x86: fix cpu hotplug with secure boot
v5 cover x86: fix cpu hotplug with secure boot
v5:
- fix hotplug on Windows when there is more than 256 possible CPUs (Windows isn't able to handle VarPackage over 255 elements
so process CPUs in batches)
- fix off-by-one in package length (Laszlo)
- fix not selecting CPU before clearing insert event (Laszlo)
- use aml_lgreater() instead of aml_lnot(aml_equal(num_added_cpus, zero) (Laszlo)
- split 'x68: acpi: trigger SMI before sending hotplug Notify event to OSPM' in samller chunks (Laszlo)
- fix comment to match spec (Laszlo)
- reorder aml_lor() and aml_land() in header (Laszlo)
v4:
- fix 5.2 machine types so they won't apply pc_compat_5_1 (Laszlo)
v3:
- rebase on top of "[PATCH v2] hw: add compat machines for 5.2"
so apply that before this patch
v2:
- AML: clean is_inserted flag only after SMI callback
- make x-smi-cpu-hotunplug false by default
- massage error hint on not supported unplug
v1:
- fix typos and some phrases (Laszlo)
- add unplug check (Laszlo)
- redo AML scan logic to avoid race when adding multiple CPUs
CPU hotplug with Secure Boot was not really supported and firmware wasn't aware of hotplugged CPUs (which might lead to guest crashes). During 4.2 we introduced locked SMI handler RAM arrea to make sure that guest OS wasn't able to inject its own SMI handler and OVMF added initial CPU hotplug support.
This series is QEMU part of that support which lets QMVF tell QEMU that CPU hotplug with SMI broadcast enabled is supported so that QEMU would be able to prevent hotplug in case it's not supported and trigger SMI on hotplug when it's necessary.
Igor Mammedov (10):
x86: lpc9: let firmware negotiate 'CPU hotplug with SMI' features
x86: cpuhp: prevent guest crash on CPU hotplug when broadcast SMI is
in use
x86: cpuhp: refuse cpu hot-unplug request earlier if not supported
acpi: add aml_land() and aml_break() primitives
tests: acpi: mark to be changed tables in
bios-tables-test-allowed-diff
x86: ich9: expose "smi_negotiated_features" as a QOM property
x86: acpi: introduce AcpiPmInfo::smi_on_cpuhp
x86: acpi: introduce the PCI0.SMI0 ACPI device
x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
tests: acpi: update acpi blobs with new AML
include/hw/acpi/aml-build.h | 2 + include/hw/acpi/cpu.h | 1 + include/hw/i386/ich9.h | 4 + hw/acpi/aml-build.c | 16 +++ hw/acpi/cpu.c | 156 ++++++++++++++++++++++++------ hw/acpi/ich9.c | 24 ++++- hw/i386/acpi-build.c | 35 ++++++- hw/i386/pc.c | 15 ++- hw/isa/lpc_ich9.c | 16 +++ tests/data/acpi/pc/DSDT | Bin 4934 -> 5060 bytes tests/data/acpi/pc/DSDT.acpihmat | Bin 6258 -> 6385 bytes tests/data/acpi/pc/DSDT.bridge | Bin 6793 -> 6919 bytes tests/data/acpi/pc/DSDT.cphp | Bin 5397 -> 5524 bytes tests/data/acpi/pc/DSDT.dimmpxm | Bin 6587 -> 6714 bytes tests/data/acpi/pc/DSDT.ipmikcs | Bin 5006 -> 5132 bytes tests/data/acpi/pc/DSDT.memhp | Bin 6293 -> 6419 bytes tests/data/acpi/pc/DSDT.numamem | Bin 4940 -> 5066 bytes tests/data/acpi/q35/DSDT | Bin 7678 -> 7804 bytes tests/data/acpi/q35/DSDT.acpihmat | Bin 9002 -> 9129 bytes tests/data/acpi/q35/DSDT.bridge | Bin 7695 -> 7821 bytes tests/data/acpi/q35/DSDT.cphp | Bin 8141 -> 8268 bytes tests/data/acpi/q35/DSDT.dimmpxm | Bin 9331 -> 9458 bytes tests/data/acpi/q35/DSDT.ipmibt | Bin 7753 -> 7879 bytes tests/data/acpi/q35/DSDT.memhp | Bin 9037 -> 9163 bytes tests/data/acpi/q35/DSDT.mmio64 | Bin 8808 -> 8934 bytes tests/data/acpi/q35/DSDT.numamem | Bin 7684 -> 7810 bytes tests/data/acpi/q35/DSDT.tis | Bin 8283 -> 8409 bytes 27 files changed, 239 insertions(+), 30 deletions(-)
Signature · 1 line
2.27.0
PATCH v5 1/10 x86: lpc9: let firmware negotiate 'CPU hotplug with SMI' features
It will allow firmware to notify QEMU that firmware requires SMI being triggered on CPU hot[un]plug, so that it would be able to account for hotplugged CPU and relocate it to new SMM base and/or safely remove CPU on unplug.
Using negotiated features, follow up patches will insert SMI upcall into AML code, to make sure that firmware processes hotplug before guest OS would attempt to use new CPU.
---
v4:
- fix 5.2 machine types so they won't apply pc_compat_5_1
(Laszlo Ersek <[email protected]>)
v3:
- rebase on top of "[PATCH v2] hw: add compat machines for 5.2"
so apply that before this patch
v2:
- rebase on top of 5.1 (move compat values to 5.1 machine)
- make "x-smi-cpu-hotunplug" false by default (Laszlo Ersek <[email protected]>)
fixup
--- include/hw/i386/ich9.h | 2 ++ hw/i386/pc.c | 4 +++- hw/isa/lpc_ich9.c | 13 +++++++++++++ 3 files changed, 18 insertions(+), 1 deletion(-)
include/hw/i386/ich9.h +2 -0
@@ -247,5 +247,7 @@ typedef struct ICH9LPCState {/* bit positions used in fw_cfg SMI feature negotiation */ #define ICH9_LPC_SMI_F_BROADCAST_BIT 0 +#define ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT 1 +#define ICH9_LPC_SMI_F_CPU_HOT_UNPLUG_BIT 2#endif /* HW_ICH9_H */
hw/i386/pc.c +3 -1
@@ -97,7 +97,9 @@#include "trace.h" #include CONFIG_DEVICES -GlobalProperty pc_compat_5_1[] = {}; +GlobalProperty pc_compat_5_1[] = { + {"ICH9-LPC" ,"x-smi-cpu-hotplug" ,"off" }, +};const size_t pc_compat_5_1_len =G_N_ELEMENTS (pc_compat_5_1);GlobalProperty pc_compat_5_0[] = {
hw/isa/lpc_ich9.c +13 -0
@@ -373,6 +373,15 @@ static void smi_features_ok_callback(void *opaque)/* guest requests invalid features, leave @features_ok at zero */ return ; } +if (!(guest_features &BIT_ULL (ICH9_LPC_SMI_F_BROADCAST_BIT )) && + guest_features & (BIT_ULL (ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT ) | +BIT_ULL (ICH9_LPC_SMI_F_CPU_HOT_UNPLUG_BIT ))) { +/* +* cpu hot-[un]plug with SMI requires SMI broadcast, +* leave @features_ok at zero +*/ +return ; + }/* valid feature subset requested, lock it down, report success */ lpc->smi_negotiated_features = guest_features;
@@ -747,6 +756,10 @@ static Property ich9_lpc_properties[] = {DEFINE_PROP_BOOL ("noreboot" , ICH9LPCState, pin_strap.spkr_hi, true),DEFINE_PROP_BIT64 ("x-smi-broadcast" , ICH9LPCState, smi_host_features,ICH9_LPC_SMI_F_BROADCAST_BIT , true), +DEFINE_PROP_BIT64 ("x-smi-cpu-hotplug" , ICH9LPCState, smi_host_features, +ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT , true), +DEFINE_PROP_BIT64 ("x-smi-cpu-hotunplug" , ICH9LPCState, smi_host_features, +ICH9_LPC_SMI_F_CPU_HOT_UNPLUG_BIT , false), DEFINE_PROP_END_OF_LIST(), };
Signature · 1 line
2.27.0
PATCH v5 2/10 x86: cpuhp: prevent guest crash on CPU hotplug when broadcast SMI is in use
In reply to Igor MammedovThere were reports of guest crash on CPU hotplug, when using q35 machine type and OVMF with SMM, due to hotplugged CPU trying to process SMI at default SMI handler location without it being relocated by firmware first.
Fix it by refusing hotplug if firmware hasn't negotiated CPU hotplug with SMI support while SMI broadcast is in use.
---
v1:
fix typos an use suggested wording in commit and error msg s/secure boot/smm/; s/hotplug SMI/hotplug with SMI/
(Laszlo Ersek <[email protected]>)
--- hw/acpi/ich9.c | 12 +++++++++++- hw/i386/pc.c | 11 +++++++++++ 2 files changed, 22 insertions(+), 1 deletion(-)
hw/acpi/ich9.c +11 -1
@@ -408,10 +408,20 @@ void ich9_pm_device_pre_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,ICH9LPCState *lpc =ICH9_LPC_DEVICE (hotplug_dev);if (object_dynamic_cast (OBJECT (dev),TYPE_PC_DIMM ) && - !lpc->pm.acpi_memory_hotplug.is_enabled) + !lpc->pm.acpi_memory_hotplug.is_enabled) {error_setg (errp,"memory hotplug is not enabled: %s.memory-hotplug-support " "is not set" ,object_get_typename (OBJECT (lpc))); + }else if (object_dynamic_cast (OBJECT (dev),TYPE_CPU )) { +uint64_t negotiated = lpc->smi_negotiated_features; + +if (negotiated &BIT_ULL (ICH9_LPC_SMI_F_BROADCAST_BIT ) && + !(negotiated &BIT_ULL (ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT ))) { +error_setg (errp,"cpu hotplug with SMI wasn't enabled by firmware" ); +error_append_hint (errp,"update machine type to newer than 5.1 " +"and firmware that suppors CPU hotplug with SMM" ); + } + } } void ich9_pm_device_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
hw/i386/pc.c +11 -0
@@ -1500,6 +1500,17 @@ static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev,return ; } +if (pcms->acpi_dev) { +Error *local_err =NULL ; + +hotplug_handler_pre_plug (HOTPLUG_HANDLER (pcms->acpi_dev), dev, + &local_err); +if (local_err) { +error_propagate (errp, local_err); +return ; + } + } +init_topo_info (&topo_info, x86ms); env->nr_dies = x86ms->smp_dies;
Signature · 1 line
2.27.0
PATCH v5 3/10 x86: cpuhp: refuse cpu hot-unplug request earlier if not supported
In reply to Igor MammedovCPU hot-unplug with SMM requires firmware participation to prevent guest crash (i.e. CPU can be removed only after OS _and_ firmware were prepared for the action).
Previous patches introduced ICH9_LPC_SMI_F_CPU_HOT_UNPLUG_BIT feature bit, which is advertised by firmware when it has support for CPU hot-unplug. Use it to check if guest is able to handle unplug and make device_del fail gracefully if hot-unplug feature hasn't been negotiated.
---
v2:
- fix typo in commit message
- drop 5.1 version from hint message (Laszlo)
--- hw/acpi/ich9.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
hw/acpi/ich9.c +12 -0
@@ -460,6 +460,18 @@ void ich9_pm_device_unplug_request_cb(HotplugHandler *hotplug_dev, errp); }else if (object_dynamic_cast (OBJECT (dev),TYPE_CPU ) && !lpc->pm.cpu_hotplug_legacy) { +uint64_t negotiated = lpc->smi_negotiated_features; + +if (negotiated &BIT_ULL (ICH9_LPC_SMI_F_BROADCAST_BIT ) && + !(negotiated &BIT_ULL (ICH9_LPC_SMI_F_CPU_HOT_UNPLUG_BIT ))) { +error_setg (errp,"cpu hot-unplug with SMI wasn't enabled " +"by firmware" ); +error_append_hint (errp,"update machine type to a version having " +"x-smi-cpu-hotunplug=on and firmware that " +"supports CPU hot-unplug with SMM" ); +return ; + } +acpi_cpu_unplug_request_cb (hotplug_dev, &lpc->pm.cpuhp_state, dev, errp); }else {
Signature · 1 line
2.27.0
PATCH v5 4/10 acpi: add aml_land() and aml_break() primitives
In reply to Igor Mammedov---
v5:
- fix comment to match spec (Laszlo Ersek <[email protected]>)
- reorder aml_lor() and aml_land() in header (Laszlo Ersek <[email protected]>)
--- include/hw/acpi/aml-build.h | 2 ++ hw/acpi/aml-build.c | 16 ++++++++++++++++ 2 files changed, 18 insertions(+)
include/hw/acpi/aml-build.h +2 -0
@@ -290,6 +290,7 @@ Aml *aml_to_buffer(Aml *src, Aml *dst);Aml *aml_store (Aml *val,Aml *target);Aml *aml_and (Aml *arg1,Aml *arg2,Aml *dst);Aml *aml_or (Aml *arg1,Aml *arg2,Aml *dst); +Aml *aml_land (Aml *arg1,Aml *arg2);Aml *aml_lor (Aml *arg1,Aml *arg2);Aml *aml_shiftleft (Aml *arg1,Aml *count);Aml *aml_shiftright (Aml *arg1,Aml *count,Aml *dst);
@@ -300,6 +301,7 @@ Aml *aml_increment(Aml *arg);Aml *aml_decrement (Aml *arg);Aml *aml_index (Aml *arg1,Aml *idx);Aml *aml_notify (Aml *arg1,Aml *arg2); +Aml *aml_break (void );Aml *aml_call0 (const char *method);Aml *aml_call1 (const char *method,Aml *arg1);Aml *aml_call2 (const char *method,Aml *arg1,Aml *arg2);
hw/acpi/aml-build.c +16 -0
@@ -556,6 +556,15 @@ Aml *aml_or(Aml *arg1, Aml *arg2, Aml *dst)return build_opcode_2arg_dst (0x7D /* OrOp */ , arg1, arg2, dst); } +/* ACPI 1.0b: 16.2.5.4 Type 2 Opcodes Encoding: DefLAnd */ +Aml *aml_land (Aml *arg1,Aml *arg2) +{ +Aml *var =aml_opcode (0x90 /* LAndOp */ ); +aml_append (var, arg1); +aml_append (var, arg2); +return var; +} +/* ACPI 1.0b: 16.2.5.4 Type 2 Opcodes Encoding: DefLOr */ Aml *aml_lor (Aml *arg1,Aml *arg2) {
@@ -629,6 +638,13 @@ Aml *aml_notify(Aml *arg1, Aml *arg2)return var; } +/* ACPI 1.0b: 16.2.5.3 Type 1 Opcodes Encoding: DefBreak */ +Aml *aml_break (void ) +{ +Aml *var =aml_opcode (0xa5 /* BreakOp */ ); +return var; +} +/* helper to call method without argument */ Aml *aml_call0 (const char *method) {
Signature · 1 line
2.27.0
PATCH v5 5/10 tests: acpi: mark to be changed tables in bios-tables-test-allowed-diff
In reply to Igor Mammedov... to let tests pass until binary blobs are updated with new AML
--- tests/qtest/bios-tables-test-allowed-diff.h | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+)
tests/qtest/bios-tables-test-allowed-diff.h +19 -0
@@ -1 +1,20 @@/* List of comma-separated changed AML files to ignore */ +"tests/data/acpi/pc/DSDT" , +"tests/data/acpi/q35/DSDT" , +"tests/data/acpi/q35/DSDT.tis" , +"tests/data/acpi/q35/DSDT.bridge" , +"tests/data/acpi/q35/DSDT.mmio64" , +"tests/data/acpi/q35/DSDT.ipmibt" , +"tests/data/acpi/q35/DSDT.cphp" , +"tests/data/acpi/q35/DSDT.memhp" , +"tests/data/acpi/q35/DSDT.numamem" , +"tests/data/acpi/q35/DSDT.dimmpxm" , +"tests/data/acpi/q35/DSDT.acpihmat" , +"tests/data/acpi/pc/DSDT.bridge" , +"tests/data/acpi/pc/DSDT.ipmikcs" , +"tests/data/acpi/pc/DSDT.cphp" , +"tests/data/acpi/pc/DSDT.memhp" , +"tests/data/acpi/pc/DSDT.numamem" , +"tests/data/acpi/pc/DSDT.dimmpxm" , +"tests/data/acpi/pc/DSDT.acpihmat" , +"tests/data/acpi/pc/DSDT.acpihmat",
Signature · 1 line
2.27.0
PATCH v5 6/10 x86: ich9: expose "smi_negotiated_features" as a QOM property
In reply to Igor MammedovExpose the "smi_negotiated_features" field of ICH9LPCState as a QOM property.
