The original modesetting code set MISC to 0x2d, which is <hpgoddev>, <clksel> and <ioaddsel>
With the conversion to atomic modesetting, <rammapen> accidentally got enabled as well. Revert this change and initialize MISC with a constant value of <hgoddev> and <ioaddsel>. The <clksel> bits are set in mga_crtc_set_plls(), sync flags are set in mgag200_set_mode_regs().
While at it, also rename the flag constant to match the nameing in the MGA Programming Manual.
Signed-off-by: Thomas Zimmermann tzimmermann@suse.de Reported-by: kernel test robot rong.a.chen@intel.com Suggested-by: Emil Velikov emil.l.velikov@gmail.com Fixes: db05f8d3dc87 ("drm/mgag200: Split MISC register update into PLL selection, SYNC and I/O") Cc: Thomas Zimmermann tzimmermann@suse.de Cc: Sam Ravnborg sam@ravnborg.org Cc: Emil Velikov emil.velikov@collabora.com Cc: Dave Airlie airlied@redhat.com Cc: Daniel Vetter daniel.vetter@ffwll.ch Cc: Gerd Hoffmann kraxel@redhat.com Cc: "José Roberto de Souza" jose.souza@intel.com Cc: Andrzej Pietrasiewicz andrzej.p@collabora.com Cc: Rong Chen rong.a.chen@intel.com Cc: John Donnelly john.p.donnelly@Oracle.com --- drivers/gpu/drm/mgag200/mgag200_mode.c | 6 ++---- drivers/gpu/drm/mgag200/mgag200_reg.h | 2 +- 2 files changed, 3 insertions(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/mgag200/mgag200_mode.c b/drivers/gpu/drm/mgag200/mgag200_mode.c index f16bd278ab7e4..3b7235bd0bcba 100644 --- a/drivers/gpu/drm/mgag200/mgag200_mode.c +++ b/drivers/gpu/drm/mgag200/mgag200_mode.c @@ -1018,10 +1018,8 @@ static void mgag200_init_regs(struct mga_device *mdev) if (mdev->type == G200_EW3) WREG_ECRT(0x34, 0x5);
- misc = RREG8(MGA_MISC_IN); - misc |= MGAREG_MISC_IOADSEL | - MGAREG_MISC_RAMMAPEN | - MGAREG_MISC_HIGH_PG_SEL; + misc = MGAREG_MISC_HPGODDEV | + MGAREG_MISC_IOADSEL; WREG8(MGA_MISC_OUT, misc); }
diff --git a/drivers/gpu/drm/mgag200/mgag200_reg.h b/drivers/gpu/drm/mgag200/mgag200_reg.h index 29f7194faadc0..f6629e0d4bdf2 100644 --- a/drivers/gpu/drm/mgag200/mgag200_reg.h +++ b/drivers/gpu/drm/mgag200/mgag200_reg.h @@ -228,7 +228,7 @@ #define MGAREG_MISC_CLK_SEL_MGA_PIX (0x2 << 2) #define MGAREG_MISC_CLK_SEL_MGA_MSK (0x3 << 2) #define MGAREG_MISC_VIDEO_DIS (0x1 << 4) -#define MGAREG_MISC_HIGH_PG_SEL (0x1 << 5) +#define MGAREG_MISC_HPGODDEV BIT(5) #define MGAREG_MISC_HSYNCPOL BIT(6) #define MGAREG_MISC_VSYNCPOL BIT(7)
Hi Thomas,
On Tue, 16 Jun 2020 at 15:26, Thomas Zimmermann tzimmermann@suse.de wrote:
The original modesetting code set MISC to 0x2d, which is <hpgoddev>, <clksel> and <ioaddsel>
With the conversion to atomic modesetting, <rammapen> accidentally got enabled as well. Revert this change and initialize MISC with a constant value of <hgoddev> and <ioaddsel>. The <clksel> bits are set in mga_crtc_set_plls(), sync flags are set in mgag200_set_mode_regs().
Let's keep the remove (restoring original functionality) and rename (cosmetics) separate patches. The read has also disappeared, which should be safe although might be better on it's own.
Thanks Emil
Hi Emil
Am 16.06.20 um 17:14 schrieb Emil Velikov:
Hi Thomas,
On Tue, 16 Jun 2020 at 15:26, Thomas Zimmermann tzimmermann@suse.de wrote:
The original modesetting code set MISC to 0x2d, which is <hpgoddev>, <clksel> and <ioaddsel>
With the conversion to atomic modesetting, <rammapen> accidentally got enabled as well. Revert this change and initialize MISC with a constant value of <hgoddev> and <ioaddsel>. The <clksel> bits are set in mga_crtc_set_plls(), sync flags are set in mgag200_set_mode_regs().
