ppc/pnv: add XIVE support for KVM guests
v3 cover ppc/pnv: add XIVE support for KVM guests
Hello,
The QEMU PowerNV machine emulates a baremetal OpenPOWER system and acts as an hypervisor (L0). Supporting emulation of KVM to run guests (L1) requires a few more extensions, among which guest support for the XIVE interrupt controller on POWER9 processor.
The following changes add new per-CPU PowerNV machines and extend the XIVE models with the new XiveFabric and XivePresenter interfaces to provide support for XIVE escalations and interrupt resend. This mechanism is used by XIVE to notify the hypervisor that a vCPU is not dispatched on a HW thread. Tested on a QEMU PowerNV machine and a simple QEMU pseries guest doing network on a local bridge.
The XIVE interrupt controller offers a way to increase the XIVE resources per chip by configuring multiple XIVE blocks on a chip. This is not currently supported by the model. However, some configurations, such as OPAL/skiboot, use one block-per-chip configuration with some optimizations. One of them is to override the hardwired chip ID by the block id in the PowerBUS operations and for CAM line compares. This patchset improves the support for this setup. Tested with 4 chips.
A series from Suraj adding guest support in the Radix MMU model of the QEMU PowerNV machine is still required and will be send later. The whole patchset can be found under :
https://github.com/legoater/qemu/tree/powernv-4.1
Thanks,
C.
Changes since v2:
- introduced the XiveFabric and XivePresenter interfaces
- removed the need of a XiveRouter pointer under XiveTCTX
Changes since v1:
- minor extra fixes
- split the escalation support in different patches
- kept the XiveRouter type for XiveTCTX back pointer (will address this in P10)
- removed pnv_xive_vst_size(). Really broken on indirect tables.
- improved the dump of the NVT table
- introduce pnv_xive_get_block_id()
Cédric Le Goater (18):
ppc/pnv: Introduce PowerNV machines with fixed CPU models tests/boot-serial-test: add support for all the PowerNV machines ppc/xive: Introduce the XiveFabric and XivePresenter interfaces ppc/pnv: Implement the XiveFabric and XivePresenter interfaces ppc/spapr: Implement the XiveFabric and XivePresenter interfaces ppc/xive: Use the XiveFabric and XivePresenter interfaces ppc/xive: Extend the TIMA operation with a XivePresenter parameter ppc/pnv: Clarify how the TIMA is accessed on a multichip system ppc/xive: Move the TIMA operations to the controller model ppc/xive: Introduce a xive_tctx_ipb_update() helper ppc/xive: Synthesize interrupt from the saved IPB in the NVT ppc/pnv: Remove pnv_xive_vst_size() routine
ppc/pnv: Dump the XIVE NVT table
ppc/pnv: Skip empty slots of the XIVE NVT table ppc/pnv: Introduce a pnv_xive_block_id() helper ppc/pnv: Extend XivePresenter with a get_block_id() handler ppc/pnv: Quiesce some XIVE errors
ppc/xive: Introduce a xive_os_cam_decode() helper
include/hw/ppc/pnv.h | 13 ++ include/hw/ppc/pnv_xive.h | 3 - include/hw/ppc/spapr_irq.h | 6 + include/hw/ppc/xive.h | 71 ++++++-- include/hw/ppc/xive_regs.h | 24 +++ hw/intc/pnv_xive.c | 356 ++++++++++++++++++++++++------------- hw/intc/spapr_xive.c | 80 ++++++++- hw/intc/xive.c | 319 ++++++++++++++++++--------------- hw/ppc/pnv.c | 102 ++++++++++- hw/ppc/spapr.c | 34 ++++ hw/ppc/spapr_irq.c | 25 +++ tests/boot-serial-test.c | 3 +- 12 files changed, 743 insertions(+), 293 deletions(-)
Signature · 1 line
2.21.0
PATCH v3 1/18 ppc/pnv: Introduce PowerNV machines with fixed CPU models
Make the current "powernv" machine an abstract type and derive from it new machines with specific CPU models: power8 and power9.
The "powernv" machine is now an alias on the "powernv9" machine.
--- hw/ppc/pnv.c | 70 ++++++++++++++++++++++++++++++++++++++++++++++------ 1 file changed, 63 insertions(+), 7 deletions(-)
hw/ppc/pnv.c +63 -7
@@ -602,9 +602,20 @@ static void pnv_chip_power9_pic_print_info(PnvChip *chip, Monitor *mon)pnv_psi_pic_print_info (&chip9->psi, mon); } +static bool pnv_match_cpu (const char *default_type,const char *cpu_type) +{ +PowerPCCPUClass *ppc_default = +POWERPC_CPU_CLASS (object_class_by_name (default_type)); +PowerPCCPUClass *ppc = +POWERPC_CPU_CLASS (object_class_by_name (cpu_type)); + +return ppc_default->pvr_match (ppc_default, ppc->pvr); +} +static void pnv_init (MachineState *machine) {PnvMachineState *pnv =PNV_MACHINE (machine); +MachineClass *mc =MACHINE_GET_CLASS (machine);MemoryRegion *ram;char *fw_filename;long fw_size;
@@ -664,13 +675,23 @@ static void pnv_init(MachineState *machine) } } +/* +* Check compatibility of the specified CPU with the machine +* default. +*/ +if (!pnv_match_cpu (mc->default_cpu_type, machine->cpu_type)) { +error_report ("invalid CPU model '%s' for %s machine" , + machine->cpu_type, mc->name); +exit (1 ); + } +/* Create the processor chips */ i =strlen (machine->cpu_type) -strlen (POWERPC_CPU_TYPE_SUFFIX ); chip_typename =g_strdup_printf (PNV_CHIP_TYPE_NAME ("%.*s" ), i, machine->cpu_type);if (!object_class_by_name (chip_typename)) { -error_report ("invalid CPU model '%.*s' for %s machine" , - i, machine->cpu_type,MACHINE_GET_CLASS (machine)->name); +error_report ("invalid chip model '%.*s' for %s machine" , + i, machine->cpu_type, mc->name);exit (1 ); }
@@ -1348,25 +1369,43 @@ static void pnv_machine_class_props_init(ObjectClass *oc) NULL); } -static void pnv_machine_class_init (ObjectClass *oc,void *data) +static void pnv_machine_power8_class_init (ObjectClass *oc,void *data) {MachineClass *mc =MACHINE_CLASS (oc);XICSFabricClass *xic =XICS_FABRIC_CLASS (oc); + + mc->desc ="IBM PowerNV (Non-Virtualized) POWER8" ; + mc->default_cpu_type =POWERPC_CPU_TYPE_NAME ("power8_v2.0" ); + + xic->icp_get = pnv_icp_get; + xic->ics_get = pnv_ics_get; + xic->ics_resend = pnv_ics_resend; +} + +static void pnv_machine_power9_class_init (ObjectClass *oc,void *data) +{ +MachineClass *mc =MACHINE_CLASS (oc); + + mc->desc ="IBM PowerNV (Non-Virtualized) POWER9" ; + mc->default_cpu_type =POWERPC_CPU_TYPE_NAME ("power9_v2.0" ); + + mc->alias ="powernv" ; +} + +static void pnv_machine_class_init (ObjectClass *oc,void *data) +{ +MachineClass *mc =MACHINE_CLASS (oc);InterruptStatsProviderClass *ispc =INTERRUPT_STATS_PROVIDER_CLASS (oc); mc->desc ="IBM PowerNV (Non-Virtualized)" ; mc->init = pnv_init; mc->reset = pnv_reset; mc->max_cpus =MAX_CPUS ; - mc->default_cpu_type =POWERPC_CPU_TYPE_NAME ("power8_v2.0" ); mc->block_default_type =IF_IDE ;/* Pnv provides a AHCI device for * storage */ mc->no_parallel =1 ; mc->default_boot_order =NULL ; mc->default_ram_size =2 * GiB; - xic->icp_get = pnv_icp_get; - xic->ics_get = pnv_ics_get; - xic->ics_resend = pnv_ics_resend; ispc->print_info = pnv_pic_print_info;pnv_machine_class_props_init (oc);
@@ -1386,10 +1425,27 @@ static void pnv_machine_class_init(ObjectClass *oc, void *data) .parent = TYPE_PNV9_CHIP, \ } +#define DEFINE_PNV_MACHINE_TYPE (cpu, class_initfn) \ + { \ + .name = MACHINE_TYPE_NAME(cpu), \ + .parent = TYPE_PNV_MACHINE, \ + .instance_size = sizeof(PnvMachineState), \ + .instance_init = pnv_machine_instance_init, \ + .class_init = class_initfn, \ + .interfaces = (InterfaceInfo[]) { \ + { TYPE_XICS_FABRIC }, \ + { TYPE_INTERRUPT_STATS_PROVIDER }, \ + { }, \ + }, \ + } +static const TypeInfo types[] = { +DEFINE_PNV_MACHINE_TYPE ("powernv8" , pnv_machine_power8_class_init), +DEFINE_PNV_MACHINE_TYPE ("powernv9" , pnv_machine_power9_class_init), { .name =TYPE_PNV_MACHINE , .parent =TYPE_MACHINE , + .abstract = true, .instance_size =sizeof (PnvMachineState), .instance_init = pnv_machine_instance_init, .class_init = pnv_machine_class_init,
Signature · 1 line
2.21.0
David Gibson
AppliedRe: [Qemu-devel] [PATCH v3 01/18] ppc/pnv: Introduce PowerNV machines with fixed CPU models
Quoted 6 lines · Cédric Le Goater
Make the current "powernv" machine an abstract type and derive from it new machines with specific CPU models: power8 and power9.
The "powernv" machine is now an alias on the "powernv9" machine.
Signed-off-by Cédric Le Goater <[email protected]>
Applied to ppc-for-4.2.
