[PATCH] x86/mce: /dev/mcelog: Dynamically allocate space for machine check records

[PATCH] x86/mce: /dev/mcelog: Dynamically allocate space for machine check records

[Tony Luck]

We have had a hard coded limit of 32 machine check records since the
dawn of time.  But as numbers of cores increase, it is possible for
more than 32 errors to be reported before a user process reads from
/dev/mcelog. In this case the additional errors are lost.

Keep 32 as the minimum. But tune the maximum value up based on the
number of processors.

Signed-off-by: Tony Luck <tony.luck@intel.com>
---
 arch/x86/include/asm/mce.h           |  6 ++--
 arch/x86/kernel/cpu/mce/dev-mcelog.c | 46 ++++++++++++++++------------
 2 files changed, 29 insertions(+), 23 deletions(-)

diff --git a/arch/x86/include/asm/mce.h b/arch/x86/include/asm/mce.h
index 4359b955e0b7..c9411a3dcdbe 100644
--- a/arch/x86/include/asm/mce.h
+++ b/arch/x86/include/asm/mce.h
@@ -102,7 +102,7 @@
 
 #define MCE_OVERFLOW 0		/* bit 0 in flags means overflow */
 
-#define MCE_LOG_LEN 32
+#define MCE_LOG_MIN_LEN 32
 #define MCE_LOG_SIGNATURE	"MACHINECHECK"
 
 /* AMD Scalable MCA */
@@ -135,11 +135,11 @@
  */
 struct mce_log_buffer {
 	char signature[12]; /* "MACHINECHECK" */
-	unsigned len;	    /* = MCE_LOG_LEN */
+	unsigned len;	    /* = elements in .mce_entry[] */
 	unsigned next;
 	unsigned flags;
 	unsigned recordlen;	/* length of struct mce */
-	struct mce entry[MCE_LOG_LEN];
+	struct mce entry[];
 };
 
 enum mce_notifier_prios {
diff --git a/arch/x86/kernel/cpu/mce/dev-mcelog.c b/arch/x86/kernel/cpu/mce/dev-mcelog.c
index 7c8958dee103..83c81bb3aec9 100644
--- a/arch/x86/kernel/cpu/mce/dev-mcelog.c
+++ b/arch/x86/kernel/cpu/mce/dev-mcelog.c
@@ -29,11 +29,7 @@ static char *mce_helper_argv[2] = { mce_helper, NULL };
  * separate MCEs from kernel messages to avoid bogus bug reports.
  */
 
-static struct mce_log_buffer mcelog = {
-	.signature	= MCE_LOG_SIGNATURE,
-	.len		= MCE_LOG_LEN,
-	.recordlen	= sizeof(struct mce),
-};
+static struct mce_log_buffer *mcelog;
 
 static DECLARE_WAIT_QUEUE_HEAD(mce_chrdev_wait);
 
@@ -45,21 +41,21 @@ static int dev_mce_log(struct notifier_block *nb, unsigned long val,
 
 	mutex_lock(&mce_chrdev_read_mutex);
 
-	entry = mcelog.next;
+	entry = mcelog->next;
 
 	/*
 	 * When the buffer fills up discard new entries. Assume that the
 	 * earlier errors are the more interesting ones:
 	 */
-	if (entry >= MCE_LOG_LEN) {
-		set_bit(MCE_OVERFLOW, (unsigned long *)&mcelog.flags);
+	if (entry >= mcelog->len) {
+		set_bit(MCE_OVERFLOW, (unsigned long *)&mcelog->flags);
 		goto unlock;
 	}
 
-	mcelog.next = entry + 1;
+	mcelog->next = entry + 1;
 
-	memcpy(mcelog.entry + entry, mce, sizeof(struct mce));
-	mcelog.entry[entry].finished = 1;
+	memcpy(mcelog->entry + entry, mce, sizeof(struct mce));
+	mcelog->entry[entry].finished = 1;
 
 	/* wake processes polling /dev/mcelog */
 	wake_up_interruptible(&mce_chrdev_wait);
@@ -214,21 +210,21 @@ static ssize_t mce_chrdev_read(struct file *filp, char __user *ubuf,
 
 	/* Only supports full reads right now */
 	err = -EINVAL;
-	if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce))
+	if (*off != 0 || usize < mcelog->len*sizeof(struct mce))
 		goto out;
 
-	next = mcelog.next;
+	next = mcelog->next;
 	err = 0;
 
 	for (i = 0; i < next; i++) {
-		struct mce *m = &mcelog.entry[i];
+		struct mce *m = &mcelog->entry[i];
 
 		err |= copy_to_user(buf, m, sizeof(*m));
 		buf += sizeof(*m);
 	}
 
-	memset(mcelog.entry, 0, next * sizeof(struct mce));
-	mcelog.next = 0;
+	memset(mcelog->entry, 0, next * sizeof(struct mce));
+	mcelog->next = 0;
 
 	if (err)
 		err = -EFAULT;
@@ -242,7 +238,7 @@ static ssize_t mce_chrdev_read(struct file *filp, char __user *ubuf,
 static __poll_t mce_chrdev_poll(struct file *file, poll_table *wait)
 {
 	poll_wait(file, &mce_chrdev_wait, wait);
-	if (READ_ONCE(mcelog.next))
+	if (READ_ONCE(mcelog->next))
 		return EPOLLIN | EPOLLRDNORM;
 	if (!mce_apei_read_done && apei_check_mce())
 		return EPOLLIN | EPOLLRDNORM;
@@ -261,13 +257,13 @@ static long mce_chrdev_ioctl(struct file *f, unsigned int cmd,
 	case MCE_GET_RECORD_LEN:
 		return put_user(sizeof(struct mce), p);
 	case MCE_GET_LOG_LEN:
-		return put_user(MCE_LOG_LEN, p);
+		return put_user(mcelog->len, p);
 	case MCE_GETCLEAR_FLAGS: {
 		unsigned flags;
 
 		do {
-			flags = mcelog.flags;
-		} while (cmpxchg(&mcelog.flags, flags, 0) != flags);
+			flags = mcelog->flags;
+		} while (cmpxchg(&mcelog->flags, flags, 0) != flags);
 
 		return put_user(flags, p);
 	}
@@ -340,6 +336,15 @@ static struct miscdevice mce_chrdev_device = {
 static __init int dev_mcelog_init_device(void)
 {
 	int err;
+	int mce_log_len;
+
+	mce_log_len = max(MCE_LOG_MIN_LEN, num_online_cpus());
+	mcelog = kzalloc(sizeof(*mcelog) + mce_log_len * sizeof(struct mce), GFP_KERNEL);
+	if (!mcelog)
+		return -ENOMEM;
+	strncpy(mcelog->signature, MCE_LOG_SIGNATURE, sizeof(mcelog->signature));
+	mcelog->len = mce_log_len;
+	mcelog->recordlen = sizeof(struct mce);
 
 	/* register character device /dev/mcelog */
 	err = misc_register(&mce_chrdev_device);
@@ -350,6 +355,7 @@ static __init int dev_mcelog_init_device(void)
 		else
 			pr_err("Unable to init device /dev/mcelog (rc: %d)\n", err);
 
+		kfree(mcelog);
 		return err;
 	}
 
-- 
2.21.1

Re: [PATCH] x86/mce: /dev/mcelog: Dynamically allocate space for machine check records

[Luck, Tony]

> On Feb 7, 2020, at 16:06, Tony Luck <tony.luck@intel.com> wrote:
> 
> We have had a hard coded limit of 32 machine check records since the
> dawn of time.  But as numbers of cores increase, it is possible for
> more than 32 errors to be reported 

Note that this is not a theoretical issue. Validation
team is hitting this in a case that doesn’t look
completely implausible.

