Skip to contentSkip to conversation

transmeta cpu code question

DISCUSSION Nico Schottelius 8 participants 14 messages Nov 20, 2003 – Nov 24, 2003

lore
Share & tools
lore
#0 lore raw

Hello!

What does this do:

printk(KERN_INFO "CPU: Processor revision %u.%u.%u.%u,
%u MHz\n",

                       (cpu_rev >> 24) & 0xff,
                       (cpu_rev >> 16) & 0xff,
                       (cpu_rev >> 8) & 0xff,
                       cpu_rev & 0xff,
                       cpu_freq);

(from arch/i386/kernel/cpu/transmeta.c)

Does not & 0xff make no sense? 0 & 1 makes 0, 1 & 1 makes 1, no changes.

And I don't understand why we do this for 8bit and shifting the cpu_rev...

Can someone enlighten me (with CC' as I am not subscribed) ?

Nico

#1 lore raw
Quoted 12 lines · Nico Schottelius

Hello!

What does this do:

printk(KERN_INFO "CPU: Processor revision %u.%u.%u.%u,
%u MHz\n",

                       (cpu_rev >> 24) & 0xff,
                       (cpu_rev >> 16) & 0xff,
                       (cpu_rev >> 8) & 0xff,
                       cpu_rev & 0xff,
                       cpu_freq);
Quoted 4 lines

(from arch/i386/kernel/cpu/transmeta.c)

Does not & 0xff make no sense? 0 & 1 makes 0, 1 & 1 makes 1, no changes.

I may be wrong, but afaik the 0xff's are there to truncate the other values to 8-bit. 0xff == 0b11111111, which is 8-bit.

Quoted 1 line

From the bitshifting above, cpu_rev seems to be a 32-bit value, so the four

sections of cpu_rev (>>24, >>16, >>8 and >>0) are being ANDed with 0xff to show only the 8 low bits (post-shift).

Quoted 12 lines

And I don't understand why we do this for 8bit and shifting the cpu_rev...

Can someone enlighten me (with CC' as I am not subscribed) ?

Nico
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Signature · 1 line
Ben
#2 lore raw

Thank you all guys for your answers, it looks like it was more obvious I would like to admit.

So to recapitulate it, the and with the 0xff 8-Bit value with an 32Bit Value let's us cut out values, to cut within a value 'with a pair of scissors' like this:

0x AA BB CC DD

   |  |  |  |  
   | 0xBB| 0xDD

0xAA 0xCC

Have a nice day, I will read on the small transmeta code...

I am still interested whether there are possibilities to
a) use the crusoe without the morphing software (to use its native command set)
b) to fine tune Linux to my specific processor, to make it use all available feautures the processor has.

Nico

#3 lore raw
Followup to:  <[email protected]>
By author:    Nico Schottelius <[email protected]>
In newsgroup: linux.dev.kernel
Quoted 4 lines

I am still interested whether there are possibilities to
a) use the crusoe without the morphing software (to use its native command set)

No. You really don't want to -- the native instruction set doesn't look like anything Linux likes to see, and worse, there are binary incompatibilities not just from processor to processor but sometimes from silicon revision to silicon revision.

It's also not faster in any meaningful way, since the dynamic translator does optimistic optimization.

Quoted 2 lines

b) to fine tune Linux to my specific processor, to make it use all available feautures the processor has.

The biggest problem we've found is that a lot of distributions will install an i586 libc even though we have (and export) all the necessary features. Crusoe reports family = 5 mainly to work around a bug in Some Other Operating System[TM], but supports all userspace-visible i686 features gcc expects.

-hpa

Signature · 4 lines
<[email protected]> at work, <[email protected]> in private!
If you send me mail in HTML format I will assume it's spam.
"Unix gives you enough rope to shoot yourself in the foot."
Architectures needed: ia64 m68k mips64 ppc ppc64 s390 s390x sh v850 x86-64
#4 lore raw
Quoted 2 lines · H. Peter Anvin

It's also not faster in any meaningful way, since the dynamic translator does optimistic optimization.

Statically compiled code for the Crusoe chips may not be faster. (Arguably statically compiled code for _any_ CPU is not the best strategy for fast code).

