Hello,
This patch series adds a new x86 arch specific BPF helper, bpf_rdtsc() which can be used for reading the hardware time stamp counter (TSC.) Currently the same counter is directly accessible from userspace (using RDTSC instruction), and kernel space using various rdtsc_*() APIs, however eBPF lacks the support.
The main usage for the TSC counter is for various profiling and timing purposes, getting accurate cycle counter values. The counter can be currently read from BPF programs by using the existing perf subsystem services (bpf_perf_event_read()), however its usage is cumbersome at best. Additionally, the perf subsystem provides relative value only for the counter, but absolute values are desired by some use cases like Wult [1]. The absolute value of TSC can be read with BPF programs currently via some kprobe / bpf_core_read() magic (see [2], [3], [4] for example), but this relies on accessing kernel internals and is not stable API, and is pretty cumbersome. Thus, this patch proposes a new arch x86 specific BPF helper to avoid the above issues.
-Tero
[1] https://github.com/intel/wult [2] https://github.com/intel/wult/blob/c92237c95b898498faf41e6644983102d1fe5156/... [3] https://github.com/intel/wult/blob/c92237c95b898498faf41e6644983102d1fe5156/... [4] https://github.com/intel/wult/blob/c92237c95b898498faf41e6644983102d1fe5156/...
Currently the raw TSC counter can be read within kernel via rdtsc_ordered() and friends, and additionally even userspace has access to it via the RDTSC assembly instruction. BPF programs on the other hand don't have direct access to the TSC counter, but alternatively must go through the performance subsystem (bpf_perf_event_read), which only provides relative value compared to the start point of the program, and is also much slower than the direct read. Add a new BPF helper definition for bpf_rdtsc() which can be used for any accurate profiling needs.
A use-case for the new API is for example wakeup latency tracing via eBPF on Intel architecture, where it is extremely beneficial to be able to get raw TSC timestamps and compare these directly to the value programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct latency value from the hardware interrupt to the execution of the interrupt handler can be calculated. Having the functionality within eBPF also has added benefits of allowing to filter any other relevant data like C-state residency values, and also to drop any irrelevant data points directly in the kernel context, without passing all the data to userspace for post-processing.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com --- arch/x86/include/asm/msr.h | 1 + arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h index 65ec1965cd28..3dde673cb563 100644 --- a/arch/x86/include/asm/msr.h +++ b/arch/x86/include/asm/msr.h @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); void msrs_free(struct msr *msrs); int msr_set_bit(u32 msr, u8 bit); int msr_clear_bit(u32 msr, u8 bit); +u64 bpf_rdtsc(void);
#ifdef CONFIG_SMP int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 344698852146..ded857abef81 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -15,6 +15,8 @@ #include <linux/timex.h> #include <linux/static_key.h> #include <linux/static_call.h> +#include <linux/btf.h> +#include <linux/btf_ids.h>
#include <asm/hpet.h> #include <asm/timer.h> @@ -29,6 +31,7 @@ #include <asm/intel-family.h> #include <asm/i8259.h> #include <asm/uv/uv.h> +#include <asm/tlbflush.h>
unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ EXPORT_SYMBOL(cpu_khz); @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) tsc_enable_sched_clock(); }
+u64 bpf_rdtsc(void) +{ + /* Check if Time Stamp is enabled only in ring 0 */ + if (cr4_read_shadow() & X86_CR4_TSD) + return 0; + + return rdtsc_ordered(); +} + +BTF_SET8_START(tsc_bpf_kfunc_ids) +BTF_ID_FLAGS(func, bpf_rdtsc) +BTF_SET8_END(tsc_bpf_kfunc_ids) + +static const struct btf_kfunc_id_set tsc_bpf_kfunc_set = { + .owner = THIS_MODULE, + .set = &tsc_bpf_kfunc_ids, +}; + void __init tsc_init(void) { if (!cpu_feature_enabled(X86_FEATURE_TSC)) { @@ -1594,6 +1615,8 @@ void __init tsc_init(void)
clocksource_register_khz(&clocksource_tsc_early, tsc_khz); detect_art(); + + register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &tsc_bpf_kfunc_set); }
#ifdef CONFIG_SMP
On 7/3/23 3:57 AM, Tero Kristo wrote:
Currently the raw TSC counter can be read within kernel via rdtsc_ordered() and friends, and additionally even userspace has access to it via the RDTSC assembly instruction. BPF programs on the other hand don't have direct access to the TSC counter, but alternatively must go through the performance subsystem (bpf_perf_event_read), which only provides relative value compared to the start point of the program, and is also much slower than the direct read. Add a new BPF helper definition for bpf_rdtsc() which can be used for any accurate profiling needs.
A use-case for the new API is for example wakeup latency tracing via eBPF on Intel architecture, where it is extremely beneficial to be able to get raw TSC timestamps and compare these directly to the value programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct latency value from the hardware interrupt to the execution of the interrupt handler can be calculated. Having the functionality within eBPF also has added benefits of allowing to filter any other relevant data like C-state residency values, and also to drop any irrelevant data points directly in the kernel context, without passing all the data to userspace for post-processing.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
arch/x86/include/asm/msr.h | 1 + arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h index 65ec1965cd28..3dde673cb563 100644 --- a/arch/x86/include/asm/msr.h +++ b/arch/x86/include/asm/msr.h @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); void msrs_free(struct msr *msrs); int msr_set_bit(u32 msr, u8 bit); int msr_clear_bit(u32 msr, u8 bit); +u64 bpf_rdtsc(void); #ifdef CONFIG_SMP int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 344698852146..ded857abef81 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -15,6 +15,8 @@ #include <linux/timex.h> #include <linux/static_key.h> #include <linux/static_call.h> +#include <linux/btf.h> +#include <linux/btf_ids.h> #include <asm/hpet.h> #include <asm/timer.h> @@ -29,6 +31,7 @@ #include <asm/intel-family.h> #include <asm/i8259.h> #include <asm/uv/uv.h> +#include <asm/tlbflush.h> unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ EXPORT_SYMBOL(cpu_khz); @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) tsc_enable_sched_clock(); } +u64 bpf_rdtsc(void)
Please see kernel/bpf/helpers.c. For kfunc definition, we should have
__diag_push(); __diag_ignore_all("-Wmissing-prototypes", "Global functions as their definitions will be in vmlinux BTF");
_bpf_kfunc u64 bpf_rdtsc(void) { ... }
__diag_pop();
+{
- /* Check if Time Stamp is enabled only in ring 0 */
- if (cr4_read_shadow() & X86_CR4_TSD)
return 0;- return rdtsc_ordered();
+}
+BTF_SET8_START(tsc_bpf_kfunc_ids) +BTF_ID_FLAGS(func, bpf_rdtsc) +BTF_SET8_END(tsc_bpf_kfunc_ids)
+static const struct btf_kfunc_id_set tsc_bpf_kfunc_set = {
- .owner = THIS_MODULE,
- .set = &tsc_bpf_kfunc_ids,
+};
- void __init tsc_init(void) { if (!cpu_feature_enabled(X86_FEATURE_TSC)) {
@@ -1594,6 +1615,8 @@ void __init tsc_init(void) clocksource_register_khz(&clocksource_tsc_early, tsc_khz); detect_art();
- register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &tsc_bpf_kfunc_set);
register_btf_kfunc_id_set() could fail, maybe you at least wants to have a warning so bpf prog users may be aware that kfunc bpf_rdtsc() not really available to bpf programs?
} #ifdef CONFIG_SMP
On 04/07/2023 07:49, Yonghong Song wrote:
On 7/3/23 3:57 AM, Tero Kristo wrote:
Currently the raw TSC counter can be read within kernel via rdtsc_ordered() and friends, and additionally even userspace has access to it via the RDTSC assembly instruction. BPF programs on the other hand don't have direct access to the TSC counter, but alternatively must go through the performance subsystem (bpf_perf_event_read), which only provides relative value compared to the start point of the program, and is also much slower than the direct read. Add a new BPF helper definition for bpf_rdtsc() which can be used for any accurate profiling needs.
A use-case for the new API is for example wakeup latency tracing via eBPF on Intel architecture, where it is extremely beneficial to be able to get raw TSC timestamps and compare these directly to the value programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct latency value from the hardware interrupt to the execution of the interrupt handler can be calculated. Having the functionality within eBPF also has added benefits of allowing to filter any other relevant data like C-state residency values, and also to drop any irrelevant data points directly in the kernel context, without passing all the data to userspace for post-processing.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
arch/x86/include/asm/msr.h | 1 + arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h index 65ec1965cd28..3dde673cb563 100644 --- a/arch/x86/include/asm/msr.h +++ b/arch/x86/include/asm/msr.h @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); void msrs_free(struct msr *msrs); int msr_set_bit(u32 msr, u8 bit); int msr_clear_bit(u32 msr, u8 bit); +u64 bpf_rdtsc(void); #ifdef CONFIG_SMP int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 344698852146..ded857abef81 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -15,6 +15,8 @@ #include <linux/timex.h> #include <linux/static_key.h> #include <linux/static_call.h> +#include <linux/btf.h> +#include <linux/btf_ids.h> #include <asm/hpet.h> #include <asm/timer.h> @@ -29,6 +31,7 @@ #include <asm/intel-family.h> #include <asm/i8259.h> #include <asm/uv/uv.h> +#include <asm/tlbflush.h> unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ EXPORT_SYMBOL(cpu_khz); @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) tsc_enable_sched_clock(); } +u64 bpf_rdtsc(void)
Please see kernel/bpf/helpers.c. For kfunc definition, we should have
__diag_push(); __diag_ignore_all("-Wmissing-prototypes", "Global functions as their definitions will be in vmlinux BTF");
_bpf_kfunc u64 bpf_rdtsc(void) { ... }
__diag_pop();
Thanks, I'll modify this for next rev.
+{ + /* Check if Time Stamp is enabled only in ring 0 */ + if (cr4_read_shadow() & X86_CR4_TSD) + return 0;
+ return rdtsc_ordered(); +}
+BTF_SET8_START(tsc_bpf_kfunc_ids) +BTF_ID_FLAGS(func, bpf_rdtsc) +BTF_SET8_END(tsc_bpf_kfunc_ids)
+static const struct btf_kfunc_id_set tsc_bpf_kfunc_set = { + .owner = THIS_MODULE, + .set = &tsc_bpf_kfunc_ids, +};
void __init tsc_init(void) { if (!cpu_feature_enabled(X86_FEATURE_TSC)) { @@ -1594,6 +1615,8 @@ void __init tsc_init(void) clocksource_register_khz(&clocksource_tsc_early, tsc_khz); detect_art();
+ register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &tsc_bpf_kfunc_set);
register_btf_kfunc_id_set() could fail, maybe you at least wants to have a warning so bpf prog users may be aware that kfunc bpf_rdtsc() not really available to bpf programs?
Yes, I'll add a warning print.
-Tero
} #ifdef CONFIG_SMP
On Mon, Jul 3, 2023 at 3:58 AM Tero Kristo tero.kristo@linux.intel.com wrote:
Currently the raw TSC counter can be read within kernel via rdtsc_ordered() and friends, and additionally even userspace has access to it via the RDTSC assembly instruction. BPF programs on the other hand don't have direct access to the TSC counter, but alternatively must go through the performance subsystem (bpf_perf_event_read), which only provides relative value compared to the start point of the program, and is also much slower than the direct read. Add a new BPF helper definition for bpf_rdtsc() which can be used for any accurate profiling needs.
A use-case for the new API is for example wakeup latency tracing via eBPF on Intel architecture, where it is extremely beneficial to be able to get raw TSC timestamps and compare these directly to the value programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct latency value from the hardware interrupt to the execution of the interrupt handler can be calculated. Having the functionality within eBPF also has added benefits of allowing to filter any other relevant data like C-state residency values, and also to drop any irrelevant data points directly in the kernel context, without passing all the data to userspace for post-processing.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
arch/x86/include/asm/msr.h | 1 + arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h index 65ec1965cd28..3dde673cb563 100644 --- a/arch/x86/include/asm/msr.h +++ b/arch/x86/include/asm/msr.h @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); void msrs_free(struct msr *msrs); int msr_set_bit(u32 msr, u8 bit); int msr_clear_bit(u32 msr, u8 bit); +u64 bpf_rdtsc(void);
#ifdef CONFIG_SMP int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 344698852146..ded857abef81 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -15,6 +15,8 @@ #include <linux/timex.h> #include <linux/static_key.h> #include <linux/static_call.h> +#include <linux/btf.h> +#include <linux/btf_ids.h>
#include <asm/hpet.h> #include <asm/timer.h> @@ -29,6 +31,7 @@ #include <asm/intel-family.h> #include <asm/i8259.h> #include <asm/uv/uv.h> +#include <asm/tlbflush.h>
unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ EXPORT_SYMBOL(cpu_khz); @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) tsc_enable_sched_clock(); }
+u64 bpf_rdtsc(void) +{
/* Check if Time Stamp is enabled only in ring 0 */if (cr4_read_shadow() & X86_CR4_TSD)return 0;
Why check this? It's always enabled in the kernel, no?
return rdtsc_ordered();
Why _ordered? Why not just rdtsc ? Especially since you want to trace latency. Extra lfence will ruin the measurements.
Alexei Starovoitov wrote:
On Mon, Jul 3, 2023 at 3:58 AM Tero Kristo tero.kristo@linux.intel.com wrote:
Currently the raw TSC counter can be read within kernel via rdtsc_ordered() and friends, and additionally even userspace has access to it via the RDTSC assembly instruction. BPF programs on the other hand don't have direct access to the TSC counter, but alternatively must go through the performance subsystem (bpf_perf_event_read), which only provides relative value compared to the start point of the program, and is also much slower than the direct read. Add a new BPF helper definition for bpf_rdtsc() which can be used for any accurate profiling needs.
