmcs); ++pmc_index) {
+		struct pmc *pmc = pmcdev->pmcs[pmc_index];
+		const struct pmc_bit_map **maps;
 
-	/* Loop over maps */
-	for (mp = 0; mp < num_maps; mp++) {
-		u32 req_mask = 0;
-		u32 lpm_status;
-		int mode, idx, i, len = 32;
+		if (!pmc)
+			continue;
+
+		maps = pmc->map->lpm_sts;
+		num_maps = pmc->map->lpm_num_maps;
+		sts_offset = pmc->map->lpm_status_offset;
+		lpm_req_regs = pmc->lpm_req_regs;
 
 		/*
-		 * Capture the requirements and create a mask so that we only
-		 * show an element if it's required for at least one of the
-		 * enabled low power modes
+		 * When there are multiple PMCs, though the PMC may exist, the
+		 * requirement register discovery could have failed so check
+		 * before accessing.
 		 */
-		pmc_for_each_mode(idx, mode, pmcdev)
-			req_mask |= lpm_req_regs[mp + (mode * num_maps)];
-
-		/* Get the last latched status for this map */
-		lpm_status = pmc_core_reg_read(pmc, sts_offset + (mp * 4));
-
-		/*  Loop over elements in this map */
-		map = maps[mp];
-		for (i = 0; map[i].name && i < len; i++) {
-			u32 bit_mask = map[i].bit_mask;
-
-			if (!(bit_mask & req_mask))
-				/*
-				 * Not required for any enabled states
-				 * so don't display
-				 */
-				continue;
-
-			/* Display the element name in the first column */
-			seq_printf(s, "%30s |", map[i].name);
-
-			/* Loop over the enabled states and display if required */
-			pmc_for_each_mode(idx, mode, pmcdev) {
-				if (lpm_req_regs[mp + (mode * num_maps)] & bit_mask)
-					seq_printf(s, " %9s |",
-						   "Required");
-				else
-					seq_printf(s, " %9s |", " ");
+		if (!lpm_req_regs)
+			continue;
+
+		/* Display the header */
+		pmc_core_substate_req_header_show(s, pmc_index);
+
+		/* Loop over maps */
+		for (mp = 0; mp < num_maps; mp++) {
+			u32 req_mask = 0;
+			u32 lpm_status;
+			const struct pmc_bit_map *map;
+			int mode, idx, i, len = 32;
+
+			/*
+			 * Capture the requirements and create a mask so that we only
+			 * show an element if it's required for at least one of the
+			 * enabled low power modes
+			 */
+			pmc_for_each_mode(idx, mode, pmcdev)
+				req_mask |= lpm_req_regs[mp + (mode * num_maps)];
+
+			/* Get the last latched status for this map */
+			lpm_status = pmc_core_reg_read(pmc, sts_offset + (mp * 4));
+
+			/*  Loop over elements in this map */
+			map = maps[mp];
+			for (i = 0; map[i].name && i < len; i++) {
+				u32 bit_mask = map[i].bit_mask;
+
+				if (!(bit_mask & req_mask)) {
+					/*
+					 * Not required for any enabled states
+					 * so don't display
+					 */
+					continue;
+				}
+
+				/* Display the element name in the first column */
+				seq_printf(s, "pmc%d: %26s |", pmc_index, map[i].name);
+
+				/* Loop over the enabled states and display if required */
+				pmc_for_each_mode(idx, mode, pmcdev) {
+					bool required = lpm_req_regs[mp + (mode * num_maps)] &
+							bit_mask;
+					seq_printf(s, " %9s |", required ? "Required" : " ");
+				}
+
+				/* In Status column, show the last captured state of this agent */
+				seq_printf(s, " %9s |", lpm_status & bit_mask ? "Yes" : " ");
+
+				seq_puts(s, "\n");
 			}
-
-			/* In Status column, show the last captured state of this agent */
-			if (lpm_status & bit_mask)
-				seq_printf(s, " %9s |", "Yes");
-			else
-				seq_printf(s, " %9s |", " ");
-
-			seq_puts(s, "\n");
 		}
 	}
-
 	return 0;
 }
 DEFINE_SHOW_ATTRIBUTE(pmc_core_substate_req_regs);
Signature · 1 line
2.34.1
author maintainer
#23 lore raw

PATCH v6 17/20 platform/x86/intel/pmc: Retrieve LPM information using Intel PMT

In reply to David E. Box

From: Xi Pardee <[email protected]>

On supported platforms, the low power mode (LPM) requirements for entering each idle substate are described in Platform Monitoring Technology (PMT) telemetry entries. Provide a function for platform code to attempt to find and read the requirements from the telemetry entries.

