Skip to contentSkip to conversation

drm/mgag200: Workaround HW bug with non-0 offset

PATCH In review Thomas Zimmermann 4 participants 12 messages Nov 26, 2019 – Dec 6, 2019

Subsystems (4)
lore
Share & tools
lore

Thomas Zimmermann

author maintainer
#0 lore raw

v2 cover drm/mgag200: Workaround HW bug with non-0 offset

We found an MGA chip that does not interpret the scanout offset correctly. This patchset works around the problem by placing all buffer objects at offset 0 on this system.

v2:
* serarate [4/4] from the rest of the series
* add missing tags to patch files

Thomas Zimmermann (3):
drm/mgag200: Extract device type from flags
drm/mgag200: Store flags from PCI driver data in device structure drm/mgag200: Add workaround for HW that does not support 'startadd'

 drivers/gpu/drm/mgag200/mgag200_drv.c  | 36 +++++++++++++++++++++++++-
 drivers/gpu/drm/mgag200/mgag200_drv.h  | 18 +++++++++++++
 drivers/gpu/drm/mgag200/mgag200_main.c |  3 ++-
 3 files changed, 55 insertions(+), 2 deletions(-)

--
2.23.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

author maintainer
#1 lore raw

PATCH v2 1/3 drm/mgag200: Extract device type from flags

Adds a conversion function that extracts the device type from the PCI id-table flags. Allows for storing additional information in the other flag bits.

Fixes 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin")
Cc John Donnelly <[email protected]>
Cc Gerd Hoffmann <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc David Airlie <[email protected]>
Cc Sam Ravnborg <[email protected]>
Cc Emil Velikov <[email protected]>
Cc "Y.C. Chen" <[email protected]>
Cc Laurent Pinchart <[email protected]>
Cc "José Roberto de Souza" <[email protected]>
Cc Andrzej Pietrasiewicz <[email protected]>
Cc <[email protected]> # v5.3+
---
 drivers/gpu/drm/mgag200/mgag200_drv.h  | 7 +++++++
 drivers/gpu/drm/mgag200/mgag200_main.c | 2 +-
 2 files changed, 8 insertions(+), 1 deletion(-)
drivers/gpu/drm/mgag200/mgag200_drv.h +7 -0
index 0ea9a525e57d..976404634092 100644
@@ -150,6 +150,8 @@ enum mga_type {
 	G200_EW3,
 };
 
+#define MGAG200_TYPE_MASK	(0x000000ff)
+
 #define IS_G200_SE(mdev) (mdev->type == G200_SE_A || mdev->type == G200_SE_B)
 
 struct mga_device {
@@ -181,6 +183,11 @@ struct mga_device {
 	u32 unique_rev_id;
 };
 
+static inline enum mga_type
+mgag200_type_from_driver_data(kernel_ulong_t driver_data)
+{
+	return (enum mga_type)(driver_data & MGAG200_TYPE_MASK);
+}
 				/* mgag200_mode.c */
 int mgag200_modeset_init(struct mga_device *mdev);
 void mgag200_modeset_fini(struct mga_device *mdev);
drivers/gpu/drm/mgag200/mgag200_main.c +1 -1
index 5f74aabcd3df..517c5693ad69 100644
@@ -94,7 +94,7 @@ static int mgag200_device_init(struct drm_device *dev,
 	struct mga_device *mdev = dev->dev_private;
 	int ret, option;
 
-	mdev->type = flags;
+	mdev->type = mgag200_type_from_driver_data(flags);
 
 	/* Hardcode the number of CRTCs to 1 */
 	mdev->num_crtc = 1;
Signature · 1 line
2.23.0
#2 lore raw

Re: [PATCH v2 1/3] drm/mgag200: Extract device type from flags

Hi Thomas,

Quoted 7 lines · Thomas Zimmermann

Adds a conversion function that extracts the device type from the PCI id-table flags. Allows for storing additional information in the other flag bits.

Fixes 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin")

Are you sure the fixes tag is correct? Neither the commit summary nor the patch itself seems related to the changes below.

