[PATCH v2 0/5] rockchip: Fix eMMC performance regression
The eMMC performance on RK3399 was reduced sigificant by the commit 2cc6cde647e2 ("mmc: rockchip_sdhci: Limit number of blocks read in a single command").
That workaround should only have been applied to RK3568 and RK3588. This series fixes that and also help boost eMMC performance on two RK3399 boards even more by enabling use of SDMA.
There is also an extra commit to help build a u-boot-rockchip-spi.bin image that can be used for SPI flash boot on RockPro64.
Changes in v2: - Rebase on top of defconfig and spi v2 series - Collect r-b and t-b tags
Jonas Karlman (5): mmc: rockchip_sdhci: Skip blocks read workaround on RK3399 mmc: rockchip_sdhci: Disable DMA mode using a device tree property rockchip: rockpro64: Use SDMA to boost eMMC performance rockchip: rock-pi-4: Use SDMA to boost eMMC performance rockchip: rockpro64: Build u-boot-rockchip-spi.bin
arch/arm/dts/rk3399-rock-pi-4-u-boot.dtsi | 6 ++++++ arch/arm/dts/rk3399-rockpro64-u-boot.dtsi | 5 +++++ arch/arm/dts/rk3399-u-boot.dtsi | 1 + arch/arm/dts/rk3588s-u-boot.dtsi | 1 + configs/rock-pi-4-rk3399_defconfig | 2 ++ configs/rock5b-rk3588_defconfig | 1 - configs/rockpro64-rk3399_defconfig | 5 +++++ drivers/mmc/rockchip_sdhci.c | 12 +++++++++++- 8 files changed, 31 insertions(+), 2 deletions(-)
The workaround to limit number of blocks to read in a single command should only be applied to RK3568 and RK3588. Change to be more strict when to apply the workaround.
Fixes: 2cc6cde647e2 ("mmc: rockchip_sdhci: Limit number of blocks read in a single command") Suggested-by: Simon Glass sjg@chromium.org Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com Tested-by: Quentin Schulz foss+uboot@0leil.net # RK3399 Puma, RK3588 Tiger --- v2: - Collect r-b and t-b tags
drivers/mmc/rockchip_sdhci.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/mmc/rockchip_sdhci.c b/drivers/mmc/rockchip_sdhci.c index 4f110976f4e8..8e4a158049a9 100644 --- a/drivers/mmc/rockchip_sdhci.c +++ b/drivers/mmc/rockchip_sdhci.c @@ -594,7 +594,9 @@ static int rockchip_sdhci_probe(struct udevice *dev) * triggers Data End Bit Error on RK3568 and RK3588. Limit to reading * max 4 blocks in one command when using PIO mode. */ - if (!(host->flags & USE_DMA)) + if (!(host->flags & USE_DMA) && + (device_is_compatible(dev, "rockchip,rk3568-dwcmshc") || + device_is_compatible(dev, "rockchip,rk3588-dwcmshc"))) cfg->b_max = 4;
return sdhci_probe(dev);
Loading part of TF-A into SRAM from eMMC using DMA fails on RK3399 similar to other Rockchip SoCs. Checksum validation fails with:
## Checking hash(es) for Image atf-2 ... sha256 error! Bad hash value for 'hash' hash node in 'atf-2' image node spl_load_simple_fit: can't load image loadables index 1 (ret = -1) mmc_load_image_raw_sector: mmc block read error SPL: failed to boot from all boot devices ### ERROR ### Please RESET the board ###
Add a device tree property, u-boot,spl-fifo-mode, to control when the rockchip_sdhci driver should disable the use of DMA and fallback on PIO mode. Same device tree property is used by the rockchip_dw_mmc driver.
In commit 2cc6cde647e2 ("mmc: rockchip_sdhci: Limit number of blocks read in a single command") the DMA mode was disabled using a CONFIG option on RK3588. Revert that and instead disable DMA using the device tree property for all RK3588 boards, also apply similar workaround for all RK3399 boards.
Fixes: 2cc6cde647e2 ("mmc: rockchip_sdhci: Limit number of blocks read in a single command") Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com Tested-by: Quentin Schulz foss+uboot@0leil.net # RK3399 Puma, RK3588 Tiger --- v2: - Rebase on top of defconfig and spi v2 series - Collect r-b and t-b tags
arch/arm/dts/rk3399-u-boot.dtsi | 1 + arch/arm/dts/rk3588s-u-boot.dtsi | 1 + configs/rock5b-rk3588_defconfig | 1 - drivers/mmc/rockchip_sdhci.c | 8 ++++++++ 4 files changed, 10 insertions(+), 1 deletion(-)
diff --git a/arch/arm/dts/rk3399-u-boot.dtsi b/arch/arm/dts/rk3399-u-boot.dtsi index e677ae678dab..3423b882c437 100644 --- a/arch/arm/dts/rk3399-u-boot.dtsi +++ b/arch/arm/dts/rk3399-u-boot.dtsi @@ -120,6 +120,7 @@ &sdhci { max-frequency = <200000000>; bootph-all; + u-boot,spl-fifo-mode; };
&sdmmc { diff --git a/arch/arm/dts/rk3588s-u-boot.dtsi b/arch/arm/dts/rk3588s-u-boot.dtsi index 64c309046587..c703e41802b6 100644 --- a/arch/arm/dts/rk3588s-u-boot.dtsi +++ b/arch/arm/dts/rk3588s-u-boot.dtsi @@ -239,6 +239,7 @@
&sdhci { bootph-pre-ram; + u-boot,spl-fifo-mode; };
&uart2 { diff --git a/configs/rock5b-rk3588_defconfig b/configs/rock5b-rk3588_defconfig index 9d0b55c01ac9..c1155c20efa8 100644 --- a/configs/rock5b-rk3588_defconfig +++ b/configs/rock5b-rk3588_defconfig @@ -66,7 +66,6 @@ CONFIG_MMC_DW=y CONFIG_MMC_DW_ROCKCHIP=y CONFIG_MMC_SDHCI=y CONFIG_MMC_SDHCI_SDMA=y -# CONFIG_SPL_MMC_SDHCI_SDMA is not set CONFIG_MMC_SDHCI_ROCKCHIP=y CONFIG_SPI_FLASH_MACRONIX=y CONFIG_SPI_FLASH_XTX=y diff --git a/drivers/mmc/rockchip_sdhci.c b/drivers/mmc/rockchip_sdhci.c index 8e4a158049a9..285332d9f4fd 100644 --- a/drivers/mmc/rockchip_sdhci.c +++ b/drivers/mmc/rockchip_sdhci.c @@ -589,6 +589,14 @@ static int rockchip_sdhci_probe(struct udevice *dev) if (ret) return ret;
+ /* + * Disable use of DMA and force use of PIO mode in SPL to fix an issue + * where loading part of TF-A into SRAM using DMA silently fails. + */ + if (IS_ENABLED(CONFIG_SPL_BUILD) && + dev_read_bool(dev, "u-boot,spl-fifo-mode")) + host->flags &= ~USE_DMA; + /* * Reading more than 4 blocks with a single CMD18 command in PIO mode * triggers Data End Bit Error on RK3568 and RK3588. Limit to reading
Enable the use of SDMA mode to boost eMMC performance on RockPro64. Also add missing flags to indicate the supported MMC modes.
