Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   kan.liang@linux.intel.com
To:     peterz@infradead.org, acme@kernel.org, tglx@linutronix.de,
        mingo@redhat.com, linux-kernel@vger.kernel.org
Cc:     eranian@google.com, jolsa@redhat.com, namhyung@kernel.org,
        ak@linux.intel.com, Kan Liang <kan.liang@linux.intel.com>
Subject: [PATCH RFC 1/7] perf/core, x86: Add PERF_SAMPLE_PAGE_SIZE
Date:   Fri, 10 Aug 2018 06:36:21 -0700
Message-Id: <1533908187-25204-1-git-send-email-kan.liang@linux.intel.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

From: Kan Liang <kan.liang@linux.intel.com>

Current perf can report both virtual address and physical address, but
it doesn't report page size. Users have no idea how large the utilized
page is. They cannot promote/demote large pages to optimize memory use.

Add a new sample type for page size.

Current perf already has a facility to collect data virtual address.
A function, to retrieve page size by full page-table walk of a given
virtual address, is introduced for x86. Other architectures can
implement their own functions later separately.
The function must be IRQ-safe. For x86, disabling IRQs over the walk is
sufficient to prevent any tear down of the page tables.

The new sample type requires collecting the virtual address. The
virtual address will not be output unless SAMPLE_ADDR is applied.

Although only a few bits are needed to indicate the page size, a u64
type is still claimed for page_size. Because struct perf_sample_data
requires cacheline_aligned.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
---
 arch/x86/events/core.c          | 25 +++++++++++++++++++++++++
 arch/x86/events/intel/ds.c      |  2 +-
 arch/x86/events/perf_event.h    |  2 +-
 include/linux/perf_event.h      |  1 +
 include/uapi/linux/perf_event.h | 13 ++++++++++++-
 kernel/events/core.c            | 15 +++++++++++++++
 6 files changed, 55 insertions(+), 3 deletions(-)

diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
index 5f4829f..719e527 100644
--- a/arch/x86/events/core.c
+++ b/arch/x86/events/core.c
@@ -2573,3 +2573,28 @@ void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap)
 	cap->events_mask_len	= x86_pmu.events_mask_len;
 }
 EXPORT_SYMBOL_GPL(perf_get_x86_pmu_capability);
+
+u64 perf_get_page_size(u64 virt)
+{
+	unsigned long flags;
+	unsigned int level;
+	pte_t *pte;
+
+	if (!virt)
+		return 0;
+
+	/*
+	 * Interrupts are disabled, so it prevents any tear down
+	 * of the page tables.
+	 * See the comment near struct mmu_table_batch.
+	 */
+	local_irq_save(flags);
+	if (virt >= TASK_SIZE)
+		pte = lookup_address(virt, &level);
+	else
+		pte = lookup_address_in_pgd(pgd_offset(current->mm, virt),
+					    virt, &level);
+	local_irq_restore(flags);
+
+	return (u64)level;
+}
diff --git a/arch/x86/events/intel/ds.c b/arch/x86/events/intel/ds.c
index b7b01d7..a3e56c7 100644
--- a/arch/x86/events/intel/ds.c
+++ b/arch/x86/events/intel/ds.c
@@ -1274,7 +1274,7 @@ static void setup_pebs_sample_data(struct perf_event *event,
 	}
 
 
-	if ((sample_type & (PERF_SAMPLE_ADDR | PERF_SAMPLE_PHYS_ADDR)) &&
+	if ((sample_type & (PERF_SAMPLE_ADDR | PERF_SAMPLE_PHYS_ADDR | PERF_SAMPLE_PAGE_SIZE)) &&
 	    x86_pmu.intel_cap.pebs_format >= 1)
 		data->addr = pebs->dla;
 
diff --git a/arch/x86/events/perf_event.h b/arch/x86/events/perf_event.h
index 1562863..affcd26 100644
--- a/arch/x86/events/perf_event.h
+++ b/arch/x86/events/perf_event.h
@@ -94,7 +94,7 @@ struct amd_nb {
 	PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \
 	PERF_SAMPLE_TRANSACTION | PERF_SAMPLE_PHYS_ADDR | \
 	PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER | \
-	PERF_SAMPLE_PERIOD)
+	PERF_SAMPLE_PERIOD | PERF_SAMPLE_PAGE_SIZE)
 
 #define PEBS_REGS \
 	(PERF_REG_X86_AX | \
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 53c500f..9d13745 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -937,6 +937,7 @@ struct perf_sample_data {
 	u64				stack_user_size;
 
 	u64				phys_addr;
+	u64				page_size;
 } ____cacheline_aligned;
 
 /* default value for data source */
diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index eeb787b..5473443 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -141,8 +141,9 @@ enum perf_event_sample_format {
 	PERF_SAMPLE_TRANSACTION			= 1U << 17,
 	PERF_SAMPLE_REGS_INTR			= 1U << 18,
 	PERF_SAMPLE_PHYS_ADDR			= 1U << 19,
+	PERF_SAMPLE_PAGE_SIZE			= 1U << 20,
 
-	PERF_SAMPLE_MAX = 1U << 20,		/* non-ABI */
+	PERF_SAMPLE_MAX = 1U << 21,		/* non-ABI */
 
 	__PERF_SAMPLE_CALLCHAIN_EARLY		= 1ULL << 63,
 };
@@ -861,6 +862,7 @@ enum perf_event_type {
 	 *	{ u64			abi; # enum perf_sample_regs_abi
 	 *	  u64			regs[weight(mask)]; } && PERF_SAMPLE_REGS_INTR
 	 *	{ u64			phys_addr;} && PERF_SAMPLE_PHYS_ADDR
+	 *	{ u64			page_size;} && PERF_SAMPLE_PAGE_SIZE
 	 * };
 	 */
 	PERF_RECORD_SAMPLE			= 9,
@@ -1099,6 +1101,15 @@ union perf_mem_data_src {
 #define PERF_MEM_S(a, s) \
 	(((__u64)PERF_MEM_##a##_##s) << PERF_MEM_##a##_SHIFT)
 
+
+enum perf_mem_page_size {
+	PERF_MEM_PAGE_SIZE_NONE,
+	PERF_MEM_PAGE_SIZE_4K,
+	PERF_MEM_PAGE_SIZE_2M,
+	PERF_MEM_PAGE_SIZE_1G,
+	PERF_MEM_PAGE_SIZE_512G,
+};
+
 /*
  * single taken branch record layout:
  *
diff --git a/kernel/events/core.c b/kernel/events/core.c
index f6ea33a..e848e9b 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -1751,6 +1751,9 @@ static void __perf_event_header_size(struct perf_event *event, u64 sample_type)
 	if (sample_type & PERF_SAMPLE_PHYS_ADDR)
 		size += sizeof(data->phys_addr);
 
+	if (sample_type & PERF_SAMPLE_PAGE_SIZE)
+		size += sizeof(data->page_size);
+
 	event->header_size = size;
 }
 
@@ -6294,6 +6297,9 @@ void perf_output_sample(struct perf_output_handle *handle,
 	if (sample_type & PERF_SAMPLE_PHYS_ADDR)
 		perf_output_put(handle, data->phys_addr);
 
+	if (sample_type & PERF_SAMPLE_PAGE_SIZE)
+		perf_output_put(handle, data->page_size);
+
 	if (!event->attr.watermark) {
 		int wakeup_events = event->attr.wakeup_events;
 
@@ -6341,6 +6347,12 @@ static u64 perf_virt_to_phys(u64 virt)
 	return phys_addr;
 }
 
+/* Return page size of given virtual address. IRQ-safe required. */
+u64 __weak perf_get_page_size(u64 virt)
+{
+	return PERF_MEM_PAGE_SIZE_NONE;
+}
+
 static struct perf_callchain_entry __empty_callchain = { .nr = 0, };
 
 struct perf_callchain_entry *
@@ -6482,6 +6494,9 @@ void perf_prepare_sample(struct perf_event_header *header,
 
 	if (sample_type & PERF_SAMPLE_PHYS_ADDR)
 		data->phys_addr = perf_virt_to_phys(data->addr);
+
+	if (sample_type & PERF_SAMPLE_PAGE_SIZE)
+		data->page_size = perf_get_page_size(data->addr);
 }
 
 static __always_inline void
-- 
2.7.4