A use-case for the new API is for example wakeup latency tracing via eBPF on Intel architecture, where it is extremely beneficial to be able to get raw TSC timestamps and compare these directly to the value programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct latency value from the hardware interrupt to the execution of the interrupt handler can be calculated. Having the functionality within eBPF also has added benefits of allowing to filter any other relevant data like C-state residency values, and also to drop any irrelevant data points directly in the kernel context, without passing all the data to userspace for post-processing.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
arch/x86/include/asm/msr.h | 1 + arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h index 65ec1965cd28..3dde673cb563 100644 --- a/arch/x86/include/asm/msr.h +++ b/arch/x86/include/asm/msr.h @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); void msrs_free(struct msr *msrs); int msr_set_bit(u32 msr, u8 bit); int msr_clear_bit(u32 msr, u8 bit); +u64 bpf_rdtsc(void);
#ifdef CONFIG_SMP int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 344698852146..ded857abef81 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -15,6 +15,8 @@ #include <linux/timex.h> #include <linux/static_key.h> #include <linux/static_call.h> +#include <linux/btf.h> +#include <linux/btf_ids.h>
#include <asm/hpet.h> #include <asm/timer.h> @@ -29,6 +31,7 @@ #include <asm/intel-family.h> #include <asm/i8259.h> #include <asm/uv/uv.h> +#include <asm/tlbflush.h>
unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ EXPORT_SYMBOL(cpu_khz); @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) tsc_enable_sched_clock(); }
+u64 bpf_rdtsc(void) +{
/* Check if Time Stamp is enabled only in ring 0 */if (cr4_read_shadow() & X86_CR4_TSD)return 0;Why check this? It's always enabled in the kernel, no?
return rdtsc_ordered();Why _ordered? Why not just rdtsc ? Especially since you want to trace latency. Extra lfence will ruin the measurements.
If we used it as a fast way to order events on multiple CPUs I guess we need the lfence? We use ktime_get_ns() now for things like this when we just need an order counter. We have also observed time going backwards with this and have heuristics to correct it but its rare.
On 06/07/2023 08:16, John Fastabend wrote:
Alexei Starovoitov wrote:
On Mon, Jul 3, 2023 at 3:58 AM Tero Kristo tero.kristo@linux.intel.com wrote:
Currently the raw TSC counter can be read within kernel via rdtsc_ordered() and friends, and additionally even userspace has access to it via the RDTSC assembly instruction. BPF programs on the other hand don't have direct access to the TSC counter, but alternatively must go through the performance subsystem (bpf_perf_event_read), which only provides relative value compared to the start point of the program, and is also much slower than the direct read. Add a new BPF helper definition for bpf_rdtsc() which can be used for any accurate profiling needs.
A use-case for the new API is for example wakeup latency tracing via eBPF on Intel architecture, where it is extremely beneficial to be able to get raw TSC timestamps and compare these directly to the value programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct latency value from the hardware interrupt to the execution of the interrupt handler can be calculated. Having the functionality within eBPF also has added benefits of allowing to filter any other relevant data like C-state residency values, and also to drop any irrelevant data points directly in the kernel context, without passing all the data to userspace for post-processing.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
arch/x86/include/asm/msr.h | 1 + arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h index 65ec1965cd28..3dde673cb563 100644 --- a/arch/x86/include/asm/msr.h +++ b/arch/x86/include/asm/msr.h @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); void msrs_free(struct msr *msrs); int msr_set_bit(u32 msr, u8 bit); int msr_clear_bit(u32 msr, u8 bit); +u64 bpf_rdtsc(void);
#ifdef CONFIG_SMP int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 344698852146..ded857abef81 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -15,6 +15,8 @@ #include <linux/timex.h> #include <linux/static_key.h> #include <linux/static_call.h> +#include <linux/btf.h> +#include <linux/btf_ids.h>
#include <asm/hpet.h> #include <asm/timer.h> @@ -29,6 +31,7 @@ #include <asm/intel-family.h> #include <asm/i8259.h> #include <asm/uv/uv.h> +#include <asm/tlbflush.h>
unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ EXPORT_SYMBOL(cpu_khz); @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) tsc_enable_sched_clock(); }
+u64 bpf_rdtsc(void) +{
/* Check if Time Stamp is enabled only in ring 0 */if (cr4_read_shadow() & X86_CR4_TSD)return 0;Why check this? It's always enabled in the kernel, no?
It is always enabled, but there are certain syscalls that can be used to disable the TSC access for oneself. prctl(PR_SET_TSC, ...) and seccomp(SET_MODE_STRICT,...). Not having the check in place would in theory allow a restricted BPF program to circumvent this (if there ever was such a thing.) But yes, I do agree this part is a bit debatable whether it should be there at all.
return rdtsc_ordered();Why _ordered? Why not just rdtsc ? Especially since you want to trace latency. Extra lfence will ruin the measurements.
If we used it as a fast way to order events on multiple CPUs I guess we need the lfence? We use ktime_get_ns() now for things like this when we just need an order counter. We have also observed time going backwards with this and have heuristics to correct it but its rare.
Yeah, I think it is better to induce some extra latency instead of having some weird ordering issues with the timestamps.
Also, things like the ftrace also use rdtsc_ordered() as its underlying clock, if you use x86-tsc as the trace clock (see arch/x86/kernel/trace_clock.c.)
-Tero
On Thu, Jul 6, 2023 at 4:59 AM Tero Kristo tero.kristo@linux.intel.com wrote:
On 06/07/2023 08:16, John Fastabend wrote:
Alexei Starovoitov wrote:
On Mon, Jul 3, 2023 at 3:58 AM Tero Kristo tero.kristo@linux.intel.com wrote:
Currently the raw TSC counter can be read within kernel via rdtsc_ordered() and friends, and additionally even userspace has access to it via the RDTSC assembly instruction. BPF programs on the other hand don't have direct access to the TSC counter, but alternatively must go through the performance subsystem (bpf_perf_event_read), which only provides relative value compared to the start point of the program, and is also much slower than the direct read. Add a new BPF helper definition for bpf_rdtsc() which can be used for any accurate profiling needs.
A use-case for the new API is for example wakeup latency tracing via eBPF on Intel architecture, where it is extremely beneficial to be able to get raw TSC timestamps and compare these directly to the value programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct latency value from the hardware interrupt to the execution of the interrupt handler can be calculated. Having the functionality within eBPF also has added benefits of allowing to filter any other relevant data like C-state residency values, and also to drop any irrelevant data points directly in the kernel context, without passing all the data to userspace for post-processing.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
arch/x86/include/asm/msr.h | 1 + arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h index 65ec1965cd28..3dde673cb563 100644 --- a/arch/x86/include/asm/msr.h +++ b/arch/x86/include/asm/msr.h @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); void msrs_free(struct msr *msrs); int msr_set_bit(u32 msr, u8 bit); int msr_clear_bit(u32 msr, u8 bit); +u64 bpf_rdtsc(void);
#ifdef CONFIG_SMP int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 344698852146..ded857abef81 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -15,6 +15,8 @@ #include <linux/timex.h> #include <linux/static_key.h> #include <linux/static_call.h> +#include <linux/btf.h> +#include <linux/btf_ids.h>
#include <asm/hpet.h> #include <asm/timer.h> @@ -29,6 +31,7 @@ #include <asm/intel-family.h> #include <asm/i8259.h> #include <asm/uv/uv.h> +#include <asm/tlbflush.h>
unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ EXPORT_SYMBOL(cpu_khz); @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) tsc_enable_sched_clock(); }
+u64 bpf_rdtsc(void) +{
/* Check if Time Stamp is enabled only in ring 0 */if (cr4_read_shadow() & X86_CR4_TSD)return 0;Why check this? It's always enabled in the kernel, no?
It is always enabled, but there are certain syscalls that can be used to disable the TSC access for oneself. prctl(PR_SET_TSC, ...) and seccomp(SET_MODE_STRICT,...). Not having the check in place would in theory allow a restricted BPF program to circumvent this (if there ever was such a thing.) But yes, I do agree this part is a bit debatable whether it should be there at all.
What do you mean 'circumvent' ? It's a tracing bpf prog running in the kernel loaded by root and reading tsc for the purpose of the kernel. There is no unprivileged access to tsc here.
return rdtsc_ordered();Why _ordered? Why not just rdtsc ? Especially since you want to trace latency. Extra lfence will ruin the measurements.
If we used it as a fast way to order events on multiple CPUs I guess we need the lfence? We use ktime_get_ns() now for things like this when we just need an order counter. We have also observed time going backwards with this and have heuristics to correct it but its rare.
Yeah, I think it is better to induce some extra latency instead of having some weird ordering issues with the timestamps.
lfence is not 'some extra latency'. I suspect rdtsc_ordered() will be slower than bpf_ktime_get_ns(). What's the point of using it then?
Also, things like the ftrace also use rdtsc_ordered() as its underlying clock, if you use x86-tsc as the trace clock (see arch/x86/kernel/trace_clock.c.)
-Tero
Alexei Starovoitov wrote:
On Thu, Jul 6, 2023 at 4:59 AM Tero Kristo tero.kristo@linux.intel.com wrote:
On 06/07/2023 08:16, John Fastabend wrote:
Alexei Starovoitov wrote:
On Mon, Jul 3, 2023 at 3:58 AM Tero Kristo tero.kristo@linux.intel.com wrote:
Currently the raw TSC counter can be read within kernel via rdtsc_ordered() and friends, and additionally even userspace has access to it via the RDTSC assembly instruction. BPF programs on the other hand don't have direct access to the TSC counter, but alternatively must go through the performance subsystem (bpf_perf_event_read), which only provides relative value compared to the start point of the program, and is also much slower than the direct read. Add a new BPF helper definition for bpf_rdtsc() which can be used for any accurate profiling needs.
A use-case for the new API is for example wakeup latency tracing via eBPF on Intel architecture, where it is extremely beneficial to be able to get raw TSC timestamps and compare these directly to the value programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct latency value from the hardware interrupt to the execution of the interrupt handler can be calculated. Having the functionality within eBPF also has added benefits of allowing to filter any other relevant data like C-state residency values, and also to drop any irrelevant data points directly in the kernel context, without passing all the data to userspace for post-processing.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
arch/x86/include/asm/msr.h | 1 + arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h index 65ec1965cd28..3dde673cb563 100644 --- a/arch/x86/include/asm/msr.h +++ b/arch/x86/include/asm/msr.h @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); void msrs_free(struct msr *msrs); int msr_set_bit(u32 msr, u8 bit); int msr_clear_bit(u32 msr, u8 bit); +u64 bpf_rdtsc(void);
#ifdef CONFIG_SMP int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 344698852146..ded857abef81 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -15,6 +15,8 @@ #include <linux/timex.h> #include <linux/static_key.h> #include <linux/static_call.h> +#include <linux/btf.h> +#include <linux/btf_ids.h>
#include <asm/hpet.h> #include <asm/timer.h> @@ -29,6 +31,7 @@ #include <asm/intel-family.h> #include <asm/i8259.h> #include <asm/uv/uv.h> +#include <asm/tlbflush.h>
unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ EXPORT_SYMBOL(cpu_khz); @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) tsc_enable_sched_clock(); }
+u64 bpf_rdtsc(void) +{
/* Check if Time Stamp is enabled only in ring 0 */if (cr4_read_shadow() & X86_CR4_TSD)return 0;Why check this? It's always enabled in the kernel, no?
It is always enabled, but there are certain syscalls that can be used to disable the TSC access for oneself. prctl(PR_SET_TSC, ...) and seccomp(SET_MODE_STRICT,...). Not having the check in place would in theory allow a restricted BPF program to circumvent this (if there ever was such a thing.) But yes, I do agree this part is a bit debatable whether it should be there at all.
What do you mean 'circumvent' ? It's a tracing bpf prog running in the kernel loaded by root and reading tsc for the purpose of the kernel. There is no unprivileged access to tsc here.
return rdtsc_ordered();Why _ordered? Why not just rdtsc ? Especially since you want to trace latency. Extra lfence will ruin the measurements.
If we used it as a fast way to order events on multiple CPUs I guess we need the lfence? We use ktime_get_ns() now for things like this when we just need an order counter. We have also observed time going backwards with this and have heuristics to correct it but its rare.
Yeah, I think it is better to induce some extra latency instead of having some weird ordering issues with the timestamps.
lfence is not 'some extra latency'. I suspect rdtsc_ordered() will be slower than bpf_ktime_get_ns(). What's the point of using it then?
I would only use it if its faster then bpf_ktime_get_ns() and have already figured out how to handle rare unordered events so I think its OK to relax somewhat strict ordering.
Also, things like the ftrace also use rdtsc_ordered() as its underlying clock, if you use x86-tsc as the trace clock (see arch/x86/kernel/trace_clock.c.)
-Tero
On 07/07/2023 08:41, John Fastabend wrote:
Alexei Starovoitov wrote:
On Thu, Jul 6, 2023 at 4:59 AM Tero Kristo tero.kristo@linux.intel.com wrote:
On 06/07/2023 08:16, John Fastabend wrote:
Alexei Starovoitov wrote:
On Mon, Jul 3, 2023 at 3:58 AM Tero Kristo tero.kristo@linux.intel.com wrote:
Currently the raw TSC counter can be read within kernel via rdtsc_ordered() and friends, and additionally even userspace has access to it via the RDTSC assembly instruction. BPF programs on the other hand don't have direct access to the TSC counter, but alternatively must go through the performance subsystem (bpf_perf_event_read), which only provides relative value compared to the start point of the program, and is also much slower than the direct read. Add a new BPF helper definition for bpf_rdtsc() which can be used for any accurate profiling needs.
