All of lore.kernel.org
 help / color / mirror / Atom feed
From: Peter Zijlstra <peterz@infradead.org>
To: Thomas Gleixner <tglx@linutronix.de>
Cc: Steven Rostedt <rostedt@goodmis.org>,
	LKML <linux-kernel@vger.kernel.org>,
	Ingo Molnar <mingo@kernel.org>,
	kan.liang@linux.intel.com, like.xu@linux.intel.com,
	Andy Lutomirski <luto@amacapital.net>
Subject: [PATCH] perf/arch: Remove perf_sample_data::regs_user_copy
Date: Fri, 30 Oct 2020 11:59:17 +0100	[thread overview]
Message-ID: <20201030105917.GI2611@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <877dr8nh6u.fsf@nanos.tec.linutronix.de>

On Fri, Oct 30, 2020 at 11:26:01AM +0100, Thomas Gleixner wrote:
> What's worse is perf_sample_data which is 384 bytes and is 64 bytes aligned.

Here; this shrinks it to 192 bytes. Combined with the static dummy this
should reduce stack usage by 360 bytes.

---
Subject: perf/arch: Remove perf_sample_data::regs_user_copy

struct perf_sample_data lives on-stack, we should be careful about it's
size. Furthermore, the pt_regs copy in there is only because x86_64 is a
trainwreck, solve it differently.

Halves sizeof(perf_sample_data).

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 arch/arm/kernel/perf_regs.c   |  3 +--
 arch/arm64/kernel/perf_regs.c |  3 +--
 arch/csky/kernel/perf_regs.c  |  3 +--
 arch/powerpc/perf/perf_regs.c |  3 +--
 arch/riscv/kernel/perf_regs.c |  3 +--
 arch/s390/kernel/perf_regs.c  |  3 +--
 arch/x86/kernel/perf_regs.c   | 15 +++++++++++----
 include/linux/perf_event.h    |  6 ------
 include/linux/perf_regs.h     |  6 ++----
 kernel/events/core.c          |  8 +++-----
 10 files changed, 22 insertions(+), 31 deletions(-)

diff --git a/arch/arm/kernel/perf_regs.c b/arch/arm/kernel/perf_regs.c
index 05fe92aa7d98..0529f90395c9 100644
--- a/arch/arm/kernel/perf_regs.c
+++ b/arch/arm/kernel/perf_regs.c
@@ -32,8 +32,7 @@ u64 perf_reg_abi(struct task_struct *task)
 }
 
 void perf_get_regs_user(struct perf_regs *regs_user,
-			struct pt_regs *regs,
-			struct pt_regs *regs_user_copy)
+			struct pt_regs *regs)
 {
 	regs_user->regs = task_pt_regs(current);
 	regs_user->abi = perf_reg_abi(current);
diff --git a/arch/arm64/kernel/perf_regs.c b/arch/arm64/kernel/perf_regs.c
index 94e8718e7229..f6f58e6265df 100644
--- a/arch/arm64/kernel/perf_regs.c
+++ b/arch/arm64/kernel/perf_regs.c
@@ -73,8 +73,7 @@ u64 perf_reg_abi(struct task_struct *task)
 }
 
 void perf_get_regs_user(struct perf_regs *regs_user,
-			struct pt_regs *regs,
-			struct pt_regs *regs_user_copy)
+			struct pt_regs *regs)
 {
 	regs_user->regs = task_pt_regs(current);
 	regs_user->abi = perf_reg_abi(current);
diff --git a/arch/csky/kernel/perf_regs.c b/arch/csky/kernel/perf_regs.c
index eb32838b8210..09b7f88a2d6a 100644
--- a/arch/csky/kernel/perf_regs.c
+++ b/arch/csky/kernel/perf_regs.c
@@ -32,8 +32,7 @@ u64 perf_reg_abi(struct task_struct *task)
 }
 
 void perf_get_regs_user(struct perf_regs *regs_user,
-			struct pt_regs *regs,
-			struct pt_regs *regs_user_copy)
+			struct pt_regs *regs)
 {
 	regs_user->regs = task_pt_regs(current);
 	regs_user->abi = perf_reg_abi(current);
diff --git a/arch/powerpc/perf/perf_regs.c b/arch/powerpc/perf/perf_regs.c
index 8e53f2fc3fe0..6f681b105eec 100644
--- a/arch/powerpc/perf/perf_regs.c
+++ b/arch/powerpc/perf/perf_regs.c
@@ -144,8 +144,7 @@ u64 perf_reg_abi(struct task_struct *task)
 }
 
 void perf_get_regs_user(struct perf_regs *regs_user,
-			struct pt_regs *regs,
-			struct pt_regs *regs_user_copy)
+			struct pt_regs *regs)
 {
 	regs_user->regs = task_pt_regs(current);
 	regs_user->abi = (regs_user->regs) ? perf_reg_abi(current) :
diff --git a/arch/riscv/kernel/perf_regs.c b/arch/riscv/kernel/perf_regs.c
index 04a38fbeb9c7..fd304a248de6 100644
--- a/arch/riscv/kernel/perf_regs.c
+++ b/arch/riscv/kernel/perf_regs.c
@@ -36,8 +36,7 @@ u64 perf_reg_abi(struct task_struct *task)
 }
 
 void perf_get_regs_user(struct perf_regs *regs_user,
-			struct pt_regs *regs,
-			struct pt_regs *regs_user_copy)
+			struct pt_regs *regs)
 {
 	regs_user->regs = task_pt_regs(current);
 	regs_user->abi = perf_reg_abi(current);
diff --git a/arch/s390/kernel/perf_regs.c b/arch/s390/kernel/perf_regs.c
index 4352a504f235..6e9e5d5e927e 100644
--- a/arch/s390/kernel/perf_regs.c
+++ b/arch/s390/kernel/perf_regs.c
@@ -53,8 +53,7 @@ u64 perf_reg_abi(struct task_struct *task)
 }
 
 void perf_get_regs_user(struct perf_regs *regs_user,
-			struct pt_regs *regs,
-			struct pt_regs *regs_user_copy)
+			struct pt_regs *regs)
 {
 	/*
 	 * Use the regs from the first interruption and let
diff --git a/arch/x86/kernel/perf_regs.c b/arch/x86/kernel/perf_regs.c
index bb7e1132290b..750129bf539c 100644
--- a/arch/x86/kernel/perf_regs.c
+++ b/arch/x86/kernel/perf_regs.c
@@ -101,8 +101,7 @@ u64 perf_reg_abi(struct task_struct *task)
 }
 
 void perf_get_regs_user(struct perf_regs *regs_user,
-			struct pt_regs *regs,
-			struct pt_regs *regs_user_copy)
+			struct pt_regs *regs)
 {
 	regs_user->regs = task_pt_regs(current);
 	regs_user->abi = perf_reg_abi(current);
@@ -129,12 +128,20 @@ u64 perf_reg_abi(struct task_struct *task)
 		return PERF_SAMPLE_REGS_ABI_64;
 }
 
+static DEFINE_PER_CPU(struct pt_regs, nmi_user_regs);
+
 void perf_get_regs_user(struct perf_regs *regs_user,
-			struct pt_regs *regs,
-			struct pt_regs *regs_user_copy)
+			struct pt_regs *regs)
 {
+	struct pt_regs *regs_user_copy = this_cpu_ptr(&nmi_user_regs);
 	struct pt_regs *user_regs = task_pt_regs(current);
 
+	if (!in_nmi()) {
+		regs_user->regs = task_pt_regs(current);
+		regs_user->abi = perf_reg_abi(current);
+		return;
+	}
+
 	/*
 	 * If we're in an NMI that interrupted task_pt_regs setup, then
 	 * we can't sample user regs at all.  This check isn't really
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 0c19d279b97f..e58f4e6b6e1c 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -1022,13 +1022,7 @@ struct perf_sample_data {
 	struct perf_callchain_entry	*callchain;
 	u64				aux_size;
 
-	/*
-	 * regs_user may point to task_pt_regs or to regs_user_copy, depending
-	 * on arch details.
-	 */
 	struct perf_regs		regs_user;
-	struct pt_regs			regs_user_copy;
-
 	struct perf_regs		regs_intr;
 	u64				stack_user_size;
 
diff --git a/include/linux/perf_regs.h b/include/linux/perf_regs.h
index 2d12e97d5e7b..5efcdb768c75 100644
--- a/include/linux/perf_regs.h
+++ b/include/linux/perf_regs.h
@@ -20,8 +20,7 @@ u64 perf_reg_value(struct pt_regs *regs, int idx);
 int perf_reg_validate(u64 mask);
 u64 perf_reg_abi(struct task_struct *task);
 void perf_get_regs_user(struct perf_regs *regs_user,
-			struct pt_regs *regs,
-			struct pt_regs *regs_user_copy);
+			struct pt_regs *regs);
 #else
 
 #define PERF_REG_EXTENDED_MASK	0
@@ -42,8 +41,7 @@ static inline u64 perf_reg_abi(struct task_struct *task)
 }
 
 static inline void perf_get_regs_user(struct perf_regs *regs_user,
-				      struct pt_regs *regs,
-				      struct pt_regs *regs_user_copy)
+				      struct pt_regs *regs);
 {
 	regs_user->regs = task_pt_regs(current);
 	regs_user->abi = perf_reg_abi(current);
diff --git a/kernel/events/core.c b/kernel/events/core.c
index da467e1dd49a..14709182d94e 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -6374,14 +6374,13 @@ perf_output_sample_regs(struct perf_output_handle *handle,
 }
 
 static void perf_sample_regs_user(struct perf_regs *regs_user,
-				  struct pt_regs *regs,
-				  struct pt_regs *regs_user_copy)
+				  struct pt_regs *regs)
 {
 	if (user_mode(regs)) {
 		regs_user->abi = perf_reg_abi(current);
 		regs_user->regs = regs;
 	} else if (!(current->flags & PF_KTHREAD)) {
-		perf_get_regs_user(regs_user, regs, regs_user_copy);
+		perf_get_regs_user(regs_user, regs);
 	} else {
 		regs_user->abi = PERF_SAMPLE_REGS_ABI_NONE;
 		regs_user->regs = NULL;
@@ -7083,8 +7082,7 @@ void perf_prepare_sample(struct perf_event_header *header,
 	}
 
 	if (sample_type & (PERF_SAMPLE_REGS_USER | PERF_SAMPLE_STACK_USER))
-		perf_sample_regs_user(&data->regs_user, regs,
-				      &data->regs_user_copy);
+		perf_sample_regs_user(&data->regs_user, regs);
 
 	if (sample_type & PERF_SAMPLE_REGS_USER) {
 		/* regs dump ABI info */

  parent reply	other threads:[~2020-10-30 10:59 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-10-30  4:27 [BUG] Stack overflow when running perf and function tracer Steven Rostedt
2020-10-30  9:00 ` Peter Zijlstra
2020-10-30 10:26   ` Thomas Gleixner
2020-10-30 10:32     ` Peter Zijlstra
2020-10-30 11:36       ` Thomas Gleixner
2020-10-30 13:42         ` Thomas Gleixner
2020-10-30 10:59     ` Peter Zijlstra [this message]
2020-10-30 12:33       ` [PATCH] perf/arch: Remove perf_sample_data::regs_user_copy Steven Rostedt
2020-10-30 11:19   ` [PATCH] perf/x86: Make dummy_iregs static Peter Zijlstra

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20201030105917.GI2611@hirez.programming.kicks-ass.net \
    --to=peterz@infradead.org \
    --cc=kan.liang@linux.intel.com \
    --cc=like.xu@linux.intel.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=luto@amacapital.net \
    --cc=mingo@kernel.org \
    --cc=rostedt@goodmis.org \
    --cc=tglx@linutronix.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body. 
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.