Clean up write-combining MTRR addition
Subsystems (12)
v2 cover Clean up write-combining MTRR addition
A fair number of drivers (mostly graphics) add write-combining MTRRs. Most ignore errors and most add the MTRR even on PAT systems which don't need to use MTRRs.
This series adds new functions arch_phys_wc_{add,del} that, on PAT-less x86 systems with MTRRs, add MTRRs and report errors, and that do nothing otherwise. (Other architectures, if any, with a similar mechanism could implement them.)
I've only tested the radeon driver, since I don't have test hardware easily available for the other drivers.
Benefits include:
- Simpler code
- No more complaints about MTRR conflict warnings on PAT systems
- Eventual unexporting of the MTRR API?
This series eliminates about half of the mtrr_add calls in drivers/.
Changes from v1:
- Helpers renamed
- Lots of bugs fixed
The series is also at:
https://git.kernel.org/cgit/linux/kernel/git/luto/linux.git/log/?h=mtrr_cleanup/rfc_v2
Andy Lutomirski (8):
Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed drm (ast,cirrus,mgag200,nouveau,savage,vmwgfx): Remove
drm_mtrr_{add,del} drm: Update drm_addmap and drm_mmap to use PAT WC instead of MTRRs drm,agpgart: Use pgprot_writecombine for AGP maps and make the MTRR
optional
i915: Use arch_phys_wc_{add,del}
radeon: Switch to arch_phys_wc_add and add a missing ..._del
uvesafb: Clean up MTRR code
drm: Remove mtrr_add and mtrr_del fallback hack for non-MTRR systems
Documentation/fb/uvesafb.txt | 16 +++----- arch/x86/include/asm/io.h | 7 ++++ arch/x86/include/asm/mtrr.h | 5 ++- arch/x86/kernel/cpu/mtrr/main.c | 48 +++++++++++++++++++++++ drivers/char/agp/frontend.c | 8 ++-- drivers/gpu/drm/ast/ast_ttm.c | 13 ++----- drivers/gpu/drm/cirrus/cirrus_ttm.c | 15 ++------ drivers/gpu/drm/drm_bufs.c | 17 +++++---- drivers/gpu/drm/drm_pci.c | 8 ++-- drivers/gpu/drm/drm_stub.c | 10 +---- drivers/gpu/drm/drm_vm.c | 22 +++++------ drivers/gpu/drm/i915/i915_dma.c | 44 +++------------------ drivers/gpu/drm/mgag200/mgag200_ttm.c | 14 ++----- drivers/gpu/drm/nouveau/nouveau_ttm.c | 13 ++----- drivers/gpu/drm/radeon/radeon_object.c | 5 ++- drivers/gpu/drm/savage/savage_bci.c | 43 ++++++++------------- drivers/gpu/drm/savage/savage_drv.h | 5 +-- drivers/gpu/drm/vmwgfx/vmwgfx_drv.c | 10 ++--- drivers/video/uvesafb.c | 70 +++++++++------------------------- include/drm/drmP.h | 34 +---------------- include/drm/drm_os_linux.h | 16 -------- include/linux/io.h | 25 ++++++++++++ include/video/uvesafb.h | 1 + 23 files changed, 181 insertions(+), 268 deletions(-)
Signature · 1 line
1.8.1.4
PATCH v2 1/8 Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed
Several drivers currently use mtrr_add through various #ifdef guards and/or drm wrappers. The vast majority of them want to add WC MTRRs on x86 systems and don't actually need the MTRR if PAT (i.e. ioremap_wc, etc) are working.
arch_phys_wc_add and arch_phys_wc_del are new functions, available on all architectures and configurations, that add WC MTRRs on x86 if needed (and handle errors) and do nothing at all otherwise. They're also easier to use than mtrr_add and mtrr_del, so the call sites can be simplified.
As an added benefit, this will avoid wasting MTRRs and possibly warning pointlessly on PAT-supporting systems.
--- arch/x86/include/asm/io.h | 7 ++++++ arch/x86/include/asm/mtrr.h | 5 ++++- arch/x86/kernel/cpu/mtrr/main.c | 48 +++++++++++++++++++++++++++++++++++++++++ include/linux/io.h | 25 +++++++++++++++++++++ 4 files changed, 84 insertions(+), 1 deletion(-)
arch/x86/include/asm/io.h +7 -0
@@ -345,4 +345,11 @@ extern bool xen_biovec_phys_mergeable(const struct bio_vec *vec1,#define IO_SPACE_LIMIT 0xffff +#ifdef CONFIG_MTRR +extern int __must_check arch_phys_wc_add (unsigned long base, +unsigned long size); +extern void arch_phys_wc_del (int handle); +#define arch_phys_wc_add arch_phys_wc_add +#endif +#endif /* _ASM_X86_IO_H */
arch/x86/include/asm/mtrr.h +4 -1
@@ -26,7 +26,10 @@#include <uapi/asm/mtrr.h> -/* The following functions are for use by other drivers */ +/* +* The following functions are for use by other drivers that cannot use +* arch_phys_wc_add and arch_phys_wc_del. +*/ # ifdef CONFIG_MTRR extern u8 mtrr_type_lookup (u64 addr,u64 end);extern void mtrr_save_fixed_ranges (void *);
arch/x86/kernel/cpu/mtrr/main.c +48 -0
60 lines not on this page · raw message
include/linux/io.h +25 -0
@@ -76,4 +76,29 @@ void devm_ioremap_release(struct device *dev, void *res);#define arch_has_dev_port () (1)#endif +/* +* Some systems (x86 without PAT) have a somewhat reliable way to mark a +* physical address range such that uncached mappings will actually +* end up write-combining. This facility should be used in conjunction +* with pgprot_writecombine, ioremap-wc, or set_memory_wc, since it has +* no effect if the per-page mechanisms are functional. +* (On x86 without PAT, these functions manipulate MTRRs.) +* +* arch_phys_del_wc(0) or arch_phys_del_wc(any error code) is guaranteed +* to have no effect. +*/ +#ifndef arch_phys_wc_add +static inline int __must_check arch_phys_wc_add (unsigned long base, +unsigned long size) +{ +return 0 ;/* It worked (i.e. did nothing). */ +} + +static inline void arch_phys_wc_del (int handle) +{ +} + +#define arch_phys_wc_add arch_phys_wc_add +#endif +#endif /* _LINUX_IO_H */
Signature · 1 line
1.8.1.4
Re: [RFC/PATCH v2 1/8] Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed
Quoted 78 lines · Andy Lutomirski
Several drivers currently use
mtrr_addthrough various #ifdef guards and/or drm wrappers. The vast majority of them want to add WC MTRRs on x86 systems and don't actually need the MTRR if PAT (i.e.ioremap_wc, etc) are working.
arch_phys_wc_addandarch_phys_wc_delare new functions, available on all architectures and configurations, that add WC MTRRs on x86 if needed (and handle errors) and do nothing at all otherwise. They're also easier to use thanmtrr_addandmtrr_del, so the call sites can be simplified.As an added benefit, this will avoid wasting MTRRs and possibly warning pointlessly on PAT-supporting systems.
Signed-off-by Andy Lutomirski <[email protected]>--- arch/x86/include/asm/io.h | 7 ++++++ arch/x86/include/asm/mtrr.h | 5 ++++- arch/x86/kernel/cpu/mtrr/main.c | 48 +++++++++++++++++++++++++++++++++++++++++ include/linux/io.h | 25 +++++++++++++++++++++ 4 files changed, 84 insertions(+), 1 deletion(-)arch/x86/include/asm/io.h +7
-0@@ -345,4 +345,11 @@ extern bool xen_biovec_phys_mergeable(const struct bio_vec *vec1,#define IO_SPACE_LIMIT 0xffff +#ifdef CONFIG_MTRR +extern int __must_check arch_phys_wc_add (unsigned long base, +unsigned long size); +extern void arch_phys_wc_del (int handle); +#define arch_phys_wc_add arch_phys_wc_add +#endif +#endif /* _ASM_X86_IO_H */ arch/x86/include/asm/mtrr.h +4
-1@@ -26,7 +26,10 @@#include <uapi/asm/mtrr.h> -/* The following functions are for use by other drivers */ +/* +* The following functions are for use by other drivers that cannot use +* arch_phys_wc_add and arch_phys_wc_del. +*/ # ifdef CONFIG_MTRR extern u8 mtrr_type_lookup (u64 addr,u64 end);extern void mtrr_save_fixed_ranges (void *);arch/x86/kernel/cpu/mtrr/main.c +9
-0@@ -51,6 +51,7 @@#include <asm/e820.h> #include <asm/mtrr.h> #include <asm/msr.h> +#include <asm/pat.h> #include "mtrr.h" @@ -524,6 +525,53 @@ int mtrr_del(int reg, unsigned long base, unsigned long size) }EXPORT_SYMBOL (mtrr_del); +/** +* arch_phys_wc_add - add a WC MTRR and handle errors if PAT is unavailable +* @base: Physical base address +* @size: Size of region +* +* If PAT is available, this does nothing. If PAT is unavailable, it +* attempts to add a WC MTRR covering size bytes starting at base and +* logs an error if this fails.
arch/x86/kernel/cpu/mtrr/main.c int mtrr_del(int reg, unsigned long base, unsigned long size) · line 545
Quoted 10 lines
arch/x86/kernel/cpu/mtrr/main.c +10
-0+* +* Drivers must store the return value to pass to mtrr_del_wc_if_needed, +* but drivers should not try to interpret that return value. +*/ +int arch_phys_wc_add (unsigned long base,unsigned long size) +{ +int ret; + +if (pat_enabled) +return 0 ;/* Success! (We don't need to do anything.) */
Shouldn't we #define a big number for this case since mtrr_add returns 0-based mtrr indices? Rather unlikely that the very first mtrr is unused I know, but still feels like a cleaner interface. And we don't need to leak that #define out at all to users of this interface.
