| Message ID | 20171106173426.8902-1-cascardo@canonical.com (mailing list archive) |
|---|---|
| State | Rejected |
| Headers | show |
| Series | [v3] ppc64 boot: Wait for boot cpu to show up if nr_cpus limit is about to hit. | expand |
On 11/06/2017 03:34 PM, Thadeu Lima de Souza Cascardo wrote: > From: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com> > > The kernel boot parameter 'nr_cpus=' allows one to specify number of > possible cpus in the system. In the normal scenario the first cpu (cpu0) > that shows up is the boot cpu and hence it gets covered under nr_cpus > limit. > > But this assumption will be broken in kdump scenario where kdump kenrel Minor nit: s/kenrel/kernel > after a crash can boot up on an non-zero boot cpu. The paca structure > allocation depends on value of nr_cpus and is indexed using logical cpu > ids. This definetly will be an issue if boot cpu id > nr_cpus Minor nit (2): s/definetly/definitely > > This patch modifies allocate_pacas() and smp_setup_cpu_maps() to > accommodate boot cpu for the case where boot_cpuid > nr_cpu_ids. > > This change would help to reduce the memory reservation requirement for > kdump on ppc64. > > Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com> > Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com> Tested-by: Guilherme G. Piccoli <gpiccoli@linux.vnet.ibm.com> Without this patch, got a crash deterministically when booting with nr_cpus=1 in P8 bare-metal. The patch fixes the issue... Thanks, Guilherme > --- > > v3: fixup signedness or nr_cpus to match nr_cpu_ids > and fix conflict due to change from %d to %u > > Resending this as it was not applied, and I can reproduce the issue with > v4.14-rc8 when booting a kdump kernel after a crash that has been given > nr_cpus=1 as a parameter. With this patch, I can't reproduce it anymore. > > --- > arch/powerpc/include/asm/paca.h | 3 +++ > arch/powerpc/include/asm/smp.h | 1 + > arch/powerpc/kernel/paca.c | 23 +++++++++++++++++----- > arch/powerpc/kernel/prom.c | 39 +++++++++++++++++++++++++++++++++++++- > arch/powerpc/kernel/setup-common.c | 25 ++++++++++++++++++++++++ > 5 files changed, 85 insertions(+), 6 deletions(-) > > diff --git a/arch/powerpc/include/asm/paca.h b/arch/powerpc/include/asm/paca.h > index 04b60af027ae..ea0dbf2bbeef 100644 > --- a/arch/powerpc/include/asm/paca.h > +++ b/arch/powerpc/include/asm/paca.h > @@ -49,6 +49,9 @@ extern unsigned int debug_smp_processor_id(void); /* from linux/smp.h */ > #define get_lppaca() (get_paca()->lppaca_ptr) > #define get_slb_shadow() (get_paca()->slb_shadow_ptr) > > +/* Maximum number of threads per core. */ > +#define MAX_SMT 8 > + > struct task_struct; > > /* > diff --git a/arch/powerpc/include/asm/smp.h b/arch/powerpc/include/asm/smp.h > index fac963e10d39..553cd22b2ccc 100644 > --- a/arch/powerpc/include/asm/smp.h > +++ b/arch/powerpc/include/asm/smp.h > @@ -30,6 +30,7 @@ > #include <asm/percpu.h> > > extern int boot_cpuid; > +extern int boot_hw_cpuid; > extern int spinning_secondaries; > > extern void cpu_die(void); > diff --git a/arch/powerpc/kernel/paca.c b/arch/powerpc/kernel/paca.c > index 2ff2b8a19f71..9c689ee4b6a3 100644 > --- a/arch/powerpc/kernel/paca.c > +++ b/arch/powerpc/kernel/paca.c > @@ -207,6 +207,7 @@ void __init allocate_pacas(void) > { > u64 limit; > int cpu; > + unsigned int nr_cpus; > > limit = ppc64_rma_size; > > @@ -219,20 +220,32 @@ void __init allocate_pacas(void) > limit = min(0x10000000ULL, limit); > #endif > > - paca_size = PAGE_ALIGN(sizeof(struct paca_struct) * nr_cpu_ids); > + /* > + * Always align up the nr_cpu_ids to SMT threads and allocate > + * the paca. This will help us to prepare for a situation where > + * boot cpu id > nr_cpus_id. We will use the last nthreads > + * slots (nthreads == threads per core) to accommodate a core > + * that contains boot cpu thread. > + * > + * Do not change nr_cpu_ids value here. Let us do that in > + * early_init_dt_scan_cpus() where we know exact value > + * of threads per core. > + */ > + nr_cpus = _ALIGN_UP(nr_cpu_ids, MAX_SMT); > + paca_size = PAGE_ALIGN(sizeof(struct paca_struct) * nr_cpus); > > paca = __va(memblock_alloc_base(paca_size, PAGE_SIZE, limit)); > memset(paca, 0, paca_size); > > printk(KERN_DEBUG "Allocated %u bytes for %u pacas at %p\n", > - paca_size, nr_cpu_ids, paca); > + paca_size, nr_cpus, paca); > > - allocate_lppacas(nr_cpu_ids, limit); > + allocate_lppacas(nr_cpus, limit); > > - allocate_slb_shadows(nr_cpu_ids, limit); > + allocate_slb_shadows(nr_cpus, limit); > > /* Can't use for_each_*_cpu, as they aren't functional yet */ > - for (cpu = 0; cpu < nr_cpu_ids; cpu++) > + for (cpu = 0; cpu < nr_cpus; cpu++) > initialise_paca(&paca[cpu], cpu); > } > > diff --git a/arch/powerpc/kernel/prom.c b/arch/powerpc/kernel/prom.c > index f83056297441..93837093c5cb 100644 > --- a/arch/powerpc/kernel/prom.c > +++ b/arch/powerpc/kernel/prom.c > @@ -302,6 +302,29 @@ static void __init check_cpu_feature_properties(unsigned long node) > } > } > > +/* > + * Adjust the logical id of a boot cpu to fall under nr_cpu_ids. Map it to > + * last core slot in the allocated paca array. > + * > + * e.g. on SMT=8 system, kernel booted with nr_cpus=1 and boot cpu = 33, > + * align nr_cpu_ids to MAX_SMT value 8. Allocate paca array to hold up-to > + * MAX_SMT=8 cpus. Since boot cpu 33 is greater than nr_cpus (8), adjust > + * its logical id so that new id becomes less than nr_cpu_ids. Make sure > + * that boot cpu's new logical id is aligned to its thread id and falls > + * under last nthreads slots available in paca array. In this case the > + * boot cpu 33 is adjusted to new boot cpu id 1. > + * > + */ > +static inline void adjust_boot_cpuid(int nthreads, int phys_id) > +{ > + boot_hw_cpuid = phys_id; > + if (boot_cpuid >= nr_cpu_ids) { > + boot_cpuid = (boot_cpuid % nthreads) + (nr_cpu_ids - nthreads); > + pr_info("Adjusted logical boot cpu id: logical %d physical %d\n", > + boot_cpuid, phys_id); > + } > +} > + > static int __init early_init_dt_scan_cpus(unsigned long node, > const char *uname, int depth, > void *data) > @@ -325,6 +348,18 @@ static int __init early_init_dt_scan_cpus(unsigned long node, > > nthreads = len / sizeof(int); > > +#ifdef CONFIG_SMP > + /* > + * Now that we know threads per core lets align nr_cpu_ids to > + * correct SMT value. > + */ > + if (nr_cpu_ids % nthreads) { > + nr_cpu_ids = _ALIGN_UP(nr_cpu_ids, nthreads); > + pr_info("Aligned nr_cpus to SMT=%d, nr_cpu_ids = %d\n", > + nthreads, nr_cpu_ids); > + } > +#endif > + > /* > * Now see if any of these threads match our boot cpu. > * NOTE: This must match the parsing done in smp_setup_cpu_maps. > @@ -363,7 +398,9 @@ static int __init early_init_dt_scan_cpus(unsigned long node, > DBG("boot cpu: logical %d physical %d\n", found, > be32_to_cpu(intserv[found_thread])); > boot_cpuid = found; > - set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread])); > + adjust_boot_cpuid(nthreads, be32_to_cpu(intserv[found_thread])); > + set_hard_smp_processor_id(boot_cpuid, > + be32_to_cpu(intserv[found_thread])); > > /* > * PAPR defines "logical" PVR values for cpus that > diff --git a/arch/powerpc/kernel/setup-common.c b/arch/powerpc/kernel/setup-common.c > index 2e3bc16d02b2..46b1e0972a20 100644 > --- a/arch/powerpc/kernel/setup-common.c > +++ b/arch/powerpc/kernel/setup-common.c > @@ -85,6 +85,7 @@ struct machdep_calls *machine_id; > EXPORT_SYMBOL(machine_id); > > int boot_cpuid = -1; > +int boot_hw_cpuid = -1; > EXPORT_SYMBOL_GPL(boot_cpuid); > > /* > @@ -473,6 +474,7 @@ void __init smp_setup_cpu_maps(void) > struct device_node *dn = NULL; > int cpu = 0; > int nthreads = 1; > + bool boot_cpu_added = false; > > DBG("smp_setup_cpu_maps()\n"); > > @@ -499,6 +501,24 @@ void __init smp_setup_cpu_maps(void) > } > > nthreads = len / sizeof(int); > + /* > + * If boot cpu hasn't been added to paca and there are only > + * last nthreads slots available in paca array then wait > + * for boot cpu to show up. > + */ > + if (!boot_cpu_added && (cpu + nthreads) >= nr_cpu_ids) { > + int found = 0; > + > + DBG("Holding last nthreads paca slots for boot cpu\n"); > + for (j = 0; j < nthreads && cpu < nr_cpu_ids; j++) { > + if (boot_hw_cpuid == be32_to_cpu(intserv[j])) { > + found = 1; > + break; > + } > + } > + if (!found) > + continue; > + } > > for (j = 0; j < nthreads && cpu < nr_cpu_ids; j++) { > bool avail; > @@ -514,6 +534,11 @@ void __init smp_setup_cpu_maps(void) > set_cpu_present(cpu, avail); > set_hard_smp_processor_id(cpu, be32_to_cpu(intserv[j])); > set_cpu_possible(cpu, true); > + if (boot_hw_cpuid == be32_to_cpu(intserv[j])) { > + DBG("Boot cpu %d (hard id %d) added to paca\n", > + cpu, be32_to_cpu(intserv[j])); > + boot_cpu_added = true; > + } > cpu++; > } > } >