Let's keep the remove (restoring original functionality) and rename (cosmetics) separate patches. The read has also disappeared, which should be safe although might be better on it's own.
I'm waiting for Rong Chen's performance results on this patch. Moving the rename into a separate patch makes sense, but removing the read is part of restoring the original behavior. I think it should be in this patch. Maybe I can write a better commit message to highlight the change.
Best regards Thomas
Thanks Emil _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
On Wed, Jun 17, 2020 at 08:28:02AM +0200, Thomas Zimmermann wrote:
Hi Emil
Am 16.06.20 um 17:14 schrieb Emil Velikov:
Hi Thomas,
On Tue, 16 Jun 2020 at 15:26, Thomas Zimmermann tzimmermann@suse.de wrote:
The original modesetting code set MISC to 0x2d, which is <hpgoddev>, <clksel> and <ioaddsel>
With the conversion to atomic modesetting, <rammapen> accidentally got enabled as well. Revert this change and initialize MISC with a constant value of <hgoddev> and <ioaddsel>. The <clksel> bits are set in mga_crtc_set_plls(), sync flags are set in mgag200_set_mode_regs().
Let's keep the remove (restoring original functionality) and rename (cosmetics) separate patches. The read has also disappeared, which should be safe although might be better on it's own.
I'm waiting for Rong Chen's performance results on this patch. Moving the rename into a separate patch makes sense, but removing the read is part of restoring the original behavior. I think it should be in this patch. Maybe I can write a better commit message to highlight the change.
Hi Thomas,
I tested the patch based on previous patch series, it seems doesn't take effect, and commit 39fb72816c4ee brings a larger regression when comparing to commit caac4dda44f37:
4606edf870927 drm/mgag200: Don't set <rammapen> in MISC 3 3 3 39fb72816c4ee drm/mgag200: Replace VRAM helpers with SHMEM helpers 3 22 caac4dda44f37 drm/mgag200: Convert to simple KMS helper 238 233 240 234 239 4b11c90431108 drm/mgag200: Use simple-display data structures 35 34 34 34 db22c903c9322 drm/mgag200: Remove out-commented suspend/resume helpers 34 34 87d4880d2aeb5 drm/mgag200: Move register initialization into separate function 35 0c51726f95396 drm/mgag200: Move hiprilvl setting into separate functions 35 8078e8b182e73 drm/mgag200: Set primary plane's format in separate helper function 35 19f7b409d95b7 drm/mgag200: Set pitch in a separate helper function 35 34 9b9a363828c60 drm/mgag200: Update mode registers after plane registers 34 e44e907dd8f93 drm/mgag200: Split MISC register update into PLL selection, SYNC and I/O 34 34 34 34 bef2303526adb drm/mgag200: Move mode-setting code into separate helper function 97 97 5cd8460e81e8f drm/mgag200: Clean up mga_crtc_do_set_base() 96 0671ca8c559ba drm/mgag200: Clean up mga_set_start_address() e82c8969a2474 drm/mgag200: Remove HW cursor 96
Best Regards, Rong Chen
Hi
Am 17.06.20 um 11:22 schrieb Rong Chen:
On Wed, Jun 17, 2020 at 08:28:02AM +0200, Thomas Zimmermann wrote:
Hi Emil
Am 16.06.20 um 17:14 schrieb Emil Velikov:
Hi Thomas,
On Tue, 16 Jun 2020 at 15:26, Thomas Zimmermann tzimmermann@suse.de wrote:
The original modesetting code set MISC to 0x2d, which is <hpgoddev>, <clksel> and <ioaddsel>
With the conversion to atomic modesetting, <rammapen> accidentally got enabled as well. Revert this change and initialize MISC with a constant value of <hgoddev> and <ioaddsel>. The <clksel> bits are set in mga_crtc_set_plls(), sync flags are set in mgag200_set_mode_regs().
Let's keep the remove (restoring original functionality) and rename (cosmetics) separate patches. The read has also disappeared, which should be safe although might be better on it's own.
I'm waiting for Rong Chen's performance results on this patch. Moving the rename into a separate patch makes sense, but removing the read is part of restoring the original behavior. I think it should be in this patch. Maybe I can write a better commit message to highlight the change.
Hi Thomas,
I tested the patch based on previous patch series, it seems doesn't take effect, and commit 39fb72816c4ee brings a larger regression when comparing to commit caac4dda44f37:
Thanks for testing!