-Daniel
Quoted 69 lines
arch/x86/kernel/cpu/mtrr/main.c +29
-0+ + ret =mtrr_add (base, size,MTRR_TYPE_WRCOMB , true); +if (ret <0 ) { +pr_warn ("Failed to add WC MTRR for [%p-%p]; performance may suffer." , + (void *)base, (void *)(base + size -1 )); +return ret; + } +return ret +1000 ; +} +EXPORT_SYMBOL (arch_phys_wc_add); + +/* +* arch_phys_wc_del - undoes arch_phys_wc_add +* @handle: Return value from arch_phys_wc_add +* +* This cleans up after mtrr_add_wc_if_needed. +* +* The API guarantees that mtrr_del_wc_if_needed(error code) and +* mtrr_del_wc_if_needed(0) do nothing. +*/ +extern void arch_phys_wc_del (int handle) +{ +if (handle >=1 ) { +WARN_ON (handle <1000 ); +mtrr_del (handle -1000 ,0 ,0 ); + } +} +EXPORT_SYMBOL (arch_phys_wc_del); +/* * HACK ALERT! * These should be called implicitly, but we can't yet until all the initcall include/linux/io.h +25
-0@@ -76,4 +76,29 @@ void devm_ioremap_release(struct device *dev, void *res);#define arch_has_dev_port () (1)#endif +/* +* Some systems (x86 without PAT) have a somewhat reliable way to mark a +* physical address range such that uncached mappings will actually +* end up write-combining. This facility should be used in conjunction +* with pgprot_writecombine, ioremap-wc, or set_memory_wc, since it has +* no effect if the per-page mechanisms are functional. +* (On x86 without PAT, these functions manipulate MTRRs.) +* +* arch_phys_del_wc(0) or arch_phys_del_wc(any error code) is guaranteed +* to have no effect. +*/ +#ifndef arch_phys_wc_add +static inline int __must_check arch_phys_wc_add (unsigned long base, +unsigned long size) +{ +return 0 ;/* It worked (i.e. did nothing). */ +} + +static inline void arch_phys_wc_del (int handle) +{ +} + +#define arch_phys_wc_add arch_phys_wc_add +#endif +#endif /* _LINUX_IO_H */ Signature · 1 line
1.8.1.4
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [RFC/PATCH v2 1/8] Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed
Quoted 86 lines · Daniel Vetter
Quoted 85 lines · Andy Lutomirski
Several drivers currently use
mtrr_addthrough various #ifdef guards and/or drm wrappers. The vast majority of them want to add WC MTRRs on x86 systems and don't actually need the MTRR if PAT (i.e.ioremap_wc, etc) are working.
arch_phys_wc_addandarch_phys_wc_delare new functions, available on all architectures and configurations, that add WC MTRRs on x86 if needed (and handle errors) and do nothing at all otherwise. They're also easier to use thanmtrr_addandmtrr_del, so the call sites can be simplified.As an added benefit, this will avoid wasting MTRRs and possibly warning pointlessly on PAT-supporting systems.
Signed-off-by Andy Lutomirski <[email protected]>--- arch/x86/include/asm/io.h | 7 ++++++ arch/x86/include/asm/mtrr.h | 5 ++++- arch/x86/kernel/cpu/mtrr/main.c | 48 +++++++++++++++++++++++++++++++++++++++++ include/linux/io.h | 25 +++++++++++++++++++++ 4 files changed, 84 insertions(+), 1 deletion(-)arch/x86/include/asm/io.h +7
-0@@ -345,4 +345,11 @@ extern bool xen_biovec_phys_mergeable(const struct bio_vec *vec1,#define IO_SPACE_LIMIT 0xffff +#ifdef CONFIG_MTRR +extern int __must_check arch_phys_wc_add (unsigned long base, +unsigned long size); +extern void arch_phys_wc_del (int handle); +#define arch_phys_wc_add arch_phys_wc_add +#endif +#endif /* _ASM_X86_IO_H */ arch/x86/include/asm/mtrr.h +4
-1@@ -26,7 +26,10 @@#include <uapi/asm/mtrr.h> -/* The following functions are for use by other drivers */ +/* +* The following functions are for use by other drivers that cannot use +* arch_phys_wc_add and arch_phys_wc_del. +*/ # ifdef CONFIG_MTRR extern u8 mtrr_type_lookup (u64 addr,u64 end);extern void mtrr_save_fixed_ranges (void *);arch/x86/kernel/cpu/mtrr/main.c +16
-0@@ -51,6 +51,7 @@#include <asm/e820.h> #include <asm/mtrr.h> #include <asm/msr.h> +#include <asm/pat.h> #include "mtrr.h" @@ -524,6 +525,53 @@ int mtrr_del(int reg, unsigned long base, unsigned long size) }EXPORT_SYMBOL (mtrr_del); +/** +* arch_phys_wc_add - add a WC MTRR and handle errors if PAT is unavailable +* @base: Physical base address +* @size: Size of region +* +* If PAT is available, this does nothing. If PAT is unavailable, it +* attempts to add a WC MTRR covering size bytes starting at base and +* logs an error if this fails. +* +* Drivers must store the return value to pass to mtrr_del_wc_if_needed, +* but drivers should not try to interpret that return value. +*/ +int arch_phys_wc_add (unsigned long base,unsigned long size) +{ +int ret;
Quoted 8 lines
arch/x86/kernel/cpu/mtrr/main.c int mtrr_del(int reg, unsigned long base, unsigned long size) · line 545
Quoted 3 lines
arch/x86/kernel/cpu/mtrr/main.c +3
-0+ +if (pat_enabled) +return 0 ;/* Success! (We don't need to do anything.) */ Shouldn't we #define a big number for this case since
mtrr_addreturns 0-based mtrr indices? Rather unlikely that the very first mtrr is unused I know, but still feels like a cleaner interface. And we don't need to leak that #define out at all to users of this interface.
I did something more like that in v1, but I like having the property that arch_phys_wc_del(0) does nothing as opposed to possibly breaking the system. Given that almost all of these things are stored in kzalloced space, this seems safer to me. The return value here could just as easily be -ENOSYS, but nothing should care.
There is an issue, though: the drm maps interface is leaking the return values to userspace via a couple of ioctls, so I should fix it to at least return the correct value. (Why that interface includes an mtrr number is a mystery to me.)
Have I convinced you that my bike shed color is okay? I'll send out a v3 later today with a fix for the leaking-to-userspace issue and I'll fix the i915 thing.
--Andy
Re: [RFC/PATCH v2 1/8] Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed
Quoted 102 lines · Andy Lutomirski
Quoted 86 lines · Daniel Vetter
Quoted 85 lines · Andy Lutomirski
Several drivers currently use
mtrr_addthrough various #ifdef guards and/or drm wrappers. The vast majority of them want to add WC MTRRs on x86 systems and don't actually need the MTRR if PAT (i.e.ioremap_wc, etc) are working.
arch_phys_wc_addandarch_phys_wc_delare new functions, available on all architectures and configurations, that add WC MTRRs on x86 if needed (and handle errors) and do nothing at all otherwise. They're also easier to use thanmtrr_addandmtrr_del, so the call sites can be simplified.As an added benefit, this will avoid wasting MTRRs and possibly warning pointlessly on PAT-supporting systems.