---
V6 - no change

V5 - no change

V4 - no change

V3 - no change

V2 - remove extra parens

 drivers/platform/x86/intel/pmc/core.h       |   3 +
 drivers/platform/x86/intel/pmc/core_ssram.c | 135 ++++++++++++++++++++
 2 files changed, 138 insertions(+)
drivers/platform/x86/intel/pmc/core.h +3 -0
index edaa70067e41..85b6f6ae4995 100644
@@ -320,6 +320,7 @@ struct pmc_reg_map {
 	const u32 lpm_status_offset;
 	const u32 lpm_live_status_offset;
 	const u32 etr3_offset;
+	const u8  *lpm_reg_index;
 };
 
 /**
@@ -329,6 +330,7 @@ struct pmc_reg_map {
  *			specific attributes
  */
 struct pmc_info {
+	u32 guid;
 	u16 devid;
 	const struct pmc_reg_map *map;
 };
@@ -486,6 +488,7 @@ extern const struct pmc_bit_map *mtl_ioem_lpm_maps[];
 extern const struct pmc_reg_map mtl_ioem_reg_map;
 
 extern void pmc_core_get_tgl_lpm_reqs(struct platform_device *pdev);
+extern int pmc_core_ssram_get_lpm_reqs(struct pmc_dev *pmcdev);
 extern int pmc_core_send_ltr_ignore(struct pmc_dev *pmcdev, u32 value);
 
 int pmc_core_resume_common(struct pmc_dev *pmcdev);
drivers/platform/x86/intel/pmc/core_ssram.c +135 -0

146 lines not on this page · raw message

Signature · 1 line
2.34.1
author maintainer
#24 lore raw

PATCH v6 18/20 platform/x86/intel/pmc: Read low power mode requirements for MTL-M and MTL-P

In reply to David E. Box

From: Xi Pardee <[email protected]>

Add support to read the low power mode requirements for Meteor Lake M and Meteor Lake P.

---
V6 - no change

V5 - no change

V4 - no change

V3 - directly return value from pmc_core_ssram_get_lpm_reqs()

V2 - fixed unused return value

 drivers/platform/x86/intel/pmc/mtl.c | 39 +++++++++++++++++++++++-----
 1 file changed, 32 insertions(+), 7 deletions(-)
drivers/platform/x86/intel/pmc/mtl.c +32 -7
index d1d3d33fb4b8..7ceeae507f4c 100644
@@ -11,6 +11,13 @@
 #include <linux/pci.h>
 #include "core.h"
 
+/* PMC SSRAM PMT Telemetry GUIDS */
+#define SOCP_LPM_REQ_GUID	0x2625030
+#define IOEM_LPM_REQ_GUID	0x4357464
+#define IOEP_LPM_REQ_GUID	0x5077612
+
+static const u8 MTL_LPM_REG_INDEX[] = {0, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, 16, 20};
+
 /*
  * Die Mapping to Product.
  * Product SOCDie IOEDie PCHDie
@@ -465,6 +472,7 @@ const struct pmc_reg_map mtl_socm_reg_map = {
 	.lpm_sts = mtl_socm_lpm_maps,
 	.lpm_status_offset = MTL_LPM_STATUS_OFFSET,
 	.lpm_live_status_offset = MTL_LPM_LIVE_STATUS_OFFSET,
+	.lpm_reg_index = MTL_LPM_REG_INDEX,
 };
 
