patch disabling use of PSE on Atom CPUs with a certain erratum

patch disabling use of PSE on Atom CPUs with a certain erratum

[Jan Beulich]

In commit 7a0fc404ae663776e96db43879a0fa24fec1fa3a you
disable use of PSE, but I wonder how this can take any effect on
64-bit CPUs, as they use large pages during early boot already.
Those pages, when use of PSE gets disabled, will get broken up
into 4k pages in phys_pmd_init(), thus actually increasing the
chances that you hit the erratum.

What am I missing?

Jan

Re: patch disabling use of PSE on Atom CPUs with a certain erratum

[H. Peter Anvin]

On 10/07/2010 02:12 AM, Jan Beulich wrote:
> In commit 7a0fc404ae663776e96db43879a0fa24fec1fa3a you
> disable use of PSE, but I wonder how this can take any effect on
> 64-bit CPUs, as they use large pages during early boot already.
> Those pages, when use of PSE gets disabled, will get broken up
> into 4k pages in phys_pmd_init(), thus actually increasing the
> chances that you hit the erratum.
> 
> What am I missing?
> 

Those Atoms don't have 64-bit support.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: patch disabling use of PSE on Atom CPUs with a certain erratum

[Jan Beulich]

>>> On 07.10.10 at 15:46, "H. Peter Anvin" <hpa@zytor.com> wrote:
> On 10/07/2010 02:12 AM, Jan Beulich wrote:
>> In commit 7a0fc404ae663776e96db43879a0fa24fec1fa3a you
>> disable use of PSE, but I wonder how this can take any effect on
>> 64-bit CPUs, as they use large pages during early boot already.
>> Those pages, when use of PSE gets disabled, will get broken up
>> into 4k pages in phys_pmd_init(), thus actually increasing the
>> chances that you hit the erratum.
>> 
>> What am I missing?
>> 
> 
> Those Atoms don't have 64-bit support.

That's different from what I found looking around on the web.

Jan

Re: patch disabling use of PSE on Atom CPUs with a certain erratum

[H. Peter Anvin]

On 10/07/2010 07:01 AM, Jan Beulich wrote:
>>
>> Those Atoms don't have 64-bit support.
> 
> That's different from what I found looking around on the web.
> 

Do tell...

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: patch disabling use of PSE on Atom CPUs with a certain erratum

[Jan Beulich]

>>> On 07.10.10 at 16:08, "H. Peter Anvin" <hpa@zytor.com> wrote:
> On 10/07/2010 07:01 AM, Jan Beulich wrote:
>>>
>>> Those Atoms don't have 64-bit support.
>> 
>> That's different from what I found looking around on the web.
>> 
> 
> Do tell...

E.g.

http://en.wikipedia.org/wiki/List_of_Intel_Atom_microprocessors

in particular the Atom 230 and Atom 330 (which to my reading match
the spec updates for the 200 and 300 series, which in turn exhibit
the erratum in question).

Jan

Re: patch disabling use of PSE on Atom CPUs with a certain erratum

[H. Peter Anvin]

On 10/07/2010 07:52 AM, Jan Beulich wrote:
>>>> On 07.10.10 at 16:08, "H. Peter Anvin" <hpa@zytor.com> wrote:
>> On 10/07/2010 07:01 AM, Jan Beulich wrote:
>>>>
>>>> Those Atoms don't have 64-bit support.
>>>
>>> That's different from what I found looking around on the web.
>>>
>>
>> Do tell...
> 
> E.g.
> 
> http://en.wikipedia.org/wiki/List_of_Intel_Atom_microprocessors
> 
> in particular the Atom 230 and Atom 330 (which to my reading match
> the spec updates for the 200 and 300 series, which in turn exhibit
> the erratum in question).
> 

You're right... I misremembered.  The question is what to do at this
point, since we're already in trouble.  A possible failure on startup
seems better than a state in which we could get data corruption at
almost any time.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: patch disabling use of PSE on Atom CPUs with a certain erratum

[Jan Beulich]

>>> On 07.10.10 at 17:10, "H. Peter Anvin" <hpa@zytor.com> wrote:
> On 10/07/2010 07:52 AM, Jan Beulich wrote:
>>>>> On 07.10.10 at 16:08, "H. Peter Anvin" <hpa@zytor.com> wrote:
>>> On 10/07/2010 07:01 AM, Jan Beulich wrote:
>>>>>
>>>>> Those Atoms don't have 64-bit support.
>>>>
>>>> That's different from what I found looking around on the web.
>>>>
>>>
>>> Do tell...
>> 
>> E.g.
>> 
>> http://en.wikipedia.org/wiki/List_of_Intel_Atom_microprocessors 
>> 
>> in particular the Atom 230 and Atom 330 (which to my reading match
>> the spec updates for the 200 and 300 series, which in turn exhibit
>> the erratum in question).
>> 
> 
> You're right... I misremembered.  The question is what to do at this
> point, since we're already in trouble.  A possible failure on startup
> seems better than a state in which we could get data corruption at
> almost any time.

The only possible (afaict) way to address this would be to check for
the erratum in the boot code, and establish 4k page mappings from
the beginning. That's likely going to be ugly though.

The main reason I was asking was not so much to trigger a code
change, but to understand the implications (since both the comment
in the code and the changeset comment don't really hint at this
leaving a problem open for 64-bit), to some degree to understand
whether e.g. Xen would also need such a workaround (if anyone
cares to run Xen on Atoms).

Jan

Re: patch disabling use of PSE on Atom CPUs with a certain erratum

[H. Peter Anvin]

On 10/07/2010 08:29 AM, Jan Beulich wrote:
>>
>> You're right... I misremembered.  The question is what to do at this
>> point, since we're already in trouble.  A possible failure on startup
>> seems better than a state in which we could get data corruption at
>> almost any time.
> 
> The only possible (afaict) way to address this would be to check for
> the erratum in the boot code, and establish 4k page mappings from
> the beginning. That's likely going to be ugly though.
> 

Well, I guess we could do on 64 bits what we do on 32 bits, and always
use 4K pages for the initial bootstrap, to then be coalesced if PSE is
supported.  Again, ugly.

> The main reason I was asking was not so much to trigger a code
> change, but to understand the implications (since both the comment
> in the code and the changeset comment don't really hint at this
> leaving a problem open for 64-bit), to some degree to understand
> whether e.g. Xen would also need such a workaround (if anyone
> cares to run Xen on Atoms).

Well, keep in mind that the right thing really is to get the microcode
update into the CPU early... i.e. in the BIOS, or in a pinch, from the
bootloader -- but before the kernel runs.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

RE: patch disabling use of PSE on Atom CPUs with a certain erratum

[Yuhong Bao]

> Well, keep in mind that the right thing really is to get the microcode
> update into the CPU early... i.e. in the BIOS, or in a pinch, from the
> bootloader -- but before the kernel runs.

So which vendors have released BIOS updates?
Good thing 64-bit was limited to the desktop Atoms in this generation.

Yuhong Bao