| Message ID | 1413367931-4224-1-git-send-email-hdegoede@redhat.com |
|---|---|
| State | Accepted |
| Headers | show |
On Wed, 15 Oct 2014 12:12:11 +0200 Hans de Goede <hdegoede@redhat.com> wrote: > This change breaks various hardcoded assumptions in the linux-sunxi-3.4 > kernels, causing a divide by 0 error in the linux-sunxi-3.4 nand driver, > rendering 3.4 kernels unbootable. Besides this problem, it also breaks LCD > output (on devices without nand). > > Fixes for these issues are being added to the linux-sunxi-3.4 kernel, but for > now it seems better to revert this. > > This reverts commit 013f2d746955147439215a4939655c9ed6bdd866. As mentioned in the commit message https://www.mail-archive.com/linux-sunxi@googlegroups.com/msg07884.html the hardcoded assumptions in the sunxi-3.4 kernel are still bad with or without this revert if somebody tries to use higher DRAM clock speeds. The patch in question just smokes out bugs, which are otherwise still there. Going forward, the mainline kernel needs the G2D driver and it needs to run at a decent clock speed to be provide reasonable performance. Low PLL5P clock speed is limiting our options. This revert does not do anyone any favour and only postpones the resolution of the problem.
Hi, On 10/15/2014 12:34 PM, Siarhei Siamashka wrote: > On Wed, 15 Oct 2014 12:12:11 +0200 > Hans de Goede <hdegoede@redhat.com> wrote: > >> This change breaks various hardcoded assumptions in the linux-sunxi-3.4 >> kernels, causing a divide by 0 error in the linux-sunxi-3.4 nand driver, >> rendering 3.4 kernels unbootable. Besides this problem, it also breaks LCD >> output (on devices without nand). >> >> Fixes for these issues are being added to the linux-sunxi-3.4 kernel, but for >> now it seems better to revert this. >> >> This reverts commit 013f2d746955147439215a4939655c9ed6bdd866. > > As mentioned in the commit message > https://www.mail-archive.com/linux-sunxi@googlegroups.com/msg07884.html > the hardcoded assumptions in the sunxi-3.4 kernel are still bad > with or without this revert if somebody tries to use higher DRAM > clock speeds. > > The patch in question just smokes out bugs, which are otherwise still > there. > > Going forward, the mainline kernel needs the G2D driver and it needs > to run at a decent clock speed to be provide reasonable performance. > Low PLL5P clock speed is limiting our options. > > This revert does not do anyone any favour and only postpones the > resolution of the problem. I agree that we need to fix the linux-sunxi-3.4 kernels to work properly with different pll5 settings, which is why I've actually send a fix for some pll5 issues to the linux-sunxi list and acked your patches which fix other bits. I can understand you want to have optimal settings for the G2D stuff, and I'm not against that. But getting the G2D stuff upstream is something which is still far in the future. I really want people to be able to start using upstream u-boot, starting today. And it will take some time for the sunxi-3.4 fixes to trickle down to users. So for now I believe we really should revert the PLL5 u-boot changes as they break some kernels from booting. Then once getting G2D support upstream really comes into the picture (e.g. upstream kernel patches are posted and being reviewed), then we can talk about re-applying the PLL5 changes, but for now I believe that reverting them is best. Regards, Hans