HTH
Emil

author maintainer
#3 lore raw

Re: [PATCH v2 1/3] drm/mgag200: Extract device type from flags

Hi Emil

Am 27.11.19 um 17:29 schrieb Emil Velikov:

Quoted 8 lines

Hi Thomas,

Quoted 5 lines · Thomas Zimmermann

Adds a conversion function that extracts the device type from the PCI id-table flags. Allows for storing additional information in the other flag bits.

Quoted 5 lines
Quoted 2 lines
Fixes 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin")

Are you sure the fixes tag is correct? Neither the commit summary nor the patch itself seems related to the changes below.

Yes, it's correct. It's part of a patch series [1][2][3] that fixes the bug.

Best regards
Thomas

[1]
https://cgit.freedesktop.org/drm/drm-tip/commit/?id=3a8a5aba142a44eaeba0cb0ec1b4a8f177b5e59a
[2]
https://cgit.freedesktop.org/drm/drm-tip/commit/?id=d6d437d97d54c85a1a93967b2745e31dff03365a
[3]
https://cgit.freedesktop.org/drm/drm-tip/commit/?id=1591fadf857cdbaf2baa55e421af99a61354713c

Quoted 8 lines

HTH
Emil
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Signature · 6 lines
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer
#4 lore raw

Re: [PATCH v2 1/3] drm/mgag200: Extract device type from flags

Quoted 22 lines · Thomas Zimmermann

Hi Emil

Am 27.11.19 um 17:29 schrieb Emil Velikov:

Quoted 8 lines

Hi Thomas,

Quoted 5 lines · Thomas Zimmermann

Adds a conversion function that extracts the device type from the PCI id-table flags. Allows for storing additional information in the other flag bits.

Quoted 5 lines
Quoted 2 lines
Fixes 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin")

Are you sure the fixes tag is correct? Neither the commit summary nor the patch itself seems related to the changes below.

Yes, it's correct. It's part of a patch series [1][2][3] that fixes the bug.

Best regards
Thomas

Quoted 7 lines

[1]
https://cgit.freedesktop.org/drm/drm-tip/commit/?id=3a8a5aba142a44eaeba0cb0ec1b4a8f177b5e59a
[2]
https://cgit.freedesktop.org/drm/drm-tip/commit/?id=d6d437d97d54c85a1a93967b2745e31dff03365a
[3]
https://cgit.freedesktop.org/drm/drm-tip/commit/?id=1591fadf857cdbaf2baa55e421af99a61354713c

I see, different alignment is required for one mga200 GPU. Personally, I would have mentioned that in the commit message.

Regardless, hats off for the prompt fixup.

Thanks
Emil
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

author maintainer
#5 lore raw

PATCH v2 2/3 drm/mgag200: Store flags from PCI driver data in device structure

In reply to Thomas Zimmermann

The flags field in struct mga_device has been unused so far. We now use it to store flag bits from the PCI driver.

Fixes 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin")
Cc John Donnelly <[email protected]>
Cc Gerd Hoffmann <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc David Airlie <[email protected]>
Cc Sam Ravnborg <[email protected]>
Cc "Y.C. Chen" <[email protected]>
Cc Neil Armstrong <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc "José Roberto de Souza" <[email protected]>
Cc Andrzej Pietrasiewicz <[email protected]>
Cc <[email protected]> # v5.3+
---
 drivers/gpu/drm/mgag200/mgag200_drv.h  | 8 ++++++++
 drivers/gpu/drm/mgag200/mgag200_main.c | 1 +
 2 files changed, 9 insertions(+)
drivers/gpu/drm/mgag200/mgag200_drv.h +8 -0
index 976404634092..4b4f9ce74a84 100644
@@ -151,6 +151,7 @@ enum mga_type {
 };
 
 #define MGAG200_TYPE_MASK	(0x000000ff)
+#define MGAG200_FLAG_MASK	(0x00ffff00)
 
 #define IS_G200_SE(mdev) (mdev->type == G200_SE_A || mdev->type == G200_SE_B)
 
