* Re: [RFC/PATCH v2 0/8] Clean up write-combining MTRR addition
From: Daniel Vetter @ 2013-05-10 9:42 UTC (permalink / raw)
To: Andy Lutomirski
Cc: linux-kernel, dri-devel, linux-fbdev, Daniel Vetter,
Jerome Glisse, Alex Deucher, Dave Airlie
In-Reply-To: <cover.1368128020.git.luto@amacapital.net>
On Thu, May 09, 2013 at 12:46:19PM -0700, Andy Lutomirski wrote:
> 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
With my two comments addressed, the entire series is
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
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
>
> 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(-)
>
> --
> 1.8.1.4
>
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
^ permalink raw reply
* Re: [RFC/PATCH v2 5/8] i915: Use arch_phys_wc_{add,del}
From: Daniel Vetter @ 2013-05-10 9:36 UTC (permalink / raw)
To: Andy Lutomirski
Cc: linux-kernel, dri-devel, linux-fbdev, Daniel Vetter,
Jerome Glisse, Alex Deucher, Dave Airlie
In-Reply-To: <2d9af02b508eaf182c72b6e60a63070de265a759.1368128020.git.luto@amacapital.net>
On Thu, May 09, 2013 at 12:46:24PM -0700, Andy Lutomirski wrote:
> i915 open-coded logic that was essentially equivalent to the new API.
>
> Signed-off-by: Andy Lutomirski <luto@amacapital.net>
> ---
>
> Changes from v1: More cleanup
>
> drivers/gpu/drm/i915/i915_dma.c | 44 ++++++-----------------------------------
> 1 file changed, 6 insertions(+), 38 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
> index 4fa6beb..cfdfb45 100644
> --- a/drivers/gpu/drm/i915/i915_dma.c
> +++ b/drivers/gpu/drm/i915/i915_dma.c
> @@ -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;
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
> 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();
>
> --
> 1.8.1.4
>
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
^ permalink raw reply
* Re: [RFC/PATCH v2 1/8] Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed
From: Daniel Vetter @ 2013-05-10 9:19 UTC (permalink / raw)
To: Andy Lutomirski
Cc: linux-kernel, dri-devel, linux-fbdev, Daniel Vetter,
Jerome Glisse, Alex Deucher, Dave Airlie
In-Reply-To: <97346b603822ffd8ab446316fc08fbf7686c9d50.1368128020.git.luto@amacapital.net>
On Thu, May 09, 2013 at 12:46:20PM -0700, Andy Lutomirski wrote:
> 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.
>
> Signed-off-by: Andy Lutomirski <luto@amacapital.net>
> ---
> 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(-)
>
> diff --git a/arch/x86/include/asm/io.h b/arch/x86/include/asm/io.h
> index d8e8eef..34f69cb 100644
> --- a/arch/x86/include/asm/io.h
> +++ b/arch/x86/include/asm/io.h
> @@ -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 */
> diff --git a/arch/x86/include/asm/mtrr.h b/arch/x86/include/asm/mtrr.h
> index e235582..10d0fba 100644
> --- a/arch/x86/include/asm/mtrr.h
> +++ b/arch/x86/include/asm/mtrr.h
> @@ -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 *);
> diff --git a/arch/x86/kernel/cpu/mtrr/main.c b/arch/x86/kernel/cpu/mtrr/main.c
> index 726bf96..23bd49a 100644
> --- a/arch/x86/kernel/cpu/mtrr/main.c
> +++ b/arch/x86/kernel/cpu/mtrr/main.c
> @@ -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;
> +
> + 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
> +
> + 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
> diff --git a/include/linux/io.h b/include/linux/io.h
> index 069e407..f4f42fa 100644
> --- a/include/linux/io.h
> +++ b/include/linux/io.h
> @@ -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 */
> --
> 1.8.1.4
>
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
^ permalink raw reply
* [RFC 34/42] drivers/video/omap2/dss: don't check resource with devm_ioremap_resource
From: Wolfram Sang @ 2013-05-10 8:17 UTC (permalink / raw)
To: linux-kernel
Cc: Wolfram Sang, Tomi Valkeinen, Florian Tobias Schandinat,
linux-omap, linux-fbdev
In-Reply-To: <1368173847-5661-1-git-send-email-wsa@the-dreams.de>
devm_ioremap_resource does sanity checks on the given resource. No need to
duplicate this in the driver.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
---
drivers/video/omap2/dss/hdmi.c | 7 +------
1 file changed, 1 insertion(+), 6 deletions(-)
diff --git a/drivers/video/omap2/dss/hdmi.c b/drivers/video/omap2/dss/hdmi.c
index 17f4d55..7bc6762 100644
