| Message ID | 1350404684-6883-1-git-send-email-lorenzo.pieralisi@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tuesday 16 October 2012 09:54 PM, Lorenzo Pieralisi wrote: > Currently, reading /proc/cpuinfo provides userspace with CPU ID of > the CPU carrying out the read from the file. This is fine as long as all > CPUs in the system are the same. With the advent of big.LITTLE and > heterogenous ARM systems this approach provides user space with incorrect > bits of information since CPU ids in the system might differ from the one > provided by the CPU reading the file. > > This patch updates the cpuinfo show function and some internal data > structures so that a read from /proc/cpuinfo prints HW information for > all online CPUs at once, mirroring x86 behaviour. > > Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> > --- > > Posting it as a way to get advice on the best way to improve /proc/cpuinfo for > heterogenous multi-cluster systems and to provide proper information to > userspace without breaking existing semantics, no more than that. > > Comments more than welcome. > > Thanks, > Lorenzo > > arch/arm/include/asm/cpu.h | 1 + > arch/arm/kernel/setup.c | 66 +++++++++++++++++++++++----------------------- > arch/arm/kernel/smp.c | 1 + > 3 files changed, 35 insertions(+), 33 deletions(-) > [...] > diff --git a/arch/arm/kernel/setup.c b/arch/arm/kernel/setup.c > index 8707cff..bf7839d08 100644 > --- a/arch/arm/kernel/setup.c > +++ b/arch/arm/kernel/setup.c > @@ -1054,13 +1054,15 @@ static const char *hwcap_str[] = { > > static int c_show(struct seq_file *m, void *v) > { > - int i; > + int i, j; > + u32 cpuid; > > - seq_printf(m, "Processor\t: %s rev %d (%s)\n", > - cpu_name, read_cpuid_id() & 15, elf_platform); > + for_each_online_cpu(i) { > + cpuid = is_smp() ? per_cpu(cpu_data, i).cpuid : read_cpuid_id(); > + seq_printf(m, "Processor\t: %s rev %d (%s)\n", > + cpu_name, cpuid & 15, elf_platform); > Not exactly related to the $subject patch, but I remember doing a patch to have cat /proc/cpuinfo spitting only online CPUs just like x86 using for_each_online_cpu(i). At that point Russell mentioned about a possibility of read() syscall spreading over the hot-plug operation and hence the above may not be safe. is that right Russell ? regards Santosh
On Tue, Oct 16, 2012 at 11:29:39PM +0530, Santosh Shilimkar wrote: > Not exactly related to the $subject patch, but I remember doing a patch > to have cat /proc/cpuinfo spitting only online CPUs just like x86 using > for_each_online_cpu(i). > At that point Russell mentioned about a possibility of read() syscall > spreading over the hot-plug operation and hence the above may not > be safe. > > is that right Russell ? That is correct, but you will notice that the code now uses the online cpus rather than the present, inspite of my objections. I gave up fighting the case, so now people get to live with the consequences of this. This will be fun with dynamic hotplugging with big.LITTLE when people come to read any of these files which change their output with the online CPUs. Oh what fun we're in for there. I guess the only way to convince people is to let them make their mistakes in the kernel.
On Tue, Oct 16, 2012 at 10:47:52PM +0100, Russell King - ARM Linux wrote: > On Tue, Oct 16, 2012 at 11:29:39PM +0530, Santosh Shilimkar wrote: > > Not exactly related to the $subject patch, but I remember doing a patch > > to have cat /proc/cpuinfo spitting only online CPUs just like x86 using > > for_each_online_cpu(i). > > At that point Russell mentioned about a possibility of read() syscall > > spreading over the hot-plug operation and hence the above may not > > be safe. > > > > is that right Russell ? > > That is correct, but you will notice that the code now uses the online > cpus rather than the present, inspite of my objections. I gave up > fighting the case, so now people get to live with the consequences of > this. > > This will be fun with dynamic hotplugging with big.LITTLE when people > come to read any of these files which change their output with the > online CPUs. Oh what fun we're in for there. I had a look at how glibc currently detects cpus in __get_nprocs() and all methods end up relying on online CPUs: - /sys/devices/system/cpu/online - /proc/stat - /proc/cpuinfo Given the thread: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/054081.html I do not think there is room to change the current behaviour of /proc/cpuinfo as far as the online vs. present argument goes, and I certainly did not try. I posted this code just to understand if it makes sense to expose all CPU ids in /proc/cpuinfo the way I did (the same way x86 does, even though the code is slightly different). > I guess the only way to convince people is to let them make their > mistakes in the kernel. That's what I wanted to prevent by posting this code, namely getting feedback on the best way to improve /proc/cpuinfo for systems where CPU ids are not homogeneous. The "online vs. present" issue is already there as Russell pointed out. Thank you very much for the review, any further feedback appreciated. Lorenzo
diff --git a/arch/arm/include/asm/cpu.h b/arch/arm/include/asm/cpu.h index d797223..2744f06 100644 --- a/arch/arm/include/asm/cpu.h +++ b/arch/arm/include/asm/cpu.h @@ -15,6 +15,7 @@ struct cpuinfo_arm { struct cpu cpu; + u32 cpuid; #ifdef CONFIG_SMP unsigned int loops_per_jiffy; #endif diff --git a/arch/arm/kernel/setup.c b/arch/arm/kernel/setup.c index 8707cff..bf7839d08 100644 --- a/arch/arm/kernel/setup.c +++ b/arch/arm/kernel/setup.c @@ -1054,13 +1054,15 @@ static const char *hwcap_str[] = { static int c_show(struct seq_file *m, void *v) { - int i; + int i, j; + u32 cpuid; - seq_printf(m, "Processor\t: %s rev %d (%s)\n", - cpu_name, read_cpuid_id() & 15, elf_platform); + for_each_online_cpu(i) { + cpuid = is_smp() ? per_cpu(cpu_data, i).cpuid : read_cpuid_id(); + seq_printf(m, "Processor\t: %s rev %d (%s)\n", + cpu_name, cpuid & 15, elf_platform); #if defined(CONFIG_SMP) - for_each_online_cpu(i) { /* * glibc reads /proc/cpuinfo to determine the number of * online processors, looking for lines beginning with @@ -1070,42 +1072,40 @@ static int c_show(struct seq_file *m, void *v) seq_printf(m, "BogoMIPS\t: %lu.%02lu\n\n", per_cpu(cpu_data, i).loops_per_jiffy / (500000UL/HZ), (per_cpu(cpu_data, i).loops_per_jiffy / (5000UL/HZ)) % 100); - } -#else /* CONFIG_SMP */ - seq_printf(m, "BogoMIPS\t: %lu.%02lu\n", - loops_per_jiffy / (500000/HZ), - (loops_per_jiffy / (5000/HZ)) % 100); +#else + seq_printf(m, "BogoMIPS\t: %lu.%02lu\n", + loops_per_jiffy / (500000/HZ), + (loops_per_jiffy / (5000/HZ)) % 100); #endif + /* dump out the processor features */ + seq_puts(m, "Features\t: "); - /* dump out the processor features */ - seq_puts(m, "Features\t: "); + for (j = 0; hwcap_str[j]; j++) + if (elf_hwcap & (1 << j)) + seq_printf(m, "%s ", hwcap_str[j]); - for (i = 0; hwcap_str[i]; i++) - if (elf_hwcap & (1 << i)) - seq_printf(m, "%s ", hwcap_str[i]); + seq_printf(m, "\nCPU implementer\t: 0x%02x\n", cpuid >> 24); + seq_printf(m, "CPU architecture: %s\n", + proc_arch[cpu_architecture()]); - seq_printf(m, "\nCPU implementer\t: 0x%02x\n", read_cpuid_id() >> 24); - seq_printf(m, "CPU architecture: %s\n", proc_arch[cpu_architecture()]); - - if ((read_cpuid_id() & 0x0008f000) == 0x00000000) { - /* pre-ARM7 */ - seq_printf(m, "CPU part\t: %07x\n", read_cpuid_id() >> 4); - } else { - if ((read_cpuid_id() & 0x0008f000) == 0x00007000) { - /* ARM7 */ - seq_printf(m, "CPU variant\t: 0x%02x\n", - (read_cpuid_id() >> 16) & 127); + if ((cpuid & 0x0008f000) == 0x00000000) { + /* pre-ARM7 */ + seq_printf(m, "CPU part\t: %07x\n", cpuid >> 4); } else { - /* post-ARM7 */ - seq_printf(m, "CPU variant\t: 0x%x\n", - (read_cpuid_id() >> 20) & 15); + if ((cpuid & 0x0008f000) == 0x00007000) { + /* ARM7 */ + seq_printf(m, "CPU variant\t: 0x%02x\n", + (cpuid >> 16) & 127); + } else { + /* post-ARM7 */ + seq_printf(m, "CPU variant\t: 0x%x\n", + (cpuid >> 20) & 15); + } + seq_printf(m, "CPU part\t: 0x%03x\n", + (cpuid >> 4) & 0xfff); } - seq_printf(m, "CPU part\t: 0x%03x\n", - (read_cpuid_id() >> 4) & 0xfff); + seq_printf(m, "CPU revision\t: %d\n\n", cpuid & 15); } - seq_printf(m, "CPU revision\t: %d\n", read_cpuid_id() & 15); - - seq_puts(m, "\n"); seq_printf(m, "Hardware\t: %s\n", machine_name); seq_printf(m, "Revision\t: %04x\n", system_rev); diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c index f0bcb1e..73f407e 100644 --- a/arch/arm/kernel/smp.c +++ b/arch/arm/kernel/smp.c @@ -211,6 +211,7 @@ static void __cpuinit smp_store_cpu_info(unsigned int cpuid) struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid); cpu_info->loops_per_jiffy = loops_per_jiffy; + cpu_info->cpuid = read_cpuid_id(); store_cpu_topology(cpuid); }
Currently, reading /proc/cpuinfo provides userspace with CPU ID of the CPU carrying out the read from the file. This is fine as long as all CPUs in the system are the same. With the advent of big.LITTLE and heterogenous ARM systems this approach provides user space with incorrect bits of information since CPU ids in the system might differ from the one provided by the CPU reading the file. This patch updates the cpuinfo show function and some internal data structures so that a read from /proc/cpuinfo prints HW information for all online CPUs at once, mirroring x86 behaviour. Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> --- Posting it as a way to get advice on the best way to improve /proc/cpuinfo for heterogenous multi-cluster systems and to provide proper information to userspace without breaking existing semantics, no more than that. Comments more than welcome. Thanks, Lorenzo arch/arm/include/asm/cpu.h | 1 + arch/arm/kernel/setup.c | 66 +++++++++++++++++++++++----------------------- arch/arm/kernel/smp.c | 1 + 3 files changed, 35 insertions(+), 33 deletions(-)