@@ -188,6 +189,13 @@ mgag200_type_from_driver_data(kernel_ulong_t driver_data)
 {
 	return (enum mga_type)(driver_data & MGAG200_TYPE_MASK);
 }
+
+static inline unsigned long
+mgag200_flags_from_driver_data(kernel_ulong_t driver_data)
+{
+	return driver_data & MGAG200_FLAG_MASK;
+}
+
 				/* mgag200_mode.c */
 int mgag200_modeset_init(struct mga_device *mdev);
 void mgag200_modeset_fini(struct mga_device *mdev);
drivers/gpu/drm/mgag200/mgag200_main.c +1 -0
index 517c5693ad69..e1bc5b0aa774 100644
@@ -94,6 +94,7 @@ static int mgag200_device_init(struct drm_device *dev,
 	struct mga_device *mdev = dev->dev_private;
 	int ret, option;
 
+	mdev->flags = mgag200_flags_from_driver_data(flags);
 	mdev->type = mgag200_type_from_driver_data(flags);
 
 	/* Hardcode the number of CRTCs to 1 */
Signature · 1 line
2.23.0
maintainer
#6 lore raw

Re: [PATCH v2 2/3] drm/mgag200: Store flags from PCI driver data in device structure

Hi,

[This is an automated email]

This commit has been processed because it contains a "Fixes:" tag,
fixing commit: 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin").

The bot has tested the following trees: v5.3.13.

v5.3.13: Failed to apply! Possible dependencies:
1c355c0ecfc0 ("drm/mgag200: Extract device type from flags")

NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

Signature · 6 lines
Thanks,
Sasha
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
author maintainer
#7 lore raw

PATCH v2 3/3 drm/mgag200: Add workaround for HW that does not support 'startadd'

In reply to Thomas Zimmermann

There's at least one system that does not interpret the value of the device's 'startadd' field correctly, which leads to incorrectly displayed scanout buffers. Always placing the active scanout buffer at offset 0 works around the problem.

Fixes 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin")
Cc Gerd Hoffmann <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc David Airlie <[email protected]>
Cc Sam Ravnborg <[email protected]>
Cc "Y.C. Chen" <[email protected]>
Cc Neil Armstrong <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc "José Roberto de Souza" <[email protected]>
Cc Andrzej Pietrasiewicz <[email protected]>
Cc <[email protected]> # v5.3+
---
 drivers/gpu/drm/mgag200/mgag200_drv.c | 36 ++++++++++++++++++++++++++-
 drivers/gpu/drm/mgag200/mgag200_drv.h |  3 +++
 2 files changed, 38 insertions(+), 1 deletion(-)
drivers/gpu/drm/mgag200/mgag200_drv.c +35 -1
index 397f8b0a9af8..d43951caeea0 100644
@@ -30,6 +30,8 @@ module_param_named(modeset, mgag200_modeset, int, 0400);
 static struct drm_driver driver;
 
 static const struct pci_device_id pciidlist[] = {
+	{ PCI_VENDOR_ID_MATROX, 0x522, PCI_VENDOR_ID_SUN, 0x4852, 0, 0,
+		G200_SE_A | MGAG200_FLAG_HW_BUG_NO_STARTADD},
 	{ PCI_VENDOR_ID_MATROX, 0x522, PCI_ANY_ID, PCI_ANY_ID, 0, 0, G200_SE_A },
 	{ PCI_VENDOR_ID_MATROX, 0x524, PCI_ANY_ID, PCI_ANY_ID, 0, 0, G200_SE_B },
 	{ PCI_VENDOR_ID_MATROX, 0x530, PCI_ANY_ID, PCI_ANY_ID, 0, 0, G200_EV },
@@ -60,6 +62,35 @@ static void mga_pci_remove(struct pci_dev *pdev)
 
 DEFINE_DRM_GEM_FOPS(mgag200_driver_fops);
 