Using mmc read command to read 32 MiB data shows following improvement:
=> time mmc read 10000000 2000 10000
Before: time: 3.178 seconds After: time: 0.402 seconds
This also enables CONFIG_SPL_FIT_SIGNATURE option to help discover any possible future issue with loading TF-A into DRAM/SRAM.
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com --- v2: - Collect r-b tag
arch/arm/dts/rk3399-rockpro64-u-boot.dtsi | 5 +++++ configs/rockpro64-rk3399_defconfig | 2 ++ 2 files changed, 7 insertions(+)
diff --git a/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi b/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi index 32a83b2855ac..bd864d067018 100644 --- a/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi +++ b/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi @@ -15,6 +15,11 @@ }; };
+&sdhci { + cap-mmc-highspeed; + mmc-ddr-1_8v; +}; + &spi1 { spi_flash: flash@0 { bootph-all; diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 2b89b1baba51..0ca2cecade25 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -20,6 +20,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y CONFIG_USE_PREBOOT=y @@ -63,6 +64,7 @@ CONFIG_ROCKCHIP_EFUSE=y CONFIG_MMC_DW=y CONFIG_MMC_DW_ROCKCHIP=y CONFIG_MMC_SDHCI=y +CONFIG_MMC_SDHCI_SDMA=y CONFIG_MMC_SDHCI_ROCKCHIP=y CONFIG_SF_DEFAULT_BUS=1 CONFIG_SPI_FLASH_GIGADEVICE=y
On Wed, May 17, 2023 at 7:41 PM Jonas Karlman jonas@kwiboo.se wrote:
Enable the use of SDMA mode to boost eMMC performance on RockPro64. Also add missing flags to indicate the supported MMC modes.
Using mmc read command to read 32 MiB data shows following improvement:
=> time mmc read 10000000 2000 10000
Before: time: 3.178 seconds After: time: 0.402 seconds
This also enables CONFIG_SPL_FIT_SIGNATURE option to help discover any possible future issue with loading TF-A into DRAM/SRAM.
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com
v2:
- Collect r-b tag
arch/arm/dts/rk3399-rockpro64-u-boot.dtsi | 5 +++++ configs/rockpro64-rk3399_defconfig | 2 ++ 2 files changed, 7 insertions(+)
diff --git a/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi b/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi index 32a83b2855ac..bd864d067018 100644 --- a/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi +++ b/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi @@ -15,6 +15,11 @@ }; };
+&sdhci {
cap-mmc-highspeed;mmc-ddr-1_8v;+};
Has this been submitted for the upstream Linux kernel DT? The u-boot.dtsi isn't meant to be a general dumping ground for things that should be going upstream. Does it work with the Linux kernel as well as those people that boot using firmware provided DT will get this as well.
Peter
&spi1 { spi_flash: flash@0 { bootph-all; diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 2b89b1baba51..0ca2cecade25 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -20,6 +20,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y CONFIG_USE_PREBOOT=y @@ -63,6 +64,7 @@ CONFIG_ROCKCHIP_EFUSE=y CONFIG_MMC_DW=y CONFIG_MMC_DW_ROCKCHIP=y CONFIG_MMC_SDHCI=y +CONFIG_MMC_SDHCI_SDMA=y CONFIG_MMC_SDHCI_ROCKCHIP=y CONFIG_SF_DEFAULT_BUS=1 CONFIG_SPI_FLASH_GIGADEVICE=y -- 2.40.1
Hi Peter,
On 2023-05-17 21:07, Peter Robinson wrote:
On Wed, May 17, 2023 at 7:41 PM Jonas Karlman jonas@kwiboo.se wrote:
Enable the use of SDMA mode to boost eMMC performance on RockPro64. Also add missing flags to indicate the supported MMC modes.
Using mmc read command to read 32 MiB data shows following improvement:
=> time mmc read 10000000 2000 10000
Before: time: 3.178 seconds After: time: 0.402 seconds
This also enables CONFIG_SPL_FIT_SIGNATURE option to help discover any possible future issue with loading TF-A into DRAM/SRAM.
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com
v2:
- Collect r-b tag
arch/arm/dts/rk3399-rockpro64-u-boot.dtsi | 5 +++++ configs/rockpro64-rk3399_defconfig | 2 ++ 2 files changed, 7 insertions(+)
diff --git a/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi b/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi index 32a83b2855ac..bd864d067018 100644 --- a/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi +++ b/arch/arm/dts/rk3399-rockpro64-u-boot.dtsi @@ -15,6 +15,11 @@ }; };
+&sdhci {
cap-mmc-highspeed;mmc-ddr-1_8v;+};
Has this been submitted for the upstream Linux kernel DT? The u-boot.dtsi isn't meant to be a general dumping ground for things that should be going upstream. Does it work with the Linux kernel as well as those people that boot using firmware provided DT will get this as well.
It has not been submitted for upstream linux device tree yet. I have patches pending for rk3399-rockpro64 and some rk35xx devices in my local tree, hoping to have them submitted any day now.
This should work with linux, however linux should pick the hs200 mode over ddr52 mode.
With hs200 mode Kconfig enabled in u-boot the above test cmd result in ~0.2 seconds.
