[RESEND PATCH v5 0/6] kvm powerpc/book3s-hv: Expose Hostwide counters as perf-events

[RESEND PATCH v5 0/6] kvm powerpc/book3s-hv: Expose Hostwide counters as perf-events

[Vaibhav Jain]

Changes from v5
Link: https://lore.kernel.org/all/20250317100834.451452-1-vaibhav@linux.ibm.com/
* Rebased to latest upstream kernel tree
* Added review-by from Athira to a patch.
============

Cover Letter
============

This patch-series adds support for reporting Hostwide(L1-Lpar) counters via
perf-events in a new kernel module named 'kvm-hv-pmu'. With the support for
running KVM Guest in a PSeries-Lpar using nested-APIv2 via [1], the
underlying L0-PowerVM hypervisor holds some state information pertaining to
all running L2-KVM Guests in an L1-Lpar. This state information is held in
a pre-allocated memory thats owned by L0-PowerVM and is termed as
Guest-Management-Area(GMA). The GMA is allocated per L1-LPAR and is only
allocated if the lpar is KVM enabled. The size of this area is a fixed
percentage of the memory assigned to the KVM enabled L1-lpar and is
composed of two major components, Guest Management Space(Host-Heap) and
Guest Page Table Management Space(Host-Pagetable).

The Host-Heap holds various data-structures allocated by L0-PowerVM for
L2-KVM Guests running in the L1-Lpar. The Host-Pagetable holds the Radix
pagetable[2] for the L2-KVM Guest which is used by L0-PowerVM to handle
page faults. Since the size of both of these areas is limited and fixed via
partition boot profile, it puts an upper bound on the number of L2-KVM
Guests that can be run in an LPAR. Also due limited size of Host-Pagetable
area, L0-PowerVM is at times forced to perform reclaim operation on
it. This reclaim operation is usually performed when running large number
of L2-KVM Guests which are memory bound and increases Host-Pagetable
utilization.

In light of the above its recommended to track usage of these areas to
ensure consistent L2-KVM Guest performance. Hence this patch-series
attempts to expose the max-size and current-usage of these areas as well as
cumulative amount of bytes reclaimed from Host-Pagetable as perf-events
that can be queried via perf-stat.

The patch series introduces a new 'kvm-hv' PMU that hosted in a new kernel
module named 'kvm-hv-pmu' which exports the perf-events mentioned
below. Since perf-events exported represents the state of the whole L1-Lpar
and not that of a specific L2-KVM guest hence the 'kvm-hv' PMU's scope is
set as PERF_PMU_SCOPE_SYS_WIDE(System-Wide).

New perf-events introduced
==========================

* kvm-hv/host_heap/		: The currently used bytes in the
				  Hypervisor's Guest Management Space
				  associated with the Host Partition.
* kvm-hv/host_heap_max/		: The maximum bytes available in the
				  Hypervisor's Guest Management Space
				  associated with the Host Partition.
* kvm-hv/host_pagetable/	: The currently used bytes in the
				  Hypervisor's Guest Page Table Management
				  Space associated with the Host Partition.
* kvm-hv/host_pagetable_max/	: The maximum bytes available in the
				  Hypervisor's Guest Page Table Management
				  Space associated with the Host Partition.
* kvm-hv/host_pagetable_reclaim/: The amount of space in bytes that has
				  been reclaimed due to overcommit in the
				  Hypervisor's Guest Page Table Management
				  Space associated with the Host Partition.

Structure of this patch series
==============================
Start with documenting and updating the KVM nested-APIv2 hcall
specifications for H_GUEST_GET_STATE hcall and Hostwide guest-state-buffer
elements.

Subsequent patches add support for adding and parsing Hostwide
guest-state-buffer elements in existing kvm-hv apiv2 infrastructure. Also
add a kunit test case to verify correctness of the changes introduced.

Next set of patches in the patch-set introduces a new PMU (kvm-hv) and
kernel module (kvm-hv-pmu), implements plumbing in the kvm-hv-pmu module to
create populate and parse a guest-state-buffer holding the Hostwide
counters returned from L0-PowerVM.

The final patch in the series creates the five new perf-events which then
leverage the kernel's perf-event infrastructure to report the Hostwide
counters returned from L0-PowerVM to perf tool.

Output
======
Once the patch-set is integrated, perf-stat should report the Hostwide
counters for a kvm-enabled pseries lpar with kvm-hv-pmu module loaded as
below:

$ sudo perf stat -e 'kvm-hv/host_heap/'  -e 'kvm-hv/host_heap_max/' \
  -e 'kvm-hv/host_pagetable/' -e 'kvm-hv/host_pagetable_max/' \
  -e 'kvm-hv/host_pagetable_reclaim/' -- sleep 0

Performance counter stats for 'system wide':

                 0      kvm-hv/host_heap/
    10,995,367,936      kvm-hv/host_heap_max/
         2,178,304      kvm-hv/host_pagetable/
     2,147,483,648      kvm-hv/host_pagetable_max/
                 0      kvm-hv/host_pagetable_reclaim/

The patch can be tested with Qemu-TCG emulation support for Book3s-HV APIv2
proposed at [3]. Currently with Qemu-TCG the values for all the Hostwide
counters is reported as '0'.

References
==========
[1] - commit 19d31c5f1157 ("KVM: PPC: Add support for nestedv2 guests")
[2] - "KVM in a PowerVM LPAR: A Power user guide Part II"
      https://ibm.biz/BdGHeY
[3] - https://lore.kernel.org/all/20250221155449.530645-1-vaibhav@linux.ibm.com

Vaibhav Jain (6):
  powerpc: Document APIv2 KVM hcall spec for Hostwide counters
  kvm powerpc/book3s-apiv2: Add support for Hostwide GSB elements
  kvm powerpc/book3s-apiv2: Add kunit tests for Hostwide GSB elements
  kvm powerpc/book3s-apiv2: Introduce kvm-hv specific PMU
  powerpc/kvm-hv-pmu: Implement GSB message-ops for hostwide counters
  powerpc/kvm-hv-pmu: Add perf-events for Hostwide counters

 Documentation/arch/powerpc/kvm-nested.rst     |  40 +-
 arch/powerpc/include/asm/guest-state-buffer.h |  35 +-
 arch/powerpc/include/asm/hvcall.h             |  13 +-
 arch/powerpc/kvm/Kconfig                      |  13 +
 arch/powerpc/kvm/book3s_hv_nestedv2.c         |   6 +
 arch/powerpc/kvm/guest-state-buffer.c         |  39 ++
 arch/powerpc/kvm/test-guest-state-buffer.c    | 214 +++++++++
 arch/powerpc/perf/Makefile                    |   2 +
 arch/powerpc/perf/kvm-hv-pmu.c                | 435 ++++++++++++++++++
 9 files changed, 775 insertions(+), 22 deletions(-)
 create mode 100644 arch/powerpc/perf/kvm-hv-pmu.c

-- 
2.49.0

[RESEND PATCH v5 1/6] powerpc: Document APIv2 KVM hcall spec for Hostwide counters

[Vaibhav Jain]

Update kvm-nested APIv2 documentation to include five new
Guest-State-Elements to fetch the hostwide counters. These counters are
per L1-Lpar and indicate the amount of Heap/Page-table memory allocated,
available and Page-table memory reclaimed for all L2-Guests active
instances

Cc: linux-doc@vger.kernel.org
Cc: linuxppc-dev@lists.ozlabs.org
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Bagas Sanjaya <bagasdotme@gmail.com>
Tested-by: Gautam Menghani <gautam@linux.ibm.com>
Signed-off-by: Vaibhav Jain <vaibhav@linux.ibm.com>
---
Changelog

v5->resend:
* Rebase the patch to latest upstream kernel tree

v4->v5:
* None

v3->v4:
* Added reviewed by [ Bagas Sanjaya ]

v2->v3:
* Minor reword for the cumulative reclaim counter [ Gautam ]

v1->v2:
* Reworded section on GSID [Gautam]
---
 Documentation/arch/powerpc/kvm-nested.rst | 40 +++++++++++++++++------
 1 file changed, 30 insertions(+), 10 deletions(-)

diff --git a/Documentation/arch/powerpc/kvm-nested.rst b/Documentation/arch/powerpc/kvm-nested.rst
index 5defd13cc6c1..574592505604 100644
--- a/Documentation/arch/powerpc/kvm-nested.rst
+++ b/Documentation/arch/powerpc/kvm-nested.rst
@@ -208,13 +208,9 @@ associated values for each ID in the GSB::
       flags:
          Bit 0: getGuestWideState: Request state of the Guest instead
            of an individual VCPU.
-         Bit 1: takeOwnershipOfVcpuState Indicate the L1 is taking
-           over ownership of the VCPU state and that the L0 can free
-           the storage holding the state. The VCPU state will need to
-           be returned to the Hypervisor via H_GUEST_SET_STATE prior
-           to H_GUEST_RUN_VCPU being called for this VCPU. The data
-           returned in the dataBuffer is in a Hypervisor internal
-           format.
+         Bit 1: getHostWideState: Request stats of the Host. This causes
+           the guestId and vcpuId parameters to be ignored and attempting
+           to get the VCPU/Guest state will cause an error.
          Bits 2-63: Reserved
       guestId: ID obtained from H_GUEST_CREATE
       vcpuId: ID of the vCPU pass to H_GUEST_CREATE_VCPU
@@ -406,9 +402,10 @@ the partition like the timebase offset and partition scoped page
 table information.
 
 +--------+-------+----+--------+----------------------------------+
-|   ID   | Size  | RW | Thread | Details                          |
-|        | Bytes |    | Guest  |                                  |
-|        |       |    | Scope  |                                  |
+|   ID   | Size  | RW |(H)ost  | Details                          |
+|        | Bytes |    |(G)uest |                                  |
+|        |       |    |(T)hread|                                  |
+|        |       |    |Scope   |                                  |
 +========+=======+====+========+==================================+
 | 0x0000 |       | RW |   TG   | NOP element                      |
 +--------+-------+----+--------+----------------------------------+
@@ -434,6 +431,29 @@ table information.
 |        |       |    |        |- 0x8 Table size.                 |
 +--------+-------+----+--------+----------------------------------+
 | 0x0007-|       |    |        | Reserved                         |
+| 0x07FF |       |    |        |                                  |
++--------+-------+----+--------+----------------------------------+
+| 0x0800 | 0x08  | R  |   H    | Current usage in bytes of the    |
+|        |       |    |        | L0's Guest Management Space      |
+|        |       |    |        | for an L1-Lpar.                  |
++--------+-------+----+--------+----------------------------------+
+| 0x0801 | 0x08  | R  |   H    | Max bytes available in the       |
+|        |       |    |        | L0's Guest Management Space for  |
+|        |       |    |        | an L1-Lpar                       |
++--------+-------+----+--------+----------------------------------+
+| 0x0802 | 0x08  | R  |   H    | Current usage in bytes of the    |
+|        |       |    |        | L0's Guest Page Table Management |
+|        |       |    |        | Space for an L1-Lpar             |
++--------+-------+----+--------+----------------------------------+
+| 0x0803 | 0x08  | R  |   H    | Max bytes available in the L0's  |
+|        |       |    |        | Guest Page Table Management      |
+|        |       |    |        | Space for an L1-Lpar             |
++--------+-------+----+--------+----------------------------------+
+| 0x0804 | 0x08  | R  |   H    | Cumulative Reclaimed bytes from  |
+|        |       |    |        | L0 Guest's Page Table Management |
+|        |       |    |        | Space due to overcommit          |
++--------+-------+----+--------+----------------------------------+
+| 0x0805-|       |    |        | Reserved                         |
 | 0x0BFF |       |    |        |                                  |
 +--------+-------+----+--------+----------------------------------+
 | 0x0C00 | 0x10  | RW |   T    |Run vCPU Input Buffer:            |
-- 
2.49.0

[RESEND PATCH v5 2/6] kvm powerpc/book3s-apiv2: Add support for Hostwide GSB elements

[Vaibhav Jain]

Add support for adding and parsing Hostwide elements to the
Guest-state-buffer data structure used in apiv2. These elements are used to
share meta-information pertaining to entire L1-Lpar and this
meta-information is maintained by L0-PowerVM hypervisor. Example of this
include the amount of the page-table memory currently used by L0-PowerVM
for hosting the Shadow-Pagetable of all active L2-Guests. More of the are
documented in kernel-documentation at [1]. The Hostwide GSB elements are
currently only support with H_GUEST_SET_STATE hcall with a special flag
namely 'KVMPPC_GS_FLAGS_HOST_WIDE'.

The patch introduces new defs for the 5 new Hostwide GSB elements including
their GSIDs as well as introduces a new class of GSB elements namely
'KVMPPC_GS_CLASS_HOSTWIDE' to indicate to GSB construction/parsing
infrastructure in 'kvm/guest-state-buffer.c'. Also
gs_msg_ops_vcpu_get_size(), kvmppc_gsid_type() and
kvmppc_gse_{flatten,unflatten}_iden() are updated to appropriately indicate
the needed size for these Hostwide GSB elements as well as how to
flatten/unflatten their GSIDs so that they can be marked as available in
GSB bitmap.

[1] Documention/arch/powerpc/kvm-nested.rst

Signed-off-by: Vaibhav Jain <vaibhav@linux.ibm.com>
---
Changelog

v5->resend:
* Rebase the patch to latest upstream kernel tree

v4->v5:
None

v3->v4:
None

v2->v3:
None

v1->v2:
None
---
 arch/powerpc/include/asm/guest-state-buffer.h | 35 ++++++++++++++---
 arch/powerpc/include/asm/hvcall.h             | 13 ++++---
 arch/powerpc/kvm/book3s_hv_nestedv2.c         |  6 +++
 arch/powerpc/kvm/guest-state-buffer.c         | 39 +++++++++++++++++++
 4 files changed, 81 insertions(+), 12 deletions(-)

diff --git a/arch/powerpc/include/asm/guest-state-buffer.h b/arch/powerpc/include/asm/guest-state-buffer.h
index d107abe1468f..acd61eb36d59 100644
--- a/arch/powerpc/include/asm/guest-state-buffer.h
+++ b/arch/powerpc/include/asm/guest-state-buffer.h
@@ -28,6 +28,21 @@
  /* Process Table Info */
 #define KVMPPC_GSID_PROCESS_TABLE		0x0006
 
+/* Guest Management Heap Size */
+#define KVMPPC_GSID_L0_GUEST_HEAP		0x0800
+
+/* Guest Management Heap Max Size */
+#define KVMPPC_GSID_L0_GUEST_HEAP_MAX		0x0801
+
+/* Guest Pagetable Size */
+#define KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE	0x0802
+
+/* Guest Pagetable Max Size */
+#define KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX	0x0803
+
+/* Guest Pagetable Reclaim in bytes */
+#define KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM	0x0804
+
 /* H_GUEST_RUN_VCPU input buffer Info */
 #define KVMPPC_GSID_RUN_INPUT			0x0C00
 /* H_GUEST_RUN_VCPU output buffer Info */
@@ -106,6 +121,11 @@
 #define KVMPPC_GSE_GUESTWIDE_COUNT \
 	(KVMPPC_GSE_GUESTWIDE_END - KVMPPC_GSE_GUESTWIDE_START + 1)
 
+#define KVMPPC_GSE_HOSTWIDE_START KVMPPC_GSID_L0_GUEST_HEAP
+#define KVMPPC_GSE_HOSTWIDE_END KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM
+#define KVMPPC_GSE_HOSTWIDE_COUNT \
+	(KVMPPC_GSE_HOSTWIDE_END - KVMPPC_GSE_HOSTWIDE_START + 1)
+
 #define KVMPPC_GSE_META_START KVMPPC_GSID_RUN_INPUT
 #define KVMPPC_GSE_META_END KVMPPC_GSID_VPA
 #define KVMPPC_GSE_META_COUNT (KVMPPC_GSE_META_END - KVMPPC_GSE_META_START + 1)
@@ -130,7 +150,8 @@
 	(KVMPPC_GSE_INTR_REGS_END - KVMPPC_GSE_INTR_REGS_START + 1)
 
 #define KVMPPC_GSE_IDEN_COUNT                                 \
-	(KVMPPC_GSE_GUESTWIDE_COUNT + KVMPPC_GSE_META_COUNT + \
+	(KVMPPC_GSE_HOSTWIDE_COUNT + \
+	 KVMPPC_GSE_GUESTWIDE_COUNT + KVMPPC_GSE_META_COUNT + \
 	 KVMPPC_GSE_DW_REGS_COUNT + KVMPPC_GSE_W_REGS_COUNT + \
 	 KVMPPC_GSE_VSRS_COUNT + KVMPPC_GSE_INTR_REGS_COUNT)
 