But if someone is able to write better code morphing software than Transmeta, that would be faster :)

-- Jamie

#5 lore raw

Quote from Jamie Lokier <[email protected]>:

Quoted 10 lines
Quoted 2 lines · H. Peter Anvin

It's also not faster in any meaningful way, since the dynamic translator does optimistic optimization.

Statically compiled code for the Crusoe chips may not be faster. (Arguably statically compiled code for _any_ CPU is not the best strategy for fast code).

But if someone is able to write better code morphing software than Transmeta, that would be faster :)

The idea of a CMS which only implemented the subset of X86 instructions that gcc actually uses was discussed on the list a few months ago, but unsuprisingly it never progressed beyond the thought experiment stage.

John.

#6 lore raw
Quoted 7 lines · John Bradford
Quoted 7 lines
Quoted 2 lines

It's also not faster in any meaningful way, since the dynamic translator does optimistic optimization.

Statically compiled code for the Crusoe chips may not be faster. (Arguably statically compiled code for _any_ CPU is not the best strategy for fast code).

Quoted 7 lines
Quoted 2 lines

But if someone is able to write better code morphing software than Transmeta, that would be faster :)

The idea of a CMS which only implemented the subset of X86 instructions that gcc actually uses was discussed on the list a few months ago, but unsuprisingly it never progressed beyond the thought experiment stage.

What would be the point in that? Surely the CMS overhead for decoding the rarely used x86 instructions is negligable, precisely because of their rarity?

-- Jamie

#7 lore raw
Followup to:  <[email protected]>
By author:    Jamie Lokier <[email protected]>
In newsgroup: linux.dev.kernel
Quoted 5 lines

What would be the point in that? Surely the CMS overhead for decoding the rarely used x86 instructions is negligable, precisely because of their rarity?

Pretty much.

The bulk of the time spent in CMS is in the backend scheduler, at which time the original x86 instructions are pretty much unrecognizable.

If there is something that one could imagine doing at the CMS level to help on Linux, it would probably be something like making it optional to actually perform stores beneath the stack pointer, in which case a lot of stack frame operations could be done purely in registers. CMS will do them in registers already, but will be forced to perform a store at the end of the translation anyway in order to keep exact x86 semantics.

-hpa

Signature · 4 lines
<[email protected]> at work, <[email protected]> in private!
If you send me mail in HTML format I will assume it's spam.
"Unix gives you enough rope to shoot yourself in the foot."
Architectures needed: ia64 m68k mips64 ppc ppc64 s390 s390x sh v850 x86-64
#8 lore raw
Quoted 7 lines · H. Peter Anvin

If there is something that one could imagine doing at the CMS level to help on Linux, it would probably be something like making it optional to actually perform stores beneath the stack pointer, in which case a lot of stack frame operations could be done purely in registers. CMS will do them in registers already, but will be forced to perform a store at the end of the translation anyway in order to keep exact x86 semantics.

You couldn't enable that globally, because it'd break userspace programs which write below the stack safely using sigaltstack(), or which even use the stack pointer as a general purpose register (I've coded games which do that in tight rendering loops), or during funky thunking code.

It sounds like a good thing to add as a per-task feature though.

-- Jamie

#9 lore raw
In reply to Nico Schottelius
Quoted 12 lines · Nico Schottelius

Hello!

What does this do:

printk(KERN_INFO "CPU: Processor revision %u.%u.%u.%u,
%u MHz\n",

                       (cpu_rev >> 24) & 0xff,
                       (cpu_rev >> 16) & 0xff,
                       (cpu_rev >> 8) & 0xff,
                       cpu_rev & 0xff,
                       cpu_freq);
Quoted 7 lines

(from arch/i386/kernel/cpu/transmeta.c)

Does not & 0xff make no sense? 0 & 1 makes 0, 1 & 1 makes 1, no changes.

And I don't understand why we do this for 8bit and shifting the cpu_rev...

You are a bit confused. The cpu_rev is a 4 byte value, each byte is a decimal of the revision.

And (0 & 1) makes 1, not 0. That's an AND, not an OR.

Think about it this way. If cpu_rev == 0x01040801, then this would produce:

(0x01040801 >> 24 & 0xff) -> (0x01 & 0xff) -> 0x01

(0x01040801 >> 16 & 0xff) -> (0x0104 & 0xff) -> 0x04

(0x01040801 >> 8 & 0xff) -> (0x010408 & 0xff) -> 0x08

(0x01040801 & 0xff) -> 0x01

Signature · 4 lines
Debian     - http://www.debian.org/
Linux 1394 - http://www.linux1394.org/
Subversion - http://subversion.tigris.org/
WatchGuard - http://www.watchguard.com/
#10 lore raw
Quoted 1 line · Ben Collins

And (0 & 1) makes 1, not 0. That's an AND, not an OR.

Not quite Ben... 0 & _anything_ is still 0.

#11 lore raw
Quoted 5 lines · Kevin P. Fleming
Quoted 1 line · Ben Collins

And (0 & 1) makes 1, not 0. That's an AND, not an OR.

Not quite Ben... 0 & _anything_ is still 0.

Typing too fast. Brain running slow. That's what I get.

Signature · 4 lines
Debian     - http://www.debian.org/
Linux 1394 - http://www.linux1394.org/
Subversion - http://subversion.tigris.org/
WatchGuard - http://www.watchguard.com/
#12 lore raw
In reply to Ben Collins

EDIT: Oops, meant to reply-all. Apologies Ben.

Quoted 4 lines · Ben Collins

You are a bit confused. The cpu_rev is a 4 byte value, each byte is a decimal of the revision.

And (0 & 1) makes 1, not 0. That's an AND, not an OR.

I'm willing to put some safe money on (0 & 1) equalling 0.

It has to, for your explanation below to work. :)

Quoted 11 lines

Think about it this way. If cpu_rev == 0x01040801, then this would produce:

(0x01040801 >> 24 & 0xff) -> (0x01 & 0xff) -> 0x01

(0x01040801 >> 16 & 0xff) -> (0x0104 & 0xff) -> 0x04

(0x01040801 >> 8 & 0xff) -> (0x010408 & 0xff) -> 0x08

(0x01040801 & 0xff) -> 0x01

Signature · 1 line
Ben
#13 lore raw
In reply to Ben Collins

Quoting Ben Collins <[email protected]>:

Quoted 18 lines · Nico Schottelius
Quoted 17 lines

Hello!

What does this do:

printk(KERN_INFO "CPU: Processor revision %u.%u.%u.%u,
%u MHz\n",

                       (cpu_rev >> 24) & 0xff,
                       (cpu_rev >> 16) & 0xff,
                       (cpu_rev >> 8) & 0xff,
                       cpu_rev & 0xff,
                       cpu_freq);

(from arch/i386/kernel/cpu/transmeta.c)

Does not & 0xff make no sense? 0 & 1 makes 0, 1 & 1 makes 1, no changes.

Quoted 7 lines
Quoted 2 lines

And I don't understand why we do this for 8bit and shifting the cpu_rev...

You are a bit confused. The cpu_rev is a 4 byte value, each byte is a decimal of the revision.

And (0 & 1) makes 1, not 0. That's an AND, not an OR.

Last I checked, 0 AND anything is 0. Think of it this way:

A B A & B
------------

T   F     F
T   T     T
F   T     F
F   F     F

So, anything ANDed with F is false.

-sandalle

Signature · 17 lines
PGP Key Fingerprint:  FCFF 26A1 BE21 08F4 BB91  FAED 1D7B 7D74 A8EF DD61
http://search.keyserver.net:11371/pks/lookup?op=get&search=0xA8EFDD61

-----BEGIN GEEK CODE BLOCK-----
Version: 3.12
GCS/E/IT$ d-- s++:+>: a-- C++(+++) BL++++VIS>$ P+(++) L+++ E-(---) W++ N+@ o?
K? w++++>-- O M-@ V-- PS+(+++) PE(-) Y++(+) PGP++(+) t+() 5++ X(+) R+(++)
tv(--)b++(+++) DI+@ D++(+++) G>+++ e>+++ h---(++) r++ y+
------END GEEK CODE BLOCK------

Eric Sandall                     |  Source Mage GNU/Linux Developer
[email protected]                  |  http://www.sourcemage.org/
http://eric.sandall.us/          |  SysAdmin @ Inst. Shock Physics @ WSU
http://counter.li.org/  #196285  |  http://www.shock.wsu.edu/

----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.