Regards, Jonas
Peter
&spi1 { spi_flash: flash@0 { bootph-all; diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 2b89b1baba51..0ca2cecade25 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -20,6 +20,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y CONFIG_USE_PREBOOT=y @@ -63,6 +64,7 @@ CONFIG_ROCKCHIP_EFUSE=y CONFIG_MMC_DW=y CONFIG_MMC_DW_ROCKCHIP=y CONFIG_MMC_SDHCI=y +CONFIG_MMC_SDHCI_SDMA=y CONFIG_MMC_SDHCI_ROCKCHIP=y CONFIG_SF_DEFAULT_BUS=1 CONFIG_SPI_FLASH_GIGADEVICE=y -- 2.40.1
Enable the use of SDMA mode to boost eMMC performance on ROCK Pi 4. Also add missing flags to indicate the supported MMC modes.
Using mmc read command to read 32 MiB data shows following improvement:
=> time mmc read 10000000 2000 10000
Before: time: 3.178 seconds After: time: 0.402 seconds
This also enables CONFIG_SPL_FIT_SIGNATURE option to help discover any possible future issue with loading TF-A into DRAM/SRAM.
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com --- v2: - Rebase to fix conflict with CONFIG_DEFAULT_FDT_FILE - Collect r-b tag
arch/arm/dts/rk3399-rock-pi-4-u-boot.dtsi | 6 ++++++ configs/rock-pi-4-rk3399_defconfig | 2 ++ 2 files changed, 8 insertions(+)
diff --git a/arch/arm/dts/rk3399-rock-pi-4-u-boot.dtsi b/arch/arm/dts/rk3399-rock-pi-4-u-boot.dtsi index c17e769f649f..60122f3bcd6c 100644 --- a/arch/arm/dts/rk3399-rock-pi-4-u-boot.dtsi +++ b/arch/arm/dts/rk3399-rock-pi-4-u-boot.dtsi @@ -12,6 +12,12 @@ }; };
+&sdhci { + cap-mmc-highspeed; + mmc-ddr-1_8v; + mmc-hs200-1_8v; +}; + &vdd_log { regulator-init-microvolt = <950000>; }; diff --git a/configs/rock-pi-4-rk3399_defconfig b/configs/rock-pi-4-rk3399_defconfig index cd93093cd2d3..4b984adc6ef8 100644 --- a/configs/rock-pi-4-rk3399_defconfig +++ b/configs/rock-pi-4-rk3399_defconfig @@ -19,6 +19,7 @@ CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y # CONFIG_ANDROID_BOOT_IMAGE is not set +CONFIG_SPL_FIT_SIGNATURE=y CONFIG_DEFAULT_FDT_FILE="rockchip/rk3399-rock-pi-4a.dtb" CONFIG_DISPLAY_BOARDINFO_LATE=y CONFIG_MISC_INIT_R=y @@ -56,6 +57,7 @@ CONFIG_ROCKCHIP_EFUSE=y CONFIG_MMC_DW=y CONFIG_MMC_DW_ROCKCHIP=y CONFIG_MMC_SDHCI=y +CONFIG_MMC_SDHCI_SDMA=y CONFIG_MMC_SDHCI_ROCKCHIP=y CONFIG_ETH_DESIGNWARE=y CONFIG_GMAC_ROCKCHIP=y
Enable CONFIG_ROCKCHIP_SPI_IMAGE to build u-boot-rockchip-spi.bin. Define CONFIG_SYS_SPI_U_BOOT_OFFS to write u-boot.itb at the expected offset. Enable CONFIG_LTO to reduce size of SPL so that the mkimage output fit before the 0x60000 offset in u-boot-rockchip-spi.bin.
=> sf probe SF: Detected gd25q128 with page size 256 Bytes, erase size 4 KiB, total 16 MiB => load mmc 1:1 10000000 u-boot-rockchip-spi.bin 1442304 bytes read in 27 ms (50.9 MiB/s) => sf update $fileaddr 0 $filesize device 0 offset 0x0, size 0x160200 1421824 bytes written, 20480 bytes skipped in 9.501s, speed 155432 B/s
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com --- v2: - Collect r-b tag
configs/rockpro64-rk3399_defconfig | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 0ca2cecade25..f41c03067903 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -11,6 +11,7 @@ CONFIG_ENV_OFFSET=0x3F8000 CONFIG_DEFAULT_DEVICE_TREE="rk3399-rockpro64" CONFIG_DM_RESET=y CONFIG_ROCKCHIP_RK3399=y +CONFIG_ROCKCHIP_SPI_IMAGE=y CONFIG_TARGET_ROCKPRO64_RK3399=y CONFIG_SPL_STACK=0x400000 CONFIG_DEBUG_UART_BASE=0xFF1A0000 @@ -20,6 +21,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y @@ -37,6 +39,7 @@ CONFIG_SPL_BSS_MAX_SIZE=0x2000 CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x10000 CONFIG_SPL_SPI_LOAD=y +CONFIG_SYS_SPI_U_BOOT_OFFS=0x60000 CONFIG_TPL=y CONFIG_CMD_BOOTZ=y CONFIG_CMD_GPT=y
On Wed, May 17, 2023 at 7:41 PM Jonas Karlman jonas@kwiboo.se wrote:
Enable CONFIG_ROCKCHIP_SPI_IMAGE to build u-boot-rockchip-spi.bin. Define CONFIG_SYS_SPI_U_BOOT_OFFS to write u-boot.itb at the expected
None of the other rk33* devices enable this offset yet my Pinebook Pro works fine booting from SPI flash, what does this fix/enable/change over the defaults?
offset. Enable CONFIG_LTO to reduce size of SPL so that the mkimage output fit before the 0x60000 offset in u-boot-rockchip-spi.bin.
The enabling of LTO seems like a separate change TBH, the changes seem to be independent and there's no mention of it in the subject.