-Tony

Re: [PATCH] x86/mce: /dev/mcelog: Dynamically allocate space for machine check records

[Borislav Petkov]

On Fri, Feb 07, 2020 at 04:05:51PM -0800, Tony Luck wrote:
> We have had a hard coded limit of 32 machine check records since the
> dawn of time.  But as numbers of cores increase, it is possible for
> more than 32 errors to be reported before a user process reads from
> /dev/mcelog. In this case the additional errors are lost.
> 
> Keep 32 as the minimum. But tune the maximum value up based on the
> number of processors.
> 
> Signed-off-by: Tony Luck <tony.luck@intel.com>
> ---
>  arch/x86/include/asm/mce.h           |  6 ++--
>  arch/x86/kernel/cpu/mce/dev-mcelog.c | 46 ++++++++++++++++------------
>  2 files changed, 29 insertions(+), 23 deletions(-)

...

> @@ -214,21 +210,21 @@ static ssize_t mce_chrdev_read(struct file *filp, char __user *ubuf,
>  
>  	/* Only supports full reads right now */
>  	err = -EINVAL;
> -	if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce))
> +	if (*off != 0 || usize < mcelog->len*sizeof(struct mce))

Add spaces around that *

>  		goto out;
>  
> -	next = mcelog.next;
> +	next = mcelog->next;
>  	err = 0;
>  
>  	for (i = 0; i < next; i++) {
> -		struct mce *m = &mcelog.entry[i];
> +		struct mce *m = &mcelog->entry[i];
>  
>  		err |= copy_to_user(buf, m, sizeof(*m));
>  		buf += sizeof(*m);
>  	}
>  
> -	memset(mcelog.entry, 0, next * sizeof(struct mce));
> -	mcelog.next = 0;
> +	memset(mcelog->entry, 0, next * sizeof(struct mce));
> +	mcelog->next = 0;
>  
>  	if (err)
>  		err = -EFAULT;

...

> @@ -340,6 +336,15 @@ static struct miscdevice mce_chrdev_device = {
>  static __init int dev_mcelog_init_device(void)
>  {
>  	int err;
> +	int mce_log_len;

Please sort function local variables declaration in a reverse christmas
tree order:

	<type A> longest_variable_name;
	<type B> shorter_var_name;
	<type C> even_shorter;
	<type D> i;

> +
> +	mce_log_len = max(MCE_LOG_MIN_LEN, num_online_cpus());

arch/x86/kernel/cpu/mce/dev-mcelog.c: In function ‘dev_mcelog_init_device’:
./include/linux/kernel.h:835:29: warning: comparison of distinct pointer types lacks a cast
  835 |   (!!(sizeof((typeof(x) *)1 == (typeof(y) *)1)))
      |                             ^~
./include/linux/kernel.h:849:4: note: in expansion of macro ‘__typecheck’
  849 |   (__typecheck(x, y) && __no_side_effects(x, y))
      |    ^~~~~~~~~~~
./include/linux/kernel.h:859:24: note: in expansion of macro ‘__safe_cmp’
  859 |  __builtin_choose_expr(__safe_cmp(x, y), \
      |                        ^~~~~~~~~~
./include/linux/kernel.h:875:19: note: in expansion of macro ‘__careful_cmp’
  875 | #define max(x, y) __careful_cmp(x, y, >)
      |                   ^~~~~~~~~~~~~
arch/x86/kernel/cpu/mce/dev-mcelog.c:341:16: note: in expansion of macro ‘max’
  341 |  mce_log_len = max(MCE_LOG_MIN_LEN, num_online_cpus());
      |                ^~~

That MCE_LOG_MIN_LEN wants to be 32U.

> +	mcelog = kzalloc(sizeof(*mcelog) + mce_log_len * sizeof(struct mce), GFP_KERNEL);
> +	if (!mcelog)
> +		return -ENOMEM;

<---- newline here.

> +	strncpy(mcelog->signature, MCE_LOG_SIGNATURE, sizeof(mcelog->signature));
> +	mcelog->len = mce_log_len;
> +	mcelog->recordlen = sizeof(struct mce);
>  
>  	/* register character device /dev/mcelog */
>  	err = misc_register(&mce_chrdev_device);

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette

[PATCH v2] x86/mce: /dev/mcelog: Dynamically allocate space for machine check records

[Luck, Tony]

We have had a hard coded limit of 32 machine check records since the
dawn of time.  But as numbers of cores increase, it is possible for
more than 32 errors to be reported before a user process reads from
/dev/mcelog. In this case the additional errors are lost.

Keep 32 as the minimum. But tune the maximum value up based on the
number of processors.

Signed-off-by: Tony Luck <tony.luck@intel.com>
---
v2: Fix comments from Boris
	1) Spaces around a multiply '*' operator
	2) Reverse Christmas tree order for local variables
	3) MCE_LOG_MIN_LEN need to be unsigned for use by max()
	4) Missing blank line

 arch/x86/include/asm/mce.h           |  6 ++--
 arch/x86/kernel/cpu/mce/dev-mcelog.c | 47 ++++++++++++++++------------
 2 files changed, 30 insertions(+), 23 deletions(-)

diff --git a/arch/x86/include/asm/mce.h b/arch/x86/include/asm/mce.h
index 4359b955e0b7..9d5b09913ef3 100644
--- a/arch/x86/include/asm/mce.h
+++ b/arch/x86/include/asm/mce.h
@@ -102,7 +102,7 @@
 
 #define MCE_OVERFLOW 0		/* bit 0 in flags means overflow */
 
-#define MCE_LOG_LEN 32
+#define MCE_LOG_MIN_LEN 32U
 #define MCE_LOG_SIGNATURE	"MACHINECHECK"
 
 /* AMD Scalable MCA */
@@ -135,11 +135,11 @@
  */
 struct mce_log_buffer {
 	char signature[12]; /* "MACHINECHECK" */
-	unsigned len;	    /* = MCE_LOG_LEN */
+	unsigned len;	    /* = elements in .mce_entry[] */
 	unsigned next;
 	unsigned flags;
 	unsigned recordlen;	/* length of struct mce */
-	struct mce entry[MCE_LOG_LEN];
+	struct mce entry[];
 };
 
 enum mce_notifier_prios {
diff --git a/arch/x86/kernel/cpu/mce/dev-mcelog.c b/arch/x86/kernel/cpu/mce/dev-mcelog.c
index 7c8958dee103..d089567a9ce8 100644
--- a/arch/x86/kernel/cpu/mce/dev-mcelog.c
+++ b/arch/x86/kernel/cpu/mce/dev-mcelog.c
@@ -29,11 +29,7 @@ static char *mce_helper_argv[2] = { mce_helper, NULL };
  * separate MCEs from kernel messages to avoid bogus bug reports.
  */
 
-static struct mce_log_buffer mcelog = {
-	.signature	= MCE_LOG_SIGNATURE,
-	.len		= MCE_LOG_LEN,
-	.recordlen	= sizeof(struct mce),
-};
+static struct mce_log_buffer *mcelog;
 
 static DECLARE_WAIT_QUEUE_HEAD(mce_chrdev_wait);
 
@@ -45,21 +41,21 @@ static int dev_mce_log(struct notifier_block *nb, unsigned long val,
 
 	mutex_lock(&mce_chrdev_read_mutex);
 
-	entry = mcelog.next;
+	entry = mcelog->next;
 
 	/*
 	 * When the buffer fills up discard new entries. Assume that the
 	 * earlier errors are the more interesting ones:
 	 */
-	if (entry >= MCE_LOG_LEN) {
-		set_bit(MCE_OVERFLOW, (unsigned long *)&mcelog.flags);
+	if (entry >= mcelog->len) {
+		set_bit(MCE_OVERFLOW, (unsigned long *)&mcelog->flags);
 		goto unlock;
 	}
 
-	mcelog.next = entry + 1;
+	mcelog->next = entry + 1;
 
-	memcpy(mcelog.entry + entry, mce, sizeof(struct mce));
-	mcelog.entry[entry].finished = 1;
+	memcpy(mcelog->entry + entry, mce, sizeof(struct mce));
+	mcelog->entry[entry].finished = 1;
 
 	/* wake processes polling /dev/mcelog */
 	wake_up_interruptible(&mce_chrdev_wait);
@@ -214,21 +210,21 @@ static ssize_t mce_chrdev_read(struct file *filp, char __user *ubuf,
 
 	/* Only supports full reads right now */
 	err = -EINVAL;
-	if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce))
+	if (*off != 0 || usize < mcelog->len * sizeof(struct mce))
 		goto out;
 
-	next = mcelog.next;
+	next = mcelog->next;
 	err = 0;
 
 	for (i = 0; i < next; i++) {
-		struct mce *m = &mcelog.entry[i];
+		struct mce *m = &mcelog->entry[i];
 
 		err |= copy_to_user(buf, m, sizeof(*m));
 		buf += sizeof(*m);
 	}
 
-	memset(mcelog.entry, 0, next * sizeof(struct mce));
-	mcelog.next = 0;
+	memset(mcelog->entry, 0, next * sizeof(struct mce));
+	mcelog->next = 0;
 
 	if (err)
 		err = -EFAULT;
@@ -242,7 +238,7 @@ static ssize_t mce_chrdev_read(struct file *filp, char __user *ubuf,
 static __poll_t mce_chrdev_poll(struct file *file, poll_table *wait)
 {
 	poll_wait(file, &mce_chrdev_wait, wait);
-	if (READ_ONCE(mcelog.next))
+	if (READ_ONCE(mcelog->next))
 		return EPOLLIN | EPOLLRDNORM;
 	if (!mce_apei_read_done && apei_check_mce())
 		return EPOLLIN | EPOLLRDNORM;
@@ -261,13 +257,13 @@ static long mce_chrdev_ioctl(struct file *f, unsigned int cmd,
 	case MCE_GET_RECORD_LEN:
 		return put_user(sizeof(struct mce), p);
 	case MCE_GET_LOG_LEN:
-		return put_user(MCE_LOG_LEN, p);
+		return put_user(mcelog->len, p);
 	case MCE_GETCLEAR_FLAGS: {
 		unsigned flags;
 
 		do {
-			flags = mcelog.flags;
-		} while (cmpxchg(&mcelog.flags, flags, 0) != flags);
+			flags = mcelog->flags;
+		} while (cmpxchg(&mcelog->flags, flags, 0) != flags);
 
 		return put_user(flags, p);
 	}
@@ -339,8 +335,18 @@ static struct miscdevice mce_chrdev_device = {
 
 static __init int dev_mcelog_init_device(void)
 {
+	int mce_log_len;
 	int err;
 
+	mce_log_len = max(MCE_LOG_MIN_LEN, num_online_cpus());
+	mcelog = kzalloc(sizeof(*mcelog) + mce_log_len * sizeof(struct mce), GFP_KERNEL);
+	if (!mcelog)
+		return -ENOMEM;
+
+	strncpy(mcelog->signature, MCE_LOG_SIGNATURE, sizeof(mcelog->signature));
+	mcelog->len = mce_log_len;
+	mcelog->recordlen = sizeof(struct mce);
+
 	/* register character device /dev/mcelog */
 	err = misc_register(&mce_chrdev_device);
 	if (err) {
@@ -350,6 +356,7 @@ static __init int dev_mcelog_init_device(void)
 		else
 			pr_err("Unable to init device /dev/mcelog (rc: %d)\n", err);
 
+		kfree(mcelog);
 		return err;
 	}
 
-- 
2.21.1

[tip: ras/core] x86/mce/dev-mcelog: Dynamically allocate space for machine check records

[tip-bot2 for Tony Luck]

The following commit has been merged into the ras/core branch of tip:

Commit-ID:     d8ecca4043f2d9d89daab7915eca8c2ec6254d0f
Gitweb:        https://git.kernel.org/tip/d8ecca4043f2d9d89daab7915eca8c2ec6254d0f
Author:        Tony Luck <tony.luck@intel.com>
AuthorDate:    Tue, 18 Feb 2020 10:44:08 -08:00
Committer:     Borislav Petkov <bp@suse.de>
CommitterDate: Tue, 10 Mar 2020 10:25:14 +01:00

x86/mce/dev-mcelog: Dynamically allocate space for machine check records

We have had a hard coded limit of 32 machine check records since the
dawn of time.  But as numbers of cores increase, it is possible for
more than 32 errors to be reported before a user process reads from
/dev/mcelog. In this case the additional errors are lost.

Keep 32 as the minimum. But tune the maximum value up based on the
number of processors.

Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20200218184408.GA23048@agluck-desk2.amr.corp.intel.com
---
 arch/x86/include/asm/mce.h           |  6 +--
 arch/x86/kernel/cpu/mce/dev-mcelog.c | 47 +++++++++++++++------------
 2 files changed, 30 insertions(+), 23 deletions(-)

diff --git a/arch/x86/include/asm/mce.h b/arch/x86/include/asm/mce.h
index 4359b95..9d5b099 100644
--- a/arch/x86/include/asm/mce.h
+++ b/arch/x86/include/asm/mce.h
@@ -102,7 +102,7 @@
 
 #define MCE_OVERFLOW 0		/* bit 0 in flags means overflow */
 
-#define MCE_LOG_LEN 32
+#define MCE_LOG_MIN_LEN 32U
 #define MCE_LOG_SIGNATURE	"MACHINECHECK"
 
 /* AMD Scalable MCA */
@@ -135,11 +135,11 @@
  */
 struct mce_log_buffer {
 	char signature[12]; /* "MACHINECHECK" */
-	unsigned len;	    /* = MCE_LOG_LEN */
+	unsigned len;	    /* = elements in .mce_entry[] */
 	unsigned next;
 	unsigned flags;
 	unsigned recordlen;	/* length of struct mce */
-	struct mce entry[MCE_LOG_LEN];
+	struct mce entry[];
 };
 
 enum mce_notifier_prios {
diff --git a/arch/x86/kernel/cpu/mce/dev-mcelog.c b/arch/x86/kernel/cpu/mce/dev-mcelog.c
index 7c8958d..d089567 100644
--- a/arch/x86/kernel/cpu/mce/dev-mcelog.c
+++ b/arch/x86/kernel/cpu/mce/dev-mcelog.c
@@ -29,11 +29,7 @@ static char *mce_helper_argv[2] = { mce_helper, NULL };
  * separate MCEs from kernel messages to avoid bogus bug reports.
  */
 
-static struct mce_log_buffer mcelog = {
-	.signature	= MCE_LOG_SIGNATURE,
-	.len		= MCE_LOG_LEN,
-	.recordlen	= sizeof(struct mce),
-};
+static struct mce_log_buffer *mcelog;
 
 static DECLARE_WAIT_QUEUE_HEAD(mce_chrdev_wait);
 
@@ -45,21 +41,21 @@ static int dev_mce_log(struct notifier_block *nb, unsigned long val,
 
 	mutex_lock(&mce_chrdev_read_mutex);
 
-	entry = mcelog.next;
+	entry = mcelog->next;
 
 	/*
 	 * When the buffer fills up discard new entries. Assume that the
 	 * earlier errors are the more interesting ones:
 	 */
-	if (entry >= MCE_LOG_LEN) {
-		set_bit(MCE_OVERFLOW, (unsigned long *)&mcelog.flags);
+	if (entry >= mcelog->len) {
+		set_bit(MCE_OVERFLOW, (unsigned long *)&mcelog->flags);
 		goto unlock;
 	}
 
-	mcelog.next = entry + 1;
+	mcelog->next = entry + 1;
 
-	memcpy(mcelog.entry + entry, mce, sizeof(struct mce));
-	mcelog.entry[entry].finished = 1;
+	memcpy(mcelog->entry + entry, mce, sizeof(struct mce));
+	mcelog->entry[entry].finished = 1;
 
 	/* wake processes polling /dev/mcelog */
 	wake_up_interruptible(&mce_chrdev_wait);
@@ -214,21 +210,21 @@ static ssize_t mce_chrdev_read(struct file *filp, char __user *ubuf,
 
 	/* Only supports full reads right now */
 	err = -EINVAL;
-	if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce))
+	if (*off != 0 || usize < mcelog->len * sizeof(struct mce))
 		goto out;
 
-	next = mcelog.next;
+	next = mcelog->next;
 	err = 0;
 
 	for (i = 0; i < next; i++) {
-		struct mce *m = &mcelog.entry[i];
+		struct mce *m = &mcelog->entry[i];
 
 		err |= copy_to_user(buf, m, sizeof(*m));
 		buf += sizeof(*m);
 	}
 
-	memset(mcelog.entry, 0, next * sizeof(struct mce));
-	mcelog.next = 0;
+	memset(mcelog->entry, 0, next * sizeof(struct mce));
+	mcelog->next = 0;
 
 	if (err)
 		err = -EFAULT;
@@ -242,7 +238,7 @@ out:
 static __poll_t mce_chrdev_poll(struct file *file, poll_table *wait)
 {
 	poll_wait(file, &mce_chrdev_wait, wait);
-	if (READ_ONCE(mcelog.next))
+	if (READ_ONCE(mcelog->next))
 		return EPOLLIN | EPOLLRDNORM;
 	if (!mce_apei_read_done && apei_check_mce())
 		return EPOLLIN | EPOLLRDNORM;
@@ -261,13 +257,13 @@ static long mce_chrdev_ioctl(struct file *f, unsigned int cmd,
 	case MCE_GET_RECORD_LEN:
 		return put_user(sizeof(struct mce), p);
 	case MCE_GET_LOG_LEN:
-		return put_user(MCE_LOG_LEN, p);
+		return put_user(mcelog->len, p);
 	case MCE_GETCLEAR_FLAGS: {
 		unsigned flags;
 
 		do {
-			flags = mcelog.flags;
-		} while (cmpxchg(&mcelog.flags, flags, 0) != flags);
+			flags = mcelog->flags;
+		} while (cmpxchg(&mcelog->flags, flags, 0) != flags);
 
 		return put_user(flags, p);
 	}
@@ -339,8 +335,18 @@ static struct miscdevice mce_chrdev_device = {
 
 static __init int dev_mcelog_init_device(void)
 {
+	int mce_log_len;
 	int err;
 
+	mce_log_len = max(MCE_LOG_MIN_LEN, num_online_cpus());
+	mcelog = kzalloc(sizeof(*mcelog) + mce_log_len * sizeof(struct mce), GFP_KERNEL);
+	if (!mcelog)
+		return -ENOMEM;
+
+	strncpy(mcelog->signature, MCE_LOG_SIGNATURE, sizeof(mcelog->signature));
+	mcelog->len = mce_log_len;
+	mcelog->recordlen = sizeof(struct mce);
+
 	/* register character device /dev/mcelog */
 	err = misc_register(&mce_chrdev_device);
 	if (err) {
@@ -350,6 +356,7 @@ static __init int dev_mcelog_init_device(void)
 		else
 			pr_err("Unable to init device /dev/mcelog (rc: %d)\n", err);
 
+		kfree(mcelog);
 		return err;
 	}