[PATCH 00/16] Refactor the architecture parts of mt7628
Weijie Gao
weijie.gao at mediatek.com
Fri Jan 10 11:48:41 CET 2020
On Thu, 2020-01-09 at 13:36 +0100, Stefan Roese wrote:
> Hi Weijie,
>
> On 09.01.20 09:49, Weijie Gao wrote:
> > On Thu, 2020-01-09 at 08:52 +0100, Stefan Roese wrote:
> >> Hi Weijie,
> >>
> >> On 08.01.20 03:59, Weijie Gao wrote:
> >>> This patch series are divided into two parts:
> >>>
> >>> The main part is to rewrite the whole architecture code of mt7628:
> >>> 1. Lock parts of the d-cache for initial stack so the rest of the code can
> >>> be reimplemented in C.
> >>> 2. Memory controller & DDR initialization have been fully written to support
> >>> detecting DDR size automatically.
> >>> 3. DDR calibration has also been reimplemented with a clear logic.
> >>> 4. Implemented a new sysreset driver to take advantage of the reset
> >>> controller so we can drop the use of syscon-based sysreset to reduce size.
> >>>
> >>> The second part is to add SPL support for mt7628:
> >>> 1. With SPL enabled we can build the ROM-bootable and RAM-bootable binary
> >>> simultaneously, and we can drop RAM boot related configs and defconfig
> >>> files.
> >>> 2. Generate compressed u-boot.bin image for SPL to reduce size of final
> >>> combined binary.
> >>> 3. Add a demo board (mt7628_rfb) to make use of the compressed image.
> >>
> >> Many thanks for working on this. Unfortunately I didn't receive these patches
> >> directly (I remember that you had problems sending patches to me before).
> >>
> >> I did a quick test on the LinkIt board and have a few questions before
> >> I start reviewing the patches in more detail:
> >>
> >> I don't see any serial output from the SPL. I assume that the muxing
> >> for UART2 is missing here in the SPL image (see board_debug_uart_init)?
> >
> > I'll do further test for this.
>
> I also did some tests and here is a WIP patch that fixes this issue for
> the LinkIt board. Please feel free to integrate it into your next patch
> set version:
>
> diff --git a/arch/mips/mach-mtmips/spl.c b/arch/mips/mach-mtmips/spl.c
> index 37172abadf..33b4485be2 100644
> --- a/arch/mips/mach-mtmips/spl.c
> +++ b/arch/mips/mach-mtmips/spl.c
> @@ -7,10 +7,25 @@
>
> #include <common.h>
> #include <spl.h>
> +#include <asm/io.h>
> +
> +#define MT76XX_GPIO1_MODE 0x10000060
> +
> +static void uart2_mux_init(void)
> +{
> + void __iomem *gpio_mode;
> +
> + /* Select UART2 mode instead of GPIO mode (default) */
> + gpio_mode = ioremap_nocache(MT76XX_GPIO1_MODE, 0x100);
> + clrbits_le32(gpio_mode, GENMASK(27, 26));
> +}
>
> void __noreturn board_init_f(ulong dummy)
> {
> #ifdef CONFIG_SPL_SERIAL_SUPPORT
> +#if CONFIG_CONS_INDEX == 3
> + uart2_mux_init();
> +#endif
> preloader_console_init();
> #endif
>
> diff --git a/include/configs/linkit-smart-7688.h b/include/configs/linkit-smart-7688.h
> index 01705ee2a5..b4bba9dcc5 100644
> --- a/include/configs/linkit-smart-7688.h
> +++ b/include/configs/linkit-smart-7688.h
> @@ -32,8 +32,10 @@
> #define CONFIG_SYS_NS16550_MEM32
> #define CONFIG_SYS_NS16550_CLK 40000000
> #define CONFIG_SYS_NS16550_REG_SIZE -4
> -#define CONFIG_SYS_NS16550_COM1 0xb0000e00
> -#define CONFIG_CONS_INDEX 1
> +/* Console index starts with 1 -> UART2 = IDX 3 */
> +#define CONFIG_SYS_NS16550_COM3 0xb0000e00
> +#undef CONFIG_CONS_INDEX
> +#define CONFIG_CONS_INDEX 3
> #endif
>
> /* UART */
>
>
> >>
> >> Did you compare the boot speed with using SPL + main U-Boot vs U-Boot
> >> only?
> >
> > As far as I tested the difference of boot time should be less than 1s.
>
> 1 second is quite a lot of time. I prefer to stay with the non compressed
> boot configuration then. Testing shows that the new SPL based version
> also boots pretty fast, when the main U-Boot proper is not compressed.
>
> >>
> >> In the current combined image the non-compressed U-Boot proper is
> >> included. Why did you not include the LZMA compressed image per default?
> >> Is the decompression too slow?
> >
> > I just kept these two boards untouched. I can switch them to compressed
> > mode if you have no concern.
>
> I did not test yet. If its really a difference in time of 1 second (or
> similar), then let stay with uncompressed for these board please.
>
> Thanks,
> Stefan
I did a precise measurement, and this is the result:
type | spl load/decompression time | boot to shell (total time)
spl + lzma | 75ms | 563ms
spl + none | 139ms | 627ms
non-spl | N/A | 626ms
Actually spl+lzma is the fastest. I think this is because it has the
smallest binary size.
It's OK to switch all of them to compressed mode.
Best Regards,
More information about the U-Boot
mailing list