On Thu, Nov 16, 2017 at 07:26:43PM -0200, Guilherme G. Piccoli wrote: > On 11/06/2017 03:34 PM, Thadeu Lima de Souza Cascardo wrote: > > From: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com> > > > > The kernel boot parameter 'nr_cpus=' allows one to specify number of > > possible cpus in the system. In the normal scenario the first cpu (cpu0) > > that shows up is the boot cpu and hence it gets covered under nr_cpus > > limit. > > > > But this assumption will be broken in kdump scenario where kdump kenrel > > Minor nit: s/kenrel/kernel > > > after a crash can boot up on an non-zero boot cpu. The paca structure > > allocation depends on value of nr_cpus and is indexed using logical cpu > > ids. This definetly will be an issue if boot cpu id > nr_cpus > > Minor nit (2): s/definetly/definitely Hi, Guilherme. Thanks a lot for testing and reviewing the patch. I will ignore those small typos, unless the maintainers tell me to fix them. I hope you don't mind. I would like to see this merged sooner rather than later. Thanks. Cascardo. > > > > This patch modifies allocate_pacas() and smp_setup_cpu_maps() to > > accommodate boot cpu for the case where boot_cpuid > nr_cpu_ids. > > > > This change would help to reduce the memory reservation requirement for > > kdump on ppc64. > > > > Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com> > > Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com> > > Tested-by: Guilherme G. Piccoli <gpiccoli@linux.vnet.ibm.com> > > Without this patch, got a crash deterministically when booting with > nr_cpus=1 in P8 bare-metal. The patch fixes the issue... > > Thanks, > > > Guilherme > > > > --- > > > > v3: fixup signedness or nr_cpus to match nr_cpu_ids > > and fix conflict due to change from %d to %u > > > > Resending this as it was not applied, and I can reproduce the issue with > > v4.14-rc8 when booting a kdump kernel after a crash that has been given > > nr_cpus=1 as a parameter. With this patch, I can't reproduce it anymore. > > > > --- > > arch/powerpc/include/asm/paca.h | 3 +++ > > arch/powerpc/include/asm/smp.h | 1 + > > arch/powerpc/kernel/paca.c | 23 +++++++++++++++++----- > > arch/powerpc/kernel/prom.c | 39 +++++++++++++++++++++++++++++++++++++- > > arch/powerpc/kernel/setup-common.c | 25 ++++++++++++++++++++++++ > > 5 files changed, 85 insertions(+), 6 deletions(-) > > > > diff --git a/arch/powerpc/include/asm/paca.h b/arch/powerpc/include/asm/paca.h > > index 04b60af027ae..ea0dbf2bbeef 100644 > > --- a/arch/powerpc/include/asm/paca.h > > +++ b/arch/powerpc/include/asm/paca.h > > @@ -49,6 +49,9 @@ extern unsigned int debug_smp_processor_id(void); /* from linux/smp.h */ > > #define get_lppaca() (get_paca()->lppaca_ptr) > > #define get_slb_shadow() (get_paca()->slb_shadow_ptr) > > > > +/* Maximum number of threads per core. */ > > +#define MAX_SMT 8 > > + > > struct task_struct; > > > > /* > > diff --git a/arch/powerpc/include/asm/smp.h b/arch/powerpc/include/asm/smp.h > > index fac963e10d39..553cd22b2ccc 100644 > > --- a/arch/powerpc/include/asm/smp.h > > +++ b/arch/powerpc/include/asm/smp.h > > @@ -30,6 +30,7 @@ > > #include <asm/percpu.h> > > > > extern int boot_cpuid; > > +extern int boot_hw_cpuid; > > extern int spinning_secondaries; > > > > extern void cpu_die(void); > > diff --git a/arch/powerpc/kernel/paca.c b/arch/powerpc/kernel/paca.c > > index 2ff2b8a19f71..9c689ee4b6a3 100644 > > --- a/arch/powerpc/kernel/paca.c > > +++ b/arch/powerpc/kernel/paca.c > > @@ -207,6 +207,7 @@ void __init allocate_pacas(void) > > { > > u64 limit; > > int cpu; > > + unsigned int nr_cpus; > > > > limit = ppc64_rma_size; > > > > @@ -219,20 +220,32 @@ void __init allocate_pacas(void) > > limit = min(0x10000000ULL, limit); > > #endif > > > > - paca_size = PAGE_ALIGN(sizeof(struct paca_struct) * nr_cpu_ids); > > + /* > > + * Always align up the nr_cpu_ids to SMT threads and allocate > > + * the paca. This will help us to prepare for a situation where > > + * boot cpu id > nr_cpus_id. We will use the last nthreads > > + * slots (nthreads == threads per core) to accommodate a core > > + * that contains boot cpu thread. > > + * > > + * Do not change