 const struct pmc_bit_map mtl_ioep_pfear_map[] = {
@@ -782,6 +790,13 @@ const struct pmc_reg_map mtl_ioep_reg_map = {
 	.ltr_show_sts = mtl_ioep_ltr_show_map,
 	.ltr_ignore_offset = CNP_PMC_LTR_IGNORE_OFFSET,
 	.ltr_ignore_max = ADL_NUM_IP_IGN_ALLOWED,
+	.lpm_num_maps = ADL_LPM_NUM_MAPS,
+	.lpm_res_counter_step_x2 = TGL_PMC_LPM_RES_COUNTER_STEP_X2,
+	.lpm_residency_offset = MTL_LPM_RESIDENCY_OFFSET,
+	.lpm_priority_offset = MTL_LPM_PRI_OFFSET,
+	.lpm_en_offset = MTL_LPM_EN_OFFSET,
+	.lpm_sts_latch_en_offset = MTL_LPM_STATUS_LATCH_EN_OFFSET,
+	.lpm_reg_index = MTL_LPM_REG_INDEX,
 };
 
 const struct pmc_bit_map mtl_ioem_pfear_map[] = {
@@ -922,6 +937,13 @@ const struct pmc_reg_map mtl_ioem_reg_map = {
 	.ltr_show_sts = mtl_ioep_ltr_show_map,
 	.ltr_ignore_offset = CNP_PMC_LTR_IGNORE_OFFSET,
 	.ltr_ignore_max = ADL_NUM_IP_IGN_ALLOWED,
+	.lpm_sts_latch_en_offset = MTL_LPM_STATUS_LATCH_EN_OFFSET,
+	.lpm_num_maps = ADL_LPM_NUM_MAPS,
+	.lpm_priority_offset = MTL_LPM_PRI_OFFSET,
+	.lpm_en_offset = MTL_LPM_EN_OFFSET,
+	.lpm_res_counter_step_x2 = TGL_PMC_LPM_RES_COUNTER_STEP_X2,
+	.lpm_residency_offset = MTL_LPM_RESIDENCY_OFFSET,
+	.lpm_reg_index = MTL_LPM_REG_INDEX,
 };
 
 #define PMC_DEVID_SOCM	0x7e7f
@@ -929,16 +951,19 @@ const struct pmc_reg_map mtl_ioem_reg_map = {
 #define PMC_DEVID_IOEM	0x7ebf
 static struct pmc_info mtl_pmc_info_list[] = {
 	{
-		.devid = PMC_DEVID_SOCM,
-		.map = &mtl_socm_reg_map,
+		.guid	= SOCP_LPM_REQ_GUID,
+		.devid	= PMC_DEVID_SOCM,
+		.map	= &mtl_socm_reg_map,
 	},
 	{
-		.devid = PMC_DEVID_IOEP,
-		.map = &mtl_ioep_reg_map,
+		.guid	= IOEP_LPM_REQ_GUID,
+		.devid	= PMC_DEVID_IOEP,
+		.map	= &mtl_ioep_reg_map,
 	},
 	{
-		.devid = PMC_DEVID_IOEM,
-		.map = &mtl_ioem_reg_map
+		.guid	= IOEM_LPM_REQ_GUID,
+		.devid	= PMC_DEVID_IOEM,
+		.map	= &mtl_ioem_reg_map
 	},
 	{}
 };
@@ -1014,5 +1039,5 @@ int mtl_core_init(struct pmc_dev *pmcdev)
 	dev_dbg(&pmcdev->pdev->dev, "ignoring GBE LTR\n");
 	pmc_core_send_ltr_ignore(pmcdev, 3);
 
-	return 0;
+	return pmc_core_ssram_get_lpm_reqs(pmcdev);
 }
Signature · 1 line
2.34.1
author maintainer
#25 lore raw

PATCH v6 19/20 platform/x86/intel/pmc: Add debug attribute for Die C6 counter

In reply to David E. Box

Add a "die_c6_us_show" debugfs attribute. Reads the counter value using Intel Platform Monitoring Technology (PMT) driver API. This counter is useful for determining the idle residency of CPUs in the compute tile. Also adds a missing forward declaration for punit_ep which was declared in an earlier upstream commit but only used for the first time in this one.