On Thu, 2014-10-16 at 10:48 +0200, Hans de Goede wrote: > So for now I believe we really should revert the PLL5 u-boot changes as they > break some kernels from booting. This seems reasonable to me. Acked-by: Ian Campbell <ijc@hellion.org.uk> > > Then once getting G2D support upstream really comes into the picture > (e.g. upstream kernel patches are posted and being reviewed), then we can talk > about re-applying the PLL5 changes, but for now I believe that reverting them > is best. > > Regards, > > Hans > > > > > > >
On Thu, 16 Oct 2014 10:48:34 +0200 Hans de Goede <hdegoede@redhat.com> wrote: > Hi, > > On 10/15/2014 12:34 PM, Siarhei Siamashka wrote: > > On Wed, 15 Oct 2014 12:12:11 +0200 > > Hans de Goede <hdegoede@redhat.com> wrote: > > > >> This change breaks various hardcoded assumptions in the linux-sunxi-3.4 > >> kernels, causing a divide by 0 error in the linux-sunxi-3.4 nand driver, > >> rendering 3.4 kernels unbootable. Besides this problem, it also breaks LCD > >> output (on devices without nand). > >> > >> Fixes for these issues are being added to the linux-sunxi-3.4 kernel, but for > >> now it seems better to revert this. > >> > >> This reverts commit 013f2d746955147439215a4939655c9ed6bdd866. > > > > As mentioned in the commit message > > https://www.mail-archive.com/linux-sunxi@googlegroups.com/msg07884.html > > the hardcoded assumptions in the sunxi-3.4 kernel are still bad > > with or without this revert if somebody tries to use higher DRAM > > clock speeds. > > > > The patch in question just smokes out bugs, which are otherwise still > > there. > > > > Going forward, the mainline kernel needs the G2D driver and it needs > > to run at a decent clock speed to be provide reasonable performance. > > Low PLL5P clock speed is limiting our options. > > > > This revert does not do anyone any favour and only postpones the > > resolution of the problem. > > I agree that we need to fix the linux-sunxi-3.4 kernels to work properly with > different pll5 settings, which is why I've actually send a fix for some pll5 > issues to the linux-sunxi list and acked your patches which fix other bits. > > I can understand you want to have optimal settings for the G2D stuff, and I'm > not against that. But getting the G2D stuff upstream is something which > is still far in the future. The future is happening right now ;-) Luc has implemented the HDMI support in u-boot and ensured that it works with the simplefb driver, providing framebuffer in the mainline kernel. This has code has been available for several months for any geeks, willing to fish out and cherry pick the patches. But right now this needs to be more easily available for the users and developers. The mainline kernel got all the basic essentials and has been ready for the graphics and multimedia since around version 3.16 As anyone can see, the sunxi-3.4 kernel provides a rather rotten foundation. And even though we have a reasonably good support for both proprietary and free reverse engineered drivers with sunxi-3.4, it still makes a lot of sense to get the graphics drivers in the mainline kernel ASAP. > I really want people to be able to start using upstream u-boot, starting > today. And it will take some time for the sunxi-3.4 fixes to trickle down to > users. > > So for now I believe we really should revert the PLL5 u-boot changes as they > break some kernels from booting. > > Then once getting G2D support upstream really comes into the picture > (e.g. upstream kernel patches are posted and being reviewed), then we can talk > about re-applying the PLL5 changes, but for now I believe that reverting them > is best. You are the boss. We are all free to have our own private branches anyway.