--- include/hw/i386/ich9.h | 2 ++ hw/isa/lpc_ich9.c | 3 +++ 2 files changed, 5 insertions(+)
include/hw/i386/ich9.h +2 -0
@@ -245,6 +245,8 @@ typedef struct ICH9LPCState {#define ICH9_SMB_HST_D1 0x06#define ICH9_SMB_HOST_BLOCK_DB 0x07 +#define ICH9_LPC_SMI_NEGOTIATED_FEAT_PROP "x-smi-negotiated-features" +/* bit positions used in fw_cfg SMI feature negotiation */ #define ICH9_LPC_SMI_F_BROADCAST_BIT 0#define ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT 1
hw/isa/lpc_ich9.c +3 -0
@@ -647,6 +647,9 @@ static void ich9_lpc_initfn(Object *obj) &acpi_enable_cmd, OBJ_PROP_FLAG_READ);object_property_add_uint8_ptr (OBJECT (lpc),ACPI_PM_PROP_ACPI_DISABLE_CMD , &acpi_disable_cmd,OBJ_PROP_FLAG_READ ); +object_property_add_uint64_ptr (obj,ICH9_LPC_SMI_NEGOTIATED_FEAT_PROP , + &lpc->smi_negotiated_features, +OBJ_PROP_FLAG_READ );ich9_pm_add_properties (obj, &lpc->pm); }
Signature · 1 line
2.27.0
Re: [PATCH v5 06/10] x86: ich9: expose "smi_negotiated_features" as a QOM property
Quoted 29 lines · Igor Mammedov
Expose the "
smi_negotiated_features" field of ICH9LPCState as a QOM property.Signed-off-by Igor Mammedov <[email protected]>Reviewed-by Laszlo Ersek <[email protected]>--- include/hw/i386/ich9.h | 2 ++ hw/isa/lpc_ich9.c | 3 +++ 2 files changed, 5 insertions(+)include/hw/i386/ich9.h +2
-0@@ -245,6 +245,8 @@ typedef struct ICH9LPCState {#define ICH9_SMB_HST_D1 0x06#define ICH9_SMB_HOST_BLOCK_DB 0x07 +#define ICH9_LPC_SMI_NEGOTIATED_FEAT_PROP "x-smi-negotiated-features" +/* bit positions used in fw_cfg SMI feature negotiation */ #define ICH9_LPC_SMI_F_BROADCAST_BIT 0#define ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT 1hw/isa/lpc_ich9.c +0
-0
hw/isa/lpc_ich9.c static void ich9_lpc_initfn(Object *obj) · line 655
Quoted 9 lines
hw/isa/lpc_ich9.c +3
-0@@ -647,6 +647,9 @@ static void ich9_lpc_initfn(Object *obj) &acpi_enable_cmd, OBJ_PROP_FLAG_READ);object_property_add_uint8_ptr (OBJECT (lpc),ACPI_PM_PROP_ACPI_DISABLE_CMD , &acpi_disable_cmd,OBJ_PROP_FLAG_READ ); +object_property_add_uint64_ptr (obj,ICH9_LPC_SMI_NEGOTIATED_FEAT_PROP , + &lpc->smi_negotiated_features, +OBJ_PROP_FLAG_READ );ich9_pm_add_properties (obj, &lpc->pm); }
PATCH v5 7/10 x86: acpi: introduce AcpiPmInfo::smi_on_cpuhp
In reply to Igor MammedovTranslate the "CPU hotplug with SMI" feature bit, from the property added in the last patch, to a dedicated boolean in AcpiPmInfo.
--- hw/i386/acpi-build.c | 6 ++++++ 1 file changed, 6 insertions(+)
hw/i386/acpi-build.c +6 -0
@@ -95,6 +95,7 @@ typedef struct AcpiPmInfo {bool s3_disabled;bool s4_disabled;bool pcihp_bridge_en; +bool smi_on_cpuhp;uint8_t s4_val;AcpiFadtData fadt;uint16_t cpu_hp_io_base;
@@ -194,6 +195,7 @@ static void acpi_get_pm_info(MachineState *machine, AcpiPmInfo *pm) pm->cpu_hp_io_base =0 ; pm->pcihp_io_base =0 ; pm->pcihp_io_len =0 ; + pm->smi_on_cpuhp = false;assert (obj);init_common_fadt_data (machine, obj, &pm->fadt);
@@ -207,12 +209,16 @@ static void acpi_get_pm_info(MachineState *machine, AcpiPmInfo *pm)object_property_get_uint (obj,ACPI_PCIHP_IO_LEN_PROP ,NULL ); }if (lpc) { +uint64_t smi_features =object_property_get_uint (lpc, +ICH9_LPC_SMI_NEGOTIATED_FEAT_PROP ,NULL );struct AcpiGenericAddress r = { .space_id =AML_AS_SYSTEM_IO , .bit_width =8 , .address =ICH9_RST_CNT_IOPORT }; pm->fadt.reset_reg = r; pm->fadt.reset_val =0xf ; pm->fadt.flags |=1 <<ACPI_FADT_F_RESET_REG_SUP ; pm->cpu_hp_io_base =ICH9_CPU_HOTPLUG_IO_BASE ; + pm->smi_on_cpuhp = + !!(smi_features &BIT_ULL (ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT )); }/* The above need not be conditional on machine type because the reset port
Signature · 1 line
2.27.0
Re: [PATCH v5 07/10] x86: acpi: introduce AcpiPmInfo::smi_on_cpuhp
Quoted 37 lines · Igor Mammedov
Translate the "CPU hotplug with SMI" feature bit, from the property added in the last patch, to a dedicated boolean in AcpiPmInfo.
Signed-off-by Igor Mammedov <[email protected]>Reviewed-by Laszlo Ersek <[email protected]>--- hw/i386/acpi-build.c | 6 ++++++ 1 file changed, 6 insertions(+)hw/i386/acpi-build.c +4
-0@@ -95,6 +95,7 @@ typedef struct AcpiPmInfo {bool s3_disabled;bool s4_disabled;bool pcihp_bridge_en; +bool smi_on_cpuhp;uint8_t s4_val;AcpiFadtData fadt;uint16_t cpu_hp_io_base;@@ -194,6 +195,7 @@ static void acpi_get_pm_info(MachineState *machine, AcpiPmInfo *pm) pm->cpu_hp_io_base =0 ; pm->pcihp_io_base =0 ; pm->pcihp_io_len =0 ; + pm->smi_on_cpuhp = false;assert (obj);init_common_fadt_data (machine, obj, &pm->fadt);@@ -207,12 +209,16 @@ static void acpi_get_pm_info(MachineState *machine, AcpiPmInfo *pm)object_property_get_uint (obj,ACPI_PCIHP_IO_LEN_PROP ,NULL ); }if (lpc) { +uint64_t smi_features =object_property_get_uint (lpc, +ICH9_LPC_SMI_NEGOTIATED_FEAT_PROP ,NULL );struct AcpiGenericAddress r = { .space_id =AML_AS_SYSTEM_IO ,
hw/i386/acpi-build.c static void acpi_get_pm_info(MachineState *machine, AcpiPmInfo *pm) · line 224
Quoted 10 lines
hw/i386/acpi-build.c +2
-0.bit_width = 8, .address = ICH9_RST_CNT_IOPORT }; pm->fadt.reset_reg = r; pm->fadt.reset_val =0xf ; pm->fadt.flags |=1 <<ACPI_FADT_F_RESET_REG_SUP ; pm->cpu_hp_io_base =ICH9_CPU_HOTPLUG_IO_BASE ; + pm->smi_on_cpuhp = + !!(smi_features &BIT_ULL (ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT )); }/* The above need not be conditional on machine type because the reset port
PATCH v5 8/10 x86: acpi: introduce the PCI0.SMI0 ACPI device
In reply to Igor MammedovWhen CPU hotplug with SMI has been negotiated, describe the SMI register block in the DSDT. Pass the ACPI name of the SMI control register to build_cpus_aml(), so that CPU_SCAN_METHOD can access the register in the next patch.
--- include/hw/acpi/cpu.h | 1 + hw/i386/acpi-build.c | 29 ++++++++++++++++++++++++++++- 2 files changed, 29 insertions(+), 1 deletion(-)
include/hw/acpi/cpu.h +1 -0
@@ -50,6 +50,7 @@ void cpu_hotplug_hw_init(MemoryRegion *as, Object *owner,typedef struct CPUHotplugFeatures {bool acpi_1_compatible;bool has_legacy_cphp; +const char *smi_path; }CPUHotplugFeatures ; void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts,
hw/i386/acpi-build.c +28 -1
@@ -1521,6 +1521,32 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,aml_append (dev,aml_name_decl ("_UID" ,aml_int (0 )));aml_append (dev,build_q35_osc_method ());aml_append (sb_scope, dev); + +if (pm->smi_on_cpuhp) { +/* reserve SMI block resources, IO ports 0xB2, 0xB3 */ + dev =aml_device ("PCI0.SMI0" ); +aml_append (dev,aml_name_decl ("_HID" ,aml_eisaid ("PNP0A06" ))); +aml_append (dev,aml_name_decl ("_UID" ,aml_string ("SMI resources" ))); + crs =aml_resource_template (); +aml_append (crs, +aml_io ( +AML_DECODE16 , +ACPI_PORT_SMI_CMD , +ACPI_PORT_SMI_CMD , +1 , +2 ) + ); +aml_append (dev,aml_name_decl ("_CRS" , crs)); +aml_append (dev,aml_operation_region ("SMIR" ,AML_SYSTEM_IO , +aml_int (ACPI_PORT_SMI_CMD ),2 )); + field =aml_field ("SMIR" ,AML_BYTE_ACC ,AML_NOLOCK , +AML_WRITE_AS_ZEROS ); +aml_append (field,aml_named_field ("SMIC" ,8 )); +aml_append (field,aml_reserved_field (8 )); +aml_append (dev, field); +aml_append (sb_scope, dev); + } +aml_append (dsdt, sb_scope);build_hpet_aml (dsdt);
@@ -1536,7 +1562,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,build_legacy_cpu_hotplug_aml (dsdt, machine, pm->cpu_hp_io_base); }else {CPUHotplugFeatures opts = { - .acpi_1_compatible = true, .has_legacy_cphp = true + .acpi_1_compatible = true, .has_legacy_cphp = true, + .smi_path = pm->smi_on_cpuhp ?"\\_SB.PCI0.SMI0.SMIC" :NULL , };build_cpus_aml (dsdt, machine, opts, pm->cpu_hp_io_base,"\\_SB.PCI0" ,"\\_GPE._E02" );
Signature · 1 line
2.27.0
Re: [PATCH v5 08/10] x86: acpi: introduce the PCI0.SMI0 ACPI device
Quoted 63 lines · Igor Mammedov
When CPU hotplug with SMI has been negotiated, describe the SMI register block in the DSDT. Pass the ACPI name of the SMI control register to
build_cpus_aml(), so thatCPU_SCAN_METHODcan access the register in the next patch.Signed-off-by Igor Mammedov <[email protected]>Reviewed-by Laszlo Ersek <[email protected]>--- include/hw/acpi/cpu.h | 1 + hw/i386/acpi-build.c | 29 ++++++++++++++++++++++++++++- 2 files changed, 29 insertions(+), 1 deletion(-)include/hw/acpi/cpu.h +1
-0@@ -50,6 +50,7 @@ void cpu_hotplug_hw_init(MemoryRegion *as, Object *owner,typedef struct CPUHotplugFeatures {bool acpi_1_compatible;bool has_legacy_cphp; +const char *smi_path; }CPUHotplugFeatures ; void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts,hw/i386/acpi-build.c +26
-0@@ -1521,6 +1521,32 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,aml_append (dev,aml_name_decl ("_UID" ,aml_int (0 )));aml_append (dev,build_q35_osc_method ());aml_append (sb_scope, dev); + +if (pm->smi_on_cpuhp) { +/* reserve SMI block resources, IO ports 0xB2, 0xB3 */ + dev =aml_device ("PCI0.SMI0" ); +aml_append (dev,aml_name_decl ("_HID" ,aml_eisaid ("PNP0A06" ))); +aml_append (dev,aml_name_decl ("_UID" ,aml_string ("SMI resources" ))); + crs =aml_resource_template (); +aml_append (crs, +aml_io ( +AML_DECODE16 , +ACPI_PORT_SMI_CMD , +ACPI_PORT_SMI_CMD , +1 , +2 ) + ); +aml_append (dev,aml_name_decl ("_CRS" , crs)); +aml_append (dev,aml_operation_region ("SMIR" ,AML_SYSTEM_IO , +aml_int (ACPI_PORT_SMI_CMD ),2 )); + field =aml_field ("SMIR" ,AML_BYTE_ACC ,AML_NOLOCK , +AML_WRITE_AS_ZEROS ); +aml_append (field,aml_named_field ("SMIC" ,8 )); +aml_append (field,aml_reserved_field (8 )); +aml_append (dev, field); +aml_append (sb_scope, dev); + } +aml_append (dsdt, sb_scope);build_hpet_aml (dsdt);
hw/i386/acpi-build.c build_dsdt(GArray *table_data, BIOSLinker *linker, · line 1569
Quoted 9 lines
hw/i386/acpi-build.c +2
-1@@ -1536,7 +1562,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,build_legacy_cpu_hotplug_aml (dsdt, machine, pm->cpu_hp_io_base); }else {CPUHotplugFeatures opts = { - .acpi_1_compatible = true, .has_legacy_cphp = true + .acpi_1_compatible = true, .has_legacy_cphp = true, + .smi_path = pm->smi_on_cpuhp ?"\\_SB.PCI0.SMI0.SMIC" :NULL , };build_cpus_aml (dsdt, machine, opts, pm->cpu_hp_io_base,"\\_SB.PCI0" ,"\\_GPE._E02" );
PATCH v5 9/10 x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
In reply to Igor MammedovIn case firmware has negotiated CPU hotplug SMI feature, generate AML to describe SMI IO port region and send SMI to firmware on each CPU hotplug SCI in case new CPUs were hotplugged.
