Skip to contentSkip to conversation

enable dual rng on VIA C7

PATCH Udo van den Heuvel 3 participants 10 messages Nov 11, 2007 – Jan 25, 2008 Hardware random number generator core

lore
Share & tools
lore
#0 lore raw

Hello,

Any reason why the second rng on the VIA C7 CPU is not enabled?

Kind regards,
Udo

drivers/char/hw_random/via-rng.c +2 -0
@@ -41,6 +41,7 @@
 	VIA_STRFILT_ENABLE	= (1 << 14),
 	VIA_RAWBITS_ENABLE	= (1 << 13),
 	VIA_RNG_ENABLE		= (1 << 6),
+	VIA_RNG_DUAL            = (1 << 9),
 	VIA_XSTORE_CNT_MASK	= 0x0F,
 
 	VIA_RNG_CHUNK_8		= 0x00,	/* 64 rand bits, 64 stored bits */
@@ -128,6 +129,7 @@
 	lo &= ~(0x7f << VIA_STRFILT_CNT_SHIFT);
 	lo &= ~VIA_XSTORE_CNT_MASK;
 	lo &= ~(VIA_STRFILT_ENABLE | VIA_STRFILT_FAIL | VIA_RAWBITS_ENABLE);
+	lo |= VIA_RNG_DUAL;
 	lo |= VIA_RNG_ENABLE;
 
 	if (lo != old_lo)
#1 lore raw
Quoted 19 lines · On Sun, 11 Nov 2007 19:49:08 +0100 Udo van den Heuvel <[email protected]> wrote:

Any reason why the second rng on the VIA C7 CPU is not enabled?

Kind regards,
Udo

[via-rng.patch text/plain (634B)]

drivers/char/hw_random/via-rng.c +1 -0
@@ -41,6 +41,7 @@
 	VIA_STRFILT_ENABLE	= (1 << 14),
 	VIA_RAWBITS_ENABLE	= (1 << 13),
 	VIA_RNG_ENABLE		= (1 << 6),
+	VIA_RNG_DUAL            = (1 << 9),
 	VIA_XSTORE_CNT_MASK	= 0x0F,
 
 	VIA_RNG_CHUNK_8		= 0x00,	/* 64 rand bits, 64 stored bits */

drivers/char/hw_random/via-rng.c line 135

Quoted 7 lines
drivers/char/hw_random/via-rng.c +1 -0
@@ -128,6 +129,7 @@
 	lo &= ~(0x7f << VIA_STRFILT_CNT_SHIFT);
 	lo &= ~VIA_XSTORE_CNT_MASK;
 	lo &= ~(VIA_STRFILT_ENABLE | VIA_STRFILT_FAIL | VIA_RAWBITS_ENABLE);
+	lo |= VIA_RNG_DUAL;
 	lo |= VIA_RNG_ENABLE;
 
 	if (lo != old_lo)

Does the patch work?

It's missing a signed-off-by:, btw

#2 lore raw
Quoted 3 lines · Andrew Morton
Quoted 1 line · On Sun, 11 Nov 2007 19:49:08 +0100 Udo van den Heuvel <[email protected]> wrote:

Any reason why the second rng on the VIA C7 CPU is not enabled?

(...)

Quoted 1 line

Does the patch work?

Yes, at least:

dd if=/dev/hwrng of=/dev/null bs=1024 count=1024
shows a higher speed than without the patch.

Quoted 1 line

It's missing a signed-off-by:, btw

I did not know we are already that far ;-)
I mean: can this patch be aplied without hurting C3/C7 CPU's with just one RNG? Maybe an expert needs to test/answer?
Maybe some logic needs to be applied around the extra bit?

Kind regards,
Udo

#3 lore raw
Quoted 6 lines · Udo van den Heuvel

I did not know we are already that far ;-)
I mean: can this patch be aplied without hurting C3/C7 CPU's with just one RNG? Maybe an expert needs to test/answer?
Maybe some logic needs to be applied around the extra bit?

From the padlock spec..

"SRC Bits[9:8] Noise source select (I): These bits control the two noise
sources on the processor that input bits to the accumulation buffers. On Nehemiah processors prior to stepping 8, these bits are reserved and undefined. The default RESET state is both bits = 0."

Something like this perhaps ?

drivers/char/hw_random/via-rng.c +14 -0
index ec435cb..95aac0f 100644
@@ -41,6 +41,8 @@ enum {
 	VIA_STRFILT_ENABLE	= (1 << 14),
 	VIA_RAWBITS_ENABLE	= (1 << 13),
 	VIA_RNG_ENABLE		= (1 << 6),
+	VIA_NOISESRC1		= (1 << 8),
+	VIA_NOISESRC2		= (1 << 9),
 	VIA_XSTORE_CNT_MASK	= 0x0F,
 
 	VIA_RNG_CHUNK_8		= 0x00,	/* 64 rand bits, 64 stored bits */
@@ -114,6 +116,7 @@ static int via_rng_data_read(struct hwrng *rng, u32 *data)
 
 static int via_rng_init(struct hwrng *rng)
 {
+	struct cpuinfo_x86 *c = &cpu_data(0);
 	u32 lo, hi, old_lo;
 
 	/* Control the RNG via MSR.  Tread lightly and pay very close
@@ -129,6 +132,17 @@ static int via_rng_init(struct hwrng *rng)
 	lo &= ~VIA_XSTORE_CNT_MASK;
 	lo &= ~(VIA_STRFILT_ENABLE | VIA_STRFILT_FAIL | VIA_RAWBITS_ENABLE);
 	lo |= VIA_RNG_ENABLE;
+	lo |= VIA_NOISESRC1;
+
+	/* Enable secondary noise source on CPUs where it is present. */
+
+	/* Nehemiah stepping 8 and higher */
+	if ((c->x86_model == 9) && (c->x86_mask > 7))
+		lo |= VIA_NOISESRC2;
+
+	/* Esther */
+	if (c->x86_model >= 10)
+		lo |= VIA_NOISESRC2;
 
 	if (lo != old_lo)
 		wrmsr(MSR_VIA_RNG, lo, hi);
Signature · 1 line
http://www.codemonkey.org.uk
#4 lore raw
Quoted 7 lines · Dave Jones
Quoted 4 lines · Udo van den Heuvel

I did not know we are already that far ;-)
I mean: can this patch be aplied without hurting C3/C7 CPU's with just one RNG? Maybe an expert needs to test/answer?
Maybe some logic needs to be applied around the extra bit?