Quoted 132 lines
--- hw/ppc/pnv.c | 70 ++++++++++++++++++++++++++++++++++++++++++++++------ 1 file changed, 63 insertions(+), 7 deletions(-)hw/ppc/pnv.c +63
-7@@ -602,9 +602,20 @@ static void pnv_chip_power9_pic_print_info(PnvChip *chip, Monitor *mon)pnv_psi_pic_print_info (&chip9->psi, mon); } +static bool pnv_match_cpu (const char *default_type,const char *cpu_type) +{ +PowerPCCPUClass *ppc_default = +POWERPC_CPU_CLASS (object_class_by_name (default_type)); +PowerPCCPUClass *ppc = +POWERPC_CPU_CLASS (object_class_by_name (cpu_type)); + +return ppc_default->pvr_match (ppc_default, ppc->pvr); +} +static void pnv_init (MachineState *machine) {PnvMachineState *pnv =PNV_MACHINE (machine); +MachineClass *mc =MACHINE_GET_CLASS (machine);MemoryRegion *ram;char *fw_filename;long fw_size;@@ -664,13 +675,23 @@ static void pnv_init(MachineState *machine) } } +/* +* Check compatibility of the specified CPU with the machine +* default. +*/ +if (!pnv_match_cpu (mc->default_cpu_type, machine->cpu_type)) { +error_report ("invalid CPU model '%s' for %s machine" , + machine->cpu_type, mc->name); +exit (1 ); + } +/* Create the processor chips */ i =strlen (machine->cpu_type) -strlen (POWERPC_CPU_TYPE_SUFFIX ); chip_typename =g_strdup_printf (PNV_CHIP_TYPE_NAME ("%.*s" ), i, machine->cpu_type);if (!object_class_by_name (chip_typename)) { -error_report ("invalid CPU model '%.*s' for %s machine" , - i, machine->cpu_type,MACHINE_GET_CLASS (machine)->name); +error_report ("invalid chip model '%.*s' for %s machine" , + i, machine->cpu_type, mc->name);exit (1 ); }@@ -1348,25 +1369,43 @@ static void pnv_machine_class_props_init(ObjectClass *oc) NULL); } -static void pnv_machine_class_init (ObjectClass *oc,void *data) +static void pnv_machine_power8_class_init (ObjectClass *oc,void *data) {MachineClass *mc =MACHINE_CLASS (oc);XICSFabricClass *xic =XICS_FABRIC_CLASS (oc); + + mc->desc ="IBM PowerNV (Non-Virtualized) POWER8" ; + mc->default_cpu_type =POWERPC_CPU_TYPE_NAME ("power8_v2.0" ); + + xic->icp_get = pnv_icp_get; + xic->ics_get = pnv_ics_get; + xic->ics_resend = pnv_ics_resend; +} + +static void pnv_machine_power9_class_init (ObjectClass *oc,void *data) +{ +MachineClass *mc =MACHINE_CLASS (oc); + + mc->desc ="IBM PowerNV (Non-Virtualized) POWER9" ; + mc->default_cpu_type =POWERPC_CPU_TYPE_NAME ("power9_v2.0" ); + + mc->alias ="powernv" ; +} + +static void pnv_machine_class_init (ObjectClass *oc,void *data) +{ +MachineClass *mc =MACHINE_CLASS (oc);InterruptStatsProviderClass *ispc =INTERRUPT_STATS_PROVIDER_CLASS (oc); mc->desc ="IBM PowerNV (Non-Virtualized)" ; mc->init = pnv_init; mc->reset = pnv_reset; mc->max_cpus =MAX_CPUS ; - mc->default_cpu_type =POWERPC_CPU_TYPE_NAME ("power8_v2.0" ); mc->block_default_type =IF_IDE ;/* Pnv provides a AHCI device for * storage */ mc->no_parallel =1 ; mc->default_boot_order =NULL ; mc->default_ram_size =2 * GiB; - xic->icp_get = pnv_icp_get; - xic->ics_get = pnv_ics_get; - xic->ics_resend = pnv_ics_resend; ispc->print_info = pnv_pic_print_info;pnv_machine_class_props_init (oc);@@ -1386,10 +1425,27 @@ static void pnv_machine_class_init(ObjectClass *oc, void *data) .parent = TYPE_PNV9_CHIP, \ } +#define DEFINE_PNV_MACHINE_TYPE (cpu, class_initfn) \ + { \ + .name = MACHINE_TYPE_NAME(cpu), \ + .parent = TYPE_PNV_MACHINE, \ + .instance_size = sizeof(PnvMachineState), \ + .instance_init = pnv_machine_instance_init, \ + .class_init = class_initfn, \ + .interfaces = (InterfaceInfo[]) { \ + { TYPE_XICS_FABRIC }, \ + { TYPE_INTERRUPT_STATS_PROVIDER }, \ + { }, \ + }, \ + } +static const TypeInfo types[] = { +DEFINE_PNV_MACHINE_TYPE ("powernv8" , pnv_machine_power8_class_init), +DEFINE_PNV_MACHINE_TYPE ("powernv9" , pnv_machine_power9_class_init), { .name =TYPE_PNV_MACHINE , .parent =TYPE_MACHINE , + .abstract = true, .instance_size =sizeof (PnvMachineState), .instance_init = pnv_machine_instance_init, .class_init = pnv_machine_class_init,
Signature · 4 lines
David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
PATCH v3 2/18 tests/boot-serial-test: add support for all the PowerNV machines
In reply to Cédric Le GoaterUse the machine names specifiying the CPU type, POWER8 and POWER9.
--- tests/boot-serial-test.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
tests/boot-serial-test.c +2 -1
@@ -103,7 +103,8 @@ static testdef_t tests[] = { {"ppc64" ,"pseries" ,"-machine cap-cfpc=broken,cap-sbbc=broken,cap-ibs=broken" , "Open Firmware" }, - { "ppc64", "powernv", "-cpu POWER8", "OPAL" }, + { "ppc64", "powernv8", "", "OPAL" }, + { "ppc64", "powernv9", "", "OPAL" }, { "ppc64", "sam460ex", "-device e1000", "8086 100e" }, { "i386", "isapc", "-cpu qemu32 -device sga", "SGABIOS" }, { "i386", "pc", "-device sga", "SGABIOS" },
Signature · 1 line
2.21.0
David Gibson
AppliedRe: [Qemu-devel] [PATCH v3 02/18] tests/boot-serial-test: add support for all the PowerNV machines
Quoted 3 lines · Cédric Le Goater
Use the machine names specifiying the CPU type, POWER8 and POWER9.
Signed-off-by Cédric Le Goater <[email protected]>
Applied to ppc-for-4.2, thanks.
Quoted 18 lines
--- tests/boot-serial-test.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)tests/boot-serial-test.c +2
-1@@ -103,7 +103,8 @@ static testdef_t tests[] = { {"ppc64" ,"pseries" ,"-machine cap-cfpc=broken,cap-sbbc=broken,cap-ibs=broken" , "Open Firmware" }, - { "ppc64", "powernv", "-cpu POWER8", "OPAL" }, + { "ppc64", "powernv8", "", "OPAL" }, + { "ppc64", "powernv9", "", "OPAL" }, { "ppc64", "sam460ex", "-device e1000", "8086 100e" }, { "i386", "isapc", "-cpu qemu32 -device sga", "SGABIOS" }, { "i386", "pc", "-device sga", "SGABIOS" },
Signature · 4 lines
David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
PATCH v3 3/18 ppc/xive: Introduce the XiveFabric and XivePresenter interfaces
In reply to Cédric Le GoaterWhen the XIVE IVRE sub-engine (XiveRouter) looks for a Notification Virtual Target (NVT) to notify, it broadcasts a message on the PowerBUS to find an XIVE IVPE sub-engine (Presenter) which has the NVT dispatched on one of its HW threads, and then forwards the notification if any response was received.
The current XIVE presenter model is sufficient for the pseries machine because it has a single interrupt controller device, but the PowerNV machine can have multiple chips each having its own interrupt controller. In this case, the XIVE presenter model is too simple and the CAM line matching should scan all chips of the system.
We introduce a XiveFabric QOM interface which needs to be implemented by the machine. It acts as the PowerBUS interface between the interrupt controller and the system. On HW, the XIVE sub-engine responsible for the communication with the other chip is the Common Queue (CQ) bridge unit. This interface offers a 'match_nvt' handler to perform the CAM line matching when looking for a XIVE Presenter with a dispatched NVT.
We also introduce a XivePresenter QOM interface to represent the XIVE Presenter (PC) sub-engine of the XIVE controller with a similar 'match_nvt' handler to perform the CAM line matching. This interface could be merged in the XiveRouter but as it is stateless and performs the XIVE IVPE sub-engine tasks, it makes sense to keep it independent. It should have its use in POWER10.
--- include/hw/ppc/xive.h | 54 +++++++++++++++++++++++++++++++++++++++++++ hw/intc/xive.c | 36 +++++++++++++++++++++-------- 2 files changed, 81 insertions(+), 9 deletions(-)
include/hw/ppc/xive.h +54 -0
@@ -368,6 +368,60 @@ int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,XiveTCTX *xive_router_get_tctx (XiveRouter *xrtr,CPUState *cs);void xive_router_notify (XiveNotifier *xn,uint32_t lisn); +/* +* XIVE Presenter +*/ + +typedef struct XiveTCTXMatch { +XiveTCTX *tctx; +uint8_t ring; +}XiveTCTXMatch ; + +typedef struct XivePresenter XivePresenter ; + +#define TYPE_XIVE_PRESENTER "xive-presenter" +#define XIVE_PRESENTER (obj) \ + INTERFACE_CHECK(XivePresenter, (obj), TYPE_XIVE_PRESENTER) +#define XIVE_PRESENTER_CLASS (klass) \ + OBJECT_CLASS_CHECK(XivePresenterClass, (klass), TYPE_XIVE_PRESENTER) +#define XIVE_PRESENTER_GET_CLASS (obj) \ + OBJECT_GET_CLASS(XivePresenterClass, (obj), TYPE_XIVE_PRESENTER) + +typedef struct XivePresenterClass { +InterfaceClass parent; +int (*match_nvt)(XivePresenter *xptr,uint8_t format, +uint8_t nvt_blk,uint32_t nvt_idx, +bool cam_ignore,uint8_t priority, +uint32_t logic_serv,XiveTCTXMatch *match); +}XivePresenterClass ; + +int xive_presenter_tctx_match (XivePresenter *xptr,XiveTCTX *tctx, +uint8_t format, +uint8_t nvt_blk,uint32_t nvt_idx, +bool cam_ignore,uint32_t logic_serv); + +/* +* XIVE Fabric (Interface between Interrupt Controller and Machine) +*/ + +typedef struct XiveFabric XiveFabric ; + +#define TYPE_XIVE_FABRIC "xive-fabric" +#define XIVE_FABRIC (obj) \ + INTERFACE_CHECK(XiveFabric, (obj), TYPE_XIVE_FABRIC) +#define XIVE_FABRIC_CLASS (klass) \ + OBJECT_CLASS_CHECK(XiveFabricClass, (klass), TYPE_XIVE_FABRIC) +#define XIVE_FABRIC_GET_CLASS (obj) \ + OBJECT_GET_CLASS(XiveFabricClass, (obj), TYPE_XIVE_FABRIC) + +typedef struct XiveFabricClass { +InterfaceClass parent; +int (*match_nvt)(XiveFabric *xfb,uint8_t format, +uint8_t nvt_blk,uint32_t nvt_idx, +bool cam_ignore,uint8_t priority, +uint32_t logic_serv,XiveTCTXMatch *match); +}XiveFabricClass ; +/* * XIVE END ESBs */
hw/intc/xive.c +27 -9
@@ -1294,9 +1294,10 @@ static uint32_t xive_tctx_hw_cam_line(XiveTCTX *tctx)/* * The thread context register words are in big-endian format. */ -static int xive_presenter_tctx_match (XiveTCTX *tctx,uint8_t format, -uint8_t nvt_blk,uint32_t nvt_idx, -bool cam_ignore,uint32_t logic_serv) +int xive_presenter_tctx_match (XivePresenter *xptr,XiveTCTX *tctx, +uint8_t format, +uint8_t nvt_blk,uint32_t nvt_idx, +bool cam_ignore,uint32_t logic_serv) {uint32_t cam =xive_nvt_cam_line (nvt_blk, nvt_idx);uint32_t qw3w2 =xive_tctx_word2 (&tctx->regs[TM_QW3_HV_PHYS ]);
@@ -1353,11 +1354,6 @@ static int xive_presenter_tctx_match(XiveTCTX *tctx, uint8_t format,return -1 ; } -typedef struct XiveTCTXMatch { -XiveTCTX *tctx; -uint8_t ring; -}XiveTCTXMatch ; - static bool xive_presenter_match(XiveRouter *xrtr, uint8_t format, uint8_t nvt_blk, uint32_t nvt_idx, bool cam_ignore, uint8_t priority,
@@ -1383,7 +1379,8 @@ static bool xive_presenter_match(XiveRouter *xrtr, uint8_t format,* Check the thread context CAM lines and record matches. We * will handle CPU exception delivery later */ - ring =xive_presenter_tctx_match (tctx, format, nvt_blk, nvt_idx, + ring =xive_presenter_tctx_match (XIVE_PRESENTER (xrtr), tctx, format, + nvt_blk, nvt_idx, cam_ignore, logic_serv);/* * Save the context and follow on to catch duplicates, that we
@@ -1670,6 +1667,7 @@ static const TypeInfo xive_router_info = { .class_init = xive_router_class_init, .interfaces = (InterfaceInfo []) { {TYPE_XIVE_NOTIFIER }, + {TYPE_XIVE_PRESENTER }, { } } };
@@ -1841,10 +1839,30 @@ static const TypeInfo xive_notifier_info = { .class_size = sizeof(XiveNotifierClass), }; +/* +* XIVE Presenter +*/ +static const TypeInfo xive_presenter_info = { + .name =TYPE_XIVE_PRESENTER , + .parent =TYPE_INTERFACE , + .class_size =sizeof (XivePresenterClass), +}; + +/* +* XIVE Fabric +*/ +static const TypeInfo xive_fabric_info = { + .name =TYPE_XIVE_FABRIC , + .parent =TYPE_INTERFACE , + .class_size =sizeof (XiveFabricClass), +}; +static void xive_register_types (void ) {type_register_static (&xive_source_info);type_register_static (&xive_notifier_info); +type_register_static (&xive_presenter_info); +type_register_static (&xive_fabric_info);type_register_static (&xive_router_info);type_register_static (&xive_end_source_info);type_register_static (&xive_tctx_info);
Signature · 1 line
2.21.0
PATCH v3 4/18 ppc/pnv: Implement the XiveFabric and XivePresenter interfaces
In reply to Cédric Le GoaterThe CAM line matching on the PowerNV machine now scans all chips of the system and all CPUs of a chip to find a dispatched NVT in the thread contexts.
As there is now easy way to loop on the CPUs belonging to a chip, the PowerNV handler loops on all CPUs and filter out the external CPUs.
--- hw/intc/pnv_xive.c | 76 ++++++++++++++++++++++++++++++++++++++++++++++ hw/ppc/pnv.c | 32 +++++++++++++++++++ 2 files changed, 108 insertions(+)
hw/intc/pnv_xive.c +76 -0
@@ -390,6 +390,80 @@ static int pnv_xive_get_eas(XiveRouter *xrtr, uint8_t blk, uint32_t idx,return pnv_xive_vst_read (xive,VST_TSEL_IVT , blk, idx, eas); } +static int cpu_pir (PowerPCCPU *cpu) +{ +CPUPPCState *env = &cpu->env; +return env->spr_cb[SPR_PIR ].default_value; +} + +static int cpu_chip_id (PowerPCCPU *cpu) +{ +int pir =cpu_pir (cpu); +return (pir >>8 ) &0x7f ; +} + +static bool pnv_xive_is_cpu_enabled (PnvXive *xive,PowerPCCPU *cpu) +{ +int pir =cpu_pir (cpu); +int thrd_id = pir &0x7f ; + +return xive->regs[PC_THREAD_EN_REG0 >>3 ] &PPC_BIT (thrd_id); +} + +static bool pnv_xive_is_ignored (PnvChip *chip,CPUState *cs) +{ +return chip->chip_id !=cpu_chip_id (POWERPC_CPU (cs)); +} + +#define PNV_CHIP_CPU_FOREACH (chip, cs) \ + CPU_FOREACH(cs) \ + if (pnv_xive_is_ignored(chip, cs)) {} else + +static int pnv_xive_match_nvt (XivePresenter *xptr,uint8_t format, +uint8_t nvt_blk,uint32_t nvt_idx, +bool cam_ignore,uint8_t priority, +uint32_t logic_serv,XiveTCTXMatch *match) +{ +PnvXive *xive =PNV_XIVE (xptr); +CPUState *cs; +int count =0 ; + +/* +* Loop on all CPUs of the machine and filter out the CPUs +* belonging to another chip. +*/ + PNV_CHIP_CPU_FOREACH(xive->chip, cs) { +PowerPCCPU *cpu =POWERPC_CPU (cs); +XiveTCTX *tctx =XIVE_TCTX (pnv_cpu_state (cpu)->intc); +int ring; + +if (!pnv_xive_is_cpu_enabled (xive, cpu)) { +continue ; + } + + ring =xive_presenter_tctx_match (xptr, tctx, format, nvt_blk, nvt_idx, + cam_ignore, logic_serv); +/* +* Save the context and follow on to catch duplicates, that we +* don't support yet. +*/ +if (ring !=-1 ) { +if (match->tctx) { +qemu_log_mask (LOG_GUEST_ERROR ,"XIVE: already found a " +"thread context NVT %x/%x\n" , + nvt_blk, nvt_idx); +return -1 ; + } + + match->ring = ring; + match->tctx = tctx; + count++; + } + } + +return count; +} +static XiveTCTX *pnv_xive_get_tctx (XiveRouter *xrtr,CPUState *cs) {PowerPCCPU *cpu =POWERPC_CPU (cs);
@@ -1795,6 +1869,7 @@ static void pnv_xive_class_init(ObjectClass *klass, void *data)PnvXScomInterfaceClass *xdc =PNV_XSCOM_INTERFACE_CLASS (klass);XiveRouterClass *xrc =XIVE_ROUTER_CLASS (klass);XiveNotifierClass *xnc =XIVE_NOTIFIER_CLASS (klass); +XivePresenterClass *xpc =XIVE_PRESENTER_CLASS (klass); xdc->dt_xscom = pnv_xive_dt_xscom;
@@ -1810,6 +1885,7 @@ static void pnv_xive_class_init(ObjectClass *klass, void *data) xrc->get_tctx = pnv_xive_get_tctx; xnc->notify = pnv_xive_notify; + xpc->match_nvt = pnv_xive_match_nvt; };static const TypeInfo pnv_xive_info = {
hw/ppc/pnv.c +32 -0
@@ -1322,6 +1322,35 @@ static void pnv_pic_print_info(InterruptStatsProvider *obj, } } +static int pnv_xive_match_nvt (XiveFabric *xfb,uint8_t format, +uint8_t nvt_blk,uint32_t nvt_idx, +bool cam_ignore,uint8_t priority, +uint32_t logic_serv, +XiveTCTXMatch *match) +{ +PnvMachineState *pnv =PNV_MACHINE (xfb); +int total_count =0 ; +int i; + +for (i =0 ; i < pnv->num_chips; i++) { +Pnv9Chip *chip9 =PNV9_CHIP (pnv->chips[i]); +XivePresenter *xptr =XIVE_PRESENTER (&chip9->xive); +XivePresenterClass *xpc =XIVE_PRESENTER_GET_CLASS (xptr); +int count; + + count = xpc->match_nvt (xptr, format, nvt_blk, nvt_idx, cam_ignore, + priority, logic_serv, match); + +if (count <0 ) { +return count; + } + + total_count += count; + } + +return total_count; +} +static void pnv_get_num_chips (Object *obj,Visitor *v,const char *name,void *opaque,Error **errp) {
@@ -1385,9 +1414,11 @@ static void pnv_machine_power8_class_init(ObjectClass *oc, void *data)static void pnv_machine_power9_class_init (ObjectClass *oc,void *data) {MachineClass *mc =MACHINE_CLASS (oc); +XiveFabricClass *xfc =XIVE_FABRIC_CLASS (oc); mc->desc ="IBM PowerNV (Non-Virtualized) POWER9" ; mc->default_cpu_type =POWERPC_CPU_TYPE_NAME ("power9_v2.0" ); + xfc->match_nvt = pnv_xive_match_nvt; mc->alias ="powernv" ; }
@@ -1435,6 +1466,7 @@ static void pnv_machine_class_init(ObjectClass *oc, void *data) .interfaces = (InterfaceInfo[]) { \ {TYPE_XICS_FABRIC }, \ {TYPE_INTERRUPT_STATS_PROVIDER }, \ + {TYPE_XIVE_FABRIC }, \ { }, \ }, \ }
Signature · 1 line
2.21.0
Re: [Qemu-devel] [PATCH v3 04/18] ppc/pnv: Implement the XiveFabric and XivePresenter interfaces
Quoted 40 lines · Cédric Le Goater
The CAM line matching on the PowerNV machine now scans all chips of the system and all CPUs of a chip to find a dispatched NVT in the thread contexts.
As there is now easy way to loop on the CPUs belonging to a chip, the PowerNV handler loops on all CPUs and filter out the external CPUs.
Fixes af53dbf6227a ("ppc/xive: introduce a simplified XIVE presenter")Signed-off-by Cédric Le Goater <[email protected]>--- hw/intc/pnv_xive.c | 76 ++++++++++++++++++++++++++++++++++++++++++++++ hw/ppc/pnv.c | 32 +++++++++++++++++++ 2 files changed, 108 insertions(+)hw/intc/pnv_xive.c +18
-0@@ -390,6 +390,80 @@ static int pnv_xive_get_eas(XiveRouter *xrtr, uint8_t blk, uint32_t idx,return pnv_xive_vst_read (xive,VST_TSEL_IVT , blk, idx, eas); } +static int cpu_pir (PowerPCCPU *cpu) +{ +CPUPPCState *env = &cpu->env; +return env->spr_cb[SPR_PIR ].default_value; +} + +static int cpu_chip_id (PowerPCCPU *cpu) +{ +int pir =cpu_pir (cpu); +return (pir >>8 ) &0x7f ; +} + +static bool pnv_xive_is_cpu_enabled (PnvXive *xive,PowerPCCPU *cpu) +{ +int pir =cpu_pir (cpu); +int thrd_id = pir &0x7f ; + +return xive->regs[PC_THREAD_EN_REG0 >>3 ] &PPC_BIT (thrd_id);
hw/intc/pnv_xive.c static int pnv_xive_get_eas(XiveRouter *xrtr, uint8_t blk, uint32_t idx, · line 420
Quoted 10 lines
hw/intc/pnv_xive.c +10
-0+} + +static bool pnv_xive_is_ignored (PnvChip *chip,CPUState *cs) +{ +return chip->chip_id !=cpu_chip_id (POWERPC_CPU (cs)); +} + +#define PNV_CHIP_CPU_FOREACH (chip, cs) \ + CPU_FOREACH(cs) \ + if (pnv_xive_is_ignored(chip, cs)) {} else
patchew will complain here but I think we can consider it as an exception since it is inspired from the macros in migration/ram.c ?
Thanks,
C.
PATCH v3 5/18 ppc/spapr: Implement the XiveFabric and XivePresenter interfaces
In reply to Cédric Le GoaterThe CAM line matching sequence in the pseries machine does not change much apart from the use of the new QOM interfaces.
--- include/hw/ppc/spapr_irq.h | 6 ++++++ hw/intc/spapr_xive.c | 41 ++++++++++++++++++++++++++++++++++++++ hw/ppc/spapr.c | 34 +++++++++++++++++++++++++++++++ hw/ppc/spapr_irq.c | 25 +++++++++++++++++++++++ 4 files changed, 106 insertions(+)
include/hw/ppc/spapr_irq.h +6 -0
@@ -30,6 +30,8 @@ int spapr_irq_msi_alloc(SpaprMachineState *spapr, uint32_t num, bool align,void spapr_irq_msi_free (SpaprMachineState *spapr,int irq,uint32_t num);void spapr_irq_msi_reset (SpaprMachineState *spapr); +struct XiveTCTXMatch ; + typedef struct SpaprIrq { uint32_t nr_irqs; uint32_t nr_msis;
@@ -49,6 +51,10 @@ typedef struct SpaprIrq {void (*set_irq)(void *opaque,int srcno,int val);const char *(*get_nodename)(SpaprMachineState *spapr);void (*init_kvm)(SpaprMachineState *spapr,Error **errp); +int (*match_nvt)(SpaprMachineState *spapr,uint8_t format, +uint8_t nvt_blk,uint32_t nvt_idx, +bool cam_ignore,uint8_t priority, +uint32_t logic_serv,struct XiveTCTXMatch *match); } SpaprIrq;extern SpaprIrq spapr_irq_xics;
hw/intc/spapr_xive.c +41 -0
59 lines not on this page · raw message
hw/ppc/spapr.c +34 -0
@@ -4302,6 +4302,37 @@ static void spapr_pic_print_info(InterruptStatsProvider *obj, spapr->irq->print_info (spapr, mon); } +static int spapr_xive_match_nvt (XiveFabric *xfb,uint8_t format, +uint8_t nvt_blk,uint32_t nvt_idx, +bool cam_ignore,uint8_t priority, +uint32_t logic_serv,XiveTCTXMatch *match) +{ +SpaprMachineState *spapr =SPAPR_MACHINE (xfb); +int count; + + count = spapr->irq->match_nvt (spapr, format, nvt_blk, nvt_idx, cam_ignore, + priority, logic_serv, match); +if (count <0 ) { +return count; + } + +/* +* When we implement the save and restore of the thread interrupt +* contexts in the enter/exit CPU handlers of the machine and the +* escalations in QEMU, we should be able to handle non dispatched +* vCPUs. +* +* Until this is done, the sPAPR machine should find at least one +* matching context always. +*/ +if (count ==0 ) { +qemu_log_mask (LOG_GUEST_ERROR ,"XIVE: NVT %x/%x is not dispatched\n" , + nvt_blk, nvt_idx); + } + +return count; +} +int spapr_get_vcpu_id (PowerPCCPU *cpu) {return cpu->vcpu_id;
@@ -4398,6 +4429,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data)PPCVirtualHypervisorClass *vhc =PPC_VIRTUAL_HYPERVISOR_CLASS (oc);XICSFabricClass *xic =XICS_FABRIC_CLASS (oc);InterruptStatsProviderClass *ispc =INTERRUPT_STATS_PROVIDER_CLASS (oc); +XiveFabricClass *xfc =XIVE_FABRIC_CLASS (oc); mc->desc ="pSeries Logical Partition (PAPR compliant)" ; mc->ignore_boot_device_suffixes = true;
@@ -4456,6 +4488,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data)*/ mc->numa_mem_align_shift =28 ; mc->numa_mem_supported = true; + xfc->match_nvt = spapr_xive_match_nvt; smc->default_caps.caps[SPAPR_CAP_HTM ] =SPAPR_CAP_OFF ; smc->default_caps.caps[SPAPR_CAP_VSX ] =SPAPR_CAP_ON ;
@@ -4488,6 +4521,7 @@ static const TypeInfo spapr_machine_info = { { TYPE_PPC_VIRTUAL_HYPERVISOR }, { TYPE_XICS_FABRIC }, { TYPE_INTERRUPT_STATS_PROVIDER }, + { TYPE_XIVE_FABRIC }, { } }, };
hw/ppc/spapr_irq.c +25 -0
@@ -262,6 +262,7 @@ SpaprIrq spapr_irq_xics = { .set_irq = spapr_irq_set_irq_xics, .get_nodename = spapr_irq_get_nodename_xics, .init_kvm = spapr_irq_init_kvm_xics, + .match_nvt = NULL, /* should not be used */ };/*
@@ -411,6 +412,18 @@ static void spapr_irq_init_kvm_xive(SpaprMachineState *spapr, Error **errp) } } +static int spapr_irq_match_nvt_xive (SpaprMachineState *spapr,uint8_t format, +uint8_t nvt_blk,uint32_t nvt_idx, +bool cam_ignore,uint8_t priority, +uint32_t logic_serv,XiveTCTXMatch *match) +{ +XivePresenter *xptr =XIVE_PRESENTER (spapr->xive); +XivePresenterClass *xpc =XIVE_PRESENTER_GET_CLASS (xptr); + +return xpc->match_nvt (xptr, format, nvt_blk, nvt_idx, cam_ignore, + priority, logic_serv, match); +} +/* * XIVE uses the full IRQ number space. Set it to 8K to be compatible * with XICS.
@@ -436,6 +449,7 @@ SpaprIrq spapr_irq_xive = { .set_irq = spapr_irq_set_irq_xive, .get_nodename = spapr_irq_get_nodename_xive, .init_kvm = spapr_irq_init_kvm_xive, + .match_nvt = spapr_irq_match_nvt_xive, };/*
@@ -590,6 +604,15 @@ static const char *spapr_irq_get_nodename_dual(SpaprMachineState *spapr)return spapr_irq_current (spapr)->get_nodename (spapr); } +static int spapr_irq_match_nvt_dual (SpaprMachineState *spapr,uint8_t format, +uint8_t nvt_blk,uint32_t nvt_idx, +bool cam_ignore,uint8_t priority, +uint32_t logic_serv,XiveTCTXMatch *match) +{ +return spapr_irq_current (spapr)->match_nvt (spapr, format, nvt_blk, nvt_idx, + cam_ignore, priority, logic_serv, match); +} +/* * Define values in sync with the XIVE and XICS backend */
@@ -613,6 +636,7 @@ SpaprIrq spapr_irq_dual = { .set_irq = spapr_irq_set_irq_dual, .get_nodename = spapr_irq_get_nodename_dual, .init_kvm = NULL, /* should not be used */ + .match_nvt = spapr_irq_match_nvt_dual, };
@@ -828,4 +852,5 @@ SpaprIrq spapr_irq_xics_legacy = { .set_irq = spapr_irq_set_irq_xics, .get_nodename = spapr_irq_get_nodename_xics, .init_kvm = spapr_irq_init_kvm_xics, + .match_nvt = NULL, /* should not be used */ };
Signature · 1 line
2.21.0
PATCH v3 6/18 ppc/xive: Use the XiveFabric and XivePresenter interfaces
In reply to Cédric Le GoaterNow that the machines have handlers implemented for the XiveFabric and XivePresenter interfaces, remove xive_presenter_match() and make use of the 'match_nvt' handler of the machine.
--- hw/intc/xive.c | 77 +++++++++++--------------------------------------- 1 file changed, 17 insertions(+), 60 deletions(-)
hw/intc/xive.c +17 -60
@@ -1354,59 +1354,6 @@ int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,return -1 ; } -static bool xive_presenter_match (XiveRouter *xrtr,uint8_t format, -uint8_t nvt_blk,uint32_t nvt_idx, -bool cam_ignore,uint8_t priority, -uint32_t logic_serv,XiveTCTXMatch *match) -{ -CPUState *cs; - -/* -* TODO (PowerNV): handle chip_id overwrite of block field for -* hardwired CAM compares -*/ - -CPU_FOREACH (cs) { -XiveTCTX *tctx =xive_router_get_tctx (xrtr, cs); -int ring; - -/* -* HW checks that the CPU is enabled in the Physical Thread -* Enable Register (PTER). -*/ - -/* -* Check the thread context CAM lines and record matches. We -* will handle CPU exception delivery later -*/ - ring =xive_presenter_tctx_match (XIVE_PRESENTER (xrtr), tctx, format, - nvt_blk, nvt_idx, - cam_ignore, logic_serv); -/* -* Save the context and follow on to catch duplicates, that we -* don't support yet. -*/ -if (ring !=-1 ) { -if (match->tctx) { -qemu_log_mask (LOG_GUEST_ERROR ,"XIVE: already found a thread " -"context NVT %x/%x\n" , nvt_blk, nvt_idx); -return false; - } - - match->ring = ring; - match->tctx = tctx; - } - } - -if (!match->tctx) { -qemu_log_mask (LOG_UNIMP ,"XIVE: NVT %x/%x is not dispatched\n" , - nvt_blk, nvt_idx); -return false; - } - -return true; -} -/* * This is our simple Xive Presenter Engine model. It is merged in the * Router as it does not require an extra object.
@@ -1422,22 +1369,32 @@ static bool xive_presenter_match(XiveRouter *xrtr, uint8_t format,* * The parameters represent what is sent on the PowerBus */ -static bool xive_presenter_notify (XiveRouter *xrtr,uint8_t format, +static bool xive_presenter_notify (uint8_t format,uint8_t nvt_blk,uint32_t nvt_idx,bool cam_ignore,uint8_t priority,uint32_t logic_serv) { +XiveFabric *xfb =XIVE_FABRIC (qdev_get_machine ()); +XiveFabricClass *xfc =XIVE_FABRIC_GET_CLASS (xfb);XiveTCTXMatch match = { .tctx =NULL , .ring =0 }; -bool found; +int count; - found =xive_presenter_match (xrtr, format, nvt_blk, nvt_idx, cam_ignore, - priority, logic_serv, &match); -if (found) { +/* +* Ask the machine to scan the interrupt controllers for a match +*/ + count = xfc->match_nvt (xfb, format, nvt_blk, nvt_idx, cam_ignore, + priority, logic_serv, &match); +if (count <0 ) { +return false; + } + +/* handle CPU exception delivery */ +if (count) {ipb_update (&match.tctx->regs[match.ring], priority);xive_tctx_notify (match.tctx, match.ring); } -return found; +return count; }/*
@@ -1550,7 +1507,7 @@ static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,return ; } - found =xive_presenter_notify (xrtr, format, nvt_blk, nvt_idx, + found =xive_presenter_notify (format, nvt_blk, nvt_idx,xive_get_field32 (END_W7_F0_IGNORE , end.w7), priority,xive_get_field32 (END_W7_F1_LOG_SERVER_ID , end.w7));
Signature · 1 line
2.21.0
PATCH v3 7/18 ppc/xive: Extend the TIMA operation with a XivePresenter parameter
In reply to Cédric Le GoaterThe TIMA operations are performed on behalf of the XIVE IVPE sub-engine (Presenter) on the thread interrupt context registers. The current operations the model supports are simple and do not require access to the controller but more complex operations we will add will need access to the controller NVT table and to its configuration.
--- include/hw/ppc/xive.h | 7 +++--- hw/intc/pnv_xive.c | 4 +-- hw/intc/xive.c | 58 ++++++++++++++++++++++++------------------- 3 files changed, 38 insertions(+), 31 deletions(-)
include/hw/ppc/xive.h +4 -3
@@ -464,9 +464,10 @@ typedef struct XiveENDSource {#define XIVE_TM_USER_PAGE 0x3extern const MemoryRegionOps xive_tm_ops; -void xive_tctx_tm_write (XiveTCTX *tctx,hwaddr offset,uint64_t value, -unsigned size); -uint64_t xive_tctx_tm_read (XiveTCTX *tctx,hwaddr offset,unsigned size); +void xive_tctx_tm_write (XivePresenter *xptr,XiveTCTX *tctx,hwaddr offset, +uint64_t value,unsigned size); +uint64_t xive_tctx_tm_read (XivePresenter *xptr,XiveTCTX *tctx,hwaddr offset, +unsigned size);void xive_tctx_pic_print_info (XiveTCTX *tctx,Monitor *mon);Object *xive_tctx_create (Object *cpu,XiveRouter *xrtr,Error **errp);
hw/intc/pnv_xive.c +2 -2
@@ -1444,7 +1444,7 @@ static void xive_tm_indirect_write(void *opaque, hwaddr offset, {XiveTCTX *tctx =pnv_xive_get_indirect_tctx (PNV_XIVE (opaque)); -xive_tctx_tm_write (tctx, offset, value, size); +xive_tctx_tm_write (XIVE_PRESENTER (opaque), tctx, offset, value, size); } static uint64_t xive_tm_indirect_read(void *opaque, hwaddr offset,
@@ -1452,7 +1452,7 @@ static uint64_t xive_tm_indirect_read(void *opaque, hwaddr offset, {XiveTCTX *tctx =pnv_xive_get_indirect_tctx (PNV_XIVE (opaque)); -return xive_tctx_tm_read (tctx, offset, size); +return xive_tctx_tm_read (XIVE_PRESENTER (opaque), tctx, offset, size); }static const MemoryRegionOps xive_tm_indirect_ops = {
hw/intc/xive.c +32 -26
158 lines not on this page · raw message
Signature · 1 line
2.21.0
PATCH v3 8/18 ppc/pnv: Clarify how the TIMA is accessed on a multichip system
In reply to Cédric Le GoaterThe TIMA MMIO space is shared among the chips and to identify the chip from which the access is being done, the PowerBUS uses a chip field in the load/store messages. QEMU does not model these messages, so instead, we extract the chip id from the CPU PIR and do a lookup at the machine level to fetch the targeted interrupt controller.
Introduce pnv_get_chip() and pnv_xive_tm_get_xive() helpers to clarify this process in pnv_xive_get_tctx(). The latter will be removed in the subsequent patches but the same principle will be kept.
--- include/hw/ppc/pnv.h | 13 +++++++++++++ hw/intc/pnv_xive.c | 40 +++++++++++++++++++++++----------------- 2 files changed, 36 insertions(+), 17 deletions(-)
include/hw/ppc/pnv.h +13 -0
@@ -182,6 +182,19 @@ static inline bool pnv_is_power9(PnvMachineState *pnv)return pnv_chip_is_power9 (pnv->chips[0 ]); } +static inline PnvChip *pnv_get_chip (PnvMachineState *pnv,uint32_t chip_id) +{ +int i; + +for (i =0 ; i < pnv->num_chips; i++) { +PnvChip *chip = pnv->chips[i]; +if (chip->chip_id == chip_id) { +return chip; + } + } +return NULL ; +} +#define PNV_FDT_ADDR 0x01000000#define PNV_TIMEBASE_FREQ 512000000ULL
hw/intc/pnv_xive.c +23 -17
@@ -464,31 +464,37 @@ static int pnv_xive_match_nvt(XivePresenter *xptr, uint8_t format,return count; } +/* +* The TIMA MMIO space is shared among the chips and to identify the +* chip from which the access is being done, we extract the chip id +* from the PIR. +*/ +static PnvXive *pnv_xive_tm_get_xive (PowerPCCPU *cpu) +{ +PnvMachineState *pnv =PNV_MACHINE (qdev_get_machine ()); +PnvChip *chip; +PnvXive *xive; + + chip =pnv_get_chip (pnv,cpu_chip_id (cpu)); +assert (chip); + xive = &PNV9_CHIP (chip)->xive; + +if (!pnv_xive_is_cpu_enabled (xive, cpu)) { +xive_error (xive,"IC: CPU %x is not enabled" ,cpu_pir (cpu)); + } +return xive; +} +static XiveTCTX *pnv_xive_get_tctx (XiveRouter *xrtr,CPUState *cs) {PowerPCCPU *cpu =POWERPC_CPU (cs); -XiveTCTX *tctx =XIVE_TCTX (pnv_cpu_state (cpu)->intc); -PnvXive *xive =NULL ; -CPUPPCState *env = &cpu->env; -int pir = env->spr_cb[SPR_PIR ].default_value; +PnvXive *xive =pnv_xive_tm_get_xive (cpu); -/* -* Perform an extra check on the HW thread enablement. -* -* The TIMA is shared among the chips and to identify the chip -* from which the access is being done, we extract the chip id -* from the PIR. -*/ - xive =pnv_xive_get_ic ((pir >>8 ) &0xf );if (!xive) {return NULL ; } -if (!(xive->regs[PC_THREAD_EN_REG0 >>3 ] &PPC_BIT (pir &0x3f ))) { -xive_error (PNV_XIVE (xrtr),"IC: CPU %x is not enabled" , pir); - } - -return tctx; +return XIVE_TCTX (pnv_cpu_state (cpu)->intc); }/*
Signature · 1 line
2.21.0
PATCH v3 9/18 ppc/xive: Move the TIMA operations to the controller model
In reply to Cédric Le GoaterThis also removes the need of the get_tctx() XiveRouter handler in the core XIVE framework.
--- include/hw/ppc/xive.h | 3 --- hw/intc/pnv_xive.c | 48 ++++++++++++++++++++++++++++++------------- hw/intc/spapr_xive.c | 41 +++++++++++++++++++++++++++--------- hw/intc/xive.c | 36 -------------------------------- 4 files changed, 65 insertions(+), 63 deletions(-)
include/hw/ppc/xive.h +0 -3
@@ -352,7 +352,6 @@ typedef struct XiveRouterClass { XiveNVT *nvt);int (*write_nvt)(XiveRouter *xrtr,uint8_t nvt_blk,uint32_t nvt_idx,XiveNVT *nvt,uint8_t word_number); - XiveTCTX *(*get_tctx)(XiveRouter *xrtr, CPUState *cs); } XiveRouterClass; int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
@@ -365,7 +364,6 @@ int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt);int xive_router_write_nvt (XiveRouter *xrtr,uint8_t nvt_blk,uint32_t nvt_idx,XiveNVT *nvt,uint8_t word_number); -XiveTCTX *xive_router_get_tctx (XiveRouter *xrtr,CPUState *cs);void xive_router_notify (XiveNotifier *xn,uint32_t lisn);/*
@@ -463,7 +461,6 @@ typedef struct XiveENDSource {#define XIVE_TM_OS_PAGE 0x2#define XIVE_TM_USER_PAGE 0x3 -extern const MemoryRegionOps xive_tm_ops;void xive_tctx_tm_write (XivePresenter *xptr,XiveTCTX *tctx,hwaddr offset,uint64_t value,unsigned size); uint64_t xive_tctx_tm_read(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
hw/intc/pnv_xive.c +34 -14
72 lines not on this page · raw message
hw/intc/spapr_xive.c +31 -10
65 lines not on this page · raw message
hw/intc/xive.c +0 -36
@@ -480,35 +480,6 @@ uint64_t xive_tctx_tm_read(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,return xive_tm_raw_read (tctx, offset, size); } -static void xive_tm_write (void *opaque,hwaddr offset, -uint64_t value,unsigned size) -{ -XiveTCTX *tctx =xive_router_get_tctx (XIVE_ROUTER (opaque), current_cpu); - -xive_tctx_tm_write (XIVE_PRESENTER (opaque), tctx, offset, value, size); -} - -static uint64_t xive_tm_read (void *opaque,hwaddr offset,unsigned size) -{ -XiveTCTX *tctx =xive_router_get_tctx (XIVE_ROUTER (opaque), current_cpu); - -return xive_tctx_tm_read (XIVE_PRESENTER (opaque), tctx, offset, size); -} - -const MemoryRegionOps xive_tm_ops = { - .read = xive_tm_read, - .write = xive_tm_write, - .endianness =DEVICE_BIG_ENDIAN , - .valid = { - .min_access_size =1 , - .max_access_size =8 , - }, - .impl = { - .min_access_size =1 , - .max_access_size =8 , - }, -}; -static char *xive_tctx_ring_print (uint8_t *ring) {uint32_t w2 =xive_tctx_word2 (ring);
@@ -1277,13 +1248,6 @@ int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,return xrc->write_nvt (xrtr, nvt_blk, nvt_idx, nvt, word_number); } -XiveTCTX *xive_router_get_tctx (XiveRouter *xrtr,CPUState *cs) -{ -XiveRouterClass *xrc =XIVE_ROUTER_GET_CLASS (xrtr); - -return xrc->get_tctx (xrtr, cs); -} -/* * Encode the HW CAM line in the block group mode format : *
Signature · 1 line
2.21.0
PATCH v3 10/18 ppc/xive: Introduce a xive_tctx_ipb_update() helper
In reply to Cédric Le GoaterWe will use it to resend missed interrupts when a vCPU context is pushed a HW thread.
--- include/hw/ppc/xive.h | 1 + hw/intc/xive.c | 15 +++++++++++---- 2 files changed, 12 insertions(+), 4 deletions(-)
include/hw/ppc/xive.h +1 -0
@@ -468,6 +468,7 @@ uint64_t xive_tctx_tm_read(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,void xive_tctx_pic_print_info (XiveTCTX *tctx,Monitor *mon);Object *xive_tctx_create (Object *cpu,XiveRouter *xrtr,Error **errp); +void xive_tctx_ipb_update (XiveTCTX *tctx,uint8_t ring,uint8_t ipb);static inline uint32_t xive_nvt_cam_line (uint8_t nvt_blk,uint32_t nvt_idx) {
hw/intc/xive.c +11 -4
@@ -132,6 +132,15 @@ static void xive_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)xive_tctx_notify (tctx, ring); } +void xive_tctx_ipb_update (XiveTCTX *tctx,uint8_t ring,uint8_t ipb) +{ +uint8_t *regs = &tctx->regs[ring]; + + regs[TM_IPB ] |= ipb; + regs[TM_PIPR ] =ipb_to_pipr (regs[TM_IPB ]); +xive_tctx_notify (tctx, ring); +} +static inline uint32_t xive_tctx_word2 (uint8_t *ring) {return *((uint32_t *) &ring[TM_WORD2 ]);
@@ -333,8 +342,7 @@ static void xive_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,static void xive_tm_set_os_pending (XivePresenter *xptr,XiveTCTX *tctx,hwaddr offset,uint64_t value,unsigned size) { -ipb_update (&tctx->regs[TM_QW1_OS ], value &0xff ); -xive_tctx_notify (tctx,TM_QW1_OS ); +xive_tctx_ipb_update (tctx,TM_QW1_OS ,priority_to_ipb (value &0xff )); } static uint64_t xive_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
@@ -1360,8 +1368,7 @@ static bool xive_presenter_notify(uint8_t format,/* handle CPU exception delivery */ if (count) { -ipb_update (&match.tctx->regs[match.ring], priority); -xive_tctx_notify (match.tctx, match.ring); +xive_tctx_ipb_update (match.tctx, match.ring,priority_to_ipb (priority)); }return count;
Signature · 1 line
2.21.0
PATCH v3 11/18 ppc/xive: Synthesize interrupt from the saved IPB in the NVT
In reply to Cédric Le GoaterWhen an interrupt can not be presented to a vCPU, the XIVE presenter updates the Interrupt Pending Buffer of the XIVE NVT if backlog is activated in the END.
Later, when the same vCPU is dispatched, its context is pushed in the thread context registers and the VO bit is set in the CAM line word to activate the context. The HW grabs the associated NVT to pull the pending bits, and merges them with the IPB of the TIMA. If interrupts were missed while the vCPU was not dispatched, these are synthesized in this sequence.
