Messages in this thread | | | From | "Liang, Kan" <> | Subject | RE: [PATCH 1/8] perf: Add PERF_SAMPLE_PHYS_ADDR | Date | Fri, 8 Jan 2016 21:12:36 +0000 |
| |
> > On Wed, Jan 6, 2016 at 3:04 AM, <kan.liang@intel.com> wrote: > > From: Kan Liang <kan.liang@intel.com> > > > > For understanding how the workload maps to memory channels and > > hardware behavior, it's useful to collect address maps with physical > addresses. > > This is not intended for detecting page sharing (which can be already > > done using the mmap inode), but for lower level hardware behavior > > studies. > > Perf supports load latency/DLA which can only collect virtual addresses. > > This patch add a new sample type PERF_SAMPLE_PHYS_ADDR to expose > the > > physical addresses. > > For kernel direct mapping addresses, the patch uses virt_to_phys to > > convert the virtual addresses from DLA to physical address. > > For user virtual addresses, __get_user_pages_fast is used to walk the > > pages tables for user physical address. > > This does not work for vmalloc addresses. Right now these are not > > resolved, but code to do that could be added. > > For security, the physical address can only be exposed to root or > > privileged user. > > > The other limitation of your patch series is that it is providing physical > address for data only. It does not cover code, i.e., it is not possible to > obtain the physical address of a sampled IP. So maybe the new > sample_type should be renamed to PERF_SAMPLE_DATA_PHYS_ADDR > instead.
We use sample_type PERF_SAMPLE_ADDR for the data virtual address. So in this patch I choose a similar name PERF_SAMPLE_PHYS_ADDR for data physical address. I think it should not lead to ambiguity.
> Should you add physical addresses for IP (no need for PEBS for this), then > you'd have to have a new name, anyway.
We use another sample_type PERF_SAMPLE_IP to get data->ip. So I think we may name PERF_SAMPLE_PHYS_IP as the new sample_type for the physical addresses of IP Also, since code and data are from different sample_type, I think it may be better to implement the PERF_SAMPLE_PHYS_IP in a separate patch later.
Thanks, Kan
> > Getting code and data physical addresses would provide better coverage > of the memory subsystem usage. > > > Signed-off-by: Kan Liang <kan.liang@intel.com> > > --- > > arch/x86/kernel/cpu/perf_event.h | 2 +- > > arch/x86/kernel/cpu/perf_event_intel_ds.c | 23 > +++++++++++++++++++++++ > > include/linux/perf_event.h | 3 +++ > > include/uapi/linux/perf_event.h | 4 +++- > > kernel/events/core.c | 11 +++++++++++ > > 5 files changed, 41 insertions(+), 2 deletions(-) > > > > diff --git a/arch/x86/kernel/cpu/perf_event.h > > b/arch/x86/kernel/cpu/perf_event.h > > index 799e6bd..164de68 100644 > > --- a/arch/x86/kernel/cpu/perf_event.h > > +++ b/arch/x86/kernel/cpu/perf_event.h > > @@ -90,7 +90,7 @@ struct amd_nb { > > (PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_ADDR | \ > > PERF_SAMPLE_ID | PERF_SAMPLE_CPU | > PERF_SAMPLE_STREAM_ID | \ > > PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \ > > - PERF_SAMPLE_TRANSACTION) > > + PERF_SAMPLE_TRANSACTION | PERF_SAMPLE_PHYS_ADDR) > > > > /* > > * A debug store configuration. > > diff --git a/arch/x86/kernel/cpu/perf_event_intel_ds.c > > b/arch/x86/kernel/cpu/perf_event_intel_ds.c > > index 5db1c77..2e333dc 100644 > > --- a/arch/x86/kernel/cpu/perf_event_intel_ds.c > > +++ b/arch/x86/kernel/cpu/perf_event_intel_ds.c > > @@ -986,6 +986,7 @@ static void setup_pebs_sample_data(struct > perf_event *event, > > u64 sample_type; > > int fll, fst, dsrc; > > int fl = event->hw.flags; > > + struct page *p = NULL; > > > > if (pebs == NULL) > > return; > > @@ -1071,6 +1072,28 @@ static void setup_pebs_sample_data(struct > perf_event *event, > > x86_pmu.intel_cap.pebs_format >= 1) > > data->addr = pebs->dla; > > > > + if ((sample_type & PERF_SAMPLE_PHYS_ADDR) && (data->addr != > 0)) { > > + if (data->addr >= TASK_SIZE) { > > + /* If it's vmalloc()d memory, leave phys_addr as 0 */ > > + if (virt_addr_valid(data->addr) && > > + !(data->addr >= VMALLOC_START && data->addr < > VMALLOC_END)) > > + data->phys_addr = (u64)virt_to_phys((void *)data- > >addr); > > + } else { > > + /* > > + * Walking the pages tables for user address. > > + * Interrupts are disabled, so it prevents any tear down > > + * of the page tables. > > + * Try IRQ-safe __get_user_pages_fast first. > > + * If failed, leave phys_addr as 0. > > + */ > > + if ((current->mm != NULL) && > > + (__get_user_pages_fast(data->addr, 1, 0, &p) == 1)) > > + data->phys_addr = page_to_phys(p) + > > + data->addr % PAGE_SIZE; > > + > > + if (p) > > + put_page(p); > > + } > > + } > > if (x86_pmu.intel_cap.pebs_format >= 2) { > > /* Only set the TSX weight when no memory weight. */ > > if ((sample_type & PERF_SAMPLE_WEIGHT) && !fll) diff > > --git a/include/linux/perf_event.h b/include/linux/perf_event.h index > > f9828a4..d9c0527 100644 > > --- a/include/linux/perf_event.h > > +++ b/include/linux/perf_event.h > > @@ -795,6 +795,8 @@ struct perf_sample_data { > > > > struct perf_regs regs_intr; > > u64 stack_user_size; > > + > > + u64 phys_addr; > > } ____cacheline_aligned; > > > > /* default value for data source */ > > @@ -815,6 +817,7 @@ static inline void perf_sample_data_init(struct > perf_sample_data *data, > > data->weight = 0; > > data->data_src.val = PERF_MEM_NA; > > data->txn = 0; > > + data->phys_addr = 0; > > } > > > > extern void perf_output_sample(struct perf_output_handle *handle, > > diff --git a/include/uapi/linux/perf_event.h > > b/include/uapi/linux/perf_event.h index 1afe962..5afc572 100644 > > --- a/include/uapi/linux/perf_event.h > > +++ b/include/uapi/linux/perf_event.h > > @@ -139,8 +139,9 @@ enum perf_event_sample_format { > > PERF_SAMPLE_IDENTIFIER = 1U << 16, > > PERF_SAMPLE_TRANSACTION = 1U << 17, > > PERF_SAMPLE_REGS_INTR = 1U << 18, > > + PERF_SAMPLE_PHYS_ADDR = 1U << 19, > > > > - PERF_SAMPLE_MAX = 1U << 19, /* non-ABI */ > > + PERF_SAMPLE_MAX = 1U << 20, /* non-ABI */ > > }; > > > > /* > > @@ -767,6 +768,7 @@ enum perf_event_type { > > * { u64 transaction; } && PERF_SAMPLE_TRANSACTION > > * { u64 abi; # enum perf_sample_regs_abi > > * u64 regs[weight(mask)]; } && > PERF_SAMPLE_REGS_INTR > > + * { u64 phys_addr;} && PERF_SAMPLE_PHYS_ADDR > > * }; > > */ > > PERF_RECORD_SAMPLE = 9, > > diff --git a/kernel/events/core.c b/kernel/events/core.c index > > a627f36..9a922a2 100644 > > --- a/kernel/events/core.c > > +++ b/kernel/events/core.c > > @@ -1334,6 +1334,9 @@ static void __perf_event_header_size(struct > perf_event *event, u64 sample_type) > > if (sample_type & PERF_SAMPLE_TRANSACTION) > > size += sizeof(data->txn); > > > > + if (sample_type & PERF_SAMPLE_PHYS_ADDR) > > + size += sizeof(data->phys_addr); > > + > > event->header_size = size; > > } > > > > @@ -5432,6 +5435,9 @@ void perf_output_sample(struct > perf_output_handle *handle, > > if (sample_type & PERF_SAMPLE_DATA_SRC) > > perf_output_put(handle, data->data_src.val); > > > > + if (sample_type & PERF_SAMPLE_PHYS_ADDR) > > + perf_output_put(handle, data->phys_addr); > > + > > if (sample_type & PERF_SAMPLE_TRANSACTION) > > perf_output_put(handle, data->txn); > > > > @@ -8269,6 +8275,11 @@ SYSCALL_DEFINE5(perf_event_open, > > return -EINVAL; > > } > > > > + /* Only privileged users can get kernel addresses */ > > + if ((attr.sample_type & PERF_SAMPLE_PHYS_ADDR) && > > + !capable(CAP_SYS_ADMIN)) > > + return -EACCES; > > + > > /* > > * In cgroup mode, the pid argument is used to pass the fd > > * opened to the cgroup directory in cgroupfs. The cpu > > argument > > -- > > 1.8.3.1 > >
| |