nr_cpu_ids value here. Let us do that in > > + * early_init_dt_scan_cpus() where we know exact value > > + * of threads per core. > > + */ > > + nr_cpus = _ALIGN_UP(nr_cpu_ids, MAX_SMT); > > + paca_size = PAGE_ALIGN(sizeof(struct paca_struct) * nr_cpus); > > > > paca = __va(memblock_alloc_base(paca_size, PAGE_SIZE, limit)); > > memset(paca, 0, paca_size); > > > > printk(KERN_DEBUG "Allocated %u bytes for %u pacas at %p\n", > > - paca_size, nr_cpu_ids, paca); > > + paca_size, nr_cpus, paca); > > > > - allocate_lppacas(nr_cpu_ids, limit); > > + allocate_lppacas(nr_cpus, limit); > > > > - allocate_slb_shadows(nr_cpu_ids, limit); > > + allocate_slb_shadows(nr_cpus, limit); > > > > /* Can't use for_each_*_cpu, as they aren't functional yet */ > > - for (cpu = 0; cpu < nr_cpu_ids; cpu++) > > + for (cpu = 0; cpu < nr_cpus; cpu++) > > initialise_paca(&paca[cpu], cpu); > > } > > > > diff --git a/arch/powerpc/kernel/prom.c b/arch/powerpc/kernel/prom.c > > index f83056297441..93837093c5cb 100644 > > --- a/arch/powerpc/kernel/prom.c > > +++ b/arch/powerpc/kernel/prom.c > > @@ -302,6 +302,29 @@ static void __init check_cpu_feature_properties(unsigned long node) > > } > > } > > > > +/* > > + * Adjust the logical id of a boot cpu to fall under nr_cpu_ids. Map it to > > + * last core slot in the allocated paca array. > > + * > > + * e.g. on SMT=8 system, kernel booted with nr_cpus=1 and boot cpu = 33, > > + * align nr_cpu_ids to MAX_SMT value 8. Allocate paca array to hold up-to > > + * MAX_SMT=8 cpus. Since boot cpu 33 is greater than nr_cpus (8), adjust > > + * its logical id so that new id becomes less than nr_cpu_ids. Make sure > > + * that boot cpu's new logical id is aligned to its thread id and falls > > + * under last nthreads slots available in paca array. In this case the > > + * boot cpu 33 is adjusted to new boot cpu id 1. > > + * > > + */ > > +static inline void adjust_boot_cpuid(int nthreads, int phys_id) > > +{ > > + boot_hw_cpuid = phys_id; > > + if (boot_cpuid >= nr_cpu_ids) { > > + boot_cpuid = (boot_cpuid % nthreads) + (nr_cpu_ids - nthreads); > > + pr_info("Adjusted logical boot cpu id: logical %d physical %d\n", > > + boot_cpuid, phys_id); > > + } > > +} > > + > > static int __init early_init_dt_scan_cpus(unsigned long node, > > const char *uname, int depth, > > void *data) > > @@ -325,6 +348,18 @@ static int __init early_init_dt_scan_cpus(unsigned long node, > > > > nthreads = len / sizeof(int); > > > > +#ifdef CONFIG_SMP > > + /* > > + * Now that we know threads per core lets align nr_cpu_ids to > > + * correct SMT value. > > + */ > > + if (nr_cpu_ids % nthreads) { > > + nr_cpu_ids = _ALIGN_UP(nr_cpu_ids, nthreads); > > + pr_info("Aligned nr_cpus to SMT=%d, nr_cpu_ids = %d\n", > > + nthreads, nr_cpu_ids); > > + } > > +#endif > > + > > /* > > * Now see if any of these threads match our boot cpu. > > * NOTE: This must match the parsing done in smp_setup_cpu_maps. > > @@ -363,7 +398,9 @@ static int __init early_init_dt_scan_cpus(unsigned long node, > > DBG("boot cpu: logical %d physical %d\n", found, > > be32_to_cpu(intserv[found_thread])); > > boot_cpuid = found; > > - set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread])); > > + adjust_boot_cpuid(nthreads, be32_to_cpu(intserv[found_thread])); > > + set_hard_smp_processor_id(boot_cpuid, > > + be32_to_cpu(intserv[found_thread])); > > > > /* > > * PAPR defines "logical" PVR values for cpus that > > diff --git a/arch/powerpc/kernel/setup-common.c b/arch/powerpc/kernel/setup-common.c > > index 2e3bc16d02b2..46b1e0972a20 100644 > > --- a/arch/powerpc/kernel/setup-common.c > > +++ b/arch/powerpc/kernel/setup-common.c > > @@ -85,6 +85,7 @@ struct machdep_calls *machine_id; > > EXPORT_SYMBOL(machine_id); > > > > int boot_cpuid = -1; > > +int boot_hw_cpuid = -1; > > EXPORT_SYMBOL_GPL(boot_cpuid); > > > > /* > > @@ -473,6 +474,7 @@ void __init smp_setup_cpu_maps(void) > > struct device_node *dn = NULL; > > int cpu = 0; > > int nthreads = 1; > > + bool boot_cpu_added = false; > > > > DBG("smp_setup_cpu_maps()\n"); > > > > @@ -499,6 +501,24 @@ void __init smp_setup_cpu_maps(void) > > } > > > > nthreads = len / sizeof(int); > > + /* > > + * If boot cpu hasn't been added to paca and there are only > > + * last nthreads slots available in paca array then wait > > + * for