Skip to contentSkip to conversation

use MTRR for write combining if PAT is not available

RFC PATCH Thomas Schlichter 3 participants 6 messages Oct 19, 2009 – Oct 21, 2009

lore
Share & tools
lore
#0 lore raw
Quoted 7 lines · Suresh Siddha

If we are going to make ioremap() and set_memory_wc() add mtrr's in non-pat case, then we need to delete the added mtrr(s) in the corresponding iounmap() and set_memory_wb() aswell.

hmm, this is becoming too complex. The way i915 and other graphics drivers are using set_memory_wc(), it is def a bad idea to start adding mtrr's behind the back for non-pat case.

Yes, maybe it's better to drop it for ioremap() and set_memory_wc(). But I'd like to keep it for mmapping the PCI region. It should help all the people with PAT-incapable CPUs and graphics chips without DRM support (for them there simply is no driver that should set up the MTRR entries...).

Quoted 3 lines

Can't we just force PAT option always and we probably don't care about ioremap_wc() on processors were PAT doesn't get enabled because of known errata.

I don't think this is a good idea, Robert Hancock wrote there may be millions of such Laptops (Core Solo/Duo erratum AE7, Pentium M erratum Y31) :
http://marc.info/?l=linux-kernel&m=125537136105246&w=2

Quoted 2 lines

or Perhaps just try to add mtrr only for the pci mmap case like the 4th patch in this series..

I'd prefer this! ;-)

Kind regards,
Thomas

#1 lore raw
Quoted 10 lines · Thomas Schlichter

I don't think this is a good idea, Robert Hancock wrote there may be millions of such Laptops (Core Solo/Duo erratum AE7, Pentium M erratum Y31) :

http://marc.info/?l=linux-kernel&m=125537136105246&w=2

Quoted 2 lines

or Perhaps just try to add mtrr only for the pci mmap case like the 4th patch in this series..

I'd prefer this! ;-)

Hm, we could perhaps do that - but i think we should only do that on systems that have PAT disabled.

Which brings up the question of how to properly QA such a narrow segment of the market. Maybe disabling CONFIG_X86_PAT should enable that logic too.

Ingo

#2 lore raw
Quoted 11 lines · Ingo Molnar
Quoted 9 lines · Thomas Schlichter

I don't think this is a good idea, Robert Hancock wrote there may be millions of such Laptops (Core Solo/Duo erratum AE7, Pentium M erratum Y31) :

http://marc.info/?l=linux-kernel&m=125537136105246&w=2

Quoted 2 lines

or Perhaps just try to add mtrr only for the pci mmap case like the 4th patch in this series..

Quoted 8 lines
Quoted 1 line

I'd prefer this! ;-)

Hm, we could perhaps do that - but i think we should only do that on systems that have PAT disabled.

Which brings up the question of how to properly QA such a narrow segment of the market. Maybe disabling CONFIG_X86_PAT should enable that logic too.

Also, I think we can simplify and avoid mtrr_add_unaligned() and worry only about the pci mmap aligned cases etc. Those are the only interesting ones.

thanks,
suresh

#3 lore raw
In reply to Ingo Molnar
Quoted 8 lines · Ingo Molnar
Quoted 4 lines · Thomas Schlichter
Quoted 2 lines

or Perhaps just try to add mtrr only for the pci mmap case like the 4th patch in this series..

I'd prefer this! ;-)

Hm, we could perhaps do that - but i think we should only do that on systems that have PAT disabled.

The patch I sent should do exactly that.

Quoted 3 lines

Which brings up the question of how to properly QA such a narrow segment of the market. Maybe disabling CONFIG_X86_PAT should enable that logic too.

When CONFIG_X86_PAT is disabled, pat_enabled is 0, and thus my new code should be active. Or do I miss something?

What do you think about the latest version of my patch series I just sent?

Kind regards,
Thomas

#4 lore raw
Quoted 1 line · Thomas Schlichter

What do you think about the latest version of my patch series I just sent?

I think we can simplify this by just using mtrr_add_page() and avoid all the complexity that comes with mtrr_add_unaligned().

pci_mmap_range() should call one mtrr_add_page() if the base and size are nicely aligned. Almost all the cases, this is the usage sequence here anyway.

thanks,
suresh

#5 lore raw
Quoted 9 lines · Suresh Siddha
Quoted 1 line · Thomas Schlichter

What do you think about the latest version of my patch series I just sent?

I think we can simplify this by just using mtrr_add_page() and avoid all the complexity that comes with mtrr_add_unaligned().

pci_mmap_range() should call one mtrr_add_page() if the base and size are nicely aligned. Almost all the cases, this is the usage sequence here anyway.

yep, keeping it simple is a must.

Ingo