Since new CPUs can be hotplugged while CPU_SCAN_METHOD is running we can't send SMI before new CPUs are fetched from QEMU as it could cause sending Notify to a CPU that firmware hasn't seen yet. So fetch new CPUs into local cache first, then send SMI and after that send Notify events to cached CPUs. This should ensure that Notify is sent only to CPUs which were processed by firmware first.
Any CPUs that were hotplugged after caching will be processed by the next CPU_SCAN_METHOD, when pending SCI is handled.
---
v5:
- fix hotplug on Windows when there is more than 256 possible CPUs (Windows isn't able to handle VarPackage over 255 elements
so process CPUs in batches)
- (Laszlo Ersek <[email protected]>)
- fix off-by-one in package length
- fix not selecting CPU before clearing insert event
- use aml_lgreater() instead of aml_lnot(aml_equal(num_added_cpus, zero)
- split off in separate patches:
- making smi_negotiated_features a property
- introduction of AcpiPmInfo.smi_on_cpuhp
- introduction of PCI0.SMI0 ACPI device
v2:
- clear insert event after firmware has returned
control from SMI. (Laszlo Ersek <[email protected]>)
v1:
- make sure that Notify is sent only to CPUs seen by FW
- (Laszlo Ersek <[email protected]>)
- use macro instead of x-smi-negotiated-features
- style fixes
- reserve whole SMI block (0xB2, 0xB3)
v0:
- s/aml_string/aml_eisaid/
/fixes Windows BSOD, on nonsense value/ (Laszlo Ersek <[email protected]>)
- put SMI device under PCI0 like the rest of hotplug IO port
- do not generate SMI AML if CPU hotplug SMI feature hasn't been negotiated
--- hw/acpi/cpu.c | 156 +++++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 129 insertions(+), 27 deletions(-)
hw/acpi/cpu.c +129 -27
@@ -14,6 +14,8 @@#define ACPI_CPU_CMD_DATA_OFFSET_RW 8#define ACPI_CPU_CMD_DATA2_OFFSET_R 0 +#define OVMF_CPUHP_SMI_CMD 4 +enum {CPHP_GET_NEXT_CPU_WITH_EVENT_CMD =0 ,CPHP_OST_EVENT_CMD =1 ,
@@ -321,6 +323,7 @@ const VMStateDescription vmstate_cpu_hotplug = {#define CPU_NOTIFY_METHOD "CTFY"#define CPU_EJECT_METHOD "CEJ0"#define CPU_OST_METHOD "COST" +#define CPU_ADDED_LIST "CNEW"#define CPU_ENABLED "CPEN"#define CPU_SELECTOR "CSEL"
@@ -465,42 +468,141 @@ void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts, method =aml_method (CPU_SCAN_METHOD ,0 ,AML_SERIALIZED ); { +const uint8_t max_cpus_per_pass =255 ;Aml *else_ctx; -Aml *while_ctx; +Aml *while_ctx, *while_ctx2;Aml *has_event =aml_local (0 );Aml *dev_chk =aml_int (1 );Aml *eject_req =aml_int (3 );Aml *next_cpu_cmd =aml_int (CPHP_GET_NEXT_CPU_WITH_EVENT_CMD ); +Aml *num_added_cpus =aml_local (1 ); +Aml *cpu_idx =aml_local (2 ); +Aml *uid =aml_local (3 ); +Aml *has_job =aml_local (4 ); +Aml *new_cpus =aml_name (CPU_ADDED_LIST );aml_append (method,aml_acquire (ctrl_lock,0xFFFF )); -aml_append (method,aml_store (one, has_event)); - while_ctx =aml_while (aml_equal (has_event, one)); + +/* +* Windows versions newer than XP (including Windows 10/Windows +* Server 2019), do support* VarPackageOp but, it is cripled to hold +* the same elements number as old PackageOp. +* For compatibility with Windows XP (so it won't crash) use ACPI1.0 +* PackageOp which can hold max 255 elements. +* +* use named package as old Windows don't support it in local var +*/ +aml_append (method,aml_name_decl (CPU_ADDED_LIST , +aml_package (max_cpus_per_pass))); + +aml_append (method,aml_store (zero, uid)); +aml_append (method,aml_store (one, has_job)); +/* +* CPU_ADDED_LIST can hold limited number of elements, outer loop +* allows to process CPUs in batches which let us to handle more +* CPUs than CPU_ADDED_LIST can hold. +*/ + while_ctx2 =aml_while (aml_equal (has_job, one)); { -/* clear loop exit condition, ins_evt/rm_evt checks -* will set it to 1 while next_cpu_cmd returns a CPU -* with events */ -aml_append (while_ctx,aml_store (zero, has_event)); -aml_append (while_ctx,aml_store (next_cpu_cmd, cpu_cmd)); - ifctx =aml_if (aml_equal (ins_evt, one)); - { -aml_append (ifctx, -aml_call2 (CPU_NOTIFY_METHOD , cpu_data, dev_chk)); -aml_append (ifctx,aml_store (one, ins_evt)); -aml_append (ifctx,aml_store (one, has_event)); - } -aml_append (while_ctx, ifctx); - else_ctx =aml_else (); - ifctx =aml_if (aml_equal (rm_evt, one)); - { -aml_append (ifctx, -aml_call2 (CPU_NOTIFY_METHOD , cpu_data, eject_req)); -aml_append (ifctx,aml_store (one, rm_evt)); -aml_append (ifctx,aml_store (one, has_event)); - } -aml_append (else_ctx, ifctx); -aml_append (while_ctx, else_ctx); +aml_append (while_ctx2,aml_store (zero, has_job)); + +aml_append (while_ctx2,aml_store (one, has_event)); +aml_append (while_ctx2,aml_store (zero, num_added_cpus)); + +/* +* Scan CPUs, till there are CPUs with events or +* CPU_ADDED_LIST capacity is exhausted +*/ + while_ctx =aml_while (aml_land (aml_equal (has_event, one), +aml_lless (uid,aml_int (arch_ids->len)))); + { +/* +* clear loop exit condition, ins_evt/rm_evt checks will +* set it to 1 while next_cpu_cmd returns a CPU with events +*/ +aml_append (while_ctx,aml_store (zero, has_event)); + +aml_append (while_ctx,aml_store (uid, cpu_selector)); +aml_append (while_ctx,aml_store (next_cpu_cmd, cpu_cmd)); + +/* +* wrap around case, scan is complete, exit loop. +* It happens since events are not cleared in scan loop, +* so next_cpu_cmd continues to find already processed CPUs +*/ + ifctx =aml_if (aml_lless (cpu_data, uid)); + { +aml_append (ifctx,aml_break ()); + } +aml_append (while_ctx, ifctx); + +/* +* if CPU_ADDED_LIST is full, exit inner loop and process +* collected CPUs +*/ + ifctx =aml_if ( +aml_equal (num_added_cpus,aml_int (max_cpus_per_pass))); + { +aml_append (ifctx,aml_store (one, has_job)); +aml_append (ifctx,aml_break ()); + } +aml_append (while_ctx, ifctx); + +aml_append (while_ctx,aml_store (cpu_data, uid)); + ifctx =aml_if (aml_equal (ins_evt, one)); + { +/* cache added CPUs to Notify/Wakeup later */ +aml_append (ifctx,aml_store (uid, +aml_index (new_cpus, num_added_cpus))); +aml_append (ifctx,aml_increment (num_added_cpus)); +aml_append (ifctx,aml_store (one, has_event)); + } +aml_append (while_ctx, ifctx); + else_ctx =aml_else (); + ifctx =aml_if (aml_equal (rm_evt, one)); + { +aml_append (ifctx, +aml_call2 (CPU_NOTIFY_METHOD , uid, eject_req)); +aml_append (ifctx,aml_store (one, rm_evt)); +aml_append (ifctx,aml_store (one, has_event)); + } +aml_append (else_ctx, ifctx); +aml_append (while_ctx, else_ctx); +aml_append (while_ctx,aml_increment (uid)); + } +aml_append (while_ctx2, while_ctx); + +/* +* in case FW negotiated ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT, +* make upcall to FW, so it can pull in new CPUs before +* OS is notified and wakes them up +*/ +if (opts.smi_path) { + ifctx =aml_if (aml_lgreater (num_added_cpus, zero)); + { +aml_append (ifctx,aml_store (aml_int (OVMF_CPUHP_SMI_CMD ), +aml_name ("%s" , opts.smi_path))); + } +aml_append (while_ctx2, ifctx); + } + +/* Notify OSPM about new CPUs and clear insert events */ +aml_append (while_ctx2,aml_store (zero, cpu_idx)); + while_ctx =aml_while (aml_lless (cpu_idx, num_added_cpus)); + { +aml_append (while_ctx, +aml_store (aml_derefof (aml_index (new_cpus, cpu_idx)), + uid)); +aml_append (while_ctx, +aml_call2 (CPU_NOTIFY_METHOD , uid, dev_chk)); +aml_append (while_ctx,aml_store (uid,aml_debug ())); +aml_append (while_ctx,aml_store (uid, cpu_selector)); +aml_append (while_ctx,aml_store (one, ins_evt)); +aml_append (while_ctx,aml_increment (cpu_idx)); + } +aml_append (while_ctx2, while_ctx); } -aml_append (method, while_ctx); +aml_append (method, while_ctx2);aml_append (method,aml_release (ctrl_lock)); }aml_append (cpus_dev, method);
Signature · 1 line
2.27.0
Re: [PATCH v5 09/10] x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
Hi Igor,
Quoted 230 lines · Igor Mammedov
In case firmware has negotiated CPU hotplug SMI feature, generate AML to describe SMI IO port region and send SMI to firmware on each CPU hotplug SCI in case new CPUs were hotplugged.
Since new CPUs can be hotplugged while
CPU_SCAN_METHODis running we can't send SMI before new CPUs are fetched from QEMU as it could cause sending Notify to a CPU that firmware hasn't seen yet. So fetch new CPUs into local cache first, then send SMI and after that send Notify events to cached CPUs. This should ensure that Notify is sent only to CPUs which were processed by firmware first.