boot cpu to show up. > > + */ > > + if (!boot_cpu_added && (cpu + nthreads) >= nr_cpu_ids) { > > + int found = 0; > > + > > + DBG("Holding last nthreads paca slots for boot cpu\n"); > > + for (j = 0; j < nthreads && cpu < nr_cpu_ids; j++) { > > + if (boot_hw_cpuid == be32_to_cpu(intserv[j])) { > > + found = 1; > > + break; > > + } > > + } > > + if (!found) > > + continue; > > + } > > > > for (j = 0; j < nthreads && cpu < nr_cpu_ids; j++) { > > bool avail; > > @@ -514,6 +534,11 @@ void __init smp_setup_cpu_maps(void) > > set_cpu_present(cpu, avail); > > set_hard_smp_processor_id(cpu, be32_to_cpu(intserv[j])); > > set_cpu_possible(cpu, true); > > + if (boot_hw_cpuid == be32_to_cpu(intserv[j])) { > > + DBG("Boot cpu %d (hard id %d) added to paca\n", > > + cpu, be32_to_cpu(intserv[j])); > > + boot_cpu_added = true; > > + } > > cpu++; > > } > > } > > >
On 11/17/2017 03:26 PM, Thadeu Lima de Souza Cascardo wrote: > On Thu, Nov 16, 2017 at 07:26:43PM -0200, Guilherme G. Piccoli wrote: >> On 11/06/2017 03:34 PM, Thadeu Lima de Souza Cascardo wrote: >>> From: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com> >>> >>> The kernel boot parameter 'nr_cpus=' allows one to specify number of >>> possible cpus in the system. In the normal scenario the first cpu (cpu0) >>> that shows up is the boot cpu and hence it gets covered under nr_cpus >>> limit. >>> >>> But this assumption will be broken in kdump scenario where kdump kenrel >> >> Minor nit: s/kenrel/kernel >> >>> after a crash can boot up on an non-zero boot cpu. The paca structure >>> allocation depends on value of nr_cpus and is indexed using logical cpu >>> ids. This definetly will be an issue if boot cpu id > nr_cpus >> >> Minor nit (2): s/definetly/definitely > > Hi, Guilherme. > > Thanks a lot for testing and reviewing the patch. I will ignore those > small typos, unless the maintainers tell me to fix them. I hope you > don't mind. > > I would like to see this merged sooner rather than later. Sure Cascardo! In fact, those kind of minor/small nits could be fixed by Michael once/if he merges heheh I don't think it'd need a respin! Cheers, Guilherme > > Thanks. > Cascardo. > >>> >>> This patch modifies allocate_pacas() and smp_setup_cpu_maps() to >>> accommodate boot cpu for the case where boot_cpuid > nr_cpu_ids. >>> >>> This change would help to reduce the memory reservation requirement for >>> kdump on ppc64. >>> >>> Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com> >>> Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com> >> >> Tested-by: Guilherme G. Piccoli <gpiccoli@linux.vnet.ibm.com> >> >> Without this patch, got a crash deterministically when booting with >> nr_cpus=1 in P8 bare-metal. The patch fixes the issue... >> >> Thanks, >> >> >> Guilherme >> >> >>> --- >>> >>> v3: fixup signedness or nr_cpus to match nr_cpu_ids >>> and fix conflict due to change from %d to %u >>> >>> Resending this as it was not applied, and I can reproduce the issue with >>> v4.14-rc8 when booting a kdump kernel after a crash that has been given >>> nr_cpus=1 as a parameter. With this patch, I can't reproduce it anymore. >>> >>> --- >>> arch/powerpc/include/asm/paca.h | 3 +++ >>> arch/powerpc/include/asm/smp.h | 1 + >>> arch/powerpc/kernel/paca.c | 23 +++++++++++++++++----- >>> arch/powerpc/kernel/prom.c | 39 +++++++++++++++++++++++++++++++++++++- >>> arch/powerpc/kernel/setup-common.c | 25 ++++++++++++++++++++++++ >>> 5 files changed, 85 insertions(+), 6 deletions(-) >>> >>> diff --git a/arch/powerpc/include/asm/paca.h b/arch/powerpc/include/asm/paca.h >>> index 04b60af027ae..ea0dbf2bbeef 100644 >>> --- a/arch/powerpc/include/asm/paca.h >>> +++ b/arch/powerpc/include/asm/paca.h >>> @@ -49,6 +49,9 @@ extern unsigned int debug_smp_processor_id(void); /* from linux/smp.h */ >>> #define get_lppaca() (get_paca()->lppaca_ptr) >>> #define get_slb_shadow() (get_paca()->slb_shadow_ptr) >>> >>> +/* Maximum number of threads per core. */ >>> +#define MAX_SMT 8 >>> + >>> struct task_struct; >>> >>> /* >>> diff --git a/arch/powerpc/include/asm/smp.h b/arch/powerpc/include/asm/smp.h >>> index fac963e10d39..553cd22b2ccc 100644 >>> --- a/arch/powerpc/include/asm/smp.h >>> +++ b/arch/powerpc/include/asm/smp.h >>> @@ -30,6 +30,7 @@ >>> #include <asm/percpu.h> >>> >>> extern int boot_cpuid; >>> +extern int boot_hw_cpuid; >>> extern int spinning_secondaries; >>> >>> extern void cpu_die(void); >>> diff --git a/arch/powerpc/kernel/paca.c b/arch/powerpc/kernel/paca.c >>> index 2ff2b8a19f71..9c689ee4b6a3 100644 >>> --- a/arch/powerpc/kernel/paca.c >>> +++ b/arch/powerpc/kernel/paca.c >>> @@ -207,6 +207,7 @@ void __init allocate_pacas(void) >>> { >>> u64 limit; >>> int cpu; >>> + unsigned int nr_cpus; >>> >>> limit = ppc64_rma_size; >>> >>> @@ -219,20 +220,32 @@ void __init allocate_pacas(void) >>> limit = min(0x10000000ULL, limit); >>> #endif >>> >>> - paca_size = PAGE_ALIGN(sizeof(struct paca_struct) * nr_cpu_ids); >>> + /* >>> + * Always align up the nr_cpu_ids to SMT threads and allocate >>> + * the paca. This will help us to prepare for a situation where >>> + * boot cpu id > nr_cpus_id. We will use the last nthreads >>> + * slots (nthreads == threads per core) to accommodate a core >>> + * that contains boot cpu thread. >>> + * >>> + * Do not change nr_cpu_ids value here. Let us do that in >>> + * early_init_dt_scan_cpus() where we know exact value >>> + * of threads per core. >>> + */ >>> + nr_cpus = _ALIGN_UP(nr_cpu_ids, MAX_SMT); >>> + paca_size = PAGE_ALIGN(sizeof(struct paca_struct) * nr_cpus); >>> >>> paca = __va(memblock_alloc_base(paca_size, PAGE_SIZE, limit)); >>> memset(paca, 0, paca_size); >>> >>> printk(KERN_DEBUG "Allocated %u bytes for %u pacas at %p\n", >>> - paca_size, nr_cpu_ids, paca); >>> + paca_size, nr_cpus, paca); >>> >>> - allocate_lppacas(nr_cpu_ids, limit); >>> + allocate_lppacas(nr_cpus, limit); >>> >>> - allocate_slb_shadows(nr_cpu_ids, limit); >>> + allocate_slb_shadows(nr_cpus, limit); >>> >>> /* Can't use for_each_*_cpu, as they aren't functional yet */ >>> - for (cpu = 0; cpu < nr_cpu_ids; cpu++) >>> + for (cpu = 0; cpu < nr_cpus; cpu++) >>> initialise_paca(&paca[cpu], cpu); >>> } >>> >>> diff --git a/arch/powerpc/kernel/prom.c b/arch/powerpc/kernel/prom.c >>> index f83056297441..93837093c5cb 100644 >>> --- a/arch/powerpc/kernel/prom.c >>> +++ b/arch/powerpc/kernel/prom.c >>> @@ -302,6 +302,29 @@ static void __init check_cpu_feature_properties(unsigned long node) >>> } >>> } >>> >>> +/* >>> + * Adjust the logical id of a boot cpu to fall under nr_cpu_ids. Map it to >>> + * last core slot in the allocated paca array. >>> + * >>> + * e.g. on SMT=8 system, kernel booted with nr_cpus=1 and boot cpu = 33, >>> + * align nr_cpu_ids to MAX_SMT value 8. Allocate paca array to hold up-to >>> + * MAX_SMT=8 cpus. Since boot cpu 33 is greater than nr_cpus (8), adjust >>> + * its logical id so that new id becomes less than nr_cpu_ids. Make sure >>> + * that boot cpu's new logical id is aligned to its thread id and falls >>> + * under last nthreads slots available in paca array. In this case the >>> + * boot cpu 33 is adjusted to new boot cpu id 1. >>> + * >>> + */ >>> +static inline void adjust_boot_cpuid(int nthreads, int phys_id) >>> +{ >>> + boot_hw_cpuid = phys_id; >>> + if (boot_cpuid >= nr_cpu_ids) { >>> + boot_cpuid = (boot_cpuid % nthreads) + (nr_cpu_ids - nthreads); >>> + pr_info("Adjusted logical boot cpu id: logical %d physical %d\n", >>> + boot_cpuid, phys_id); >>> + } >>> +} >>> + >>> static int __init early_init_dt_scan_cpus(unsigned long node, >>> const char *uname, int depth, >>> void *data) >>> @@ -325,6 +348,18 @@ static int __init early_init_dt_scan_cpus(unsigned long node, >>> >>> nthreads = len / sizeof(int); >>> >>> +#ifdef CONFIG_SMP >>> + /* >>> + * Now that we know threads per core lets align nr_cpu_ids to >>> + * correct SMT value. >>> + */ >>> + if (nr_cpu_ids % nthreads) { >>> + nr_cpu_ids = _ALIGN_UP(nr_cpu_ids, nthreads); >>> + pr_info("Aligned nr_cpus to SMT=%d, nr_cpu_ids = %d\n", >>> + nthreads, nr_cpu_ids); >>> + } >>> +#endif >>> + >>> /* >>> * Now see if any of these threads match our boot cpu. >>> * NOTE: This must match the parsing done in smp_setup_cpu_maps. >>> @@ -363,7 +398,9 @@ static int __init early_init_dt_scan_cpus(unsigned long node, >>> DBG("boot cpu: logical %d physical %d\n", found, >>> be32_to_cpu(intserv[found_thread])); >>> boot_cpuid = found; >>> - set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread])); >>> + adjust_boot_cpuid(nthreads, be32_to_cpu(intserv[found_thread])); >>> + set_hard_smp_processor_id(boot_cpuid, >>> + be32_to_cpu(intserv[found_thread])); >>> >>> /* >>> * PAPR defines "logical" PVR values for cpus that >>> diff --git a/arch/powerpc/kernel/setup-common.c b/arch/powerpc/kernel/setup-common.c >>> index 2e3bc16d02b2..46b1e0972a20 100644 >>> --- a/arch/powerpc/kernel/setup-common.c >>> +++ b/arch/powerpc/kernel/setup-common.c >>> @@ -85,6 +85,7 @@ struct machdep_calls *machine_id; >>> EXPORT_SYMBOL(machine_id); >>> >>> int boot_cpuid = -1; >>> +int boot_hw_cpuid = -1; >>> EXPORT_SYMBOL_GPL(boot_cpuid); >>> >>> /* >>> @@ -473,6 +474,7 @@ void __init smp_setup_cpu_maps(void) >>> struct device_node *dn = NULL; >>> int cpu = 0; >>> int nthreads = 1; >>> + bool boot_cpu_added = false; >>> >>> DBG("smp_setup_cpu_maps()\n"); >>> >>> @@ -499,6 +501,24 @@ void __init smp_setup_cpu_maps(void) >>> } >>> >>> nthreads = len / sizeof(int); >>> + /* >>> + * If boot cpu hasn't been added to paca and there are only >>> + * last nthreads slots available in paca array then wait >>> + * for boot cpu to show up. >>> + */ >>> + if (!boot_cpu_added && (cpu + nthreads) >= nr_cpu_ids) { >>> + int found = 0; >>> + >>> + DBG("Holding last nthreads paca slots for boot cpu\n"); >>> + for (j = 0; j < nthreads && cpu < nr_cpu_ids; j++) { >>> + if (boot_hw_cpuid == be32_to_cpu(intserv[j])) { >>> + found = 1; >>> + break; >>> + } >>> + } >>> + if (!found) >>> + continue; >>> + } >>> >>> for (j = 0; j < nthreads && cpu < nr_cpu_ids; j++) { >>> bool avail; >>> @@ -514,6 +534,11 @@ void __init smp_setup_cpu_maps(void) >>> set_cpu_present(cpu, avail); >>> set_hard_smp_processor_id(cpu, be32_to_cpu(intserv[j])); >>> set_cpu_possible(cpu, true); >>> + if (boot_hw_cpuid == be32_to_cpu(intserv[j])) { >>> + DBG("Boot cpu %d (hard id %d) added to paca\n", >>> + cpu, be32_to_cpu(intserv[j])); >>> + boot_cpu_added = true; >>> + } >>> cpu++; >>> } >>> } >>> >> >
Ping. Any further reviews/comments? Or any chance of this getting applied soon? Thanks. Cascardo.
On 11/29/2017 10:40 AM, Thadeu Lima de Souza Cascardo wrote: > Ping. Any further reviews/comments? Or any chance of this getting > applied soon? > > Thanks. > Cascardo. > Cross-referencing 2 threads on the same topic: https://lists.ozlabs.org/pipermail/linuxppc-dev/2017-November/166328.html Liu ping fan asked the state of this patch in the above link of mailing list thread. Let's choose one of the threads to further discuss the solution to this long-term bug! Thanks, Guilherme
diff --git a/arch/powerpc/include/asm/paca.h b/arch/powerpc/include/asm/paca.h index 04b60af027ae..ea0dbf2bbeef 100644 --- a/arch/powerpc/include/asm/paca.h +++ b/arch/powerpc/include/asm/paca.h @@ -49,6 +49,9 @@ extern unsigned int debug_smp_processor_id(void); /* from linux/smp.h */ #define get_lppaca() (get_paca()->lppaca_ptr) #define get_slb_shadow() (get_paca()->slb_shadow_ptr) +/* Maximum number of threads per core. */ +#define MAX_SMT 8 + struct task_struct; /* diff --git a/arch/powerpc/include/asm/smp.h b/arch/powerpc/include/asm/smp.h index fac963e10d39..553cd22b2ccc 100644 --- a/arch/powerpc/include/asm/smp.h +++ b/arch/powerpc/include/asm/smp.h @@ -30,6 +30,7 @@ #include <asm/percpu.h> extern int boot_cpuid; +extern int boot_hw_cpuid; extern int spinning_secondaries; extern void cpu_die(void); diff --git a/arch/powerpc/kernel/paca.c b/arch/powerpc/kernel/paca.c index 2ff2b8a19f71..9c689ee4b6a3 100644 --- a/arch/powerpc/kernel/paca.c +++ b/arch/powerpc/kernel/paca.c @@ -207,6 +207,7 @@ void __init allocate_pacas(void) { u64 limit; int cpu; + unsigned int nr_cpus; limit = ppc64_rma_size; @@ -219,20 +220,32 @@ void __init allocate_pacas(void) limit = min(0x10000000ULL, limit); #endif - paca_size = PAGE_ALIGN(sizeof(struct paca_struct) * nr_cpu_ids); + /* + * Always align up the nr_cpu_ids to SMT threads and allocate + * the paca. This will help us to prepare for a situation where + * boot cpu id > nr_cpus_id. We will use the last nthreads + * slots (nthreads == threads per core) to accommodate a core + * that contains boot cpu thread. + * + * Do not change nr_cpu_ids value here. Let us do that in + * early_init_dt_scan_cpus() where we know exact value + * of threads per core. + */ + nr_cpus = _ALIGN_UP(nr_cpu_ids, MAX_SMT); + paca_size = PAGE_ALIGN(sizeof(struct paca_struct) * nr_cpus); paca = __va(memblock_alloc_base(paca_size, PAGE_SIZE, limit)); memset(paca, 0, paca_size); printk(KERN_DEBUG "Allocated %u bytes for %u pacas at %p\n", - paca_size, nr_cpu_ids, paca); + paca_size, nr_cpus, paca); - allocate_lppacas(nr_cpu_ids, limit); + allocate_lppacas(nr_cpus, limit); - allocate_slb_shadows(nr_cpu_ids, limit); + allocate_slb_shadows(nr_cpus, limit); /* Can't use for_each_*_cpu, as they aren't functional yet */ - for (cpu = 0; cpu < nr_cpu_ids; cpu++) + for (cpu = 0; cpu < nr_cpus; cpu++) initialise_paca(&paca[cpu], cpu); } diff --git a/arch/powerpc/kernel/prom.c b/arch/powerpc/kernel/prom.c index f83056297441..93837093c5cb 100644 --- a/arch/powerpc/kernel/prom.c +++ b/arch/powerpc/kernel/prom.c @@ -302,6 +302,29 @@ static void __init check_cpu_feature_properties(unsigned long node) } } +/* + * Adjust the logical id of a boot cpu to fall under nr_cpu_ids. Map it to + * last core slot in the allocated paca array. + * + * e.g. on SMT=8 system, kernel booted with nr_cpus=1 and boot cpu = 33, + * align nr_cpu_ids to MAX_SMT value 8. Allocate paca array to hold up-to + * MAX_SMT=8 cpus. Since boot cpu 33 is greater than nr_cpus (8), adjust + * its logical id so that new id becomes less than nr_cpu_ids. Make sure + * that boot cpu's new logical id is aligned to its thread id and falls + * under last nthreads slots available in paca array. In this case the + * boot cpu 33 is adjusted to new boot cpu id 1. + * + */ +static inline void adjust_boot_cpuid(int nthreads, int phys_id) +{ + boot_hw_cpuid = phys_id; + if (boot_cpuid >= nr_cpu_ids) { + boot_cpuid = (boot_cpuid % nthreads) + (nr_cpu_ids - nthreads); + pr_info("Adjusted logical boot cpu id: logical %d physical %d\n", + boot_cpuid, phys_id); + } +} + static int __init early_init_dt_scan_cpus(unsigned long node, const char *uname, int depth, void *data) @@ -325,6 +348,18 @@ static int __init early_init_dt_scan_cpus(unsigned long node, nthreads = len / sizeof(int); +#ifdef CONFIG_SMP + /* + * Now that we know threads per core lets align nr_cpu_ids to + * correct SMT value. + */ + if (nr_cpu_ids % nthreads) { + nr_cpu_ids = _ALIGN_UP(nr_cpu_ids, nthreads); + pr_info("Aligned nr_cpus to SMT=%d, nr_cpu_ids = %d\n", + nthreads, nr_cpu_ids); + } +#endif + /* * Now see if any of these threads match our boot cpu. * NOTE: This must match the parsing done in smp_setup_cpu_maps. @@ -363,7 +398,9 @@ static int __init early_init_dt_scan_cpus(unsigned long node, DBG("boot cpu: logical %d physical %d\n", found, be32_to_cpu(intserv[found_thread])); boot_cpuid = found; - set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread])); + adjust_boot_cpuid(nthreads, be32_to_cpu(intserv[found_thread])); + set_hard_smp_processor_id(boot_cpuid, + be32_to_cpu(intserv[found_thread])); /* * PAPR defines "logical" PVR values for cpus that diff --git a/arch/powerpc/kernel/setup-common.c b/arch/powerpc/kernel/setup-common.c index 2e3bc16d02b2..46b1e0972a20 100644 --- a/arch/powerpc/kernel/setup-common.c +++ b/arch/powerpc/kernel/setup-common.c @@ -85,6 +85,7 @@ struct machdep_calls *machine_id; EXPORT_SYMBOL(machine_id); int boot_cpuid = -1; +int boot_hw_cpuid = -1; EXPORT_SYMBOL_GPL(boot_cpuid); /* @@ -473,6 +474,7 @@ void __init smp_setup_cpu_maps(void) struct device_node *dn = NULL; int cpu = 0; int nthreads = 1; + bool boot_cpu_added = false; DBG("smp_setup_cpu_maps()\n"); @@ -499,6 +501,24 @@ void __init smp_setup_cpu_maps(void) } nthreads = len / sizeof(int); + /* + * If boot cpu hasn't been added to paca and there are only + * last nthreads slots available in paca array then wait + * for boot cpu to show up. + */ + if (!boot_cpu_added && (cpu + nthreads) >= nr_cpu_ids) { + int found = 0; + + DBG("Holding last nthreads paca slots for boot cpu\n"); + for (j = 0; j < nthreads && cpu < nr_cpu_ids; j++) { + if (boot_hw_cpuid == be32_to_cpu(intserv[j])) { + found = 1; + break; + } + } + if (!found) + continue; + } for (j = 0; j < nthreads && cpu < nr_cpu_ids; j++) { bool avail; @@ -514,6 +534,11 @@ void __init smp_setup_cpu_maps(void) set_cpu_present(cpu, avail); set_hard_smp_processor_id(cpu, be32_to_cpu(intserv[j])); set_cpu_possible(cpu, true); + if (boot_hw_cpuid == be32_to_cpu(intserv[j])) { + DBG("Boot cpu %d (hard id %d) added to paca\n", + cpu, be32_to_cpu(intserv[j])); + boot_cpu_added = true; + } cpu++; } }