Skip to contentSkip to conversation

sh/irq: use seq_put_decimal_ull_width() for decimal values

PATCH Applied David Wang 2 participants 6 messages Nov 8, 2024 – Jan 30, 2025 SuperH

lore
Share & tools
lore

David Wang

author
#0 lore raw

PATCH 13/13 sh/irq: use seq_put_decimal_ull_width() for decimal values

Performance improvement for reading /proc/interrupts on arch sh

---
 arch/sh/kernel/irq.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/sh/kernel/irq.c +2 -2
index 4e6835de54cf..9022d8af9d68 100644
@@ -43,9 +43,9 @@ int arch_show_interrupts(struct seq_file *p, int prec)
 {
 	int j;
 
-	seq_printf(p, "%*s: ", prec, "NMI");
+	seq_printf(p, "%*s:", prec, "NMI");
 	for_each_online_cpu(j)
-		seq_printf(p, "%10u ", per_cpu(irq_stat.__nmi_count, j));
+		seq_put_decimal_ull_width(p, " ", per_cpu(irq_stat.__nmi_count, j), 10);
 	seq_printf(p, "  Non-maskable interrupts\n");
 
 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
Signature · 1 line
2.39.2
#1 lore raw

Hi David,

Quoted 13 lines · David Wang

Performance improvement for reading /proc/interrupts on arch sh

---
 arch/sh/kernel/irq.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/sh/kernel/irq.c +0 -0
index 4e6835de54cf..9022d8af9d68 100644
@@ -43,9 +43,9 @@ int arch_show_interrupts(struct seq_file *p, int prec)
 {

arch/sh/kernel/irq.c int arch_show_interrupts(struct seq_file *p, int prec) · line 51

Quoted 10 lines
arch/sh/kernel/irq.c +2 -2
 	int j;
 
-	seq_printf(p, "%*s: ", prec, "NMI");
+	seq_printf(p, "%*s:", prec, "NMI");
 	for_each_online_cpu(j)
-		seq_printf(p, "%10u ", per_cpu(irq_stat.__nmi_count, j));
+		seq_put_decimal_ull_width(p, " ", per_cpu(irq_stat.__nmi_count, j), 10);
 	seq_printf(p, "  Non-maskable interrupts\n");
 
 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));

Sorry for the very late reply!

I don't quite understand why seq_put_decimal_ull_width() should be faster than seq_printf().

Can you elaborate on this a bit more?

Adrian

Signature · 4 lines
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer
`. `'   Physicist
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913
#2 lore raw

Hi David,

Quoted 23 lines · John Paul Adrian Glaubitz
Quoted 22 lines · David Wang

Performance improvement for reading /proc/interrupts on arch sh

---
 arch/sh/kernel/irq.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/sh/kernel/irq.c +2 -2
index 4e6835de54cf..9022d8af9d68 100644
@@ -43,9 +43,9 @@ int arch_show_interrupts(struct seq_file *p, int prec)
 {
 	int j;
 
-	seq_printf(p, "%*s: ", prec, "NMI");
+	seq_printf(p, "%*s:", prec, "NMI");
 	for_each_online_cpu(j)
-		seq_printf(p, "%10u ", per_cpu(irq_stat.__nmi_count, j));
+		seq_put_decimal_ull_width(p, " ", per_cpu(irq_stat.__nmi_count, j), 10);
 	seq_printf(p, "  Non-maskable interrupts\n");
 
Quoted 7 lines

arch/sh/kernel/irq.c int arch_show_interrupts(struct seq_file *p, int prec) · line 51

Quoted 1 line
arch/sh/kernel/irq.c +0 -0
 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));

Sorry for the very late reply!

I don't quite understand why seq_put_decimal_ull_width() should be faster than seq_printf().

Can you elaborate on this a bit more?

I just checked existing merges of this patch for other architectures which include a more elaborate patch description. If you could do that for SH as well, I can pick the patch for v6.14.

Adrian

Signature · 4 lines
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer
`. `'   Physicist
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913
author
#3 lore raw
Quoted 32 lines · John Paul Adrian Glaubitz

Hi David,

Quoted 29 lines · John Paul Adrian Glaubitz
Quoted 13 lines · David Wang

Performance improvement for reading /proc/interrupts on arch sh

---
 arch/sh/kernel/irq.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/sh/kernel/irq.c +0 -0
index 4e6835de54cf..9022d8af9d68 100644
@@ -43,9 +43,9 @@ int arch_show_interrupts(struct seq_file *p, int prec)
 {

arch/sh/kernel/irq.c int arch_show_interrupts(struct seq_file *p, int prec) · line 51

Quoted 10 lines
arch/sh/kernel/irq.c +2 -2
 	int j;
 
-	seq_printf(p, "%*s: ", prec, "NMI");
+	seq_printf(p, "%*s:", prec, "NMI");
 	for_each_online_cpu(j)
-		seq_printf(p, "%10u ", per_cpu(irq_stat.__nmi_count, j));
+		seq_put_decimal_ull_width(p, " ", per_cpu(irq_stat.__nmi_count, j), 10);
 	seq_printf(p, "  Non-maskable interrupts\n");
 
 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));

Sorry for the very late reply!

I don't quite understand why seq_put_decimal_ull_width() should be faster than seq_printf().

Quoted 13 lines
Quoted 1 line

Can you elaborate on this a bit more?

I just checked existing merges of this patch for other architectures which include a more elaborate patch description. If you could do that for SH as well, I can pick the patch for v6.14.

Adrian

Signature · 4 lines
.''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer
`. `'   Physicist
 `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

Sure, I will make the change.

To be honest, I do not have a SH platform to confirm the performance changes. If convenient, a simple strace -T -e read cat /proc/interrupts > /dev/null can be used to verify any improvement, and the core change in commit f9ed1f7c2e26fcd1978(genirq/proc: Use seq_put_decimal_ull_width() for decimal values) is needed.

Thanks
David

author
#4 lore raw

PATCH v2 1/1 sh/irq: use seq_put_decimal_ull_width() for decimal values

In reply to John Paul Adrian Glaubitz

On a system with n CPUs and m interrupts, there will be n*m decimal values yielded via seq_printf(.."%10u "..) which has significant costs parsing format string and is less efficient than seq_put_decimal_ull_width(). Stress reading /proc/interrupts indicates ~30% performance improvement with this patch.

---
 arch/sh/kernel/irq.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/sh/kernel/irq.c +2 -2
index 4e6835de54cf..9022d8af9d68 100644
@@ -43,9 +43,9 @@ int arch_show_interrupts(struct seq_file *p, int prec)
 {
 	int j;
 
-	seq_printf(p, "%*s: ", prec, "NMI");
+	seq_printf(p, "%*s:", prec, "NMI");
 	for_each_online_cpu(j)
-		seq_printf(p, "%10u ", per_cpu(irq_stat.__nmi_count, j));
+		seq_put_decimal_ull_width(p, " ", per_cpu(irq_stat.__nmi_count, j), 10);
 	seq_printf(p, "  Non-maskable interrupts\n");
 
 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
Signature · 1 line
2.39.2
maintainer Reviewed-by on v2 1/1
#5 lore raw

Hi David,

Quoted 17 lines · David Wang

On a system with n CPUs and m interrupts, there will be n*m decimal values yielded via seq_printf(.."%10u "..) which has significant costs parsing format string and is less efficient than seq_put_decimal_ull_width(). Stress reading /proc/interrupts indicates ~30% performance improvement with this patch.

---
 arch/sh/kernel/irq.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/sh/kernel/irq.c +0 -0
index 4e6835de54cf..9022d8af9d68 100644
@@ -43,9 +43,9 @@ int arch_show_interrupts(struct seq_file *p, int prec)
 {

arch/sh/kernel/irq.c int arch_show_interrupts(struct seq_file *p, int prec) · line 51

Quoted 10 lines
arch/sh/kernel/irq.c +2 -2
 	int j;
 
-	seq_printf(p, "%*s: ", prec, "NMI");
+	seq_printf(p, "%*s:", prec, "NMI");
 	for_each_online_cpu(j)
-		seq_printf(p, "%10u ", per_cpu(irq_stat.__nmi_count, j));
+		seq_put_decimal_ull_width(p, " ", per_cpu(irq_stat.__nmi_count, j), 10);
 	seq_printf(p, "  Non-maskable interrupts\n");
 
 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));

Thanks,
Adrian

Signature · 4 lines
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer
`. `'   Physicist
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913