Any CPUs that were hotplugged after caching will be processed by the nextCPU_SCAN_METHOD, when pending SCI is handled.Signed-off-by Igor Mammedov <[email protected]>---
v5:
- fix hotplug on Windows when there is more than 256 possible CPUs (Windows isn't able to handle VarPackage over 255 elements so process CPUs in batches)
- (Laszlo Ersek <[email protected]>)
- fix off-by-one in package length
- fix not selecting CPU before clearing insert event
- useaml_lgreater()instead of aml_lnot(aml_equal(num_added_cpus, zero)
- split off in separate patches:
- makingsmi_negotiated_featuresa property
- introduction of AcpiPmInfo.smi_on_cpuhp
- introduction of PCI0.SMI0 ACPI device
v2:
- clear insert event after firmware has returned
control from SMI. (Laszlo Ersek <[email protected]>)
v1:
- make sure that Notify is sent only to CPUs seen by FW
- (Laszlo Ersek <[email protected]>)
- use macro instead of x-smi-negotiated-features
- style fixes
- reserve whole SMI block (0xB2,0xB3)
v0:
- s/aml_string/aml_eisaid/
/fixes Windows BSOD, on nonsense value/ (Laszlo Ersek <[email protected]>)
- put SMI device under PCI0 like the rest of hotplug IO port
- do not generate SMI AML if CPU hotplug SMI feature hasn't been negotiated--- hw/acpi/cpu.c | 156 +++++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 129 insertions(+), 27 deletions(-)hw/acpi/cpu.c +124
-26@@ -14,6 +14,8 @@#define ACPI_CPU_CMD_DATA_OFFSET_RW 8#define ACPI_CPU_CMD_DATA2_OFFSET_R 0 +#define OVMF_CPUHP_SMI_CMD 4 +enum {CPHP_GET_NEXT_CPU_WITH_EVENT_CMD =0 ,CPHP_OST_EVENT_CMD =1 ,@@ -321,6 +323,7 @@ const VMStateDescription vmstate_cpu_hotplug = {#define CPU_NOTIFY_METHOD "CTFY"#define CPU_EJECT_METHOD "CEJ0"#define CPU_OST_METHOD "COST" +#define CPU_ADDED_LIST "CNEW"#define CPU_ENABLED "CPEN"#define CPU_SELECTOR "CSEL"@@ -465,42 +468,141 @@ void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts, method =aml_method (CPU_SCAN_METHOD ,0 ,AML_SERIALIZED ); { +const uint8_t max_cpus_per_pass =255 ;Aml *else_ctx; -Aml *while_ctx; +Aml *while_ctx, *while_ctx2;Aml *has_event =aml_local (0 );Aml *dev_chk =aml_int (1 );Aml *eject_req =aml_int (3 );Aml *next_cpu_cmd =aml_int (CPHP_GET_NEXT_CPU_WITH_EVENT_CMD ); +Aml *num_added_cpus =aml_local (1 ); +Aml *cpu_idx =aml_local (2 ); +Aml *uid =aml_local (3 ); +Aml *has_job =aml_local (4 ); +Aml *new_cpus =aml_name (CPU_ADDED_LIST );aml_append (method,aml_acquire (ctrl_lock,0xFFFF )); -aml_append (method,aml_store (one, has_event)); - while_ctx =aml_while (aml_equal (has_event, one)); + +/* +* Windows versions newer than XP (including Windows 10/Windows +* Server 2019), do support* VarPackageOp but, it is cripled to hold +* the same elements number as old PackageOp. +* For compatibility with Windows XP (so it won't crash) use ACPI1.0 +* PackageOp which can hold max 255 elements. +* +* use named package as old Windows don't support it in local var +*/ +aml_append (method,aml_name_decl (CPU_ADDED_LIST , +aml_package (max_cpus_per_pass))); + +aml_append (method,aml_store (zero, uid)); +aml_append (method,aml_store (one, has_job)); +/* +* CPU_ADDED_LIST can hold limited number of elements, outer loop +* allows to process CPUs in batches which let us to handle more +* CPUs than CPU_ADDED_LIST can hold. +*/ + while_ctx2 =aml_while (aml_equal (has_job, one)); { -/* clear loop exit condition, ins_evt/rm_evt checks -* will set it to 1 while next_cpu_cmd returns a CPU -* with events */ -aml_append (while_ctx,aml_store (zero, has_event)); -aml_append (while_ctx,aml_store (next_cpu_cmd, cpu_cmd)); - ifctx =aml_if (aml_equal (ins_evt, one)); - { -aml_append (ifctx, -aml_call2 (CPU_NOTIFY_METHOD , cpu_data, dev_chk)); -aml_append (ifctx,aml_store (one, ins_evt)); -aml_append (ifctx,aml_store (one, has_event)); - } -aml_append (while_ctx, ifctx); - else_ctx =aml_else (); - ifctx =aml_if (aml_equal (rm_evt, one)); - { -aml_append (ifctx, -aml_call2 (CPU_NOTIFY_METHOD , cpu_data, eject_req)); -aml_append (ifctx,aml_store (one, rm_evt)); -aml_append (ifctx,aml_store (one, has_event)); - } -aml_append (else_ctx, ifctx); -aml_append (while_ctx, else_ctx); +aml_append (while_ctx2,aml_store (zero, has_job)); + +aml_append (while_ctx2,aml_store (one, has_event)); +aml_append (while_ctx2,aml_store (zero, num_added_cpus)); + +/* +* Scan CPUs, till there are CPUs with events or +* CPU_ADDED_LIST capacity is exhausted +*/ + while_ctx =aml_while (aml_land (aml_equal (has_event, one), +aml_lless (uid,aml_int (arch_ids->len)))); + { +/* +* clear loop exit condition, ins_evt/rm_evt checks will +* set it to 1 while next_cpu_cmd returns a CPU with events +*/ +aml_append (while_ctx,aml_store (zero, has_event)); + +aml_append (while_ctx,aml_store (uid, cpu_selector)); +aml_append (while_ctx,aml_store (next_cpu_cmd, cpu_cmd)); + +/* +* wrap around case, scan is complete, exit loop. +* It happens since events are not cleared in scan loop, +* so next_cpu_cmd continues to find already processed CPUs +*/ + ifctx =aml_if (aml_lless (cpu_data, uid)); + { +aml_append (ifctx,aml_break ()); + } +aml_append (while_ctx, ifctx); + +/* +* if CPU_ADDED_LIST is full, exit inner loop and process +* collected CPUs +*/ + ifctx =aml_if ( +aml_equal (num_added_cpus,aml_int (max_cpus_per_pass))); + { +aml_append (ifctx,aml_store (one, has_job)); +aml_append (ifctx,aml_break ()); + } +aml_append (while_ctx, ifctx); + +aml_append (while_ctx,aml_store (cpu_data, uid)); + ifctx =aml_if (aml_equal (ins_evt, one)); + { +/* cache added CPUs to Notify/Wakeup later */ +aml_append (ifctx,aml_store (uid, +aml_index (new_cpus, num_added_cpus))); +aml_append (ifctx,aml_increment (num_added_cpus)); +aml_append (ifctx,aml_store (one, has_event)); + } +aml_append (while_ctx, ifctx); + else_ctx =aml_else (); + ifctx =aml_if (aml_equal (rm_evt, one)); + { +aml_append (ifctx, +aml_call2 (CPU_NOTIFY_METHOD , uid, eject_req)); +aml_append (ifctx,aml_store (one, rm_evt)); +aml_append (ifctx,aml_store (one, has_event)); + } +aml_append (else_ctx, ifctx); +aml_append (while_ctx, else_ctx); +aml_append (while_ctx,aml_increment (uid)); + } +aml_append (while_ctx2, while_ctx); + +/* +* in case FW negotiated ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT, +* make upcall to FW, so it can pull in new CPUs before +* OS is notified and wakes them up +*/ +if (opts.smi_path) { + ifctx =aml_if (aml_lgreater (num_added_cpus, zero)); + { +aml_append (ifctx,aml_store (aml_int (OVMF_CPUHP_SMI_CMD ), +aml_name ("%s" , opts.smi_path))); + } +aml_append (while_ctx2, ifctx); + } + +/* Notify OSPM about new CPUs and clear insert events */ +aml_append (while_ctx2,aml_store (zero, cpu_idx)); + while_ctx =aml_while (aml_lless (cpu_idx, num_added_cpus)); + { +aml_append (while_ctx, +aml_store (aml_derefof (aml_index (new_cpus, cpu_idx)), + uid)); +aml_append (while_ctx, +aml_call2 (CPU_NOTIFY_METHOD , uid, dev_chk)); +aml_append (while_ctx,aml_store (uid,aml_debug ())); +aml_append (while_ctx,aml_store (uid, cpu_selector));
hw/acpi/cpu.c void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts, · line 608
Quoted 10 lines
hw/acpi/cpu.c +5
-1+aml_append (while_ctx,aml_store (one, ins_evt)); +aml_append (while_ctx,aml_increment (cpu_idx)); + } +aml_append (while_ctx2, while_ctx); } -aml_append (method, while_ctx); +aml_append (method, while_ctx2);aml_append (method,aml_release (ctrl_lock)); }aml_append (cpus_dev, method);
Here's the ASL decompiled, and using the meaningful local variable names (and 8 possible VCPUs):
1 Method (CSCN, 0, Serialized)
2 {
3 Acquire (\_SB.PCI0.PRES.CPLK, 0xFFFF)
4 Name (CNEW, Package (0xFF){})
5 Local_Uid = Zero
6 Local_HasJob = One
7 While ((Local_HasJob == One))
8 {
9 Local_HasJob = Zero
10 Local_HasEvent = One
11 Local_NumAddedCpus = Zero
12 While (((Local_HasEvent == One) && (Local_Uid < 0x08)))
13 {
14 Local_HasEvent = Zero
15 \_SB.PCI0.PRES.CSEL = Local_Uid
16 \_SB.PCI0.PRES.CCMD = Zero
17 If ((\_SB.PCI0.PRES.CDAT < Local_Uid))
18 {
19 Break
20 }
21
22 If ((Local_NumAddedCpus == 0xFF))
23 {
24 Local_HasJob = One
25 Break
26 }
27
28 Local_Uid = \_SB.PCI0.PRES.CDAT
29 If ((\_SB.PCI0.PRES.CINS == One))
30 {
31 CNEW [Local_NumAddedCpus] = Local_Uid
32 Local_NumAddedCpus++
33 Local_HasEvent = One
34 }
35 ElseIf ((\_SB.PCI0.PRES.CRMV == One))
36 {
37 CTFY (Local_Uid, 0x03)
38 \_SB.PCI0.PRES.CRMV = One
39 Local_HasEvent = One
40 }
41
42 Local_Uid++
43 }
44
45 If ((Local_NumAddedCpus > Zero))
46 {
47 \_SB.PCI0.SMI0.SMIC = 0x04
48 }
49
50 Local_CpuIdx = Zero
51 While ((Local_CpuIdx < Local_NumAddedCpus))
52 {
53 Local_Uid = DerefOf (CNEW [Local_CpuIdx])
54 CTFY (Local_Uid, One)
55 Debug = Local_Uid
56 \_SB.PCI0.PRES.CSEL = Local_Uid
57 \_SB.PCI0.PRES.CINS = One
58 Local_CpuIdx++
59 }
60 }
61
62 Release (\_SB.PCI0.PRES.CPLK)
63 }When we take the Break on line 25, then:
(a) on line 25, the following equality holds:
Local_Uid == CNEW[Local_NumAddedCpus - 1] + 1
(b) on line 60, the following equality holds:
Local_Uid == CNEW[Local_NumAddedCpus - 1]
This means that, when we write Local_Uid to CSEL on line 15 again, then:
- we effectively re-investigate the last-cached CPU (with selector value CNEW[Local_NumAddedCpus - 1])
- rather than resuming the scanning right after it -- that is, with selector value (CNEW[Local_NumAddedCpus - 1] + 1) --, in the spirit of line 42.
My question is: is this "rewind" intentional?
Now, I don't see a functionality problem with this, as on line 57, we clear the pending insert event for the last-cached CPU, so when we re-check it, the "get pending" command will simply seek past it.
But I'd like to understand if this is *precisely* your intent, or if it's an oversight and it just ends up working OK.
Basically my question is whether we should add, on top:
cscn.v5.incr 2020-09-07 16:52:22.133843710 +0200 line 66
Quoted 13 lines
cscn.v5.incr 2020-09-07 16:52:22.133843710 +0200 renamed from cscn.v5 2020-09-07 15:02:13.401663487 +0200 +4
-0@@ -57,6 +57,10 @@ \_SB.PCI0.PRES.CINS = One Local_CpuIdx++ } + + if ((Local_HasJob == One)) { + Local_Uid++ + } } Release (\_SB.PCI0.PRES.CPLK)
If not -- that is, the currently proposed patch is intentional --, then I think we should add a comment, about the effective "rewind" being intentional & okay.
(Note: it's certainly valid and necessary to re-write CSEL on line 15 after raising the SMI on line 47; the question is not whether we should or should not re-write CSEL (we must!), but the specific value that we write to CSEL.)
So:
- If the outer loop body is entered only once, then the patch looks fine, from both the review side, and the testing side (I tested it with 4-8 possible VCPUs).
- If the outer loop body is entered twice or more, then from the review side, please my question above. From the testing side: do you have an environment where I could test this with OVMF?
(I expect it to work OK. Upon the first SMI, the firmware will likely pick up more VCPUs than what's in CNEW. But edk2 commit 020bb4b46d6f ("OvmfPkg/CpuHotplugSmm: fix CPU hotplug race just before SMI broadcast", 2020-08-27) should deal with that.)
Hmm, actually, there's no need for a special environment: I can patch QEMU and lower "max_cpus_per_pass" to something small, such as "3" or whatever, for testing the outer loop multiple times. But first I'd like to know your thoughts on the "rewind".
Thanks!
Laszlo
Re: [PATCH v5 09/10] x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
Quoted 330 lines · On Mon, 7 Sep 2020 17:17:52 +0200 Laszlo Ersek <[email protected]> wrote:
Hi Igor,
Quoted 230 lines · Igor Mammedov
In case firmware has negotiated CPU hotplug SMI feature, generate AML to describe SMI IO port region and send SMI to firmware on each CPU hotplug SCI in case new CPUs were hotplugged.
Since new CPUs can be hotplugged while
CPU_SCAN_METHODis running we can't send SMI before new CPUs are fetched from QEMU as it could cause sending Notify to a CPU that firmware hasn't seen yet. So fetch new CPUs into local cache first, then send SMI and after that send Notify events to cached CPUs. This should ensure that Notify is sent only to CPUs which were processed by firmware first.
