* [RFC PATCH 0/3] CXL Performance Monitoring Unit QEMU emulation.
@ 2022-08-31 15:33 ` Jonathan Cameron via
0 siblings, 0 replies; 8+ messages in thread
From: Jonathan Cameron @ 2022-08-31 15:33 UTC (permalink / raw)
To: qemu-devel, linux-cxl; +Cc: Ben Widawsky, Dan Williams
CXL 3.0 introduces a CXL Performance Monitoring Unit.
(13.2 Performance Monitoring)
These unit may be present in any CXL component but for now the kernel
code only supports them in Type 3 Memory devices.
Add basic emulation of such units and instantiate 2 in each CXL Type 3 device.
Kernel driver was developed against this fairly minimal emulation.
https://lore.kernel.org/linux-cxl/20220824103617.21781-1-Jonathan.Cameron@huawei.com/T/#t
For now, this RFC is both to seek general feedback on the approach used and to
provide a platform for testing the kernel driver against.
Note that the CPMU definition is extremely flexible and configuring the
available options via the QEMU command line may be impractical.
To avoid that complexity, this configuration is currently fixed.
This code provides a pair of identical CPMU instances on each CXL type 3
device which were constructed to hit some of the corner cases in the
enumeration code.
The counters themselves just get bigger in a fashion designed to overflow
fairly quickly. Actual values are meaningless. For now there is no
control over the existence of CPMU instances. Likely we will add at least
an option to enable them or not.
Precursor patch to add MSI/MSIX support may well merge as part of a
different series, but is needed to apply this to 7.1.
Jonathan Cameron (3):
hw/mem/cxl-type3: Add MSI/MSIX support
hw/cxl: Switch to using an array for CXLRegisterLocator base
addresses.
hw/cxl: CXL Performance Monitoring Unit (CPMU) Emulation
hw/cxl/cxl-cpmu.c | 295 +++++++++++++++++++++++++++++++++
hw/cxl/meson.build | 1 +
hw/mem/cxl_type3.c | 31 +++-
hw/pci-bridge/cxl_downstream.c | 4 +-
hw/pci-bridge/cxl_root_port.c | 4 +-
hw/pci-bridge/cxl_upstream.c | 4 +-
include/hw/cxl/cxl.h | 1 +
include/hw/cxl/cxl_cpmu.h | 99 +++++++++++
include/hw/cxl/cxl_device.h | 31 +++-
include/hw/cxl/cxl_pci.h | 15 +-
10 files changed, 463 insertions(+), 22 deletions(-)
create mode 100644 hw/cxl/cxl-cpmu.c
create mode 100644 include/hw/cxl/cxl_cpmu.h
--
2.32.0
^ permalink raw reply [flat|nested] 8+ messages in thread
* [RFC PATCH 0/3] CXL Performance Monitoring Unit QEMU emulation.
@ 2022-08-31 15:33 ` Jonathan Cameron via
0 siblings, 0 replies; 8+ messages in thread
From: Jonathan Cameron via @ 2022-08-31 15:33 UTC (permalink / raw)
To: qemu-devel, linux-cxl; +Cc: Ben Widawsky, Dan Williams
CXL 3.0 introduces a CXL Performance Monitoring Unit.
(13.2 Performance Monitoring)
These unit may be present in any CXL component but for now the kernel
code only supports them in Type 3 Memory devices.
Add basic emulation of such units and instantiate 2 in each CXL Type 3 device.
Kernel driver was developed against this fairly minimal emulation.
https://lore.kernel.org/linux-cxl/20220824103617.21781-1-Jonathan.Cameron@huawei.com/T/#t
For now, this RFC is both to seek general feedback on the approach used and to
provide a platform for testing the kernel driver against.
Note that the CPMU definition is extremely flexible and configuring the
available options via the QEMU command line may be impractical.
To avoid that complexity, this configuration is currently fixed.
This code provides a pair of identical CPMU instances on each CXL type 3
device which were constructed to hit some of the corner cases in the
enumeration code.
The counters themselves just get bigger in a fashion designed to overflow
fairly quickly. Actual values are meaningless. For now there is no
control over the existence of CPMU instances. Likely we will add at least
an option to enable them or not.
Precursor patch to add MSI/MSIX support may well merge as part of a
different series, but is needed to apply this to 7.1.
Jonathan Cameron (3):
hw/mem/cxl-type3: Add MSI/MSIX support
hw/cxl: Switch to using an array for CXLRegisterLocator base
addresses.
hw/cxl: CXL Performance Monitoring Unit (CPMU) Emulation
hw/cxl/cxl-cpmu.c | 295 +++++++++++++++++++++++++++++++++
hw/cxl/meson.build | 1 +
hw/mem/cxl_type3.c | 31 +++-
hw/pci-bridge/cxl_downstream.c | 4 +-
hw/pci-bridge/cxl_root_port.c | 4 +-
hw/pci-bridge/cxl_upstream.c | 4 +-
include/hw/cxl/cxl.h | 1 +
include/hw/cxl/cxl_cpmu.h | 99 +++++++++++
include/hw/cxl/cxl_device.h | 31 +++-
include/hw/cxl/cxl_pci.h | 15 +-
10 files changed, 463 insertions(+), 22 deletions(-)
create mode 100644 hw/cxl/cxl-cpmu.c
create mode 100644 include/hw/cxl/cxl_cpmu.h
--
2.32.0
^ permalink raw reply [flat|nested] 8+ messages in thread
* [RFC PATCH 1/3] hw/mem/cxl-type3: Add MSI/MSIX support
2022-08-31 15:33 ` Jonathan Cameron via
@ 2022-08-31 15:33 ` Jonathan Cameron via
-1 siblings, 0 replies; 8+ messages in thread
From: Jonathan Cameron @ 2022-08-31 15:33 UTC (permalink / raw)
To: qemu-devel, linux-cxl; +Cc: Ben Widawsky, Dan Williams
This will be used by several upcoming patch sets so break it out
such that it doesn't matter which one lands first.
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
---
hw/mem/cxl_type3.c | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/hw/mem/cxl_type3.c b/hw/mem/cxl_type3.c
index e0c1535b73..68d200144b 100644
--- a/hw/mem/cxl_type3.c
+++ b/hw/mem/cxl_type3.c
@@ -13,6 +13,8 @@
#include "qemu/rcu.h"
#include "sysemu/hostmem.h"
#include "hw/cxl/cxl.h"
+#include "hw/pci/msi.h"
+#include "hw/pci/msix.h"
/*
* Null value of all Fs suggested by IEEE RA guidelines for use of
@@ -146,6 +148,8 @@ static void ct3_realize(PCIDevice *pci_dev, Error **errp)
ComponentRegisters *regs = &cxl_cstate->crb;
MemoryRegion *mr = ®s->component_registers;
uint8_t *pci_conf = pci_dev->config;
+ unsigned short msix_num = 1;
+ int i;
if (!cxl_setup_memory(ct3d, errp)) {
return;
@@ -180,6 +184,12 @@ static void ct3_realize(PCIDevice *pci_dev, Error **errp)
PCI_BASE_ADDRESS_SPACE_MEMORY |
PCI_BASE_ADDRESS_MEM_TYPE_64,
&ct3d->cxl_dstate.device_registers);
+
+ /* MSI(-X) Initailization */
+ msix_init_exclusive_bar(pci_dev, msix_num, 4, NULL);
+ for (i = 0; i < msix_num; i++) {
+ msix_vector_use(pci_dev, i);
+ }
}
static void ct3_exit(PCIDevice *pci_dev)
--
2.32.0
^ permalink raw reply related [flat|nested] 8+ messages in thread
* [RFC PATCH 1/3] hw/mem/cxl-type3: Add MSI/MSIX support
@ 2022-08-31 15:33 ` Jonathan Cameron via
0 siblings, 0 replies; 8+ messages in thread
From: Jonathan Cameron via @ 2022-08-31 15:33 UTC (permalink / raw)
To: qemu-devel, linux-cxl; +Cc: Ben Widawsky, Dan Williams
This will be used by several upcoming patch sets so break it out
such that it doesn't matter which one lands first.
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
---
hw/mem/cxl_type3.c | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/hw/mem/cxl_type3.c b/hw/mem/cxl_type3.c
index e0c1535b73..68d200144b 100644
--- a/hw/mem/cxl_type3.c
+++ b/hw/mem/cxl_type3.c
@@ -13,6 +13,8 @@
#include "qemu/rcu.h"
#include "sysemu/hostmem.h"
#include "hw/cxl/cxl.h"
+#include "hw/pci/msi.h"
+#include "hw/pci/msix.h"
/*
* Null value of all Fs suggested by IEEE RA guidelines for use of
@@ -146,6 +148,8 @@ static void ct3_realize(PCIDevice *pci_dev, Error **errp)
ComponentRegisters *regs = &cxl_cstate->crb;
MemoryRegion *mr = ®s->component_registers;
uint8_t *pci_conf = pci_dev->config;
+ unsigned short msix_num = 1;
+ int i;
if (!cxl_setup_memory(ct3d, errp)) {
return;
@@ -180,6 +184,12 @@ static void ct3_realize(PCIDevice *pci_dev, Error **errp)
PCI_BASE_ADDRESS_SPACE_MEMORY |
PCI_BASE_ADDRESS_MEM_TYPE_64,
&ct3d->cxl_dstate.device_registers);
+
+ /* MSI(-X) Initailization */
+ msix_init_exclusive_bar(pci_dev, msix_num, 4, NULL);
+ for (i = 0; i < msix_num; i++) {
+ msix_vector_use(pci_dev, i);
+ }
}
static void ct3_exit(PCIDevice *pci_dev)
--
2.32.0
^ permalink raw reply related [flat|nested] 8+ messages in thread
* [RFC PATCH 2/3] hw/cxl: Switch to using an array for CXLRegisterLocator base addresses.
2022-08-31 15:33 ` Jonathan Cameron via
@ 2022-08-31 15:33 ` Jonathan Cameron via
-1 siblings, 0 replies; 8+ messages in thread
From: Jonathan Cameron @ 2022-08-31 15:33 UTC (permalink / raw)
To: qemu-devel, linux-cxl; +Cc: Ben Widawsky, Dan Williams
Allows for easier looping over entries when adding CPMU instances.
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
---
hw/mem/cxl_type3.c | 8 ++++----
hw/pci-bridge/cxl_downstream.c | 4 ++--
hw/pci-bridge/cxl_root_port.c | 4 ++--
hw/pci-bridge/cxl_upstream.c | 4 ++--
include/hw/cxl/cxl_pci.h | 10 ++++------
5 files changed, 14 insertions(+), 16 deletions(-)
diff --git a/hw/mem/cxl_type3.c b/hw/mem/cxl_type3.c
index 68d200144b..5d29d2595c 100644
--- a/hw/mem/cxl_type3.c
+++ b/hw/mem/cxl_type3.c
@@ -44,10 +44,10 @@ static void build_dvsecs(CXLType3Dev *ct3d)
dvsec = (uint8_t *)&(CXLDVSECRegisterLocator){
.rsvd = 0,
- .reg0_base_lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
- .reg0_base_hi = 0,
- .reg1_base_lo = RBI_CXL_DEVICE_REG | CXL_DEVICE_REG_BAR_IDX,
- .reg1_base_hi = 0,
+ .reg_base[0].lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
+ .reg_base[0].hi = 0,
+ .reg_base[1].lo = RBI_CXL_DEVICE_REG | CXL_DEVICE_REG_BAR_IDX,
+ .reg_base[1].hi = 0,
};
cxl_component_create_dvsec(cxl_cstate, CXL2_TYPE3_DEVICE,
REG_LOC_DVSEC_LENGTH, REG_LOC_DVSEC,
diff --git a/hw/pci-bridge/cxl_downstream.c b/hw/pci-bridge/cxl_downstream.c
index a361e519d0..7822ccd5de 100644
--- a/hw/pci-bridge/cxl_downstream.c
+++ b/hw/pci-bridge/cxl_downstream.c
@@ -126,8 +126,8 @@ static void build_dvsecs(CXLComponentState *cxl)
dvsec = (uint8_t *)&(CXLDVSECRegisterLocator){
.rsvd = 0,
- .reg0_base_lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
- .reg0_base_hi = 0,
+ .reg_base[0].lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
+ .reg_base[0].hi = 0,
};
cxl_component_create_dvsec(cxl, CXL2_DOWNSTREAM_PORT,
REG_LOC_DVSEC_LENGTH, REG_LOC_DVSEC,
diff --git a/hw/pci-bridge/cxl_root_port.c b/hw/pci-bridge/cxl_root_port.c
index fb213fa06e..08c2441dab 100644
--- a/hw/pci-bridge/cxl_root_port.c
+++ b/hw/pci-bridge/cxl_root_port.c
@@ -87,8 +87,8 @@ static void build_dvsecs(CXLComponentState *cxl)
dvsec = (uint8_t *)&(CXLDVSECRegisterLocator){
.rsvd = 0,
- .reg0_base_lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
- .reg0_base_hi = 0,
+ .reg_base[0].lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
+ .reg_base[0].hi = 0,
};
cxl_component_create_dvsec(cxl, CXL2_ROOT_PORT,
REG_LOC_DVSEC_LENGTH, REG_LOC_DVSEC,
diff --git a/hw/pci-bridge/cxl_upstream.c b/hw/pci-bridge/cxl_upstream.c
index a83a3e81e4..45ee6ba884 100644
--- a/hw/pci-bridge/cxl_upstream.c
+++ b/hw/pci-bridge/cxl_upstream.c
@@ -111,8 +111,8 @@ static void build_dvsecs(CXLComponentState *cxl)
dvsec = (uint8_t *)&(CXLDVSECRegisterLocator){
.rsvd = 0,
- .reg0_base_lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
- .reg0_base_hi = 0,
+ .reg_base[0].lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
+ .reg_base[0].hi = 0,
};
cxl_component_create_dvsec(cxl, CXL2_UPSTREAM_PORT,
REG_LOC_DVSEC_LENGTH, REG_LOC_DVSEC,
diff --git a/include/hw/cxl/cxl_pci.h b/include/hw/cxl/cxl_pci.h
index 01cf002096..8cbeb61142 100644
--- a/include/hw/cxl/cxl_pci.h
+++ b/include/hw/cxl/cxl_pci.h
@@ -141,12 +141,10 @@ QEMU_BUILD_BUG_ON(sizeof(CXLDVSECPortFlexBus) != 0x14);
typedef struct CXLDVSECRegisterLocator {
DVSECHeader hdr;
uint16_t rsvd;
- uint32_t reg0_base_lo;
- uint32_t reg0_base_hi;
- uint32_t reg1_base_lo;
- uint32_t reg1_base_hi;
- uint32_t reg2_base_lo;
- uint32_t reg2_base_hi;
+ struct {
+ uint32_t lo;
+ uint32_t hi;
+ } reg_base[3];
} CXLDVSECRegisterLocator;
QEMU_BUILD_BUG_ON(sizeof(CXLDVSECRegisterLocator) != 0x24);
--
2.32.0
^ permalink raw reply related [flat|nested] 8+ messages in thread
* [RFC PATCH 2/3] hw/cxl: Switch to using an array for CXLRegisterLocator base addresses.
@ 2022-08-31 15:33 ` Jonathan Cameron via
0 siblings, 0 replies; 8+ messages in thread
From: Jonathan Cameron via @ 2022-08-31 15:33 UTC (permalink / raw)
To: qemu-devel, linux-cxl; +Cc: Ben Widawsky, Dan Williams
Allows for easier looping over entries when adding CPMU instances.
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
---
hw/mem/cxl_type3.c | 8 ++++----
hw/pci-bridge/cxl_downstream.c | 4 ++--
hw/pci-bridge/cxl_root_port.c | 4 ++--
hw/pci-bridge/cxl_upstream.c | 4 ++--
include/hw/cxl/cxl_pci.h | 10 ++++------
5 files changed, 14 insertions(+), 16 deletions(-)
diff --git a/hw/mem/cxl_type3.c b/hw/mem/cxl_type3.c
index 68d200144b..5d29d2595c 100644
--- a/hw/mem/cxl_type3.c
+++ b/hw/mem/cxl_type3.c
@@ -44,10 +44,10 @@ static void build_dvsecs(CXLType3Dev *ct3d)
dvsec = (uint8_t *)&(CXLDVSECRegisterLocator){
.rsvd = 0,
- .reg0_base_lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
- .reg0_base_hi = 0,
- .reg1_base_lo = RBI_CXL_DEVICE_REG | CXL_DEVICE_REG_BAR_IDX,
- .reg1_base_hi = 0,
+ .reg_base[0].lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
+ .reg_base[0].hi = 0,
+ .reg_base[1].lo = RBI_CXL_DEVICE_REG | CXL_DEVICE_REG_BAR_IDX,
+ .reg_base[1].hi = 0,
};
cxl_component_create_dvsec(cxl_cstate, CXL2_TYPE3_DEVICE,
REG_LOC_DVSEC_LENGTH, REG_LOC_DVSEC,
diff --git a/hw/pci-bridge/cxl_downstream.c b/hw/pci-bridge/cxl_downstream.c
index a361e519d0..7822ccd5de 100644
--- a/hw/pci-bridge/cxl_downstream.c
+++ b/hw/pci-bridge/cxl_downstream.c
@@ -126,8 +126,8 @@ static void build_dvsecs(CXLComponentState *cxl)
dvsec = (uint8_t *)&(CXLDVSECRegisterLocator){
.rsvd = 0,
- .reg0_base_lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
- .reg0_base_hi = 0,
+ .reg_base[0].lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
+ .reg_base[0].hi = 0,
};
cxl_component_create_dvsec(cxl, CXL2_DOWNSTREAM_PORT,
REG_LOC_DVSEC_LENGTH, REG_LOC_DVSEC,
diff --git a/hw/pci-bridge/cxl_root_port.c b/hw/pci-bridge/cxl_root_port.c
index fb213fa06e..08c2441dab 100644
--- a/hw/pci-bridge/cxl_root_port.c
+++ b/hw/pci-bridge/cxl_root_port.c
@@ -87,8 +87,8 @@ static void build_dvsecs(CXLComponentState *cxl)
dvsec = (uint8_t *)&(CXLDVSECRegisterLocator){
.rsvd = 0,
- .reg0_base_lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
- .reg0_base_hi = 0,
+ .reg_base[0].lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
+ .reg_base[0].hi = 0,
};
cxl_component_create_dvsec(cxl, CXL2_ROOT_PORT,
REG_LOC_DVSEC_LENGTH, REG_LOC_DVSEC,
diff --git a/hw/pci-bridge/cxl_upstream.c b/hw/pci-bridge/cxl_upstream.c
index a83a3e81e4..45ee6ba884 100644
--- a/hw/pci-bridge/cxl_upstream.c
+++ b/hw/pci-bridge/cxl_upstream.c
@@ -111,8 +111,8 @@ static void build_dvsecs(CXLComponentState *cxl)
dvsec = (uint8_t *)&(CXLDVSECRegisterLocator){
.rsvd = 0,
- .reg0_base_lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
- .reg0_base_hi = 0,
+ .reg_base[0].lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
+ .reg_base[0].hi = 0,
};
cxl_component_create_dvsec(cxl, CXL2_UPSTREAM_PORT,
REG_LOC_DVSEC_LENGTH, REG_LOC_DVSEC,
diff --git a/include/hw/cxl/cxl_pci.h b/include/hw/cxl/cxl_pci.h
index 01cf002096..8cbeb61142 100644
--- a/include/hw/cxl/cxl_pci.h
+++ b/include/hw/cxl/cxl_pci.h
@@ -141,12 +141,10 @@ QEMU_BUILD_BUG_ON(sizeof(CXLDVSECPortFlexBus) != 0x14);
typedef struct CXLDVSECRegisterLocator {
DVSECHeader hdr;
uint16_t rsvd;
- uint32_t reg0_base_lo;
- uint32_t reg0_base_hi;
- uint32_t reg1_base_lo;
- uint32_t reg1_base_hi;
- uint32_t reg2_base_lo;
- uint32_t reg2_base_hi;
+ struct {
+ uint32_t lo;
+ uint32_t hi;
+ } reg_base[3];
} CXLDVSECRegisterLocator;
QEMU_BUILD_BUG_ON(sizeof(CXLDVSECRegisterLocator) != 0x24);
--
2.32.0
^ permalink raw reply related [flat|nested] 8+ messages in thread
* [RFC PATCH 3/3] hw/cxl: CXL Performance Monitoring Unit (CPMU) Emulation
2022-08-31 15:33 ` Jonathan Cameron via
@ 2022-08-31 15:33 ` Jonathan Cameron via
-1 siblings, 0 replies; 8+ messages in thread
From: Jonathan Cameron @ 2022-08-31 15:33 UTC (permalink / raw)
To: qemu-devel, linux-cxl; +Cc: Ben Widawsky, Dan Williams
CXL 3.0 introduces a CXL Performance Monitoring Unit.
Add basic emulation of such units and instantiate 2 in
CXL Type 3 devices.
Limitations:
- Random selection of supported events.
- Mixture of fixed purpose and configurable counters is intended
to hit some edge cases in the discovery code, not to make any
practical sense.
- Currently no per CPMU configuration so the two instances are
currently identical. Not sure it's worth putting a configuration
interface in place as it would be very complex to use.
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
---
hw/cxl/cxl-cpmu.c | 295 ++++++++++++++++++++++++++++++++++++
hw/cxl/meson.build | 1 +
hw/mem/cxl_type3.c | 15 +-
include/hw/cxl/cxl.h | 1 +
include/hw/cxl/cxl_cpmu.h | 99 ++++++++++++
include/hw/cxl/cxl_device.h | 31 +++-
include/hw/cxl/cxl_pci.h | 7 +-
7 files changed, 441 insertions(+), 8 deletions(-)
diff --git a/hw/cxl/cxl-cpmu.c b/hw/cxl/cxl-cpmu.c
new file mode 100644
index 0000000000..ee647de607
--- /dev/null
+++ b/hw/cxl/cxl-cpmu.c
@@ -0,0 +1,295 @@
+/*
+ * CXL Performance Monitoring Unit
+ *
+ * Copyright(C) 2022 Jonathan Cameron - Huawei
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/log.h"
+#include "hw/cxl/cxl.h"
+
+#include "hw/pci/msi.h"
+#include "hw/pci/msix.h"
+
+static uint64_t cpmu_read(void *opaque, hwaddr offset, unsigned size)
+{
+ CPMUState *cpmu = opaque;
+ uint64_t retval = 0;
+ int index;
+
+ /* I'm lazy - lets assume 4 or 8 byte reads only - fix that up later. */
+ switch (offset) {
+ case A_CXL_CPMU_CAP:
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP,
+ NUM_COUNTERS, CPMU_NUM_COUNTERS - 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, COUNTER_WIDTH, 48);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP,
+ NUM_EVENT_GROUPS, CPMU_NUM_EVENT_GROUPS - 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, FILTERS_SUP, 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, MSI_N, cpmu->msi_n);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, WRITEABLE_WHEN_EN, 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, FREEZE, 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, INT, 1);
+ break;
+ case A_CXL_CPMU_OVERFLOW_STS:
+ retval = cpmu->overflow_status_bm;
+ break;
+ case A_CXL_CPMU_FREEZE:
+ retval = cpmu->freeze_status_bm;
+ break;
+ case A_CXL_CPMU_EVENT_CAP0:
+ /* Event group 0, clock ticks */
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP0, SUPPORTED_EVENTS, 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP0, EVENT_GROUP_ID, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP0,
+ EVENT_VENDOR_ID, 0x1e98);
+ break;
+ case A_CXL_CPMU_EVENT_CAP1:
+ /* Random mashup */
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, SUPPORTED_EVENTS,
+ 0xFF);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, EVENT_GROUP_ID, 0x10);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, EVENT_VENDOR_ID,
+ 0x1e98);
+ break;
+ case A_CXL_CPMU_EVENT_CAP2:
+ /* Random mashup */
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, SUPPORTED_EVENTS,
+ 0xFF);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, EVENT_GROUP_ID, 0x12);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, EVENT_VENDOR_ID,
+ 0x1e98);
+ break;
+
+ case A_CXL_CPMU_CNT0_CFG:
+ /* Lets' make this a fixed function counter doing only the first set */
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, TYPE,
+ CXL_CPMU_CNT_CFG_TYPE_FIXED_FUN);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, ENABLE,
+ cpmu->counter_enable[0]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, INT_ON_OVERFLOW,
+ cpmu->int_en[0]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, FREEZE_ON_OVERFLOW,
+ cpmu->freeze_en[0]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, EDGE, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, INVERT, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, THRESHOLD, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, EVENTS, 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, EVENT_GROUP_ID_IDX, 0);
+ break;
+ case A_CXL_CPMU_CNT1_CFG:
+ case A_CXL_CPMU_CNT2_CFG:
+ /* A couple of configurable counters */
+ index = (offset - A_CXL_CPMU_CNT0_CFG) / sizeof(uint64_t);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, TYPE,
+ CXL_CPMU_CNT_CFG_TYPE_CONFIGURABLE);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, ENABLE,
+ cpmu->counter_enable[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, INT_ON_OVERFLOW,
+ cpmu->int_en[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, FREEZE_ON_OVERFLOW,
+ cpmu->freeze_en[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, EDGE, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, INVERT, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, THRESHOLD, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, EVENTS, 0xFF);
+ /* Fixme - perhaps make this flexible at somepoint */
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG,
+ EVENT_GROUP_ID_IDX, 1);
+ break;
+ case A_CXL_CPMU_CNT3_CFG:
+ /* Try to break my code! */
+ index = (offset - A_CXL_CPMU_CNT0_CFG) / sizeof(uint64_t);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG,
+ TYPE, CXL_CPMU_CNT_CFG_TYPE_FIXED_FUN);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG,
+ ENABLE, cpmu->counter_enable[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG,
+ INT_ON_OVERFLOW, cpmu->int_en[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG,
+ FREEZE_ON_OVERFLOW, cpmu->freeze_en[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG, EDGE, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG, INVERT, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG, THRESHOLD, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG, EVENTS, 1 << 3);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG,
+ EVENT_GROUP_ID_IDX, 1);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT0_CFG:
+ retval = FIELD_DP64(retval, CXL_CPMU_FILTER0_CNT0_CFG, VALUE, 0xffff);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT1_CFG:
+ index = 1;
+ retval = FIELD_DP64(retval, CXL_CPMU_FILTER0_CNT1_CFG, VALUE,
+ cpmu->filter_value[0][index]);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT2_CFG:
+ index = 2;
+ retval = FIELD_DP64(retval, CXL_CPMU_FILTER0_CNT2_CFG, VALUE,
+ cpmu->filter_value[0][index]);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT3_CFG:
+ index = 3;
+ retval = FIELD_DP64(retval, CXL_CPMU_FILTER0_CNT3_CFG, VALUE,
+ cpmu->filter_value[0][index]);
+ break;
+ case A_CXL_CPMU_CNT0:
+ case A_CXL_CPMU_CNT1:
+ case A_CXL_CPMU_CNT2:
+ case A_CXL_CPMU_CNT3:
+ index = (offset - A_CXL_CPMU_CNT0) / sizeof(uint64_t);
+ retval = cpmu->counter[index] & 0xffffffffffff;
+
+ break;
+ default:
+ break;
+ }
+
+ return retval;
+}
+
+static void cpmu_write(void *opaque, hwaddr offset, uint64_t value,
+ unsigned size)
+{
+ CPMUState *cpmu = opaque;
+ int index;
+
+ /* Lazy and assume correct size reads and writes for now */
+ switch (offset) {
+ case 0:
+ break;
+ case 8:
+ break;
+ case A_CXL_CPMU_FREEZE:
+ cpmu->freeze_status_bm = value;
+ break;
+ case A_CXL_CPMU_OVERFLOW_STS:
+ cpmu->overflow_status_bm &= ~value;
+ break;
+ case A_CXL_CPMU_CNT0_CFG:
+ case A_CXL_CPMU_CNT1_CFG:
+ case A_CXL_CPMU_CNT2_CFG:
+ case A_CXL_CPMU_CNT3_CFG:
+ index = (offset - A_CXL_CPMU_CNT0_CFG) / sizeof(uint64_t);
+ cpmu->int_en[index] =
+ FIELD_EX32(value, CXL_CPMU_CNT0_CFG, INT_ON_OVERFLOW);
+ cpmu->freeze_en[index] =
+ FIELD_EX32(value, CXL_CPMU_CNT0_CFG, FREEZE_ON_OVERFLOW);
+ cpmu->counter_enable[index] =
+ FIELD_EX32(value, CXL_CPMU_CNT0_CFG, ENABLE);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT1_CFG:
+ index = 1;
+ cpmu->filter_value[0][index] =
+ FIELD_EX32(value, CXL_CPMU_FILTER0_CNT1_CFG, VALUE);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT2_CFG:
+ index = 2;
+ break;
+ case A_CXL_CPMU_FILTER0_CNT3_CFG:
+ index = 3;
+ cpmu->filter_value[0][index] =
+ FIELD_EX32(value, CXL_CPMU_FILTER0_CNT1_CFG, VALUE);
+ break;
+ case A_CXL_CPMU_CNT0:
+ case A_CXL_CPMU_CNT1:
+ case A_CXL_CPMU_CNT2:
+ case A_CXL_CPMU_CNT3:
+ index = (offset - A_CXL_CPMU_CNT0) / sizeof(uint64_t);
+ cpmu->counter[index] = value & 0xffffffffffff;
+ break;
+ }
+
+ return;
+}
+
+static const MemoryRegionOps cpmu_ops = {
+ .read = cpmu_read,
+ .write = cpmu_write,
+ .endianness = DEVICE_LITTLE_ENDIAN,
+ .valid = {
+ .min_access_size = 1,
+ .max_access_size = 8,
+ .unaligned = false,
+ },
+ .impl = {
+ .min_access_size = 4,
+ .max_access_size = 8,
+ },
+};
+
+
+static void cpmu_counter_update(void *opaque)
+{
+ CPMUState *cpmu = opaque;
+ CXLDeviceState *cxl_dstate = cpmu->private;
+ bool interrupt_needed = false;
+ int i;
+
+ timer_del(cpmu->timer);
+
+ for (i = 0; i < CPMU_NUM_COUNTERS; i++) {
+ /*
+ * TODO: check enabled, not frozen etc
+ *
+ * Hack to make the numbers get bigger! Ideally for at least
+ * some types of event we could hook it up to actual accesses.
+ */
+ uint64_t previous = cpmu->counter[i];
+ if (cpmu->counter_enable[i]) {
+ switch (i) {
+ case 0:
+ cpmu->counter[i] += (1ULL << 44) * 1 + 7;
+ break;
+ case 1:
+ cpmu->counter[i] += (1ULL << 43) * 1 + 3;
+ break;
+ case 2:
+ cpmu->counter[i] += (1ULL << 43) * 1 + 7;
+ break;
+ default:
+ cpmu->counter[i] += 30;
+ break;
+ }
+ if (cpmu->counter[i] / (1ULL << 48) != previous / (1ULL << 48)) {
+ cpmu->overflow_status_bm |= (1 << i);
+ if (cpmu->int_en[i]) {
+ interrupt_needed = true;
+ }
+ }
+ }
+ }
+ timer_mod(cpmu->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 100000000);
+ if (interrupt_needed) {
+ CXLType3Dev *dev = container_of(cxl_dstate, CXLType3Dev, cxl_dstate);
+
+ if (msix_enabled(&dev->parent_obj)) {
+ msix_notify(&dev->parent_obj, cpmu->msi_n);
+ } else if (msi_enabled(&dev->parent_obj)) {
+ msi_notify(&dev->parent_obj, cpmu->msi_n);
+ }
+ }
+}
+
+void cxl_cpmu_register_block_init(Object *obj, CXLDeviceState *cxl_dstate,
+ int id, uint8_t msi_n)
+{
+ CPMUState *cpmu = &cxl_dstate->cpmu[id];
+ MemoryRegion *registers = &cxl_dstate->cpmu_registers[id];
+ g_autofree gchar *name = g_strdup_printf("cpmu%d-registers", id);
+
+ cpmu->msi_n = msi_n;
+ cpmu->private = cxl_dstate;
+ memory_region_init_io(registers, obj, &cpmu_ops, cpmu,
+ name, pow2ceil(CXL_CPMU_SIZE));
+ cpmu->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, cpmu_counter_update,
+ &cxl_dstate->cpmu[id]);
+ timer_mod(cpmu->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 100000000);
+
+ /* Need to force 64k Alignment in the bar */
+ memory_region_add_subregion(&cxl_dstate->device_registers,
+ CXL_CPMU_OFFSET(id), registers);
+}
diff --git a/hw/cxl/meson.build b/hw/cxl/meson.build
index f117b99949..26ac763e8f 100644
--- a/hw/cxl/meson.build
+++ b/hw/cxl/meson.build
@@ -4,6 +4,7 @@ softmmu_ss.add(when: 'CONFIG_CXL',
'cxl-device-utils.c',
'cxl-mailbox-utils.c',
'cxl-host.c',
+ 'cxl-cpmu.c',
),
if_false: files(
'cxl-host-stubs.c',
diff --git a/hw/mem/cxl_type3.c b/hw/mem/cxl_type3.c
index 5d29d2595c..9242290bfe 100644
--- a/hw/mem/cxl_type3.c
+++ b/hw/mem/cxl_type3.c
@@ -25,7 +25,9 @@
static void build_dvsecs(CXLType3Dev *ct3d)
{
CXLComponentState *cxl_cstate = &ct3d->cxl_cstate;
+ CXLDVSECRegisterLocator *regloc_dvsec;
uint8_t *dvsec;
+ int i;
dvsec = (uint8_t *)&(CXLDVSECDevice){
.cap = 0x1e,
@@ -42,16 +44,21 @@ static void build_dvsecs(CXLType3Dev *ct3d)
PCIE_CXL_DEVICE_DVSEC,
PCIE_CXL2_DEVICE_DVSEC_REVID, dvsec);
- dvsec = (uint8_t *)&(CXLDVSECRegisterLocator){
+ regloc_dvsec = &(CXLDVSECRegisterLocator){
.rsvd = 0,
.reg_base[0].lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
.reg_base[0].hi = 0,
.reg_base[1].lo = RBI_CXL_DEVICE_REG | CXL_DEVICE_REG_BAR_IDX,
.reg_base[1].hi = 0,
};
+ for (i = 0; i < CXL_NUM_CPMU_INSTANCES; i++) {
+ regloc_dvsec->reg_base[2 + i].lo = CXL_CPMU_OFFSET(i) |
+ RBI_CXL_CPMU_REG | CXL_DEVICE_REG_BAR_IDX;
+ regloc_dvsec->reg_base[2 + i].hi = 0;
+ }
cxl_component_create_dvsec(cxl_cstate, CXL2_TYPE3_DEVICE,
REG_LOC_DVSEC_LENGTH, REG_LOC_DVSEC,
- REG_LOC_DVSEC_REVID, dvsec);
+ REG_LOC_DVSEC_REVID, (uint8_t *)regloc_dvsec);
dvsec = (uint8_t *)&(CXLDVSECDeviceGPF){
.phase2_duration = 0x603, /* 3 seconds */
.phase2_power = 0x33, /* 0x33 miliwatts */
@@ -148,7 +155,7 @@ static void ct3_realize(PCIDevice *pci_dev, Error **errp)
ComponentRegisters *regs = &cxl_cstate->crb;
MemoryRegion *mr = ®s->component_registers;
uint8_t *pci_conf = pci_dev->config;
- unsigned short msix_num = 1;
+ unsigned short msix_num = 4;
int i;
if (!cxl_setup_memory(ct3d, errp)) {
@@ -180,6 +187,8 @@ static void ct3_realize(PCIDevice *pci_dev, Error **errp)
PCI_BASE_ADDRESS_SPACE_MEMORY | PCI_BASE_ADDRESS_MEM_TYPE_64, mr);
cxl_device_register_block_init(OBJECT(pci_dev), &ct3d->cxl_dstate);
+ cxl_cpmu_register_block_init(OBJECT(pci_dev), &ct3d->cxl_dstate, 0, 3);
+ cxl_cpmu_register_block_init(OBJECT(pci_dev), &ct3d->cxl_dstate, 1, 3);
pci_register_bar(pci_dev, CXL_DEVICE_REG_BAR_IDX,
PCI_BASE_ADDRESS_SPACE_MEMORY |
PCI_BASE_ADDRESS_MEM_TYPE_64,
diff --git a/include/hw/cxl/cxl.h b/include/hw/cxl/cxl.h
index 38e0e271d5..97aa379d18 100644
--- a/include/hw/cxl/cxl.h
+++ b/include/hw/cxl/cxl.h
@@ -17,6 +17,7 @@
#include "hw/pci/pci_host.h"
#include "cxl_pci.h"
#include "cxl_component.h"
+#include "cxl_cpmu.h"
#include "cxl_device.h"
#define CXL_COMPONENT_REG_BAR_IDX 0
diff --git a/include/hw/cxl/cxl_cpmu.h b/include/hw/cxl/cxl_cpmu.h
new file mode 100644
index 0000000000..d68b15c965
--- /dev/null
+++ b/include/hw/cxl/cxl_cpmu.h
@@ -0,0 +1,99 @@
+/*
+ * QEMU CXL Performance Monitoring Unit
+ *
+ * Copyright (c) 2021 Jonathan Cameron - Huawei
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "hw/register.h"
+
+#ifndef _CXL_CPMU_H_
+#define _CXL_CPMU_H_
+
+REG32(CXL_CPMU_CAP, 0)
+ FIELD(CXL_CPMU_CAP, NUM_COUNTERS, 0, 6)
+ FIELD(CXL_CPMU_CAP, COUNTER_WIDTH, 8, 8)
+ FIELD(CXL_CPMU_CAP, NUM_EVENT_GROUPS, 20, 5)
+ FIELD(CXL_CPMU_CAP, FILTERS_SUP, 32, 8)
+ FIELD(CXL_CPMU_CAP, MSI_N, 44, 4)
+ FIELD(CXL_CPMU_CAP, WRITEABLE_WHEN_EN, 48, 1)
+ FIELD(CXL_CPMU_CAP, FREEZE, 49, 1)
+ FIELD(CXL_CPMU_CAP, INT, 50, 1)
+/* Register at offset 0x8 is reserved */
+REG32(CXL_CPMU_OVERFLOW_STS, 0x10)
+REG32(CXL_CPMU_FREEZE, 0x18)
+/* Registers at offset 20 + reserved */
+
+#define CXL_CPMU_EVENT_CAP(n) \
+ REG64(CXL_CPMU_EVENT_CAP##n, 0x100 + 8 * (n)) \
+ FIELD(CXL_CPMU_EVENT_CAP##n, SUPPORTED_EVENTS, 0, 32) \
+ FIELD(CXL_CPMU_EVENT_CAP##n, EVENT_GROUP_ID, 32, 16) \
+ FIELD(CXL_CPMU_EVENT_CAP##n, EVENT_VENDOR_ID, 48, 16)
+