Quoted 8 lines
Quoted 1 line

From the padlock spec..

"SRC Bits[9:8] Noise source select (I): These bits control the two noise
sources on the processor that input bits to the accumulation buffers. On Nehemiah processors prior to stepping 8, these bits are reserved and undefined. The default RESET state is both bits = 0."

Something like this perhaps ?

Yes, I think that's a big step in the right direction!

But I am no expert and cannot really judge how necessary or correct the implementation is w.r.t. the 'undefined' function bits for CPU's that lack a certain feature.

#5 lore raw
Quoted 15 lines · Udo van den Heuvel
Quoted 14 lines · Dave Jones
Quoted 6 lines · Udo van den Heuvel

I did not know we are already that far ;-)
I mean: can this patch be aplied without hurting C3/C7 CPU's with just one RNG? Maybe an expert needs to test/answer?
Maybe some logic needs to be applied around the extra bit?

From the padlock spec..

"SRC Bits[9:8] Noise source select (I): These bits control the two noise
sources on the processor that input bits to the accumulation buffers. On Nehemiah processors prior to stepping 8, these bits are reserved and undefined. The default RESET state is both bits = 0."

Quoted 7 lines
Quoted 1 line

Something like this perhaps ?

Yes, I think that's a big step in the right direction!

But I am no expert and cannot really judge how necessary or correct the implementation is w.r.t. the 'undefined' function bits for CPU's that lack a certain feature.

The checks at the end of the patch for the x86_mask/model ensure we only enable the 2nd noise source on CPUs documented to have it, so we should be safe.

Andrew, want to throw that in the -mm pile for a while?

Dave

Signature · 1 line
http://www.codemonkey.org.uk
#6 lore raw
Quoted 8 lines · Dave Jones
Quoted 7 lines
Quoted 1 line

Something like this perhaps ?

Yes, I think that's a big step in the right direction!

But I am no expert and cannot really judge how necessary or correct the implementation is w.r.t. the 'undefined' function bits for CPU's that lack a certain feature.

Quoted 5 lines

The checks at the end of the patch for the x86_mask/model ensure we only enable the 2nd noise source on CPUs documented to have it, so we should be safe.

Andrew, want to throw that in the -mm pile for a while?

Thanks for assuring we are 'safe'.
Sounds OK to me.
Thanks for picking up this tiny improvement.

Any ideas on the power consumption increase question I received w.r.t. this patch? (or why VIA would have made the 2nd RNG switchable)

#7 lore raw
In reply to Dave Jones
Quoted 24 lines · On Tue, 27 Nov 2007 13:50:53 -0500 Dave Jones <[email protected]> wrote:
Quoted 15 lines · Udo van den Heuvel
Quoted 14 lines · Dave Jones
Quoted 6 lines · Udo van den Heuvel

I did not know we are already that far ;-)
I mean: can this patch be aplied without hurting C3/C7 CPU's with just one RNG? Maybe an expert needs to test/answer?
Maybe some logic needs to be applied around the extra bit?

From the padlock spec..

"SRC Bits[9:8] Noise source select (I): These bits control the two noise
sources on the processor that input bits to the accumulation buffers. On Nehemiah processors prior to stepping 8, these bits are reserved and undefined. The default RESET state is both bits = 0."

Quoted 7 lines
Quoted 1 line

Something like this perhaps ?

Yes, I think that's a big step in the right direction!

But I am no expert and cannot really judge how necessary or correct the implementation is w.r.t. the 'undefined' function bits for CPU's that lack a certain feature.

Quoted 6 lines

The checks at the end of the patch for the x86_mask/model ensure we only enable the 2nd noise source on CPUs documented to have it, so we should be safe.

Andrew, want to throw that in the -mm pile for a while?

Did that, renamed to "via-rng: enable secondary noise source on CPUs where it is present".

Has anyone tested it?

#8 lore raw
Quoted 6 lines · Andrew Morton
Quoted 1 line

Andrew, want to throw that in the -mm pile for a while?

Did that, renamed to "via-rng: enable secondary noise source on CPUs where it is present".

Has anyone tested it?

I hope so; I did not yet get that far since the patche does not compile on 2.6.23.8. I was assured by Dave that the patch should work on 2.6.24 (and mm-tree).

#9 lore raw
In reply to Andrew Morton
Quoted 4 lines · Andrew Morton
Quoted 1 line · On Tue, 27 Nov 2007 13:50:53 -0500 Dave Jones <[email protected]> wrote:

On Tue, Nov 27, 2007 at 05:08:26PM +0100, Udo van den Heuvel wrote:

(...)

Quoted 1 line
Quoted 1 line
Quoted 1 line

Yes, I think that's a big step in the right direction!

(...)

Quoted 7 lines
Quoted 2 lines

Andrew, want to throw that in the -mm pile for a while?

Did that, renamed to "via-rng: enable secondary noise source on CPUs where it is present".

Has anyone tested it?

Works for me.
Did anyone else have a look?