Any CPUs that were hotplugged after caching will be processed by the nextCPU_SCAN_METHOD, when pending SCI is handled.Signed-off-by Igor Mammedov <[email protected]>---
v5:
- fix hotplug on Windows when there is more than 256 possible CPUs (Windows isn't able to handle VarPackage over 255 elements so process CPUs in batches)
- (Laszlo Ersek <[email protected]>)
- fix off-by-one in package length
- fix not selecting CPU before clearing insert event
- useaml_lgreater()instead of aml_lnot(aml_equal(num_added_cpus, zero)
- split off in separate patches:
- makingsmi_negotiated_featuresa property
- introduction of AcpiPmInfo.smi_on_cpuhp
- introduction of PCI0.SMI0 ACPI device
v2:
- clear insert event after firmware has returned
control from SMI. (Laszlo Ersek <[email protected]>)
v1:
- make sure that Notify is sent only to CPUs seen by FW
- (Laszlo Ersek <[email protected]>)
- use macro instead of x-smi-negotiated-features
- style fixes
- reserve whole SMI block (0xB2,0xB3)
v0:
- s/aml_string/aml_eisaid/
/fixes Windows BSOD, on nonsense value/ (Laszlo Ersek <[email protected]>)
- put SMI device under PCI0 like the rest of hotplug IO port
- do not generate SMI AML if CPU hotplug SMI feature hasn't been negotiated--- hw/acpi/cpu.c | 156 +++++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 129 insertions(+), 27 deletions(-)hw/acpi/cpu.c +124
-26@@ -14,6 +14,8 @@#define ACPI_CPU_CMD_DATA_OFFSET_RW 8#define ACPI_CPU_CMD_DATA2_OFFSET_R 0 +#define OVMF_CPUHP_SMI_CMD 4 +enum {CPHP_GET_NEXT_CPU_WITH_EVENT_CMD =0 ,CPHP_OST_EVENT_CMD =1 ,@@ -321,6 +323,7 @@ const VMStateDescription vmstate_cpu_hotplug = {#define CPU_NOTIFY_METHOD "CTFY"#define CPU_EJECT_METHOD "CEJ0"#define CPU_OST_METHOD "COST" +#define CPU_ADDED_LIST "CNEW"#define CPU_ENABLED "CPEN"#define CPU_SELECTOR "CSEL"@@ -465,42 +468,141 @@ void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts, method =aml_method (CPU_SCAN_METHOD ,0 ,AML_SERIALIZED ); { +const uint8_t max_cpus_per_pass =255 ;Aml *else_ctx; -Aml *while_ctx; +Aml *while_ctx, *while_ctx2;Aml *has_event =aml_local (0 );Aml *dev_chk =aml_int (1 );Aml *eject_req =aml_int (3 );Aml *next_cpu_cmd =aml_int (CPHP_GET_NEXT_CPU_WITH_EVENT_CMD ); +Aml *num_added_cpus =aml_local (1 ); +Aml *cpu_idx =aml_local (2 ); +Aml *uid =aml_local (3 ); +Aml *has_job =aml_local (4 ); +Aml *new_cpus =aml_name (CPU_ADDED_LIST );aml_append (method,aml_acquire (ctrl_lock,0xFFFF )); -aml_append (method,aml_store (one, has_event)); - while_ctx =aml_while (aml_equal (has_event, one)); + +/* +* Windows versions newer than XP (including Windows 10/Windows +* Server 2019), do support* VarPackageOp but, it is cripled to hold +* the same elements number as old PackageOp. +* For compatibility with Windows XP (so it won't crash) use ACPI1.0 +* PackageOp which can hold max 255 elements. +* +* use named package as old Windows don't support it in local var +*/ +aml_append (method,aml_name_decl (CPU_ADDED_LIST , +aml_package (max_cpus_per_pass))); + +aml_append (method,aml_store (zero, uid)); +aml_append (method,aml_store (one, has_job)); +/* +* CPU_ADDED_LIST can hold limited number of elements, outer loop +* allows to process CPUs in batches which let us to handle more +* CPUs than CPU_ADDED_LIST can hold. +*/ + while_ctx2 =aml_while (aml_equal (has_job, one)); { -/* clear loop exit condition, ins_evt/rm_evt checks -* will set it to 1 while next_cpu_cmd returns a CPU -* with events */ -aml_append (while_ctx,aml_store (zero, has_event)); -aml_append (while_ctx,aml_store (next_cpu_cmd, cpu_cmd)); - ifctx =aml_if (aml_equal (ins_evt, one)); - { -aml_append (ifctx, -aml_call2 (CPU_NOTIFY_METHOD , cpu_data, dev_chk)); -aml_append (ifctx,aml_store (one, ins_evt)); -aml_append (ifctx,aml_store (one, has_event)); - } -aml_append (while_ctx, ifctx); - else_ctx =aml_else (); - ifctx =aml_if (aml_equal (rm_evt, one)); - { -aml_append (ifctx, -aml_call2 (CPU_NOTIFY_METHOD , cpu_data, eject_req)); -aml_append (ifctx,aml_store (one, rm_evt)); -aml_append (ifctx,aml_store (one, has_event)); - } -aml_append (else_ctx, ifctx); -aml_append (while_ctx, else_ctx); +aml_append (while_ctx2,aml_store (zero, has_job)); + +aml_append (while_ctx2,aml_store (one, has_event)); +aml_append (while_ctx2,aml_store (zero, num_added_cpus)); + +/* +* Scan CPUs, till there are CPUs with events or +* CPU_ADDED_LIST capacity is exhausted +*/ + while_ctx =aml_while (aml_land (aml_equal (has_event, one), +aml_lless (uid,aml_int (arch_ids->len)))); + { +/* +* clear loop exit condition, ins_evt/rm_evt checks will +* set it to 1 while next_cpu_cmd returns a CPU with events +*/ +aml_append (while_ctx,aml_store (zero, has_event)); + +aml_append (while_ctx,aml_store (uid, cpu_selector)); +aml_append (while_ctx,aml_store (next_cpu_cmd, cpu_cmd)); + +/* +* wrap around case, scan is complete, exit loop. +* It happens since events are not cleared in scan loop, +* so next_cpu_cmd continues to find already processed CPUs +*/ + ifctx =aml_if (aml_lless (cpu_data, uid)); + { +aml_append (ifctx,aml_break ()); + } +aml_append (while_ctx, ifctx); + +/* +* if CPU_ADDED_LIST is full, exit inner loop and process +* collected CPUs +*/ + ifctx =aml_if ( +aml_equal (num_added_cpus,aml_int (max_cpus_per_pass))); + { +aml_append (ifctx,aml_store (one, has_job)); +aml_append (ifctx,aml_break ()); + } +aml_append (while_ctx, ifctx); + +aml_append (while_ctx,aml_store (cpu_data, uid)); + ifctx =aml_if (aml_equal (ins_evt, one)); + { +/* cache added CPUs to Notify/Wakeup later */ +aml_append (ifctx,aml_store (uid, +aml_index (new_cpus, num_added_cpus))); +aml_append (ifctx,aml_increment (num_added_cpus)); +aml_append (ifctx,aml_store (one, has_event)); + } +aml_append (while_ctx, ifctx); + else_ctx =aml_else (); + ifctx =aml_if (aml_equal (rm_evt, one)); + { +aml_append (ifctx, +aml_call2 (CPU_NOTIFY_METHOD , uid, eject_req)); +aml_append (ifctx,aml_store (one, rm_evt)); +aml_append (ifctx,aml_store (one, has_event)); + } +aml_append (else_ctx, ifctx); +aml_append (while_ctx, else_ctx); +aml_append (while_ctx,aml_increment (uid)); + } +aml_append (while_ctx2, while_ctx); + +/* +* in case FW negotiated ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT, +* make upcall to FW, so it can pull in new CPUs before +* OS is notified and wakes them up +*/ +if (opts.smi_path) { + ifctx =aml_if (aml_lgreater (num_added_cpus, zero)); + { +aml_append (ifctx,aml_store (aml_int (OVMF_CPUHP_SMI_CMD ), +aml_name ("%s" , opts.smi_path))); + } +aml_append (while_ctx2, ifctx); + } + +/* Notify OSPM about new CPUs and clear insert events */ +aml_append (while_ctx2,aml_store (zero, cpu_idx)); + while_ctx =aml_while (aml_lless (cpu_idx, num_added_cpus)); + { +aml_append (while_ctx, +aml_store (aml_derefof (aml_index (new_cpus, cpu_idx)), + uid)); +aml_append (while_ctx, +aml_call2 (CPU_NOTIFY_METHOD , uid, dev_chk)); +aml_append (while_ctx,aml_store (uid,aml_debug ())); +aml_append (while_ctx,aml_store (uid, cpu_selector));hw/acpi/cpu.c void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts, · line 608
Quoted 10 lines
hw/acpi/cpu.c +5
-1+aml_append (while_ctx,aml_store (one, ins_evt)); +aml_append (while_ctx,aml_increment (cpu_idx)); + } +aml_append (while_ctx2, while_ctx); } -aml_append (method, while_ctx); +aml_append (method, while_ctx2);aml_append (method,aml_release (ctrl_lock)); }aml_append (cpus_dev, method);Here's the ASL decompiled, and using the meaningful local variable names (and 8 possible VCPUs):
1 Method (CSCN, 0, Serialized) 2 { 3 Acquire (\_SB.PCI0.PRES.CPLK, 0xFFFF) 4 Name (CNEW, Package (0xFF){}) 5 Local_Uid = Zero 6 Local_HasJob = One 7 While ((Local_HasJob == One)) 8 { 9 Local_HasJob = Zero 10 Local_HasEvent = One 11 Local_NumAddedCpus = Zero 12 While (((Local_HasEvent == One) && (Local_Uid < 0x08))) 13 { 14 Local_HasEvent = Zero 15 \_SB.PCI0.PRES.CSEL = Local_Uid 16 \_SB.PCI0.PRES.CCMD = Zero 17 If ((\_SB.PCI0.PRES.CDAT < Local_Uid)) 18 { 19 Break 20 } 21 22 If ((Local_NumAddedCpus == 0xFF)) 23 { 24 Local_HasJob = One 25 Break 26 } 27 28 Local_Uid = \_SB.PCI0.PRES.CDAT 29 If ((\_SB.PCI0.PRES.CINS == One)) 30 { 31 CNEW [Local_NumAddedCpus] = Local_Uid 32 Local_NumAddedCpus++ 33 Local_HasEvent = One 34 } 35 ElseIf ((\_SB.PCI0.PRES.CRMV == One)) 36 { 37 CTFY (Local_Uid, 0x03) 38 \_SB.PCI0.PRES.CRMV = One 39 Local_HasEvent = One 40 } 41 42 Local_Uid++ 43 } 44 45 If ((Local_NumAddedCpus > Zero)) 46 { 47 \_SB.PCI0.SMI0.SMIC = 0x04 48 } 49 50 Local_CpuIdx = Zero 51 While ((Local_CpuIdx < Local_NumAddedCpus)) 52 { 53 Local_Uid = DerefOf (CNEW [Local_CpuIdx]) 54 CTFY (Local_Uid, One) 55 Debug = Local_Uid 56 \_SB.PCI0.PRES.CSEL = Local_Uid 57 \_SB.PCI0.PRES.CINS = One 58 Local_CpuIdx++ 59 } 60 } 61 62 Release (\_SB.PCI0.PRES.CPLK) 63 }When we take the Break on line 25, then:
(a) on line 25, the following equality holds:
Local_Uid== CNEW[Local_NumAddedCpus - 1] + 1(b) on line 60, the following equality holds:
Local_Uid== CNEW[Local_NumAddedCpus - 1]This means that, when we write
Local_Uidto CSEL on line 15 again, then:- we effectively re-investigate the last-cached CPU (with selector value CNEW[Local_NumAddedCpus - 1])
- rather than resuming the scanning right after it -- that is, with selector value (CNEW[Local_NumAddedCpus - 1] + 1) --, in the spirit of line 42.
Quoted 8 lines
My question is: is this "rewind" intentional?
Now, I don't see a functionality problem with this, as on line 57, we clear the pending insert event for the last-cached CPU, so when we re-check it, the "get pending" command will simply seek past it.
But I'd like to understand if this is *precisely* your intent, or if it's an oversight and it just ends up working OK.
it's the later (it should not have any adverse effects) so I didn't care much about restarting from the last processed CPU.
how about moving
22 If ((Local_NumAddedCpus == 0xFF))
23 {
24 Local_HasJob = One
25 Break
26 }right after
40 }
41
42 Local_Uid++
instead of adding extra 'if' at the end of outer loop?
Quoted 32 lines
Basically my question is whether we should add, on top:
cscn.v5.incr 2020-09-07 16:52:22.133843710 +0200 line 66
Quoted 13 lines
cscn.v5.incr 2020-09-07 16:52:22.133843710 +0200 renamed from cscn.v5 2020-09-07 15:02:13.401663487 +0200 +4
-0@@ -57,6 +57,10 @@ \_SB.PCI0.PRES.CINS = One Local_CpuIdx++ } + + if ((Local_HasJob == One)) { + Local_Uid++ + } } Release (\_SB.PCI0.PRES.CPLK)If not -- that is, the currently proposed patch is intentional --, then I think we should add a comment, about the effective "rewind" being intentional & okay.
(Note: it's certainly valid and necessary to re-write CSEL on line 15 after raising the SMI on line 47; the question is not whether we should or should not re-write CSEL (we must!), but the specific value that we write to CSEL.)
So:
- If the outer loop body is entered only once, then the patch looks fine, from both the review side, and the testing side (I tested it with 4-8 possible VCPUs).
Quoted 8 lines
- If the outer loop body is entered twice or more, then from the review side, please my question above. From the testing side: do you have an environment where I could test this with OVMF?
(I expect it to work OK. Upon the first SMI, the firmware will likely pick up more VCPUs than what's in CNEW. But edk2 commit
020bb4b46d6f("OvmfPkg/CpuHotplugSmm: fix CPU hotplug race just before SMI broadcast", 2020-08-27) should deal with that.)
it should work, as firmware doesn't have to jump over hoops to accomodate Windows quirks.
Quoted 4 lines
Hmm, actually, there's no need for a special environment: I can patch QEMU and lower "
max_cpus_per_pass" to something small, such as "3" or whatever, for testing the outer loop multiple times. But first I'd like to know your thoughts on the "rewind".
modulo OVMF, I've tested both:
- hacking max_cpus_per_pass to simulate multiple passes of the outer loop
- with maxcpus=288 to cross 255 thresold (2 passes of outer loop), to check that WS2019 is still working.
Quoted 4 lines
Thanks!
Laszlo
Laszlo Ersek
Reviewed-byRe: [PATCH v5 09/10] x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
Quoted 105 lines · Igor Mammedov
Quoted 83 lines · On Mon, 7 Sep 2020 17:17:52 +0200 Laszlo Ersek <[email protected]> wrote:
1 Method (CSCN, 0, Serialized) 2 { 3 Acquire (\_SB.PCI0.PRES.CPLK, 0xFFFF) 4 Name (CNEW, Package (0xFF){}) 5 Local_Uid = Zero 6 Local_HasJob = One 7 While ((Local_HasJob == One)) 8 { 9 Local_HasJob = Zero 10 Local_HasEvent = One 11 Local_NumAddedCpus = Zero 12 While (((Local_HasEvent == One) && (Local_Uid < 0x08))) 13 { 14 Local_HasEvent = Zero 15 \_SB.PCI0.PRES.CSEL = Local_Uid 16 \_SB.PCI0.PRES.CCMD = Zero 17 If ((\_SB.PCI0.PRES.CDAT < Local_Uid)) 18 { 19 Break 20 } 21 22 If ((Local_NumAddedCpus == 0xFF)) 23 { 24 Local_HasJob = One 25 Break 26 } 27 28 Local_Uid = \_SB.PCI0.PRES.CDAT 29 If ((\_SB.PCI0.PRES.CINS == One)) 30 { 31 CNEW [Local_NumAddedCpus] = Local_Uid 32 Local_NumAddedCpus++ 33 Local_HasEvent = One 34 } 35 ElseIf ((\_SB.PCI0.PRES.CRMV == One)) 36 { 37 CTFY (Local_Uid, 0x03) 38 \_SB.PCI0.PRES.CRMV = One 39 Local_HasEvent = One 40 } 41 42 Local_Uid++ 43 } 44 45 If ((Local_NumAddedCpus > Zero)) 46 { 47 \_SB.PCI0.SMI0.SMIC = 0x04 48 } 49 50 Local_CpuIdx = Zero 51 While ((Local_CpuIdx < Local_NumAddedCpus)) 52 { 53 Local_Uid = DerefOf (CNEW [Local_CpuIdx]) 54 CTFY (Local_Uid, One) 55 Debug = Local_Uid 56 \_SB.PCI0.PRES.CSEL = Local_Uid 57 \_SB.PCI0.PRES.CINS = One 58 Local_CpuIdx++ 59 } 60 } 61 62 Release (\_SB.PCI0.PRES.CPLK) 63 }When we take the Break on line 25, then:
(a) on line 25, the following equality holds:
Local_Uid== CNEW[Local_NumAddedCpus - 1] + 1(b) on line 60, the following equality holds:
Local_Uid== CNEW[Local_NumAddedCpus - 1]This means that, when we write
Local_Uidto CSEL on line 15 again, then:- we effectively re-investigate the last-cached CPU (with selector value CNEW[Local_NumAddedCpus - 1])
- rather than resuming the scanning right after it -- that is, with selector value (CNEW[Local_NumAddedCpus - 1] + 1) --, in the spirit of line 42.
Quoted 8 lines
My question is: is this "rewind" intentional?
Now, I don't see a functionality problem with this, as on line 57, we clear the pending insert event for the last-cached CPU, so when we re-check it, the "get pending" command will simply seek past it.
But I'd like to understand if this is *precisely* your intent, or if it's an oversight and it just ends up working OK.
it's the later (it should not have any adverse effects) so I didn't care much about restarting from the last processed CPU.
how about moving
22 If ((Local_NumAddedCpus == 0xFF)) 23 { 24 Local_HasJob = One 25 Break 26 }
Quoted 6 lines
right after
40 }
41
42 Local_Uid++instead of adding extra 'if' at the end of outer loop?
That only seems to save a CSEL write on line 15, during the first iteration of the outer loop. And we would still re-select the last selector from CNEW, in the second iteration of the outer loop.
But, again, there's no bug; I just wanted to understand your intent.
Can you please squash the following patch:
Quoted 10 lines
hw/acpi/cpu.c +2
-0@@ -601,6 +601,15 @@ void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts,aml_append (while_ctx,aml_increment (cpu_idx)); }aml_append (while_ctx2, while_ctx); +/* +* If another batch is needed, then it will resume scanning
hw/acpi/cpu.c void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts, · line 615
Quoted 10 lines
hw/acpi/cpu.c +7
-0+* exactly at -- and not after -- the last CPU that's currently +* in CPU_ADDED_LIST. In other words, the last CPU in +* CPU_ADDED_LIST is going to be re-checked. That's OK: we've +* just cleared the insert event for *all* CPUs in +* CPU_ADDED_LIST, including the last one. So the scan will +* simply seek past it. +*/ }aml_append (method, while_ctx2);aml_append (method,aml_release (ctrl_lock));
With that:
I'll also follow up with test results for this patch (meaning a lowered "max_cpus_per_pass").
Thanks!
Laszlo
Laszlo Ersek
Tested-byRe: [PATCH v5 09/10] x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
Quoted 144 lines · Laszlo Ersek
Quoted 105 lines · Igor Mammedov
Quoted 83 lines · On Mon, 7 Sep 2020 17:17:52 +0200 Laszlo Ersek <[email protected]> wrote:
1 Method (CSCN, 0, Serialized) 2 { 3 Acquire (\_SB.PCI0.PRES.CPLK, 0xFFFF) 4 Name (CNEW, Package (0xFF){}) 5 Local_Uid = Zero 6 Local_HasJob = One 7 While ((Local_HasJob == One)) 8 { 9 Local_HasJob = Zero 10 Local_HasEvent = One 11 Local_NumAddedCpus = Zero 12 While (((Local_HasEvent == One) && (Local_Uid < 0x08))) 13 { 14 Local_HasEvent = Zero 15 \_SB.PCI0.PRES.CSEL = Local_Uid 16 \_SB.PCI0.PRES.CCMD = Zero 17 If ((\_SB.PCI0.PRES.CDAT < Local_Uid)) 18 { 19 Break 20 } 21 22 If ((Local_NumAddedCpus == 0xFF)) 23 { 24 Local_HasJob = One 25 Break 26 } 27 28 Local_Uid = \_SB.PCI0.PRES.CDAT 29 If ((\_SB.PCI0.PRES.CINS == One)) 30 { 31 CNEW [Local_NumAddedCpus] = Local_Uid 32 Local_NumAddedCpus++ 33 Local_HasEvent = One 34 } 35 ElseIf ((\_SB.PCI0.PRES.CRMV == One)) 36 { 37 CTFY (Local_Uid, 0x03) 38 \_SB.PCI0.PRES.CRMV = One 39 Local_HasEvent = One 40 } 41 42 Local_Uid++ 43 } 44 45 If ((Local_NumAddedCpus > Zero)) 46 { 47 \_SB.PCI0.SMI0.SMIC = 0x04 48 } 49 50 Local_CpuIdx = Zero 51 While ((Local_CpuIdx < Local_NumAddedCpus)) 52 { 53 Local_Uid = DerefOf (CNEW [Local_CpuIdx]) 54 CTFY (Local_Uid, One) 55 Debug = Local_Uid 56 \_SB.PCI0.PRES.CSEL = Local_Uid 57 \_SB.PCI0.PRES.CINS = One 58 Local_CpuIdx++ 59 } 60 } 61 62 Release (\_SB.PCI0.PRES.CPLK) 63 }When we take the Break on line 25, then:
(a) on line 25, the following equality holds:
Local_Uid== CNEW[Local_NumAddedCpus - 1] + 1(b) on line 60, the following equality holds:
Local_Uid== CNEW[Local_NumAddedCpus - 1]This means that, when we write
Local_Uidto CSEL on line 15 again, then:- we effectively re-investigate the last-cached CPU (with selector value CNEW[Local_NumAddedCpus - 1])
- rather than resuming the scanning right after it -- that is, with selector value (CNEW[Local_NumAddedCpus - 1] + 1) --, in the spirit of line 42.
Quoted 8 lines
My question is: is this "rewind" intentional?
Now, I don't see a functionality problem with this, as on line 57, we clear the pending insert event for the last-cached CPU, so when we re-check it, the "get pending" command will simply seek past it.
But I'd like to understand if this is *precisely* your intent, or if it's an oversight and it just ends up working OK.
it's the later (it should not have any adverse effects) so I didn't care much about restarting from the last processed CPU.
how about moving
22 If ((Local_NumAddedCpus == 0xFF)) 23 { 24 Local_HasJob = One 25 Break 26 }Quoted 6 lines
right after
40 }
41
42 Local_Uid++instead of adding extra 'if' at the end of outer loop?
That only seems to save a CSEL write on line 15, during the first iteration of the outer loop. And we would still re-select the last selector from CNEW, in the second iteration of the outer loop.
But, again, there's no bug; I just wanted to understand your intent.
Can you please squash the following patch:
Quoted 10 lines
hw/acpi/cpu.c +2
-0@@ -601,6 +601,15 @@ void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts,aml_append (while_ctx,aml_increment (cpu_idx)); }aml_append (while_ctx2, while_ctx); +/* +* If another batch is needed, then it will resume scanning hw/acpi/cpu.c void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts, · line 615
Quoted 10 lines
hw/acpi/cpu.c +7
-0+* exactly at -- and not after -- the last CPU that's currently +* in CPU_ADDED_LIST. In other words, the last CPU in +* CPU_ADDED_LIST is going to be re-checked. That's OK: we've +* just cleared the insert event for *all* CPUs in +* CPU_ADDED_LIST, including the last one. So the scan will +* simply seek past it. +*/ }aml_append (method, while_ctx2);aml_append (method,aml_release (ctrl_lock));With that:
Quoted 4 lines
Reviewed-by Laszlo Ersek <[email protected]>I'll also follow up with test results for this patch (meaning a lowered "
max_cpus_per_pass").
PATCH v5 9/10 fixup! x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
In reply to Igor MammedovAdd comment explaining why while_ctx2 restarts from the last processed CPU.
--- hw/acpi/cpu.c | 9 +++++++++ 1 file changed, 9 insertions(+)
hw/acpi/cpu.c +9 -0
@@ -601,6 +601,15 @@ void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts,aml_append (while_ctx,aml_increment (cpu_idx)); }aml_append (while_ctx2, while_ctx); +/* +* If another batch is needed, then it will resume scanning +* exactly at -- and not after -- the last CPU that's currently +* in CPU_ADDED_LIST. In other words, the last CPU in +* CPU_ADDED_LIST is going to be re-checked. That's OK: we've +* just cleared the insert event for *all* CPUs in +* CPU_ADDED_LIST, including the last one. So the scan will +* simply seek past it. +*/ }aml_append (method, while_ctx2);aml_append (method,aml_release (ctrl_lock));
Signature · 1 line
2.27.0
Re: [PATCH v5 9/10] fixup! x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
Quoted 20 lines · Igor Mammedov
Add comment explaining why
while_ctx2restarts from the last processed CPU.Signed-off-by Igor Mammedov <[email protected]>Reviewed-by Laszlo Ersek <[email protected]>Tested-by Laszlo Ersek <[email protected]>--- hw/acpi/cpu.c | 9 +++++++++ 1 file changed, 9 insertions(+)hw/acpi/cpu.c +2
-0@@ -601,6 +601,15 @@ void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts,aml_append (while_ctx,aml_increment (cpu_idx)); }aml_append (while_ctx2, while_ctx); +/* +* If another batch is needed, then it will resume scanning
hw/acpi/cpu.c void build_cpus_aml(Aml *table, MachineState *machine, CPUHotplugFeatures opts, · line 615
Quoted 10 lines
hw/acpi/cpu.c +7
-0+* exactly at -- and not after -- the last CPU that's currently +* in CPU_ADDED_LIST. In other words, the last CPU in +* CPU_ADDED_LIST is going to be re-checked. That's OK: we've +* just cleared the insert event for *all* CPUs in +* CPU_ADDED_LIST, including the last one. So the scan will +* simply seek past it. +*/ }aml_append (method, while_ctx2);aml_append (method,aml_release (ctrl_lock));
Thank you!
Laszlo
PATCH v5 10/10 tests: acpi: update acpi blobs with new AML
In reply to Igor Mammedovhere is diff against tests/data/acpi/q35/DSDT
for currently shipped ovmf binary.
(once firmware blob is updated, it will negotiate CPU hotplug feature which will ad extra hunk sending SMI and Q35 tests will need to be updated), but otherwise diff shows new CPU hotplug AML that is shared between q35 and pc machines.
Method (CSCN, 0, Serialized)
{
Acquire (\_SB.PCI0.PRES.CPLK, 0xFFFF)
- Local0 = One
- While ((Local0 == One))
- {
- Local0 = Zero
- \_SB.PCI0.PRES.CCMD = Zero
- If ((\_SB.PCI0.PRES.CINS == One))
+ Name (CNEW, Package (0xFF){})
+ Local3 = Zero
+ Local4 = One
+ While ((Local4 == One))
+ {
+ Local4 = Zero
+ Local0 = One
+ Local1 = Zero
+ While (((Local0 == One) && (Local3 < One)))
{
- CTFY (\_SB.PCI0.PRES.CDAT, One)
- \_SB.PCI0.PRES.CINS = One
- Local0 = One
+ Local0 = Zero
+ \_SB.PCI0.PRES.CSEL = Local3
+ \_SB.PCI0.PRES.CCMD = Zero
+ If ((\_SB.PCI0.PRES.CDAT < Local3))
+ {+ Break
+ }
+
+ If ((Local1 == 0xFF))
+ {
+ Local4 = One
+ Break
+ }
+
+ Local3 = \_SB.PCI0.PRES.CDAT
+ If ((\_SB.PCI0.PRES.CINS == One))
+ {
+ CNEW [Local1] = Local3
+ Local1++
+ Local0 = One
+ }
+ ElseIf ((\_SB.PCI0.PRES.CRMV == One))
+ {
+ CTFY (Local3, 0x03)
+ \_SB.PCI0.PRES.CRMV = One
+ Local0 = One
+ }
+
+ Local3++
}- ElseIf ((\_SB.PCI0.PRES.CRMV == One))
+
+ Local2 = Zero
+ While ((Local2 < Local1))
{
- CTFY (\_SB.PCI0.PRES.CDAT, 0x03)
- \_SB.PCI0.PRES.CRMV = One
- Local0 = One
+ Local3 = DerefOf (CNEW [Local2])
+ CTFY (Local3, One)
+ Debug = Local3
+ \_SB.PCI0.PRES.CSEL = Local3
+ \_SB.PCI0.PRES.CINS = One
+ Local2++
}
}--- tests/qtest/bios-tables-test-allowed-diff.h | 19 ------------------- tests/data/acpi/pc/DSDT | Bin 4934 -> 5060 bytes tests/data/acpi/pc/DSDT.acpihmat | Bin 6258 -> 6385 bytes tests/data/acpi/pc/DSDT.bridge | Bin 6793 -> 6919 bytes tests/data/acpi/pc/DSDT.cphp | Bin 5397 -> 5524 bytes tests/data/acpi/pc/DSDT.dimmpxm | Bin 6587 -> 6714 bytes tests/data/acpi/pc/DSDT.ipmikcs | Bin 5006 -> 5132 bytes tests/data/acpi/pc/DSDT.memhp | Bin 6293 -> 6419 bytes tests/data/acpi/pc/DSDT.numamem | Bin 4940 -> 5066 bytes tests/data/acpi/q35/DSDT | Bin 7678 -> 7804 bytes tests/data/acpi/q35/DSDT.acpihmat | Bin 9002 -> 9129 bytes tests/data/acpi/q35/DSDT.bridge | Bin 7695 -> 7821 bytes tests/data/acpi/q35/DSDT.cphp | Bin 8141 -> 8268 bytes tests/data/acpi/q35/DSDT.dimmpxm | Bin 9331 -> 9458 bytes tests/data/acpi/q35/DSDT.ipmibt | Bin 7753 -> 7879 bytes tests/data/acpi/q35/DSDT.memhp | Bin 9037 -> 9163 bytes tests/data/acpi/q35/DSDT.mmio64 | Bin 8808 -> 8934 bytes tests/data/acpi/q35/DSDT.numamem | Bin 7684 -> 7810 bytes tests/data/acpi/q35/DSDT.tis | Bin 8283 -> 8409 bytes 19 files changed, 19 deletions(-)
tests/qtest/bios-tables-test-allowed-diff.h +0 -19
@@ -1,20 +1 @@/* List of comma-separated changed AML files to ignore */ -"tests/data/acpi/pc/DSDT" , -"tests/data/acpi/q35/DSDT" , -"tests/data/acpi/q35/DSDT.tis" , -"tests/data/acpi/q35/DSDT.bridge" , -"tests/data/acpi/q35/DSDT.mmio64" , -"tests/data/acpi/q35/DSDT.ipmibt" , -"tests/data/acpi/q35/DSDT.cphp" , -"tests/data/acpi/q35/DSDT.memhp" , -"tests/data/acpi/q35/DSDT.numamem" , -"tests/data/acpi/q35/DSDT.dimmpxm" , -"tests/data/acpi/q35/DSDT.acpihmat" , -"tests/data/acpi/pc/DSDT.bridge" , -"tests/data/acpi/pc/DSDT.ipmikcs" , -"tests/data/acpi/pc/DSDT.cphp" , -"tests/data/acpi/pc/DSDT.memhp" , -"tests/data/acpi/pc/DSDT.numamem" , -"tests/data/acpi/pc/DSDT.dimmpxm" , -"tests/data/acpi/pc/DSDT.acpihmat" , -"tests/data/acpi/pc/DSDT.acpihmat",
tests/data/acpi/pc/DSDT binary +0 -0
tests/data/acpi/pc/DSDT.acpihmat binary +0 -0
tests/data/acpi/pc/DSDT.bridge binary +0 -0
tests/data/acpi/pc/DSDT.cphp binary +0 -0
tests/data/acpi/pc/DSDT.dimmpxm binary +0 -0
tests/data/acpi/pc/DSDT.ipmikcs binary +0 -0
tests/data/acpi/pc/DSDT.memhp binary +0 -0
tests/data/acpi/pc/DSDT.numamem binary +0 -0
tests/data/acpi/q35/DSDT binary +0 -0
tests/data/acpi/q35/DSDT.acpihmat binary +0 -0
tests/data/acpi/q35/DSDT.bridge binary +0 -0
tests/data/acpi/q35/DSDT.cphp binary +0 -0
tests/data/acpi/q35/DSDT.dimmpxm binary +0 -0
tests/data/acpi/q35/DSDT.ipmibt binary +0 -0
tests/data/acpi/q35/DSDT.memhp binary +0 -0
tests/data/acpi/q35/DSDT.mmio64 binary +0 -0
tests/data/acpi/q35/DSDT.numamem binary +0 -0
tests/data/acpi/q35/DSDT.tis binary +0 -0
Signature · 1 line
2.27.0
Quoted 30 lines · On Mon, 7 Sep 2020 07:23:38 -0400 Igor Mammedov <[email protected]> wrote:
v5:
- fix hotplug on Windows when there is more than 256 possible CPUs (Windows isn't able to handle VarPackage over 255 elements
so process CPUs in batches)
- fix off-by-one in package length (Laszlo)
- fix not selecting CPU before clearing insert event (Laszlo)
- useaml_lgreater()instead of aml_lnot(aml_equal(num_added_cpus, zero) (Laszlo)
- split 'x68: acpi: trigger SMI before sending hotplug Notify event to OSPM' in samller chunks (Laszlo)
- fix comment to match spec (Laszlo)
- reorderaml_lor()andaml_land()in header (Laszlo)
v4:
- fix 5.2 machine types so they won't applypc_compat_5_1(Laszlo)
v3:
- rebase on top of "[PATCH v2] hw: add compat machines for 5.2"
so apply that before this patch
v2:
- AML: cleanis_insertedflag only after SMI callback
- make x-smi-cpu-hotunplug false by default
- massage error hint on not supported unplug
v1:
- fix typos and some phrases (Laszlo)
- add unplug check (Laszlo)
- redo AML scan logic to avoid race when adding multiple CPUsCPU hotplug with Secure Boot was not really supported and firmware wasn't aware of hotplugged CPUs (which might lead to guest crashes). During 4.2 we introduced locked SMI handler RAM arrea to make sure that guest OS wasn't able to inject its own SMI handler and OVMF added initial CPU hotplug support.
Quoted 4 lines
This series is QEMU part of that support which lets QMVF tell QEMU that CPU hotplug with SMI broadcast enabled is supported so that QEMU would be able to prevent hotplug in case it's not supported and trigger SMI on hotplug when it's necessary.
Michael,
Can you merge this along with Laszlo's
[PATCH 00/10] edk2: adopt the edk2-stable202008 release
via PCI tree, preferably in the same pull req.
Quoted 44 lines
Igor Mammedov (10):
x86: lpc9: let firmware negotiate 'CPU hotplug with SMI' features
x86: cpuhp: prevent guest crash on CPU hotplug when broadcast SMI is
in use
x86: cpuhp: refuse cpu hot-unplug request earlier if not supported
acpi: addaml_land()andaml_break()primitives
tests: acpi: mark to be changed tables in
bios-tables-test-allowed-diff
x86: ich9: expose "smi_negotiated_features" as a QOM property
x86: acpi: introduceAcpiPmInfo::smi_on_cpuhp
x86: acpi: introduce the PCI0.SMI0 ACPI device
x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
tests: acpi: update acpi blobs with new AMLinclude/hw/acpi/aml-build.h | 2 + include/hw/acpi/cpu.h | 1 + include/hw/i386/ich9.h | 4 + hw/acpi/aml-build.c | 16 +++ hw/acpi/cpu.c | 156 ++++++++++++++++++++++++------ hw/acpi/ich9.c | 24 ++++- hw/i386/acpi-build.c | 35 ++++++- hw/i386/pc.c | 15 ++- hw/isa/lpc_ich9.c | 16 +++ tests/data/acpi/pc/DSDT | Bin 4934 -> 5060 bytes tests/data/acpi/pc/DSDT.acpihmat | Bin 6258 -> 6385 bytes tests/data/acpi/pc/DSDT.bridge | Bin 6793 -> 6919 bytes tests/data/acpi/pc/DSDT.cphp | Bin 5397 -> 5524 bytes tests/data/acpi/pc/DSDT.dimmpxm | Bin 6587 -> 6714 bytes tests/data/acpi/pc/DSDT.ipmikcs | Bin 5006 -> 5132 bytes tests/data/acpi/pc/DSDT.memhp | Bin 6293 -> 6419 bytes tests/data/acpi/pc/DSDT.numamem | Bin 4940 -> 5066 bytes tests/data/acpi/q35/DSDT | Bin 7678 -> 7804 bytes tests/data/acpi/q35/DSDT.acpihmat | Bin 9002 -> 9129 bytes tests/data/acpi/q35/DSDT.bridge | Bin 7695 -> 7821 bytes tests/data/acpi/q35/DSDT.cphp | Bin 8141 -> 8268 bytes tests/data/acpi/q35/DSDT.dimmpxm | Bin 9331 -> 9458 bytes tests/data/acpi/q35/DSDT.ipmibt | Bin 7753 -> 7879 bytes tests/data/acpi/q35/DSDT.memhp | Bin 9037 -> 9163 bytes tests/data/acpi/q35/DSDT.mmio64 | Bin 8808 -> 8934 bytes tests/data/acpi/q35/DSDT.numamem | Bin 7684 -> 7810 bytes tests/data/acpi/q35/DSDT.tis | Bin 8283 -> 8409 bytes 27 files changed, 239 insertions(+), 30 deletions(-)
Quoted 38 lines · Igor Mammedov
Quoted 30 lines · On Mon, 7 Sep 2020 07:23:38 -0400 Igor Mammedov <[email protected]> wrote:
v5:
- fix hotplug on Windows when there is more than 256 possible CPUs (Windows isn't able to handle VarPackage over 255 elements
so process CPUs in batches)
- fix off-by-one in package length (Laszlo)
- fix not selecting CPU before clearing insert event (Laszlo)
- useaml_lgreater()instead of aml_lnot(aml_equal(num_added_cpus, zero) (Laszlo)
- split 'x68: acpi: trigger SMI before sending hotplug Notify event to OSPM' in samller chunks (Laszlo)
- fix comment to match spec (Laszlo)
- reorderaml_lor()andaml_land()in header (Laszlo)
v4:
- fix 5.2 machine types so they won't applypc_compat_5_1(Laszlo)
v3:
- rebase on top of "[PATCH v2] hw: add compat machines for 5.2"
so apply that before this patch
v2:
- AML: cleanis_insertedflag only after SMI callback
- make x-smi-cpu-hotunplug false by default
- massage error hint on not supported unplug
v1:
- fix typos and some phrases (Laszlo)
- add unplug check (Laszlo)
- redo AML scan logic to avoid race when adding multiple CPUsCPU hotplug with Secure Boot was not really supported and firmware wasn't aware of hotplugged CPUs (which might lead to guest crashes). During 4.2 we introduced locked SMI handler RAM arrea to make sure that guest OS wasn't able to inject its own SMI handler and OVMF added initial CPU hotplug support.
Quoted 4 lines
This series is QEMU part of that support which lets QMVF tell QEMU that CPU hotplug with SMI broadcast enabled is supported so that QEMU would be able to prevent hotplug in case it's not supported and trigger SMI on hotplug when it's necessary.
Quoted 5 lines
Michael,
Can you merge this along with Laszlo's
[PATCH 00/10] edk2: adopt the edk2-stable202008 release
via PCI tree, preferably in the same pull req.
That would be very kind, thank you -- however, please give some time to Phil for reviewing the (rest of the) edk2 update series.
Cheers!
Laszlo
Quoted 47 lines
Quoted 46 lines
Igor Mammedov (10):
x86: lpc9: let firmware negotiate 'CPU hotplug with SMI' features
x86: cpuhp: prevent guest crash on CPU hotplug when broadcast SMI is
in use
x86: cpuhp: refuse cpu hot-unplug request earlier if not supported
acpi: addaml_land()andaml_break()primitives
tests: acpi: mark to be changed tables in
bios-tables-test-allowed-diff
x86: ich9: expose "smi_negotiated_features" as a QOM property
x86: acpi: introduceAcpiPmInfo::smi_on_cpuhp
x86: acpi: introduce the PCI0.SMI0 ACPI device
x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
tests: acpi: update acpi blobs with new AMLinclude/hw/acpi/aml-build.h | 2 + include/hw/acpi/cpu.h | 1 + include/hw/i386/ich9.h | 4 + hw/acpi/aml-build.c | 16 +++ hw/acpi/cpu.c | 156 ++++++++++++++++++++++++------ hw/acpi/ich9.c | 24 ++++- hw/i386/acpi-build.c | 35 ++++++- hw/i386/pc.c | 15 ++- hw/isa/lpc_ich9.c | 16 +++ tests/data/acpi/pc/DSDT | Bin 4934 -> 5060 bytes tests/data/acpi/pc/DSDT.acpihmat | Bin 6258 -> 6385 bytes tests/data/acpi/pc/DSDT.bridge | Bin 6793 -> 6919 bytes tests/data/acpi/pc/DSDT.cphp | Bin 5397 -> 5524 bytes tests/data/acpi/pc/DSDT.dimmpxm | Bin 6587 -> 6714 bytes tests/data/acpi/pc/DSDT.ipmikcs | Bin 5006 -> 5132 bytes tests/data/acpi/pc/DSDT.memhp | Bin 6293 -> 6419 bytes tests/data/acpi/pc/DSDT.numamem | Bin 4940 -> 5066 bytes tests/data/acpi/q35/DSDT | Bin 7678 -> 7804 bytes tests/data/acpi/q35/DSDT.acpihmat | Bin 9002 -> 9129 bytes tests/data/acpi/q35/DSDT.bridge | Bin 7695 -> 7821 bytes tests/data/acpi/q35/DSDT.cphp | Bin 8141 -> 8268 bytes tests/data/acpi/q35/DSDT.dimmpxm | Bin 9331 -> 9458 bytes tests/data/acpi/q35/DSDT.ipmibt | Bin 7753 -> 7879 bytes tests/data/acpi/q35/DSDT.memhp | Bin 9037 -> 9163 bytes tests/data/acpi/q35/DSDT.mmio64 | Bin 8808 -> 8934 bytes tests/data/acpi/q35/DSDT.numamem | Bin 7684 -> 7810 bytes tests/data/acpi/q35/DSDT.tis | Bin 8283 -> 8409 bytes 27 files changed, 239 insertions(+), 30 deletions(-)
Quoted 16 lines · On Mon, 7 Sep 2020 07:23:38 -0400 Igor Mammedov <[email protected]> wrote:
v5:
- fix hotplug on Windows when there is more than 256 possible CPUs (Windows isn't able to handle VarPackage over 255 elements
so process CPUs in batches)
- fix off-by-one in package length (Laszlo)
- fix not selecting CPU before clearing insert event (Laszlo)
- useaml_lgreater()instead of aml_lnot(aml_equal(num_added_cpus, zero) (Laszlo)
- split 'x68: acpi: trigger SMI before sending hotplug Notify event to OSPM' in samller chunks (Laszlo)
- fix comment to match spec (Laszlo)
- reorderaml_lor()andaml_land()in header (Laszlo)
v4:
- fix 5.2 machine types so they won't applypc_compat_5_1(Laszlo)
v3:
- rebase on top of "[PATCH v2] hw: add compat machines for 5.2"
so apply that before this patch
Quoted 8 lines
v2:
- AML: cleanis_insertedflag only after SMI callback
- make x-smi-cpu-hotunplug false by default
- massage error hint on not supported unplug
v1:
- fix typos and some phrases (Laszlo)
- add unplug check (Laszlo)
- redo AML scan logic to avoid race when adding multiple CPUs
Michael,
just saw your pull request which missed this series. Is there any plans to queue it for the next pull request?
Quoted 53 lines
CPU hotplug with Secure Boot was not really supported and firmware wasn't aware of hotplugged CPUs (which might lead to guest crashes). During 4.2 we introduced locked SMI handler RAM arrea to make sure that guest OS wasn't able to inject its own SMI handler and OVMF added initial CPU hotplug support.
This series is QEMU part of that support which lets QMVF tell QEMU that CPU hotplug with SMI broadcast enabled is supported so that QEMU would be able to prevent hotplug in case it's not supported and trigger SMI on hotplug when it's necessary.
Igor Mammedov (10):
x86: lpc9: let firmware negotiate 'CPU hotplug with SMI' features
x86: cpuhp: prevent guest crash on CPU hotplug when broadcast SMI is
in use
x86: cpuhp: refuse cpu hot-unplug request earlier if not supported
acpi: addaml_land()andaml_break()primitives
tests: acpi: mark to be changed tables in
bios-tables-test-allowed-diff
x86: ich9: expose "smi_negotiated_features" as a QOM property
x86: acpi: introduceAcpiPmInfo::smi_on_cpuhp
x86: acpi: introduce the PCI0.SMI0 ACPI device
x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
tests: acpi: update acpi blobs with new AMLinclude/hw/acpi/aml-build.h | 2 + include/hw/acpi/cpu.h | 1 + include/hw/i386/ich9.h | 4 + hw/acpi/aml-build.c | 16 +++ hw/acpi/cpu.c | 156 ++++++++++++++++++++++++------ hw/acpi/ich9.c | 24 ++++- hw/i386/acpi-build.c | 35 ++++++- hw/i386/pc.c | 15 ++- hw/isa/lpc_ich9.c | 16 +++ tests/data/acpi/pc/DSDT | Bin 4934 -> 5060 bytes tests/data/acpi/pc/DSDT.acpihmat | Bin 6258 -> 6385 bytes tests/data/acpi/pc/DSDT.bridge | Bin 6793 -> 6919 bytes tests/data/acpi/pc/DSDT.cphp | Bin 5397 -> 5524 bytes tests/data/acpi/pc/DSDT.dimmpxm | Bin 6587 -> 6714 bytes tests/data/acpi/pc/DSDT.ipmikcs | Bin 5006 -> 5132 bytes tests/data/acpi/pc/DSDT.memhp | Bin 6293 -> 6419 bytes tests/data/acpi/pc/DSDT.numamem | Bin 4940 -> 5066 bytes tests/data/acpi/q35/DSDT | Bin 7678 -> 7804 bytes tests/data/acpi/q35/DSDT.acpihmat | Bin 9002 -> 9129 bytes tests/data/acpi/q35/DSDT.bridge | Bin 7695 -> 7821 bytes tests/data/acpi/q35/DSDT.cphp | Bin 8141 -> 8268 bytes tests/data/acpi/q35/DSDT.dimmpxm | Bin 9331 -> 9458 bytes tests/data/acpi/q35/DSDT.ipmibt | Bin 7753 -> 7879 bytes tests/data/acpi/q35/DSDT.memhp | Bin 9037 -> 9163 bytes tests/data/acpi/q35/DSDT.mmio64 | Bin 8808 -> 8934 bytes tests/data/acpi/q35/DSDT.numamem | Bin 7684 -> 7810 bytes tests/data/acpi/q35/DSDT.tis | Bin 8283 -> 8409 bytes 27 files changed, 239 insertions(+), 30 deletions(-)
Quoted 28 lines · Igor Mammedov
Quoted 16 lines · On Mon, 7 Sep 2020 07:23:38 -0400 Igor Mammedov <[email protected]> wrote:
v5:
- fix hotplug on Windows when there is more than 256 possible CPUs (Windows isn't able to handle VarPackage over 255 elements
so process CPUs in batches)
- fix off-by-one in package length (Laszlo)
- fix not selecting CPU before clearing insert event (Laszlo)
- useaml_lgreater()instead of aml_lnot(aml_equal(num_added_cpus, zero) (Laszlo)
- split 'x68: acpi: trigger SMI before sending hotplug Notify event to OSPM' in samller chunks (Laszlo)
- fix comment to match spec (Laszlo)
- reorderaml_lor()andaml_land()in header (Laszlo)
v4:
- fix 5.2 machine types so they won't applypc_compat_5_1(Laszlo)
v3:
- rebase on top of "[PATCH v2] hw: add compat machines for 5.2"
so apply that before this patchQuoted 8 lines
v2:
- AML: cleanis_insertedflag only after SMI callback
- make x-smi-cpu-hotunplug false by default
- massage error hint on not supported unplug
v1:
- fix typos and some phrases (Laszlo)
- add unplug check (Laszlo)
- redo AML scan logic to avoid race when adding multiple CPUs
Quoted 4 lines
Michael,
just saw your pull request which missed this series. Is there any plans to queue it for the next pull request?
Oh.
You didn't Cc me on most patches so I assumed this is targeting some other tree. Sorry.
Will review and queue, thanks.
Quoted 55 lines
Quoted 55 lines
CPU hotplug with Secure Boot was not really supported and firmware wasn't aware of hotplugged CPUs (which might lead to guest crashes). During 4.2 we introduced locked SMI handler RAM arrea to make sure that guest OS wasn't able to inject its own SMI handler and OVMF added initial CPU hotplug support.
This series is QEMU part of that support which lets QMVF tell QEMU that CPU hotplug with SMI broadcast enabled is supported so that QEMU would be able to prevent hotplug in case it's not supported and trigger SMI on hotplug when it's necessary.
Igor Mammedov (10):
x86: lpc9: let firmware negotiate 'CPU hotplug with SMI' features
x86: cpuhp: prevent guest crash on CPU hotplug when broadcast SMI is
in use
x86: cpuhp: refuse cpu hot-unplug request earlier if not supported
acpi: addaml_land()andaml_break()primitives
tests: acpi: mark to be changed tables in
bios-tables-test-allowed-diff
x86: ich9: expose "smi_negotiated_features" as a QOM property
x86: acpi: introduceAcpiPmInfo::smi_on_cpuhp
x86: acpi: introduce the PCI0.SMI0 ACPI device
x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
tests: acpi: update acpi blobs with new AMLinclude/hw/acpi/aml-build.h | 2 + include/hw/acpi/cpu.h | 1 + include/hw/i386/ich9.h | 4 + hw/acpi/aml-build.c | 16 +++ hw/acpi/cpu.c | 156 ++++++++++++++++++++++++------ hw/acpi/ich9.c | 24 ++++- hw/i386/acpi-build.c | 35 ++++++- hw/i386/pc.c | 15 ++- hw/isa/lpc_ich9.c | 16 +++ tests/data/acpi/pc/DSDT | Bin 4934 -> 5060 bytes tests/data/acpi/pc/DSDT.acpihmat | Bin 6258 -> 6385 bytes tests/data/acpi/pc/DSDT.bridge | Bin 6793 -> 6919 bytes tests/data/acpi/pc/DSDT.cphp | Bin 5397 -> 5524 bytes tests/data/acpi/pc/DSDT.dimmpxm | Bin 6587 -> 6714 bytes tests/data/acpi/pc/DSDT.ipmikcs | Bin 5006 -> 5132 bytes tests/data/acpi/pc/DSDT.memhp | Bin 6293 -> 6419 bytes tests/data/acpi/pc/DSDT.numamem | Bin 4940 -> 5066 bytes tests/data/acpi/q35/DSDT | Bin 7678 -> 7804 bytes tests/data/acpi/q35/DSDT.acpihmat | Bin 9002 -> 9129 bytes tests/data/acpi/q35/DSDT.bridge | Bin 7695 -> 7821 bytes tests/data/acpi/q35/DSDT.cphp | Bin 8141 -> 8268 bytes tests/data/acpi/q35/DSDT.dimmpxm | Bin 9331 -> 9458 bytes tests/data/acpi/q35/DSDT.ipmibt | Bin 7753 -> 7879 bytes tests/data/acpi/q35/DSDT.memhp | Bin 9037 -> 9163 bytes tests/data/acpi/q35/DSDT.mmio64 | Bin 8808 -> 8934 bytes tests/data/acpi/q35/DSDT.numamem | Bin 7684 -> 7810 bytes tests/data/acpi/q35/DSDT.tis | Bin 8283 -> 8409 bytes 27 files changed, 239 insertions(+), 30 deletions(-)
Quoted 31 lines · On Mon, 21 Sep 2020 08:34:31 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 30 lines · Igor Mammedov
Quoted 16 lines · On Mon, 7 Sep 2020 07:23:38 -0400 Igor Mammedov <[email protected]> wrote:
v5:
- fix hotplug on Windows when there is more than 256 possible CPUs (Windows isn't able to handle VarPackage over 255 elements
so process CPUs in batches)
- fix off-by-one in package length (Laszlo)
- fix not selecting CPU before clearing insert event (Laszlo)
- useaml_lgreater()instead of aml_lnot(aml_equal(num_added_cpus, zero) (Laszlo)
- split 'x68: acpi: trigger SMI before sending hotplug Notify event to OSPM' in samller chunks (Laszlo)
- fix comment to match spec (Laszlo)
- reorderaml_lor()andaml_land()in header (Laszlo)
v4:
- fix 5.2 machine types so they won't applypc_compat_5_1(Laszlo)
v3:
- rebase on top of "[PATCH v2] hw: add compat machines for 5.2"
so apply that before this patchQuoted 8 lines
v2:
- AML: cleanis_insertedflag only after SMI callback
- make x-smi-cpu-hotunplug false by default
- massage error hint on not supported unplug
v1:
- fix typos and some phrases (Laszlo)
- add unplug check (Laszlo)
- redo AML scan logic to avoid race when adding multiple CPUsMichael,
Quoted 7 lines
Quoted 2 lines
just saw your pull request which missed this series. Is there any plans to queue it for the next pull request?
Oh.
You didn't Cc me on most patches so I assumed this is targeting some other tree. Sorry.
Will review and queue, thanks.
Thanks,
Series doesn't apply anymore,
I've rebased fixing conflicts and resent it as v6.
Quoted 58 lines
Quoted 56 lines
Quoted 56 lines
CPU hotplug with Secure Boot was not really supported and firmware wasn't aware of hotplugged CPUs (which might lead to guest crashes). During 4.2 we introduced locked SMI handler RAM arrea to make sure that guest OS wasn't able to inject its own SMI handler and OVMF added initial CPU hotplug support.
This series is QEMU part of that support which lets QMVF tell QEMU that CPU hotplug with SMI broadcast enabled is supported so that QEMU would be able to prevent hotplug in case it's not supported and trigger SMI on hotplug when it's necessary.
Igor Mammedov (10):
x86: lpc9: let firmware negotiate 'CPU hotplug with SMI' features
x86: cpuhp: prevent guest crash on CPU hotplug when broadcast SMI is
in use
x86: cpuhp: refuse cpu hot-unplug request earlier if not supported
acpi: addaml_land()andaml_break()primitives
tests: acpi: mark to be changed tables in
bios-tables-test-allowed-diff
x86: ich9: expose "smi_negotiated_features" as a QOM property
x86: acpi: introduceAcpiPmInfo::smi_on_cpuhp
x86: acpi: introduce the PCI0.SMI0 ACPI device
x68: acpi: trigger SMI before sending hotplug Notify event to OSPM
tests: acpi: update acpi blobs with new AMLinclude/hw/acpi/aml-build.h | 2 + include/hw/acpi/cpu.h | 1 + include/hw/i386/ich9.h | 4 + hw/acpi/aml-build.c | 16 +++ hw/acpi/cpu.c | 156 ++++++++++++++++++++++++------ hw/acpi/ich9.c | 24 ++++- hw/i386/acpi-build.c | 35 ++++++- hw/i386/pc.c | 15 ++- hw/isa/lpc_ich9.c | 16 +++ tests/data/acpi/pc/DSDT | Bin 4934 -> 5060 bytes tests/data/acpi/pc/DSDT.acpihmat | Bin 6258 -> 6385 bytes tests/data/acpi/pc/DSDT.bridge | Bin 6793 -> 6919 bytes tests/data/acpi/pc/DSDT.cphp | Bin 5397 -> 5524 bytes tests/data/acpi/pc/DSDT.dimmpxm | Bin 6587 -> 6714 bytes tests/data/acpi/pc/DSDT.ipmikcs | Bin 5006 -> 5132 bytes tests/data/acpi/pc/DSDT.memhp | Bin 6293 -> 6419 bytes tests/data/acpi/pc/DSDT.numamem | Bin 4940 -> 5066 bytes tests/data/acpi/q35/DSDT | Bin 7678 -> 7804 bytes tests/data/acpi/q35/DSDT.acpihmat | Bin 9002 -> 9129 bytes tests/data/acpi/q35/DSDT.bridge | Bin 7695 -> 7821 bytes tests/data/acpi/q35/DSDT.cphp | Bin 8141 -> 8268 bytes tests/data/acpi/q35/DSDT.dimmpxm | Bin 9331 -> 9458 bytes tests/data/acpi/q35/DSDT.ipmibt | Bin 7753 -> 7879 bytes tests/data/acpi/q35/DSDT.memhp | Bin 9037 -> 9163 bytes tests/data/acpi/q35/DSDT.mmio64 | Bin 8808 -> 8934 bytes tests/data/acpi/q35/DSDT.numamem | Bin 7684 -> 7810 bytes tests/data/acpi/q35/DSDT.tis | Bin 8283 -> 8409 bytes 27 files changed, 239 insertions(+), 30 deletions(-)