| Message ID | 1454586455-10202-7-git-send-email-imammedo@redhat.com |
|---|---|
| State | New |
| Headers | show |
On Thu, Feb 04, 2016 at 12:47:33PM +0100, Igor Mammedov wrote: > cpu->found_cpus bitmap is used for setting present > flag in CPON AML package at start up. But it takes > a bunch of code to fill bitmap and cloud be simplified > by calling qemu_get_cpu_by_arch_id(apic_id) directly. > > Hence do so and remove not used anymore bitmap > with related utilities, which saves us ~32LOC > and also would simplify consolidating APCI parts > of CPU hotplug. > > Signed-off-by: Igor Mammedov <imammedo@redhat.com> This makes the code loops through all smp_cpus CPUs max_cpus times, instead of just looping through the smp_cpus CPUs once. I'm all for making the code simpler, and there's an upper bound: looking at CPU objects 64k times (because MAX_CPUMASK_BITS is 255). But I want to make sure we agree this is an optimization we want to drop.
On Fri, 5 Feb 2016 13:39:07 -0200 Eduardo Habkost <ehabkost@redhat.com> wrote: > On Thu, Feb 04, 2016 at 12:47:33PM +0100, Igor Mammedov wrote: > > cpu->found_cpus bitmap is used for setting present > > flag in CPON AML package at start up. But it takes > > a bunch of code to fill bitmap and cloud be simplified > > by calling qemu_get_cpu_by_arch_id(apic_id) directly. > > > > Hence do so and remove not used anymore bitmap > > with related utilities, which saves us ~32LOC > > and also would simplify consolidating APCI parts > > of CPU hotplug. > > > > Signed-off-by: Igor Mammedov <imammedo@redhat.com> > > This makes the code loops through all smp_cpus CPUs max_cpus > times, instead of just looping through the smp_cpus CPUs once. Yep, that looks bad. I'll redo it using possible_cpu_arch_ids(), it is a little bit bigger refactoring. > > I'm all for making the code simpler, and there's an upper bound: > looking at CPU objects 64k times (because MAX_CPUMASK_BITS is > 255). But I want to make sure we agree this is an optimization we > want to drop. >
On Fri, 5 Feb 2016 17:19:50 +0100 Igor Mammedov <imammedo@redhat.com> wrote: > On Fri, 5 Feb 2016 13:39:07 -0200 > Eduardo Habkost <ehabkost@redhat.com> wrote: > > > On Thu, Feb 04, 2016 at 12:47:33PM +0100, Igor Mammedov wrote: > > > cpu->found_cpus bitmap is used for setting present > > > flag in CPON AML package at start up. But it takes > > > a bunch of code to fill bitmap and cloud be simplified > > > by calling qemu_get_cpu_by_arch_id(apic_id) directly. > > > > > > Hence do so and remove not used anymore bitmap > > > with related utilities, which saves us ~32LOC > > > and also would simplify consolidating APCI parts > > > of CPU hotplug. > > > > > > Signed-off-by: Igor Mammedov <imammedo@redhat.com> > > > > This makes the code loops through all smp_cpus CPUs max_cpus > > times, instead of just looping through the smp_cpus CPUs once. > Yep, that looks bad. > I'll redo it using possible_cpu_arch_ids(), it is a little > bit bigger refactoring. I think I'll make bitmap local to build_processor_devices() as it's needed only for building CPON package, that way it will go along with related legacy hotplug and won't get in the way of new hotplug. > > > > > I'm all for making the code simpler, and there's an upper bound: > > looking at CPU objects 64k times (because MAX_CPUMASK_BITS is > > 255). But I want to make sure we agree this is an optimization we > > want to drop. > > > >
On Fri, Feb 05, 2016 at 05:44:49PM +0100, Igor Mammedov wrote: > On Fri, 5 Feb 2016 17:19:50 +0100 > Igor Mammedov <imammedo@redhat.com> wrote: > > > On Fri, 5 Feb 2016 13:39:07 -0200 > > Eduardo Habkost <ehabkost@redhat.com> wrote: > > > > > On Thu, Feb 04, 2016 at 12:47:33PM +0100, Igor Mammedov wrote: > > > > cpu->found_cpus bitmap is used for setting present > > > > flag in CPON AML package at start up. But it takes > > > > a bunch of code to fill bitmap and cloud be simplified > > > > by calling qemu_get_cpu_by_arch_id(apic_id) directly. > > > > > > > > Hence do so and remove not used anymore bitmap > > > > with related utilities, which saves us ~32LOC > > > > and also would simplify consolidating APCI parts > > > > of CPU hotplug. > > > > > > > > Signed-off-by: Igor Mammedov <imammedo@redhat.com> > > > > > > This makes the code loops through all smp_cpus CPUs max_cpus > > > times, instead of just looping through the smp_cpus CPUs once. > > Yep, that looks bad. > > I'll redo it using possible_cpu_arch_ids(), it is a little > > bit bigger refactoring. > I think I'll make bitmap local to build_processor_devices() as it's > needed only for building CPON package, that way it will go along > with related legacy hotplug and won't get in the way of new hotplug. What about pc_possible_cpu_arch_ids() itself? It has exactly the same problem: it looks at CPU objects smp_cpus*max_cpus times, to fill the CPUArchId.cpu field. If that's a problem for build_processor_devices(), I assume that's a problem for all other possible_cpu_arch_ids() callers?
On Thu, 11 Feb 2016 13:59:33 -0200 Eduardo Habkost <ehabkost@redhat.com> wrote: > On Fri, Feb 05, 2016 at 05:44:49PM +0100, Igor Mammedov wrote: > > On Fri, 5 Feb 2016 17:19:50 +0100 > > Igor Mammedov <imammedo@redhat.com> wrote: > > > > > On Fri, 5 Feb 2016 13:39:07 -0200 > > > Eduardo Habkost <ehabkost@redhat.com> wrote: > > > > > > > On Thu, Feb 04, 2016 at 12:47:33PM +0100, Igor Mammedov wrote: > > > > > cpu->found_cpus bitmap is used for setting present > > > > > flag in CPON AML package at start up. But it takes > > > > > a bunch of code to fill bitmap and cloud be simplified > > > > > by calling qemu_get_cpu_by_arch_id(apic_id) directly. > > > > > > > > > > Hence do so and remove not used anymore bitmap > > > > > with related utilities, which saves us ~32LOC > > > > > and also would simplify consolidating APCI parts > > > > > of CPU hotplug. > > > > > > > > > > Signed-off-by: Igor Mammedov <imammedo@redhat.com> > > > > > > > > This makes the code loops through all smp_cpus CPUs max_cpus > > > > times, instead of just looping through the smp_cpus CPUs once. > > > Yep, that looks bad. > > > I'll redo it using possible_cpu_arch_ids(), it is a little > > > bit bigger refactoring. > > I think I'll make bitmap local to build_processor_devices() as it's > > needed only for building CPON package, that way it will go along > > with related legacy hotplug and won't get in the way of new hotplug. > > What about pc_possible_cpu_arch_ids() itself? It has exactly the > same problem: it looks at CPU objects smp_cpus*max_cpus times, to > fill the CPUArchId.cpu field. If that's a problem for > build_processor_devices(), I assume that's a problem for all > other possible_cpu_arch_ids() callers? pls see v2 https://www.mail-archive.com/qemu-devel@nongnu.org/msg351298.html
diff --git a/hw/i386/acpi-build.c b/hw/i386/acpi-build.c index 9eeeffa..921830e 100644 --- a/hw/i386/acpi-build.c +++ b/hw/i386/acpi-build.c @@ -76,10 +76,6 @@ #define ACPI_BUILD_DPRINTF(fmt, ...) #endif -typedef struct AcpiCpuInfo { - DECLARE_BITMAP(found_cpus, ACPI_CPU_HOTPLUG_ID_LIMIT); -} AcpiCpuInfo; - typedef struct AcpiMcfgInfo { uint64_t mcfg_base; uint32_t mcfg_size; @@ -121,31 +117,6 @@ typedef struct AcpiBuildPciBusHotplugState { bool pcihp_bridge_en; } AcpiBuildPciBusHotplugState; -static -int acpi_add_cpu_info(Object *o, void *opaque) -{ - AcpiCpuInfo *cpu = opaque; - uint64_t apic_id; - - if (object_dynamic_cast(o, TYPE_CPU)) { - apic_id = object_property_get_int(o, "apic-id", NULL); - assert(apic_id < ACPI_CPU_HOTPLUG_ID_LIMIT); - - set_bit(apic_id, cpu->found_cpus); - } - - object_child_foreach(o, acpi_add_cpu_info, opaque); - return 0; -} - -static void acpi_get_cpu_info(AcpiCpuInfo *cpu) -{ - Object *root = object_get_root(); - - memset(cpu->found_cpus, 0, sizeof cpu->found_cpus); - object_child_foreach(root, acpi_add_cpu_info, cpu); -} - static void acpi_get_pm_info(AcpiPmInfo *pm) { Object *piix = piix4_pm_find(); @@ -967,8 +938,7 @@ static Aml *build_crs(PCIHostState *host, } static void build_processor_devices(Aml *sb_scope, unsigned acpi_cpus, - AcpiCpuInfo *cpu, AcpiPmInfo *pm, - MachineState *machine) + AcpiPmInfo *pm, MachineState *machine) { int i; Aml *dev; @@ -1063,7 +1033,7 @@ static void build_processor_devices(Aml *sb_scope, unsigned acpi_cpus, aml_varpackage(acpi_cpus); for (i = 0; i < acpi_cpus; i++) { - uint8_t b = test_bit(i, cpu->found_cpus) ? 0x01 : 0x00; + uint8_t b = qemu_get_cpu_by_arch_id(i) ? 0x01 : 0x00; aml_append(pkg, aml_int(b)); } aml_append(sb_scope, aml_name_decl(CPU_ON_BITMAP, pkg)); @@ -1955,7 +1925,7 @@ static Aml *build_q35_osc_method(void) static void build_dsdt(GArray *table_data, GArray *linker, - AcpiCpuInfo *cpu, AcpiPmInfo *pm, AcpiMiscInfo *misc, + AcpiPmInfo *pm, AcpiMiscInfo *misc, PcPciInfo *pci, PcGuestInfo *guest_info, MachineState *machine) { @@ -2263,7 +2233,7 @@ build_dsdt(GArray *table_data, GArray *linker, sb_scope = aml_scope("\\_SB"); { - build_processor_devices(sb_scope, guest_info->apic_id_limit, cpu, pm, + build_processor_devices(sb_scope, guest_info->apic_id_limit, pm, machine); build_memory_devices(sb_scope, nr_mem, pm->mem_hp_io_base, @@ -2618,7 +2588,6 @@ void acpi_build(PcGuestInfo *guest_info, AcpiBuildTables *tables) { GArray *table_offsets; unsigned facs, dsdt, rsdt, fadt; - AcpiCpuInfo cpu; AcpiPmInfo pm; AcpiMiscInfo misc; AcpiMcfgInfo mcfg; @@ -2628,7 +2597,6 @@ void acpi_build(PcGuestInfo *guest_info, AcpiBuildTables *tables) GArray *tables_blob = tables->table_data; MachineState *machine = MACHINE(qdev_get_machine()); - acpi_get_cpu_info(&cpu); acpi_get_pm_info(&pm); acpi_get_misc_info(&misc); acpi_get_pci_info(&pci); @@ -2651,7 +2619,7 @@ void acpi_build(PcGuestInfo *guest_info, AcpiBuildTables *tables) /* DSDT is pointed to by FADT */ dsdt = tables_blob->len; - build_dsdt(tables_blob, tables->linker, &cpu, &pm, &misc, &pci, + build_dsdt(tables_blob, tables->linker, &pm, &misc, &pci, guest_info, machine); /* Count the size of the DSDT and SSDT, we will need it for legacy
cpu->found_cpus bitmap is used for setting present flag in CPON AML package at start up. But it takes a bunch of code to fill bitmap and cloud be simplified by calling qemu_get_cpu_by_arch_id(apic_id) directly. Hence do so and remove not used anymore bitmap with related utilities, which saves us ~32LOC and also would simplify consolidating APCI parts of CPU hotplug. Signed-off-by: Igor Mammedov <imammedo@redhat.com> --- hw/i386/acpi-build.c | 42 +++++------------------------------------- 1 file changed, 5 insertions(+), 37 deletions(-)