| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20230424001956.21434-2-ricardo.neri-calderon@linux.intel.com>
Date: Sun, 23 Apr 2023 17:19:55 -0700
From: Ricardo Neri <ricardo.neri-calderon@...ux.intel.com>
To: x86@...nel.org
Cc: Andreas Herrmann <aherrmann@...e.com>,
Chen Yu <yu.c.chen@...el.com>, Len Brown <len.brown@...el.com>,
Pu Wen <puwen@...on.cn>,
"Rafael J. Wysocki" <rafael.j.wysocki@...el.com>,
Srinivas Pandruvada <srinivas.pandruvada@...ux.intel.com>,
Zhang Rui <rui.zhang@...el.com>,
Ricardo Neri <ricardo.neri@...el.com>,
linux-kernel@...r.kernel.org,
Ricardo Neri <ricardo.neri-calderon@...ux.intel.com>,
stable@...r.kernel.org
Subject: [PATCH v2 1/2] x86/cacheinfo: Delete global num_cache_leaves
Linux remembers cpu_cachinfo::num_leaves per CPU, but x86 initializes all
CPUs from the same global "num_cache_leaves".
This is erroneous on systems like Meteor Lake, which has different
num_leaves per CPU. Delete the global "num_cache_leaves" and initialize
num_leaves accurately on each CPU.
Cc: Andreas Herrmann <aherrmann@...e.com>
Cc: Chen Yu <yu.c.chen@...el.com>
Cc: Len Brown <len.brown@...el.com>
Cc: Pu Wen <puwen@...on.cn>
Cc: "Rafael J. Wysocki" <rafael.j.wysocki@...el.com>
Cc: Srinivas Pandruvada <srinivas.pandruvada@...ux.intel.com>
Cc: Zhang Rui <rui.zhang@...el.com>
Cc: stable@...r.kernel.org
Reviewed-by: Len Brown <len.brown@...el.com>
Signed-off-by: Ricardo Neri <ricardo.neri-calderon@...ux.intel.com>
---
After this change, all CPUs will traverse CPUID leaf 0x4 when booted for
the first time. On systems with asymmetric cache topologies this is
useless work.
Creating a list of processor models that have asymmetric cache topologies
was considered. The burden of maintaining such list would outweigh the
performance benefit of skipping this extra step.
---
Changes since v1:
* Do not make num_cache_leaves a per-CPU variable. Instead, reuse the
existing per-CPU ci_cpu_cacheinfo variable. (Dave Hansen)
---
arch/x86/kernel/cpu/cacheinfo.c | 45 ++++++++++++++++++---------------
1 file changed, 25 insertions(+), 20 deletions(-)
diff --git a/arch/x86/kernel/cpu/cacheinfo.c b/arch/x86/kernel/cpu/cacheinfo.c
index 4063e8991211..45c4e9daf3f1 100644
--- a/arch/x86/kernel/cpu/cacheinfo.c
+++ b/arch/x86/kernel/cpu/cacheinfo.c
@@ -176,7 +176,16 @@ struct _cpuid4_info_regs {
struct amd_northbridge *nb;
};
-static unsigned short num_cache_leaves;
+static inline unsigned int get_num_cache_leaves(unsigned int cpu)
+{
+ return get_cpu_cacheinfo(cpu)->num_leaves;
+}
+
+static inline void
+set_num_cache_leaves(unsigned int nr_leaves, unsigned int cpu)
+{
+ get_cpu_cacheinfo(cpu)->num_leaves = nr_leaves;
+}
/* AMD doesn't have CPUID4. Emulate it here to report the same
information to the user. This makes some assumptions about the machine:
@@ -716,19 +725,21 @@ void cacheinfo_hygon_init_llc_id(struct cpuinfo_x86 *c, int cpu)
void init_amd_cacheinfo(struct cpuinfo_x86 *c)
{
+ unsigned int cpu = c->cpu_index;
+
if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
- num_cache_leaves = find_num_cache_leaves(c);
+ set_num_cache_leaves(find_num_cache_leaves(c), cpu);
} else if (c->extended_cpuid_level >= 0x80000006) {
if (cpuid_edx(0x80000006) & 0xf000)
- num_cache_leaves = 4;
+ set_num_cache_leaves(4, cpu);
else
- num_cache_leaves = 3;
+ set_num_cache_leaves(3, cpu);
}
}
void init_hygon_cacheinfo(struct cpuinfo_x86 *c)
{
- num_cache_leaves = find_num_cache_leaves(c);
+ set_num_cache_leaves(find_num_cache_leaves(c), c->cpu_index);
}
void init_intel_cacheinfo(struct cpuinfo_x86 *c)
@@ -738,24 +749,21 @@ void init_intel_cacheinfo(struct cpuinfo_x86 *c)
unsigned int new_l1d = 0, new_l1i = 0; /* Cache sizes from cpuid(4) */
unsigned int new_l2 = 0, new_l3 = 0, i; /* Cache sizes from cpuid(4) */
unsigned int l2_id = 0, l3_id = 0, num_threads_sharing, index_msb;
-#ifdef CONFIG_SMP
unsigned int cpu = c->cpu_index;
-#endif
if (c->cpuid_level > 3) {
- static int is_initialized;
-
- if (is_initialized == 0) {
- /* Init num_cache_leaves from boot CPU */
- num_cache_leaves = find_num_cache_leaves(c);
- is_initialized++;
- }
+ /*
+ * There should be at least one leaf. A non-zero value means
+ * that the number of leaves has been initialized.
+ */
+ if (!get_num_cache_leaves(cpu))
+ set_num_cache_leaves(find_num_cache_leaves(c), cpu);
/*
* Whenever possible use cpuid(4), deterministic cache
* parameters cpuid leaf to find the cache details
*/
- for (i = 0; i < num_cache_leaves; i++) {
+ for (i = 0; i < get_num_cache_leaves(cpu); i++) {
struct _cpuid4_info_regs this_leaf = {};
int retval;
@@ -791,14 +799,14 @@ void init_intel_cacheinfo(struct cpuinfo_x86 *c)
* Don't use cpuid2 if cpuid4 is supported. For P4, we use cpuid2 for
* trace cache
*/
- if ((num_cache_leaves == 0 || c->x86 == 15) && c->cpuid_level > 1) {
+ if ((!get_num_cache_leaves(cpu) || c->x86 == 15) && c->cpuid_level > 1) {
/* supports eax=2 call */
int j, n;
unsigned int regs[4];
unsigned char *dp = (unsigned char *)regs;
int only_trace = 0;
- if (num_cache_leaves != 0 && c->x86 == 15)
+ if (get_num_cache_leaves(cpu) && c->x86 == 15)
only_trace = 1;
/* Number of times to iterate */
@@ -1000,12 +1008,9 @@ int init_cache_level(unsigned int cpu)
{
struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
- if (!num_cache_leaves)
- return -ENOENT;
if (!this_cpu_ci)
return -EINVAL;
this_cpu_ci->num_levels = 3;
- this_cpu_ci->num_leaves = num_cache_leaves;
return 0;
}
--
2.25.1
Powered by blists - more mailing lists