--- a/drivers/video/omap2/dss/hdmi.c
+++ b/drivers/video/omap2/dss/hdmi.c
@@ -1064,13 +1064,8 @@ static int omapdss_hdmihw_probe(struct platform_device *pdev)
mutex_init(&hdmi.lock);
mutex_init(&hdmi.ip_data.lock);
- res = platform_get_resource(hdmi.pdev, IORESOURCE_MEM, 0);
- if (!res) {
- DSSERR("can't get IORESOURCE_MEM HDMI\n");
- return -EINVAL;
- }
-
/* Base address taken from platform */
+ res = platform_get_resource(hdmi.pdev, IORESOURCE_MEM, 0);
hdmi.ip_data.base_wp = devm_ioremap_resource(&pdev->dev, res);
if (IS_ERR(hdmi.ip_data.base_wp))
return PTR_ERR(hdmi.ip_data.base_wp);
--
1.7.10.4
^ permalink raw reply related
* [RFC 33/42] drivers/video/omap2: don't check resource with devm_ioremap_resource
From: Wolfram Sang @ 2013-05-10 8:17 UTC (permalink / raw)
To: linux-kernel
Cc: Wolfram Sang, Tomi Valkeinen, Florian Tobias Schandinat,
linux-omap, linux-fbdev
In-Reply-To: <1368173847-5661-1-git-send-email-wsa@the-dreams.de>
devm_ioremap_resource does sanity checks on the given resource. No need to
duplicate this in the driver.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
---
drivers/video/omap2/vrfb.c | 5 -----
1 file changed, 5 deletions(-)
diff --git a/drivers/video/omap2/vrfb.c b/drivers/video/omap2/vrfb.c
index 5261229..f346b02 100644
--- a/drivers/video/omap2/vrfb.c
+++ b/drivers/video/omap2/vrfb.c
@@ -353,11 +353,6 @@ static int __init vrfb_probe(struct platform_device *pdev)
/* first resource is the register res, the rest are vrfb contexts */
mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
- if (!mem) {
- dev_err(&pdev->dev, "can't get vrfb base address\n");
- return -EINVAL;
- }
-
vrfb_base = devm_ioremap_resource(&pdev->dev, mem);
if (IS_ERR(vrfb_base))
return PTR_ERR(vrfb_base);
--
1.7.10.4
^ permalink raw reply related
* [RFC 32/42] drivers/video: don't check resource with devm_ioremap_resource
From: Wolfram Sang @ 2013-05-10 8:17 UTC (permalink / raw)
To: linux-kernel; +Cc: Wolfram Sang, Florian Tobias Schandinat, linux-fbdev
In-Reply-To: <1368173847-5661-1-git-send-email-wsa@the-dreams.de>
devm_ioremap_resource does sanity checks on the given resource. No need to
duplicate this in the driver.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
---
drivers/video/mxsfb.c | 7 +------
1 file changed, 1 insertion(+), 6 deletions(-)
diff --git a/drivers/video/mxsfb.c b/drivers/video/mxsfb.c
index 1b2c26d..3dc3265 100644
--- a/drivers/video/mxsfb.c
+++ b/drivers/video/mxsfb.c
@@ -883,12 +883,6 @@ static int mxsfb_probe(struct platform_device *pdev)
if (of_id)
pdev->id_entry = of_id->data;
- res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
- if (!res) {
- dev_err(&pdev->dev, "Cannot get memory IO resource\n");
- return -ENODEV;
- }
-
fb_info = framebuffer_alloc(sizeof(struct mxsfb_info), &pdev->dev);
if (!fb_info) {
dev_err(&pdev->dev, "Failed to allocate fbdev\n");
@@ -897,6 +891,7 @@ static int mxsfb_probe(struct platform_device *pdev)
host = to_imxfb_host(fb_info);
+ res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
host->base = devm_ioremap_resource(&pdev->dev, res);
if (IS_ERR(host->base)) {
dev_err(&pdev->dev, "ioremap failed\n");
--
1.7.10.4
^ permalink raw reply related
* Re: [RFC/PATCH v2 0/8] Clean up write-combining MTRR addition
From: Andy Lutomirski @ 2013-05-10 1:21 UTC (permalink / raw)
To: Jerome Glisse
Cc: linux-kernel, dri-devel, linux-fbdev, Daniel Vetter, Alex Deucher,
Dave Airlie
In-Reply-To: <CAH3drwa_ypKKqZ8eayeR4fA4Ounj55R3R9y0_b5crp99YMFQ=w@mail.gmail.com>
On Thu, May 9, 2013 at 4:44 PM, Jerome Glisse <j.glisse@gmail.com> wrote:
> On Thu, May 9, 2013 at 3:46 PM, Andy Lutomirski <luto@amacapital.net> wrote:
>> 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 or
DRM_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
^ permalink raw reply
* Re: [RFC/PATCH v2 0/8] Clean up write-combining MTRR addition
From: Jerome Glisse @ 2013-05-09 23:44 UTC (permalink / raw)
To: Andy Lutomirski; +Cc: linux-fbdev, Daniel Vetter, linux-kernel, dri-devel
In-Reply-To: <cover.1368128020.git.luto@amacapital.net>
On Thu, May 9, 2013 at 3:46 PM, Andy Lutomirski <luto@amacapital.net> wrote:
> 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
> 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(-)
>
> --
> 1.8.1.4
>
^ permalink raw reply
* [PATCH 01/12] fbdev: FB_GOLDFISH should depend on HAS_DMA
From: Geert Uytterhoeven @ 2013-05-09 21:04 UTC (permalink / raw)
To: linux-kernel; +Cc: Geert Uytterhoeven, Florian Tobias Schandinat, linux-fbdev
If NO_DMA=y:
drivers/built-in.o: In function `goldfish_fb_remove':
drivers/video/goldfishfb.c:301: undefined reference to `dma_free_coherent'
drivers/built-in.o: In function `goldfish_fb_probe':
drivers/video/goldfishfb.c:247: undefined reference to `dma_alloc_coherent'
drivers/video/goldfishfb.c:280: undefined reference to `dma_free_coherent'
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>
Cc: linux-fbdev@vger.kernel.org
---
drivers/video/Kconfig | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/drivers/video/Kconfig b/drivers/video/Kconfig
index 9669c9c..da01cec 100644
--- a/drivers/video/Kconfig
+++ b/drivers/video/Kconfig
@@ -2199,7 +2199,7 @@ config FB_XILINX
config FB_GOLDFISH
tristate "Goldfish Framebuffer"
- depends on FB
+ depends on FB && HAS_DMA
select FB_CFB_FILLRECT
select FB_CFB_COPYAREA
select FB_CFB_IMAGEBLIT
--
1.7.0.4
^ permalink raw reply related
* [RFC/PATCH v2 8/8] drm: Remove mtrr_add and mtrr_del fallback hack for non-MTRR systems
From: Andy Lutomirski @ 2013-05-09 19:46 UTC (permalink / raw)
To: linux-kernel, dri-devel, linux-fbdev
Cc: Daniel Vetter, Jerome Glisse, Alex Deucher, Dave Airlie,
Andy Lutomirski
In-Reply-To: <cover.1368128020.git.luto@amacapital.net>
There are no users left in drivers/gpu.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---
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(-)
diff --git a/include/drm/drmP.h b/include/drm/drmP.h
index 3e6cfa0..7a9fef5 100644
--- a/include/drm/drmP.h
+++ b/include/drm/drmP.h
@@ -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>
diff --git a/include/drm/drm_os_linux.h b/include/drm/drm_os_linux.h
index 3933691..35c7c2b 100644
--- a/include/drm/drm_os_linux.h
+++ b/include/drm/drm_os_linux.h
@@ -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)
--
1.8.1.4
^ permalink raw reply related
* [RFC/PATCH v2 7/8] uvesafb: Clean up MTRR code
From: Andy Lutomirski @ 2013-05-09 19:46 UTC (permalink / raw)
To: linux-kernel, dri-devel, linux-fbdev
Cc: Daniel Vetter, Jerome Glisse, Alex Deucher, Dave Airlie,
Andy Lutomirski, Michal Januszewski
In-Reply-To: <cover.1368128020.git.luto@amacapital.net>
The 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.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---
Documentation/fb/uvesafb.txt | 16 ++++------
drivers/video/uvesafb.c | 70 +++++++++++---------------------------------
include/video/uvesafb.h | 1 +
3 files changed, 23 insertions(+), 64 deletions(-)
diff --git a/Documentation/fb/uvesafb.txt b/Documentation/fb/uvesafb.txt
index eefdd91..f6362d8 100644
--- a/Documentation/fb/uvesafb.txt
+++ b/Documentation/fb/uvesafb.txt
@@ -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.
diff --git a/drivers/video/uvesafb.c b/drivers/video/uvesafb.c
index d428445..8701f96 100644
--- a/drivers/video/uvesafb.c
+++ b/drivers/video/uvesafb.c
@@ -24,9 +24,6 @@
#ifdef CONFIG_X86
#include <video/vga.h>
#endif
-#ifdef CONFIG_MTRR
-#include <asm/mtrr.h>
-#endif
#include "edid.h"
static struct cb_id uvesafb_cn_id = {
@@ -1540,67 +1537,30 @@ static void uvesafb_init_info(struct fb_info *info, struct vbe_mode_ib *mode)
static void uvesafb_init_mtrr(struct fb_info *info)
{
-#ifdef CONFIG_MTRR
+ struct uvesafb_par *par = info->par;
+
if (mtrr && !(info->fix.smem_start & (PAGE_SIZE - 1))) {
int temp_size = info->fix.smem_len;
- unsigned int type = 0;
- switch (mtrr) {
- case 1:
- type = MTRR_TYPE_UNCACHABLE;
- break;
- case 2:
- type = MTRR_TYPE_WRBACK;
- break;
- case 3:
- type = MTRR_TYPE_WRCOMB;
- break;
- case 4:
- type = MTRR_TYPE_WRTHROUGH;
- break;
- default:
- type = 0;
- break;
- }
+ int rc;
- if (type) {
- int rc;
+ /* Find the largest power-of-two */
+ temp_size = roundup_pow_of_two(temp_size);
- /* Find the largest power-of-two */
- temp_size = roundup_pow_of_two(temp_size);
+ /* Try and find a power of two to add */
+ do {
+ rc = arch_phys_wc_add(info->fix.smem_start, temp_size);
+ temp_size >>= 1;
+ } while (temp_size >= PAGE_SIZE && rc = -EINVAL);
- /* Try and find a power of two to add */
- do {
- rc = mtrr_add(info->fix.smem_start,
- temp_size, type, 1);
- temp_size >>= 1;
- } while (temp_size >= PAGE_SIZE && rc = -EINVAL);
- }
+ if (rc >= 0)
+ par->mtrr_handle = rc;
}
-#endif /* CONFIG_MTRR */
}
static void uvesafb_ioremap(struct fb_info *info)
{
-#ifdef CONFIG_X86
- switch (mtrr) {
- case 1: /* uncachable */
- info->screen_base = ioremap_nocache(info->fix.smem_start, info->fix.smem_len);
- break;
- case 2: /* write-back */
- info->screen_base = ioremap_cache(info->fix.smem_start, info->fix.smem_len);
- break;
- case 3: /* write-combining */
- info->screen_base = ioremap_wc(info->fix.smem_start, info->fix.smem_len);
- break;
- case 4: /* write-through */
- default:
- info->screen_base = ioremap(info->fix.smem_start, info->fix.smem_len);
- break;
- }
-#else
- info->screen_base = ioremap(info->fix.smem_start, info->fix.smem_len);
-#endif /* CONFIG_X86 */
+ info->screen_base = ioremap_wc(info->fix.smem_start, info->fix.smem_len);
}
static ssize_t uvesafb_show_vbe_ver(struct device *dev,
@@ -1851,6 +1811,7 @@ static int uvesafb_remove(struct platform_device *dev)
unregister_framebuffer(info);
release_region(0x3c0, 32);
iounmap(info->screen_base);
+ arch_phys_wc_del(par->mtrr_handle);
release_mem_region(info->fix.smem_start, info->fix.smem_len);
fb_destroy_modedb(info->monspecs.modedb);
fb_dealloc_cmap(&info->cmap);
@@ -1930,6 +1891,9 @@ static int uvesafb_setup(char *options)
}
}
+ if (mtrr != 3 && mtrr != 1)
+ pr_warn("uvesafb: mtrr should be set to 0 or 3; %d is unsupported", mtrr);
+
return 0;
}
#endif /* !MODULE */
diff --git a/include/video/uvesafb.h b/include/video/uvesafb.h
index 1a91850..30f5362 100644
--- a/include/video/uvesafb.h
+++ b/include/video/uvesafb.h
@@ -134,6 +134,7 @@ struct uvesafb_par {
int mode_idx;
struct vbe_crtc_ib crtc;
+ int mtrr_handle;
};
#endif /* _UVESAFB_H */
--
1.8.1.4
^ permalink raw reply related
* [RFC/PATCH v2 6/8] radeon: Switch to arch_phys_wc_add and add a missing ..._del
From: Andy Lutomirski @ 2013-05-09 19:46 UTC (permalink / raw)
To: linux-kernel, dri-devel, linux-fbdev
Cc: Daniel Vetter, Jerome Glisse, Alex Deucher, Dave Airlie,
Andy Lutomirski
In-Reply-To: <cover.1368128020.git.luto@amacapital.net>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---
drivers/gpu/drm/radeon/radeon_object.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index d3aface..15cd34b 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -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,
--
1.8.1.4
^ permalink raw reply related
* [RFC/PATCH v2 5/8] i915: Use arch_phys_wc_{add,del}
From: Andy Lutomirski @ 2013-05-09 19:46 UTC (permalink / raw)
To: linux-kernel, dri-devel, linux-fbdev
Cc: Daniel Vetter, Jerome Glisse, Alex Deucher, Dave Airlie,
Andy Lutomirski
In-Reply-To: <cover.1368128020.git.luto@amacapital.net>
i915 open-coded logic that was essentially equivalent to the new API.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---
Changes from v1: More cleanup
drivers/gpu/drm/i915/i915_dma.c | 44 ++++++-----------------------------------
1 file changed, 6 insertions(+), 38 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
index 4fa6beb..cfdfb45 100644
--- a/drivers/gpu/drm/i915/i915_dma.c
+++ b/drivers/gpu/drm/i915/i915_dma.c
@@ -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();
--
1.8.1.4
^ permalink raw reply related
* [RFC/PATCH v2 4/8] drm,agpgart: Use pgprot_writecombine for AGP maps and make the MTRR optional
From: Andy Lutomirski @ 2013-05-09 19:46 UTC (permalink / raw)
To: linux-kernel, dri-devel, linux-fbdev
Cc: Daniel Vetter, Jerome Glisse, Alex Deucher, Dave Airlie,
Andy Lutomirski
In-Reply-To: <cover.1368128020.git.luto@amacapital.net>
I'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.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---
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(-)
diff --git a/drivers/char/agp/frontend.c b/drivers/char/agp/frontend.c
index 2e04433..1b19239 100644
--- a/drivers/char/agp/frontend.c
+++ b/drivers/char/agp/frontend.c
@@ -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));
diff --git a/drivers/gpu/drm/drm_pci.c b/drivers/gpu/drm/drm_pci.c
index bd719e9..d0f6699 100644
--- a/drivers/gpu/drm/drm_pci.c
+++ b/drivers/gpu/drm/drm_pci.c
@@ -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;
diff --git a/drivers/gpu/drm/drm_stub.c b/drivers/gpu/drm/drm_stub.c
index 7d30802..9e2acdf 100644
--- a/drivers/gpu/drm/drm_stub.c
+++ b/drivers/gpu/drm/drm_stub.c
@@ -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);
diff --git a/drivers/gpu/drm/drm_vm.c b/drivers/gpu/drm/drm_vm.c
index 163f436..b1e4ec8 100644
--- a/drivers/gpu/drm/drm_vm.c
+++ b/drivers/gpu/drm/drm_vm.c
@@ -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)
--
1.8.1.4
^ permalink raw reply related
* [RFC/PATCH v2 3/8] drm: Update drm_addmap and drm_mmap to use PAT WC instead of MTRRs
From: Andy Lutomirski @ 2013-05-09 19:46 UTC (permalink / raw)
To: linux-kernel, dri-devel, linux-fbdev
Cc: Daniel Vetter, Jerome Glisse, Alex Deucher, Dave Airlie,
Andy Lutomirski
In-Reply-To: <cover.1368128020.git.luto@amacapital.net>
Previously, 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.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---
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(-)
diff --git a/drivers/gpu/drm/drm_bufs.c b/drivers/gpu/drm/drm_bufs.c
index 0128147..0190fce 100644
--- a/drivers/gpu/drm/drm_bufs.c
+++ b/drivers/gpu/drm/drm_bufs.c
@@ -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);
diff --git a/drivers/gpu/drm/drm_vm.c b/drivers/gpu/drm/drm_vm.c
index db7bd29..163f436 100644
--- a/drivers/gpu/drm/drm_vm.c
+++ b/drivers/gpu/drm/drm_vm.c
@@ -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,
--
1.8.1.4
^ permalink raw reply related
* [RFC/PATCH v2 2/8] drm (ast,cirrus,mgag200,nouveau,savage,vmwgfx): Remove drm_mtrr_{add,del}
From: Andy Lutomirski @ 2013-05-09 19:46 UTC (permalink / raw)
To: linux-kernel, dri-devel, linux-fbdev
Cc: Daniel Vetter, Jerome Glisse, Alex Deucher, Dave Airlie,
Andy Lutomirski
In-Reply-To: <cover.1368128020.git.luto@amacapital.net>
This 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.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---
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(-)
diff --git a/drivers/gpu/drm/ast/ast_ttm.c b/drivers/gpu/drm/ast/ast_ttm.c
index 3602731..c4574fd 100644
--- a/drivers/gpu/drm/ast/ast_ttm.c
+++ b/drivers/gpu/drm/ast/ast_ttm.c
@@ -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)
diff --git a/drivers/gpu/drm/cirrus/cirrus_ttm.c b/drivers/gpu/drm/cirrus/cirrus_ttm.c
index 1413a26..09f06d3 100644
--- a/drivers/gpu/drm/cirrus/cirrus_ttm.c
+++ b/drivers/gpu/drm/cirrus/cirrus_ttm.c
@@ -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)
diff --git a/drivers/gpu/drm/mgag200/mgag200_ttm.c b/drivers/gpu/drm/mgag200/mgag200_ttm.c
index 8fc9d92..5c6f3c8 100644
--- a/drivers/gpu/drm/mgag200/mgag200_ttm.c
+++ b/drivers/gpu/drm/mgag200/mgag200_ttm.c
@@ -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)
diff --git a/drivers/gpu/drm/nouveau/nouveau_ttm.c b/drivers/gpu/drm/nouveau/nouveau_ttm.c
index 9be9cb5..63fa6a5 100644
--- a/drivers/gpu/drm/nouveau/nouveau_ttm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_ttm.c
@@ -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;
}
diff --git a/drivers/gpu/drm/savage/savage_bci.c b/drivers/gpu/drm/savage/savage_bci.c
index b55c1d6..bd6b2cf 100644
--- a/drivers/gpu/drm/savage/savage_bci.c
+++ b/drivers/gpu/drm/savage/savage_bci.c
@@ -570,9 +570,6 @@ int savage_driver_firstopen(struct drm_device *dev)
unsigned int fb_rsrc, aper_rsrc;
int ret = 0;
- dev_priv->mtrr[0].handle = -1;
- dev_priv->mtrr[1].handle = -1;
- dev_priv->mtrr[2].handle = -1;
if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
fb_rsrc = 0;
fb_base = pci_resource_start(dev->pdev, 0);
@@ -584,21 +581,14 @@ int savage_driver_firstopen(struct drm_device *dev)
if (pci_resource_len(dev->pdev, 0) = 0x08000000) {
/* Don't make MMIO write-cobining! We need 3
* MTRRs. */
- dev_priv->mtrr[0].base = fb_base;
- dev_priv->mtrr[0].size = 0x01000000;
- dev_priv->mtrr[0].handle - drm_mtrr_add(dev_priv->mtrr[0].base,
- dev_priv->mtrr[0].size, DRM_MTRR_WC);
- dev_priv->mtrr[1].base = fb_base + 0x02000000;
- dev_priv->mtrr[1].size = 0x02000000;
- dev_priv->mtrr[1].handle - drm_mtrr_add(dev_priv->mtrr[1].base,
- dev_priv->mtrr[1].size, DRM_MTRR_WC);
- dev_priv->mtrr[2].base = fb_base + 0x04000000;
- dev_priv->mtrr[2].size = 0x04000000;
- dev_priv->mtrr[2].handle - drm_mtrr_add(dev_priv->mtrr[2].base,
- dev_priv->mtrr[2].size, DRM_MTRR_WC);
+ dev_priv->mtrr_handles[0] + arch_phys_wc_add(fb_base, 0x01000000);
+ dev_priv->mtrr_handles[1] + arch_phys_wc_add(fb_base + 0x02000000,
+ 0x02000000);
+ dev_priv->mtrr_handles[2] + arch_phys_wc_add(fb_base + 0x04000000,
+ 0x04000000);
} else {
DRM_ERROR("strange pci_resource_len %08llx\n",
(unsigned long long)
@@ -616,11 +606,9 @@ int savage_driver_firstopen(struct drm_device *dev)
if (pci_resource_len(dev->pdev, 1) = 0x08000000) {
/* Can use one MTRR to cover both fb and
* aperture. */
- dev_priv->mtrr[0].base = fb_base;
- dev_priv->mtrr[0].size = 0x08000000;
- dev_priv->mtrr[0].handle - drm_mtrr_add(dev_priv->mtrr[0].base,
- dev_priv->mtrr[0].size, DRM_MTRR_WC);
+ dev_priv->mtrr_handles[0] + arch_phys_wc_add(fb_base,
+ 0x08000000);
} else {
DRM_ERROR("strange pci_resource_len %08llx\n",
(unsigned long long)
@@ -660,11 +648,10 @@ void savage_driver_lastclose(struct drm_device *dev)
drm_savage_private_t *dev_priv = dev->dev_private;
int i;
- for (i = 0; i < 3; ++i)
- if (dev_priv->mtrr[i].handle >= 0)
- drm_mtrr_del(dev_priv->mtrr[i].handle,
- dev_priv->mtrr[i].base,
- dev_priv->mtrr[i].size, DRM_MTRR_WC);
+ for (i = 0; i < 3; ++i) {
+ arch_phys_wc_del(dev_priv->mtrr_handles[i]);
+ dev_priv->mtrr_handles[i] = 0;
+ }
}
int savage_driver_unload(struct drm_device *dev)
diff --git a/drivers/gpu/drm/savage/savage_drv.h b/drivers/gpu/drm/savage/savage_drv.h
index df2aac6..c05082a 100644
--- a/drivers/gpu/drm/savage/savage_drv.h
+++ b/drivers/gpu/drm/savage/savage_drv.h
@@ -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;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 07dfd82..78e2164 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -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);
diff --git a/include/drm/drmP.h b/include/drm/drmP.h
index 2d94d74..3e6cfa0 100644
--- a/include/drm/drmP.h
+++ b/include/drm/drmP.h
@@ -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)
--
1.8.1.4
^ permalink raw reply related
* [RFC/PATCH v2 1/8] Add arch_phys_wc_{add,del} to manipulate WC MTRRs if needed
From: Andy Lutomirski @ 2013-05-09 19:46 UTC (permalink / raw)
To: linux-kernel, dri-devel, linux-fbdev
Cc: Daniel Vetter, Jerome Glisse, Alex Deucher, Dave Airlie,
Andy Lutomirski
In-Reply-To: <cover.1368128020.git.luto@amacapital.net>
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.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---
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(-)
diff --git a/arch/x86/include/asm/io.h b/arch/x86/include/asm/io.h
index d8e8eef..34f69cb 100644
--- a/arch/x86/include/asm/io.h
+++ b/arch/x86/include/asm/io.h
@@ -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 */
diff --git a/arch/x86/include/asm/mtrr.h b/arch/x86/include/asm/mtrr.h
index e235582..10d0fba 100644
--- a/arch/x86/include/asm/mtrr.h
+++ b/arch/x86/include/asm/mtrr.h
@@ -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 *);
diff --git a/arch/x86/kernel/cpu/mtrr/main.c b/arch/x86/kernel/cpu/mtrr/main.c
index 726bf96..23bd49a 100644
--- a/arch/x86/kernel/cpu/mtrr/main.c
+++ b/arch/x86/kernel/cpu/mtrr/main.c
@@ -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;
+
+ if (pat_enabled)
+ return 0; /* Success! (We don't need to do anything.) */
+
+ 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
diff --git a/include/linux/io.h b/include/linux/io.h
index 069e407..f4f42fa 100644
--- a/include/linux/io.h
+++ b/include/linux/io.h
@@ -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 */
--
1.8.1.4
^ permalink raw reply related
* [RFC/PATCH v2 0/8] Clean up write-combining MTRR addition
From: Andy Lutomirski @ 2013-05-09 19:46 UTC (permalink / raw)
To: linux-kernel, dri-devel, linux-fbdev
Cc: Daniel Vetter, Jerome Glisse, Alex Deucher, Dave Airlie,
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_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(-)
--
1.8.1.4
^ permalink raw reply
* Re: [PATCH V2] video: implement a simple framebuffer driver
From: Rob Landley @ 2013-05-08 20:58 UTC (permalink / raw)
To: linux-arm-kernel
In-Reply-To: <5189BA36.9000105@wwwdotorg.org>
On 05/07/2013 09:36:38 PM, Stephen Warren wrote:
> On 05/07/2013 03:33 PM, Andrew Morton wrote:
...
> Subject: drivers/video: implement a simple framebuffer driver
>
> A simple frame-buffer describes a raw memory region that may be
> rendered
> to, with the assumption that the display hardware has already been set
> up to scan out from that buffer.
>
> This is useful in cases where a bootloader exists and has set up the
> display hardware, but a Linux driver doesn't yet exist for the display
> hardware.
>
> Examples use-cases include:
Virtual environments such as QEMU.
Rob
^ permalink raw reply
* Re: [PATCH V2] video: implement a simple framebuffer driver
From: Olof Johansson @ 2013-05-08 19:28 UTC (permalink / raw)
To: linux-arm-kernel
In-Reply-To: <5189BA36.9000105@wwwdotorg.org>
On Tue, May 7, 2013 at 7:36 PM, Stephen Warren <swarren@wwwdotorg.org> wrote:
> On 05/07/2013 03:33 PM, Andrew Morton wrote:
>>
>> We don't seem to have a well-defined path to travel here, and I don't
>> get the feeling that anyone has signed up to walk it?
>>
>> So I'm inclined to merge Stephen's patch as-is into 3.10. It's pretty
>> simple and standalone. Any strenuous objections?
>>
>> BTW, Olof told me off-list that this patch is a "critical piece" for
>> running mainline Linux on ARM Chromebooks. That was important
>> information - more important than anything else we were told about this
>> patch!
>>
>> So Stephen, could I please have a new changelog for this patch which
>> explains this application, and any other interesting/important things
>> which were left out? Quickly, please.
>
> OK, how about the version below:
>
> ----------
> Subject: drivers/video: implement a simple framebuffer driver
>
> A simple frame-buffer describes a raw memory region that may be rendered
> to, with the assumption that the display hardware has already been set
> up to scan out from that buffer.
>
> This is useful in cases where a bootloader exists and has set up the
> display hardware, but a Linux driver doesn't yet exist for the display
> hardware.
>
> Examples use-cases include:
>
> * The built-in LCD panels on the Samsung ARM chromebook, and Tegra
> devices, and likely many other ARM or embedded systems. These cannot yet
> be supported using a full graphics driver, since the panel control
> should be provided by the CDF (Common Display Framework), which has been
> stuck in design/review for quite some time. One could support these
> panels using custom SoC-specific code, but there is a desire to use
> common infra-structure rather than having each SoC vendor invent their
> own code, hence the desire to wait for CDF.
>
> * Hardware for which a full graphics driver is not yet available, and
> the path to obtain one upstream isn't yet clear. For example, the
> Raspberry Pi.
>
> * Any hardware in early stages of upstreaming, before a full graphics
> driver has been tackled. This driver can provide a graphical boot
> console (even full X support) much earlier in the upstreaming process,
> thus making new SoC or board support more generally useful earlier.
> ----------
>
> I assume you can just edit the patch you have and don't need me to
> resend with the adjusted description.
Oh, looks like it never got a:
Acked-by: Olof Johansson <olof@lixom.net>
Either. Feel free to add, Andrew.
-Olof
^ permalink raw reply
* [PATCH v5, part3 07/15] mm, acornfb: use free_reserved_area() to simplify code
From: Jiang Liu @ 2013-05-08 15:17 UTC (permalink / raw)
To: Andrew Morton
Cc: Jiang Liu, David Rientjes, Wen Congyang, Mel Gorman, Minchan Kim,
KAMEZAWA Hiroyuki, Michal Hocko, James Bottomley, Sergei Shtylyov,
David Howells, Mark Salter, Jianguo Wu, linux-mm, linux-arch,
linux-kernel, Florian Tobias Schandinat, linux-fbdev
In-Reply-To: <1368026235-5976-1-git-send-email-jiang.liu@huawei.com>
Use common help function free_reserved_area() to simplify code.
Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>
Cc: linux-fbdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
---
drivers/video/acornfb.c | 28 ++--------------------------
1 file changed, 2 insertions(+), 26 deletions(-)
diff --git a/drivers/video/acornfb.c b/drivers/video/acornfb.c
index 6488a73..344f2bb 100644
--- a/drivers/video/acornfb.c
+++ b/drivers/video/acornfb.c
@@ -1188,32 +1188,8 @@ static int acornfb_detect_monitortype(void)
static inline void
free_unused_pages(unsigned int virtual_start, unsigned int virtual_end)
{
- int mb_freed = 0;
-
- /*
- * Align addresses
- */
- virtual_start = PAGE_ALIGN(virtual_start);
- virtual_end = PAGE_ALIGN(virtual_end);
-
- while (virtual_start < virtual_end) {
- struct page *page;
-
- /*
- * Clear page reserved bit,
- * set count to 1, and free
- * the page.
- */
- page = virt_to_page(virtual_start);
- ClearPageReserved(page);
- init_page_count(page);
- free_page(virtual_start);
-
- virtual_start += PAGE_SIZE;
- mb_freed += PAGE_SIZE / 1024;
- }
-
- printk("acornfb: freed %dK memory\n", mb_freed);
+ free_reserved_area(virtual_start, PAGE_ALIGN(virtual_end),
+ -1, "acornfb");
}
static int acornfb_probe(struct platform_device *dev)
--
1.7.9.5
^ permalink raw reply related
* Re: [PATCH 1/4] ARM: cfa10036: dt: Change i2c0 clock frequency
From: Shawn Guo @ 2013-05-08 14:01 UTC (permalink / raw)
To: linux-arm-kernel
In-Reply-To: <1366624945-5708-2-git-send-email-maxime.ripard@free-electrons.com>
On Mon, Apr 22, 2013 at 12:02:22PM +0200, Maxime Ripard wrote:
> The OLED display can work faster. Change the i2c controller clock
> frequency to remove the tearing effect that can be seen on the display.
>
> Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Applied, thanks.
^ permalink raw reply
* Re: [PATCH 2/2] ARM: dts: cfa10036: Change the OLED display to SSD1306
From: Shawn Guo @ 2013-05-08 14:00 UTC (permalink / raw)
To: linux-arm-kernel
In-Reply-To: <1366624555-5616-3-git-send-email-maxime.ripard@free-electrons.com>
On Mon, Apr 22, 2013 at 11:55:55AM +0200, Maxime Ripard wrote:
> The SSD1307 was used in an early prototype that will never get
> distributed. The final board now has a SSD1306 instead, that has its own
> power generation unit and thus doesn't need any PWM. The panel attached
> to it also changed.
>
> Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Applied, thanks.
^ permalink raw reply
* dringender Vorschlag
From: John P. Goldman @ 2013-05-08 8:46 UTC (permalink / raw)
To: linux-fbdev
Entschuldigungen für kommen in Ihre Privatsphäre! Ich bin Rechtsanwalt
Werner Erich Zeller; Ich habe einen meiner einflussreichen und
wohlhabenden Kunden zum Tode; und er hatte eine sehr geheime und private
Investitionen von €15,000,000.00 bei einer privaten Bank in Großbritannien
hier zu Lebzeiten. Diese Investition wurde ohne einen deklarierten
nächsten Angehörigen und begünstigte. Jetzt brauche ich Sie arbeiten mit
mir als mein Partner zu erholen und zu je 50 % Aktienfonds. Alle Dokumente
werden rechtlich beantragt und beschafft, und in 5 Werktage, wird diese
Transaktion auftreten. Aber ich brauche einen ernsten, treuen und
glaubwürdigen Partner.
Bitte senden Sie mir eine vertrauliche Antwort, wenn Sie denken, Sie
vertraut werden können und sind von den Qualitäten! Ich warte auf Ihre
schnelle Antwort.
Werner Erich Zeller (Rechtsanwalt)
Rufen Sie + 44-702-409-0820 (Office)
^ permalink raw reply
* Re: [PATCH 0/4] drivers: video: Optimisations for the ssd1307fb driver
From: Tomi Valkeinen @ 2013-05-08 6:57 UTC (permalink / raw)
To: linux-arm-kernel
In-Reply-To: <1366624945-5708-1-git-send-email-maxime.ripard@free-electrons.com>
On 22/04/13 13:02, Maxime Ripard wrote:
> Hi everyone,
>
> This is a patchset to improve significantly (by an order of magnitude)
> the refreshing time of the SSD1306 and SSD1307.
>
> We do so by sending the pixels by batches, instead of 8 at a time,
> combined with some additionnal features found on these controllers
> that allows to send a whole screen content at once.
>
> This obviously removes the tearing effect that was previously seen
> with these controllers.
>
> This patchset depends on the "Add support for the Solomon SSD1306 OLED
> Controller" I sent previously.
>
> Thanks,
> Maxime
>
> Maxime Ripard (4):
> ARM: cfa10036: dt: Change i2c0 clock frequency
> ssd1307fb: Rework the communication functions
> ssd1307fb: Speed up the communication with the controller
> ssd1307fb: Make use of horizontal addressing mode
>
> arch/arm/boot/dts/imx28-cfa10036.dts | 1 +
> drivers/video/ssd1307fb.c | 123 ++++++++++++++++++++++++----------
> 2 files changed, 89 insertions(+), 35 deletions(-)
>
I've taken the patches 2-4 into fbdev 3.11 branch. The first patch "ARM:
cfa10036: dt: Change i2c0 clock frequency" should preferably go through
arch tree.
Tomi
^ permalink raw reply
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox