| Message ID | 20230226110802.103134-1-usama.arif@bytedance.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Parallel CPU bringup for x86_64 | expand |
Hello. On neděle 26. února 2023 12:07:51 CET Usama Arif wrote: > The main code change over v11 is the build error fix by Brian Gerst and > acquiring tr_lock in trampoline_64.S whenever the stack is setup. > > The git history is also rewritten to move the commits that removed > initial_stack, early_gdt_descr and initial_gs earlier in the patchset. > > Thanks, > Usama > > Changes across versions: > v2: Cut it back to just INIT/SIPI/SIPI in parallel for now, nothing more > v3: Clean up x2apic patch, add MTRR optimisation, lock topology update > in preparation for more parallelisation. > v4: Fixes to the real mode parallelisation patch spotted by SeanC, to > avoid scribbling on initial_gs in common_cpu_up(), and to allow all > 24 bits of the physical X2APIC ID to be used. That patch still needs > a Signed-off-by from its original author, who once claimed not to > remember writing it at all. But now we've fixed it, hopefully he'll > admit it now :) > v5: rebase to v6.1 and remeasure performance, disable parallel bringup > for AMD CPUs. > v6: rebase to v6.2-rc6, disabled parallel boot on amd as a cpu bug and > reused timer calibration for secondary CPUs. > v7: [David Woodhouse] iterate over all possible CPUs to find any existing > cluster mask in alloc_clustermask. (patch 1/9) > Keep parallel AMD support enabled in AMD, using APIC ID in CPUID leaf > 0x0B (for x2APIC mode) or CPUID leaf 0x01 where 8 bits are sufficient. > Included sanity checks for APIC id from 0x0B. (patch 6/9) > Removed patch for reusing timer calibration for secondary CPUs. > commit message and code improvements. > v8: Fix CPU0 hotplug by setting up the initial_gs, initial_stack and > early_gdt_descr. > Drop trampoline lock and bail if APIC ID not found in find_cpunr. > Code comments improved and debug prints added. > v9: Drop patch to avoid repeated saves of MTRR at boot time. > rebased and retested at v6.2-rc8. > added kernel doc for no_parallel_bringup and made do_parallel_bringup > __ro_after_init. > v10: Fixed suspend/resume not working with parallel smpboot. > rebased and retested to 6.2. > fixed checkpatch errors. > v11: Added patches from Brian Gerst to remove the global variables initial_gs, > initial_stack, and early_gdt_descr from the 64-bit boot code > (https://lore.kernel.org/all/20230222221301.245890-1-brgerst@gmail.com/). > v12: Fixed compilation errors, acquire tr_lock for every stack setup in > trampoline_64.S. > Rearranged commits for a cleaner git history. > > Brian Gerst (3): > x86/smpboot: Remove initial_stack on 64-bit > x86/smpboot: Remove early_gdt_descr on 64-bit > x86/smpboot: Remove initial_gs > > David Woodhouse (8): > x86/apic/x2apic: Allow CPU cluster_mask to be populated in parallel > cpu/hotplug: Move idle_thread_get() to <linux/smpboot.h> > cpu/hotplug: Add dynamic parallel bringup states before > CPUHP_BRINGUP_CPU > x86/smpboot: Reference count on smpboot_setup_warm_reset_vector() > x86/smpboot: Split up native_cpu_up into separate phases and document > them > x86/smpboot: Support parallel startup of secondary CPUs > x86/smpboot: Send INIT/SIPI/SIPI to secondary CPUs in parallel > x86/smpboot: Serialize topology updates for secondary bringup > > .../admin-guide/kernel-parameters.txt | 3 + > arch/x86/include/asm/processor.h | 6 +- > arch/x86/include/asm/realmode.h | 4 +- > arch/x86/include/asm/smp.h | 15 +- > arch/x86/include/asm/topology.h | 2 - > arch/x86/kernel/acpi/sleep.c | 15 +- > arch/x86/kernel/apic/apic.c | 2 +- > arch/x86/kernel/apic/x2apic_cluster.c | 126 ++++--- > arch/x86/kernel/asm-offsets.c | 1 + > arch/x86/kernel/cpu/common.c | 6 +- > arch/x86/kernel/head_64.S | 129 +++++-- > arch/x86/kernel/smpboot.c | 350 +++++++++++++----- > arch/x86/realmode/init.c | 3 + > arch/x86/realmode/rm/trampoline_64.S | 27 +- > arch/x86/xen/smp_pv.c | 4 +- > arch/x86/xen/xen-head.S | 2 +- > include/linux/cpuhotplug.h | 2 + > include/linux/smpboot.h | 7 + > kernel/cpu.c | 31 +- > kernel/smpboot.h | 2 - > 20 files changed, 537 insertions(+), 200 deletions(-) With `CONFIG_FORCE_NR_CPUS=y` this results in: ``` ld: vmlinux.o: in function `secondary_startup_64_no_verify': (.head.text+0x10c): undefined reference to `nr_cpu_ids' ``` That's because in `arch/x86/kernel/head_64.S` `secondary_startup_64_no_verify()` refers to `nr_cpu_ids` under `#ifdef CONFIG_SMP`, but this symbol is available under the following conditions: ``` 38 #if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) 39 #define nr_cpu_ids ((unsigned int)NR_CPUS) 40 #else 41 extern unsigned int nr_cpu_ids; 42 #endif 1090 #if (NR_CPUS > 1) && !defined(CONFIG_FORCE_NR_CPUS) 1091 /* Setup number of possible processor ids */ 1092 unsigned int nr_cpu_ids __read_mostly = NR_CPUS; 1093 EXPORT_SYMBOL(nr_cpu_ids); 1094 #endif ``` So having `CONFIG_SMP=y` and, for instance, `CONFIG_NR_CPUS=320`, it is possible to compile out `EXPORT_SYMBOL(nr_cpu_ids);` if `CONFIG_FORCE_NR_CPUS=y` is set.