Hi Hans, On 16-10-14 10:48, Hans de Goede wrote: > [snip] > > So for now I believe we really should revert the PLL5 u-boot changes as they > break some kernels from booting. > U-boot and the kernel should be independent. Hence u-boot cannot break kernels from booting or the kernel is bugged and should be fixed. Regards, Jeroen
On Thu, Oct 16, 2014 at 10:48:34AM +0200, Hans de Goede wrote: > Hi, > > On 10/15/2014 12:34 PM, Siarhei Siamashka wrote: > > On Wed, 15 Oct 2014 12:12:11 +0200 > > Hans de Goede <hdegoede@redhat.com> wrote: > > > >> This change breaks various hardcoded assumptions in the linux-sunxi-3.4 > >> kernels, causing a divide by 0 error in the linux-sunxi-3.4 nand driver, > >> rendering 3.4 kernels unbootable. Besides this problem, it also breaks LCD > >> output (on devices without nand). > >> > >> Fixes for these issues are being added to the linux-sunxi-3.4 kernel, but for > >> now it seems better to revert this. > >> > >> This reverts commit 013f2d746955147439215a4939655c9ed6bdd866. > > > > As mentioned in the commit message > > https://www.mail-archive.com/linux-sunxi@googlegroups.com/msg07884.html > > the hardcoded assumptions in the sunxi-3.4 kernel are still bad > > with or without this revert if somebody tries to use higher DRAM > > clock speeds. > > > > The patch in question just smokes out bugs, which are otherwise still > > there. > > > > Going forward, the mainline kernel needs the G2D driver and it needs > > to run at a decent clock speed to be provide reasonable performance. > > Low PLL5P clock speed is limiting our options. > > > > This revert does not do anyone any favour and only postpones the > > resolution of the problem. > > I agree that we need to fix the linux-sunxi-3.4 kernels to work properly with > different pll5 settings, which is why I've actually send a fix for some pll5 > issues to the linux-sunxi list and acked your patches which fix other bits. > > I can understand you want to have optimal settings for the G2D stuff, and I'm > not against that. But getting the G2D stuff upstream is something which > is still far in the future. I really want people to be able to start using > upstream u-boot, starting today. And it will take some time for the sunxi-3.4 > fixes to trickle down to users. We should be doing things right, in mainline. To bring up a different example, on TI OMAP4 parts at least for a long time in order to use mainline U-Boot on older kernels you had to manually add CONFIG_SOMETHING_OR_ANOTHER to enable additional clocks/mux that the old kernels had incorrectly relied on U-Boot to set. If we must do strange things to support old and incorrect but in the wild kernels we need to (a) make it opt-in (easier now with Kconfig!) and (b) schedule a removal of the hack all the same.
On Tue, 2014-10-21 at 16:58 -0400, Tom Rini wrote: > We should be doing things right, in mainline. To bring up a different > example, on TI OMAP4 parts at least for a long time in order to use > mainline U-Boot on older kernels you had to manually add > CONFIG_SOMETHING_OR_ANOTHER to enable additional clocks/mux that the old > kernels had incorrectly relied on U-Boot to set. If we must do strange > things to support old and incorrect but in the wild kernels we need to > (a) make it opt-in (easier now with Kconfig!) and (b) schedule a removal > of the hack all the same. A Kconfig option does sound like a reasonable compromise. Ian.
Hi, On 10/22/2014 10:14 AM, Ian Campbell wrote: > On Tue, 2014-10-21 at 16:58 -0400, Tom Rini wrote: >> We should be doing things right, in mainline. To bring up a different >> example, on TI OMAP4 parts at least for a long time in order to use >> mainline U-Boot on older kernels you had to manually add >> CONFIG_SOMETHING_OR_ANOTHER to enable additional clocks/mux that the old >> kernels had incorrectly relied on U-Boot to set. If we must do strange >> things to support old and incorrect but in the wild kernels we need to >> (a) make it opt-in (easier now with Kconfig!) and (b) schedule a removal >> of the hack all the same. > > A Kconfig option does sound like a reasonable compromise. Ok, I will look into this, my plan for now is to call it OLD_KERNEL_COMPAT, so that if we come across more cases like this we've one config option for them, rather then a ton of small isolated config options. Regards, Hans
On Wed, 2014-10-22 at 10:35 +0200, Hans de Goede wrote: > Hi, > > On 10/22/2014 10:14 AM, Ian Campbell wrote: > > On Tue, 2014-10-21 at 16:58 -0400, Tom Rini wrote: > >> We should be doing things right, in mainline. To bring up a different > >> example, on TI OMAP4 parts at least for a long time in order to use > >> mainline U-Boot on older kernels you had to manually add > >> CONFIG_SOMETHING_OR_ANOTHER to enable additional clocks/mux that the old > >> kernels had incorrectly relied on U-Boot to set. If we must do strange > >> things to support old and incorrect but in the wild kernels we need to > >> (a) make it opt-in (easier now with Kconfig!) and (b) schedule a removal > >> of the hack all the same. > > > > A Kconfig option does sound like a reasonable compromise. > > Ok, I will look into this, my plan for now is to call it OLD_KERNEL_COMPAT, > so that if we come across more cases like this we've one config option for > them, rather then a ton of small isolated config options. Is it particular to "old" kernels as such, or is it more to do with Allwinner SDK (and derived) kernels? Is it worth trying to keep workarounds for such kernels separated from workarounds for old mainline kernels? Ian.