---
V6 - No change

V5 - Change comment for crystal error and return value

V4 - no change

V3 - Split previous PATCH V2 13. Separates implementation (this patch) from
platform specific use (next patch)

V2 - Remove use of __func__
- Use HZ_PER_MHZ
- Fix missing newlines in printks

 drivers/platform/x86/intel/pmc/core.c | 55 +++++++++++++++++++++++++++
 drivers/platform/x86/intel/pmc/core.h |  4 ++
 2 files changed, 59 insertions(+)
drivers/platform/x86/intel/pmc/core.c +55 -0
index 9d3a1b6ef622..983e3a8f4910 100644
@@ -20,6 +20,7 @@
 #include <linux/pci.h>
 #include <linux/slab.h>
 #include <linux/suspend.h>
+#include <linux/units.h>
 
 #include <asm/cpu_device_id.h>
 #include <asm/intel-family.h>
@@ -27,6 +28,7 @@
 #include <asm/tsc.h>
 
 #include "core.h"
+#include "../pmt/telemetry.h"
 
 /* Maximum number of modes supported by platfoms that has low power mode capability */
 const char *pmc_lpm_modes[] = {
@@ -822,6 +824,47 @@ static int pmc_core_substate_req_regs_show(struct seq_file *s, void *unused)
 }
 DEFINE_SHOW_ATTRIBUTE(pmc_core_substate_req_regs);
 
+static unsigned int pmc_core_get_crystal_freq(void)
+{
+	unsigned int eax_denominator, ebx_numerator, ecx_hz, edx;
+
+	if (boot_cpu_data.cpuid_level < 0x15)
+		return 0;
+
+	eax_denominator = ebx_numerator = ecx_hz = edx = 0;
+
+	/* CPUID 15H TSC/Crystal ratio, plus optionally Crystal Hz */
+	cpuid(0x15, &eax_denominator, &ebx_numerator, &ecx_hz, &edx);
+
+	if (ebx_numerator == 0 || eax_denominator == 0)
+		return 0;
+
+	return ecx_hz;
+}
+
+static int pmc_core_die_c6_us_show(struct seq_file *s, void *unused)
+{
+	struct pmc_dev *pmcdev = s->private;
+	u64 die_c6_res, count;
+	int ret;
+
+	if (!pmcdev->crystal_freq) {
+		dev_warn_once(&pmcdev->pdev->dev, "Crystal frequency unavailable\n");
+		return -ENXIO;
+	}
+
+	ret = pmt_telem_read(pmcdev->punit_ep, pmcdev->die_c6_offset,
+			     &count, 1);
+	if (ret)
+		return ret;
+
+	die_c6_res = div64_u64(count * HZ_PER_MHZ, pmcdev->crystal_freq);
+	seq_printf(s, "%llu\n", die_c6_res);
+
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(pmc_core_die_c6_us);
+
 static int pmc_core_lpm_latch_mode_show(struct seq_file *s, void *unused)
 {
 	struct pmc_dev *pmcdev = s->private;
@@ -1118,6 +1161,12 @@ static void pmc_core_dbgfs_register(struct pmc_dev *pmcdev)
 				    pmcdev->dbgfs_dir, pmcdev,
 				    &pmc_core_substate_req_regs_fops);
 	}
+
+	if (pmcdev->has_die_c6) {
+		debugfs_create_file("die_c6_us_show", 0444,
+				    pmcdev->dbgfs_dir, pmcdev,
+				    &pmc_core_die_c6_us_fops);
+	}
 }
 
 static const struct x86_cpu_id intel_pmc_core_ids[] = {
@@ -1212,6 +1261,10 @@ static void pmc_core_clean_structure(struct platform_device *pdev)
 		pci_dev_put(pmcdev->ssram_pcidev);
 		pci_disable_device(pmcdev->ssram_pcidev);
 	}
+
+	if (pmcdev->punit_ep)
+		pmt_telem_unregister_endpoint(pmcdev->punit_ep);
+
 	platform_set_drvdata(pdev, NULL);
 	mutex_destroy(&pmcdev->lock);
 }
@@ -1232,6 +1285,8 @@ static int pmc_core_probe(struct platform_device *pdev)
 	if (!pmcdev)
 		return -ENOMEM;
 
+	pmcdev->crystal_freq = pmc_core_get_crystal_freq();
+
 	platform_set_drvdata(pdev, pmcdev);
 	pmcdev->pdev = pdev;
 
drivers/platform/x86/intel/pmc/core.h +4 -0
index 85b6f6ae4995..6d7673145f90 100644
@@ -16,6 +16,8 @@
 #include <linux/bits.h>
 #include <linux/platform_device.h>
 
+struct telem_endpoint;
+
 #define SLP_S0_RES_COUNTER_MASK			GENMASK(31, 0)
 
 #define PMC_BASE_ADDR_DEFAULT			0xFE000000
@@ -357,6 +359,7 @@ struct pmc {
  * @devs:		pointer to an array of pmc pointers
  * @pdev:		pointer to platform_device struct
  * @ssram_pcidev:	pointer to pci device struct for the PMC SSRAM
+ * @crystal_freq:	crystal frequency from cpuid
  * @dbgfs_dir:		path to debugfs interface
  * @pmc_xram_read_bit:	flag to indicate whether PMC XRAM shadow registers
  *			used to read MPHY PG and PLL status are available
@@ -374,6 +377,7 @@ struct pmc_dev {
 	struct dentry *dbgfs_dir;
 	struct platform_device *pdev;
 	struct pci_dev *ssram_pcidev;
+	unsigned int crystal_freq;
 	int pmc_xram_read_bit;
 	struct mutex lock; /* generic mutex lock for PMC Core */
 
Signature · 1 line
2.34.1
author maintainer
#26 lore raw

PATCH v6 20/20 platform/x86/intel/pmc: Show Die C6 counter on Meteor Lake

In reply to David E. Box

Expose the Die C6 counter on Meteor Lake.

---
V6 - no change

V5 - no change

V4 - no change

V3 - Split PATCH V2 13. Separates implementation (previous patch)
from platform specific use (this patch)

 drivers/platform/x86/intel/pmc/mtl.c | 32 ++++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)
drivers/platform/x86/intel/pmc/mtl.c +32 -0
index 7ceeae507f4c..38c2f946ec23 100644
@@ -10,12 +10,17 @@
 
 #include <linux/pci.h>
 #include "core.h"
+#include "../pmt/telemetry.h"
 
 /* PMC SSRAM PMT Telemetry GUIDS */
 #define SOCP_LPM_REQ_GUID	0x2625030
 #define IOEM_LPM_REQ_GUID	0x4357464
 #define IOEP_LPM_REQ_GUID	0x5077612
 
+/* Die C6 from PUNIT telemetry */
+#define MTL_PMT_DMU_DIE_C6_OFFSET	15
+#define MTL_PMT_DMU_GUID		0x1A067102
+
 static const u8 MTL_LPM_REG_INDEX[] = {0, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, 16, 20};
 
 /*
@@ -968,6 +973,32 @@ static struct pmc_info mtl_pmc_info_list[] = {
 	{}
 };
 
+static void mtl_punit_pmt_init(struct pmc_dev *pmcdev)
+{
+	struct telem_endpoint *ep;
+	struct pci_dev *pcidev;
+
+	pcidev = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(10, 0));
+	if (!pcidev) {
+		dev_err(&pmcdev->pdev->dev, "PUNIT PMT device not found.\n");
+		return;
+	}
+
+	ep = pmt_telem_find_and_register_endpoint(pcidev, MTL_PMT_DMU_GUID, 0);
+	if (IS_ERR(ep)) {
+		dev_err(&pmcdev->pdev->dev,
+			"pmc_core: couldn't get DMU telem endpoint, %ld\n",
+			PTR_ERR(ep));
+		return;
+	}
+
+	pci_dev_put(pcidev);
+	pmcdev->punit_ep = ep;
+
+	pmcdev->has_die_c6 = true;
+	pmcdev->die_c6_offset = MTL_PMT_DMU_DIE_C6_OFFSET;
+}
+
 #define MTL_GNA_PCI_DEV	0x7e4c
 #define MTL_IPU_PCI_DEV	0x7d19
 #define MTL_VPU_PCI_DEV	0x7d1d
@@ -1032,6 +1063,7 @@ int mtl_core_init(struct pmc_dev *pmcdev)
 	}
 
 	pmc_core_get_low_power_modes(pmcdev);
+	mtl_punit_pmt_init(pmcdev);
 
 	/* Due to a hardware limitation, the GBE LTR blocks PC10
 	 * when a cable is attached. Tell the PMC to ignore it.
Signature · 1 line
2.34.1
#27 lore raw
In reply to David E. Box

Hi,

Quoted 33 lines · David E. Box

On newer Intel silicon, more IP counters are being added in Intel Platform Monitoring Technology (PMT) telemetry spaces hosted in MMIO. There is a need for the intel_pmc_core driver and other drivers to access PMT hosted telemetry in the kernel using an API. This patchset adds driver APIs to allow registering and reading telemetry entries. It makes changes to the intel_pmc_core driver to use these interfaces to access the low power mode counters that are now exclusively available from PMT.

David E. Box (15):
platform/x86/intel/vsec: Fix xa_alloc memory leak platform/x86/intel/vsec: Remove unnecessary return platform/x86/intel/vsec: Move structures to header platform/x86/intel/vsec: remove platform_info from vsec device
structure
platform/x86/intel/vsec: Use cleanup.h platform/x86/intel/vsec: Assign auxdev parent by argument platform/x86/intel/vsec: Add base address field platform/x86/intel/pmt: Add header to struct intel_pmt_entry platform/x86/intel/pmt: telemetry: Export API to read telemetry platform/x86/intel/pmc: Allow pmc_core_ssram_init to fail platform/x86/intel/pmc: Cleanup SSRAM discovery platform/x86/intel/pmc/mtl: Use return value from
pmc_core_ssram_init()
platform/x86/intel/pmc: Find and register PMC telemetry entries platform/x86/intel/pmc: Add debug attribute for Die C6 counter platform/x86/intel/pmc: Show Die C6 counter on Meteor Lake

Gayatri Kammela (1):
platform/x86/intel/vsec: Add intel_vsec_register

Rajvi Jingar (1):
platform/x86/intel/pmc: Display LPM requirements for multiple PMCs

Quoted 6 lines

Xi Pardee (3):
platform/x86:intel/pmc: Call pmc_get_low_power_modes from platform
init
platform/x86/intel/pmc: Retrieve LPM information using Intel PMT platform/x86/intel/pmc: Read low power mode requirements for MTL-M and
MTL-P

Thank you for your patch-series, I've applied the series to my review-hans branch:
https://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86.git/log/?h=review-hans

Once I've run some tests on this branch the patches there will be added to the platform-drivers-x86/for-next branch and eventually will be included in the pdx86 pull-request to Linus for the next merge-window.

Regards,

Hans

Quoted 23 lines
 drivers/platform/x86/intel/pmc/Kconfig      |   1 +
 drivers/platform/x86/intel/pmc/adl.c        |   2 +
 drivers/platform/x86/intel/pmc/cnp.c        |   2 +
 drivers/platform/x86/intel/pmc/core.c       | 185 +++++++++-----
 drivers/platform/x86/intel/pmc/core.h       |  10 +-
 drivers/platform/x86/intel/pmc/core_ssram.c | 265 +++++++++++++++++---
 drivers/platform/x86/intel/pmc/icl.c        |  10 +-
 drivers/platform/x86/intel/pmc/mtl.c        |  87 ++++++-
 drivers/platform/x86/intel/pmc/spt.c        |  10 +-
 drivers/platform/x86/intel/pmc/tgl.c        |   1 +
 drivers/platform/x86/intel/pmt/class.c      |  43 +++-
 drivers/platform/x86/intel/pmt/class.h      |  30 ++-
 drivers/platform/x86/intel/pmt/crashlog.c   |   2 +-
 drivers/platform/x86/intel/pmt/telemetry.c  | 193 +++++++++++++-
 drivers/platform/x86/intel/pmt/telemetry.h  | 126 ++++++++++
 drivers/platform/x86/intel/vsec.c           | 131 +++++-----
 drivers/platform/x86/intel/vsec.h           |  45 +++-
 17 files changed, 939 insertions(+), 204 deletions(-)
 create mode 100644 drivers/platform/x86/intel/pmt/telemetry.h

base-commit: b85ea95d086471afb4ad062012a4d73cd328fa86