On 26/02/2023 18:31, Oleksandr Natalenko wrote: > Hello. > > On neděle 26. února 2023 12:07:51 CET Usama Arif wrote: >> The main code change over v11 is the build error fix by Brian Gerst and >> acquiring tr_lock in trampoline_64.S whenever the stack is setup. >> >> The git history is also rewritten to move the commits that removed >> initial_stack, early_gdt_descr and initial_gs earlier in the patchset. >> >> Thanks, >> Usama >> >> Changes across versions: >> v2: Cut it back to just INIT/SIPI/SIPI in parallel for now, nothing more >> v3: Clean up x2apic patch, add MTRR optimisation, lock topology update >> in preparation for more parallelisation. >> v4: Fixes to the real mode parallelisation patch spotted by SeanC, to >> avoid scribbling on initial_gs in common_cpu_up(), and to allow all >> 24 bits of the physical X2APIC ID to be used. That patch still needs >> a Signed-off-by from its original author, who once claimed not to >> remember writing it at all. But now we've fixed it, hopefully he'll >> admit it now :) >> v5: rebase to v6.1 and remeasure performance, disable parallel bringup >> for AMD CPUs. >> v6: rebase to v6.2-rc6, disabled parallel boot on amd as a cpu bug and >> reused timer calibration for secondary CPUs. >> v7: [David Woodhouse] iterate over all possible CPUs to find any existing >> cluster mask in alloc_clustermask. (patch 1/9) >> Keep parallel AMD support enabled in AMD, using APIC ID in CPUID leaf >> 0x0B (for x2APIC mode) or CPUID leaf 0x01 where 8 bits are sufficient. >> Included sanity checks for APIC id from 0x0B. (patch 6/9) >> Removed patch for reusing timer calibration for secondary CPUs. >> commit message and code improvements. >> v8: Fix CPU0 hotplug by setting up the initial_gs, initial_stack and >> early_gdt_descr. >> Drop trampoline lock and bail if APIC ID not found in find_cpunr. >> Code comments improved and debug prints added. >> v9: Drop patch to avoid repeated saves of MTRR at boot time. >> rebased and retested at v6.2-rc8. >> added kernel doc for no_parallel_bringup and made do_parallel_bringup >> __ro_after_init. >> v10: Fixed suspend/resume not working with parallel smpboot. >> rebased and retested to 6.2. >> fixed checkpatch errors. >> v11: Added patches from Brian Gerst to remove the global variables initial_gs, >> initial_stack, and early_gdt_descr from the 64-bit boot code >> (https://lore.kernel.org/all/20230222221301.245890-1-brgerst@gmail.com/). >> v12: Fixed compilation errors, acquire tr_lock for every stack setup in >> trampoline_64.S. >> Rearranged commits for a cleaner git history. >> >> Brian Gerst (3): >> x86/smpboot: Remove initial_stack on 64-bit >> x86/smpboot: Remove early_gdt_descr on 64-bit >> x86/smpboot: Remove initial_gs >> >> David Woodhouse (8): >> x86/apic/x2apic: Allow CPU cluster_mask to be populated in parallel >> cpu/hotplug: Move idle_thread_get() to <linux/smpboot.h> >> cpu/hotplug: Add dynamic parallel bringup states before >> CPUHP_BRINGUP_CPU >> x86/smpboot: Reference count on smpboot_setup_warm_reset_vector() >> x86/smpboot: Split up native_cpu_up into separate phases and document >> them >> x86/smpboot: Support parallel startup of secondary CPUs >> x86/smpboot: Send INIT/SIPI/SIPI to secondary CPUs in parallel >> x86/smpboot: Serialize topology updates for secondary bringup >> >> .../admin-guide/kernel-parameters.txt | 3 + >> arch/x86/include/asm/processor.h | 6 +- >> arch/x86/include/asm/realmode.h | 4 +- >> arch/x86/include/asm/smp.h | 15 +- >> arch/x86/include/asm/topology.h | 2 - >> arch/x86/kernel/acpi/sleep.c | 15 +- >> arch/x86/kernel/apic/apic.c | 2 +- >> arch/x86/kernel/apic/x2apic_cluster.c | 126 ++++--- >> arch/x86/kernel/asm-offsets.c | 1 + >> arch/x86/kernel/cpu/common.c | 6 +- >> arch/x86/kernel/head_64.S | 129 +++++-- >> arch/x86/kernel/smpboot.c | 350 +++++++++++++----- >> arch/x86/realmode/init.c | 3 + >> arch/x86/realmode/rm/trampoline_64.S | 27 +- >> arch/x86/xen/smp_pv.c | 4 +- >> arch/x86/xen/xen-head.S | 2 +- >> include/linux/cpuhotplug.h | 2 + >> include/linux/smpboot.h | 7 + >> kernel/cpu.c | 31 +- >> kernel/smpboot.h | 2 - >> 20 files changed, 537 insertions(+), 200 deletions(-) > > With `CONFIG_FORCE_NR_CPUS=y` this results in: > > ``` > ld: vmlinux.o: in function `secondary_startup_64_no_verify': > (.head.text+0x10c): undefined reference to `nr_cpu_ids' > ``` > > That's because in `arch/x86/kernel/head_64.S` `secondary_startup_64_no_verify()` refers to `nr_cpu_ids` under `#ifdef CONFIG_SMP`, but this symbol is available under the following conditions: > > ``` > 38 #if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) > 39 #define nr_cpu_ids ((unsigned int)NR_CPUS) > 40 #else > 41 extern unsigned int nr_cpu_ids; > 42 #endif > > 1090 #if (NR_CPUS > 1) && !defined(CONFIG_FORCE_NR_CPUS) > 1091 /* Setup number of possible processor ids */ > 1092 unsigned int nr_cpu_ids __read_mostly = NR_CPUS; > 1093 EXPORT_SYMBOL(nr_cpu_ids); > 1094 #endif > ``` > > So having `CONFIG_SMP=y` and, for instance, `CONFIG_NR_CPUS=320`, it is possible to compile out `EXPORT_SYMBOL(nr_cpu_ids);` if `CONFIG_FORCE_NR_CPUS=y` is set. > I think something like below diff should work in all scenarios? diff --git a/include/linux/cpumask.h b/include/linux/cpumask.h index c2aa0aa26b45..e3727dab9cab 100644 --- a/include/linux/cpumask.h +++ b/include/linux/cpumask.h @@ -35,7 +35,7 @@ typedef struct cpumask { DECLARE_BITMAP(bits, NR_CPUS); } cpumask_t; */ #define cpumask_pr_args(maskp) nr_cpu_ids, cpumask_bits(maskp) -#if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) +#if ((NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS)) && !defined(CONFIG_SMP) #define nr_cpu_ids ((unsigned int)NR_CPUS) #else extern unsigned int nr_cpu_ids; @@ -43,7 +43,7 @@ extern unsigned int nr_cpu_ids; static inline void set_nr_cpu_ids(unsigned int nr) { -#if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) +#if ((NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS)) && !defined(CONFIG_SMP) WARN_ON(nr != nr_cpu_ids); #else nr_cpu_ids = nr; diff --git a/kernel/smp.c b/kernel/smp.c index 06a413987a14..a051b16d4a24 100644 --- a/kernel/smp.c +++ b/kernel/smp.c @@ -1087,11 +1087,9 @@ static int __init maxcpus(char *str) early_param("maxcpus", maxcpus); -#if (NR_CPUS > 1) && !defined(CONFIG_FORCE_NR_CPUS) /* Setup number of possible processor ids */ unsigned int nr_cpu_ids __read_mostly = NR_CPUS; EXPORT_SYMBOL(nr_cpu_ids); -#endif /* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */ void __init setup_nr_cpu_ids(void)
On 26 February 2023 20:59:17 GMT, Usama Arif <usama.arif@bytedance.com> wrote: > > >On 26/02/2023 18:31, Oleksandr Natalenko wrote: >> Hello. >> >> On neděle 26. února 2023 12:07:51 CET Usama Arif wrote: >>> The main code change over v11 is the build error fix by Brian Gerst and >>> acquiring tr_lock in trampoline_64.S whenever the stack is setup. >>> >>> The git history is also rewritten to move the commits that removed >>> initial_stack, early_gdt_descr and initial_gs earlier in the patchset. >>> >>> Thanks, >>> Usama >>> >>> Changes across versions: >>> v2: Cut it back to just INIT/SIPI/SIPI in parallel for now, nothing more >>> v3: Clean up x2apic patch, add MTRR optimisation, lock topology update >>> in preparation for more parallelisation. >>> v4: Fixes to the real mode parallelisation patch spotted by SeanC, to >>> avoid scribbling on initial_gs in common_cpu_up(), and to allow all >>> 24 bits of the physical X2APIC ID to be used. That patch still needs >>> a Signed-off-by from its original author, who once claimed not to >>> remember writing it at all. But now we've fixed it, hopefully he'll >>> admit it now :) >>> v5: rebase to v6.1 and remeasure performance, disable parallel bringup >>> for AMD CPUs. >>> v6: rebase to v6.2-rc6, disabled parallel boot on amd as a cpu bug and >>> reused timer calibration for secondary CPUs. >>> v7: [David Woodhouse] iterate over all possible CPUs to find any existing >>> cluster mask in alloc_clustermask. (patch 1/9) >>> Keep parallel AMD support enabled in AMD, using APIC ID in CPUID leaf >>> 0x0B (for x2APIC mode) or CPUID leaf 0x01 where 8 bits are sufficient. >>> Included sanity checks for APIC id from 0x0B. (patch 6/9) >>> Removed patch for reusing timer calibration for secondary CPUs. >>> commit message and code improvements. >>> v8: Fix CPU0 hotplug by setting up the initial_gs, initial_stack and >>> early_gdt_descr. >>> Drop trampoline lock and bail if APIC ID not found in find_cpunr. >>> Code comments improved and debug prints added. >>> v9: Drop patch to avoid repeated saves of MTRR at boot time. >>> rebased and retested at v6.2-rc8. >>> added kernel doc for no_parallel_bringup and made do_parallel_bringup >>> __ro_after_init. >>> v10: Fixed suspend/resume not working with parallel smpboot. >>> rebased and retested to 6.2. >>> fixed checkpatch errors. >>> v11: Added patches from Brian Gerst to remove the global variables initial_gs, >>> initial_stack, and early_gdt_descr from the 64-bit boot code >>> (https://lore.kernel.org/all/20230222221301.245890-1-brgerst@gmail.com/). >>> v12: Fixed compilation errors, acquire tr_lock for every stack setup in >>> trampoline_64.S. >>> Rearranged commits for a cleaner git history. >>> >>> Brian Gerst (3): >>> x86/smpboot: Remove initial_stack on 64-bit >>> x86/smpboot: Remove early_gdt_descr on 64-bit >>> x86/smpboot: Remove initial_gs >>> >>> David Woodhouse (8): >>> x86/apic/x2apic: Allow CPU cluster_mask to be populated in parallel >>> cpu/hotplug: Move idle_thread_get() to <linux/smpboot.h> >>> cpu/hotplug: Add dynamic parallel bringup states before >>> CPUHP_BRINGUP_CPU >>> x86/smpboot: Reference count on smpboot_setup_warm_reset_vector() >>> x86/smpboot: Split up native_cpu_up into separate phases and document >>> them >>> x86/smpboot: Support parallel startup of secondary CPUs >>> x86/smpboot: Send INIT/SIPI/SIPI to secondary CPUs in parallel >>> x86/smpboot: Serialize topology updates for secondary bringup >>> >>> .../admin-guide/kernel-parameters.txt | 3 + >>> arch/x86/include/asm/processor.h | 6 +- >>> arch/x86/include/asm/realmode.h | 4 +- >>> arch/x86/include/asm/smp.h | 15 +- >>> arch/x86/include/asm/topology.h | 2 - >>> arch/x86/kernel/acpi/sleep.c | 15 +- >>> arch/x86/kernel/apic/apic.c | 2 +- >>> arch/x86/kernel/apic/x2apic_cluster.c | 126 ++++--- >>> arch/x86/kernel/asm-offsets.c | 1 + >>> arch/x86/kernel/cpu/common.c | 6 +- >>> arch/x86/kernel/head_64.S | 129 +++++-- >>> arch/x86/kernel/smpboot.c | 350 +++++++++++++----- >>> arch/x86/realmode/init.c | 3 + >>> arch/x86/realmode/rm/trampoline_64.S | 27 +- >>> arch/x86/xen/smp_pv.c | 4 +- >>> arch/x86/xen/xen-head.S | 2 +- >>> include/linux/cpuhotplug.h | 2 + >>> include/linux/smpboot.h | 7 + >>> kernel/cpu.c | 31 +- >>> kernel/smpboot.h | 2 - >>> 20 files changed, 537 insertions(+), 200 deletions(-) >> >> With `CONFIG_FORCE_NR_CPUS=y` this results in: >> >> ``` >> ld: vmlinux.o: in function `secondary_startup_64_no_verify': >> (.head.text+0x10c): undefined reference to `nr_cpu_ids' >> ``` >> >> That's because in `arch/x86/kernel/head_64.S` `secondary_startup_64_no_verify()` refers to `nr_cpu_ids` under `#ifdef CONFIG_SMP`, but this symbol is available under the following conditions: >> >> ``` >> 38 #if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) >> 39 #define nr_cpu_ids ((unsigned int)NR_CPUS) >> 40 #else >> 41 extern unsigned int nr_cpu_ids; >> 42 #endif >> >> 1090 #if (NR_CPUS > 1) && !defined(CONFIG_FORCE_NR_CPUS) >> 1091 /* Setup number of possible processor ids */ >> 1092 unsigned int nr_cpu_ids __read_mostly = NR_CPUS; >> 1093 EXPORT_SYMBOL(nr_cpu_ids); >> 1094 #endif >> ``` >> >> So having `CONFIG_SMP=y` and, for instance, `CONFIG_NR_CPUS=320`, it is possible to compile out `EXPORT_SYMBOL(nr_cpu_ids);` if `CONFIG_FORCE_NR_CPUS=y` is set. >> > >I think something like below diff should work in all scenarios? I'd've changed the asm side to use the constant limit.
On 26/02/2023 20:59, Usama Arif wrote: > > > On 26/02/2023 18:31, Oleksandr Natalenko wrote: >> Hello. >> >> On neděle 26. února 2023 12:07:51 CET Usama Arif wrote: >>> The main code change over v11 is the build error fix by Brian Gerst and >>> acquiring tr_lock in trampoline_64.S whenever the stack is setup. >>> >>> The git history is also rewritten to move the commits that removed >>> initial_stack, early_gdt_descr and initial_gs earlier in the patchset. >>> >>> Thanks, >>> Usama >>> >>> Changes across versions: >>> v2: Cut it back to just INIT/SIPI/SIPI in parallel for now, nothing more >>> v3: Clean up x2apic patch, add MTRR optimisation, lock topology update >>> in preparation for more parallelisation. >>> v4: Fixes to the real mode parallelisation patch spotted by SeanC, to >>> avoid scribbling on initial_gs in common_cpu_up(), and to allow all >>> 24 bits of the physical X2APIC ID to be used. That patch still >>> needs >>> a Signed-off-by from its original author, who once claimed not to >>> remember writing it at all. But now we've fixed it, hopefully he'll >>> admit it now :) >>> v5: rebase to v6.1 and remeasure performance, disable parallel bringup >>> for AMD CPUs. >>> v6: rebase to v6.2-rc6, disabled parallel boot on amd as a cpu bug and >>> reused timer calibration for secondary CPUs. >>> v7: [David Woodhouse] iterate over all possible CPUs to find any >>> existing >>> cluster mask in alloc_clustermask. (patch 1/9) >>> Keep parallel AMD support enabled in AMD, using APIC ID in CPUID >>> leaf >>> 0x0B (for x2APIC mode) or CPUID leaf 0x01 where 8 bits are >>> sufficient. >>> Included sanity checks for APIC id from 0x0B. (patch 6/9) >>> Removed patch for reusing timer calibration for secondary CPUs. >>> commit message and code improvements. >>> v8: Fix CPU0 hotplug by setting up the initial_gs, initial_stack and >>> early_gdt_descr. >>> Drop trampoline lock and bail if APIC ID not found in find_cpunr. >>> Code comments improved and debug prints added. >>> v9: Drop patch to avoid repeated saves of MTRR at boot time. >>> rebased and retested at v6.2-rc8. >>> added kernel doc for no_parallel_bringup and made >>> do_parallel_bringup >>> __ro_after_init. >>> v10: Fixed suspend/resume not working with parallel smpboot. >>> rebased and retested to 6.2. >>> fixed checkpatch errors. >>> v11: Added patches from Brian Gerst to remove the global variables >>> initial_gs, >>> initial_stack, and early_gdt_descr from the 64-bit boot code >>> >>> (https://lore.kernel.org/all/20230222221301.245890-1-brgerst@gmail.com/). >>> v12: Fixed compilation errors, acquire tr_lock for every stack setup in >>> trampoline_64.S. >>> Rearranged commits for a cleaner git history. >>> >>> Brian Gerst (3): >>> x86/smpboot: Remove initial_stack on 64-bit >>> x86/smpboot: Remove early_gdt_descr on 64-bit >>> x86/smpboot: Remove initial_gs >>> >>> David Woodhouse (8): >>> x86/apic/x2apic: Allow CPU cluster_mask to be populated in parallel >>> cpu/hotplug: Move idle_thread_get() to <linux/smpboot.h> >>> cpu/hotplug: Add dynamic parallel bringup states before >>> CPUHP_BRINGUP_CPU >>> x86/smpboot: Reference count on smpboot_setup_warm_reset_vector() >>> x86/smpboot: Split up native_cpu_up into separate phases and document >>> them >>> x86/smpboot: Support parallel startup of secondary CPUs >>> x86/smpboot: Send INIT/SIPI/SIPI to secondary CPUs in parallel >>> x86/smpboot: Serialize topology updates for secondary bringup >>> >>> .../admin-guide/kernel-parameters.txt | 3 + >>> arch/x86/include/asm/processor.h | 6 +- >>> arch/x86/include/asm/realmode.h | 4 +- >>> arch/x86/include/asm/smp.h | 15 +- >>> arch/x86/include/asm/topology.h | 2 - >>> arch/x86/kernel/acpi/sleep.c | 15 +- >>> arch/x86/kernel/apic/apic.c | 2 +- >>> arch/x86/kernel/apic/x2apic_cluster.c | 126 ++++--- >>> arch/x86/kernel/asm-offsets.c | 1 + >>> arch/x86/kernel/cpu/common.c | 6 +- >>> arch/x86/kernel/head_64.S | 129 +++++-- >>> arch/x86/kernel/smpboot.c | 350 +++++++++++++----- >>> arch/x86/realmode/init.c | 3 + >>> arch/x86/realmode/rm/trampoline_64.S | 27 +- >>> arch/x86/xen/smp_pv.c | 4 +- >>> arch/x86/xen/xen-head.S | 2 +- >>> include/linux/cpuhotplug.h | 2 + >>> include/linux/smpboot.h | 7 + >>> kernel/cpu.c | 31 +- >>> kernel/smpboot.h | 2 - >>> 20 files changed, 537 insertions(+), 200 deletions(-) >> >> With `CONFIG_FORCE_NR_CPUS=y` this results in: >> >> ``` >> ld: vmlinux.o: in function `secondary_startup_64_no_verify': >> (.head.text+0x10c): undefined reference to `nr_cpu_ids' >> ``` >> >> That's because in `arch/x86/kernel/head_64.S` >> `secondary_startup_64_no_verify()` refers to `nr_cpu_ids` under >> `#ifdef CONFIG_SMP`, but this symbol is available under the following >> conditions: >> >> ``` >> 38 #if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) >> 39 #define nr_cpu_ids ((unsigned int)NR_CPUS) >> 40 #else >> 41 extern unsigned int nr_cpu_ids; >> 42 #endif >> >> 1090 #if (NR_CPUS > 1) && !defined(CONFIG_FORCE_NR_CPUS) >> 1091 /* Setup number of possible processor ids */ >> 1092 unsigned int nr_cpu_ids __read_mostly = NR_CPUS; >> 1093 EXPORT_SYMBOL(nr_cpu_ids); >> 1094 #endif >> ``` >> >> So having `CONFIG_SMP=y` and, for instance, `CONFIG_NR_CPUS=320`, it >> is possible to compile out `EXPORT_SYMBOL(nr_cpu_ids);` if >> `CONFIG_FORCE_NR_CPUS=y` is set. >> > > I think something like below diff should work in all scenarios? > > diff --git a/include/linux/cpumask.h b/include/linux/cpumask.h > index c2aa0aa26b45..e3727dab9cab 100644 > --- a/include/linux/cpumask.h > +++ b/include/linux/cpumask.h > @@ -35,7 +35,7 @@ typedef struct cpumask { DECLARE_BITMAP(bits, > NR_CPUS); } cpumask_t; > */ > #define cpumask_pr_args(maskp) nr_cpu_ids, cpumask_bits(maskp) > > -#if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) > +#if ((NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS)) && > !defined(CONFIG_SMP) > #define nr_cpu_ids ((unsigned int)NR_CPUS) > #else > extern unsigned int nr_cpu_ids; > @@ -43,7 +43,7 @@ extern unsigned int nr_cpu_ids; > > static inline void set_nr_cpu_ids(unsigned int nr) > { > -#if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) > +#if ((NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS)) && > !defined(CONFIG_SMP) > WARN_ON(nr != nr_cpu_ids); > #else > nr_cpu_ids = nr; > diff --git a/kernel/smp.c b/kernel/smp.c > index 06a413987a14..a051b16d4a24 100644 > --- a/kernel/smp.c > +++ b/kernel/smp.c > @@ -1087,11 +1087,9 @@ static int __init maxcpus(char *str) > > early_param("maxcpus", maxcpus); > > -#if (NR_CPUS > 1) && !defined(CONFIG_FORCE_NR_CPUS) > /* Setup number of possible processor ids */ > unsigned int nr_cpu_ids __read_mostly = NR_CPUS; > EXPORT_SYMBOL(nr_cpu_ids); > -#endif > > /* An arch may set nr_cpu_ids earlier if needed, so this would be > redundant */ > void __init setup_nr_cpu_ids(void) Or better just do below? diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S index 17bdd6122dca..5d709aa67df4 100644 --- a/arch/x86/kernel/head_64.S +++ b/arch/x86/kernel/head_64.S @@ -273,7 +273,11 @@ SYM_INNER_LABEL(secondary_startup_64_no_verify, SYM_L_GLOBAL) cmpl (%rbx,%rcx,4), %edx jz .Lsetup_cpu inc %ecx +#if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) + cmpl $NR_CPUS, %ecx +#else cmpl nr_cpu_ids(%rip), %ecx +#endif jb .Lfind_cpunr /* APIC ID not found in the table. Drop the trampoline lock and bail. */
On 27/02/2023 06:13, David Woodhouse wrote: > > > On 26 February 2023 20:59:17 GMT, Usama Arif <usama.arif@bytedance.com> wrote: >> >> >> On 26/02/2023 18:31, Oleksandr Natalenko wrote: >>> Hello. >>> >>> On neděle 26. února 2023 12:07:51 CET Usama Arif wrote: >>>> The main code change over v11 is the build error fix by Brian Gerst and >>>> acquiring tr_lock in trampoline_64.S whenever the stack is setup. >>>> >>>> The git history is also rewritten to move the commits that removed >>>> initial_stack, early_gdt_descr and initial_gs earlier in the patchset. >>>> >>>> Thanks, >>>> Usama >>>> >>>> Changes across versions: >>>> v2: Cut it back to just INIT/SIPI/SIPI in parallel for now, nothing more >>>> v3: Clean up x2apic patch, add MTRR optimisation, lock topology update >>>> in preparation for more parallelisation. >>>> v4: Fixes to the real mode parallelisation patch spotted by SeanC, to >>>> avoid scribbling on initial_gs in common_cpu_up(), and to allow all >>>> 24 bits of the physical X2APIC ID to be used. That patch still needs >>>> a Signed-off-by from its original author, who once claimed not to >>>> remember writing it at all. But now we've fixed it, hopefully he'll >>>> admit it now :) >>>> v5: rebase to v6.1 and remeasure performance, disable parallel bringup >>>> for AMD CPUs. >>>> v6: rebase to v6.2-rc6, disabled parallel boot on amd as a cpu bug and >>>> reused timer calibration for secondary CPUs. >>>> v7: [David Woodhouse] iterate over all possible CPUs to find any existing >>>> cluster mask in alloc_clustermask. (patch 1/9) >>>> Keep parallel AMD support enabled in AMD, using APIC ID in CPUID leaf >>>> 0x0B (for x2APIC mode) or CPUID leaf 0x01 where 8 bits are sufficient. >>>> Included sanity checks for APIC id from 0x0B. (patch 6/9) >>>> Removed patch for reusing timer calibration for secondary CPUs. >>>> commit message and code improvements. >>>> v8: Fix CPU0 hotplug by setting up the initial_gs, initial_stack and >>>> early_gdt_descr. >>>> Drop trampoline lock and bail if APIC ID not found in find_cpunr. >>>> Code comments improved and debug prints added. >>>> v9: Drop patch to avoid repeated saves of MTRR at boot time. >>>> rebased and retested at v6.2-rc8. >>>> added kernel doc for no_parallel_bringup and made do_parallel_bringup >>>> __ro_after_init. >>>> v10: Fixed suspend/resume not working with parallel smpboot. >>>> rebased and retested to 6.2. >>>> fixed checkpatch errors. >>>> v11: Added patches from Brian Gerst to remove the global variables initial_gs, >>>> initial_stack, and early_gdt_descr from the 64-bit boot code >>>> (https://lore.kernel.org/all/20230222221301.245890-1-brgerst@gmail.com/). >>>> v12: Fixed compilation errors, acquire tr_lock for every stack setup in >>>> trampoline_64.S. >>>> Rearranged commits for a cleaner git history. >>>> >>>> Brian Gerst (3): >>>> x86/smpboot: Remove initial_stack on 64-bit >>>> x86/smpboot: Remove early_gdt_descr on 64-bit >>>> x86/smpboot: Remove initial_gs >>>> >>>> David Woodhouse (8): >>>> x86/apic/x2apic: Allow CPU cluster_mask to be populated in parallel >>>> cpu/hotplug: Move idle_thread_get() to <linux/smpboot.h> >>>> cpu/hotplug: Add dynamic parallel bringup states before >>>> CPUHP_BRINGUP_CPU >>>> x86/smpboot: Reference count on smpboot_setup_warm_reset_vector() >>>> x86/smpboot: Split up native_cpu_up into separate phases and document >>>> them >>>> x86/smpboot: Support parallel startup of secondary CPUs >>>> x86/smpboot: Send INIT/SIPI/SIPI to secondary CPUs in parallel >>>> x86/smpboot: Serialize topology updates for secondary bringup >>>> >>>> .../admin-guide/kernel-parameters.txt | 3 + >>>> arch/x86/include/asm/processor.h | 6 +- >>>> arch/x86/include/asm/realmode.h | 4 +- >>>> arch/x86/include/asm/smp.h | 15 +- >>>> arch/x86/include/asm/topology.h | 2 - >>>> arch/x86/kernel/acpi/sleep.c | 15 +- >>>> arch/x86/kernel/apic/apic.c | 2 +- >>>> arch/x86/kernel/apic/x2apic_cluster.c | 126 ++++--- >>>> arch/x86/kernel/asm-offsets.c | 1 + >>>> arch/x86/kernel/cpu/common.c | 6 +- >>>> arch/x86/kernel/head_64.S | 129 +++++-- >>>> arch/x86/kernel/smpboot.c | 350 +++++++++++++----- >>>> arch/x86/realmode/init.c | 3 + >>>> arch/x86/realmode/rm/trampoline_64.S | 27 +- >>>> arch/x86/xen/smp_pv.c | 4 +- >>>> arch/x86/xen/xen-head.S | 2 +- >>>> include/linux/cpuhotplug.h | 2 + >>>> include/linux/smpboot.h | 7 + >>>> kernel/cpu.c | 31 +- >>>> kernel/smpboot.h | 2 - >>>> 20 files changed, 537 insertions(+), 200 deletions(-) >>> >>> With `CONFIG_FORCE_NR_CPUS=y` this results in: >>> >>> ``` >>> ld: vmlinux.o: in function `secondary_startup_64_no_verify': >>> (.head.text+0x10c): undefined reference to `nr_cpu_ids' >>> ``` >>> >>> That's because in `arch/x86/kernel/head_64.S` `secondary_startup_64_no_verify()` refers to `nr_cpu_ids` under `#ifdef CONFIG_SMP`, but this symbol is available under the following conditions: >>> >>> ``` >>> 38 #if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) >>> 39 #define nr_cpu_ids ((unsigned int)NR_CPUS) >>> 40 #else >>> 41 extern unsigned int nr_cpu_ids; >>> 42 #endif >>> >>> 1090 #if (NR_CPUS > 1) && !defined(CONFIG_FORCE_NR_CPUS) >>> 1091 /* Setup number of possible processor ids */ >>> 1092 unsigned int nr_cpu_ids __read_mostly = NR_CPUS; >>> 1093 EXPORT_SYMBOL(nr_cpu_ids); >>> 1094 #endif >>> ``` >>> >>> So having `CONFIG_SMP=y` and, for instance, `CONFIG_NR_CPUS=320`, it is possible to compile out `EXPORT_SYMBOL(nr_cpu_ids);` if `CONFIG_FORCE_NR_CPUS=y` is set. >>> >> >> I think something like below diff should work in all scenarios? > > I'd've changed the asm side to use the constant limit. Yup, just needed the morning coffee :) Had sent the proper fix in https://lore.kernel.org/all/5e8ad90a-1dc6-95c2-e020-5e95da6f9eda@bytedance.com/#t I guess the diff is still small over v12 (including the cosmetic changes) to send out a new version so soon, probably better to wait a couple of days incase something else comes up as well? Thanks, Usama
On Mon, 2023-02-27 at 06:14 +0000, Usama Arif wrote: > > diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S > index 17bdd6122dca..5d709aa67df4 100644 > --- a/arch/x86/kernel/head_64.S > +++ b/arch/x86/kernel/head_64.S > @@ -273,7 +273,11 @@ SYM_INNER_LABEL(secondary_startup_64_no_verify, > SYM_L_GLOBAL) > cmpl (%rbx,%rcx,4), %edx > jz .Lsetup_cpu > inc %ecx > +#if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) > + cmpl $NR_CPUS, %ecx > +#else > cmpl nr_cpu_ids(%rip), %ecx > +#endif > jb .Lfind_cpunr > > /* APIC ID not found in the table. Drop the trampoline lock > and bail. */ The whitespace looks dodgy there but maybe that's just your mail client? Given this code is already in #ifdef CONFIG_SMP, can NR_CPUS be 1?
On 27/02/2023 15:29, David Woodhouse wrote: > On Mon, 2023-02-27 at 06:14 +0000, Usama Arif wrote: >> >> diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S >> index 17bdd6122dca..5d709aa67df4 100644 >> --- a/arch/x86/kernel/head_64.S >> +++ b/arch/x86/kernel/head_64.S >> @@ -273,7 +273,11 @@ SYM_INNER_LABEL(secondary_startup_64_no_verify, >> SYM_L_GLOBAL) >> cmpl (%rbx,%rcx,4), %edx >> jz .Lsetup_cpu >> inc %ecx >> +#if (NR_CPUS == 1) || defined(CONFIG_FORCE_NR_CPUS) >> + cmpl $NR_CPUS, %ecx >> +#else >> cmpl nr_cpu_ids(%rip), %ecx >> +#endif >> jb .Lfind_cpunr >> >> /* APIC ID not found in the table. Drop the trampoline lock >> and bail. */ > > The whitespace looks dodgy there but maybe that's just your mail client? > > Given this code is already in #ifdef CONFIG_SMP, can NR_CPUS be 1? Ah yes, we have config NR_CPUS_RANGE_BEGIN int default NR_CPUS_RANGE_END if MAXSMP default 1 if !SMP default 2 in arch/x86/Kconfig which doesn't let us select 1 for NR_CPUS if SMP is enabled, so this should be enough diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S index 17bdd6122dca..c79ae67492e1 100644 --- a/arch/x86/kernel/head_64.S +++ b/arch/x86/kernel/head_64.S @@ -273,7 +273,11 @@ SYM_INNER_LABEL(secondary_startup_64_no_verify, SYM_L_GLOBAL) cmpl (%rbx,%rcx,4), %edx jz .Lsetup_cpu inc %ecx +#if defined(CONFIG_FORCE_NR_CPUS) + cmpl $NR_CPUS, %ebx +#else cmpl nr_cpu_ids(%rip), %ecx +#endif jb .Lfind_cpunr
Hi Usama and David, thanks for the great series!
I've tested it on Steam Deck (with and without the "no_parallel_bringup"
parameter), it works fine - also tested S3/deep sleep-resume cycle.
Feel free to add (to the series):
Tested-by: Guilherme G. Piccoli <gpiccoli@igalia.com>
Also, just taking the opportunity since I'm already replying here: on
patch 09, found two typos:
s/correect/correct (commit message)
s/brinugp/bring-up (kernel-parameters.txt)
Cheers,
Guilherme
On Mon, 2023-02-27 at 18:39 -0300, Guilherme G. Piccoli wrote: > Hi Usama and David, thanks for the great series! > > I've tested it on Steam Deck (with and without the > "no_parallel_bringup" > parameter), it works fine - also tested S3/deep sleep-resume cycle. > > Feel free to add (to the series): > Tested-by: Guilherme G. Piccoli <gpiccoli@igalia.com> > > Also, just taking the opportunity since I'm already replying here: on > patch 09, found two typos: > > s/correect/correct (commit message) > s/brinugp/bring-up (kernel-parameters.txt) > > Cheers, Thanks. I've done that and pushed it out to https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/parallel-6.2-rc8-v12bis ready for the next round.
Dear Guilherme, Am 27.02.23 um 22:39 schrieb Guilherme G. Piccoli: > I've tested it on Steam Deck (with and without the "no_parallel_bringup" > parameter), it works fine - also tested S3/deep sleep-resume cycle. > > Feel free to add (to the series): > Tested-by: Guilherme G. Piccoli <gpiccoli@igalia.com> Thank you for testing the series. It’d be great if you could share the timing differences. […] Kind regards, Paul
On 28/02/2023 07:13, Paul Menzel wrote: > Dear Guilherme, > > > Am 27.02.23 um 22:39 schrieb Guilherme G. Piccoli: > >> I've tested it on Steam Deck (with and without the "no_parallel_bringup" >> parameter), it works fine - also tested S3/deep sleep-resume cycle. >> >> Feel free to add (to the series): >> Tested-by: Guilherme G. Piccoli <gpiccoli@igalia.com> > > Thank you for testing the series. It’d be great if you could share the > timing differences. > > […] > > > Kind regards, > > Paul Hi Paul! The results...weren't so great, I felt no difference heh Which is also not bad, it seems the series favors big SMP systems, Deck has only 8 CPUs. But maybe the way I measured is not ideal? I just compared timestamps on dmesg from the first SMP message up to the one that says the boot of secondary CPUs is complete. Do you have a better suggestion? I can try things here. Cheers, Guilherme
The main code change over v11 is the build error fix by Brian Gerst and acquiring tr_lock in trampoline_64.S whenever the stack is setup. The git history is also rewritten to move the commits that removed initial_stack, early_gdt_descr and initial_gs earlier in the patchset. Thanks, Usama Changes across versions: v2: Cut it back to just INIT/SIPI/SIPI in parallel for now, nothing more v3: Clean up x2apic patch, add MTRR optimisation, lock topology update in preparation for more parallelisation. v4: Fixes to the real mode parallelisation patch spotted by SeanC, to avoid scribbling on initial_gs in common_cpu_up(), and to allow all 24 bits of the physical X2APIC ID to be used. That patch still needs a Signed-off-by from its original author, who once claimed not to remember writing it at all. But now we've fixed it, hopefully he'll admit it now :) v5: rebase to v6.1 and remeasure performance, disable parallel bringup for AMD CPUs. v6: rebase to v6.2-rc6, disabled parallel boot on amd as a cpu bug and reused timer calibration for secondary CPUs. v7: [David Woodhouse] iterate over all possible CPUs to find any existing cluster mask in alloc_clustermask. (patch 1/9) Keep parallel AMD support enabled in AMD, using APIC ID in CPUID leaf 0x0B (for x2APIC mode) or CPUID leaf 0x01 where 8 bits are sufficient. Included sanity checks for APIC id from 0x0B. (patch 6/9) Removed patch for reusing timer calibration for secondary CPUs. commit message and code improvements. v8: Fix CPU0 hotplug by setting up the initial_gs, initial_stack and early_gdt_descr. Drop trampoline lock and bail if APIC ID not found in find_cpunr. Code comments improved and debug prints added. v9: Drop patch to avoid repeated saves of MTRR at boot time. rebased and retested at v6.2-rc8. added kernel doc for no_parallel_bringup and made do_parallel_bringup __ro_after_init. v10: Fixed suspend/resume not working with parallel smpboot. rebased and retested to 6.2. fixed checkpatch errors. v11: Added patches from Brian Gerst to remove the global variables initial_gs, initial_stack, and early_gdt_descr from the 64-bit boot code (https://lore.kernel.org/all/20230222221301.245890-1-brgerst@gmail.com/). v12: Fixed compilation errors, acquire tr_lock for every stack setup in trampoline_64.S. Rearranged commits for a cleaner git history. Brian Gerst (3): x86/smpboot: Remove initial_stack on 64-bit x86/smpboot: Remove early_gdt_descr on 64-bit x86/smpboot: Remove initial_gs David Woodhouse (8): x86/apic/x2apic: Allow CPU cluster_mask to be populated in parallel cpu/hotplug: Move idle_thread_get() to <linux/smpboot.h> cpu/hotplug: Add dynamic parallel bringup states before CPUHP_BRINGUP_CPU x86/smpboot: Reference count on smpboot_setup_warm_reset_vector() x86/smpboot: Split up native_cpu_up into separate phases and document them x86/smpboot: Support parallel startup of secondary CPUs x86/smpboot: Send INIT/SIPI/SIPI to secondary CPUs in parallel x86/smpboot: Serialize topology updates for secondary bringup .../admin-guide/kernel-parameters.txt | 3 + arch/x86/include/asm/processor.h | 6 +- arch/x86/include/asm/realmode.h | 4 +- arch/x86/include/asm/smp.h | 15 +- arch/x86/include/asm/topology.h | 2 - arch/x86/kernel/acpi/sleep.c | 15 +- arch/x86/kernel/apic/apic.c | 2 +- arch/x86/kernel/apic/x2apic_cluster.c | 126 ++++--- arch/x86/kernel/asm-offsets.c | 1 + arch/x86/kernel/cpu/common.c | 6 +- arch/x86/kernel/head_64.S | 129 +++++-- arch/x86/kernel/smpboot.c | 350 +++++++++++++----- arch/x86/realmode/init.c | 3 + arch/x86/realmode/rm/trampoline_64.S | 27 +- arch/x86/xen/smp_pv.c | 4 +- arch/x86/xen/xen-head.S | 2 +- include/linux/cpuhotplug.h | 2 + include/linux/smpboot.h | 7 + kernel/cpu.c | 31 +- kernel/smpboot.h | 2 - 20 files changed, 537 insertions(+), 200 deletions(-)