=> sf probe SF: Detected gd25q128 with page size 256 Bytes, erase size 4 KiB, total 16 MiB => load mmc 1:1 10000000 u-boot-rockchip-spi.bin 1442304 bytes read in 27 ms (50.9 MiB/s) => sf update $fileaddr 0 $filesize device 0 offset 0x0, size 0x160200 1421824 bytes written, 20480 bytes skipped in 9.501s, speed 155432 B/s
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com
v2:
- Collect r-b tag
configs/rockpro64-rk3399_defconfig | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 0ca2cecade25..f41c03067903 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -11,6 +11,7 @@ CONFIG_ENV_OFFSET=0x3F8000 CONFIG_DEFAULT_DEVICE_TREE="rk3399-rockpro64" CONFIG_DM_RESET=y CONFIG_ROCKCHIP_RK3399=y +CONFIG_ROCKCHIP_SPI_IMAGE=y CONFIG_TARGET_ROCKPRO64_RK3399=y CONFIG_SPL_STACK=0x400000 CONFIG_DEBUG_UART_BASE=0xFF1A0000 @@ -20,6 +21,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y @@ -37,6 +39,7 @@ CONFIG_SPL_BSS_MAX_SIZE=0x2000 CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x10000 CONFIG_SPL_SPI_LOAD=y +CONFIG_SYS_SPI_U_BOOT_OFFS=0x60000 CONFIG_TPL=y CONFIG_CMD_BOOTZ=y CONFIG_CMD_GPT=y -- 2.40.1
Hi Peter,
On 2023-05-17 21:14, Peter Robinson wrote:
On Wed, May 17, 2023 at 7:41 PM Jonas Karlman jonas@kwiboo.se wrote:
Enable CONFIG_ROCKCHIP_SPI_IMAGE to build u-boot-rockchip-spi.bin. Define CONFIG_SYS_SPI_U_BOOT_OFFS to write u-boot.itb at the expected
None of the other rk33* devices enable this offset yet my Pinebook Pro works fine booting from SPI flash, what does this fix/enable/change over the defaults?
Most other RK3399 devices define the offset in the device tree, even for the rockpro64. However, the CONFIG_SYS_SPI_U_BOOT_OFFS is used as the fallback when device tree value is missing, and also as the offset used when generating the u-boot-rockchip-spi.bin using binman.
puma-rk3399_defconfig is the only other RK3399 board that also generates a u-boot-rockchip-spi.bin, has CONFIG_ROCKCHIP_SPI_IMAGE=y. That board override the simple-bin-spi fit offset in rk3399-puma-haikou-u-boot.dtsi instead of using the Kconfig option.
offset. Enable CONFIG_LTO to reduce size of SPL so that the mkimage output fit before the 0x60000 offset in u-boot-rockchip-spi.bin.
The enabling of LTO seems like a separate change TBH, the changes seem to be independent and there's no mention of it in the subject.
Without LTO enabled the idbloader.img grows too large that it does not fit before the u-boot.itb payload at 0x60000 and bulding of u-boot-rockchip-spi.bin fails the u-boot build.
In order to generate a valid u-boot-rockchip-spi.bin, LTO was required to be enabled, there is also a short mention of it in the commit message. The alternative would be to move the payload offset to e.g. 0x80000 but that felt like a too big and risky change.
Regards, Jonas
=> sf probe SF: Detected gd25q128 with page size 256 Bytes, erase size 4 KiB, total 16 MiB => load mmc 1:1 10000000 u-boot-rockchip-spi.bin 1442304 bytes read in 27 ms (50.9 MiB/s) => sf update $fileaddr 0 $filesize device 0 offset 0x0, size 0x160200 1421824 bytes written, 20480 bytes skipped in 9.501s, speed 155432 B/s
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com
v2:
- Collect r-b tag
configs/rockpro64-rk3399_defconfig | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 0ca2cecade25..f41c03067903 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -11,6 +11,7 @@ CONFIG_ENV_OFFSET=0x3F8000 CONFIG_DEFAULT_DEVICE_TREE="rk3399-rockpro64" CONFIG_DM_RESET=y CONFIG_ROCKCHIP_RK3399=y +CONFIG_ROCKCHIP_SPI_IMAGE=y CONFIG_TARGET_ROCKPRO64_RK3399=y CONFIG_SPL_STACK=0x400000 CONFIG_DEBUG_UART_BASE=0xFF1A0000 @@ -20,6 +21,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y @@ -37,6 +39,7 @@ CONFIG_SPL_BSS_MAX_SIZE=0x2000 CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x10000 CONFIG_SPL_SPI_LOAD=y +CONFIG_SYS_SPI_U_BOOT_OFFS=0x60000 CONFIG_TPL=y CONFIG_CMD_BOOTZ=y CONFIG_CMD_GPT=y -- 2.40.1
Hi Jonas,
This regresses the Rockpro64 build for me when building with gcc 12/13 with the following error, if I remove CONFIG_LTO it builds, but overlaps.
/usr/bin/aarch64-linux-gnu-ld: /usr/lib/gcc/aarch64-linux-gnu/13/libgcc.a(lse-init.o): in function `init_have_lse_atomics': /builddir/build/BUILD/gcc-13.1.1-20230519/aarch64-linux-gnu/aarch64-linux-gnu/libgcc/../../../gcc-13.1.1-20230519/libgcc/config/aarch64/lse-init.c:46: undefined reference to `__getauxval' collect2: fatal error: ld terminated with signal 11 [Segmentation fault], core dumped compilation terminated.
Peter
On Wed, May 17, 2023 at 7:41 PM Jonas Karlman jonas@kwiboo.se wrote:
Enable CONFIG_ROCKCHIP_SPI_IMAGE to build u-boot-rockchip-spi.bin. Define CONFIG_SYS_SPI_U_BOOT_OFFS to write u-boot.itb at the expected offset. Enable CONFIG_LTO to reduce size of SPL so that the mkimage output fit before the 0x60000 offset in u-boot-rockchip-spi.bin.