A use-case for the new API is for example wakeup latency tracing via eBPF on Intel architecture, where it is extremely beneficial to be able to get raw TSC timestamps and compare these directly to the value programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct latency value from the hardware interrupt to the execution of the interrupt handler can be calculated. Having the functionality within eBPF also has added benefits of allowing to filter any other relevant data like C-state residency values, and also to drop any irrelevant data points directly in the kernel context, without passing all the data to userspace for post-processing.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
arch/x86/include/asm/msr.h | 1 + arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h index 65ec1965cd28..3dde673cb563 100644 --- a/arch/x86/include/asm/msr.h +++ b/arch/x86/include/asm/msr.h @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); void msrs_free(struct msr *msrs); int msr_set_bit(u32 msr, u8 bit); int msr_clear_bit(u32 msr, u8 bit); +u64 bpf_rdtsc(void);
#ifdef CONFIG_SMP int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 344698852146..ded857abef81 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -15,6 +15,8 @@ #include <linux/timex.h> #include <linux/static_key.h> #include <linux/static_call.h> +#include <linux/btf.h> +#include <linux/btf_ids.h>
#include <asm/hpet.h> #include <asm/timer.h> @@ -29,6 +31,7 @@ #include <asm/intel-family.h> #include <asm/i8259.h> #include <asm/uv/uv.h> +#include <asm/tlbflush.h>
unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ EXPORT_SYMBOL(cpu_khz); @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) tsc_enable_sched_clock(); }
+u64 bpf_rdtsc(void) +{
/* Check if Time Stamp is enabled only in ring 0 */if (cr4_read_shadow() & X86_CR4_TSD)return 0;Why check this? It's always enabled in the kernel, no?
It is always enabled, but there are certain syscalls that can be used to disable the TSC access for oneself. prctl(PR_SET_TSC, ...) and seccomp(SET_MODE_STRICT,...). Not having the check in place would in theory allow a restricted BPF program to circumvent this (if there ever was such a thing.) But yes, I do agree this part is a bit debatable whether it should be there at all.
What do you mean 'circumvent' ? It's a tracing bpf prog running in the kernel loaded by root and reading tsc for the purpose of the kernel. There is no unprivileged access to tsc here.
This was based on some discussions with the security team at Intel, I don't pretend to know anything about security myself. But I can drop the check. It is probably not needed because of the fact that it is already possible to read the TSC counter with the approach I mention in the cover letter; via perf and bpf_core_read().
return rdtsc_ordered();Why _ordered? Why not just rdtsc ? Especially since you want to trace latency. Extra lfence will ruin the measurements.
If we used it as a fast way to order events on multiple CPUs I guess we need the lfence? We use ktime_get_ns() now for things like this when we just need an order counter. We have also observed time going backwards with this and have heuristics to correct it but its rare.
Yeah, I think it is better to induce some extra latency instead of having some weird ordering issues with the timestamps.
lfence is not 'some extra latency'. I suspect rdtsc_ordered() will be slower than bpf_ktime_get_ns(). What's the point of using it then?
I would only use it if its faster then bpf_ktime_get_ns() and have already figured out how to handle rare unordered events so I think its OK to relax somewhat strict ordering.
I believe that on x86-arch using bpf_ktime_get_ns() also ends up calling rdtsc_odered() under the hood.
I just did some measurements on an Intel(R) Xeon(R) Platinum 8360Y CPU @ 2.40GHz, with a simple BPF code:
t1 = bpf_ktime_get_ns();
for (i = 0; i < NUM_CYC; i++) { bpf_rdtsc(); // or bpf_ktime_get_ns() here }
t2 = bpf_ktime_get_ns();
The results I got with the CPU locked at 2.4GHz (average execution times per a call within the loop, this with some 10M executions):
bpf_rdtsc() ordered : 45ns
bpf_rdtsc() un-ordered : 23ns
bpf_ktime_get_ns() : 49ns
Locking the CPU at 800MHz the results are:
bpf_rdtsc() ordered : 55ns
bpf_rdtsc() un-ordered : 33ns
bpf_ktime_get_ns() : 71ns
The bpf_rdtsc() in these results contains some extra latency caused by conditional execution, I added a flag to the call to select whether it should use _ordered() or not, and it also still contains the CR4_TSD check in place.
-Tero
Also, things like the ftrace also use rdtsc_ordered() as its underlying clock, if you use x86-tsc as the trace clock (see arch/x86/kernel/trace_clock.c.)
-Tero
On Fri, Jul 7, 2023 at 1:28 AM Tero Kristo tero.kristo@linux.intel.com wrote:
On 07/07/2023 08:41, John Fastabend wrote:
Alexei Starovoitov wrote:
On Thu, Jul 6, 2023 at 4:59 AM Tero Kristo tero.kristo@linux.intel.com wrote:
On 06/07/2023 08:16, John Fastabend wrote:
Alexei Starovoitov wrote:
On Mon, Jul 3, 2023 at 3:58 AM Tero Kristo tero.kristo@linux.intel.com wrote: > Currently the raw TSC counter can be read within kernel via rdtsc_ordered() > and friends, and additionally even userspace has access to it via the > RDTSC assembly instruction. BPF programs on the other hand don't have > direct access to the TSC counter, but alternatively must go through the > performance subsystem (bpf_perf_event_read), which only provides relative > value compared to the start point of the program, and is also much slower > than the direct read. Add a new BPF helper definition for bpf_rdtsc() which > can be used for any accurate profiling needs. > > A use-case for the new API is for example wakeup latency tracing via > eBPF on Intel architecture, where it is extremely beneficial to be able > to get raw TSC timestamps and compare these directly to the value > programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct > latency value from the hardware interrupt to the execution of the > interrupt handler can be calculated. Having the functionality within > eBPF also has added benefits of allowing to filter any other relevant > data like C-state residency values, and also to drop any irrelevant > data points directly in the kernel context, without passing all the > data to userspace for post-processing. > > Signed-off-by: Tero Kristo tero.kristo@linux.intel.com > --- > arch/x86/include/asm/msr.h | 1 + > arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ > 2 files changed, 24 insertions(+) > > diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h > index 65ec1965cd28..3dde673cb563 100644 > --- a/arch/x86/include/asm/msr.h > +++ b/arch/x86/include/asm/msr.h > @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); > void msrs_free(struct msr *msrs); > int msr_set_bit(u32 msr, u8 bit); > int msr_clear_bit(u32 msr, u8 bit); > +u64 bpf_rdtsc(void); > > #ifdef CONFIG_SMP > int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); > diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c > index 344698852146..ded857abef81 100644 > --- a/arch/x86/kernel/tsc.c > +++ b/arch/x86/kernel/tsc.c > @@ -15,6 +15,8 @@ > #include <linux/timex.h> > #include <linux/static_key.h> > #include <linux/static_call.h> > +#include <linux/btf.h> > +#include <linux/btf_ids.h> > > #include <asm/hpet.h> > #include <asm/timer.h> > @@ -29,6 +31,7 @@ > #include <asm/intel-family.h> > #include <asm/i8259.h> > #include <asm/uv/uv.h> > +#include <asm/tlbflush.h> > > unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ > EXPORT_SYMBOL(cpu_khz); > @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) > tsc_enable_sched_clock(); > } > > +u64 bpf_rdtsc(void) > +{ > + /* Check if Time Stamp is enabled only in ring 0 */ > + if (cr4_read_shadow() & X86_CR4_TSD) > + return 0; Why check this? It's always enabled in the kernel, no?
It is always enabled, but there are certain syscalls that can be used to disable the TSC access for oneself. prctl(PR_SET_TSC, ...) and seccomp(SET_MODE_STRICT,...). Not having the check in place would in theory allow a restricted BPF program to circumvent this (if there ever was such a thing.) But yes, I do agree this part is a bit debatable whether it should be there at all.
What do you mean 'circumvent' ? It's a tracing bpf prog running in the kernel loaded by root and reading tsc for the purpose of the kernel. There is no unprivileged access to tsc here.
This was based on some discussions with the security team at Intel, I don't pretend to know anything about security myself. But I can drop the check. It is probably not needed because of the fact that it is already possible to read the TSC counter with the approach I mention in the cover letter; via perf and bpf_core_read().
> + > + return rdtsc_ordered(); Why _ordered? Why not just rdtsc ? Especially since you want to trace latency. Extra lfence will ruin the measurements.
If we used it as a fast way to order events on multiple CPUs I guess we need the lfence? We use ktime_get_ns() now for things like this when we just need an order counter. We have also observed time going backwards with this and have heuristics to correct it but its rare.
Yeah, I think it is better to induce some extra latency instead of having some weird ordering issues with the timestamps.
lfence is not 'some extra latency'. I suspect rdtsc_ordered() will be slower than bpf_ktime_get_ns(). What's the point of using it then?
I would only use it if its faster then bpf_ktime_get_ns() and have already figured out how to handle rare unordered events so I think its OK to relax somewhat strict ordering.
I believe that on x86-arch using bpf_ktime_get_ns() also ends up calling rdtsc_odered() under the hood.
I just did some measurements on an Intel(R) Xeon(R) Platinum 8360Y CPU @ 2.40GHz, with a simple BPF code:
t1 = bpf_ktime_get_ns(); for (i = 0; i < NUM_CYC; i++) { bpf_rdtsc(); // or bpf_ktime_get_ns() here } t2 = bpf_ktime_get_ns();The results I got with the CPU locked at 2.4GHz (average execution times per a call within the loop, this with some 10M executions):
bpf_rdtsc() ordered : 45ns
bpf_rdtsc() un-ordered : 23ns
bpf_ktime_get_ns() : 49ns
Thanks for crunching the numbers. Based on them it's hard to justify adding the ordered variant. We already have ktime_get_ns, ktime_get_boot_ns, ktime_get_coarse_ns, ktime_get_tai_ns with pretty close performance and different time constraints. rdtsc_ordered doesn't bring anything new to the table. bpf_rdtsc() would be justified if it's significantly faster than traditional ktime*() helpers.
Locking the CPU at 800MHz the results are:
bpf_rdtsc() ordered : 55ns
bpf_rdtsc() un-ordered : 33ns
bpf_ktime_get_ns() : 71ns
The bpf_rdtsc() in these results contains some extra latency caused by conditional execution, I added a flag to the call to select whether it should use _ordered() or not, and it also still contains the CR4_TSD check in place.
-Tero
Also, things like the ftrace also use rdtsc_ordered() as its underlying clock, if you use x86-tsc as the trace clock (see arch/x86/kernel/trace_clock.c.)
-Tero
Hi,
Coming back to this bit late, I was on vacation for a few weeks.
On 07/07/2023 17:42, Alexei Starovoitov wrote:
On Fri, Jul 7, 2023 at 1:28 AM Tero Kristo tero.kristo@linux.intel.com wrote:
On 07/07/2023 08:41, John Fastabend wrote:
Alexei Starovoitov wrote:
On Thu, Jul 6, 2023 at 4:59 AM Tero Kristo tero.kristo@linux.intel.com wrote:
On 06/07/2023 08:16, John Fastabend wrote:
Alexei Starovoitov wrote: > On Mon, Jul 3, 2023 at 3:58 AM Tero Kristo tero.kristo@linux.intel.com wrote: >> Currently the raw TSC counter can be read within kernel via rdtsc_ordered() >> and friends, and additionally even userspace has access to it via the >> RDTSC assembly instruction. BPF programs on the other hand don't have >> direct access to the TSC counter, but alternatively must go through the >> performance subsystem (bpf_perf_event_read), which only provides relative >> value compared to the start point of the program, and is also much slower >> than the direct read. Add a new BPF helper definition for bpf_rdtsc() which >> can be used for any accurate profiling needs. >> >> A use-case for the new API is for example wakeup latency tracing via >> eBPF on Intel architecture, where it is extremely beneficial to be able >> to get raw TSC timestamps and compare these directly to the value >> programmed to the MSR_IA32_TSC_DEADLINE register. This way a direct >> latency value from the hardware interrupt to the execution of the >> interrupt handler can be calculated. Having the functionality within >> eBPF also has added benefits of allowing to filter any other relevant >> data like C-state residency values, and also to drop any irrelevant >> data points directly in the kernel context, without passing all the >> data to userspace for post-processing. >> >> Signed-off-by: Tero Kristo tero.kristo@linux.intel.com >> --- >> arch/x86/include/asm/msr.h | 1 + >> arch/x86/kernel/tsc.c | 23 +++++++++++++++++++++++ >> 2 files changed, 24 insertions(+) >> >> diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h >> index 65ec1965cd28..3dde673cb563 100644 >> --- a/arch/x86/include/asm/msr.h >> +++ b/arch/x86/include/asm/msr.h >> @@ -309,6 +309,7 @@ struct msr *msrs_alloc(void); >> void msrs_free(struct msr *msrs); >> int msr_set_bit(u32 msr, u8 bit); >> int msr_clear_bit(u32 msr, u8 bit); >> +u64 bpf_rdtsc(void); >> >> #ifdef CONFIG_SMP >> int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); >> diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c >> index 344698852146..ded857abef81 100644 >> --- a/arch/x86/kernel/tsc.c >> +++ b/arch/x86/kernel/tsc.c >> @@ -15,6 +15,8 @@ >> #include <linux/timex.h> >> #include <linux/static_key.h> >> #include <linux/static_call.h> >> +#include <linux/btf.h> >> +#include <linux/btf_ids.h> >> >> #include <asm/hpet.h> >> #include <asm/timer.h> >> @@ -29,6 +31,7 @@ >> #include <asm/intel-family.h> >> #include <asm/i8259.h> >> #include <asm/uv/uv.h> >> +#include <asm/tlbflush.h> >> >> unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ >> EXPORT_SYMBOL(cpu_khz); >> @@ -1551,6 +1554,24 @@ void __init tsc_early_init(void) >> tsc_enable_sched_clock(); >> } >> >> +u64 bpf_rdtsc(void) >> +{ >> + /* Check if Time Stamp is enabled only in ring 0 */ >> + if (cr4_read_shadow() & X86_CR4_TSD) >> + return 0; > Why check this? It's always enabled in the kernel, no?
It is always enabled, but there are certain syscalls that can be used to disable the TSC access for oneself. prctl(PR_SET_TSC, ...) and seccomp(SET_MODE_STRICT,...). Not having the check in place would in theory allow a restricted BPF program to circumvent this (if there ever was such a thing.) But yes, I do agree this part is a bit debatable whether it should be there at all.
What do you mean 'circumvent' ? It's a tracing bpf prog running in the kernel loaded by root and reading tsc for the purpose of the kernel. There is no unprivileged access to tsc here.
This was based on some discussions with the security team at Intel, I don't pretend to know anything about security myself. But I can drop the check. It is probably not needed because of the fact that it is already possible to read the TSC counter with the approach I mention in the cover letter; via perf and bpf_core_read().
>> + >> + return rdtsc_ordered(); > Why _ordered? Why not just rdtsc ? > Especially since you want to trace latency. Extra lfence will ruin > the measurements. > If we used it as a fast way to order events on multiple CPUs I guess we need the lfence? We use ktime_get_ns() now for things like this when we just need an order counter. We have also observed time going backwards with this and have heuristics to correct it but its rare.
Yeah, I think it is better to induce some extra latency instead of having some weird ordering issues with the timestamps.
lfence is not 'some extra latency'. I suspect rdtsc_ordered() will be slower than bpf_ktime_get_ns(). What's the point of using it then?
I would only use it if its faster then bpf_ktime_get_ns() and have already figured out how to handle rare unordered events so I think its OK to relax somewhat strict ordering.
I believe that on x86-arch using bpf_ktime_get_ns() also ends up calling rdtsc_odered() under the hood.
I just did some measurements on an Intel(R) Xeon(R) Platinum 8360Y CPU @ 2.40GHz, with a simple BPF code:
t1 = bpf_ktime_get_ns(); for (i = 0; i < NUM_CYC; i++) { bpf_rdtsc(); // or bpf_ktime_get_ns() here } t2 = bpf_ktime_get_ns();The results I got with the CPU locked at 2.4GHz (average execution times per a call within the loop, this with some 10M executions):
bpf_rdtsc() ordered : 45ns
bpf_rdtsc() un-ordered : 23ns
bpf_ktime_get_ns() : 49ns
Thanks for crunching the numbers. Based on them it's hard to justify adding the ordered variant. We already have ktime_get_ns, ktime_get_boot_ns, ktime_get_coarse_ns, ktime_get_tai_ns with pretty close performance and different time constraints. rdtsc_ordered doesn't bring anything new to the table. bpf_rdtsc() would be justified if it's significantly faster than traditional ktime*() helpers.
The only other justification I can use here is that the TSC counter is useful if you are dealing with any other counters that use TSC as a reference; mainly the Intel power management residency counters use same time base / resolution as TSC.
Converting between the TSC / ktime can get cumbersome, and you would need to get the magic conversion factors from somewhere.
-Tero
Locking the CPU at 800MHz the results are:
bpf_rdtsc() ordered : 55ns
bpf_rdtsc() un-ordered : 33ns
bpf_ktime_get_ns() : 71ns
The bpf_rdtsc() in these results contains some extra latency caused by conditional execution, I added a flag to the call to select whether it should use _ordered() or not, and it also still contains the CR4_TSD check in place.
-Tero
Also, things like the ftrace also use rdtsc_ordered() as its underlying clock, if you use x86-tsc as the trace clock (see arch/x86/kernel/trace_clock.c.)
-Tero
Add selftest for bpf_rdtsc() which reads the TSC (Time Stamp Counter) on x86_64 architectures. The test reads the TSC from both userspace and the BPF program, and verifies the TSC values are in incremental order as expected. The test is automatically skipped on architectures that do not support the feature.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com --- .../selftests/bpf/prog_tests/test_rdtsc.c | 67 +++++++++++++++++++ .../testing/selftests/bpf/progs/test_rdtsc.c | 21 ++++++ 2 files changed, 88 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/test_rdtsc.c create mode 100644 tools/testing/selftests/bpf/progs/test_rdtsc.c
diff --git a/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c b/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c new file mode 100644 index 000000000000..2b26deb5b35a --- /dev/null +++ b/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c @@ -0,0 +1,67 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2023 Intel Corporation */ + +#include "test_progs.h" +#include "test_rdtsc.skel.h" + +#ifdef __x86_64__ + +static inline u64 _rdtsc(void) +{ + u32 low, high; + + __asm__ __volatile__("rdtscp" : "=a" (low), "=d" (high)); + return ((u64)high << 32) | low; +} + +static int rdtsc(struct test_rdtsc *skel) +{ + int err, prog_fd; + u64 user_c1, user_c2; + + LIBBPF_OPTS(bpf_test_run_opts, topts); + + err = test_rdtsc__attach(skel); + if (!ASSERT_OK(err, "test_rdtsc_attach")) + return err; + + user_c1 = _rdtsc(); + + prog_fd = bpf_program__fd(skel->progs.test1); + err = bpf_prog_test_run_opts(prog_fd, &topts); + + user_c2 = _rdtsc(); + + ASSERT_OK(err, "test_run"); + ASSERT_EQ(topts.retval, 0, "test_run"); + + test_rdtsc__detach(skel); + + ASSERT_GE(skel->bss->c1, user_c1, "bpf c1 > user c1"); + ASSERT_GE(user_c2, skel->bss->c2, "user c2 > bpf c2"); + ASSERT_GE(skel->bss->c2, user_c1, "bpf c2 > bpf c1"); + ASSERT_GE(user_c2, user_c1, "user c2 > user c1"); + + return 0; +} +#endif + +void test_rdtsc(void) +{ +#ifdef __x86_64__ + struct test_rdtsc *skel; + int err; + + skel = test_rdtsc__open_and_load(); + if (!ASSERT_OK_PTR(skel, "test_rdtsc_skel_load")) + goto cleanup; + err = rdtsc(skel); + ASSERT_OK(err, "rdtsc"); + +cleanup: + test_rdtsc__destroy(skel); +#else + printf("%s:SKIP:bpf_rdtsc() kfunc not supported\n", __func__); + test__skip(); +#endif +} diff --git a/tools/testing/selftests/bpf/progs/test_rdtsc.c b/tools/testing/selftests/bpf/progs/test_rdtsc.c new file mode 100644 index 000000000000..14776b83bd3e --- /dev/null +++ b/tools/testing/selftests/bpf/progs/test_rdtsc.c @@ -0,0 +1,21 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2023 Intel Corporation */ +#include <linux/bpf.h> +#include <bpf/bpf_tracing.h> +#include <bpf/bpf_helpers.h> + +char _license[] SEC("license") = "GPL"; + +__u64 c1; +__u64 c2; + +extern __u64 bpf_rdtsc(void) __ksym; + +SEC("fentry/bpf_fentry_test1") +int BPF_PROG2(test1, int, a) +{ + c1 = bpf_rdtsc(); + c2 = bpf_rdtsc(); + + return 0; +}
Tero Kristo wrote:
Add selftest for bpf_rdtsc() which reads the TSC (Time Stamp Counter) on x86_64 architectures. The test reads the TSC from both userspace and the BPF program, and verifies the TSC values are in incremental order as expected. The test is automatically skipped on architectures that do not support the feature.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
.../selftests/bpf/prog_tests/test_rdtsc.c | 67 +++++++++++++++++++ .../testing/selftests/bpf/progs/test_rdtsc.c | 21 ++++++ 2 files changed, 88 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/test_rdtsc.c create mode 100644 tools/testing/selftests/bpf/progs/test_rdtsc.c
diff --git a/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c b/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c new file mode 100644 index 000000000000..2b26deb5b35a --- /dev/null +++ b/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c @@ -0,0 +1,67 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2023 Intel Corporation */
+#include "test_progs.h" +#include "test_rdtsc.skel.h"
+#ifdef __x86_64__
+static inline u64 _rdtsc(void) +{
- u32 low, high;
- __asm__ __volatile__("rdtscp" : "=a" (low), "=d" (high));
I think its ok but note this could fail if user doesn't have access to rdtscp and iirc that can be restricted?
- return ((u64)high << 32) | low;
+}
On 04/07/2023 01:00, John Fastabend wrote:
Tero Kristo wrote:
Add selftest for bpf_rdtsc() which reads the TSC (Time Stamp Counter) on x86_64 architectures. The test reads the TSC from both userspace and the BPF program, and verifies the TSC values are in incremental order as expected. The test is automatically skipped on architectures that do not support the feature.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
.../selftests/bpf/prog_tests/test_rdtsc.c | 67 +++++++++++++++++++ .../testing/selftests/bpf/progs/test_rdtsc.c | 21 ++++++ 2 files changed, 88 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/test_rdtsc.c create mode 100644 tools/testing/selftests/bpf/progs/test_rdtsc.c
diff --git a/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c b/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c new file mode 100644 index 000000000000..2b26deb5b35a --- /dev/null +++ b/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c @@ -0,0 +1,67 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2023 Intel Corporation */
+#include "test_progs.h" +#include "test_rdtsc.skel.h"
+#ifdef __x86_64__
+static inline u64 _rdtsc(void) +{
- u32 low, high;
- __asm__ __volatile__("rdtscp" : "=a" (low), "=d" (high));
I think its ok but note this could fail if user doesn't have access to rdtscp and iirc that can be restricted?
It is possible to restrict RDTSC access from userspace by enabling the TSD bit in CR4 register, and it will cause the userspace process to trap with general protection fault.
However, the usage of RDTSC appears to be built-in to C standard libraries (probably some timer routines) and enabling the CR4 TSD makes the system near unusable. Things like sshd + systemd also start generating the same general protection faults if RDTSC is blocked. Also, attempting to run anything at all with the BPF selftest suite causes the same general protection fault; not only the rdtsc test.
I tried this with couple of setups, one system running a minimalistic buildroot and another one running a fedora37 installation and the results were similar.
-Tero
- return ((u64)high << 32) | low;
+}
Tero Kristo wrote:
On 04/07/2023 01:00, John Fastabend wrote:
Tero Kristo wrote:
Add selftest for bpf_rdtsc() which reads the TSC (Time Stamp Counter) on x86_64 architectures. The test reads the TSC from both userspace and the BPF program, and verifies the TSC values are in incremental order as expected. The test is automatically skipped on architectures that do not support the feature.
Signed-off-by: Tero Kristo tero.kristo@linux.intel.com
.../selftests/bpf/prog_tests/test_rdtsc.c | 67 +++++++++++++++++++ .../testing/selftests/bpf/progs/test_rdtsc.c | 21 ++++++ 2 files changed, 88 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/test_rdtsc.c create mode 100644 tools/testing/selftests/bpf/progs/test_rdtsc.c
diff --git a/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c b/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c new file mode 100644 index 000000000000..2b26deb5b35a --- /dev/null +++ b/tools/testing/selftests/bpf/prog_tests/test_rdtsc.c @@ -0,0 +1,67 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2023 Intel Corporation */
+#include "test_progs.h" +#include "test_rdtsc.skel.h"
+#ifdef __x86_64__
+static inline u64 _rdtsc(void) +{
- u32 low, high;
- __asm__ __volatile__("rdtscp" : "=a" (low), "=d" (high));
I think its ok but note this could fail if user doesn't have access to rdtscp and iirc that can be restricted?
It is possible to restrict RDTSC access from userspace by enabling the TSD bit in CR4 register, and it will cause the userspace process to trap with general protection fault.
However, the usage of RDTSC appears to be built-in to C standard libraries (probably some timer routines) and enabling the CR4 TSD makes the system near unusable. Things like sshd + systemd also start generating the same general protection faults if RDTSC is blocked. Also, attempting to run anything at all with the BPF selftest suite causes the same general protection fault; not only the rdtsc test.
I tried this with couple of setups, one system running a minimalistic buildroot and another one running a fedora37 installation and the results were similar.
Thanks. Good enough for me.
-Tero
Tero Kristo wrote:
Hello,
This patch series adds a new x86 arch specific BPF helper, bpf_rdtsc() which can be used for reading the hardware time stamp counter (TSC.) Currently the same counter is directly accessible from userspace (using RDTSC instruction), and kernel space using various rdtsc_*() APIs, however eBPF lacks the support.
The main usage for the TSC counter is for various profiling and timing purposes, getting accurate cycle counter values. The counter can be currently read from BPF programs by using the existing perf subsystem services (bpf_perf_event_read()), however its usage is cumbersome at best. Additionally, the perf subsystem provides relative value only for the counter, but absolute values are desired by some use cases like Wult [1]. The absolute value of TSC can be read with BPF programs currently via some kprobe / bpf_core_read() magic (see [2], [3], [4] for example), but this relies on accessing kernel internals and is not stable API, and is pretty cumbersome. Thus, this patch proposes a new arch x86 specific BPF helper to avoid the above issues.
-Tero
[1] https://github.com/intel/wult [2] https://github.com/intel/wult/blob/c92237c95b898498faf41e6644983102d1fe5156/... [3] https://github.com/intel/wult/blob/c92237c95b898498faf41e6644983102d1fe5156/... [4] https://github.com/intel/wult/blob/c92237c95b898498faf41e6644983102d1fe5156/...
Makes a lot of sense to me.
Acked-by: John Fastabend john.fastabend@gmail.com
linux-kselftest-mirror@lists.linaro.org
-
Alexei Starovoitov -
John Fastabend -
Tero Kristo -
Yonghong Song