Hi, On 10/22/2014 11:19 AM, Ian Campbell wrote: > On Wed, 2014-10-22 at 10:35 +0200, Hans de Goede wrote: >> Hi, >> >> On 10/22/2014 10:14 AM, Ian Campbell wrote: >>> On Tue, 2014-10-21 at 16:58 -0400, Tom Rini wrote: >>>> We should be doing things right, in mainline. To bring up a different >>>> example, on TI OMAP4 parts at least for a long time in order to use >>>> mainline U-Boot on older kernels you had to manually add >>>> CONFIG_SOMETHING_OR_ANOTHER to enable additional clocks/mux that the old >>>> kernels had incorrectly relied on U-Boot to set. If we must do strange >>>> things to support old and incorrect but in the wild kernels we need to >>>> (a) make it opt-in (easier now with Kconfig!) and (b) schedule a removal >>>> of the hack all the same. >>> >>> A Kconfig option does sound like a reasonable compromise. >> >> Ok, I will look into this, my plan for now is to call it OLD_KERNEL_COMPAT, >> so that if we come across more cases like this we've one config option for >> them, rather then a ton of small isolated config options. > > Is it particular to "old" kernels as such, or is it more to do with > Allwinner SDK (and derived) kernels? Is it worth trying to keep > workarounds for such kernels separated from workarounds for old mainline > kernels? AFAIK the first mainline kernels with sunxi support are recent enough that they don't need any workarounds. TBH I don't think differentiating between the 2 brings us anything. Regards, Hans
On Wed, 2014-10-22 at 11:26 +0200, Hans de Goede wrote: > Hi, > > On 10/22/2014 11:19 AM, Ian Campbell wrote: > > On Wed, 2014-10-22 at 10:35 +0200, Hans de Goede wrote: > >> Hi, > >> > >> On 10/22/2014 10:14 AM, Ian Campbell wrote: > >>> On Tue, 2014-10-21 at 16:58 -0400, Tom Rini wrote: > >>>> We should be doing things right, in mainline. To bring up a different > >>>> example, on TI OMAP4 parts at least for a long time in order to use > >>>> mainline U-Boot on older kernels you had to manually add > >>>> CONFIG_SOMETHING_OR_ANOTHER to enable additional clocks/mux that the old > >>>> kernels had incorrectly relied on U-Boot to set. If we must do strange > >>>> things to support old and incorrect but in the wild kernels we need to > >>>> (a) make it opt-in (easier now with Kconfig!) and (b) schedule a removal > >>>> of the hack all the same. > >>> > >>> A Kconfig option does sound like a reasonable compromise. > >> > >> Ok, I will look into this, my plan for now is to call it OLD_KERNEL_COMPAT, > >> so that if we come across more cases like this we've one config option for > >> them, rather then a ton of small isolated config options. > > > > Is it particular to "old" kernels as such, or is it more to do with > > Allwinner SDK (and derived) kernels? Is it worth trying to keep > > workarounds for such kernels separated from workarounds for old mainline > > kernels? > > AFAIK the first mainline kernels with sunxi support are recent enough that they > don't need any workarounds. TBH I don't think differentiating between the 2 > brings us anything. OK. Ian.
diff --git a/arch/arm/cpu/armv7/sunxi/dram.c b/arch/arm/cpu/armv7/sunxi/dram.c index 584f742..8c4dd51 100644 --- a/arch/arm/cpu/armv7/sunxi/dram.c +++ b/arch/arm/cpu/armv7/sunxi/dram.c @@ -256,7 +256,7 @@ static void mctl_setup_dram_clock(u32 clk, u32 mbus_clk) /* PLL5P and PLL6 are the potential clock sources for MBUS */ u32 pll6x_div, pll5p_div; u32 pll6x_clk = clock_get_pll6() / 1000000; - u32 pll5p_clk = clk / 24 * 48; + u32 pll5p_clk = clk / 24 * 24; u32 pll5p_rate, pll6x_rate; #ifdef CONFIG_SUN7I pll6x_clk *= 2; /* sun7i uses PLL6*2, sun5i uses just PLL6 */ @@ -269,40 +269,46 @@ static void mctl_setup_dram_clock(u32 clk, u32 mbus_clk) reg_val &= ~CCM_PLL5_CTRL_N_MASK; /* set N to 0 (x0) */ reg_val &= ~CCM_PLL5_CTRL_P_MASK; /* set P to 0 (x1) */ if (clk >= 540 && clk < 552) { - /* dram = 540MHz, pll5p = 1080MHz */ - pll5p_clk = 1080; + /* dram = 540MHz, pll5p = 540MHz */ + pll5p_clk = 540; reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(2)); reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(3)); reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(15)); + reg_val |= CCM_PLL5_CTRL_P(1); } else if (clk >= 512 && clk < 528) { - /* dram = 512MHz, pll5p = 1536MHz */ - pll5p_clk = 1536; + /* dram = 512MHz, pll5p = 384MHz */ + pll5p_clk = 384; reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(3)); reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(4)); reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(16)); + reg_val |= CCM_PLL5_CTRL_P(2); } else if (clk >= 496 && clk < 504) { - /* dram = 496MHz, pll5p = 1488MHz */ - pll5p_clk = 1488; + /* dram = 496MHz, pll5p = 372MHz */ + pll5p_clk = 372; reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(3)); reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(2)); reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(31)); + reg_val |= CCM_PLL5_CTRL_P(2); } else if (clk >= 468 && clk < 480) { - /* dram = 468MHz, pll5p = 936MHz */ - pll5p_clk = 936; + /* dram = 468MHz, pll5p = 468MHz */ + pll5p_clk = 468; reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(2)); reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(3)); reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(13)); + reg_val |= CCM_PLL5_CTRL_P(1); } else if (clk >= 396 && clk < 408) { - /* dram = 396MHz, pll5p = 792MHz */ - pll5p_clk = 792; + /* dram = 396MHz, pll5p = 396MHz */ + pll5p_clk = 396; reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(2)); reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(3)); reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(11)); + reg_val |= CCM_PLL5_CTRL_P(1); } else { /* any other frequency that is a multiple of 24 */ reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(2)); reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(2)); reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(clk / 24)); + reg_val |= CCM_PLL5_CTRL_P(CCM_PLL5_CTRL_P_X(2)); } reg_val &= ~CCM_PLL5_CTRL_VCO_GAIN; /* PLL VCO Gain off */ reg_val |= CCM_PLL5_CTRL_EN; /* PLL On */
This change breaks various hardcoded assumptions in the linux-sunxi-3.4 kernels, causing a divide by 0 error in the linux-sunxi-3.4 nand driver, rendering 3.4 kernels unbootable. Besides this problem, it also breaks LCD output (on devices without nand). Fixes for these issues are being added to the linux-sunxi-3.4 kernel, but for now it seems better to revert this. This reverts commit 013f2d746955147439215a4939655c9ed6bdd866. Signed-off-by: Hans de Goede <hdegoede@redhat.com> --- arch/arm/cpu/armv7/sunxi/dram.c | 28 +++++++++++++++++----------- 1 file changed, 17 insertions(+), 11 deletions(-)