+static bool mgag200_pin_bo_at_0(const struct mga_device *mdev)
+{
+	return mdev->flags & MGAG200_FLAG_HW_BUG_NO_STARTADD;
+}
+
+int mgag200_driver_dumb_create(struct drm_file *file,
+			       struct drm_device *dev,
+			       struct drm_mode_create_dumb *args)
+{
+	struct mga_device *mdev = dev->dev_private;
+	unsigned long pg_align;
+
+	if (WARN_ONCE(!dev->vram_mm, "VRAM MM not initialized"))
+		return -EINVAL;
+
+	pg_align = 0ul;
+
+	/*
+	 * Aligning scanout buffers to the size of the video ram forces
+	 * placement at offset 0. Works around a bug where HW does not
+	 * respect 'startadd' field.
+	 */
+	if (mgag200_pin_bo_at_0(mdev))
+		pg_align = PFN_UP(mdev->mc.vram_size);
+
+	return drm_gem_vram_fill_create_dumb(file, dev, &dev->vram_mm->bdev,
+					     pg_align, false, args);
+}
+
 static struct drm_driver driver = {
 	.driver_features = DRIVER_GEM | DRIVER_MODESET,
 	.load = mgag200_driver_load,
@@ -71,7 +102,10 @@ static struct drm_driver driver = {
 	.major = DRIVER_MAJOR,
 	.minor = DRIVER_MINOR,
 	.patchlevel = DRIVER_PATCHLEVEL,
-	DRM_GEM_VRAM_DRIVER
+	.debugfs_init = drm_vram_mm_debugfs_init,
+	.dumb_create = mgag200_driver_dumb_create,
+	.dumb_map_offset = drm_gem_vram_driver_dumb_mmap_offset,
+	.gem_prime_mmap = drm_gem_prime_mmap,
 };
 
 static struct pci_driver mgag200_pci_driver = {
drivers/gpu/drm/mgag200/mgag200_drv.h +3 -0
index 4b4f9ce74a84..aa32aad222c2 100644
@@ -150,6 +150,9 @@ enum mga_type {
 	G200_EW3,
 };
 
+/* HW does not handle 'startadd' field correct. */
+#define MGAG200_FLAG_HW_BUG_NO_STARTADD	(1ul << 8)
+
 #define MGAG200_TYPE_MASK	(0x000000ff)
 #define MGAG200_FLAG_MASK	(0x00ffff00)
 
Signature · 1 line
2.23.0
maintainer
#8 lore raw

Re: [PATCH v2 3/3] drm/mgag200: Add workaround for HW that does not support 'startadd'

Hi,

[This is an automated email]

This commit has been processed because it contains a "Fixes:" tag,
fixing commit: 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin").

The bot has tested the following trees: v5.3.13.

v5.3.13: Build failed! Errors:
drivers/gpu/drm/mgag200/mgag200_drv.c:104:18: error: ‘drm_vram_mm_debugfs_init’ undeclared here (not in a function); did you mean ‘drm_client_debugfs_init’?

NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

Signature · 2 lines
Thanks,
Sasha

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#9 lore raw

Re: FAILED - [PATCH v2 3/3] drm/mgag200: Add workaround for HW that does not support 'startadd'

Hello Sasha and Thomas ,

This particular patch has failed on one class of servers that has a slightly different Sun Vendor. ID for the BMC video device:

I will follow up with additional details in the review comments for the original message,.

Quoted 11 lines · Sasha Levin

Hi,

[This is an automated email]

This commit has been processed because it contains a "Fixes:" tag,
fixing commit: 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin").

The bot has tested the following trees: v5.3.13.

Quoted 15 lines

v5.3.13: Build failed! Errors:
drivers/gpu/drm/mgag200/mgag200_drv.c:104:18: error: ‘drm_vram_mm_debugfs_init’ undeclared here (not in a function); did you mean ‘drm_client_debugfs_init’?

NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

Signature · 6 lines
Thanks,
Sasha
_______________________________________________
dri-devel mailing list
[email protected]
https://urldefense.proofpoint.com/v2/url?u=https-3A__lists.freedesktop.org_mailman_listinfo_dri-2Ddevel&d=DwIGaQ&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=t2fPg9D87F7D8jm0_3CG9yoiIKdRg4qc_thBw4bzMhc&m=vxMDOLV77rRe2ekdNFH9IxMSBQrTccltZd8A1H6xYCc&s=efHs2lc_RQYvzLC82c-D3wa8MpX5DCU_YsIo6XruAQg&e=

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#10 lore raw

Re: FAILED - [PATCH v2 3/3] drm/mgag200: Add workaround for HW that does not support 'startadd'

Quoted 29 lines · John Donnelly

Hello Sasha and Thomas ,

This particular patch has failed on one class of servers that has a slightly different Sun Vendor. ID for the BMC video device:

I will follow up with additional details in the review comments for the original message,.

Quoted 14 lines · Sasha Levin

Hi,

[This is an automated email]

This commit has been processed because it contains a "Fixes:" tag,
fixing commit: 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin").

The bot has tested the following trees: v5.3.13.

v5.3.13: Build failed! Errors:
drivers/gpu/drm/mgag200/mgag200_drv.c:104:18: error: ‘drm_vram_mm_debugfs_init’ undeclared here (not in a function); did you mean ‘drm_client_debugfs_init’?

I had this same issue and removed that from my local 5.4.0-rc8 build

Quoted 5 lines
Quoted 5 lines

NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

Quoted 8 lines
Quoted 7 lines
Signature · 6 lines
Thanks,
Sasha
_______________________________________________
dri-devel mailing list
[email protected]
https://urldefense.proofpoint.com/v2/url?u=https-3A__lists.freedesktop.org_mailman_listinfo_dri-2Ddevel&d=DwIGaQ&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=t2fPg9D87F7D8jm0_3CG9yoiIKdRg4qc_thBw4bzMhc&m=vxMDOLV77rRe2ekdNFH9IxMSBQrTccltZd8A1H6xYCc&s=efHs2lc_RQYvzLC82c-D3wa8MpX5DCU_YsIo6XruAQg&e=

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

author maintainer
#11 lore raw

Re: FAILED - [PATCH v2 3/3] drm/mgag200: Add workaround for HW that does not support 'startadd'

Hi

Am 06.12.19 um 01:48 schrieb John Donnelly:

Quoted 24 lines
Quoted 24 lines · John Donnelly

Hello Sasha and Thomas ,

This particular patch has failed on one class of servers that has a slightly different Sun Vendor. ID for the BMC video device:

I will follow up with additional details in the review comments for the original message,.

Quoted 9 lines · Sasha Levin

Hi,

[This is an automated email]

This commit has been processed because it contains a "Fixes:" tag,
fixing commit: 81da87f63a1e ("drm: Replace drm_gem_vram_push_to_system() with kunmap + unpin").

Quoted 8 lines
Quoted 5 lines
Quoted 5 lines

The bot has tested the following trees: v5.3.13.

v5.3.13: Build failed! Errors:
drivers/gpu/drm/mgag200/mgag200_drv.c:104:18: error: ‘drm_vram_mm_debugfs_init’ undeclared here (not in a function); did you mean ‘drm_client_debugfs_init’?

I had this same issue and removed that from my local 5.4.0-rc8 build

The bot used the wrong tree. The patch has been written against drm-tip.

Best regards
Thomas

Quoted 23 lines
Quoted 9 lines
Quoted 9 lines

NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

Quoted 8 lines
Quoted 7 lines
Signature · 6 lines
Thanks,
Sasha
_______________________________________________
dri-devel mailing list
[email protected]
https://urldefense.proofpoint.com/v2/url?u=https-3A__lists.freedesktop.org_mailman_listinfo_dri-2Ddevel&d=DwIGaQ&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=t2fPg9D87F7D8jm0_3CG9yoiIKdRg4qc_thBw4bzMhc&m=vxMDOLV77rRe2ekdNFH9IxMSBQrTccltZd8A1H6xYCc&s=efHs2lc_RQYvzLC82c-D3wa8MpX5DCU_YsIo6XruAQg&e=

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Signature · 6 lines
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer