Skip to contentSkip to conversation

random: add a way to get some random bits into the entropy pools early on

PATCH In review Arjan van de Ven 5 participants 6 messages Nov 25, 2008 – Nov 26, 2008

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 random: add a way to get some random bits into the entropy pools early on

Quoted 1 line

From 5b3b09ac82316c2d4000460d586ebe59303c12c2 Mon Sep 17 00:00:00 2001

From: Arjan van de Ven <[email protected]>
Date: Tue, 25 Nov 2008 11:41:37 -0800
Subject: [PATCH] random: add a way to get some random bits into the entropy pools early on

currently the entropy pool gets seeded on the module_init() level, but there is at least one consumer of random bits (the oops ID that is printed as part of the oops).
As a result of this, kerneloops.org is seeing a lot of oopses that all share the same 'random' number; which used to get filed away as "duplicate".

This patch adds a function to the random driver so that various pieces of the kernel can add random bits (but not entropy!) to the pool, to avoid this dupicate ID problem.

---
 arch/x86/kernel/apic.c |    7 +++++++
 drivers/char/random.c  |   29 +++++++++++++++++++++++++++++
 include/linux/random.h |    3 +++
 kernel/panic.c         |    1 +
 4 files changed, 40 insertions(+), 0 deletions(-)
arch/x86/kernel/apic.c +7 -0
index 04a7f96..11f7c7d 100644
@@ -30,6 +30,7 @@
 #include <linux/module.h>
 #include <linux/dmi.h>
 #include <linux/dmar.h>
+#include <linux/random.h>
 
 #include <asm/atomic.h>
 #include <asm/smp.h>
@@ -691,6 +692,12 @@ static int __init calibrate_APIC_clock(void)
 	} else
 		local_irq_enable();
 
+	/*
+	 * by now we have enough time spent that there is some value in
+	 * pushing the timestamp to the random pools for early kernel use.
+	 */
+	seed_random_pools();
+
 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
 		printk(KERN_WARNING
 		       "APIC timer disabled due to verification failure.\n");
drivers/char/random.c +29 -0
index 675076f..236d285 100644
@@ -929,6 +929,35 @@ static int rand_initialize(void)
 }
 module_init(rand_initialize);
 
+/*
+ * seed_std_data - seed pool with system timing data
+ *
+ * @r: pool to initialize
+ *
+ * This function mixes some system data into the pool to prepare it for use.
+ * This function can be called really early during boot to at least get some
+ * randomness from the system if random numbers are needed early, for example
+ * as part of the early oops printing.
+ *
+ * No entropy is credited for this seeding.
+ */
+static void seed_std_data(struct entropy_store *r)
+{
+	ktime_t now;
+	unsigned long flags;
+
+	now = ktime_get_real();
+	mix_pool_bytes(r, &now, sizeof(now));
+}
+
+int seed_random_pools(void)
+{
+	seed_std_data(&input_pool);
+	seed_std_data(&blocking_pool);
+	seed_std_data(&nonblocking_pool);
+	return 0;
+}
+
 void rand_initialize_irq(int irq)
 {
 	struct timer_rand_state *state;
include/linux/random.h +3 -0
index 36f125c..d23d284 100644
@@ -72,6 +72,9 @@ unsigned long randomize_range(unsigned long start, unsigned long end, unsigned l
 u32 random32(void);
 void srandom32(u32 seed);
 
+int seed_random_pools(void);
+
+
 #endif /* __KERNEL___ */
 
 #endif /* _LINUX_RANDOM_H */
kernel/panic.c +1 -0
index 6513aac..d71b167 100644
@@ -295,6 +295,7 @@ static u64 oops_id;
 
 static int init_oops_id(void)
 {
+	seed_random_pools();
 	if (!oops_id)
 		get_random_bytes(&oops_id, sizeof(oops_id));
 
Signature · 8 lines
1.5.5.1



-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org
#1 lore raw
Quoted 3 lines · On Tue, 25 Nov 2008 13:26:00 -0800 Arjan van de Ven <[email protected]> wrote:

currently the entropy pool gets seeded on the module_init() level, but there is at least one consumer of random bits (the oops ID that is printed as part of the oops).

There seems to be some missing text here.

Quoted 6 lines

As a result of this, kerneloops.org is seeing a lot of oopses that all share the same 'random' number; which used to get filed away as "duplicate".

This patch adds a function to the random driver so that various pieces of the kernel can add random bits (but not entropy!) to the pool, to avoid this dupicate ID problem.

Acked-by on v1 1/1
#2 lore raw
In reply to Arjan van de Ven
Quoted 15 lines · Arjan van de Ven
Quoted 1 line

From 5b3b09ac82316c2d4000460d586ebe59303c12c2 Mon Sep 17 00:00:00 2001

From: Arjan van de Ven <[email protected]>
Date: Tue, 25 Nov 2008 11:41:37 -0800
Subject: [PATCH] random: add a way to get some random bits into the entropy pools early on

currently the entropy pool gets seeded on the module_init() level, but there is at least one consumer of random bits (the oops ID that is printed as part of the oops).
As a result of this, kerneloops.org is seeing a lot of oopses that all share the same 'random' number; which used to get filed away as "duplicate".

This patch adds a function to the random driver so that various pieces of the kernel can add random bits (but not entropy!) to the pool, to avoid this dupicate ID problem.

Quoted 7 lines
---
 arch/x86/kernel/apic.c |    7 +++++++
 drivers/char/random.c  |   29 +++++++++++++++++++++++++++++
 include/linux/random.h |    3 +++
 kernel/panic.c         |    1 +
 4 files changed, 40 insertions(+), 0 deletions(-)

good catch ...

Ingo

#3 lore raw
In reply to Arjan van de Ven
Quoted 11 lines · Arjan van de Ven
Quoted 1 line

From 5b3b09ac82316c2d4000460d586ebe59303c12c2 Mon Sep 17 00:00:00 2001

From: Arjan van de Ven <[email protected]>
Date: Tue, 25 Nov 2008 11:41:37 -0800
Subject: [PATCH] random: add a way to get some random bits into the entropy pools early on

currently the entropy pool gets seeded on the module_init() level, but there is at least one consumer of random bits (the oops ID that is printed as part of the oops).
As a result of this, kerneloops.org is seeing a lot of oopses that all share the same 'random' number; which used to get filed away as "duplicate".

Quoted 3 lines

This patch adds a function to the random driver so that various pieces of the kernel can add random bits (but not entropy!) to the pool, to avoid this dupicate ID problem.

This appears to be equivalent to making random initialize earlier, except with more code?

If we're going down a route like this, I'd like to see the ability to actually mix in data (ie MAC addresses, DMI data, etc.) rather than just timing.

Signature · 1 line
Mathematics is the supreme nostalgia of our time.
#4 lore raw
Quoted 7 lines · On Tue, 25 Nov 2008 15:41:22 -0600 Matt Mackall <[email protected]> wrote:
Quoted 4 lines

This patch adds a function to the random driver so that various pieces of the kernel can add random bits (but not entropy!) to the pool, to avoid this dupicate ID problem.

This appears to be equivalent to making random initialize earlier, except with more code?

well sort of kinda, but arbitrarily early (eg I made the oops ID just call this whenever it's called first)

Quoted 4 lines

If we're going down a route like this, I'd like to see the ability to actually mix in data (ie MAC addresses, DMI data, etc.) rather than just timing.

I'll make a second patch for that; it does make sense to do this, but it also makes sense to just have a "just do the time" helper

Signature · 3 lines
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org
#5 lore raw
In reply to Arjan van de Ven
Quoted 10 lines · Arjan van de Ven

From 5b3b09ac82316c2d4000460d586ebe59303c12c2 Mon Sep 17 00:00:00 2001
From: Arjan van de Ven <[email protected]>
Date: Tue, 25 Nov 2008 11:41:37 -0800
Subject: [PATCH] random: add a way to get some random bits into the entropy pools early on

currently the entropy pool gets seeded on the module_init() level, but there is at least one consumer of random bits (the oops ID that is printed as part of the oops).
As a result of this, kerneloops.org is seeing a lot of oopses that all share the same 'random' number; which used to get filed away as "duplicate".

This should also help networking, which also relies on some non existant randomness early on currently.

But I think i would prefer to move the complete random initialization earlier instead of having special case calls for this.

Feeding in DMI and some other system data and perhaps some other register values would be also a good idea.
The utsname initialization that is currently in there I always found amusing because there is no way utsname is set that early.

-Andi

Signature · 1 line
[email protected]