aboutsummaryrefslogtreecommitdiff
path: root/sysdeps/x86/cpu-tunables.c
diff options
context:
space:
mode:
authorAdhemerval Zanella <adhemerval.zanella@linaro.org>2023-12-06 10:24:01 -0300
committerAdhemerval Zanella <adhemerval.zanella@linaro.org>2023-12-19 13:25:45 -0300
commit2a969b53c0b02fed7e43473a92f219d737fd217a (patch)
treeb55eda5dc496c260e9757a5fc3856838d85b38fd /sysdeps/x86/cpu-tunables.c
parent5275fc784c8113c84c85ca028ce621f68fe6642b (diff)
downloadglibc-2a969b53c0b02fed7e43473a92f219d737fd217a.zip
glibc-2a969b53c0b02fed7e43473a92f219d737fd217a.tar.gz
glibc-2a969b53c0b02fed7e43473a92f219d737fd217a.tar.bz2
elf: Do not duplicate the GLIBC_TUNABLES string
The tunable parsing duplicates the tunable environment variable so it null-terminates each one since it simplifies the later parsing. It has the drawback of adding another point of failure (__minimal_malloc failing), and the memory copy requires tuning the compiler to avoid mem operations calls. The parsing now tracks the tunable start and its size. The dl-tunable-parse.h adds helper functions to help parsing, like a strcmp that also checks for size and an iterator for suboptions that are comma-separated (used on hwcap parsing by x86, powerpc, and s390x). Since the environment variable is allocated on the stack by the kernel, it is safe to keep the references to the suboptions for later parsing of string tunables (as done by set_hwcaps by multiple architectures). Checked on x86_64-linux-gnu, powerpc64le-linux-gnu, and aarch64-linux-gnu. Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Diffstat (limited to 'sysdeps/x86/cpu-tunables.c')
-rw-r--r--sysdeps/x86/cpu-tunables.c118
1 files changed, 43 insertions, 75 deletions
diff --git a/sysdeps/x86/cpu-tunables.c b/sysdeps/x86/cpu-tunables.c
index 5697885..ef96148 100644
--- a/sysdeps/x86/cpu-tunables.c
+++ b/sysdeps/x86/cpu-tunables.c
@@ -24,11 +24,12 @@
#include <string.h>
#include <cpu-features.h>
#include <ldsodefs.h>
+#include <dl-tunables-parse.h>
#include <dl-symbol-redir-ifunc.h>
#define CHECK_GLIBC_IFUNC_CPU_OFF(f, cpu_features, name, len) \
_Static_assert (sizeof (#name) - 1 == len, #name " != " #len); \
- if (memcmp (f, #name, len) == 0) \
+ if (tunable_str_comma_strcmp_cte (&f, #name)) \
{ \
CPU_FEATURE_UNSET (cpu_features, name) \
break; \
@@ -38,7 +39,7 @@
which isn't available. */
#define CHECK_GLIBC_IFUNC_PREFERRED_OFF(f, cpu_features, name, len) \
_Static_assert (sizeof (#name) - 1 == len, #name " != " #len); \
- if (memcmp (f, #name, len) == 0) \
+ if (tunable_str_comma_strcmp_cte (&f, #name) == 0) \
{ \
cpu_features->preferred[index_arch_##name] \
&= ~bit_arch_##name; \
@@ -46,12 +47,11 @@
}
/* Enable/disable a preferred feature NAME. */
-#define CHECK_GLIBC_IFUNC_PREFERRED_BOTH(f, cpu_features, name, \
- disable, len) \
+#define CHECK_GLIBC_IFUNC_PREFERRED_BOTH(f, cpu_features, name, len) \
_Static_assert (sizeof (#name) - 1 == len, #name " != " #len); \
- if (memcmp (f, #name, len) == 0) \
+ if (tunable_str_comma_strcmp_cte (&f, #name)) \
{ \
- if (disable) \
+ if (f.disable) \
cpu_features->preferred[index_arch_##name] &= ~bit_arch_##name; \
else \
cpu_features->preferred[index_arch_##name] |= bit_arch_##name; \
@@ -61,11 +61,11 @@
/* Enable/disable a preferred feature NAME. Enable a preferred feature
only if the feature NEED is usable. */
#define CHECK_GLIBC_IFUNC_PREFERRED_NEED_BOTH(f, cpu_features, name, \
- need, disable, len) \
+ need, len) \
_Static_assert (sizeof (#name) - 1 == len, #name " != " #len); \
- if (memcmp (f, #name, len) == 0) \
+ if (tunable_str_comma_strcmp_cte (&f, #name)) \
{ \
- if (disable) \
+ if (f.disable) \
cpu_features->preferred[index_arch_##name] &= ~bit_arch_##name; \
else if (CPU_FEATURE_USABLE_P (cpu_features, need)) \
cpu_features->preferred[index_arch_##name] |= bit_arch_##name; \
@@ -93,38 +93,20 @@ TUNABLE_CALLBACK (set_hwcaps) (tunable_val_t *valp)
NOTE: the IFUNC selection may change over time. Please check all
multiarch implementations when experimenting. */
- const char *p = valp->strval, *c;
struct cpu_features *cpu_features = &GLRO(dl_x86_cpu_features);
- size_t len;
- do
- {
- const char *n;
- bool disable;
- size_t nl;
-
- for (c = p; *c != ','; c++)
- if (*c == '\0')
- break;
+ struct tunable_str_comma_state_t ts;
+ tunable_str_comma_init (&ts, valp);
- len = c - p;
- disable = *p == '-';
- if (disable)
- {
- n = p + 1;
- nl = len - 1;
- }
- else
- {
- n = p;
- nl = len;
- }
- switch (nl)
+ struct tunable_str_comma_t n;
+ while (tunable_str_comma_next (&ts, &n))
+ {
+ switch (n.len)
{
default:
break;
case 3:
- if (disable)
+ if (n.disable)
{
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, AVX, 3);
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, CX8, 3);
@@ -135,7 +117,7 @@ TUNABLE_CALLBACK (set_hwcaps) (tunable_val_t *valp)
}
break;
case 4:
- if (disable)
+ if (n.disable)
{
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, AVX2, 4);
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, BMI1, 4);
@@ -149,7 +131,7 @@ TUNABLE_CALLBACK (set_hwcaps) (tunable_val_t *valp)
}
break;
case 5:
- if (disable)
+ if (n.disable)
{
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, LZCNT, 5);
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, MOVBE, 5);
@@ -159,12 +141,12 @@ TUNABLE_CALLBACK (set_hwcaps) (tunable_val_t *valp)
}
break;
case 6:
- if (disable)
+ if (n.disable)
{
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, POPCNT, 6);
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, SSE4_1, 6);
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, SSE4_2, 6);
- if (memcmp (n, "XSAVEC", 6) == 0)
+ if (memcmp (n.str, "XSAVEC", 6) == 0)
{
/* Update xsave_state_size to XSAVE state size. */
cpu_features->xsave_state_size
@@ -174,14 +156,14 @@ TUNABLE_CALLBACK (set_hwcaps) (tunable_val_t *valp)
}
break;
case 7:
- if (disable)
+ if (n.disable)
{
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, AVX512F, 7);
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, OSXSAVE, 7);
}
break;
case 8:
- if (disable)
+ if (n.disable)
{
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, AVX512CD, 8);
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, AVX512BW, 8);
@@ -190,86 +172,72 @@ TUNABLE_CALLBACK (set_hwcaps) (tunable_val_t *valp)
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, AVX512PF, 8);
CHECK_GLIBC_IFUNC_CPU_OFF (n, cpu_features, AVX512VL, 8);
}
- CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features, Slow_BSF,
- disable, 8);
+ CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features, Slow_BSF, 8);
break;
case 11:
{
- CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features,
- Prefer_ERMS,
- disable, 11);
- CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features,
- Prefer_FSRM,
- disable, 11);
+ CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features, Prefer_ERMS,
+ 11);
+ CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features, Prefer_FSRM,
+ 11);
CHECK_GLIBC_IFUNC_PREFERRED_NEED_BOTH (n, cpu_features,
Slow_SSE4_2,
SSE4_2,
- disable, 11);
+ 11);
}
break;
case 15:
{
CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features,
- Fast_Rep_String,
- disable, 15);
+ Fast_Rep_String, 15);
}
break;
case 16:
{
CHECK_GLIBC_IFUNC_PREFERRED_NEED_BOTH
- (n, cpu_features, Prefer_No_AVX512, AVX512F,
- disable, 16);
+ (n, cpu_features, Prefer_No_AVX512, AVX512F, 16);
}
break;
case 18:
{
CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features,
- Fast_Copy_Backward,
- disable, 18);
+ Fast_Copy_Backward, 18);
}
break;
case 19:
{
CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features,
- Fast_Unaligned_Load,
- disable, 19);
+ Fast_Unaligned_Load, 19);
CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features,
- Fast_Unaligned_Copy,
- disable, 19);
+ Fast_Unaligned_Copy, 19);
}
break;
case 20:
{
CHECK_GLIBC_IFUNC_PREFERRED_NEED_BOTH
- (n, cpu_features, Prefer_No_VZEROUPPER, AVX, disable,
- 20);
+ (n, cpu_features, Prefer_No_VZEROUPPER, AVX, 20);
}
break;
case 23:
{
CHECK_GLIBC_IFUNC_PREFERRED_NEED_BOTH
- (n, cpu_features, AVX_Fast_Unaligned_Load, AVX,
- disable, 23);
+ (n, cpu_features, AVX_Fast_Unaligned_Load, AVX, 23);
}
break;
case 24:
{
CHECK_GLIBC_IFUNC_PREFERRED_NEED_BOTH
- (n, cpu_features, MathVec_Prefer_No_AVX512, AVX512F,
- disable, 24);
+ (n, cpu_features, MathVec_Prefer_No_AVX512, AVX512F, 24);
}
break;
case 26:
{
CHECK_GLIBC_IFUNC_PREFERRED_NEED_BOTH
- (n, cpu_features, Prefer_PMINUB_for_stringop, SSE2,
- disable, 26);
+ (n, cpu_features, Prefer_PMINUB_for_stringop, SSE2, 26);
}
break;
}
- p += len + 1;
}
- while (*c != '\0');
}
#if CET_ENABLED
@@ -277,11 +245,11 @@ attribute_hidden
void
TUNABLE_CALLBACK (set_x86_ibt) (tunable_val_t *valp)
{
- if (memcmp (valp->strval, "on", sizeof ("on")) == 0)
+ if (tunable_strcmp_cte (valp, "on"))
GL(dl_x86_feature_control).ibt = cet_always_on;
- else if (memcmp (valp->strval, "off", sizeof ("off")) == 0)
+ else if (tunable_strcmp_cte (valp, "off"))
GL(dl_x86_feature_control).ibt = cet_always_off;
- else if (memcmp (valp->strval, "permissive", sizeof ("permissive")) == 0)
+ else if (tunable_strcmp_cte (valp, "permissive"))
GL(dl_x86_feature_control).ibt = cet_permissive;
}
@@ -289,11 +257,11 @@ attribute_hidden
void
TUNABLE_CALLBACK (set_x86_shstk) (tunable_val_t *valp)
{
- if (memcmp (valp->strval, "on", sizeof ("on")) == 0)
+ if (tunable_strcmp_cte (valp, "on"))
GL(dl_x86_feature_control).shstk = cet_always_on;
- else if (memcmp (valp->strval, "off", sizeof ("off")) == 0)
+ else if (tunable_strcmp_cte (valp, "off"))
GL(dl_x86_feature_control).shstk = cet_always_off;
- else if (memcmp (valp->strval, "permissive", sizeof ("permissive")) == 0)
+ else if (tunable_strcmp_cte (valp, "permissive"))
GL(dl_x86_feature_control).shstk = cet_permissive;
}
#endif