[PATCH v2 2/2] x86, cpuid: allow cpuid_read() to schedule

From: Eric Dumazet
Date: Mon Mar 19 2018 - 11:25:24 EST


I noticed high latencies caused by a daemon periodically reading various
MSR and cpuid on all cpus. KASAN kernels would see ~10ms latencies
simply reading one cpuid. Even without KASAN, sending IPI to CPU
in deep sleep state or blocking hard IRQ in a a long section,
then waiting for the answer can consume hundreds of usec or more.

Switching to smp_call_function_single_async() and a completion
allows to reschedule and not burn cpu cycles.

Signed-off-by: Eric Dumazet <edumazet@xxxxxxxxxx>
Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Cc: Ingo Molnar <mingo@xxxxxxxxxx>
Cc: Hugh Dickins <hughd@xxxxxxxxxx>
---
arch/x86/kernel/cpuid.c | 34 ++++++++++++++++++++++++++--------
1 file changed, 26 insertions(+), 8 deletions(-)

diff --git a/arch/x86/kernel/cpuid.c b/arch/x86/kernel/cpuid.c
index 0931a105ffe16cde4640e759efa600b23a756d84..1d300f96df4b316dbe3392c8221467cfd8593272 100644
--- a/arch/x86/kernel/cpuid.c
+++ b/arch/x86/kernel/cpuid.c
@@ -40,6 +40,7 @@
#include <linux/notifier.h>
#include <linux/uaccess.h>
#include <linux/gfp.h>
+#include <linux/completion.h>

#include <asm/processor.h>
#include <asm/msr.h>
@@ -47,19 +48,27 @@
static struct class *cpuid_class;
static enum cpuhp_state cpuhp_cpuid_state;

+struct cpuid_regs_done {
+ struct cpuid_regs regs;
+ struct completion done;
+};
+
static void cpuid_smp_cpuid(void *cmd_block)
{
- struct cpuid_regs *cmd = (struct cpuid_regs *)cmd_block;
+ struct cpuid_regs_done *cmd = cmd_block;
+
+ cpuid_count(cmd->regs.eax, cmd->regs.ecx,
+ &cmd->regs.eax, &cmd->regs.ebx,
+ &cmd->regs.ecx, &cmd->regs.edx);

- cpuid_count(cmd->eax, cmd->ecx,
- &cmd->eax, &cmd->ebx, &cmd->ecx, &cmd->edx);
+ complete(&cmd->done);
}

static ssize_t cpuid_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)
{
char __user *tmp = buf;
- struct cpuid_regs cmd;
+ struct cpuid_regs_done cmd;
int cpu = iminor(file_inode(file));
u64 pos = *ppos;
ssize_t bytes = 0;
@@ -68,19 +77,28 @@ static ssize_t cpuid_read(struct file *file, char __user *buf,
if (count % 16)
return -EINVAL; /* Invalid chunk size */

+ init_completion(&cmd.done);
for (; count; count -= 16) {
- cmd.eax = pos;
- cmd.ecx = pos >> 32;
- err = smp_call_function_single(cpu, cpuid_smp_cpuid, &cmd, 1);
+ call_single_data_t csd = {
+ .func = cpuid_smp_cpuid,
+ .info = &cmd,
+ };
+
+ cmd.regs.eax = pos;
+ cmd.regs.ecx = pos >> 32;
+
+ err = smp_call_function_single_async(cpu, &csd);
if (err)
break;
- if (copy_to_user(tmp, &cmd, 16)) {
+ wait_for_completion(&cmd.done);
+ if (copy_to_user(tmp, &cmd.regs, 16)) {
err = -EFAULT;
break;
}
tmp += 16;
bytes += 16;
*ppos = ++pos;
+ reinit_completion(&cmd.done);
}

return bytes ? bytes : err;
--
2.16.2.804.g6dcf76e118-goog