--- include/hw/ppc/xive.h | 5 --- include/hw/ppc/xive_regs.h | 22 +++++++++++++ hw/intc/xive.c | 67 +++++++++++++++++++++++++++++++++----- 3 files changed, 81 insertions(+), 13 deletions(-)
include/hw/ppc/xive.h +0 -5
@@ -470,11 +470,6 @@ void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon);Object *xive_tctx_create (Object *cpu,XiveRouter *xrtr,Error **errp);void xive_tctx_ipb_update (XiveTCTX *tctx,uint8_t ring,uint8_t ipb); -static inline uint32_t xive_nvt_cam_line (uint8_t nvt_blk,uint32_t nvt_idx) -{ -return (nvt_blk <<19 ) | nvt_idx; -} -/* * KVM XIVE device helpers */
include/hw/ppc/xive_regs.h +22 -0
@@ -232,6 +232,7 @@ typedef struct XiveNVT {uint32_t w2;uint32_t w3;uint32_t w4; +#define NVT_W4_IPB PPC_BITMASK32(16, 23)uint32_t w5;uint32_t w6;uint32_t w7;
@@ -248,4 +249,25 @@ typedef struct XiveNVT {#define xive_nvt_is_valid (nvt) (be32_to_cpu((nvt)->w0) & NVT_W0_VALID) +/* +* The VP number space in a block is defined by the END_W6_NVT_INDEX +* field of the XIVE END +*/ +#define XIVE_NVT_SHIFT 19 + +static inline uint32_t xive_nvt_cam_line (uint8_t nvt_blk,uint32_t nvt_idx) +{ +return (nvt_blk <<XIVE_NVT_SHIFT ) | nvt_idx; +} + +static inline uint32_t xive_nvt_idx (uint32_t cam_line) +{ +return cam_line & ((1 <<XIVE_NVT_SHIFT ) -1 ); +} + +static inline uint32_t xive_nvt_blk (uint32_t cam_line) +{ +return (cam_line >>XIVE_NVT_SHIFT ) &0xf ; +} +#endif /* PPC_XIVE_REGS_H */
hw/intc/xive.c +59 -8
97 lines not on this page · raw message
Signature · 1 line
2.21.0
PATCH v3 12/18 ppc/pnv: Remove pnv_xive_vst_size() routine
In reply to Cédric Le Goaterpnv_xive_vst_size() tries to compute the size of a VSD table from the information given by FW. The number of entries of the table are deduced from the result and the MMIO regions of the ESBs and the ENDS are also resized with the computed value.
But for indirect tables, the result is incorrect. An indirect table is a one page array of VSDs pointing to subpages containing XIVE virtual structures. The number of first level VSD entries is page aligned and the overall computed size of the table is too large. It can also be completely wrong when the first VSD entry is not yet initialized.
Remove pnv_xive_vst_size() and use a simpler form for direct tables. This is only useful when outputting the XIVE sources on the monitor and to resize the ESB MMIO window.
--- hw/intc/pnv_xive.c | 112 +++++++++++++++++---------------------------- 1 file changed, 43 insertions(+), 69 deletions(-)
hw/intc/pnv_xive.c +43 -69
@@ -121,36 +121,22 @@ static uint64_t pnv_xive_vst_page_size_allowed(uint32_t page_shift) page_shift ==21 || page_shift ==24 ; } -static uint64_t pnv_xive_vst_size (uint64_t vsd) -{ -uint64_t vst_tsize =1ull << (GETFIELD (VSD_TSIZE , vsd) +12 ); - -/* -* Read the first descriptor to get the page size of the indirect -* table. -*/ -if (VSD_INDIRECT & vsd) { -uint32_t nr_pages = vst_tsize /XIVE_VSD_SIZE ; -uint32_t page_shift; - - vsd =ldq_be_dma (&address_space_memory, vsd &VSD_ADDRESS_MASK ); - page_shift =GETFIELD (VSD_TSIZE , vsd) +12 ; - -if (!pnv_xive_vst_page_size_allowed (page_shift)) { -return 0 ; - } - -return nr_pages * (1ull << page_shift); - } - -return vst_tsize; -} -static uint64_t pnv_xive_vst_addr_direct (PnvXive *xive,uint32_t type,uint64_t vsd,uint32_t idx) {const XiveVstInfo *info = &vst_infos[type];uint64_t vst_addr = vsd &VSD_ADDRESS_MASK ; +uint64_t vst_tsize =1ull << (GETFIELD (VSD_TSIZE , vsd) +12 ); +uint32_t idx_max; + + idx_max = vst_tsize / info->size -1 ; +if (idx > idx_max) { +#ifdef XIVE_DEBUG +xive_error (xive,"VST: %s entry %x out of range [ 0 .. %x ] !?" , + info->name, idx, idx_max); +#endif +return 0 ; + }return vst_addr + idx * info->size; }
@@ -213,7 +199,6 @@ static uint64_t pnv_xive_vst_addr(PnvXive *xive, uint32_t type, uint8_t blk, {const XiveVstInfo *info = &vst_infos[type];uint64_t vsd; -uint32_t idx_max;if (blk >= info->max_blocks) { xive_error(xive, "VST: invalid block id %d for VST %s %d !?",
@@ -230,15 +215,6 @@ static uint64_t pnv_xive_vst_addr(PnvXive *xive, uint32_t type, uint8_t blk,return xive ?pnv_xive_vst_addr (xive, type, blk, idx) :0 ; } - idx_max =pnv_xive_vst_size (vsd) / info->size -1 ; -if (idx > idx_max) { -#ifdef XIVE_DEBUG -xive_error (xive,"VST: %s entry %x/%x out of range [ 0 .. %x ] !?" , - info->name, blk, idx, idx_max); -#endif -return 0 ; - } -if (VSD_INDIRECT & vsd) {return pnv_xive_vst_addr_indirect (xive, type, vsd, idx); }
@@ -519,19 +495,12 @@ static uint64_t pnv_xive_pc_size(PnvXive *xive)return (~xive->regs[CQ_PC_BARM >>3 ] +1 ) &CQ_PC_BARM_MASK ; } -static uint32_t pnv_xive_nr_ipis (PnvXive *xive) +static uint32_t pnv_xive_nr_ipis (PnvXive *xive,uint8_t blk) { -uint8_t blk = xive->chip->chip_id; - -return pnv_xive_vst_size (xive->vsds[VST_TSEL_SBE ][blk]) *SBE_PER_BYTE ; -} - -static uint32_t pnv_xive_nr_ends (PnvXive *xive) -{ -uint8_t blk = xive->chip->chip_id; +uint64_t vsd = xive->vsds[VST_TSEL_SBE ][blk]; +uint64_t vst_tsize =1ull << (GETFIELD (VSD_TSIZE , vsd) +12 ); -return pnv_xive_vst_size (xive->vsds[VST_TSEL_EQDT ][blk]) - / vst_infos[VST_TSEL_EQDT ].size; +return VSD_INDIRECT & vsd ?0 : vst_tsize *SBE_PER_BYTE ; }/*
@@ -664,6 +633,7 @@ static void pnv_xive_vst_set_exclusive(PnvXive *xive, uint8_t type,XiveSource *xsrc = &xive->ipi_source;const XiveVstInfo *info = &vst_infos[type];uint32_t page_shift =GETFIELD (VSD_TSIZE , vsd) +12 ; +uint64_t vst_tsize =1ull << page_shift;uint64_t vst_addr = vsd &VSD_ADDRESS_MASK ;/* Basic checks */
@@ -699,11 +669,16 @@ static void pnv_xive_vst_set_exclusive(PnvXive *xive, uint8_t type,case VST_TSEL_EQDT :/* -* Backing store pages for the END. Compute the number of ENDs -* provisioned by FW and resize the END ESB window accordingly. +* Backing store pages for the END. +* +* If the table is direct, we can compute the number of PQ +* entries provisioned by FW (such as skiboot) and resize the +* END ESB window accordingly. */ -memory_region_set_size (&end_xsrc->esb_mmio,pnv_xive_nr_ends (xive) * - (1ull << (end_xsrc->esb_shift +1 ))); +if (!(VSD_INDIRECT & vsd)) { +memory_region_set_size (&end_xsrc->esb_mmio, (vst_tsize / info->size) + * (1ull << xsrc->esb_shift)); + }memory_region_add_subregion (&xive->end_edt_mmio,0 , &end_xsrc->esb_mmio);break ;
@@ -712,11 +687,16 @@ static void pnv_xive_vst_set_exclusive(PnvXive *xive, uint8_t type,/* * Backing store pages for the source PQ bits. The model does * not use these PQ bits backed in RAM because the XiveSource -* model has its own. Compute the number of IRQs provisioned -* by FW and resize the IPI ESB window accordingly. +* model has its own. +* +* If the table is direct, we can compute the number of PQ +* entries provisioned by FW (such as skiboot) and resize the +* ESB window accordingly. */ -memory_region_set_size (&xsrc->esb_mmio,pnv_xive_nr_ipis (xive) * - (1ull << xsrc->esb_shift)); +if (!(VSD_INDIRECT & vsd)) { +memory_region_set_size (&xsrc->esb_mmio, vst_tsize *SBE_PER_BYTE + * (1ull << xsrc->esb_shift)); + }memory_region_add_subregion (&xive->ipi_edt_mmio,0 , &xsrc->esb_mmio);break ;
@@ -1666,8 +1646,7 @@ void pnv_xive_pic_print_info(PnvXive *xive, Monitor *mon)XiveRouter *xrtr =XIVE_ROUTER (xive);uint8_t blk = xive->chip->chip_id;uint32_t srcno0 =XIVE_SRCNO (blk,0 ); -uint32_t nr_ipis =pnv_xive_nr_ipis (xive); -uint32_t nr_ends =pnv_xive_nr_ends (xive); +uint32_t nr_ipis =pnv_xive_nr_ipis (xive, blk);XiveEAS eas;XiveEND end;int i;
@@ -1687,21 +1666,16 @@ void pnv_xive_pic_print_info(PnvXive *xive, Monitor *mon) } } -monitor_printf (mon,"XIVE[%x] ENDT %08x .. %08x\n" , blk,0 , nr_ends -1 ); -for (i =0 ; i < nr_ends; i++) { -if (xive_router_get_end (xrtr, blk, i, &end)) { -break ; - } -xive_end_pic_print_info (&end, i, mon); +monitor_printf (mon,"XIVE[%x] ENDT\n" , blk); + i =0 ; +while (!xive_router_get_end (xrtr, blk, i, &end)) { +xive_end_pic_print_info (&end, i++, mon); } -monitor_printf (mon,"XIVE[%x] END Escalation %08x .. %08x\n" , blk,0 , - nr_ends -1 ); -for (i =0 ; i < nr_ends; i++) { -if (xive_router_get_end (xrtr, blk, i, &end)) { -break ; - } -xive_end_eas_pic_print_info (&end, i, mon); +monitor_printf (mon,"XIVE[%x] END Escalation EAT\n" , blk); + i =0 ; +while (!xive_router_get_end (xrtr, blk, i, &end)) { +xive_end_eas_pic_print_info (&end, i++, mon); } }
Signature · 1 line
2.21.0
PATCH v3 13/18 ppc/pnv: Dump the XIVE NVT table
In reply to Cédric Le GoaterThis is to track the configuration of the base END index of the vCPU and the Interrupt Pending Buffer. The NVT IPB is updated when an interrupt can not be presented to a vCPU.
--- include/hw/ppc/xive_regs.h | 2 ++ hw/intc/pnv_xive.c | 22 ++++++++++++++++++++++ 2 files changed, 24 insertions(+)
include/hw/ppc/xive_regs.h +2 -0
@@ -229,6 +229,8 @@ typedef struct XiveNVT {uint32_t w0;#define NVT_W0_VALID PPC_BIT32(0)uint32_t w1; +#define NVT_W1_EQ_BLOCK PPC_BITMASK32(0, 3) +#define NVT_W1_EQ_INDEX PPC_BITMASK32(4, 31)uint32_t w2;uint32_t w3;uint32_t w4;
hw/intc/pnv_xive.c +22 -0
@@ -1641,6 +1641,21 @@ static const MemoryRegionOps pnv_xive_pc_ops = { }, }; +static void xive_nvt_pic_print_info (XiveNVT *nvt,uint32_t nvt_idx, +Monitor *mon) +{ +uint8_t eq_blk =xive_get_field32 (NVT_W1_EQ_BLOCK , nvt->w1); +uint32_t eq_idx =xive_get_field32 (NVT_W1_EQ_INDEX , nvt->w1); + +if (!xive_nvt_is_valid (nvt)) { +return ; + } + +monitor_printf (mon," %08x end:%02x/%04x IPB:%02x\n" , nvt_idx, + eq_blk, eq_idx, +xive_get_field32 (NVT_W4_IPB , nvt->w4)); +} +void pnv_xive_pic_print_info (PnvXive *xive,Monitor *mon) {XiveRouter *xrtr =XIVE_ROUTER (xive);
@@ -1649,6 +1664,7 @@ void pnv_xive_pic_print_info(PnvXive *xive, Monitor *mon)uint32_t nr_ipis =pnv_xive_nr_ipis (xive, blk);XiveEAS eas;XiveEND end; +XiveNVT nvt;int i; monitor_printf(mon, "XIVE[%x] Source %08x .. %08x\n", blk, srcno0,
@@ -1677,6 +1693,12 @@ void pnv_xive_pic_print_info(PnvXive *xive, Monitor *mon)while (!xive_router_get_end (xrtr, blk, i, &end)) {xive_end_eas_pic_print_info (&end, i++, mon); } + +monitor_printf (mon,"XIVE[%x] NVTT\n" , blk); + i =0 ; +while (!xive_router_get_nvt (xrtr, blk, i, &nvt)) { +xive_nvt_pic_print_info (&nvt, i++, mon); + } }static void pnv_xive_reset (void *dev)
Signature · 1 line
2.21.0
PATCH v3 14/18 ppc/pnv: Skip empty slots of the XIVE NVT table
In reply to Cédric Le GoaterWe try to loop on the full table skipping empty indirect pages which are not necessarily allocated. This is useful to dump the contexts of the KVM vCPUs.
--- hw/intc/pnv_xive.c | 16 ++++++++++++---- 1 file changed, 12 insertions(+), 4 deletions(-)
hw/intc/pnv_xive.c +12 -4
@@ -1641,6 +1641,12 @@ static const MemoryRegionOps pnv_xive_pc_ops = { }, }; +/* +* skiboot uses an indirect NVT table with 64k subpages +*/ +#define XIVE_NVT_COUNT (1 << XIVE_NVT_SHIFT) +#define XIVE_NVT_PER_PAGE (0x10000 / sizeof(XiveNVT)) +static void xive_nvt_pic_print_info (XiveNVT *nvt,uint32_t nvt_idx,Monitor *mon) {
@@ -1694,10 +1700,12 @@ void pnv_xive_pic_print_info(PnvXive *xive, Monitor *mon)xive_end_eas_pic_print_info (&end, i++, mon); } -monitor_printf (mon,"XIVE[%x] NVTT\n" , blk); - i =0 ; -while (!xive_router_get_nvt (xrtr, blk, i, &nvt)) { -xive_nvt_pic_print_info (&nvt, i++, mon); +monitor_printf (mon,"XIVE[%x] NVTT %08x .. %08x\n" , blk,0 , +XIVE_NVT_COUNT -1 ); +for (i =0 ; i <XIVE_NVT_COUNT ; i +=XIVE_NVT_PER_PAGE ) { +while (!xive_router_get_nvt (xrtr, blk, i, &nvt)) { +xive_nvt_pic_print_info (&nvt, i++, mon); + } } }
Signature · 1 line
2.21.0
PATCH v3 15/18 ppc/pnv: Introduce a pnv_xive_block_id() helper
In reply to Cédric Le GoaterWhen PC_TCTXT_CHIPID_OVERRIDE is configured, the PC_TCTXT_CHIPID field overrides the hardwired chip ID in the Powerbus operations and for CAM compares. This is typically used in the one block-per-chip configuration to associate a unique block id number to each IC of the system.
Simplify the model with a pnv_xive_block_id() helper and remove 'tctx_chipid' which becomes useless.
The model does support multiple blocks per chip.
--- include/hw/ppc/pnv_xive.h | 3 -- hw/intc/pnv_xive.c | 68 ++++++++++++++++++++------------------- 2 files changed, 35 insertions(+), 36 deletions(-)
include/hw/ppc/pnv_xive.h +0 -3
@@ -72,9 +72,6 @@ typedef struct PnvXive {/* Interrupt controller registers */ uint64_t regs[0x300 ]; -/* Can be configured by FW */ -uint32_t tctx_chipid; -/* * Virtual Structure Descriptor tables : EAT, SBE, ENDT, NVTT, IRQ * These are in a SRAM protected by ECC.
hw/intc/pnv_xive.c +35 -33
147 lines not on this page · raw message
Signature · 1 line
2.21.0
PATCH v3 16/18 ppc/pnv: Extend XivePresenter with a get_block_id() handler
In reply to Cédric Le GoaterWhen doing CAM line compares, fetch the block id from the interrupt controller which can have set the PC_TCTXT_CHIPID field.
--- include/hw/ppc/xive.h | 1 + hw/intc/pnv_xive.c | 6 ++++++ hw/intc/spapr_xive.c | 6 ++++++ hw/intc/xive.c | 8 +++++--- 4 files changed, 18 insertions(+), 3 deletions(-)
include/hw/ppc/xive.h +1 -0
@@ -391,6 +391,7 @@ typedef struct XivePresenterClass { uint8_t nvt_blk, uint32_t nvt_idx, bool cam_ignore, uint8_t priority, uint32_t logic_serv, XiveTCTXMatch *match); +uint8_t (*get_block_id)(XivePresenter *xptr); } XivePresenterClass; int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
hw/intc/pnv_xive.c +6 -0
@@ -460,6 +460,11 @@ static int pnv_xive_match_nvt(XivePresenter *xptr, uint8_t format,return count; } +static uint8_t pnv_xive_get_block_id (XivePresenter *xptr) +{ +return pnv_xive_block_id (PNV_XIVE (xptr)); +} +/* * The TIMA MMIO space is shared among the chips and to identify the * chip from which the access is being done, we extract the chip id
@@ -1918,6 +1923,7 @@ static void pnv_xive_class_init(ObjectClass *klass, void *data) xnc->notify = pnv_xive_notify; xpc->match_nvt = pnv_xive_match_nvt; + xpc->get_block_id = pnv_xive_get_block_id; };static const TypeInfo pnv_xive_info = {
hw/intc/spapr_xive.c +6 -0
@@ -479,6 +479,11 @@ static int spapr_xive_match_nvt(XivePresenter *xptr, uint8_t format,return count; } +static uint8_t spapr_xive_get_block_id (XivePresenter *xrtr) +{ +return SPAPR_XIVE_BLOCK_ID ; +} +static const VMStateDescription vmstate_spapr_xive_end = { .name =TYPE_SPAPR_XIVE "/end" , .version_id =1 ,
@@ -570,6 +575,7 @@ static void spapr_xive_class_init(ObjectClass *klass, void *data) xrc->write_nvt = spapr_xive_write_nvt; xpc->match_nvt = spapr_xive_match_nvt; + xpc->get_block_id = spapr_xive_get_block_id; }static const TypeInfo spapr_xive_info = {
hw/intc/xive.c +5 -3
@@ -1305,12 +1305,14 @@ int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,* * chip << 19 | 0000000 0 0001 thread (7Bit) */ -static uint32_t xive_tctx_hw_cam_line (XiveTCTX *tctx) +static uint32_t xive_tctx_hw_cam_line (XivePresenter *xptr,XiveTCTX *tctx) {CPUPPCState *env = &POWERPC_CPU (tctx->cs)->env;uint32_t pir = env->spr_cb[SPR_PIR ].default_value; +XivePresenterClass *xpc =XIVE_PRESENTER_GET_CLASS (xptr); +uint8_t blk = xpc->get_block_id (xptr); -return xive_nvt_cam_line ((pir >>8 ) &0xf ,1 <<7 | (pir &0x7f )); +return xive_nvt_cam_line (blk,1 <<7 | (pir &0x7f )); }/*
@@ -1347,7 +1349,7 @@ int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,/* PHYS ring */ if ((be32_to_cpu (qw3w2) &TM_QW3W2_VT ) && - cam ==xive_tctx_hw_cam_line (tctx)) { + cam ==xive_tctx_hw_cam_line (xptr, tctx)) {return TM_QW3_HV_PHYS ; }
Signature · 1 line
2.21.0
Re: [Qemu-devel] [PATCH v3 16/18] ppc/pnv: Extend XivePresenter with a get_block_id() handler
Hello David,
Quoted 2 lines · Cédric Le Goater
When doing CAM line compares, fetch the block id from the interrupt controller which can have set the
PC_TCTXT_CHIPIDfield.
I don't know if you had time to review the rest of this patchset.
I would prefer to move the get_block_id() in XiveRouter, we could get rid of the 'block_id' field in the XiveEndSource.
Thanks,
C.
Quoted 100 lines
Signed-off-by Cédric Le Goater <[email protected]>--- include/hw/ppc/xive.h | 1 + hw/intc/pnv_xive.c | 6 ++++++ hw/intc/spapr_xive.c | 6 ++++++ hw/intc/xive.c | 8 +++++--- 4 files changed, 18 insertions(+), 3 deletions(-)include/hw/ppc/xive.h +1
-0@@ -391,6 +391,7 @@ typedef struct XivePresenterClass { uint8_t nvt_blk, uint32_t nvt_idx, bool cam_ignore, uint8_t priority, uint32_t logic_serv, XiveTCTXMatch *match); +uint8_t (*get_block_id)(XivePresenter *xptr); } XivePresenterClass; int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,hw/intc/pnv_xive.c +6
-0@@ -460,6 +460,11 @@ static int pnv_xive_match_nvt(XivePresenter *xptr, uint8_t format,return count; } +static uint8_t pnv_xive_get_block_id (XivePresenter *xptr) +{ +return pnv_xive_block_id (PNV_XIVE (xptr)); +} +/* * The TIMA MMIO space is shared among the chips and to identify the * chip from which the access is being done, we extract the chip id @@ -1918,6 +1923,7 @@ static void pnv_xive_class_init(ObjectClass *klass, void *data) xnc->notify = pnv_xive_notify; xpc->match_nvt = pnv_xive_match_nvt; + xpc->get_block_id = pnv_xive_get_block_id; };static const TypeInfo pnv_xive_info = {hw/intc/spapr_xive.c +6
-0@@ -479,6 +479,11 @@ static int spapr_xive_match_nvt(XivePresenter *xptr, uint8_t format,return count; } +static uint8_t spapr_xive_get_block_id (XivePresenter *xrtr) +{ +return SPAPR_XIVE_BLOCK_ID ; +} +static const VMStateDescription vmstate_spapr_xive_end = { .name =TYPE_SPAPR_XIVE "/end" , .version_id =1 ,@@ -570,6 +575,7 @@ static void spapr_xive_class_init(ObjectClass *klass, void *data) xrc->write_nvt = spapr_xive_write_nvt; xpc->match_nvt = spapr_xive_match_nvt; + xpc->get_block_id = spapr_xive_get_block_id; }static const TypeInfo spapr_xive_info = {hw/intc/xive.c +5
-3@@ -1305,12 +1305,14 @@ int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,* * chip << 19 | 0000000 0 0001 thread (7Bit) */ -static uint32_t xive_tctx_hw_cam_line (XiveTCTX *tctx) +static uint32_t xive_tctx_hw_cam_line (XivePresenter *xptr,XiveTCTX *tctx) {CPUPPCState *env = &POWERPC_CPU (tctx->cs)->env;uint32_t pir = env->spr_cb[SPR_PIR ].default_value; +XivePresenterClass *xpc =XIVE_PRESENTER_GET_CLASS (xptr); +uint8_t blk = xpc->get_block_id (xptr); -return xive_nvt_cam_line ((pir >>8 ) &0xf ,1 <<7 | (pir &0x7f )); +return xive_nvt_cam_line (blk,1 <<7 | (pir &0x7f )); }/* @@ -1347,7 +1349,7 @@ int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,/* PHYS ring */ if ((be32_to_cpu (qw3w2) &TM_QW3W2_VT ) && - cam ==xive_tctx_hw_cam_line (tctx)) { + cam ==xive_tctx_hw_cam_line (xptr, tctx)) {return TM_QW3_HV_PHYS ; }
PATCH v3 17/18 ppc/pnv: Quiesce some XIVE errors
In reply to Cédric Le GoaterWhen dumping the END and NVT tables, the error logging is too noisy.
--- hw/intc/pnv_xive.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
hw/intc/pnv_xive.c +5 -1
@@ -27,7 +27,7 @@#include "pnv_xive_regs.h" -#define XIVE_DEBUG +#undef XIVE_DEBUG/* * Virtual structures table (VST)
@@ -172,7 +172,9 @@ static uint64_t pnv_xive_vst_addr_indirect(PnvXive *xive, uint32_t type, vsd =ldq_be_dma (&address_space_memory, vsd_addr);if (!(vsd &VSD_ADDRESS_MASK )) { +#ifdef XIVE_DEBUG xive_error (xive,"VST: invalid %s entry %x !?" , info->name, idx); +#endif return 0 ; }
@@ -193,7 +195,9 @@ static uint64_t pnv_xive_vst_addr_indirect(PnvXive *xive, uint32_t type, vsd =ldq_be_dma (&address_space_memory, vsd_addr);if (!(vsd &VSD_ADDRESS_MASK )) { +#ifdef XIVE_DEBUG xive_error (xive,"VST: invalid %s entry %x !?" , info->name, idx); +#endif return 0 ; }
Signature · 1 line
2.21.0
PATCH v3 18/18 ppc/xive: Introduce a xive_os_cam_decode() helper
In reply to Cédric Le GoaterThe OS CAM line has a special encoding exploited by the HW. Provide a helper routine to hide the details to the TIMA command handlers. This also clarifies the endianness of different variables : 'qw1w2' is big-endian and 'cam' is native.
Also add a check on the V bit in the TM_PULL_POOL_CTX special operation.
--- hw/intc/xive.c | 40 +++++++++++++++++++++++++++++++--------- 1 file changed, 31 insertions(+), 9 deletions(-)
hw/intc/xive.c +31 -9
@@ -339,14 +339,34 @@ static void xive_tm_set_os_pending(XivePresenter *xptr, XiveTCTX *tctx,xive_tctx_ipb_update (tctx,TM_QW1_OS ,priority_to_ipb (value &0xff )); } +static void xive_os_cam_decode (uint32_t cam,uint8_t *nvt_blk, +uint32_t *nvt_idx,bool *vo) +{ + *nvt_blk =xive_nvt_blk (cam); + *nvt_idx =xive_nvt_idx (cam); + *vo = !!(cam &TM_QW1W2_VO ); +} +static uint64_t xive_tm_pull_os_ctx (XivePresenter *xptr,XiveTCTX *tctx,hwaddr offset,unsigned size) { -uint32_t qw1w2_prev =xive_tctx_word2 (&tctx->regs[TM_QW1_OS ]); -uint32_t qw1w2; +uint32_t qw1w2 =xive_tctx_word2 (&tctx->regs[TM_QW1_OS ]); +uint32_t qw1w2_new; +uint32_t cam =be32_to_cpu (qw1w2); +uint8_t nvt_blk; +uint32_t nvt_idx; +bool vo; - qw1w2 =xive_set_field32 (TM_QW1W2_VO , qw1w2_prev,0 ); -memcpy (&tctx->regs[TM_QW1_OS +TM_WORD2 ], &qw1w2,4 ); +xive_os_cam_decode (cam, &nvt_blk, &nvt_idx, &vo); + +if (!vo) { +qemu_log_mask (LOG_GUEST_ERROR ,"XIVE: pulling invalid NVT %x/%x !?\n" , + nvt_blk, nvt_idx); + } + +/* Invalidate CAM line */ + qw1w2_new =xive_set_field32 (TM_QW1W2_VO , qw1w2,0 ); +memcpy (&tctx->regs[TM_QW1_OS +TM_WORD2 ], &qw1w2_new,4 );return qw1w2; }
@@ -384,13 +404,15 @@ static void xive_tctx_need_resend(XiveRouter *xrtr, XiveTCTX *tctx,static void xive_tm_push_os_ctx (XivePresenter *xptr,XiveTCTX *tctx,hwaddr offset,uint64_t value,unsigned size) { -uint32_t qw1w2 = value; -uint8_t nvt_blk =xive_nvt_blk (qw1w2); -uint32_t nvt_idx =xive_nvt_idx (qw1w2); -bool vo = !!(qw1w2 &TM_QW1W2_VO ); +uint32_t cam = value; +uint32_t qw1w2 =cpu_to_be32 (cam); +uint8_t nvt_blk; +uint32_t nvt_idx; +bool vo; + +xive_os_cam_decode (cam, &nvt_blk, &nvt_idx, &vo);/* First update the registers */ - qw1w2 =cpu_to_be32 (qw1w2);memcpy (&tctx->regs[TM_QW1_OS +TM_WORD2 ], &qw1w2,4 );/* Check the interrupt pending bits */
Signature · 1 line
2.21.0
Quoted 14 lines · Cédric Le Goater
Hello,
The QEMU PowerNV machine emulates a baremetal OpenPOWER system and acts as an hypervisor (L0). Supporting emulation of KVM to run guests (L1) requires a few more extensions, among which guest support for the XIVE interrupt controller on POWER9 processor.
The following changes add new per-CPU PowerNV machines and extend the XIVE models with the new XiveFabric and XivePresenter interfaces to provide support for XIVE escalations and interrupt resend. This mechanism is used by XIVE to notify the hypervisor that a vCPU is not dispatched on a HW thread. Tested on a QEMU PowerNV machine and a simple QEMU pseries guest doing network on a local bridge.
Quoted 7 lines
The XIVE interrupt controller offers a way to increase the XIVE resources per chip by configuring multiple XIVE blocks on a chip. This is not currently supported by the model. However, some configurations, such as OPAL/skiboot, use one block-per-chip configuration with some optimizations. One of them is to override the hardwired chip ID by the block id in the PowerBUS operations and for CAM line compares. This patchset improves the support for this setup. Tested with 4 chips.
David,
Do you want me to resend this patchset ? or you just didn't have time to look at it ?
Patch 16 has changed a little since. The get_block_id() handler has moved to the XiveRouterClass.
Thanks,
Thanks,
C.
Quoted 23 lines · Cédric Le Goater
Quoted 14 lines · Cédric Le Goater
Hello,
The QEMU PowerNV machine emulates a baremetal OpenPOWER system and acts as an hypervisor (L0). Supporting emulation of KVM to run guests (L1) requires a few more extensions, among which guest support for the XIVE interrupt controller on POWER9 processor.
The following changes add new per-CPU PowerNV machines and extend the XIVE models with the new XiveFabric and XivePresenter interfaces to provide support for XIVE escalations and interrupt resend. This mechanism is used by XIVE to notify the hypervisor that a vCPU is not dispatched on a HW thread. Tested on a QEMU PowerNV machine and a simple QEMU pseries guest doing network on a local bridge.
Quoted 7 lines
The XIVE interrupt controller offers a way to increase the XIVE resources per chip by configuring multiple XIVE blocks on a chip. This is not currently supported by the model. However, some configurations, such as OPAL/skiboot, use one block-per-chip configuration with some optimizations. One of them is to override the hardwired chip ID by the block id in the PowerBUS operations and for CAM line compares. This patchset improves the support for this setup. Tested with 4 chips.
Quoted 4 lines
David,
Do you want me to resend this patchset ? or you just didn't have time to look at it ?
Mostly, I just haven't had time. I'm also finding the patches pretty difficult to read and review. I don't think that's an indication they're bad, just that what they're doing is necessarily complex, but it's still made it hard to tackle them.
Quoted 2 lines
Patch 16 has changed a little since. The
get_block_id()handler has moved to the XiveRouterClass.
You, might as well repost, so I'm looking at the latest stuff. I can't promise I'll be able to look at the new set terribly soon though.
Signature · 4 lines
David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
Quoted 26 lines · David Gibson
Quoted 25 lines · Cédric Le Goater
Quoted 14 lines · Cédric Le Goater
Hello,
The QEMU PowerNV machine emulates a baremetal OpenPOWER system and acts as an hypervisor (L0). Supporting emulation of KVM to run guests (L1) requires a few more extensions, among which guest support for the XIVE interrupt controller on POWER9 processor.
The following changes add new per-CPU PowerNV machines and extend the XIVE models with the new XiveFabric and XivePresenter interfaces to provide support for XIVE escalations and interrupt resend. This mechanism is used by XIVE to notify the hypervisor that a vCPU is not dispatched on a HW thread. Tested on a QEMU PowerNV machine and a simple QEMU pseries guest doing network on a local bridge.
Quoted 7 lines
The XIVE interrupt controller offers a way to increase the XIVE resources per chip by configuring multiple XIVE blocks on a chip. This is not currently supported by the model. However, some configurations, such as OPAL/skiboot, use one block-per-chip configuration with some optimizations. One of them is to override the hardwired chip ID by the block id in the PowerBUS operations and for CAM line compares. This patchset improves the support for this setup. Tested with 4 chips.
David,
Quoted 7 lines
Quoted 2 lines
Do you want me to resend this patchset ? or you just didn't have time to look at it ?
Mostly, I just haven't had time. I'm also finding the patches pretty difficult to read and review. I don't think that's an indication they're bad, just that what they're doing is necessarily complex, but it's still made it hard to tackle them.
I will try to split the initial patches on the presenter a little more.
Quoted 6 lines
Quoted 2 lines
Patch 16 has changed a little since. The
get_block_id()handler has moved to the XiveRouterClass.You, might as well repost, so I'm looking at the latest stuff. I can't promise I'll be able to look at the new set terribly soon though.
ok.
thanks,
C.