+CXL_CPMU_EVENT_CAP(0)
+CXL_CPMU_EVENT_CAP(1)
+CXL_CPMU_EVENT_CAP(2)
+CXL_CPMU_EVENT_CAP(3)
+
+#define CXL_CPMU_CNT_CFG(n) \
+ REG64(CXL_CPMU_CNT##n##_CFG, 0x200 + 8 * (n)) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, TYPE, 0, 2) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, ENABLE, 8, 1) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, INT_ON_OVERFLOW, 9, 1) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, FREEZE_ON_OVERFLOW, 10, 1) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, EDGE, 11, 1) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, INVERT, 12, 1) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, THRESHOLD, 16, 8) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, EVENTS, 24, 32) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, EVENT_GROUP_ID_IDX, 59, 5)
+
+#define CXL_CPMU_CNT_CFG_TYPE_FREE_RUN 0
+#define CXL_CPMU_CNT_CFG_TYPE_FIXED_FUN 1
+#define CXL_CPMU_CNT_CFG_TYPE_CONFIGURABLE 2
+CXL_CPMU_CNT_CFG(0)
+CXL_CPMU_CNT_CFG(1)
+CXL_CPMU_CNT_CFG(2)
+CXL_CPMU_CNT_CFG(3)
+
+#define CXL_CPMU_FILTER_CFG(n, f) \
+ REG64(CXL_CPMU_FILTER##f##_CNT##n##_CFG, 0x400 + 4 * ((f) + (n) * 8)) \
+ FIELD(CXL_CPMU_FILTER##f##_CNT##n##_CFG, VALUE, 0, 16)
+
+/* Only HDM decoder filter suppored - no effect on first counter */
+CXL_CPMU_FILTER_CFG(0, 0)
+CXL_CPMU_FILTER_CFG(1, 0)
+CXL_CPMU_FILTER_CFG(2, 0)
+CXL_CPMU_FILTER_CFG(3, 0)
+
+#define CXL_CPMU_CNT(n) \
+ REG64(CXL_CPMU_CNT##n, 0xc00 + 8 * (n))
+
+CXL_CPMU_CNT(0)
+CXL_CPMU_CNT(1)
+CXL_CPMU_CNT(2)
+CXL_CPMU_CNT(3)
+
+typedef struct CPMUState {
+#define CPMU_NUM_COUNTERS 4
+#define CPMU_NUM_EVENT_GROUPS 3
+#define CPMU_NUM_FILTERS 1
+ bool counter_enable[CPMU_NUM_COUNTERS];
+ bool int_en[CPMU_NUM_COUNTERS];
+ bool freeze_en[CPMU_NUM_COUNTERS];
+ bool filter_value[CPMU_NUM_FILTERS][CPMU_NUM_COUNTERS];
+ uint64_t counter[CPMU_NUM_COUNTERS];
+ uint64_t overflow_status_bm;
+ uint64_t freeze_status_bm;
+ uint8_t msi_n;
+ QEMUTimer *timer;
+ void *private;
+} CPMUState;
+
+typedef struct cxl_device_state CXLDeviceState;
+void cxl_cpmu_register_block_init(Object *obj,
+ CXLDeviceState *cxl_dstate,
+ int id, uint8_t msi_n);
+
+#endif
diff --git a/include/hw/cxl/cxl_device.h b/include/hw/cxl/cxl_device.h
index e4d221cdb3..84c7728a96 100644
--- a/include/hw/cxl/cxl_device.h
+++ b/include/hw/cxl/cxl_device.h
@@ -12,6 +12,7 @@
#include "hw/register.h"
+#include "hw/cxl/cxl_cpmu.h"
/*
* The following is how a CXL device's Memory Device registers are laid out.
* The only requirement from the spec is that the capabilities array and the
@@ -24,6 +25,16 @@
*
* +---------------------------------+
* | |
+ * | CPMU1 |
+ * | |
+ * +---------------------------------+
+ * | |
+ * | CPMU0 |
+ * | |
+ * +---------------------------------+
+ * | Padding to 64k Aligned |
+ * +---------------------------------+
+ * | |
* | Memory Device Registers |
* | |
* n + PAYLOAD_SIZE_MAX -----------------------------------
@@ -76,13 +87,26 @@
(CXL_MAILBOX_REGISTERS_OFFSET + CXL_MAILBOX_REGISTERS_LENGTH)
#define CXL_MEMORY_DEVICE_REGISTERS_LENGTH 0x8
+#define CXL_NUM_CPMU_INSTANCES 2
+#define CXL_CPMU_SIZE 0x8f8
+#define CXL_CPMU_OFFSET(x) \
+ QEMU_ALIGN_UP(CXL_MEMORY_DEVICE_REGISTERS_OFFSET + \
+ CXL_MEMORY_DEVICE_REGISTERS_LENGTH + \
+ (x) * (1 << 16), \
+ 1 << 16)
+
#define CXL_MMIO_SIZE \
- (CXL_DEVICE_CAP_REG_SIZE + CXL_DEVICE_STATUS_REGISTERS_LENGTH + \
- CXL_MAILBOX_REGISTERS_LENGTH + CXL_MEMORY_DEVICE_REGISTERS_LENGTH)
+ QEMU_ALIGN_UP(CXL_DEVICE_CAP_REG_SIZE + \
+ CXL_DEVICE_STATUS_REGISTERS_LENGTH + \
+ CXL_MAILBOX_REGISTERS_LENGTH + \
+ CXL_MEMORY_DEVICE_REGISTERS_LENGTH + \
+ CXL_NUM_CPMU_INSTANCES * (1 << 16), \
+ (1 << 16))
typedef struct cxl_device_state {
MemoryRegion device_registers;
+ MemoryRegion cpmu_registers[CXL_NUM_CPMU_INSTANCES];
/* mmio for device capabilities array - 8.2.8.2 */
MemoryRegion device;
MemoryRegion memory_device;
@@ -119,6 +143,9 @@ typedef struct cxl_device_state {
/* memory region for persistent memory, HDM */
uint64_t pmem_size;
+
+ /* Move me later */
+ CPMUState cpmu[CXL_NUM_CPMU_INSTANCES];
} CXLDeviceState;
/* Initialize the register block for a device */
diff --git a/include/hw/cxl/cxl_pci.h b/include/hw/cxl/cxl_pci.h
index 8cbeb61142..b7b3620a9f 100644
--- a/include/hw/cxl/cxl_pci.h
+++ b/include/hw/cxl/cxl_pci.h
@@ -35,7 +35,7 @@
#define PCIE_FLEXBUS_PORT_DVSEC_LENGTH_2_0 0x14
#define PCIE_FLEXBUS_PORT_DVSEC_REVID_2_0 1
-#define REG_LOC_DVSEC_LENGTH 0x24
+#define REG_LOC_DVSEC_LENGTH 0x2c
#define REG_LOC_DVSEC_REVID 0
enum {
@@ -144,9 +144,9 @@ typedef struct CXLDVSECRegisterLocator {
struct {
uint32_t lo;
uint32_t hi;
- } reg_base[3];
+ } reg_base[4];
} CXLDVSECRegisterLocator;
-QEMU_BUILD_BUG_ON(sizeof(CXLDVSECRegisterLocator) != 0x24);
+QEMU_BUILD_BUG_ON(sizeof(CXLDVSECRegisterLocator) != 0x2C);
/* BAR Equivalence Indicator */
#define BEI_BAR_10H 0
@@ -161,5 +161,6 @@ QEMU_BUILD_BUG_ON(sizeof(CXLDVSECRegisterLocator) != 0x24);
#define RBI_COMPONENT_REG (1 << 8)
#define RBI_BAR_VIRT_ACL (2 << 8)
#define RBI_CXL_DEVICE_REG (3 << 8)
+#define RBI_CXL_CPMU_REG (4 << 8)
#endif
--
2.32.0
^ permalink raw reply related [flat|nested] 8+ messages in thread
* [RFC PATCH 3/3] hw/cxl: CXL Performance Monitoring Unit (CPMU) Emulation
@ 2022-08-31 15:33 ` Jonathan Cameron via
0 siblings, 0 replies; 8+ messages in thread
From: Jonathan Cameron via @ 2022-08-31 15:33 UTC (permalink / raw)
To: qemu-devel, linux-cxl; +Cc: Ben Widawsky, Dan Williams
CXL 3.0 introduces a CXL Performance Monitoring Unit.
Add basic emulation of such units and instantiate 2 in
CXL Type 3 devices.
Limitations:
- Random selection of supported events.
- Mixture of fixed purpose and configurable counters is intended
to hit some edge cases in the discovery code, not to make any
practical sense.
- Currently no per CPMU configuration so the two instances are
currently identical. Not sure it's worth putting a configuration
interface in place as it would be very complex to use.
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
---
hw/cxl/cxl-cpmu.c | 295 ++++++++++++++++++++++++++++++++++++
hw/cxl/meson.build | 1 +
hw/mem/cxl_type3.c | 15 +-
include/hw/cxl/cxl.h | 1 +
include/hw/cxl/cxl_cpmu.h | 99 ++++++++++++
include/hw/cxl/cxl_device.h | 31 +++-
include/hw/cxl/cxl_pci.h | 7 +-
7 files changed, 441 insertions(+), 8 deletions(-)
diff --git a/hw/cxl/cxl-cpmu.c b/hw/cxl/cxl-cpmu.c
new file mode 100644
index 0000000000..ee647de607
--- /dev/null
+++ b/hw/cxl/cxl-cpmu.c
@@ -0,0 +1,295 @@
+/*
+ * CXL Performance Monitoring Unit
+ *
+ * Copyright(C) 2022 Jonathan Cameron - Huawei
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/log.h"
+#include "hw/cxl/cxl.h"
+
+#include "hw/pci/msi.h"
+#include "hw/pci/msix.h"
+
+static uint64_t cpmu_read(void *opaque, hwaddr offset, unsigned size)
+{
+ CPMUState *cpmu = opaque;
+ uint64_t retval = 0;
+ int index;
+
+ /* I'm lazy - lets assume 4 or 8 byte reads only - fix that up later. */
+ switch (offset) {
+ case A_CXL_CPMU_CAP:
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP,
+ NUM_COUNTERS, CPMU_NUM_COUNTERS - 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, COUNTER_WIDTH, 48);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP,
+ NUM_EVENT_GROUPS, CPMU_NUM_EVENT_GROUPS - 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, FILTERS_SUP, 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, MSI_N, cpmu->msi_n);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, WRITEABLE_WHEN_EN, 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, FREEZE, 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CAP, INT, 1);
+ break;
+ case A_CXL_CPMU_OVERFLOW_STS:
+ retval = cpmu->overflow_status_bm;
+ break;
+ case A_CXL_CPMU_FREEZE:
+ retval = cpmu->freeze_status_bm;
+ break;
+ case A_CXL_CPMU_EVENT_CAP0:
+ /* Event group 0, clock ticks */
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP0, SUPPORTED_EVENTS, 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP0, EVENT_GROUP_ID, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP0,
+ EVENT_VENDOR_ID, 0x1e98);
+ break;
+ case A_CXL_CPMU_EVENT_CAP1:
+ /* Random mashup */
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, SUPPORTED_EVENTS,
+ 0xFF);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, EVENT_GROUP_ID, 0x10);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, EVENT_VENDOR_ID,
+ 0x1e98);
+ break;
+ case A_CXL_CPMU_EVENT_CAP2:
+ /* Random mashup */
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, SUPPORTED_EVENTS,
+ 0xFF);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, EVENT_GROUP_ID, 0x12);
+ retval = FIELD_DP64(retval, CXL_CPMU_EVENT_CAP1, EVENT_VENDOR_ID,
+ 0x1e98);
+ break;
+
+ case A_CXL_CPMU_CNT0_CFG:
+ /* Lets' make this a fixed function counter doing only the first set */
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, TYPE,
+ CXL_CPMU_CNT_CFG_TYPE_FIXED_FUN);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, ENABLE,
+ cpmu->counter_enable[0]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, INT_ON_OVERFLOW,
+ cpmu->int_en[0]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, FREEZE_ON_OVERFLOW,
+ cpmu->freeze_en[0]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, EDGE, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, INVERT, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, THRESHOLD, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, EVENTS, 1);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT0_CFG, EVENT_GROUP_ID_IDX, 0);
+ break;
+ case A_CXL_CPMU_CNT1_CFG:
+ case A_CXL_CPMU_CNT2_CFG:
+ /* A couple of configurable counters */
+ index = (offset - A_CXL_CPMU_CNT0_CFG) / sizeof(uint64_t);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, TYPE,
+ CXL_CPMU_CNT_CFG_TYPE_CONFIGURABLE);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, ENABLE,
+ cpmu->counter_enable[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, INT_ON_OVERFLOW,
+ cpmu->int_en[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, FREEZE_ON_OVERFLOW,
+ cpmu->freeze_en[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, EDGE, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, INVERT, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, THRESHOLD, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG, EVENTS, 0xFF);
+ /* Fixme - perhaps make this flexible at somepoint */
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT1_CFG,
+ EVENT_GROUP_ID_IDX, 1);
+ break;
+ case A_CXL_CPMU_CNT3_CFG:
+ /* Try to break my code! */
+ index = (offset - A_CXL_CPMU_CNT0_CFG) / sizeof(uint64_t);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG,
+ TYPE, CXL_CPMU_CNT_CFG_TYPE_FIXED_FUN);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG,
+ ENABLE, cpmu->counter_enable[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG,
+ INT_ON_OVERFLOW, cpmu->int_en[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG,
+ FREEZE_ON_OVERFLOW, cpmu->freeze_en[index]);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG, EDGE, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG, INVERT, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG, THRESHOLD, 0);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG, EVENTS, 1 << 3);
+ retval = FIELD_DP64(retval, CXL_CPMU_CNT3_CFG,
+ EVENT_GROUP_ID_IDX, 1);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT0_CFG:
+ retval = FIELD_DP64(retval, CXL_CPMU_FILTER0_CNT0_CFG, VALUE, 0xffff);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT1_CFG:
+ index = 1;
+ retval = FIELD_DP64(retval, CXL_CPMU_FILTER0_CNT1_CFG, VALUE,
+ cpmu->filter_value[0][index]);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT2_CFG:
+ index = 2;
+ retval = FIELD_DP64(retval, CXL_CPMU_FILTER0_CNT2_CFG, VALUE,
+ cpmu->filter_value[0][index]);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT3_CFG:
+ index = 3;
+ retval = FIELD_DP64(retval, CXL_CPMU_FILTER0_CNT3_CFG, VALUE,
+ cpmu->filter_value[0][index]);
+ break;
+ case A_CXL_CPMU_CNT0:
+ case A_CXL_CPMU_CNT1:
+ case A_CXL_CPMU_CNT2:
+ case A_CXL_CPMU_CNT3:
+ index = (offset - A_CXL_CPMU_CNT0) / sizeof(uint64_t);
+ retval = cpmu->counter[index] & 0xffffffffffff;
+
+ break;
+ default:
+ break;
+ }
+
+ return retval;
+}
+
+static void cpmu_write(void *opaque, hwaddr offset, uint64_t value,
+ unsigned size)
+{
+ CPMUState *cpmu = opaque;
+ int index;
+
+ /* Lazy and assume correct size reads and writes for now */
+ switch (offset) {
+ case 0:
+ break;
+ case 8:
+ break;
+ case A_CXL_CPMU_FREEZE:
+ cpmu->freeze_status_bm = value;
+ break;
+ case A_CXL_CPMU_OVERFLOW_STS:
+ cpmu->overflow_status_bm &= ~value;
+ break;
+ case A_CXL_CPMU_CNT0_CFG:
+ case A_CXL_CPMU_CNT1_CFG:
+ case A_CXL_CPMU_CNT2_CFG:
+ case A_CXL_CPMU_CNT3_CFG:
+ index = (offset - A_CXL_CPMU_CNT0_CFG) / sizeof(uint64_t);
+ cpmu->int_en[index] =
+ FIELD_EX32(value, CXL_CPMU_CNT0_CFG, INT_ON_OVERFLOW);
+ cpmu->freeze_en[index] =
+ FIELD_EX32(value, CXL_CPMU_CNT0_CFG, FREEZE_ON_OVERFLOW);
+ cpmu->counter_enable[index] =
+ FIELD_EX32(value, CXL_CPMU_CNT0_CFG, ENABLE);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT1_CFG:
+ index = 1;
+ cpmu->filter_value[0][index] =
+ FIELD_EX32(value, CXL_CPMU_FILTER0_CNT1_CFG, VALUE);
+ break;
+ case A_CXL_CPMU_FILTER0_CNT2_CFG:
+ index = 2;
+ break;
+ case A_CXL_CPMU_FILTER0_CNT3_CFG:
+ index = 3;
+ cpmu->filter_value[0][index] =
+ FIELD_EX32(value, CXL_CPMU_FILTER0_CNT1_CFG, VALUE);
+ break;
+ case A_CXL_CPMU_CNT0:
+ case A_CXL_CPMU_CNT1:
+ case A_CXL_CPMU_CNT2:
+ case A_CXL_CPMU_CNT3:
+ index = (offset - A_CXL_CPMU_CNT0) / sizeof(uint64_t);
+ cpmu->counter[index] = value & 0xffffffffffff;
+ break;
+ }
+
+ return;
+}
+
+static const MemoryRegionOps cpmu_ops = {
+ .read = cpmu_read,
+ .write = cpmu_write,
+ .endianness = DEVICE_LITTLE_ENDIAN,
+ .valid = {
+ .min_access_size = 1,
+ .max_access_size = 8,
+ .unaligned = false,
+ },
+ .impl = {
+ .min_access_size = 4,
+ .max_access_size = 8,
+ },
+};
+
+
+static void cpmu_counter_update(void *opaque)
+{
+ CPMUState *cpmu = opaque;
+ CXLDeviceState *cxl_dstate = cpmu->private;
+ bool interrupt_needed = false;
+ int i;
+
+ timer_del(cpmu->timer);
+
+ for (i = 0; i < CPMU_NUM_COUNTERS; i++) {
+ /*
+ * TODO: check enabled, not frozen etc
+ *
+ * Hack to make the numbers get bigger! Ideally for at least
+ * some types of event we could hook it up to actual accesses.
+ */
+ uint64_t previous = cpmu->counter[i];
+ if (cpmu->counter_enable[i]) {
+ switch (i) {
+ case 0:
+ cpmu->counter[i] += (1ULL << 44) * 1 + 7;
+ break;
+ case 1:
+ cpmu->counter[i] += (1ULL << 43) * 1 + 3;
+ break;
+ case 2:
+ cpmu->counter[i] += (1ULL << 43) * 1 + 7;
+ break;
+ default:
+ cpmu->counter[i] += 30;
+ break;
+ }
+ if (cpmu->counter[i] / (1ULL << 48) != previous / (1ULL << 48)) {
+ cpmu->overflow_status_bm |= (1 << i);
+ if (cpmu->int_en[i]) {
+ interrupt_needed = true;
+ }
+ }
+ }
+ }
+ timer_mod(cpmu->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 100000000);
+ if (interrupt_needed) {
+ CXLType3Dev *dev = container_of(cxl_dstate, CXLType3Dev, cxl_dstate);
+
+ if (msix_enabled(&dev->parent_obj)) {
+ msix_notify(&dev->parent_obj, cpmu->msi_n);
+ } else if (msi_enabled(&dev->parent_obj)) {
+ msi_notify(&dev->parent_obj, cpmu->msi_n);
+ }
+ }
+}
+
+void cxl_cpmu_register_block_init(Object *obj, CXLDeviceState *cxl_dstate,
+ int id, uint8_t msi_n)
+{
+ CPMUState *cpmu = &cxl_dstate->cpmu[id];
+ MemoryRegion *registers = &cxl_dstate->cpmu_registers[id];
+ g_autofree gchar *name = g_strdup_printf("cpmu%d-registers", id);
+
+ cpmu->msi_n = msi_n;
+ cpmu->private = cxl_dstate;
+ memory_region_init_io(registers, obj, &cpmu_ops, cpmu,
+ name, pow2ceil(CXL_CPMU_SIZE));
+ cpmu->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, cpmu_counter_update,
+ &cxl_dstate->cpmu[id]);
+ timer_mod(cpmu->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 100000000);
+
+ /* Need to force 64k Alignment in the bar */
+ memory_region_add_subregion(&cxl_dstate->device_registers,
+ CXL_CPMU_OFFSET(id), registers);
+}
diff --git a/hw/cxl/meson.build b/hw/cxl/meson.build
index f117b99949..26ac763e8f 100644
--- a/hw/cxl/meson.build
+++ b/hw/cxl/meson.build
@@ -4,6 +4,7 @@ softmmu_ss.add(when: 'CONFIG_CXL',
'cxl-device-utils.c',
'cxl-mailbox-utils.c',
'cxl-host.c',
+ 'cxl-cpmu.c',
),
if_false: files(
'cxl-host-stubs.c',
diff --git a/hw/mem/cxl_type3.c b/hw/mem/cxl_type3.c
index 5d29d2595c..9242290bfe 100644
--- a/hw/mem/cxl_type3.c
+++ b/hw/mem/cxl_type3.c
@@ -25,7 +25,9 @@
static void build_dvsecs(CXLType3Dev *ct3d)
{
CXLComponentState *cxl_cstate = &ct3d->cxl_cstate;
+ CXLDVSECRegisterLocator *regloc_dvsec;
uint8_t *dvsec;
+ int i;
dvsec = (uint8_t *)&(CXLDVSECDevice){
.cap = 0x1e,
@@ -42,16 +44,21 @@ static void build_dvsecs(CXLType3Dev *ct3d)
PCIE_CXL_DEVICE_DVSEC,
PCIE_CXL2_DEVICE_DVSEC_REVID, dvsec);
- dvsec = (uint8_t *)&(CXLDVSECRegisterLocator){
+ regloc_dvsec = &(CXLDVSECRegisterLocator){
.rsvd = 0,
.reg_base[0].lo = RBI_COMPONENT_REG | CXL_COMPONENT_REG_BAR_IDX,
.reg_base[0].hi = 0,
.reg_base[1].lo = RBI_CXL_DEVICE_REG | CXL_DEVICE_REG_BAR_IDX,
.reg_base[1].hi = 0,
};
+ for (i = 0; i < CXL_NUM_CPMU_INSTANCES; i++) {
+ regloc_dvsec->reg_base[2 + i].lo = CXL_CPMU_OFFSET(i) |
+ RBI_CXL_CPMU_REG | CXL_DEVICE_REG_BAR_IDX;
+ regloc_dvsec->reg_base[2 + i].hi = 0;
+ }
cxl_component_create_dvsec(cxl_cstate, CXL2_TYPE3_DEVICE,
REG_LOC_DVSEC_LENGTH, REG_LOC_DVSEC,
- REG_LOC_DVSEC_REVID, dvsec);
+ REG_LOC_DVSEC_REVID, (uint8_t *)regloc_dvsec);
dvsec = (uint8_t *)&(CXLDVSECDeviceGPF){
.phase2_duration = 0x603, /* 3 seconds */
.phase2_power = 0x33, /* 0x33 miliwatts */
@@ -148,7 +155,7 @@ static void ct3_realize(PCIDevice *pci_dev, Error **errp)
ComponentRegisters *regs = &cxl_cstate->crb;
MemoryRegion *mr = ®s->component_registers;
uint8_t *pci_conf = pci_dev->config;
- unsigned short msix_num = 1;
+ unsigned short msix_num = 4;
int i;
if (!cxl_setup_memory(ct3d, errp)) {
@@ -180,6 +187,8 @@ static void ct3_realize(PCIDevice *pci_dev, Error **errp)
PCI_BASE_ADDRESS_SPACE_MEMORY | PCI_BASE_ADDRESS_MEM_TYPE_64, mr);
cxl_device_register_block_init(OBJECT(pci_dev), &ct3d->cxl_dstate);
+ cxl_cpmu_register_block_init(OBJECT(pci_dev), &ct3d->cxl_dstate, 0, 3);
+ cxl_cpmu_register_block_init(OBJECT(pci_dev), &ct3d->cxl_dstate, 1, 3);
pci_register_bar(pci_dev, CXL_DEVICE_REG_BAR_IDX,
PCI_BASE_ADDRESS_SPACE_MEMORY |
PCI_BASE_ADDRESS_MEM_TYPE_64,
diff --git a/include/hw/cxl/cxl.h b/include/hw/cxl/cxl.h
index 38e0e271d5..97aa379d18 100644
--- a/include/hw/cxl/cxl.h
+++ b/include/hw/cxl/cxl.h
@@ -17,6 +17,7 @@
#include "hw/pci/pci_host.h"
#include "cxl_pci.h"
#include "cxl_component.h"
+#include "cxl_cpmu.h"
#include "cxl_device.h"
#define CXL_COMPONENT_REG_BAR_IDX 0
diff --git a/include/hw/cxl/cxl_cpmu.h b/include/hw/cxl/cxl_cpmu.h
new file mode 100644
index 0000000000..d68b15c965
--- /dev/null
+++ b/include/hw/cxl/cxl_cpmu.h
@@ -0,0 +1,99 @@
+/*
+ * QEMU CXL Performance Monitoring Unit
+ *
+ * Copyright (c) 2021 Jonathan Cameron - Huawei
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "hw/register.h"
+
+#ifndef _CXL_CPMU_H_
+#define _CXL_CPMU_H_
+
+REG32(CXL_CPMU_CAP, 0)
+ FIELD(CXL_CPMU_CAP, NUM_COUNTERS, 0, 6)
+ FIELD(CXL_CPMU_CAP, COUNTER_WIDTH, 8, 8)
+ FIELD(CXL_CPMU_CAP, NUM_EVENT_GROUPS, 20, 5)
+ FIELD(CXL_CPMU_CAP, FILTERS_SUP, 32, 8)
+ FIELD(CXL_CPMU_CAP, MSI_N, 44, 4)
+ FIELD(CXL_CPMU_CAP, WRITEABLE_WHEN_EN, 48, 1)
+ FIELD(CXL_CPMU_CAP, FREEZE, 49, 1)
+ FIELD(CXL_CPMU_CAP, INT, 50, 1)
+/* Register at offset 0x8 is reserved */
+REG32(CXL_CPMU_OVERFLOW_STS, 0x10)
+REG32(CXL_CPMU_FREEZE, 0x18)
+/* Registers at offset 20 + reserved */
+
+#define CXL_CPMU_EVENT_CAP(n) \
+ REG64(CXL_CPMU_EVENT_CAP##n, 0x100 + 8 * (n)) \
+ FIELD(CXL_CPMU_EVENT_CAP##n, SUPPORTED_EVENTS, 0, 32) \
+ FIELD(CXL_CPMU_EVENT_CAP##n, EVENT_GROUP_ID, 32, 16) \
+ FIELD(CXL_CPMU_EVENT_CAP##n, EVENT_VENDOR_ID, 48, 16)
+
+CXL_CPMU_EVENT_CAP(0)
+CXL_CPMU_EVENT_CAP(1)
+CXL_CPMU_EVENT_CAP(2)
+CXL_CPMU_EVENT_CAP(3)
+
+#define CXL_CPMU_CNT_CFG(n) \
+ REG64(CXL_CPMU_CNT##n##_CFG, 0x200 + 8 * (n)) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, TYPE, 0, 2) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, ENABLE, 8, 1) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, INT_ON_OVERFLOW, 9, 1) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, FREEZE_ON_OVERFLOW, 10, 1) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, EDGE, 11, 1) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, INVERT, 12, 1) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, THRESHOLD, 16, 8) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, EVENTS, 24, 32) \
+ FIELD(CXL_CPMU_CNT##n##_CFG, EVENT_GROUP_ID_IDX, 59, 5)
+
+#define CXL_CPMU_CNT_CFG_TYPE_FREE_RUN 0
+#define CXL_CPMU_CNT_CFG_TYPE_FIXED_FUN 1
+#define CXL_CPMU_CNT_CFG_TYPE_CONFIGURABLE 2
+CXL_CPMU_CNT_CFG(0)
+CXL_CPMU_CNT_CFG(1)
+CXL_CPMU_CNT_CFG(2)
+CXL_CPMU_CNT_CFG(3)
+
+#define CXL_CPMU_FILTER_CFG(n, f) \
+ REG64(CXL_CPMU_FILTER##f##_CNT##n##_CFG, 0x400 + 4 * ((f) + (n) * 8)) \
+ FIELD(CXL_CPMU_FILTER##f##_CNT##n##_CFG, VALUE, 0, 16)
+
+/* Only HDM decoder filter suppored - no effect on first counter */
+CXL_CPMU_FILTER_CFG(0, 0)
+CXL_CPMU_FILTER_CFG(1, 0)
+CXL_CPMU_FILTER_CFG(2, 0)
+CXL_CPMU_FILTER_CFG(3, 0)
+
+#define CXL_CPMU_CNT(n) \
+ REG64(CXL_CPMU_CNT##n, 0xc00 + 8 * (n))
+
+CXL_CPMU_CNT(0)
+CXL_CPMU_CNT(1)
+CXL_CPMU_CNT(2)
+CXL_CPMU_CNT(3)
+
+typedef struct CPMUState {
+#define CPMU_NUM_COUNTERS 4
+#define CPMU_NUM_EVENT_GROUPS 3
+#define CPMU_NUM_FILTERS 1
+ bool counter_enable[CPMU_NUM_COUNTERS];
+ bool int_en[CPMU_NUM_COUNTERS];
+ bool freeze_en[CPMU_NUM_COUNTERS];
+ bool filter_value[CPMU_NUM_FILTERS][CPMU_NUM_COUNTERS];
+ uint64_t counter[CPMU_NUM_COUNTERS];
+ uint64_t overflow_status_bm;
+ uint64_t freeze_status_bm;
+ uint8_t msi_n;
+ QEMUTimer *timer;
+ void *private;
+} CPMUState;
+
+typedef struct cxl_device_state CXLDeviceState;
+void cxl_cpmu_register_block_init(Object *obj,
+ CXLDeviceState *cxl_dstate,
+ int id, uint8_t msi_n);
+
+#endif
diff --git a/include/hw/cxl/cxl_device.h b/include/hw/cxl/cxl_device.h
index e4d221cdb3..84c7728a96 100644
--- a/include/hw/cxl/cxl_device.h
+++ b/include/hw/cxl/cxl_device.h
@@ -12,6 +12,7 @@
#include "hw/register.h"
+#include "hw/cxl/cxl_cpmu.h"
/*
* The following is how a CXL device's Memory Device registers are laid out.
* The only requirement from the spec is that the capabilities array and the
@@ -24,6 +25,16 @@
*
* +---------------------------------+
* | |
+ * | CPMU1 |
+ * | |
+ * +---------------------------------+
+ * | |
+ * | CPMU0 |
+ * | |
+ * +---------------------------------+
+ * | Padding to 64k Aligned |
+ * +---------------------------------+
+ * | |
* | Memory Device Registers |
* | |
* n + PAYLOAD_SIZE_MAX -----------------------------------
@@ -76,13 +87,26 @@
(CXL_MAILBOX_REGISTERS_OFFSET + CXL_MAILBOX_REGISTERS_LENGTH)
#define CXL_MEMORY_DEVICE_REGISTERS_LENGTH 0x8
+#define CXL_NUM_CPMU_INSTANCES 2
+#define CXL_CPMU_SIZE 0x8f8
+#define CXL_CPMU_OFFSET(x) \
+ QEMU_ALIGN_UP(CXL_MEMORY_DEVICE_REGISTERS_OFFSET + \
+ CXL_MEMORY_DEVICE_REGISTERS_LENGTH + \
+ (x) * (1 << 16), \
+ 1 << 16)
+
#define CXL_MMIO_SIZE \
- (CXL_DEVICE_CAP_REG_SIZE + CXL_DEVICE_STATUS_REGISTERS_LENGTH + \
- CXL_MAILBOX_REGISTERS_LENGTH + CXL_MEMORY_DEVICE_REGISTERS_LENGTH)
+ QEMU_ALIGN_UP(CXL_DEVICE_CAP_REG_SIZE + \
+ CXL_DEVICE_STATUS_REGISTERS_LENGTH + \
+ CXL_MAILBOX_REGISTERS_LENGTH + \
+ CXL_MEMORY_DEVICE_REGISTERS_LENGTH + \
+ CXL_NUM_CPMU_INSTANCES * (1 << 16), \
+ (1 << 16))
typedef struct cxl_device_state {
MemoryRegion device_registers;
+ MemoryRegion cpmu_registers[CXL_NUM_CPMU_INSTANCES];
/* mmio for device capabilities array - 8.2.8.2 */
MemoryRegion device;
MemoryRegion memory_device;
@@ -119,6 +143,9 @@ typedef struct cxl_device_state {
/* memory region for persistent memory, HDM */
uint64_t pmem_size;
+
+ /* Move me later */
+ CPMUState cpmu[CXL_NUM_CPMU_INSTANCES];
} CXLDeviceState;
/* Initialize the register block for a device */
diff --git a/include/hw/cxl/cxl_pci.h b/include/hw/cxl/cxl_pci.h
index 8cbeb61142..b7b3620a9f 100644
--- a/include/hw/cxl/cxl_pci.h
+++ b/include/hw/cxl/cxl_pci.h
@@ -35,7 +35,7 @@
#define PCIE_FLEXBUS_PORT_DVSEC_LENGTH_2_0 0x14
#define PCIE_FLEXBUS_PORT_DVSEC_REVID_2_0 1
-#define REG_LOC_DVSEC_LENGTH 0x24
+#define REG_LOC_DVSEC_LENGTH 0x2c
#define REG_LOC_DVSEC_REVID 0
enum {
@@ -144,9 +144,9 @@ typedef struct CXLDVSECRegisterLocator {
struct {
uint32_t lo;
uint32_t hi;
- } reg_base[3];
+ } reg_base[4];
} CXLDVSECRegisterLocator;
-QEMU_BUILD_BUG_ON(sizeof(CXLDVSECRegisterLocator) != 0x24);
+QEMU_BUILD_BUG_ON(sizeof(CXLDVSECRegisterLocator) != 0x2C);
/* BAR Equivalence Indicator */
#define BEI_BAR_10H 0
@@ -161,5 +161,6 @@ QEMU_BUILD_BUG_ON(sizeof(CXLDVSECRegisterLocator) != 0x24);
#define RBI_COMPONENT_REG (1 << 8)
#define RBI_BAR_VIRT_ACL (2 << 8)
#define RBI_CXL_DEVICE_REG (3 << 8)
+#define RBI_CXL_CPMU_REG (4 << 8)
#endif
--
2.32.0
^ permalink raw reply related [flat|nested] 8+ messages in thread
end of thread, other threads:[~2022-08-31 15:39 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-08-31 15:33 [RFC PATCH 0/3] CXL Performance Monitoring Unit QEMU emulation Jonathan Cameron
2022-08-31 15:33 ` Jonathan Cameron via
2022-08-31 15:33 ` [RFC PATCH 1/3] hw/mem/cxl-type3: Add MSI/MSIX support Jonathan Cameron
2022-08-31 15:33 ` Jonathan Cameron via
2022-08-31 15:33 ` [RFC PATCH 2/3] hw/cxl: Switch to using an array for CXLRegisterLocator base addresses Jonathan Cameron
2022-08-31 15:33 ` Jonathan Cameron via
2022-08-31 15:33 ` [RFC PATCH 3/3] hw/cxl: CXL Performance Monitoring Unit (CPMU) Emulation Jonathan Cameron
2022-08-31 15:33 ` Jonathan Cameron via
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.