I still don't understand these numbers, but are you sure that e44e907dd8f93 is really the regression?
It would make sense that 39fb72816c4ee has an impact on performance, as it changes memory management and the way the driver updates the display. I'll take a closer look at that patch and maybe send out an update.
Best regards Thomas
4606edf870927 drm/mgag200: Don't set <rammapen> in MISC 3 3 3 39fb72816c4ee drm/mgag200: Replace VRAM helpers with SHMEM helpers 3 22 caac4dda44f37 drm/mgag200: Convert to simple KMS helper 238 233 240 234 239 4b11c90431108 drm/mgag200: Use simple-display data structures 35 34 34 34 db22c903c9322 drm/mgag200: Remove out-commented suspend/resume helpers 34 34 87d4880d2aeb5 drm/mgag200: Move register initialization into separate function 35 0c51726f95396 drm/mgag200: Move hiprilvl setting into separate functions 35 8078e8b182e73 drm/mgag200: Set primary plane's format in separate helper function 35 19f7b409d95b7 drm/mgag200: Set pitch in a separate helper function 35 34 9b9a363828c60 drm/mgag200: Update mode registers after plane registers 34 e44e907dd8f93 drm/mgag200: Split MISC register update into PLL selection, SYNC and I/O 34 34 34 34 bef2303526adb drm/mgag200: Move mode-setting code into separate helper function 97 97 5cd8460e81e8f drm/mgag200: Clean up mga_crtc_do_set_base() 96 0671ca8c559ba drm/mgag200: Clean up mga_set_start_address() e82c8969a2474 drm/mgag200: Remove HW cursor 96
Best Regards, Rong Chen
On 6/17/20 8:29 PM, Thomas Zimmermann wrote:
Hi
Am 17.06.20 um 11:22 schrieb Rong Chen:
On Wed, Jun 17, 2020 at 08:28:02AM +0200, Thomas Zimmermann wrote:
Hi Emil
Am 16.06.20 um 17:14 schrieb Emil Velikov:
Hi Thomas,
On Tue, 16 Jun 2020 at 15:26, Thomas Zimmermann tzimmermann@suse.de wrote:
The original modesetting code set MISC to 0x2d, which is <hpgoddev>, <clksel> and <ioaddsel>
With the conversion to atomic modesetting, <rammapen> accidentally got enabled as well. Revert this change and initialize MISC with a constant value of <hgoddev> and <ioaddsel>. The <clksel> bits are set in mga_crtc_set_plls(), sync flags are set in mgag200_set_mode_regs().
Let's keep the remove (restoring original functionality) and rename (cosmetics) separate patches. The read has also disappeared, which should be safe although might be better on it's own.
I'm waiting for Rong Chen's performance results on this patch. Moving the rename into a separate patch makes sense, but removing the read is part of restoring the original behavior. I think it should be in this patch. Maybe I can write a better commit message to highlight the change.
Hi Thomas,
I tested the patch based on previous patch series, it seems doesn't take effect, and commit 39fb72816c4ee brings a larger regression when comparing to commit caac4dda44f37:
Thanks for testing!
I still don't understand these numbers, but are you sure that e44e907dd8f93 is really the regression?
Yes, it's a regression too, according to the previous report https://www.spinics.net/lists/dri-devel/msg260034.html the bot bisected to commit e44e907dd8f93 which let the glmark2.800x600.score reduced by 64.9%, but the bot doesn't notice the improvement of commit caac4dda44f37, and commit 39fb72816c4ee causes another regression.
Best Regards, Rong Chen
It would make sense that 39fb72816c4ee has an impact on performance, as it changes memory management and the way the driver updates the display. I'll take a closer look at that patch and maybe send out an update.
Best regards Thomas
4606edf870927 drm/mgag200: Don't set <rammapen> in MISC 3 3 3 39fb72816c4ee drm/mgag200: Replace VRAM helpers with SHMEM helpers 3 22 caac4dda44f37 drm/mgag200: Convert to simple KMS helper 238 233 240 234 239 4b11c90431108 drm/mgag200: Use simple-display data structures 35 34 34 34 db22c903c9322 drm/mgag200: Remove out-commented suspend/resume helpers 34 34 87d4880d2aeb5 drm/mgag200: Move register initialization into separate function 35 0c51726f95396 drm/mgag200: Move hiprilvl setting into separate functions 35 8078e8b182e73 drm/mgag200: Set primary plane's format in separate helper function 35 19f7b409d95b7 drm/mgag200: Set pitch in a separate helper function 35 34 9b9a363828c60 drm/mgag200: Update mode registers after plane registers 34 e44e907dd8f93 drm/mgag200: Split MISC register update into PLL selection, SYNC and I/O 34 34 34 34 bef2303526adb drm/mgag200: Move mode-setting code into separate helper function 97 97 5cd8460e81e8f drm/mgag200: Clean up mga_crtc_do_set_base() 96 0671ca8c559ba drm/mgag200: Clean up mga_set_start_address() e82c8969a2474 drm/mgag200: Remove HW cursor 96
Best Regards, Rong Chen
Hi
Am 18.06.20 um 03:40 schrieb Rong Chen:
On 6/17/20 8:29 PM, Thomas Zimmermann wrote:
Hi
Am 17.06.20 um 11:22 schrieb Rong Chen:
On Wed, Jun 17, 2020 at 08:28:02AM +0200, Thomas Zimmermann wrote:
Hi Emil
Am 16.06.20 um 17:14 schrieb Emil Velikov:
Hi Thomas,
On Tue, 16 Jun 2020 at 15:26, Thomas Zimmermann tzimmermann@suse.de wrote:
The original modesetting code set MISC to 0x2d, which is <hpgoddev>, <clksel> and <ioaddsel>
With the conversion to atomic modesetting, <rammapen> accidentally got enabled as well. Revert this change and initialize MISC with a constant value of <hgoddev> and <ioaddsel>. The <clksel> bits are set in mga_crtc_set_plls(), sync flags are set in mgag200_set_mode_regs().
Let's keep the remove (restoring original functionality) and rename (cosmetics) separate patches. The read has also disappeared, which should be safe although might be better on it's own.
I'm waiting for Rong Chen's performance results on this patch. Moving the rename into a separate patch makes sense, but removing the read is part of restoring the original behavior. I think it should be in this patch. Maybe I can write a better commit message to highlight the change.
Hi Thomas,
I tested the patch based on previous patch series, it seems doesn't take effect, and commit 39fb72816c4ee brings a larger regression when comparing to commit caac4dda44f37:
Thanks for testing!
I still don't understand these numbers, but are you sure that e44e907dd8f93 is really the regression?
Yes, it's a regression too, according to the previous report https://www.spinics.net/lists/dri-devel/msg260034.html the bot bisected to commit e44e907dd8f93 which let the glmark2.800x600.score reduced by 64.9%, but the bot doesn't notice the improvement of commit caac4dda44f37, and commit 39fb72816c4ee causes another regression.
I sent out a patch for 39fb72816c4ee. It enables caching on the framebuffer's pages.
Best regards Thomas
Best Regards, Rong Chen
It would make sense that 39fb72816c4ee has an impact on performance, as it changes memory management and the way the driver updates the display. I'll take a closer look at that patch and maybe send out an update.
Best regards Thomas
4606edf870927 drm/mgag200: Don't set <rammapen> in MISC 3 3 3 39fb72816c4ee drm/mgag200: Replace VRAM helpers with SHMEM helpers 3 22 caac4dda44f37 drm/mgag200: Convert to simple KMS helper 238 233 240 234 239 4b11c90431108 drm/mgag200: Use simple-display data structures 35 34 34 34 db22c903c9322 drm/mgag200: Remove out-commented suspend/resume helpers 34 34 87d4880d2aeb5 drm/mgag200: Move register initialization into separate function 35 0c51726f95396 drm/mgag200: Move hiprilvl setting into separate functions 35 8078e8b182e73 drm/mgag200: Set primary plane's format in separate helper function 35 19f7b409d95b7 drm/mgag200: Set pitch in a separate helper function 35 34 9b9a363828c60 drm/mgag200: Update mode registers after plane registers 34 e44e907dd8f93 drm/mgag200: Split MISC register update into PLL selection, SYNC and I/O 34 34 34 34 bef2303526adb drm/mgag200: Move mode-setting code into separate helper function 97 97 5cd8460e81e8f drm/mgag200: Clean up mga_crtc_do_set_base() 96 0671ca8c559ba drm/mgag200: Clean up mga_set_start_address() e82c8969a2474 drm/mgag200: Remove HW cursor 96
Best Regards, Rong Chen
On Wed, 17 Jun 2020 at 07:28, Thomas Zimmermann tzimmermann@suse.de wrote:
Maybe I can write a better commit message to highlight the change.
That would be amazing, thank you.
Emil
dri-devel@lists.freedesktop.org
-
Emil Velikov -
Rong Chen -
Thomas Zimmermann