Re: [PATCH] radeon: DMA unmap dummy page during unload/unbind.

From: Alex Deucher
Date: Thu Feb 24 2011 - 03:25:50 EST


On Wed, Feb 23, 2011 at 3:48 PM, Konrad Rzeszutek Wilk
<konrad.wilk@xxxxxxxxxx> wrote:
> git commit 82568565683b4991964a5fc89a9ca0c7122818e8 adds a dummy page
> so that if something goes wrong it will at least fetch/write data
> _only_ to this location.
>
> The radeon_gart_init sets it up.. but there is no corresponding
> free-ing of this dummy page. There is one invocation to free-ing
> the dummy page in r600.c but nothing in the radeon_gart_fini path.
>
> This patch adds in the unloading/unbind path the free-ing of said
> dummy page.

This patch looks good, but we might as go one step further and just
remove the dummy page stuff from r600.c, rv770.c and evergreen.c and
just consolidate it in gart_init() and gart_fini(). See the attached
patch.

Alex

>
> Without this patch and if we have CONFIG_DMA_API_DEBUG=y set we
> get this:
>
> [   22.113420] drm: unregistered panic notifier
> [   22.126500] [drm] radeon: finishing device.
> [   22.130868] [drm] radeon: cp finalized
> [   22.135384] [TTM] Finalizing pool allocator.
> [   22.139669] [TTM] Zone  kernel: Used memory at exit: 0 kiB.
> [   22.145229] [drm] radeon: ttm finalized
> [   22.149055] vga_switcheroo: disabled
> [   22.153393] ------------[ cut here ]------------
> [   22.158021] WARNING: at /home/konrad/ssd/linux/lib/dma-debug.c:689 dma_debug_device_change+0x12e/0x161()
> [   22.167475] Hardware name: PowerEdge T105
> [   22.171562] pci 0000:01:08.0: DMA-API: device driver has pending DMA allocations while released from device [count=1]
>
> With this patch we don't see this anymore.
>
> Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
> ---
>  drivers/gpu/drm/radeon/radeon_gart.c |    2 ++
>  1 files changed, 2 insertions(+), 0 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon_gart.c b/drivers/gpu/drm/radeon/radeon_gart.c
> index 6501611..dc04c7b 100644
> --- a/drivers/gpu/drm/radeon/radeon_gart.c
> +++ b/drivers/gpu/drm/radeon/radeon_gart.c
> @@ -269,4 +269,6 @@ void radeon_gart_fini(struct radeon_device *rdev)
>        kfree(rdev->gart.pages_addr);
>        rdev->gart.pages = NULL;
>        rdev->gart.pages_addr = NULL;
> +
> +       radeon_dummy_page_fini(rdev);
>  }
> --
> 1.7.1
>
> _______________________________________________
> dri-devel mailing list
> dri-devel@xxxxxxxxxxxxxxxxxxxxx
> http://lists.freedesktop.org/mailman/listinfo/dri-devel
>
From d22bba30a66ba95cabeed7a71c982dfa96521384 Mon Sep 17 00:00:00 2001
From: Alex Deucher <alexdeucher@xxxxxxxxx>
Date: Thu, 24 Feb 2011 03:16:21 -0500
Subject: [PATCH] drm/radeon/kms: clean up gart dummy page handling

As per Konrad's original patch, the dummy page used
by the gart code and allocated in radeon_gart_init()
was not freed properly in radeon_gart_fini().

At the same time r6xx and newer allocated and freed the
dummy page on their own. So to do Konrad's patch one
better, just remove the allocation and freeing of the
dummy page in the r6xx, 7xx, evergreen, and ni code and
allocate and free in the gart_init/fini() functions for
all asics.

Cc: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
Cc: Jerome Glisse <jglisse@xxxxxxxxxx>
Signed-off-by: Alex Deucher <alexdeucher@xxxxxxxxx>
---
drivers/gpu/drm/radeon/evergreen.c | 4 ----
drivers/gpu/drm/radeon/r600.c | 4 ----
drivers/gpu/drm/radeon/radeon_gart.c | 2 ++
drivers/gpu/drm/radeon/rv770.c | 4 ----
4 files changed, 2 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/radeon/evergreen.c b/drivers/gpu/drm/radeon/evergreen.c
index 789441e..9ecb2b6 100644
--- a/drivers/gpu/drm/radeon/evergreen.c
+++ b/drivers/gpu/drm/radeon/evergreen.c
@@ -3048,9 +3048,6 @@ int evergreen_init(struct radeon_device *rdev)
{
int r;

- r = radeon_dummy_page_init(rdev);
- if (r)
- return r;
/* This don't do much */
r = radeon_gem_init(rdev);
if (r)
@@ -3162,7 +3159,6 @@ void evergreen_fini(struct radeon_device *rdev)
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
- radeon_dummy_page_fini(rdev);
}

static void evergreen_pcie_gen2_enable(struct radeon_device *rdev)
diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c
index 1252198..88eaffc 100644
--- a/drivers/gpu/drm/radeon/r600.c
+++ b/drivers/gpu/drm/radeon/r600.c
@@ -2510,9 +2510,6 @@ int r600_init(struct radeon_device *rdev)
{
int r;

- r = radeon_dummy_page_init(rdev);
- if (r)
- return r;
if (r600_debugfs_mc_info_init(rdev)) {
DRM_ERROR("Failed to register debugfs file for mc !\n");
}
@@ -2626,7 +2623,6 @@ void r600_fini(struct radeon_device *rdev)
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
- radeon_dummy_page_fini(rdev);
}


diff --git a/drivers/gpu/drm/radeon/radeon_gart.c b/drivers/gpu/drm/radeon/radeon_gart.c
index 6501611..dc04c7b 100644
--- a/drivers/gpu/drm/radeon/radeon_gart.c
+++ b/drivers/gpu/drm/radeon/radeon_gart.c
@@ -269,4 +269,6 @@ void radeon_gart_fini(struct radeon_device *rdev)
kfree(rdev->gart.pages_addr);
rdev->gart.pages = NULL;
rdev->gart.pages_addr = NULL;
+
+ radeon_dummy_page_fini(rdev);
}
diff --git a/drivers/gpu/drm/radeon/rv770.c b/drivers/gpu/drm/radeon/rv770.c
index d8ba676..6a312e6 100644
--- a/drivers/gpu/drm/radeon/rv770.c
+++ b/drivers/gpu/drm/radeon/rv770.c
@@ -1256,9 +1256,6 @@ int rv770_init(struct radeon_device *rdev)
{
int r;

- r = radeon_dummy_page_init(rdev);
- if (r)
- return r;
/* This don't do much */
r = radeon_gem_init(rdev);
if (r)
@@ -1373,7 +1370,6 @@ void rv770_fini(struct radeon_device *rdev)
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
- radeon_dummy_page_fini(rdev);
}

static void rv770_pcie_gen2_enable(struct radeon_device *rdev)
--
1.7.1.1