=> sf probe SF: Detected gd25q128 with page size 256 Bytes, erase size 4 KiB, total 16 MiB => load mmc 1:1 10000000 u-boot-rockchip-spi.bin 1442304 bytes read in 27 ms (50.9 MiB/s) => sf update $fileaddr 0 $filesize device 0 offset 0x0, size 0x160200 1421824 bytes written, 20480 bytes skipped in 9.501s, speed 155432 B/s
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com
v2:
- Collect r-b tag
configs/rockpro64-rk3399_defconfig | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 0ca2cecade25..f41c03067903 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -11,6 +11,7 @@ CONFIG_ENV_OFFSET=0x3F8000 CONFIG_DEFAULT_DEVICE_TREE="rk3399-rockpro64" CONFIG_DM_RESET=y CONFIG_ROCKCHIP_RK3399=y +CONFIG_ROCKCHIP_SPI_IMAGE=y CONFIG_TARGET_ROCKPRO64_RK3399=y CONFIG_SPL_STACK=0x400000 CONFIG_DEBUG_UART_BASE=0xFF1A0000 @@ -20,6 +21,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y @@ -37,6 +39,7 @@ CONFIG_SPL_BSS_MAX_SIZE=0x2000 CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x10000 CONFIG_SPL_SPI_LOAD=y +CONFIG_SYS_SPI_U_BOOT_OFFS=0x60000 CONFIG_TPL=y CONFIG_CMD_BOOTZ=y CONFIG_CMD_GPT=y -- 2.40.1
Hi Peter,
On 2023-06-11 22:22, Peter Robinson wrote:
Hi Jonas,
This regresses the Rockpro64 build for me when building with gcc 12/13 with the following error, if I remove CONFIG_LTO it builds, but overlaps.
/usr/bin/aarch64-linux-gnu-ld: /usr/lib/gcc/aarch64-linux-gnu/13/libgcc.a(lse-init.o): in function `init_have_lse_atomics': /builddir/build/BUILD/gcc-13.1.1-20230519/aarch64-linux-gnu/aarch64-linux-gnu/libgcc/../../../gcc-13.1.1-20230519/libgcc/config/aarch64/lse-init.c:46: undefined reference to `__getauxval' collect2: fatal error: ld terminated with signal 11 [Segmentation fault], core dumped compilation terminated.
Interesting, my test build using ubuntu-22.04 and gcc-12 did not seem to fail, see [1]. Compiling with -mno-outline-atomics flag may work, not sure that is a good workaround.
An alternative could be to move the payload to @ 512KB instead of @ 384KB.
configs/rockpro64-rk3399_defconfig: CONFIG_SYS_SPI_U_BOOT_OFFS=0x80000
arch/arm/dts/rk3399-rockpro64-u-boot.dtsi: u-boot,spl-payload-offset = <0x80000>;
[1] https://github.com/Kwiboo/u-boot-build/actions/runs/5248398191/jobs/94798279...
Regards, Jonas
Peter
On Wed, May 17, 2023 at 7:41 PM Jonas Karlman jonas@kwiboo.se wrote:
Enable CONFIG_ROCKCHIP_SPI_IMAGE to build u-boot-rockchip-spi.bin. Define CONFIG_SYS_SPI_U_BOOT_OFFS to write u-boot.itb at the expected offset. Enable CONFIG_LTO to reduce size of SPL so that the mkimage output fit before the 0x60000 offset in u-boot-rockchip-spi.bin.
=> sf probe SF: Detected gd25q128 with page size 256 Bytes, erase size 4 KiB, total 16 MiB => load mmc 1:1 10000000 u-boot-rockchip-spi.bin 1442304 bytes read in 27 ms (50.9 MiB/s) => sf update $fileaddr 0 $filesize device 0 offset 0x0, size 0x160200 1421824 bytes written, 20480 bytes skipped in 9.501s, speed 155432 B/s
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com
v2:
- Collect r-b tag
configs/rockpro64-rk3399_defconfig | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 0ca2cecade25..f41c03067903 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -11,6 +11,7 @@ CONFIG_ENV_OFFSET=0x3F8000 CONFIG_DEFAULT_DEVICE_TREE="rk3399-rockpro64" CONFIG_DM_RESET=y CONFIG_ROCKCHIP_RK3399=y +CONFIG_ROCKCHIP_SPI_IMAGE=y CONFIG_TARGET_ROCKPRO64_RK3399=y CONFIG_SPL_STACK=0x400000 CONFIG_DEBUG_UART_BASE=0xFF1A0000 @@ -20,6 +21,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y @@ -37,6 +39,7 @@ CONFIG_SPL_BSS_MAX_SIZE=0x2000 CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x10000 CONFIG_SPL_SPI_LOAD=y +CONFIG_SYS_SPI_U_BOOT_OFFS=0x60000 CONFIG_TPL=y CONFIG_CMD_BOOTZ=y CONFIG_CMD_GPT=y -- 2.40.1
On Mon, Jun 12, 2023 at 11:20 PM Jonas Karlman jonas@kwiboo.se wrote:
Hi Peter,
On 2023-06-11 22:22, Peter Robinson wrote:
Hi Jonas,
This regresses the Rockpro64 build for me when building with gcc 12/13 with the following error, if I remove CONFIG_LTO it builds, but overlaps.
/usr/bin/aarch64-linux-gnu-ld: /usr/lib/gcc/aarch64-linux-gnu/13/libgcc.a(lse-init.o): in function `init_have_lse_atomics': /builddir/build/BUILD/gcc-13.1.1-20230519/aarch64-linux-gnu/aarch64-linux-gnu/libgcc/../../../gcc-13.1.1-20230519/libgcc/config/aarch64/lse-init.c:46: undefined reference to `__getauxval' collect2: fatal error: ld terminated with signal 11 [Segmentation fault], core dumped compilation terminated.
Interesting, my test build using ubuntu-22.04 and gcc-12 did not seem to fail, see [1]. Compiling with -mno-outline-atomics flag may work, not sure that is a good workaround.
I would prefer a fix.
An alternative could be to move the payload to @ 512KB instead of @ 384KB.
What impact will moving this around have on upgrades? I am not sure we should be randomly moving stuff without knowing the impact TBH. Will this break existing users vs what you feel is appropriate for your usecase?
configs/rockpro64-rk3399_defconfig: CONFIG_SYS_SPI_U_BOOT_OFFS=0x80000
arch/arm/dts/rk3399-rockpro64-u-boot.dtsi: u-boot,spl-payload-offset = <0x80000>;
[1] https://github.com/Kwiboo/u-boot-build/actions/runs/5248398191/jobs/94798279...
Regards, Jonas
Peter
On Wed, May 17, 2023 at 7:41 PM Jonas Karlman jonas@kwiboo.se wrote:
Enable CONFIG_ROCKCHIP_SPI_IMAGE to build u-boot-rockchip-spi.bin. Define CONFIG_SYS_SPI_U_BOOT_OFFS to write u-boot.itb at the expected offset. Enable CONFIG_LTO to reduce size of SPL so that the mkimage output fit before the 0x60000 offset in u-boot-rockchip-spi.bin.