@@ -139,10 +160,11 @@
  */
 enum {
 	KVMPPC_GS_CLASS_GUESTWIDE = 0x01,
-	KVMPPC_GS_CLASS_META = 0x02,
-	KVMPPC_GS_CLASS_DWORD_REG = 0x04,
-	KVMPPC_GS_CLASS_WORD_REG = 0x08,
-	KVMPPC_GS_CLASS_VECTOR = 0x10,
+	KVMPPC_GS_CLASS_HOSTWIDE = 0x02,
+	KVMPPC_GS_CLASS_META = 0x04,
+	KVMPPC_GS_CLASS_DWORD_REG = 0x08,
+	KVMPPC_GS_CLASS_WORD_REG = 0x10,
+	KVMPPC_GS_CLASS_VECTOR = 0x18,
 	KVMPPC_GS_CLASS_INTR = 0x20,
 };
 
@@ -164,6 +186,7 @@ enum {
  */
 enum {
 	KVMPPC_GS_FLAGS_WIDE = 0x01,
+	KVMPPC_GS_FLAGS_HOST_WIDE = 0x02,
 };
 
 /**
@@ -287,7 +310,7 @@ struct kvmppc_gs_msg_ops {
  * struct kvmppc_gs_msg - a guest state message
  * @bitmap: the guest state ids that should be included
  * @ops: modify message behavior for reading and writing to buffers
- * @flags: guest wide or thread wide
+ * @flags: host wide, guest wide or thread wide
  * @data: location where buffer data will be written to or from.
  *
  * A guest state message is allows flexibility in sending in receiving data
diff --git a/arch/powerpc/include/asm/hvcall.h b/arch/powerpc/include/asm/hvcall.h
index eeef13db2770..6df6dbbe1e7c 100644
--- a/arch/powerpc/include/asm/hvcall.h
+++ b/arch/powerpc/include/asm/hvcall.h
@@ -490,14 +490,15 @@
 #define H_RPTI_PAGE_ALL (-1UL)
 
 /* Flags for H_GUEST_{S,G}_STATE */
-#define H_GUEST_FLAGS_WIDE     (1UL<<(63-0))
+#define H_GUEST_FLAGS_WIDE     (1UL << (63 - 0))
+#define H_GUEST_FLAGS_HOST_WIDE	(1UL << (63 - 1))
 
 /* Flag values used for H_{S,G}SET_GUEST_CAPABILITIES */
-#define H_GUEST_CAP_COPY_MEM	(1UL<<(63-0))
-#define H_GUEST_CAP_POWER9	(1UL<<(63-1))
-#define H_GUEST_CAP_POWER10	(1UL<<(63-2))
-#define H_GUEST_CAP_POWER11	(1UL<<(63-3))
-#define H_GUEST_CAP_BITMAP2	(1UL<<(63-63))
+#define H_GUEST_CAP_COPY_MEM	(1UL << (63 - 0))
+#define H_GUEST_CAP_POWER9	(1UL << (63 - 1))
+#define H_GUEST_CAP_POWER10	(1UL << (63 - 2))
+#define H_GUEST_CAP_POWER11	(1UL << (63 - 3))
+#define H_GUEST_CAP_BITMAP2	(1UL << (63 - 63))
 
 /*
  * Defines for H_HTM - Macros for hardware trace macro (HTM) function.
diff --git a/arch/powerpc/kvm/book3s_hv_nestedv2.c b/arch/powerpc/kvm/book3s_hv_nestedv2.c
index e5c7ce1fb761..87691cf86cae 100644
--- a/arch/powerpc/kvm/book3s_hv_nestedv2.c
+++ b/arch/powerpc/kvm/book3s_hv_nestedv2.c
@@ -123,6 +123,12 @@ static size_t gs_msg_ops_vcpu_get_size(struct kvmppc_gs_msg *gsm)
 		case KVMPPC_GSID_PROCESS_TABLE:
 		case KVMPPC_GSID_RUN_INPUT:
 		case KVMPPC_GSID_RUN_OUTPUT:
+		  /* Host wide counters */
+		case KVMPPC_GSID_L0_GUEST_HEAP:
+		case KVMPPC_GSID_L0_GUEST_HEAP_MAX:
+		case KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE:
+		case KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX:
+		case KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM:
 			break;
 		default:
 			size += kvmppc_gse_total_size(kvmppc_gsid_size(iden));
diff --git a/arch/powerpc/kvm/guest-state-buffer.c b/arch/powerpc/kvm/guest-state-buffer.c
index b80dbc58621f..871cf60ddeb6 100644
--- a/arch/powerpc/kvm/guest-state-buffer.c
+++ b/arch/powerpc/kvm/guest-state-buffer.c
@@ -92,6 +92,10 @@ static int kvmppc_gsid_class(u16 iden)
 	    (iden <= KVMPPC_GSE_GUESTWIDE_END))
 		return KVMPPC_GS_CLASS_GUESTWIDE;
 
+	if ((iden >= KVMPPC_GSE_HOSTWIDE_START) &&
+	    (iden <= KVMPPC_GSE_HOSTWIDE_END))
+		return KVMPPC_GS_CLASS_HOSTWIDE;
+
 	if ((iden >= KVMPPC_GSE_META_START) && (iden <= KVMPPC_GSE_META_END))
 		return KVMPPC_GS_CLASS_META;
 
@@ -118,6 +122,21 @@ static int kvmppc_gsid_type(u16 iden)
 	int type = -1;
 
 	switch (kvmppc_gsid_class(iden)) {
+	case KVMPPC_GS_CLASS_HOSTWIDE:
+		switch (iden) {
+		case KVMPPC_GSID_L0_GUEST_HEAP:
+			fallthrough;
+		case KVMPPC_GSID_L0_GUEST_HEAP_MAX:
+			fallthrough;
+		case KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE:
+			fallthrough;
+		case KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX:
+			fallthrough;
+		case KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM:
+			type = KVMPPC_GSE_BE64;
+			break;
+		}
+		break;
 	case KVMPPC_GS_CLASS_GUESTWIDE:
 		switch (iden) {
 		case KVMPPC_GSID_HOST_STATE_SIZE:
@@ -187,6 +206,9 @@ unsigned long kvmppc_gsid_flags(u16 iden)
 	case KVMPPC_GS_CLASS_GUESTWIDE:
 		flags = KVMPPC_GS_FLAGS_WIDE;
 		break;
+	case KVMPPC_GS_CLASS_HOSTWIDE:
+		flags = KVMPPC_GS_FLAGS_HOST_WIDE;
+		break;
 	case KVMPPC_GS_CLASS_META:
 	case KVMPPC_GS_CLASS_DWORD_REG:
 	case KVMPPC_GS_CLASS_WORD_REG:
@@ -310,6 +332,13 @@ static inline int kvmppc_gse_flatten_iden(u16 iden)
 
 	bit += KVMPPC_GSE_GUESTWIDE_COUNT;
 
+	if (class == KVMPPC_GS_CLASS_HOSTWIDE) {
+		bit += iden - KVMPPC_GSE_HOSTWIDE_START;
+		return bit;
+	}
+
+	bit += KVMPPC_GSE_HOSTWIDE_COUNT;
+
 	if (class == KVMPPC_GS_CLASS_META) {
 		bit += iden - KVMPPC_GSE_META_START;
 		return bit;
@@ -356,6 +385,12 @@ static inline u16 kvmppc_gse_unflatten_iden(int bit)
 	}
 	bit -= KVMPPC_GSE_GUESTWIDE_COUNT;
 
+	if (bit < KVMPPC_GSE_HOSTWIDE_COUNT) {
+		iden = KVMPPC_GSE_HOSTWIDE_START + bit;
+		return iden;
+	}
+	bit -= KVMPPC_GSE_HOSTWIDE_COUNT;
+
 	if (bit < KVMPPC_GSE_META_COUNT) {
 		iden = KVMPPC_GSE_META_START + bit;
 		return iden;
@@ -588,6 +623,8 @@ int kvmppc_gsb_send(struct kvmppc_gs_buff *gsb, unsigned long flags)
 
 	if (flags & KVMPPC_GS_FLAGS_WIDE)
 		hflags |= H_GUEST_FLAGS_WIDE;
+	if (flags & KVMPPC_GS_FLAGS_HOST_WIDE)
+		hflags |= H_GUEST_FLAGS_HOST_WIDE;
 
 	rc = plpar_guest_set_state(hflags, gsb->guest_id, gsb->vcpu_id,
 				   __pa(gsb->hdr), gsb->capacity, &i);
@@ -613,6 +650,8 @@ int kvmppc_gsb_recv(struct kvmppc_gs_buff *gsb, unsigned long flags)
 
 	if (flags & KVMPPC_GS_FLAGS_WIDE)
 		hflags |= H_GUEST_FLAGS_WIDE;
+	if (flags & KVMPPC_GS_FLAGS_HOST_WIDE)
+		hflags |= H_GUEST_FLAGS_HOST_WIDE;
 
 	rc = plpar_guest_get_state(hflags, gsb->guest_id, gsb->vcpu_id,
 				   __pa(gsb->hdr), gsb->capacity, &i);
-- 
2.49.0

[RESEND PATCH v5 3/6] kvm powerpc/book3s-apiv2: Add kunit tests for Hostwide GSB elements

[Vaibhav Jain]

Update 'test-guest-state-buffer.c' to add two new KUNIT test cases for
validating correctness of changes to Guest-state-buffer management
infrastructure for adding support for Hostwide GSB elements.

The newly introduced test test_gs_hostwide_msg() checks if the Hostwide
elements can be set and parsed from a Guest-state-buffer. The second kunit
test test_gs_hostwide_counters() checks if the Hostwide GSB elements can be
send to the L0-PowerVM hypervisor via the H_GUEST_SET_STATE hcall and
ensures that the returned guest-state-buffer has all the 5 Hostwide stat
counters present.

Below is the KATP test report with the newly added KUNIT tests:

KTAP version 1
    # Subtest: guest_state_buffer_test
    # module: test_guest_state_buffer
    1..7
    ok 1 test_creating_buffer
    ok 2 test_adding_element
    ok 3 test_gs_bitmap
    ok 4 test_gs_parsing
    ok 5 test_gs_msg
    ok 6 test_gs_hostwide_msg
    # test_gs_hostwide_counters: Guest Heap Size=0 bytes
    # test_gs_hostwide_counters: Guest Heap Size Max=10995367936 bytes
    # test_gs_hostwide_counters: Guest Page-table Size=2178304 bytes
    # test_gs_hostwide_counters: Guest Page-table Size Max=2147483648 bytes
    # test_gs_hostwide_counters: Guest Page-table Reclaim Size=0 bytes
    ok 7 test_gs_hostwide_counters
 # guest_state_buffer_test: pass:7 fail:0 skip:0 total:7
 # Totals: pass:7 fail:0 skip:0 total:7
 ok 1 guest_state_buffer_test

Signed-off-by: Vaibhav Jain <vaibhav@linux.ibm.com>
---
Changelog

v5->resend:
* Rebase the patch to latest upstream kernel tree

v4->v5:
None

v3->v4:
* Force skip of kunit test 'test_gs_hostwide_counters' on baremetal kvm-hv
as it was causing a kernel exception.

v2->v3:
None

v1->v2:
None
---
 arch/powerpc/kvm/test-guest-state-buffer.c | 214 +++++++++++++++++++++
 1 file changed, 214 insertions(+)

diff --git a/arch/powerpc/kvm/test-guest-state-buffer.c b/arch/powerpc/kvm/test-guest-state-buffer.c
index bfd225329a18..5ccca306997a 100644
--- a/arch/powerpc/kvm/test-guest-state-buffer.c
+++ b/arch/powerpc/kvm/test-guest-state-buffer.c
@@ -5,6 +5,7 @@
 #include <kunit/test.h>
 
 #include <asm/guest-state-buffer.h>
+#include <asm/kvm_ppc.h>
 
 static void test_creating_buffer(struct kunit *test)
 {
@@ -141,6 +142,16 @@ static void test_gs_bitmap(struct kunit *test)
 		i++;
 	}
 
+	for (u16 iden = KVMPPC_GSID_L0_GUEST_HEAP;
+	     iden <= KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM; iden++) {
+		kvmppc_gsbm_set(&gsbm, iden);
+		kvmppc_gsbm_set(&gsbm1, iden);
+		KUNIT_EXPECT_TRUE(test, kvmppc_gsbm_test(&gsbm, iden));
+		kvmppc_gsbm_clear(&gsbm, iden);
+		KUNIT_EXPECT_FALSE(test, kvmppc_gsbm_test(&gsbm, iden));
+		i++;
+	}
+
 	for (u16 iden = KVMPPC_GSID_RUN_INPUT; iden <= KVMPPC_GSID_VPA;
 	     iden++) {
 		kvmppc_gsbm_set(&gsbm, iden);
@@ -309,12 +320,215 @@ static void test_gs_msg(struct kunit *test)
 	kvmppc_gsm_free(gsm);
 }
 
+/* Test data struct for hostwide/L0 counters */
+struct kvmppc_gs_msg_test_hostwide_data {
+	u64 guest_heap;
+	u64 guest_heap_max;
+	u64 guest_pgtable_size;
+	u64 guest_pgtable_size_max;
+	u64 guest_pgtable_reclaim;
+};
+
+static size_t test_hostwide_get_size(struct kvmppc_gs_msg *gsm)
+
+{
+	size_t size = 0;
+	u16 ids[] = {
+		KVMPPC_GSID_L0_GUEST_HEAP,
+		KVMPPC_GSID_L0_GUEST_HEAP_MAX,
+		KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE,
+		KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX,
+		KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM
+	};
+
+	for (int i = 0; i < ARRAY_SIZE(ids); i++)
+		size += kvmppc_gse_total_size(kvmppc_gsid_size(ids[i]));
+	return size;
+}
+
+static int test_hostwide_fill_info(struct kvmppc_gs_buff *gsb,
+				   struct kvmppc_gs_msg *gsm)
+{
+	struct kvmppc_gs_msg_test_hostwide_data *data = gsm->data;
+
+	if (kvmppc_gsm_includes(gsm, KVMPPC_GSID_L0_GUEST_HEAP))
+		kvmppc_gse_put_u64(gsb, KVMPPC_GSID_L0_GUEST_HEAP,
+				   data->guest_heap);
+	if (kvmppc_gsm_includes(gsm, KVMPPC_GSID_L0_GUEST_HEAP_MAX))
+		kvmppc_gse_put_u64(gsb, KVMPPC_GSID_L0_GUEST_HEAP_MAX,
+				   data->guest_heap_max);
+	if (kvmppc_gsm_includes(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE))
+		kvmppc_gse_put_u64(gsb, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE,
+				   data->guest_pgtable_size);
+	if (kvmppc_gsm_includes(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX))
+		kvmppc_gse_put_u64(gsb, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX,
+				   data->guest_pgtable_size_max);
+	if (kvmppc_gsm_includes(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM))
+		kvmppc_gse_put_u64(gsb, KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM,
+				   data->guest_pgtable_reclaim);
+
+	return 0;
+}
+
+static int test_hostwide_refresh_info(struct kvmppc_gs_msg *gsm,
+				      struct kvmppc_gs_buff *gsb)
+{
+	struct kvmppc_gs_parser gsp = { 0 };
+	struct kvmppc_gs_msg_test_hostwide_data *data = gsm->data;
+	struct kvmppc_gs_elem *gse;
+	int rc;
+
+	rc = kvmppc_gse_parse(&gsp, gsb);
+	if (rc < 0)
+		return rc;
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_HEAP);
+	if (gse)
+		data->guest_heap = kvmppc_gse_get_u64(gse);
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_HEAP_MAX);
+	if (gse)
+		data->guest_heap_max = kvmppc_gse_get_u64(gse);
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE);
+	if (gse)
+		data->guest_pgtable_size = kvmppc_gse_get_u64(gse);
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX);
+	if (gse)
+		data->guest_pgtable_size_max = kvmppc_gse_get_u64(gse);
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM);
+	if (gse)
+		data->guest_pgtable_reclaim = kvmppc_gse_get_u64(gse);
+
+	return 0;
+}
+
+static struct kvmppc_gs_msg_ops gs_msg_test_hostwide_ops = {
+	.get_size = test_hostwide_get_size,
+	.fill_info = test_hostwide_fill_info,
+	.refresh_info = test_hostwide_refresh_info,
+};
+
+static void test_gs_hostwide_msg(struct kunit *test)
+{
+	struct kvmppc_gs_msg_test_hostwide_data test_data = {
+		.guest_heap = 0xdeadbeef,
+		.guest_heap_max = ~0ULL,
+		.guest_pgtable_size = 0xff,
+		.guest_pgtable_size_max = 0xffffff,
+		.guest_pgtable_reclaim = 0xdeadbeef,
+	};
+	struct kvmppc_gs_msg *gsm;
+	struct kvmppc_gs_buff *gsb;
+
+	gsm = kvmppc_gsm_new(&gs_msg_test_hostwide_ops, &test_data, GSM_SEND,
+			     GFP_KERNEL);
+	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, gsm);
+
+	gsb = kvmppc_gsb_new(kvmppc_gsm_size(gsm), 0, 0, GFP_KERNEL);
+	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, gsb);
+
+	kvmppc_gsm_include(gsm, KVMPPC_GSID_L0_GUEST_HEAP);
+	kvmppc_gsm_include(gsm, KVMPPC_GSID_L0_GUEST_HEAP_MAX);
+	kvmppc_gsm_include(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE);
+	kvmppc_gsm_include(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX);
+	kvmppc_gsm_include(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM);
+
+	kvmppc_gsm_fill_info(gsm, gsb);
+
+	memset(&test_data, 0, sizeof(test_data));
+
+	kvmppc_gsm_refresh_info(gsm, gsb);
+	KUNIT_EXPECT_EQ(test, test_data.guest_heap, 0xdeadbeef);
+	KUNIT_EXPECT_EQ(test, test_data.guest_heap_max, ~0ULL);
+	KUNIT_EXPECT_EQ(test, test_data.guest_pgtable_size, 0xff);
+	KUNIT_EXPECT_EQ(test, test_data.guest_pgtable_size_max, 0xffffff);
+	KUNIT_EXPECT_EQ(test, test_data.guest_pgtable_reclaim, 0xdeadbeef);
+
+	kvmppc_gsm_free(gsm);
+}
+
+/* Test if the H_GUEST_GET_STATE for hostwide counters works */
+static void test_gs_hostwide_counters(struct kunit *test)
+{
+	struct kvmppc_gs_msg_test_hostwide_data test_data;
+	struct kvmppc_gs_parser gsp = { 0 };
+
+	struct kvmppc_gs_msg *gsm;
+	struct kvmppc_gs_buff *gsb;
+	struct kvmppc_gs_elem *gse;
+	int rc;
+
+	if (!kvmhv_on_pseries())
+		kunit_skip(test, "This test need a kmv-hv guest");
+
+	gsm = kvmppc_gsm_new(&gs_msg_test_hostwide_ops, &test_data, GSM_SEND,
+			     GFP_KERNEL);
+	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, gsm);
+
+	gsb = kvmppc_gsb_new(kvmppc_gsm_size(gsm), 0, 0, GFP_KERNEL);
+	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, gsb);
+
+	kvmppc_gsm_include(gsm, KVMPPC_GSID_L0_GUEST_HEAP);
+
+	kvmppc_gsm_include(gsm, KVMPPC_GSID_L0_GUEST_HEAP_MAX);
+
+	kvmppc_gsm_include(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE);
+
+	kvmppc_gsm_include(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX);
+
+	kvmppc_gsm_include(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM);
+
+	kvmppc_gsm_fill_info(gsm, gsb);
+
+	/* With HOST_WIDE flags guestid and vcpuid will be ignored */
+	rc = kvmppc_gsb_recv(gsb, KVMPPC_GS_FLAGS_HOST_WIDE);
+	KUNIT_ASSERT_EQ(test, rc, 0);
+
+	/* Parse the guest state buffer is successful */
+	rc = kvmppc_gse_parse(&gsp, gsb);
+	KUNIT_ASSERT_EQ(test, rc, 0);
+
+	/* Parse the GSB and get the counters */
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_HEAP);
+	KUNIT_ASSERT_NOT_NULL_MSG(test, gse, "L0 Heap counter missing");
+	kunit_info(test, "Guest Heap Size=%llu bytes",
+		   kvmppc_gse_get_u64(gse));
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_HEAP_MAX);
+	KUNIT_ASSERT_NOT_NULL_MSG(test, gse, "L0 Heap counter max missing");
+	kunit_info(test, "Guest Heap Size Max=%llu bytes",
+		   kvmppc_gse_get_u64(gse));
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE);
+	KUNIT_ASSERT_NOT_NULL_MSG(test, gse, "L0 page-table size missing");
+	kunit_info(test, "Guest Page-table Size=%llu bytes",
+		   kvmppc_gse_get_u64(gse));
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX);
+	KUNIT_ASSERT_NOT_NULL_MSG(test, gse, "L0 page-table size-max missing");
+	kunit_info(test, "Guest Page-table Size Max=%llu bytes",
+		   kvmppc_gse_get_u64(gse));
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM);
+	KUNIT_ASSERT_NOT_NULL_MSG(test, gse, "L0 page-table reclaim size missing");
+	kunit_info(test, "Guest Page-table Reclaim Size=%llu bytes",
+		   kvmppc_gse_get_u64(gse));
+
+	kvmppc_gsm_free(gsm);
+	kvmppc_gsb_free(gsb);
+}
+
 static struct kunit_case guest_state_buffer_testcases[] = {
 	KUNIT_CASE(test_creating_buffer),
 	KUNIT_CASE(test_adding_element),
 	KUNIT_CASE(test_gs_bitmap),
 	KUNIT_CASE(test_gs_parsing),
 	KUNIT_CASE(test_gs_msg),
+	KUNIT_CASE(test_gs_hostwide_msg),
+	KUNIT_CASE(test_gs_hostwide_counters),
 	{}
 };
 
-- 
2.49.0

[RESEND PATCH v5 4/6] kvm powerpc/book3s-apiv2: Introduce kvm-hv specific PMU

[Vaibhav Jain]

Introduce a new PMU named 'kvm-hv' inside a new module named 'kvm-hv-pmu'
to report Book3s kvm-hv specific performance counters. This will expose
KVM-HV specific performance attributes to user-space via kernel's PMU
infrastructure and would enableusers to monitor active kvm-hv based guests.

The patch creates necessary scaffolding to for the new PMU callbacks and
introduces the new kernel module name 'kvm-hv-pmu' which is built with
CONFIG_KVM_BOOK3S_HV_PMU. The patch doesn't introduce any perf-events yet,
which will be introduced in later patches

Signed-off-by: Vaibhav Jain <vaibhav@linux.ibm.com>

---
Changelog

v5->resend:
* Rebase the patch to latest upstream kernel tree

v4->v5:
* Fixed a kismet warning reported by build-robot by removing dependency to
CONFIG_HV_PERF_CTRS
https://lore.kernel.org/oe-kbuild-all/202502280218.Jdd4jjlZ-lkp@intel.com/

* Removed a redundant check in kvmppc_register_pmu() [ Athira ]

v3->v4:
* Introduced a new kernel module named 'kmv-hv-pmu' to host the new PMU
instead of building the as part of KVM-HV module. [ Maddy ]
* Moved the code from arch/powerpc/kvm to arch/powerpc/perf [ Athira ]
* Added a new config named KVM_BOOK3S_HV_PMU to arch/powerpc/kvm/Kconfig

v2->v3:
* Fixed a build warning reported by kernel build robot.
Link:
https://lore.kernel.org/oe-kbuild-all/202501171030.3x0gqW8G-lkp@intel.com

v1->v2:
* Fixed an issue of kvm-hv not loading on baremetal kvm [Gautam]
---
 arch/powerpc/kvm/Kconfig       |  13 ++++
 arch/powerpc/perf/Makefile     |   2 +
 arch/powerpc/perf/kvm-hv-pmu.c | 138 +++++++++++++++++++++++++++++++++
 3 files changed, 153 insertions(+)
 create mode 100644 arch/powerpc/perf/kvm-hv-pmu.c

diff --git a/arch/powerpc/kvm/Kconfig b/arch/powerpc/kvm/Kconfig
index dbfdc126bf14..2f2702c867f7 100644
--- a/arch/powerpc/kvm/Kconfig
+++ b/arch/powerpc/kvm/Kconfig
@@ -83,6 +83,7 @@ config KVM_BOOK3S_64_HV
 	depends on KVM_BOOK3S_64 && PPC_POWERNV
 	select KVM_BOOK3S_HV_POSSIBLE
 	select KVM_GENERIC_MMU_NOTIFIER
+	select KVM_BOOK3S_HV_PMU
 	select CMA
 	help
 	  Support running unmodified book3s_64 guest kernels in
@@ -171,6 +172,18 @@ config KVM_BOOK3S_HV_NESTED_PMU_WORKAROUND
 	  those buggy L1s which saves the L2 state, at the cost of performance
 	  in all nested-capable guest entry/exit.
 
+config KVM_BOOK3S_HV_PMU
+	tristate "Hypervisor Perf events for KVM Book3s-HV"
+	depends on KVM_BOOK3S_64_HV
+	help
+	  Enable Book3s-HV Hypervisor Perf events PMU named 'kvm-hv'. These
+	  Perf events give an overview of hypervisor performance overall
+	  instead of a specific guests. Currently the PMU reports
+	  L0-Hypervisor stats on a kvm-hv enabled PSeries LPAR like:
+	  * Total/Used Guest-Heap
+	  * Total/Used Guest Page-table Memory
+	  * Total amount of Guest Page-table Memory reclaimed
+
 config KVM_BOOKE_HV
 	bool
 
diff --git a/arch/powerpc/perf/Makefile b/arch/powerpc/perf/Makefile
index ac2cf58d62db..7f53fcb7495a 100644
--- a/arch/powerpc/perf/Makefile
+++ b/arch/powerpc/perf/Makefile
@@ -18,6 +18,8 @@ obj-$(CONFIG_HV_PERF_CTRS) += hv-24x7.o hv-gpci.o hv-common.o
 
 obj-$(CONFIG_VPA_PMU) += vpa-pmu.o
 
+obj-$(CONFIG_KVM_BOOK3S_HV_PMU) += kvm-hv-pmu.o
+
 obj-$(CONFIG_PPC_8xx) += 8xx-pmu.o
 
 obj-$(CONFIG_PPC64)		+= $(obj64-y)
diff --git a/arch/powerpc/perf/kvm-hv-pmu.c b/arch/powerpc/perf/kvm-hv-pmu.c
new file mode 100644
index 000000000000..12f40a7b3ced
--- /dev/null
+++ b/arch/powerpc/perf/kvm-hv-pmu.c
@@ -0,0 +1,138 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Description: PMUs specific to running nested KVM-HV guests
+ * on Book3S processors (specifically POWER9 and later).
+ */
+
+#define pr_fmt(fmt)  "kvmppc-pmu: " fmt
+
+#include "asm-generic/local64.h"
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/ratelimit.h>
+#include <linux/kvm_host.h>
+#include <linux/gfp_types.h>
+#include <linux/pgtable.h>
+#include <linux/perf_event.h>
+#include <linux/spinlock_types.h>
+#include <linux/spinlock.h>
+
+#include <asm/types.h>
+#include <asm/kvm_ppc.h>
+#include <asm/kvm_book3s.h>
+#include <asm/mmu.h>
+#include <asm/pgalloc.h>
+#include <asm/pte-walk.h>
+#include <asm/reg.h>
+#include <asm/plpar_wrappers.h>
+#include <asm/firmware.h>
+
+enum kvmppc_pmu_eventid {
+	KVMPPC_EVENT_MAX,
+};
+
+static struct attribute *kvmppc_pmu_events_attr[] = {
+	NULL,
+};
+
+static const struct attribute_group kvmppc_pmu_events_group = {
+	.name = "events",
+	.attrs = kvmppc_pmu_events_attr,
+};
+
+PMU_FORMAT_ATTR(event, "config:0");
+static struct attribute *kvmppc_pmu_format_attr[] = {
+	&format_attr_event.attr,
+	NULL,
+};
+
+static struct attribute_group kvmppc_pmu_format_group = {
+	.name = "format",
+	.attrs = kvmppc_pmu_format_attr,
+};
+
+static const struct attribute_group *kvmppc_pmu_attr_groups[] = {
+	&kvmppc_pmu_events_group,
+	&kvmppc_pmu_format_group,
+	NULL,
+};
+
+static int kvmppc_pmu_event_init(struct perf_event *event)
+{
+	unsigned int config = event->attr.config;
+
+	pr_debug("%s: Event(%p) id=%llu cpu=%x on_cpu=%x config=%u",
+		 __func__, event, event->id, event->cpu,
+		 event->oncpu, config);
+
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	if (config >= KVMPPC_EVENT_MAX)
+		return -EINVAL;
+
+	local64_set(&event->hw.prev_count, 0);
+	local64_set(&event->count, 0);
+
+	return 0;
+}
+
+static void kvmppc_pmu_del(struct perf_event *event, int flags)
+{
+}
+
+static int kvmppc_pmu_add(struct perf_event *event, int flags)
+{
+	return 0;
+}
+
+static void kvmppc_pmu_read(struct perf_event *event)
+{
+}
+
+/* L1 wide counters PMU */
+static struct pmu kvmppc_pmu = {
+	.module = THIS_MODULE,
+	.task_ctx_nr = perf_sw_context,
+	.name = "kvm-hv",
+	.event_init = kvmppc_pmu_event_init,
+	.add = kvmppc_pmu_add,
+	.del = kvmppc_pmu_del,
+	.read = kvmppc_pmu_read,
+	.attr_groups = kvmppc_pmu_attr_groups,
+	.type = -1,
+};
+
+static int __init kvmppc_register_pmu(void)
+{
+	int rc = -EOPNOTSUPP;
+
+	/* only support events for nestedv2 right now */
+	if (kvmhv_is_nestedv2()) {
+		/* Register the pmu */
+		rc = perf_pmu_register(&kvmppc_pmu, kvmppc_pmu.name, -1);
+		if (rc)
+			goto out;
+
+		pr_info("Registered kvm-hv pmu");
+	}
+
+out:
+	return rc;
+}
+
+static void __exit kvmppc_unregister_pmu(void)
+{
+	if (kvmhv_is_nestedv2()) {
+		if (kvmppc_pmu.type != -1)
+			perf_pmu_unregister(&kvmppc_pmu);
+
+		pr_info("kvmhv_pmu unregistered.\n");
+	}
+}
+
+module_init(kvmppc_register_pmu);
+module_exit(kvmppc_unregister_pmu);
+MODULE_DESCRIPTION("KVM PPC Book3s-hv PMU");
+MODULE_AUTHOR("Vaibhav Jain <vaibhav@linux.ibm.com>");
+MODULE_LICENSE("GPL");
-- 
2.49.0

[RESEND PATCH v5 5/6] powerpc/kvm-hv-pmu: Implement GSB message-ops for hostwide counters

[Vaibhav Jain]

Implement and setup necessary structures to send a prepolulated
Guest-State-Buffer(GSB) requesting hostwide counters to L0-PowerVM and have
the returned GSB holding the values of these counters parsed. This is done
via existing GSB implementation and with the newly added support of
Hostwide elements in GSB.

The request to L0-PowerVM to return Hostwide counters is done using a
pre-allocated GSB named 'gsb_l0_stats'. To be able to populate this GSB
with the needed Guest-State-Elements (GSIDs) a instance of 'struct
kvmppc_gs_msg' named 'gsm_l0_stats' is introduced. The 'gsm_l0_stats' is
tied to an instance of 'struct kvmppc_gs_msg_ops' named  'gsb_ops_l0_stats'
which holds various callbacks to be compute the size ( hostwide_get_size()
), populate the GSB ( hostwide_fill_info() ) and
refresh ( hostwide_refresh_info() ) the contents of
'l0_stats' that holds the Hostwide counters returned from L0-PowerVM.

To protect these structures from simultaneous access a spinlock
'lock_l0_stats' has been introduced. The allocation and initialization of
the above structures is done in newly introduced kvmppc_init_hostwide() and
similarly the cleanup is performed in newly introduced
kvmppc_cleanup_hostwide().

Signed-off-by: Vaibhav Jain <vaibhav@linux.ibm.com>

---
Changelog

v5->resend:
* Rebase the patch to latest upstream kernel tree

v4->v5:
* Update kvmppc_register_pmu() to refactor the module init path.

v3->v4:
* Minor tweaks to code and patch description since this code is now being
built as a kernel module.
* Introduced kvmppc_events_sysfs_show() as power_events_sysfs_show() is not
exported to modules.

v2->v3:
Removed a redundant branch in kvmppc_init_hostwide() [Gautam]

v1->v2:
* Added error handling to hostwide_fill_info() [Gautam]
---
 arch/powerpc/perf/kvm-hv-pmu.c | 207 +++++++++++++++++++++++++++++++++
 1 file changed, 207 insertions(+)

diff --git a/arch/powerpc/perf/kvm-hv-pmu.c b/arch/powerpc/perf/kvm-hv-pmu.c
index 12f40a7b3ced..705be24ccb43 100644
--- a/arch/powerpc/perf/kvm-hv-pmu.c
+++ b/arch/powerpc/perf/kvm-hv-pmu.c
@@ -27,10 +27,41 @@
 #include <asm/plpar_wrappers.h>
 #include <asm/firmware.h>
 
+#include "asm/guest-state-buffer.h"
+
 enum kvmppc_pmu_eventid {
 	KVMPPC_EVENT_MAX,
 };
 
+#define KVMPPC_PMU_EVENT_ATTR(_name, _id) \
+	PMU_EVENT_ATTR_ID(_name, kvmppc_events_sysfs_show, _id)
+
+static ssize_t kvmppc_events_sysfs_show(struct device *dev,
+					struct device_attribute *attr,
+					char *page)
+{
+	struct perf_pmu_events_attr *pmu_attr;
+
+	pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
+	return sprintf(page, "event=0x%02llx\n", pmu_attr->id);
+}
+
+/* Holds the hostwide stats */
+static struct kvmppc_hostwide_stats {
+	u64 guest_heap;
+	u64 guest_heap_max;
+	u64 guest_pgtable_size;
+	u64 guest_pgtable_size_max;
+	u64 guest_pgtable_reclaim;
+} l0_stats;
+
+/* Protect access to l0_stats */
+static DEFINE_SPINLOCK(lock_l0_stats);
+
+/* GSB related structs needed to talk to L0 */
+static struct kvmppc_gs_msg *gsm_l0_stats;
+static struct kvmppc_gs_buff *gsb_l0_stats;
+
 static struct attribute *kvmppc_pmu_events_attr[] = {
 	NULL,
 };
@@ -90,6 +121,176 @@ static void kvmppc_pmu_read(struct perf_event *event)
 {
 }
 
+/* Return the size of the needed guest state buffer */
+static size_t hostwide_get_size(struct kvmppc_gs_msg *gsm)
+
+{
+	size_t size = 0;
+	const u16 ids[] = {
+		KVMPPC_GSID_L0_GUEST_HEAP,
+		KVMPPC_GSID_L0_GUEST_HEAP_MAX,
+		KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE,
+		KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX,
+		KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM
+	};
+
+	for (int i = 0; i < ARRAY_SIZE(ids); i++)
+		size += kvmppc_gse_total_size(kvmppc_gsid_size(ids[i]));
+	return size;
+}
+
+/* Populate the request guest state buffer */
+static int hostwide_fill_info(struct kvmppc_gs_buff *gsb,
+			      struct kvmppc_gs_msg *gsm)
+{
+	int rc = 0;
+	struct kvmppc_hostwide_stats  *stats = gsm->data;
+
+	/*
+	 * It doesn't matter what values are put into request buffer as
+	 * they are going to be overwritten anyways. But for the sake of
+	 * testcode and symmetry contents of existing stats are put
+	 * populated into the request guest state buffer.
+	 */
+	if (kvmppc_gsm_includes(gsm, KVMPPC_GSID_L0_GUEST_HEAP))
+		rc = kvmppc_gse_put_u64(gsb,
+					KVMPPC_GSID_L0_GUEST_HEAP,
+					stats->guest_heap);
+
+	if (!rc && kvmppc_gsm_includes(gsm, KVMPPC_GSID_L0_GUEST_HEAP_MAX))
+		rc = kvmppc_gse_put_u64(gsb,
+					KVMPPC_GSID_L0_GUEST_HEAP_MAX,
+					stats->guest_heap_max);
+
+	if (!rc && kvmppc_gsm_includes(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE))
+		rc = kvmppc_gse_put_u64(gsb,
+					KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE,
+					stats->guest_pgtable_size);
+	if (!rc &&
+	    kvmppc_gsm_includes(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX))
+		rc = kvmppc_gse_put_u64(gsb,
+					KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX,
+					stats->guest_pgtable_size_max);
+	if (!rc &&
+	    kvmppc_gsm_includes(gsm, KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM))
+		rc = kvmppc_gse_put_u64(gsb,
+					KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM,
+					stats->guest_pgtable_reclaim);
+
+	return rc;
+}
+
+/* Parse and update the host wide stats from returned gsb */
+static int hostwide_refresh_info(struct kvmppc_gs_msg *gsm,
+				 struct kvmppc_gs_buff *gsb)
+{
+	struct kvmppc_gs_parser gsp = { 0 };
+	struct kvmppc_hostwide_stats *stats = gsm->data;
+	struct kvmppc_gs_elem *gse;
+	int rc;
+
+	rc = kvmppc_gse_parse(&gsp, gsb);
+	if (rc < 0)
+		return rc;
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_HEAP);
+	if (gse)
+		stats->guest_heap = kvmppc_gse_get_u64(gse);
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_HEAP_MAX);
+	if (gse)
+		stats->guest_heap_max = kvmppc_gse_get_u64(gse);
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE);
+	if (gse)
+		stats->guest_pgtable_size = kvmppc_gse_get_u64(gse);
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX);
+	if (gse)
+		stats->guest_pgtable_size_max = kvmppc_gse_get_u64(gse);
+
+	gse = kvmppc_gsp_lookup(&gsp, KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM);
+	if (gse)
+		stats->guest_pgtable_reclaim = kvmppc_gse_get_u64(gse);
+
+	return 0;
+}
+
+/* gsb-message ops for setting up/parsing */
+static struct kvmppc_gs_msg_ops gsb_ops_l0_stats = {
+	.get_size = hostwide_get_size,
+	.fill_info = hostwide_fill_info,
+	.refresh_info = hostwide_refresh_info,
+};
+
+static int kvmppc_init_hostwide(void)
+{
+	int rc = 0;
+	unsigned long flags;
+
+	spin_lock_irqsave(&lock_l0_stats, flags);
+
+	/* already registered ? */
+	if (gsm_l0_stats) {
+		rc = 0;
+		goto out;
+	}
+
+	/* setup the Guest state message/buffer to talk to L0 */
+	gsm_l0_stats = kvmppc_gsm_new(&gsb_ops_l0_stats, &l0_stats,
+				      GSM_SEND, GFP_KERNEL);
+	if (!gsm_l0_stats) {
+		rc = -ENOMEM;
+		goto out;
+	}
+
+	/* Populate the Idents */
+	kvmppc_gsm_include(gsm_l0_stats, KVMPPC_GSID_L0_GUEST_HEAP);
+	kvmppc_gsm_include(gsm_l0_stats, KVMPPC_GSID_L0_GUEST_HEAP_MAX);
+	kvmppc_gsm_include(gsm_l0_stats, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE);
+	kvmppc_gsm_include(gsm_l0_stats, KVMPPC_GSID_L0_GUEST_PGTABLE_SIZE_MAX);
+	kvmppc_gsm_include(gsm_l0_stats, KVMPPC_GSID_L0_GUEST_PGTABLE_RECLAIM);
+
+	/* allocate GSB. Guest/Vcpu Id is ignored */
+	gsb_l0_stats = kvmppc_gsb_new(kvmppc_gsm_size(gsm_l0_stats), 0, 0,
+				      GFP_KERNEL);
+	if (!gsb_l0_stats) {
+		rc = -ENOMEM;
+		goto out;
+	}
+
+	/* ask the ops to fill in the info */
+	rc = kvmppc_gsm_fill_info(gsm_l0_stats, gsb_l0_stats);
+
+out:
+	if (rc) {
+		if (gsm_l0_stats)
+			kvmppc_gsm_free(gsm_l0_stats);
+		if (gsb_l0_stats)
+			kvmppc_gsb_free(gsb_l0_stats);
+		gsm_l0_stats = NULL;
+		gsb_l0_stats = NULL;
+	}
+	spin_unlock_irqrestore(&lock_l0_stats, flags);
+	return rc;
+}
+
+static void kvmppc_cleanup_hostwide(void)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&lock_l0_stats, flags);
+
+	if (gsm_l0_stats)
+		kvmppc_gsm_free(gsm_l0_stats);
+	if (gsb_l0_stats)
+		kvmppc_gsb_free(gsb_l0_stats);
+	gsm_l0_stats = NULL;
+	gsb_l0_stats = NULL;
+
+	spin_unlock_irqrestore(&lock_l0_stats, flags);
+}
+
 /* L1 wide counters PMU */
 static struct pmu kvmppc_pmu = {
 	.module = THIS_MODULE,
@@ -109,6 +310,10 @@ static int __init kvmppc_register_pmu(void)
 
 	/* only support events for nestedv2 right now */
 	if (kvmhv_is_nestedv2()) {
+		rc = kvmppc_init_hostwide();
+		if (rc)
+			goto out;
+
 		/* Register the pmu */
 		rc = perf_pmu_register(&kvmppc_pmu, kvmppc_pmu.name, -1);
 		if (rc)
@@ -124,6 +329,8 @@ static int __init kvmppc_register_pmu(void)
 static void __exit kvmppc_unregister_pmu(void)
 {
 	if (kvmhv_is_nestedv2()) {
+		kvmppc_cleanup_hostwide();
+
 		if (kvmppc_pmu.type != -1)
 			perf_pmu_unregister(&kvmppc_pmu);
 
-- 
2.49.0

[RESEND PATCH v5 6/6] powerpc/kvm-hv-pmu: Add perf-events for Hostwide counters

[Vaibhav Jain]

Update 'kvm-hv-pmu.c' to add five new perf-events mapped to the five
Hostwide counters. Since these newly introduced perf events are at system
wide scope and can be read from any L1-Lpar CPU, 'kvmppc_pmu' scope and
capabilities are updated appropriately.

Also introduce two new helpers. First is kvmppc_update_l0_stats() that uses
the infrastructure introduced in previous patches to issues the
H_GUEST_GET_STATE hcall L0-PowerVM to fetch guest-state-buffer holding the
latest values of these counters which is then parsed and 'l0_stats'
variable updated.

Second helper is kvmppc_pmu_event_update() which is called from
'kvmppv_pmu' callbacks and uses kvmppc_update_l0_stats() to update
'l0_stats' and the update the 'struct perf_event's event-counter.

Some minor updates to kvmppc_pmu_{add, del, read}() to remove some debug
scaffolding code.

Signed-off-by: Vaibhav Jain <vaibhav@linux.ibm.com>
Reviewed-by: Athira Rajeev <atrajeev@linux.ibm.com>
---
Changelog

v5->resend:
* Rebase the patch to latest upstream kernel tree
* Added Athira's reviewed-by.

v4->v5:
* Call kvmppc_pmu_event_update() during pmu's 'del()' callback [ Athira ]

v3->v4:
* Minor tweaks to patch description and code as its now being built as a
separate kernel module.

v2->v3:
None

v1->v2:
None
---
 arch/powerpc/perf/kvm-hv-pmu.c | 92 +++++++++++++++++++++++++++++++++-
 1 file changed, 91 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/perf/kvm-hv-pmu.c b/arch/powerpc/perf/kvm-hv-pmu.c
index 705be24ccb43..ae264c9080ef 100644
--- a/arch/powerpc/perf/kvm-hv-pmu.c
+++ b/arch/powerpc/perf/kvm-hv-pmu.c
@@ -30,6 +30,11 @@
 #include "asm/guest-state-buffer.h"
 
 enum kvmppc_pmu_eventid {
+	KVMPPC_EVENT_HOST_HEAP,
+	KVMPPC_EVENT_HOST_HEAP_MAX,
+	KVMPPC_EVENT_HOST_PGTABLE,
+	KVMPPC_EVENT_HOST_PGTABLE_MAX,
+	KVMPPC_EVENT_HOST_PGTABLE_RECLAIM,
 	KVMPPC_EVENT_MAX,
 };
 
@@ -61,8 +66,14 @@ static DEFINE_SPINLOCK(lock_l0_stats);
 /* GSB related structs needed to talk to L0 */
 static struct kvmppc_gs_msg *gsm_l0_stats;
 static struct kvmppc_gs_buff *gsb_l0_stats;
+static struct kvmppc_gs_parser gsp_l0_stats;
 
 static struct attribute *kvmppc_pmu_events_attr[] = {
+	KVMPPC_PMU_EVENT_ATTR(host_heap, KVMPPC_EVENT_HOST_HEAP),
+	KVMPPC_PMU_EVENT_ATTR(host_heap_max, KVMPPC_EVENT_HOST_HEAP_MAX),
+	KVMPPC_PMU_EVENT_ATTR(host_pagetable, KVMPPC_EVENT_HOST_PGTABLE),
+	KVMPPC_PMU_EVENT_ATTR(host_pagetable_max, KVMPPC_EVENT_HOST_PGTABLE_MAX),
+	KVMPPC_PMU_EVENT_ATTR(host_pagetable_reclaim, KVMPPC_EVENT_HOST_PGTABLE_RECLAIM),
 	NULL,
 };
 
@@ -71,7 +82,7 @@ static const struct attribute_group kvmppc_pmu_events_group = {
 	.attrs = kvmppc_pmu_events_attr,
 };
 
-PMU_FORMAT_ATTR(event, "config:0");
+PMU_FORMAT_ATTR(event, "config:0-5");
 static struct attribute *kvmppc_pmu_format_attr[] = {
 	&format_attr_event.attr,
 	NULL,
@@ -88,6 +99,79 @@ static const struct attribute_group *kvmppc_pmu_attr_groups[] = {
 	NULL,
 };
 
+/*
+ * Issue the hcall to get the L0-host stats.
+ * Should be called with l0-stat lock held
+ */
+static int kvmppc_update_l0_stats(void)
+{
+	int rc;
+
+	/* With HOST_WIDE flags guestid and vcpuid will be ignored */
+	rc = kvmppc_gsb_recv(gsb_l0_stats, KVMPPC_GS_FLAGS_HOST_WIDE);
+	if (rc)
+		goto out;
+
+	/* Parse the guest state buffer is successful */
+	rc = kvmppc_gse_parse(&gsp_l0_stats, gsb_l0_stats);
+	if (rc)
+		goto out;
+
+	/* Update the l0 returned stats*/
+	memset(&l0_stats, 0, sizeof(l0_stats));
+	rc = kvmppc_gsm_refresh_info(gsm_l0_stats, gsb_l0_stats);
+
+out:
+	return rc;
+}
+
+/* Update the value of the given perf_event */
+static int kvmppc_pmu_event_update(struct perf_event *event)
+{
+	int rc;
+	u64 curr_val, prev_val;
+	unsigned long flags;
+	unsigned int config = event->attr.config;
+
+	/* Ensure no one else is modifying the l0_stats */
+	spin_lock_irqsave(&lock_l0_stats, flags);
+
+	rc = kvmppc_update_l0_stats();
+	if (!rc) {
+		switch (config) {
+		case KVMPPC_EVENT_HOST_HEAP:
+			curr_val = l0_stats.guest_heap;
+			break;
+		case KVMPPC_EVENT_HOST_HEAP_MAX:
+			curr_val = l0_stats.guest_heap_max;
+			break;
+		case KVMPPC_EVENT_HOST_PGTABLE:
+			curr_val = l0_stats.guest_pgtable_size;
+			break;
+		case KVMPPC_EVENT_HOST_PGTABLE_MAX:
+			curr_val = l0_stats.guest_pgtable_size_max;
+			break;
+		case KVMPPC_EVENT_HOST_PGTABLE_RECLAIM:
+			curr_val = l0_stats.guest_pgtable_reclaim;
+			break;
+		default:
+			rc = -ENOENT;
+			break;
+		}
+	}
+
+	spin_unlock_irqrestore(&lock_l0_stats, flags);
+
+	/* If no error than update the perf event */
+	if (!rc) {
+		prev_val = local64_xchg(&event->hw.prev_count, curr_val);
+		if (curr_val > prev_val)
+			local64_add(curr_val - prev_val, &event->count);
+	}
+
+	return rc;
+}
+
 static int kvmppc_pmu_event_init(struct perf_event *event)
 {
 	unsigned int config = event->attr.config;
@@ -110,15 +194,19 @@ static int kvmppc_pmu_event_init(struct perf_event *event)
 
 static void kvmppc_pmu_del(struct perf_event *event, int flags)
 {
+	kvmppc_pmu_event_update(event);
 }
 
 static int kvmppc_pmu_add(struct perf_event *event, int flags)
 {
+	if (flags & PERF_EF_START)
+		return kvmppc_pmu_event_update(event);
 	return 0;
 }
 
 static void kvmppc_pmu_read(struct perf_event *event)
 {
+	kvmppc_pmu_event_update(event);
 }
 
 /* Return the size of the needed guest state buffer */
@@ -302,6 +390,8 @@ static struct pmu kvmppc_pmu = {
 	.read = kvmppc_pmu_read,
 	.attr_groups = kvmppc_pmu_attr_groups,
 	.type = -1,
+	.scope = PERF_PMU_SCOPE_SYS_WIDE,
+	.capabilities = PERF_PMU_CAP_NO_EXCLUDE | PERF_PMU_CAP_NO_INTERRUPT,
 };
 
 static int __init kvmppc_register_pmu(void)
-- 
2.49.0