Re: [PATCH] x86/cpu: sort cpuinfo flags

From: Kirill A. Shutemov
Date: Thu Dec 20 2018 - 07:02:46 EST


On Wed, Dec 19, 2018 at 07:50:14PM +0000, Dave Hansen wrote:
>
> From: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
>
> I frequently find myself contemplating my life choices as I try to
> find 3-character entries in the 1,000-character, unsorted "flags:"
> field of /proc/cpuinfo.
>
> Sort that field, giving me hours back in my day.
>
> This eats up ~1200 bytes (NCAPINTS*2*32) of space for the sorted
> array. I used an 'unsigned short' to use 1/4 of the space on 64-bit
> that would have been needed had pointers been used in the array.
>
> An alternatve, requiring no array, would be to do the sort at runtime,
> but it seems ridiculous for a 500-cpu system to do 500 sorts for each
> 'cat /proc/cpuinfo'.
>
> Another would be to just cache the *string* that results from this,
> which would be even faster at runtime because it could do a single
> seq_printf() and would consume less space. But, that would
> require a bit more infrastructure to make sure that the produced
> string never changed and was consistent across all CPUs, unless
> we want to store a string per 'struct cpuinfo_x86'.
>
> Signed-off-by: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
> Cc: x86@xxxxxxxxxx
> Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> Cc: Borislav Petkov <bp@xxxxxxxxx>
> Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
> Cc: Jia Zhang <qianyue.zj@xxxxxxxxxxxxxxx>
> Cc: "Gustavo A. R. Silva" <garsilva@xxxxxxxxxxxxxx>
> Cc: linux-kernel@xxxxxxxxxxxxxxx
> ---

Below is my attempt on doing the same. The key difference is that the
sorted array is generated at compile-time by mkcapflags.sh.

The patch also sorts bugs.

diff --git a/arch/x86/include/asm/cpufeature.h b/arch/x86/include/asm/cpufeature.h
index 7d442722ef24..37cc8ccaec8f 100644
--- a/arch/x86/include/asm/cpufeature.h
+++ b/arch/x86/include/asm/cpufeature.h
@@ -34,6 +34,7 @@ enum cpuid_leafs

#ifdef CONFIG_X86_FEATURE_NAMES
extern const char * const x86_cap_flags[NCAPINTS*32];
+extern const unsigned short x86_sorted_cap_flags[];
extern const char * const x86_power_flags[32];
#define X86_CAP_FMT "%s"
#define x86_cap_flag(flag) x86_cap_flags[flag]
@@ -47,6 +48,7 @@ extern const char * const x86_power_flags[32];
* X86_BUG_<name> - NCAPINTS*32.
*/
extern const char * const x86_bug_flags[NBUGINTS*32];
+extern const unsigned short x86_sorted_bug_flags[];

#define test_cpu_cap(c, bit) \
test_bit(bit, (unsigned long *)((c)->x86_capability))
diff --git a/arch/x86/kernel/cpu/mkcapflags.sh b/arch/x86/kernel/cpu/mkcapflags.sh
index d0dfb892c72f..e40c3ee9cd58 100644
--- a/arch/x86/kernel/cpu/mkcapflags.sh
+++ b/arch/x86/kernel/cpu/mkcapflags.sh
@@ -7,17 +7,9 @@
IN=$1
OUT=$2

-dump_array()
+dump_values()
{
- ARRAY=$1
- SIZE=$2
- PFX=$3
- POSTFIX=$4
-
- PFX_SZ=$(echo $PFX | wc -c)
- TABS="$(printf '\t\t\t\t\t')"
-
- echo "const char * const $ARRAY[$SIZE] = {"
+ PFX=$1

# Iterate through any input lines starting with #define $PFX
sed -n -e 's/\t/ /g' -e "s/^ *# *define *$PFX//p" $IN |
@@ -34,19 +26,58 @@ dump_array()
# Name is uppercase, VALUE is all lowercase
VALUE="$(echo "$VALUE" | tr A-Z a-z)"

- if [ -n "$POSTFIX" ]; then
- T=$(( $PFX_SZ + $(echo $POSTFIX | wc -c) + 2 ))
- TABS="$(printf '\t\t\t\t\t\t')"
- TABCOUNT=$(( ( 6*8 - ($T + 1) - $(echo "$NAME" | wc -c) ) / 8 ))
- printf "\t[%s - %s]%.*s = %s,\n" "$PFX$NAME" "$POSTFIX" "$TABCOUNT" "$TABS" "$VALUE"
- else
- TABCOUNT=$(( ( 5*8 - ($PFX_SZ + 1) - $(echo "$NAME" | wc -c) ) / 8 ))
- printf "\t[%s]%.*s = %s,\n" "$PFX$NAME" "$TABCOUNT" "$TABS" "$VALUE"
- fi
+ echo "$NAME $VALUE"
+ done
+}
+
+dump_array()
+{
+ ARRAY=$1
+ SIZE=$2
+ PFX=$3
+ POSTFIX=$4
+
+ PFX_SZ=$(echo $PFX | wc -c)
+ TABS="$(printf '\t\t\t\t\t')"
+
+ echo "const char * const $ARRAY[$SIZE] = {"
+
+ dump_values "$PFX" |
+ while read NAME VALUE
+ do
+ if [ -n "$POSTFIX" ]; then
+ T=$(( $PFX_SZ + $(echo $POSTFIX | wc -c) + 2 ))
+ TABS="$(printf '\t\t\t\t\t\t')"
+ TABCOUNT=$(( ( 6*8 - ($T + 1) - $(echo "$NAME" | wc -c) ) / 8 ))
+ printf "\t[%s - %s]%.*s = %s,\n" "$PFX$NAME" "$POSTFIX" "$TABCOUNT" "$TABS" "$VALUE"
+ else
+ TABCOUNT=$(( ( 5*8 - ($PFX_SZ + 1) - $(echo "$NAME" | wc -c) ) / 8 ))
+ printf "\t[%s]%.*s = %s,\n" "$PFX$NAME" "$TABCOUNT" "$TABS" "$VALUE"
+ fi
done
echo "};"
}

+dump_sorted_array()
+{
+ ARRAY=$1
+ PFX=$2
+
+ PFX_SZ=$(echo $PFX | wc -c)
+ TABS="$(printf '\t\t\t\t\t')"
+
+ echo "const unsigned short $ARRAY[] = {"
+
+ dump_values "$PFX" |
+ sort -k 2 |
+ while read NAME VALUE
+ do
+ TABCOUNT=$(( ( 6*8 - $PFX_SZ - $(echo "$NAME" | wc -c) ) / 8 ))
+ printf "\t%s%s,%.*s/* %s */\n" "$PFX" "$NAME" "$TABCOUNT" "$TABS" "$VALUE"
+ done
+ printf "\t-1\n};"
+}
+
trap 'rm "$OUT"' EXIT

(
@@ -58,8 +89,13 @@ trap 'rm "$OUT"' EXIT
dump_array "x86_cap_flags" "NCAPINTS*32" "X86_FEATURE_" ""
echo ""

+ dump_sorted_array "x86_sorted_cap_flags" "X86_FEATURE_"
+ echo ""
+
dump_array "x86_bug_flags" "NBUGINTS*32" "X86_BUG_" "NCAPINTS*32"
+ echo ""

+ dump_sorted_array "x86_sorted_bug_flags" "X86_BUG_"
) > $OUT

trap - EXIT
diff --git a/arch/x86/kernel/cpu/proc.c b/arch/x86/kernel/cpu/proc.c
index 2c8522a39ed5..4a5c1a4c7330 100644
--- a/arch/x86/kernel/cpu/proc.c
+++ b/arch/x86/kernel/cpu/proc.c
@@ -54,6 +54,51 @@ static void show_cpuinfo_misc(struct seq_file *m, struct cpuinfo_x86 *c)
}
#endif

+
+static void show_cpuinfo_flags(struct seq_file *m, struct cpuinfo_x86 *c)
+{
+ int i;
+
+ seq_puts(m, "flags\t\t:");
+
+ for (i = 0; i < 32*NCAPINTS; i++) {
+ /*
+ * Go through the flag list in alphabetical
+ * order to make reading this field easier.
+ */
+ unsigned int cap = x86_sorted_cap_flags[i];
+
+ if (cap == (unsigned short)-1)
+ break;
+
+ if (cpu_has(c, cap) && x86_cap_flags[cap] != NULL)
+ seq_printf(m, " %s", x86_cap_flags[cap]);
+ }
+}
+
+
+static void show_cpuinfo_bugs(struct seq_file *m, struct cpuinfo_x86 *c)
+{
+ int i;
+ seq_puts(m, "\nbugs\t\t:");
+ for (i = 0; i < 32*NBUGINTS; i++) {
+ /*
+ * Go through the flag list in alphabetical
+ * order to make reading this field easier.
+ */
+ unsigned int bug_bit = x86_sorted_bug_flags[i];
+ unsigned int bug;
+
+ if (bug_bit == (unsigned short)-1)
+ break;
+
+ bug = bug_bit - 32*NCAPINTS;
+
+ if (cpu_has_bug(c, bug_bit) && x86_bug_flags[bug])
+ seq_printf(m, " %s", x86_bug_flags[bug]);
+ }
+}
+
static int show_cpuinfo(struct seq_file *m, void *v)
{
struct cpuinfo_x86 *c = v;
@@ -96,19 +141,8 @@ static int show_cpuinfo(struct seq_file *m, void *v)

show_cpuinfo_core(m, c, cpu);
show_cpuinfo_misc(m, c);
-
- seq_puts(m, "flags\t\t:");
- for (i = 0; i < 32*NCAPINTS; i++)
- if (cpu_has(c, i) && x86_cap_flags[i] != NULL)
- seq_printf(m, " %s", x86_cap_flags[i]);
-
- seq_puts(m, "\nbugs\t\t:");
- for (i = 0; i < 32*NBUGINTS; i++) {
- unsigned int bug_bit = 32*NCAPINTS + i;
-
- if (cpu_has_bug(c, bug_bit) && x86_bug_flags[i])
- seq_printf(m, " %s", x86_bug_flags[i]);
- }
+ show_cpuinfo_flags(m, c);
+ show_cpuinfo_bugs(m, c);

seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
c->loops_per_jiffy/(500000/HZ),
--
Kirill A. Shutemov