=> sf probe SF: Detected gd25q128 with page size 256 Bytes, erase size 4 KiB, total 16 MiB => load mmc 1:1 10000000 u-boot-rockchip-spi.bin 1442304 bytes read in 27 ms (50.9 MiB/s) => sf update $fileaddr 0 $filesize device 0 offset 0x0, size 0x160200 1421824 bytes written, 20480 bytes skipped in 9.501s, speed 155432 B/s
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com
v2:
- Collect r-b tag
configs/rockpro64-rk3399_defconfig | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 0ca2cecade25..f41c03067903 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -11,6 +11,7 @@ CONFIG_ENV_OFFSET=0x3F8000 CONFIG_DEFAULT_DEVICE_TREE="rk3399-rockpro64" CONFIG_DM_RESET=y CONFIG_ROCKCHIP_RK3399=y +CONFIG_ROCKCHIP_SPI_IMAGE=y CONFIG_TARGET_ROCKPRO64_RK3399=y CONFIG_SPL_STACK=0x400000 CONFIG_DEBUG_UART_BASE=0xFF1A0000 @@ -20,6 +21,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y @@ -37,6 +39,7 @@ CONFIG_SPL_BSS_MAX_SIZE=0x2000 CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x10000 CONFIG_SPL_SPI_LOAD=y +CONFIG_SYS_SPI_U_BOOT_OFFS=0x60000 CONFIG_TPL=y CONFIG_CMD_BOOTZ=y CONFIG_CMD_GPT=y -- 2.40.1
Hi Peter,
On 2023-06-13 09:58, Peter Robinson wrote:
On Mon, Jun 12, 2023 at 11:20 PM Jonas Karlman jonas@kwiboo.se wrote:
Hi Peter,
On 2023-06-11 22:22, Peter Robinson wrote:
Hi Jonas,
This regresses the Rockpro64 build for me when building with gcc 12/13 with the following error, if I remove CONFIG_LTO it builds, but overlaps.
/usr/bin/aarch64-linux-gnu-ld: /usr/lib/gcc/aarch64-linux-gnu/13/libgcc.a(lse-init.o): in function `init_have_lse_atomics': /builddir/build/BUILD/gcc-13.1.1-20230519/aarch64-linux-gnu/aarch64-linux-gnu/libgcc/../../../gcc-13.1.1-20230519/libgcc/config/aarch64/lse-init.c:46: undefined reference to `__getauxval' collect2: fatal error: ld terminated with signal 11 [Segmentation fault], core dumped compilation terminated.
Interesting, my test build using ubuntu-22.04 and gcc-12 did not seem to fail, see [1]. Compiling with -mno-outline-atomics flag may work, not sure that is a good workaround.
I would prefer a fix.
Not sure what such fix could be, cross compiling in CI and under ubuntu 22.04 does not trigger this linking issue, not sure what is different in your build environment to not allow build with CONFIG_LTO=y.
Searching for similar linking issue suggest it could be related to outline-atomics, do you use any special C/LDFLAGS or similar?
An alternative could be to move the payload to @ 512KB instead of @ 384KB.
What impact will moving this around have on upgrades? I am not sure we should be randomly moving stuff without knowing the impact TBH. Will this break existing users vs what you feel is appropriate for your usecase?
I did a re-calculation and using @ 512 KB offset may not be enough for a worst-case scenario.
mkimage will enforce a 184 KB limit for the TPL on RK3399. With SPL loaded at 0x0 and TF-A loaded to 0x40000, we have a 256 KB size limit of SPL. BootRom that loads both TPL and SPL only read first 2 KB of every 4 KB page of SPI flash, so idbloader.img (TPL+SPL) could grow up to ~880 KB in a worst-case scenario. (184+256) x 2 = 880
Using a payload offset of @ 896 KB (0xE0000) is probably the safest default option if we need to revert the CONFIG_LTO=y change.
As long as u-boot-rockchip-spi.bin is used to update SPI flash changing offset should not have an impact. However, there are guides and scripts that write idbloader.img and u-boot.itb separately, those could break.
Such guides and scripts can already lead to users overwriting part of SPL when the produced idbloader.img is over 384 KB in size. With CONFIG_ROCKCHIP_SPI_IMAGE=y the size of idbloader.img is instead enforced to not cause an overlap of the u-boot.itb at build time.
Regards, Jonas
configs/rockpro64-rk3399_defconfig: CONFIG_SYS_SPI_U_BOOT_OFFS=0x80000
arch/arm/dts/rk3399-rockpro64-u-boot.dtsi: u-boot,spl-payload-offset = <0x80000>;
[1] https://github.com/Kwiboo/u-boot-build/actions/runs/5248398191/jobs/94798279...
Regards, Jonas
Peter
On Wed, May 17, 2023 at 7:41 PM Jonas Karlman jonas@kwiboo.se wrote:
Enable CONFIG_ROCKCHIP_SPI_IMAGE to build u-boot-rockchip-spi.bin. Define CONFIG_SYS_SPI_U_BOOT_OFFS to write u-boot.itb at the expected offset. Enable CONFIG_LTO to reduce size of SPL so that the mkimage output fit before the 0x60000 offset in u-boot-rockchip-spi.bin.
=> sf probe SF: Detected gd25q128 with page size 256 Bytes, erase size 4 KiB, total 16 MiB => load mmc 1:1 10000000 u-boot-rockchip-spi.bin 1442304 bytes read in 27 ms (50.9 MiB/s) => sf update $fileaddr 0 $filesize device 0 offset 0x0, size 0x160200 1421824 bytes written, 20480 bytes skipped in 9.501s, speed 155432 B/s
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com
v2:
- Collect r-b tag
configs/rockpro64-rk3399_defconfig | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 0ca2cecade25..f41c03067903 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -11,6 +11,7 @@ CONFIG_ENV_OFFSET=0x3F8000 CONFIG_DEFAULT_DEVICE_TREE="rk3399-rockpro64" CONFIG_DM_RESET=y CONFIG_ROCKCHIP_RK3399=y +CONFIG_ROCKCHIP_SPI_IMAGE=y CONFIG_TARGET_ROCKPRO64_RK3399=y CONFIG_SPL_STACK=0x400000 CONFIG_DEBUG_UART_BASE=0xFF1A0000 @@ -20,6 +21,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y @@ -37,6 +39,7 @@ CONFIG_SPL_BSS_MAX_SIZE=0x2000 CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x10000 CONFIG_SPL_SPI_LOAD=y +CONFIG_SYS_SPI_U_BOOT_OFFS=0x60000 CONFIG_TPL=y CONFIG_CMD_BOOTZ=y CONFIG_CMD_GPT=y -- 2.40.1
On Tue, Jun 13, 2023 at 7:48 PM Jonas Karlman jonas@kwiboo.se wrote:
Hi Peter,
On 2023-06-13 09:58, Peter Robinson wrote:
On Mon, Jun 12, 2023 at 11:20 PM Jonas Karlman jonas@kwiboo.se wrote:
Hi Peter,
On 2023-06-11 22:22, Peter Robinson wrote:
Hi Jonas,
This regresses the Rockpro64 build for me when building with gcc 12/13 with the following error, if I remove CONFIG_LTO it builds, but overlaps.
/usr/bin/aarch64-linux-gnu-ld: /usr/lib/gcc/aarch64-linux-gnu/13/libgcc.a(lse-init.o): in function `init_have_lse_atomics': /builddir/build/BUILD/gcc-13.1.1-20230519/aarch64-linux-gnu/aarch64-linux-gnu/libgcc/../../../gcc-13.1.1-20230519/libgcc/config/aarch64/lse-init.c:46: undefined reference to `__getauxval' collect2: fatal error: ld terminated with signal 11 [Segmentation fault], core dumped compilation terminated.
Interesting, my test build using ubuntu-22.04 and gcc-12 did not seem to fail, see [1]. Compiling with -mno-outline-atomics flag may work, not sure that is a good workaround.
I would prefer a fix.
Not sure what such fix could be, cross compiling in CI and under ubuntu 22.04 does not trigger this linking issue, not sure what is different in your build environment to not allow build with CONFIG_LTO=y.
Searching for similar linking issue suggest it could be related to outline-atomics, do you use any special C/LDFLAGS or similar?
An alternative could be to move the payload to @ 512KB instead of @ 384KB.
What impact will moving this around have on upgrades? I am not sure we should be randomly moving stuff without knowing the impact TBH. Will this break existing users vs what you feel is appropriate for your usecase?
I did a re-calculation and using @ 512 KB offset may not be enough for a worst-case scenario.
mkimage will enforce a 184 KB limit for the TPL on RK3399. With SPL loaded at 0x0 and TF-A loaded to 0x40000, we have a 256 KB size limit of SPL. BootRom that loads both TPL and SPL only read first 2 KB of every 4 KB page of SPI flash, so idbloader.img (TPL+SPL) could grow up to ~880 KB in a worst-case scenario. (184+256) x 2 = 880
Using a payload offset of @ 896 KB (0xE0000) is probably the safest default option if we need to revert the CONFIG_LTO=y change.
As long as u-boot-rockchip-spi.bin is used to update SPI flash changing offset should not have an impact. However, there are guides and scripts that write idbloader.img and u-boot.itb separately, those could break.
Well I believe your patch only enabled that option by default so most users I suspect write them separately up until now, does the new offsets affect existing env variables at all?
Such guides and scripts can already lead to users overwriting part of SPL when the produced idbloader.img is over 384 KB in size. With CONFIG_ROCKCHIP_SPI_IMAGE=y the size of idbloader.img is instead enforced to not cause an overlap of the u-boot.itb at build time.
Ultimately I would prefer to not actively break existing users if they're not aware of changes, and that is hard to communicate. People with broken devices tend to scream a lot.
Regards, Jonas
configs/rockpro64-rk3399_defconfig: CONFIG_SYS_SPI_U_BOOT_OFFS=0x80000
arch/arm/dts/rk3399-rockpro64-u-boot.dtsi: u-boot,spl-payload-offset = <0x80000>;
[1] https://github.com/Kwiboo/u-boot-build/actions/runs/5248398191/jobs/94798279...
Regards, Jonas
Peter
On Wed, May 17, 2023 at 7:41 PM Jonas Karlman jonas@kwiboo.se wrote:
Enable CONFIG_ROCKCHIP_SPI_IMAGE to build u-boot-rockchip-spi.bin. Define CONFIG_SYS_SPI_U_BOOT_OFFS to write u-boot.itb at the expected offset. Enable CONFIG_LTO to reduce size of SPL so that the mkimage output fit before the 0x60000 offset in u-boot-rockchip-spi.bin.
=> sf probe SF: Detected gd25q128 with page size 256 Bytes, erase size 4 KiB, total 16 MiB => load mmc 1:1 10000000 u-boot-rockchip-spi.bin 1442304 bytes read in 27 ms (50.9 MiB/s) => sf update $fileaddr 0 $filesize device 0 offset 0x0, size 0x160200 1421824 bytes written, 20480 bytes skipped in 9.501s, speed 155432 B/s
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com
v2:
- Collect r-b tag
configs/rockpro64-rk3399_defconfig | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 0ca2cecade25..f41c03067903 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -11,6 +11,7 @@ CONFIG_ENV_OFFSET=0x3F8000 CONFIG_DEFAULT_DEVICE_TREE="rk3399-rockpro64" CONFIG_DM_RESET=y CONFIG_ROCKCHIP_RK3399=y +CONFIG_ROCKCHIP_SPI_IMAGE=y CONFIG_TARGET_ROCKPRO64_RK3399=y CONFIG_SPL_STACK=0x400000 CONFIG_DEBUG_UART_BASE=0xFF1A0000 @@ -20,6 +21,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y @@ -37,6 +39,7 @@ CONFIG_SPL_BSS_MAX_SIZE=0x2000 CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x10000 CONFIG_SPL_SPI_LOAD=y +CONFIG_SYS_SPI_U_BOOT_OFFS=0x60000 CONFIG_TPL=y CONFIG_CMD_BOOTZ=y CONFIG_CMD_GPT=y -- 2.40.1
On 2023-06-14 09:46, Peter Robinson wrote:
On Tue, Jun 13, 2023 at 7:48 PM Jonas Karlman jonas@kwiboo.se wrote:
Hi Peter,
On 2023-06-13 09:58, Peter Robinson wrote:
On Mon, Jun 12, 2023 at 11:20 PM Jonas Karlman jonas@kwiboo.se wrote:
Hi Peter,
On 2023-06-11 22:22, Peter Robinson wrote:
Hi Jonas,
This regresses the Rockpro64 build for me when building with gcc 12/13 with the following error, if I remove CONFIG_LTO it builds, but overlaps.