Signed-off-by Andy Lutomirski <[email protected]>--- arch/x86/include/asm/io.h | 7 ++++++ arch/x86/include/asm/mtrr.h | 5 ++++- arch/x86/kernel/cpu/mtrr/main.c | 48 +++++++++++++++++++++++++++++++++++++++++ include/linux/io.h | 25 +++++++++++++++++++++ 4 files changed, 84 insertions(+), 1 deletion(-)arch/x86/include/asm/io.h +7
-0@@ -345,4 +345,11 @@ extern bool xen_biovec_phys_mergeable(const struct bio_vec *vec1,#define IO_SPACE_LIMIT 0xffff +#ifdef CONFIG_MTRR +extern int __must_check arch_phys_wc_add (unsigned long base, +unsigned long size); +extern void arch_phys_wc_del (int handle); +#define arch_phys_wc_add arch_phys_wc_add +#endif +#endif /* _ASM_X86_IO_H */ arch/x86/include/asm/mtrr.h +4
-1@@ -26,7 +26,10 @@#include <uapi/asm/mtrr.h> -/* The following functions are for use by other drivers */ +/* +* The following functions are for use by other drivers that cannot use +* arch_phys_wc_add and arch_phys_wc_del. +*/ # ifdef CONFIG_MTRR extern u8 mtrr_type_lookup (u64 addr,u64 end);extern void mtrr_save_fixed_ranges (void *);arch/x86/kernel/cpu/mtrr/main.c +16
-0@@ -51,6 +51,7 @@#include <asm/e820.h> #include <asm/mtrr.h> #include <asm/msr.h> +#include <asm/pat.h> #include "mtrr.h" @@ -524,6 +525,53 @@ int mtrr_del(int reg, unsigned long base, unsigned long size) }EXPORT_SYMBOL (mtrr_del); +/** +* arch_phys_wc_add - add a WC MTRR and handle errors if PAT is unavailable +* @base: Physical base address +* @size: Size of region +* +* If PAT is available, this does nothing. If PAT is unavailable, it +* attempts to add a WC MTRR covering size bytes starting at base and +* logs an error if this fails. +* +* Drivers must store the return value to pass to mtrr_del_wc_if_needed, +* but drivers should not try to interpret that return value. +*/ +int arch_phys_wc_add (unsigned long base,unsigned long size) +{ +int ret;Quoted 8 lines
arch/x86/kernel/cpu/mtrr/main.c int mtrr_del(int reg, unsigned long base, unsigned long size) · line 545
Quoted 3 lines
arch/x86/kernel/cpu/mtrr/main.c +3
-0+ +if (pat_enabled) +return 0 ;/* Success! (We don't need to do anything.) */ Shouldn't we #define a big number for this case since
mtrr_addreturns 0-based mtrr indices? Rather unlikely that the very first mtrr is unused I know, but still feels like a cleaner interface. And we don't need to leak that #define out at all to users of this interface.I did something more like that in v1, but I like having the property that
arch_phys_wc_del(0)does nothing as opposed to possibly breaking the system. Given that almost all of these things are stored in kzalloced space, this seems safer to me. The return value here could just as easily be-ENOSYS, but nothing should care.
Quoted 4 lines
There is an issue, though: the drm maps interface is leaking the return values to userspace via a couple of ioctls, so I should fix it to at least return the correct value. (Why that interface includes an mtrr number is a mystery to me.)
Old drm interfaces are horrible. Best approach is to simply stay far away from them ...
Quoted 3 lines
Have I convinced you that my bike shed color is okay? I'll send out a v3 later today with a fix for the leaking-to-userspace issue and I'll fix the i915 thing.
A comment in your arch_phys_wc_add/del functions that we do have an mtrr with index 0, but won't ever get that would be fine with me. Still feels a bit irky though ;-)
Cheers, Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [RFC/PATCH v2 1/8] Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed
Quoted 111 lines · Daniel Vetter
Quoted 102 lines · Andy Lutomirski
Quoted 86 lines · Daniel Vetter
Quoted 85 lines · Andy Lutomirski
Several drivers currently use
mtrr_addthrough various #ifdef guards and/or drm wrappers. The vast majority of them want to add WC MTRRs on x86 systems and don't actually need the MTRR if PAT (i.e.ioremap_wc, etc) are working.
arch_phys_wc_addandarch_phys_wc_delare new functions, available on all architectures and configurations, that add WC MTRRs on x86 if needed (and handle errors) and do nothing at all otherwise. They're also easier to use thanmtrr_addandmtrr_del, so the call sites can be simplified.As an added benefit, this will avoid wasting MTRRs and possibly warning pointlessly on PAT-supporting systems.
Signed-off-by Andy Lutomirski <[email protected]>--- arch/x86/include/asm/io.h | 7 ++++++ arch/x86/include/asm/mtrr.h | 5 ++++- arch/x86/kernel/cpu/mtrr/main.c | 48 +++++++++++++++++++++++++++++++++++++++++ include/linux/io.h | 25 +++++++++++++++++++++ 4 files changed, 84 insertions(+), 1 deletion(-)arch/x86/include/asm/io.h +7
-0@@ -345,4 +345,11 @@ extern bool xen_biovec_phys_mergeable(const struct bio_vec *vec1,#define IO_SPACE_LIMIT 0xffff +#ifdef CONFIG_MTRR +extern int __must_check arch_phys_wc_add (unsigned long base, +unsigned long size); +extern void arch_phys_wc_del (int handle); +#define arch_phys_wc_add arch_phys_wc_add +#endif +#endif /* _ASM_X86_IO_H */ arch/x86/include/asm/mtrr.h +4
-1@@ -26,7 +26,10 @@#include <uapi/asm/mtrr.h> -/* The following functions are for use by other drivers */ +/* +* The following functions are for use by other drivers that cannot use +* arch_phys_wc_add and arch_phys_wc_del. +*/ # ifdef CONFIG_MTRR extern u8 mtrr_type_lookup (u64 addr,u64 end);extern void mtrr_save_fixed_ranges (void *);arch/x86/kernel/cpu/mtrr/main.c +16
-0@@ -51,6 +51,7 @@#include <asm/e820.h> #include <asm/mtrr.h> #include <asm/msr.h> +#include <asm/pat.h> #include "mtrr.h" @@ -524,6 +525,53 @@ int mtrr_del(int reg, unsigned long base, unsigned long size) }EXPORT_SYMBOL (mtrr_del); +/** +* arch_phys_wc_add - add a WC MTRR and handle errors if PAT is unavailable +* @base: Physical base address +* @size: Size of region +* +* If PAT is available, this does nothing. If PAT is unavailable, it +* attempts to add a WC MTRR covering size bytes starting at base and +* logs an error if this fails. +* +* Drivers must store the return value to pass to mtrr_del_wc_if_needed, +* but drivers should not try to interpret that return value. +*/ +int arch_phys_wc_add (unsigned long base,unsigned long size) +{ +int ret;Quoted 8 lines
arch/x86/kernel/cpu/mtrr/main.c int mtrr_del(int reg, unsigned long base, unsigned long size) · line 545
Quoted 3 lines
arch/x86/kernel/cpu/mtrr/main.c +3
-0+ +if (pat_enabled) +return 0 ;/* Success! (We don't need to do anything.) */ Shouldn't we #define a big number for this case since
mtrr_addreturns 0-based mtrr indices? Rather unlikely that the very first mtrr is unused I know, but still feels like a cleaner interface. And we don't need to leak that #define out at all to users of this interface.I did something more like that in v1, but I like having the property that
arch_phys_wc_del(0)does nothing as opposed to possibly breaking the system. Given that almost all of these things are stored in kzalloced space, this seems safer to me. The return value here could just as easily be-ENOSYS, but nothing should care.Quoted 4 lines
There is an issue, though: the drm maps interface is leaking the return values to userspace via a couple of ioctls, so I should fix it to at least return the correct value. (Why that interface includes an mtrr number is a mystery to me.)
Old drm interfaces are horrible. Best approach is to simply stay far away from them ...
Quoted 7 lines
Quoted 3 lines
Have I convinced you that my bike shed color is okay? I'll send out a v3 later today with a fix for the leaking-to-userspace issue and I'll fix the i915 thing.
A comment in your arch_phys_wc_add/del functions that we do have an mtrr with index 0, but won't ever get that would be fine with me. Still feels a bit irky though ;-)
Hmm. Maybe you missed the hack I ended up using a couple of lines below. I'm always using strictly positive values to indicate a real MTRR -- I'm adding 1000 to the result if we actually do anything. So a return value of 0 is genuinely safe. (I still have to fix up the two places in the drm code that pass that hacked-up value back to userspace.)
--Andy
Daniel Vetter
Reviewed-byRe: [RFC/PATCH v2 1/8] Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed
Quoted 120 lines · Andy Lutomirski
Quoted 111 lines · Daniel Vetter
Quoted 102 lines · Andy Lutomirski
Quoted 86 lines · Daniel Vetter
Quoted 85 lines · Andy Lutomirski
Several drivers currently use
mtrr_addthrough various #ifdef guards and/or drm wrappers. The vast majority of them want to add WC MTRRs on x86 systems and don't actually need the MTRR if PAT (i.e.ioremap_wc, etc) are working.
arch_phys_wc_addandarch_phys_wc_delare new functions, available on all architectures and configurations, that add WC MTRRs on x86 if needed (and handle errors) and do nothing at all otherwise. They're also easier to use thanmtrr_addandmtrr_del, so the call sites can be simplified.As an added benefit, this will avoid wasting MTRRs and possibly warning pointlessly on PAT-supporting systems.
Signed-off-by Andy Lutomirski <[email protected]>--- arch/x86/include/asm/io.h | 7 ++++++ arch/x86/include/asm/mtrr.h | 5 ++++- arch/x86/kernel/cpu/mtrr/main.c | 48 +++++++++++++++++++++++++++++++++++++++++ include/linux/io.h | 25 +++++++++++++++++++++ 4 files changed, 84 insertions(+), 1 deletion(-)arch/x86/include/asm/io.h +7
-0@@ -345,4 +345,11 @@ extern bool xen_biovec_phys_mergeable(const struct bio_vec *vec1,#define IO_SPACE_LIMIT 0xffff +#ifdef CONFIG_MTRR +extern int __must_check arch_phys_wc_add (unsigned long base, +unsigned long size); +extern void arch_phys_wc_del (int handle); +#define arch_phys_wc_add arch_phys_wc_add +#endif +#endif /* _ASM_X86_IO_H */ arch/x86/include/asm/mtrr.h +4
-1@@ -26,7 +26,10 @@#include <uapi/asm/mtrr.h> -/* The following functions are for use by other drivers */ +/* +* The following functions are for use by other drivers that cannot use +* arch_phys_wc_add and arch_phys_wc_del. +*/ # ifdef CONFIG_MTRR extern u8 mtrr_type_lookup (u64 addr,u64 end);extern void mtrr_save_fixed_ranges (void *);arch/x86/kernel/cpu/mtrr/main.c +16
-0@@ -51,6 +51,7 @@#include <asm/e820.h> #include <asm/mtrr.h> #include <asm/msr.h> +#include <asm/pat.h> #include "mtrr.h" @@ -524,6 +525,53 @@ int mtrr_del(int reg, unsigned long base, unsigned long size) }EXPORT_SYMBOL (mtrr_del); +/** +* arch_phys_wc_add - add a WC MTRR and handle errors if PAT is unavailable +* @base: Physical base address +* @size: Size of region +* +* If PAT is available, this does nothing. If PAT is unavailable, it +* attempts to add a WC MTRR covering size bytes starting at base and +* logs an error if this fails. +* +* Drivers must store the return value to pass to mtrr_del_wc_if_needed, +* but drivers should not try to interpret that return value. +*/ +int arch_phys_wc_add (unsigned long base,unsigned long size) +{ +int ret;Quoted 8 lines
arch/x86/kernel/cpu/mtrr/main.c int mtrr_del(int reg, unsigned long base, unsigned long size) · line 545
Quoted 3 lines
arch/x86/kernel/cpu/mtrr/main.c +3
-0+ +if (pat_enabled) +return 0 ;/* Success! (We don't need to do anything.) */ Shouldn't we #define a big number for this case since
mtrr_addreturns 0-based mtrr indices? Rather unlikely that the very first mtrr is unused I know, but still feels like a cleaner interface. And we don't need to leak that #define out at all to users of this interface.I did something more like that in v1, but I like having the property that
arch_phys_wc_del(0)does nothing as opposed to possibly breaking the system. Given that almost all of these things are stored in kzalloced space, this seems safer to me. The return value here could just as easily be-ENOSYS, but nothing should care.Quoted 4 lines
There is an issue, though: the drm maps interface is leaking the return values to userspace via a couple of ioctls, so I should fix it to at least return the correct value. (Why that interface includes an mtrr number is a mystery to me.)
Old drm interfaces are horrible. Best approach is to simply stay far away from them ...
Quoted 7 lines
Quoted 3 lines
Have I convinced you that my bike shed color is okay? I'll send out a v3 later today with a fix for the leaking-to-userspace issue and I'll fix the i915 thing.
A comment in your arch_phys_wc_add/del functions that we do have an mtrr with index 0, but won't ever get that would be fine with me. Still feels a bit irky though ;-)
Quoted 6 lines
Hmm. Maybe you missed the hack I ended up using a couple of lines below. I'm always using strictly positive values to indicate a real MTRR -- I'm adding 1000 to the result if we actually do anything. So a return value of 0 is genuinely safe. (I still have to fix up the two places in the drm code that pass that hacked-up value back to userspace.)
Oops, I've indeed missed that one. Only noticed the special 0 and started hunting down the semantics of mtrr_add. With that cleared up this is
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH v2 2/8 drm (ast,cirrus,mgag200,nouveau,savage,vmwgfx): Remove drm_mtrr_{add,del}
In reply to Andy LutomirskiThis replaces drm_mtrr_{add,del} with arch_phys_wc_{add,del}. The interface is simplified (because the base and size parameters to drm_mtrr_del never did anything), and it no longer adds MTRRs on systems that don't need them.
--- drivers/gpu/drm/ast/ast_ttm.c | 13 +++-------- drivers/gpu/drm/cirrus/cirrus_ttm.c | 15 ++++-------- drivers/gpu/drm/mgag200/mgag200_ttm.c | 14 ++++-------- drivers/gpu/drm/nouveau/nouveau_ttm.c | 13 ++++------- drivers/gpu/drm/savage/savage_bci.c | 43 ++++++++++++----------------------- drivers/gpu/drm/savage/savage_drv.h | 5 +--- drivers/gpu/drm/vmwgfx/vmwgfx_drv.c | 10 ++++---- include/drm/drmP.h | 29 ----------------------- 8 files changed, 35 insertions(+), 107 deletions(-)
drivers/gpu/drm/ast/ast_ttm.c +3 -10
@@ -271,26 +271,19 @@ int ast_mm_init(struct ast_private *ast)return ret; } - ast->fb_mtrr =drm_mtrr_add (pci_resource_start (dev->pdev,0 ), -pci_resource_len (dev->pdev,0 ), -DRM_MTRR_WC ); + ast->fb_mtrr =arch_phys_wc_add (pci_resource_start (dev->pdev,0 ), +pci_resource_len (dev->pdev,0 ));return 0 ; }void ast_mm_fini (struct ast_private *ast) { -struct drm_device *dev = ast->dev;ttm_bo_device_release (&ast->ttm.bdev);ast_ttm_global_release (ast); -if (ast->fb_mtrr >=0 ) { -drm_mtrr_del (ast->fb_mtrr, -pci_resource_start (dev->pdev,0 ), -pci_resource_len (dev->pdev,0 ),DRM_MTRR_WC ); - ast->fb_mtrr =-1 ; - } +arch_phys_wc_del (ast->fb_mtrr); }void ast_ttm_placement (struct ast_bo *bo,int domain)
drivers/gpu/drm/cirrus/cirrus_ttm.c +4 -11
@@ -271,9 +271,8 @@ int cirrus_mm_init(struct cirrus_device *cirrus)return ret; } - cirrus->fb_mtrr =drm_mtrr_add (pci_resource_start (dev->pdev,0 ), -pci_resource_len (dev->pdev,0 ), -DRM_MTRR_WC ); + cirrus->fb_mtrr =arch_phys_wc_add (pci_resource_start (dev->pdev,0 ), +pci_resource_len (dev->pdev,0 )); cirrus->mm_inited = true;return 0 ;
@@ -281,8 +280,6 @@ int cirrus_mm_init(struct cirrus_device *cirrus)void cirrus_mm_fini (struct cirrus_device *cirrus) { -struct drm_device *dev = cirrus->dev; -if (!cirrus->mm_inited)return ;
@@ -290,12 +287,8 @@ void cirrus_mm_fini(struct cirrus_device *cirrus)cirrus_ttm_global_release (cirrus); -if (cirrus->fb_mtrr >=0 ) { -drm_mtrr_del (cirrus->fb_mtrr, -pci_resource_start (dev->pdev,0 ), -pci_resource_len (dev->pdev,0 ),DRM_MTRR_WC ); - cirrus->fb_mtrr =-1 ; - } +arch_phys_wc_del (cirrus->fb_mtrr); + cirrus->fb_mtrr =0 ; }void cirrus_ttm_placement (struct cirrus_bo *bo,int domain)
drivers/gpu/drm/mgag200/mgag200_ttm.c +4 -10
@@ -270,26 +270,20 @@ int mgag200_mm_init(struct mga_device *mdev)return ret; } - mdev->fb_mtrr =drm_mtrr_add (pci_resource_start (dev->pdev,0 ), -pci_resource_len (dev->pdev,0 ), -DRM_MTRR_WC ); + mdev->fb_mtrr =arch_phys_wc_add (pci_resource_start (dev->pdev,0 ), +pci_resource_len (dev->pdev,0 ));return 0 ; }void mgag200_mm_fini (struct mga_device *mdev) { -struct drm_device *dev = mdev->dev;ttm_bo_device_release (&mdev->ttm.bdev);mgag200_ttm_global_release (mdev); -if (mdev->fb_mtrr >=0 ) { -drm_mtrr_del (mdev->fb_mtrr, -pci_resource_start (dev->pdev,0 ), -pci_resource_len (dev->pdev,0 ),DRM_MTRR_WC ); - mdev->fb_mtrr =-1 ; - } +arch_phys_wc_del (mdev->fb_mtrr); + mdev->fb_mtrr =0 ; }void mgag200_ttm_placement (struct mgag200_bo *bo,int domain)
drivers/gpu/drm/nouveau/nouveau_ttm.c +4 -9
@@ -377,9 +377,8 @@ nouveau_ttm_init(struct nouveau_drm *drm)return ret; } - drm->ttm.mtrr =drm_mtrr_add (pci_resource_start (dev->pdev,1 ), -pci_resource_len (dev->pdev,1 ), -DRM_MTRR_WC ); + drm->ttm.mtrr =arch_phys_wc_add (pci_resource_start (dev->pdev,1 ), +pci_resource_len (dev->pdev,1 ));/* GART init */ if (drm->agp.stat !=ENABLED ) {
@@ -414,10 +413,6 @@ nouveau_ttm_fini(struct nouveau_drm *drm)nouveau_ttm_global_release (drm); -if (drm->ttm.mtrr >=0 ) { -drm_mtrr_del (drm->ttm.mtrr, -pci_resource_start (drm->dev->pdev,1 ), -pci_resource_len (drm->dev->pdev,1 ),DRM_MTRR_WC ); - drm->ttm.mtrr =-1 ; - } +arch_phys_wc_del (drm->ttm.mtrr); + drm->ttm.mtrr =0 ; }
drivers/gpu/drm/savage/savage_bci.c +15 -28
67 lines not on this page · raw message
drivers/gpu/drm/savage/savage_drv.h +1 -4
@@ -160,10 +160,7 @@ typedef struct drm_savage_private {drm_local_map_t *cmd_dma;drm_local_map_t fake_dma; -struct { -int handle; -unsigned long base, size; - } mtrr[3 ]; +int mtrr_handles[3 ];/* BCI and status-related stuff */ volatile uint32_t *status_ptr, *bci_ptr;
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +4 -6
@@ -565,8 +565,8 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) dev_priv->has_gmr = false; } - dev_priv->mmio_mtrr =drm_mtrr_add (dev_priv->mmio_start, - dev_priv->mmio_size,DRM_MTRR_WC ); + dev_priv->mmio_mtrr =arch_phys_wc_add (dev_priv->mmio_start, + dev_priv->mmio_size); dev_priv->mmio_virt =ioremap_wc (dev_priv->mmio_start, dev_priv->mmio_size);
@@ -664,8 +664,7 @@ out_no_device: out_err4:iounmap (dev_priv->mmio_virt); out_err3: -drm_mtrr_del (dev_priv->mmio_mtrr, dev_priv->mmio_start, - dev_priv->mmio_size,DRM_MTRR_WC ); +arch_phys_wc_del (dev_priv->mmio_mtrr);if (dev_priv->has_gmr) (void )ttm_bo_clean_mm (&dev_priv->bdev,VMW_PL_GMR ); (void )ttm_bo_clean_mm (&dev_priv->bdev,TTM_PL_VRAM );
@@ -709,8 +708,7 @@ static int vmw_driver_unload(struct drm_device *dev)ttm_object_device_release (&dev_priv->tdev);iounmap (dev_priv->mmio_virt); -drm_mtrr_del (dev_priv->mmio_mtrr, dev_priv->mmio_start, - dev_priv->mmio_size,DRM_MTRR_WC ); +arch_phys_wc_del (dev_priv->mmio_mtrr);if (dev_priv->has_gmr) (void )ttm_bo_clean_mm (&dev_priv->bdev,VMW_PL_GMR ); (void )ttm_bo_clean_mm (&dev_priv->bdev,TTM_PL_VRAM );
include/drm/drmP.h +0 -29
@@ -1249,37 +1249,8 @@ static inline int drm_core_has_MTRR(struct drm_device *dev) {return drm_core_check_feature (dev,DRIVER_USE_MTRR ); } - -#define DRM_MTRR_WC MTRR_TYPE_WRCOMB - -static inline int drm_mtrr_add (unsigned long offset,unsigned long size, -unsigned int flags) -{ -return mtrr_add (offset, size, flags,1 ); -} - -static inline int drm_mtrr_del (int handle,unsigned long offset, -unsigned long size,unsigned int flags) -{ -return mtrr_del (handle, offset, size); -} -#else #define drm_core_has_MTRR (dev) (0) - -#define DRM_MTRR_WC 0 - -static inline int drm_mtrr_add (unsigned long offset,unsigned long size, -unsigned int flags) -{ -return 0 ; -} - -static inline int drm_mtrr_del (int handle,unsigned long offset, -unsigned long size,unsigned int flags) -{ -return 0 ; -}#endif static inline void drm_device_set_unplugged (struct drm_device *dev)
Signature · 1 line
1.8.1.4
PATCH v2 3/8 drm: Update drm_addmap and drm_mmap to use PAT WC instead of MTRRs
In reply to Andy LutomirskiPreviously, DRM_FRAME_BUFFER mappings, as well as DRM_REGISTERS mappings with DRM_WRITE_COMBINING set, resulted in an unconditional MTRR being added but the actual mappings being created as UC-.
Now these mappings have the MTRR added only if needed, but they will be mapped with pgprot_writecombine.
The non-WC DRM_REGISTERS case now uses pgprot_noncached instead of hardcoding the bit twiddling.
The DRM_AGP case is unchanged for now.
---
Unlike v1, this should work well even for drivers that call drmAddMap from userspace. I didn't understand how the flags were supposed to work last time around.
drivers/gpu/drm/drm_bufs.c | 17 +++++++++-------- drivers/gpu/drm/drm_vm.c | 23 +++++++++++------------ 2 files changed, 20 insertions(+), 20 deletions(-)
drivers/gpu/drm/drm_bufs.c +9 -8
@@ -210,12 +210,16 @@ static int drm_addmap_core(struct drm_device * dev, resource_size_t offset,if (drm_core_has_MTRR (dev)) {if (map->type == _DRM_FRAME_BUFFER || (map->flags & _DRM_WRITE_COMBINING)) { - map->mtrr =mtrr_add (map->offset, map->size, -MTRR_TYPE_WRCOMB ,1 ); + map->mtrr = +arch_phys_wc_add (map->offset, map->size); } }if (map->type == _DRM_REGISTERS) { - map->handle =ioremap (map->offset, map->size); +if (map->flags & _DRM_WRITE_COMBINING) + map->handle =ioremap_wc (map->offset, + map->size); +else + map->handle =ioremap (map->offset, map->size);if (!map->handle) {kfree (map);return -ENOMEM ;
@@ -451,11 +455,8 @@ int drm_rmmap_locked(struct drm_device *dev, struct drm_local_map *map)iounmap (map->handle);/* FALLTHROUGH */ case _DRM_FRAME_BUFFER: -if (drm_core_has_MTRR (dev) && map->mtrr >=0 ) { -int retcode; - retcode =mtrr_del (map->mtrr, map->offset, map->size); -DRM_DEBUG ("mtrr_del=%d\n" , retcode); - } +if (drm_core_has_MTRR (dev)) +arch_phys_wc_del (map->mtrr);break ;case _DRM_SHM:vfree (map->handle);
drivers/gpu/drm/drm_vm.c +11 -12
@@ -43,14 +43,18 @@static void drm_vm_open (struct vm_area_struct *vma);static void drm_vm_close (struct vm_area_struct *vma); -static pgprot_t drm_io_prot (uint32_t map_type,struct vm_area_struct *vma) +static pgprot_t drm_io_prot (struct drm_local_map *map, +struct vm_area_struct *vma) {pgprot_t tmp =vm_get_page_prot (vma->vm_flags);#if defined(__i386__) || defined(__x86_64__) -if (boot_cpu_data.x86 >3 && map_type != _DRM_AGP) { -pgprot_val (tmp) |= _PAGE_PCD; -pgprot_val (tmp) &= ~_PAGE_PWT; +if (map->type != _DRM_AGP) { +if (map->type == _DRM_FRAME_BUFFER || + map->flags & _DRM_WRITE_COMBINING) + tmp =pgprot_writecombine (tmp); +else + tmp =pgprot_noncached (tmp); }#elif defined(__powerpc__)pgprot_val (tmp) |= _PAGE_NO_CACHE;
@@ -250,13 +254,8 @@ static void drm_vm_shm_close(struct vm_area_struct *vma)switch (map->type) {case _DRM_REGISTERS:case _DRM_FRAME_BUFFER: -if (drm_core_has_MTRR (dev) && map->mtrr >=0 ) { -int retcode; - retcode =mtrr_del (map->mtrr, - map->offset, - map->size); -DRM_DEBUG ("mtrr_del = %d\n" , retcode); - } +if (drm_core_has_MTRR (dev)) +arch_phys_wc_del (map->mtrr);iounmap (map->handle);break ;case _DRM_SHM:
@@ -617,7 +616,7 @@ int drm_mmap_locked(struct file *filp, struct vm_area_struct *vma)case _DRM_REGISTERS: offset =drm_core_get_reg_ofs (dev); vma->vm_flags |=VM_IO ;/* not in core dump */ - vma->vm_page_prot =drm_io_prot (map->type, vma); + vma->vm_page_prot =drm_io_prot (map, vma); if (io_remap_pfn_range(vma, vma->vm_start, (map->offset + offset) >> PAGE_SHIFT, vma->vm_end - vma->vm_start,
Signature · 1 line
1.8.1.4
PATCH v2 4/8 drm,agpgart: Use pgprot_writecombine for AGP maps and make the MTRR optional
In reply to Andy LutomirskiI'm not sure I understand the intent of the previous behavior. mmap on /dev/agpgart and DRM_AGP maps had no cache flags set, so they would be fully cacheable. But the DRM code (most of the time) would add a write-combining MTRR that would change the effective memory type to WC.
The new behavior just requests WC explicitly for all AGP maps.
If there is any code out there that expects cacheable access to the AGP aperture (because the drm driver doesn't request an MTRR or because it's using /dev/agpgart directly), then it will now end up with a UC or WC mapping, depending on the architecture and PAT availability. But cacheable access to the aperture seems like it's asking for trouble, because, AIUI, the aperture is an alias of RAM.
---
It's conceivable that libpciaccess could have issues with this due to memtype conflicts, but I think this is unlikely (as long as a WC mapping gets there first, I think the conflict resolution rules will all work out). In any case, everything that maps the AGP aperture ought to be using /dev/agpgart or the DRM API, in which case everything should be okay.
I don't have anything with an AGP slot to test AFAIK.
drivers/char/agp/frontend.c | 8 +++++--- drivers/gpu/drm/drm_pci.c | 8 ++++---- drivers/gpu/drm/drm_stub.c | 10 ++-------- drivers/gpu/drm/drm_vm.c | 11 ++++------- 4 files changed, 15 insertions(+), 22 deletions(-)
drivers/char/agp/frontend.c +5 -3
@@ -603,7 +603,8 @@ static int agp_mmap(struct file *file, struct vm_area_struct *vma) vma->vm_ops = kerninfo.vm_ops; }else if (io_remap_pfn_range (vma, vma->vm_start, (kerninfo.aper_base + offset) >>PAGE_SHIFT , - size, vma->vm_page_prot)) { + size, +pgprot_writecombine (vma->vm_page_prot))) { goto out_again; }mutex_unlock (&(agp_fe.agp_mutex));
@@ -618,8 +619,9 @@ static int agp_mmap(struct file *file, struct vm_area_struct *vma)if (kerninfo.vm_ops) { vma->vm_ops = kerninfo.vm_ops; }else if (io_remap_pfn_range (vma, vma->vm_start, - kerninfo.aper_base >>PAGE_SHIFT , - size, vma->vm_page_prot)) { + kerninfo.aper_base >>PAGE_SHIFT , + size, +pgprot_writecombine (vma->vm_page_prot))) { goto out_again; }mutex_unlock (&(agp_fe.agp_mutex));
drivers/gpu/drm/drm_pci.c +4 -4
@@ -278,10 +278,10 @@ int drm_pci_agp_init(struct drm_device *dev) }if (drm_core_has_MTRR (dev)) {if (dev->agp) - dev->agp->agp_mtrr = -mtrr_add (dev->agp->agp_info.aper_base, - dev->agp->agp_info.aper_size * -1024 *1024 ,MTRR_TYPE_WRCOMB ,1 ); + dev->agp->agp_mtrr =arch_phys_wc_add ( + dev->agp->agp_info.aper_base, + dev->agp->agp_info.aper_size * +1024 *1024 ); } }return 0 ;
drivers/gpu/drm/drm_stub.c +2 -8
@@ -451,14 +451,8 @@ void drm_put_dev(struct drm_device *dev)drm_lastclose (dev); -if (drm_core_has_MTRR (dev) &&drm_core_has_AGP (dev) && - dev->agp && dev->agp->agp_mtrr >=0 ) { -int retval; - retval =mtrr_del (dev->agp->agp_mtrr, - dev->agp->agp_info.aper_base, - dev->agp->agp_info.aper_size *1024 *1024 ); -DRM_DEBUG ("mtrr_del=%d\n" , retval); - } +if (drm_core_has_MTRR (dev) &&drm_core_has_AGP (dev) && dev->agp) +arch_phys_wc_del (dev->agp->agp_mtrr);if (dev->driver->unload) dev->driver->unload (dev);
drivers/gpu/drm/drm_vm.c +4 -7
@@ -49,13 +49,10 @@ static pgprot_t drm_io_prot(struct drm_local_map *map,pgprot_t tmp =vm_get_page_prot (vma->vm_flags);#if defined(__i386__) || defined(__x86_64__) -if (map->type != _DRM_AGP) { -if (map->type == _DRM_FRAME_BUFFER || - map->flags & _DRM_WRITE_COMBINING) - tmp =pgprot_writecombine (tmp); -else - tmp =pgprot_noncached (tmp); - } +if (map->type == _DRM_REGISTERS && !(map->flags & _DRM_WRITE_COMBINING)) + tmp =pgprot_noncached (tmp); +else + tmp =pgprot_writecombine (tmp);#elif defined(__powerpc__)pgprot_val (tmp) |= _PAGE_NO_CACHE;if (map_type == _DRM_REGISTERS)
Signature · 1 line
1.8.1.4
PATCH v2 5/8 i915: Use arch_phys_wc_{add,del}
In reply to Andy Lutomirskii915 open-coded logic that was essentially equivalent to the new API.
---
Changes from v1: More cleanup
drivers/gpu/drm/i915/i915_dma.c | 44 ++++++----------------------------------- 1 file changed, 6 insertions(+), 38 deletions(-)
drivers/gpu/drm/i915/i915_dma.c +6 -38
@@ -42,7 +42,6 @@#include <linux/vga_switcheroo.h> #include <linux/slab.h> #include <acpi/video.h> -#include <asm/pat.h> #define LP_RING (d) (&((struct drm_i915_private *)(d))->ring[RCS])
@@ -1393,29 +1392,6 @@ void i915_master_destroy(struct drm_device *dev, struct drm_master *master) master->driver_priv =NULL ; } -static void -i915_mtrr_setup (struct drm_i915_private *dev_priv,unsigned long base, -unsigned long size) -{ - dev_priv->mm.gtt_mtrr =-1 ; - -#if defined(CONFIG_X86_PAT ) -if (cpu_has_pat) -return ; -#endif - -/* Set up a WC MTRR for non-PAT systems. This is more common than -* one would think, because the kernel disables PAT on first -* generation Core chips because WC PAT gets overridden by a UC -* MTRR if present. Even if a UC MTRR isn't present. -*/ - dev_priv->mm.gtt_mtrr =mtrr_add (base, size,MTRR_TYPE_WRCOMB ,1 ); -if (dev_priv->mm.gtt_mtrr <0 ) { -DRM_INFO ("MTRR allocation failed. Graphics " -"performance may suffer.\n" ); - } -} -static void i915_kick_out_firmware_fb (struct drm_i915_private *dev_priv) {struct apertures_struct *ap;
@@ -1552,8 +1528,8 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags) goto out_rmmap; } -i915_mtrr_setup (dev_priv, dev_priv->gtt.mappable_base, - aperture_size); + dev_priv->mm.gtt_mtrr =arch_phys_wc_add (dev_priv->gtt.mappable_base, + aperture_size);/* The i915 workqueue is primarily used for batched retirement of * requests (and thus managing bo) once the task has been completed
@@ -1656,12 +1632,8 @@ out_gem_unload:intel_teardown_mchbar (dev);destroy_workqueue (dev_priv->wq); out_mtrrfree: -if (dev_priv->mm.gtt_mtrr >=0 ) { -mtrr_del (dev_priv->mm.gtt_mtrr, - dev_priv->gtt.mappable_base, - aperture_size); - dev_priv->mm.gtt_mtrr =-1 ; - } +arch_phys_wc_del (dev_priv->mm.gtt_mtrr); + dev_priv->mm.gtt_mtrr =0 ;io_mapping_free (dev_priv->gtt.mappable); out_rmmap:pci_iounmap (dev->pdev, dev_priv->regs);
@@ -1697,12 +1669,8 @@ int i915_driver_unload(struct drm_device *dev)cancel_delayed_work_sync (&dev_priv->mm.retire_work);io_mapping_free (dev_priv->gtt.mappable); -if (dev_priv->mm.gtt_mtrr >=0 ) { -mtrr_del (dev_priv->mm.gtt_mtrr, - dev_priv->gtt.mappable_base, - dev_priv->gtt.mappable_end); - dev_priv->mm.gtt_mtrr =-1 ; - } +arch_phys_wc_del (dev_priv->mm.gtt_mtrr); + dev_priv->mm.gtt_mtrr =0 ;acpi_video_unregister ();
Signature · 1 line
1.8.1.4
Re: [RFC/PATCH v2 5/8] i915: Use arch_phys_wc_{add,del}
Quoted 65 lines · Andy Lutomirski
i915 open-coded logic that was essentially equivalent to the new API.
Signed-off-by Andy Lutomirski <[email protected]>---
Changes from v1: More cleanup
drivers/gpu/drm/i915/i915_dma.c | 44 ++++++----------------------------------- 1 file changed, 6 insertions(+), 38 deletions(-)drivers/gpu/drm/i915/i915_dma.c +2
-26@@ -42,7 +42,6 @@#include <linux/vga_switcheroo.h> #include <linux/slab.h> #include <acpi/video.h> -#include <asm/pat.h> #define LP_RING (d) (&((struct drm_i915_private *)(d))->ring[RCS])@@ -1393,29 +1392,6 @@ void i915_master_destroy(struct drm_device *dev, struct drm_master *master) master->driver_priv =NULL ; } -static void -i915_mtrr_setup (struct drm_i915_private *dev_priv,unsigned long base, -unsigned long size) -{ - dev_priv->mm.gtt_mtrr =-1 ; - -#if defined(CONFIG_X86_PAT ) -if (cpu_has_pat) -return ; -#endif - -/* Set up a WC MTRR for non-PAT systems. This is more common than -* one would think, because the kernel disables PAT on first -* generation Core chips because WC PAT gets overridden by a UC -* MTRR if present. Even if a UC MTRR isn't present. -*/ - dev_priv->mm.gtt_mtrr =mtrr_add (base, size,MTRR_TYPE_WRCOMB ,1 ); -if (dev_priv->mm.gtt_mtrr <0 ) { -DRM_INFO ("MTRR allocation failed. Graphics " -"performance may suffer.\n" ); - } -} -static void i915_kick_out_firmware_fb (struct drm_i915_private *dev_priv) {struct apertures_struct *ap;@@ -1552,8 +1528,8 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags) goto out_rmmap; } -i915_mtrr_setup (dev_priv, dev_priv->gtt.mappable_base, - aperture_size); + dev_priv->mm.gtt_mtrr =arch_phys_wc_add (dev_priv->gtt.mappable_base, + aperture_size);/* The i915 workqueue is primarily used for batched retirement of * requests (and thus managing bo) once the task has been completed @@ -1656,12 +1632,8 @@ out_gem_unload:intel_teardown_mchbar (dev);
drivers/gpu/drm/i915/i915_dma.c out_gem_unload: · line 1636
Quoted 10 lines
drivers/gpu/drm/i915/i915_dma.c +2
-6destroy_workqueue (dev_priv->wq); out_mtrrfree: -if (dev_priv->mm.gtt_mtrr >=0 ) { -mtrr_del (dev_priv->mm.gtt_mtrr, - dev_priv->gtt.mappable_base, - aperture_size); - dev_priv->mm.gtt_mtrr =-1 ; - } +arch_phys_wc_del (dev_priv->mm.gtt_mtrr); + dev_priv->mm.gtt_mtrr =0 ;
I think you can drop this gtt_mtrr = 0 asignment (and the one below) since the driver will be unloaded anyway and no longer care. And with my bikeshed on the first 1 patch it'd be wrong, too.
-Daniel
Quoted 21 lines
drivers/gpu/drm/i915/i915_dma.c +2
-6io_mapping_free (dev_priv->gtt.mappable); out_rmmap:pci_iounmap (dev->pdev, dev_priv->regs);@@ -1697,12 +1669,8 @@ int i915_driver_unload(struct drm_device *dev)cancel_delayed_work_sync (&dev_priv->mm.retire_work);io_mapping_free (dev_priv->gtt.mappable); -if (dev_priv->mm.gtt_mtrr >=0 ) { -mtrr_del (dev_priv->mm.gtt_mtrr, - dev_priv->gtt.mappable_base, - dev_priv->gtt.mappable_end); - dev_priv->mm.gtt_mtrr =-1 ; - } +arch_phys_wc_del (dev_priv->mm.gtt_mtrr); + dev_priv->mm.gtt_mtrr =0 ;acpi_video_unregister ();Signature · 1 line
1.8.1.4
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH v2 6/8 radeon: Switch to arch_phys_wc_add and add a missing ..._del
In reply to Andy Lutomirski--- drivers/gpu/drm/radeon/radeon_object.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/gpu/drm/radeon/radeon_object.c +3 -2
@@ -321,8 +321,8 @@ void radeon_bo_force_delete(struct radeon_device *rdev)int radeon_bo_init (struct radeon_device *rdev) {/* Add an MTRR for the VRAM */ - rdev->mc.vram_mtrr =mtrr_add (rdev->mc.aper_base, rdev->mc.aper_size, -MTRR_TYPE_WRCOMB ,1 ); + rdev->mc.vram_mtrr =arch_phys_wc_add (rdev->mc.aper_base, + rdev->mc.aper_size);DRM_INFO ("Detected VRAM RAM=%lluM, BAR=%lluM\n" , rdev->mc.mc_vram_size >>20 , (unsigned long long )rdev->mc.aper_size >>20 );
@@ -334,6 +334,7 @@ int radeon_bo_init(struct radeon_device *rdev)void radeon_bo_fini (struct radeon_device *rdev) {radeon_ttm_fini (rdev); +arch_phys_wc_del (rdev->mc.vram_mtrr); } void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
Signature · 1 line
1.8.1.4
PATCH v2 7/8 uvesafb: Clean up MTRR code
In reply to Andy LutomirskiThe old code allowed very strange memory types. Now it works like all the other video drivers: ioremap_wc is used unconditionally, and MTRRs are set if PAT is unavailable (unless MTRR is disabled by a module parameter).
UC, WB, and WT support is gone. If there are MTRR conflicts that prevent addition of a WC MTRR, adding a non-conflicting MTRR is pointless; it's better to just turn off MTRR support entirely.
As an added bonus, any MTRR added is freed on unload.
--- Documentation/fb/uvesafb.txt | 16 ++++------ drivers/video/uvesafb.c | 70 +++++++++++--------------------------------- include/video/uvesafb.h | 1 + 3 files changed, 23 insertions(+), 64 deletions(-)
Documentation/fb/uvesafb.txt +5 -11
@@ -81,17 +81,11 @@ pmipal Use the protected mode interface for palette changes. mtrr:n Setup memory type range registers for the framebuffer where n: - 0 - disabled (equivalent to nomtrr) (default) - 1 - uncachable - 2 - write-back - 3 - write-combining - 4 - write-through - - If you see the following in dmesg, choose the type that matches - the old one. In this example, use "mtrr:2". -... -mtrr: type mismatch for e0000000,8000000 old: write-back new: write-combining -... + 0 - disabled (equivalent to nomtrr) + 3 - write-combining (default) + + Values other than 0 and 3 will result in a warning and will be + treated just like 3. nomtrr Do not use memory type range registers.
drivers/video/uvesafb.c +17 -53
105 lines not on this page · raw message
include/video/uvesafb.h +1 -0
@@ -134,6 +134,7 @@ struct uvesafb_par {int mode_idx;struct vbe_crtc_ib crtc; + int mtrr_handle; };#endif /* _UVESAFB_H */
Signature · 1 line
1.8.1.4
PATCH v2 8/8 drm: Remove mtrr_add and mtrr_del fallback hack for non-MTRR systems
In reply to Andy LutomirskiThere are no users left in drivers/gpu.
---
This is new in v2. The code I'm deleting is kind of gross.
include/drm/drmP.h | 5 +---- include/drm/drm_os_linux.h | 16 ---------------- 2 files changed, 1 insertion(+), 20 deletions(-)
include/drm/drmP.h +1 -4
@@ -55,16 +55,13 @@#include <linux/mm.h> #include <linux/cdev.h> #include <linux/mutex.h> +#include <linux/io.h> #include <linux/slab.h> #if defined(__alpha__) || defined(__powerpc__)#include <asm/pgtable.h> /* For pte_wrprotect */ #endif -#include <asm/io.h> #include <asm/mman.h> #include <asm/uaccess.h> -#ifdef CONFIG_MTRR -#include <asm/mtrr.h> -#endif #if defined(CONFIG_AGP ) || defined(CONFIG_AGP_MODULE )#include <linux/types.h> #include <linux/agp_backend.h>
include/drm/drm_os_linux.h +0 -16
@@ -65,22 +65,6 @@ struct no_agp_kern {#define DRM_AGP_KERN struct no_agp_kern#endif -#if !(__OS_HAS_MTRR) -static __inline__int mtrr_add (unsigned long base,unsigned long size, -unsigned int type,char increment) -{ -return -ENODEV ; -} - -static __inline__int mtrr_del (int reg,unsigned long base,unsigned long size) -{ -return -ENODEV ; -} - -#define MTRR_TYPE_WRCOMB 1 - -#endif -/** Other copying of data to kernel space */ #define DRM_COPY_FROM_USER (arg1, arg2, arg3) \ copy_from_user(arg1, arg2, arg3)
Signature · 1 line
1.8.1.4
Quoted 3 lines · Andy Lutomirski
A fair number of drivers (mostly graphics) add write-combining MTRRs. Most ignore errors and most add the MTRR even on PAT systems which don't need to use MTRRs.
This comment is wrong, as i said we need MTRR on PAT system for VRAM.
Cheers,
Jerome
Quoted 62 lines
This series adds new functions arch_phys_wc_{add,del} that, on PAT-less x86 systems with MTRRs, add MTRRs and report errors, and that do nothing otherwise. (Other architectures, if any, with a similar mechanism could implement them.)
I've only tested the radeon driver, since I don't have test hardware easily available for the other drivers.
Benefits include:
- Simpler code
- No more complaints about MTRR conflict warnings on PAT systems
- Eventual unexporting of the MTRR API?This series eliminates about half of the
mtrr_addcalls indrivers/.Changes from v1:
- Helpers renamed
- Lots of bugs fixedThe series is also at:
https://git.kernel.org/cgit/linux/kernel/git/luto/linux.git/log/?h=mtrr_cleanup/rfc_v2Andy Lutomirski (8):
Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed drm (ast,cirrus,mgag200,nouveau,savage,vmwgfx): Removedrm_mtrr_{add,del}drm: Update
drm_addmapanddrm_mmapto use PAT WC instead of MTRRs drm,agpgart: Usepgprot_writecombinefor AGP maps and make the MTRR
optional
i915: Use arch_phys_wc_{add,del}
radeon: Switch toarch_phys_wc_addand add a missing ..._del
uvesafb: Clean up MTRR code
drm: Removemtrr_addandmtrr_delfallback hack for non-MTRR systemsDocumentation/fb/uvesafb.txt | 16 +++----- arch/x86/include/asm/io.h | 7 ++++ arch/x86/include/asm/mtrr.h | 5 ++- arch/x86/kernel/cpu/mtrr/main.c | 48 +++++++++++++++++++++++ drivers/char/agp/frontend.c | 8 ++-- drivers/gpu/drm/ast/ast_ttm.c | 13 ++----- drivers/gpu/drm/cirrus/cirrus_ttm.c | 15 ++------ drivers/gpu/drm/drm_bufs.c | 17 +++++---- drivers/gpu/drm/drm_pci.c | 8 ++-- drivers/gpu/drm/drm_stub.c | 10 +---- drivers/gpu/drm/drm_vm.c | 22 +++++------ drivers/gpu/drm/i915/i915_dma.c | 44 +++------------------ drivers/gpu/drm/mgag200/mgag200_ttm.c | 14 ++----- drivers/gpu/drm/nouveau/nouveau_ttm.c | 13 ++----- drivers/gpu/drm/radeon/radeon_object.c | 5 ++- drivers/gpu/drm/savage/savage_bci.c | 43 ++++++++------------- drivers/gpu/drm/savage/savage_drv.h | 5 +-- drivers/gpu/drm/vmwgfx/vmwgfx_drv.c | 10 ++--- drivers/video/uvesafb.c | 70 +++++++++------------------------- include/drm/drmP.h | 34 +---------------- include/drm/drm_os_linux.h | 16 -------- include/linux/io.h | 25 ++++++++++++ include/video/uvesafb.h | 1 + 23 files changed, 181 insertions(+), 268 deletions(-)--
1.8.1.4
Quoted 6 lines · Jerome Glisse
Quoted 3 lines · Andy Lutomirski
A fair number of drivers (mostly graphics) add write-combining MTRRs. Most ignore errors and most add the MTRR even on PAT systems which don't need to use MTRRs.
This comment is wrong, as i said we need MTRR on PAT system for VRAM.
I didn't follow it last time.
- If userspace is setting an MTRR directly, then it will work exactly
as before -- this patch has no effect on the userspace MTRR APIs.
- If userspace uses the drm map interface with DRM_FRAME_BUFFER orDRM_WRITE_COMBINING, there won't be an MTRR but the range will still be WC.
- If userspace uses GEM or TTM, then everything should still use WC
(TTM has explicit handling for this, which I presume is correct, and i915, the major GEM user, already doesn't set an MTRR).
- If userspace does not map the range, then either the kernel driver
is buggy or it doesn't matter because there's no map.
Is there a case I've missed?
--Andy
Quoted 29 lines · Andy Lutomirski
A fair number of drivers (mostly graphics) add write-combining MTRRs. Most ignore errors and most add the MTRR even on PAT systems which don't need to use MTRRs.
This series adds new functions arch_phys_wc_{add,del} that, on PAT-less x86 systems with MTRRs, add MTRRs and report errors, and that do nothing otherwise. (Other architectures, if any, with a similar mechanism could implement them.)
I've only tested the radeon driver, since I don't have test hardware easily available for the other drivers.
Benefits include:
- Simpler code
- No more complaints about MTRR conflict warnings on PAT systems
- Eventual unexporting of the MTRR API?This series eliminates about half of the
mtrr_addcalls indrivers/.Changes from v1:
- Helpers renamed
- Lots of bugs fixedThe series is also at:
https://git.kernel.org/cgit/linux/kernel/git/luto/linux.git/log/?h=mtrr_cleanup/rfc_v2Andy Lutomirski (8):
Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed drm (ast,cirrus,mgag200,nouveau,savage,vmwgfx): Remove
Quoted 8 lines
drm_mtrr_{add,del}drm: Update
drm_addmapanddrm_mmapto use PAT WC instead of MTRRs drm,agpgart: Usepgprot_writecombinefor AGP maps and make the MTRR
optional
i915: Use arch_phys_wc_{add,del}
radeon: Switch toarch_phys_wc_addand add a missing ..._del
uvesafb: Clean up MTRR code
drm: Removemtrr_addandmtrr_delfallback hack for non-MTRR systems
With my two comments addressed, the entire series is
My knowledge for the userspace driver created drm maps in drm_vm.c is a bit shoddy. Otoh most of the code I've seen using that is _horribly_ broken (as in userspace creates a register map that the kernel uses, with no refcounting at all). Your changes look correct though, so I think this is about as good as it gets.
Yours, Daniel
Quoted 29 lines
Documentation/fb/uvesafb.txt | 16 +++----- arch/x86/include/asm/io.h | 7 ++++ arch/x86/include/asm/mtrr.h | 5 ++- arch/x86/kernel/cpu/mtrr/main.c | 48 +++++++++++++++++++++++ drivers/char/agp/frontend.c | 8 ++-- drivers/gpu/drm/ast/ast_ttm.c | 13 ++----- drivers/gpu/drm/cirrus/cirrus_ttm.c | 15 ++------ drivers/gpu/drm/drm_bufs.c | 17 +++++---- drivers/gpu/drm/drm_pci.c | 8 ++-- drivers/gpu/drm/drm_stub.c | 10 +---- drivers/gpu/drm/drm_vm.c | 22 +++++------ drivers/gpu/drm/i915/i915_dma.c | 44 +++------------------ drivers/gpu/drm/mgag200/mgag200_ttm.c | 14 ++----- drivers/gpu/drm/nouveau/nouveau_ttm.c | 13 ++----- drivers/gpu/drm/radeon/radeon_object.c | 5 ++- drivers/gpu/drm/savage/savage_bci.c | 43 ++++++++------------- drivers/gpu/drm/savage/savage_drv.h | 5 +-- drivers/gpu/drm/vmwgfx/vmwgfx_drv.c | 10 ++--- drivers/video/uvesafb.c | 70 +++++++++------------------------- include/drm/drmP.h | 34 +---------------- include/drm/drm_os_linux.h | 16 -------- include/linux/io.h | 25 ++++++++++++ include/video/uvesafb.h | 1 + 23 files changed, 181 insertions(+), 268 deletions(-)Signature · 1 line
1.8.1.4
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch