Re: Crash with old Windows on new CPU

On Wed, Mar 9, 2016 at 5:48 AM, Christian Ullrich <chris(at)chrullrich(dot)net>
wrote:

> * Peter Eisentraut wrote:
>
> On 2/12/16 11:24 AM, Christian Ullrich wrote:
>>
>
> Otherwise, it may be time to update the manual (15.6 Supported
>>> Platforms) where it says PostgreSQL "can be expected to work on these
>>> operating systems: [...] Windows (Win2000 SP4 and later), [...]".
>>> Perhaps we could add "except Windows before 7 SP1/2008R2 SP1 when
>>> running in x64 mode on Intel CPUs introduced after May 2013 (Haswell and
>>> later)"?
>>>
>>
> Wouldn't the fix be for users to upgrade their service packs?
>>
>
> Windows 2008 and 2008R2 are entirely different things: 2008 is the server
> sibling of Vista (internal version 6.0), 2008R2 is that of Windows 7 (6.1).
> There is no version of 2008 that supports AVX2.
>
>
Windows 2008 went out of mainstream support in January last year, but is on
extended support until 2020. Extended support means both paid support and
security support is still there (what you don't get is new hardware support
and generic non-security related updates). That means we're going to see
these versions around for a long time. (And Vista is in extended support
until 2017).

And exactly the type of upgrade scenario outlined in the OP is only going
to be come more common as old hardware gets replaced. If we didn't patch
this, the reasonable thing would be to say we don't support Visual Studio
2013, rather than a specific version of Windows, I think.

Given the small and localized change of this (and hey it even goes inside a
function labeled hacks), I definitely think it's worth doing.

I've commited this one, including a reference to the MS bug report where
they say they're not going to fix it in existing versions. I didn't
actually test it on mingw, but as it doesn't define MSC_VER it shouldn't be
affected (by the bug or by the patch).

I did notice the #ifdef's are actually different in the header and body
section of the patch, which seems wrong. I used the one from the actual
implementation (_M_AMD64) for the header includes as, and also merged the
#ifdef's together to a single #if in each section. Please verify!

Thanks for a very good analysis and patch, and for good explanations of the
details! :)

--
Magnus Hagander
Me: http://www.hagander.net/
Work: http://www.redpill-linpro.com/