/usr/bin/aarch64-linux-gnu-ld: /usr/lib/gcc/aarch64-linux-gnu/13/libgcc.a(lse-init.o): in function `init_have_lse_atomics': /builddir/build/BUILD/gcc-13.1.1-20230519/aarch64-linux-gnu/aarch64-linux-gnu/libgcc/../../../gcc-13.1.1-20230519/libgcc/config/aarch64/lse-init.c:46: undefined reference to `__getauxval' collect2: fatal error: ld terminated with signal 11 [Segmentation fault], core dumped compilation terminated.
Interesting, my test build using ubuntu-22.04 and gcc-12 did not seem to fail, see [1]. Compiling with -mno-outline-atomics flag may work, not sure that is a good workaround.
I would prefer a fix.
Not sure what such fix could be, cross compiling in CI and under ubuntu 22.04 does not trigger this linking issue, not sure what is different in your build environment to not allow build with CONFIG_LTO=y.
Searching for similar linking issue suggest it could be related to outline-atomics, do you use any special C/LDFLAGS or similar?
An alternative could be to move the payload to @ 512KB instead of @ 384KB.
What impact will moving this around have on upgrades? I am not sure we should be randomly moving stuff without knowing the impact TBH. Will this break existing users vs what you feel is appropriate for your usecase?
I did a re-calculation and using @ 512 KB offset may not be enough for a worst-case scenario.
mkimage will enforce a 184 KB limit for the TPL on RK3399. With SPL loaded at 0x0 and TF-A loaded to 0x40000, we have a 256 KB size limit of SPL. BootRom that loads both TPL and SPL only read first 2 KB of every 4 KB page of SPI flash, so idbloader.img (TPL+SPL) could grow up to ~880 KB in a worst-case scenario. (184+256) x 2 = 880
Using a payload offset of @ 896 KB (0xE0000) is probably the safest default option if we need to revert the CONFIG_LTO=y change.
As long as u-boot-rockchip-spi.bin is used to update SPI flash changing offset should not have an impact. However, there are guides and scripts that write idbloader.img and u-boot.itb separately, those could break.
Well I believe your patch only enabled that option by default so most users I suspect write them separately up until now, does the new offsets affect existing env variables at all?
It does not look like it will overwrite env offset.
CONFIG_ENV_SIZE=0x8000 CONFIG_ENV_OFFSET=0x3F8000
There will be room for a 3168 KB u-boot.itb payload until it would overlap env in SPI flash with a 0xE0000 offset for u-boot.itb.
Also look like many RK3399 defconfig have a SPL_MAX_SIZE=0x2e000, something that could be increased to SPL_MAX_SIZE=0x40000 with an updated offset for u-boot.itb.
I will prepare a small fix series that address your LTO issue and use proper offsets and max size options.
Such guides and scripts can already lead to users overwriting part of SPL when the produced idbloader.img is over 384 KB in size. With CONFIG_ROCKCHIP_SPI_IMAGE=y the size of idbloader.img is instead enforced to not cause an overlap of the u-boot.itb at build time.
Ultimately I would prefer to not actively break existing users if they're not aware of changes, and that is hard to communicate. People with broken devices tend to scream a lot.
Such change will not actively break existing users, SPL look for a FIT header and will fall back to next media in u-boot,spl-boot-order.
Will include an update to SPI flash section of rockchip.rst to mention how to properly flash to correct offset using u-boot-rockchip-spi.bin in the fix series.
Regards, Jonas
Regards, Jonas
configs/rockpro64-rk3399_defconfig: CONFIG_SYS_SPI_U_BOOT_OFFS=0x80000
arch/arm/dts/rk3399-rockpro64-u-boot.dtsi: u-boot,spl-payload-offset = <0x80000>;
[1] https://github.com/Kwiboo/u-boot-build/actions/runs/5248398191/jobs/94798279...
Regards, Jonas
Peter
On Wed, May 17, 2023 at 7:41 PM Jonas Karlman jonas@kwiboo.se wrote:
Enable CONFIG_ROCKCHIP_SPI_IMAGE to build u-boot-rockchip-spi.bin. Define CONFIG_SYS_SPI_U_BOOT_OFFS to write u-boot.itb at the expected offset. Enable CONFIG_LTO to reduce size of SPL so that the mkimage output fit before the 0x60000 offset in u-boot-rockchip-spi.bin.
=> sf probe SF: Detected gd25q128 with page size 256 Bytes, erase size 4 KiB, total 16 MiB => load mmc 1:1 10000000 u-boot-rockchip-spi.bin 1442304 bytes read in 27 ms (50.9 MiB/s) => sf update $fileaddr 0 $filesize device 0 offset 0x0, size 0x160200 1421824 bytes written, 20480 bytes skipped in 9.501s, speed 155432 B/s
Signed-off-by: Jonas Karlman jonas@kwiboo.se Reviewed-by: Kever Yang kever.yang@rock-chips.com
v2:
- Collect r-b tag
configs/rockpro64-rk3399_defconfig | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/configs/rockpro64-rk3399_defconfig b/configs/rockpro64-rk3399_defconfig index 0ca2cecade25..f41c03067903 100644 --- a/configs/rockpro64-rk3399_defconfig +++ b/configs/rockpro64-rk3399_defconfig @@ -11,6 +11,7 @@ CONFIG_ENV_OFFSET=0x3F8000 CONFIG_DEFAULT_DEVICE_TREE="rk3399-rockpro64" CONFIG_DM_RESET=y CONFIG_ROCKCHIP_RK3399=y +CONFIG_ROCKCHIP_SPI_IMAGE=y CONFIG_TARGET_ROCKPRO64_RK3399=y CONFIG_SPL_STACK=0x400000 CONFIG_DEBUG_UART_BASE=0xFF1A0000 @@ -20,6 +21,7 @@ CONFIG_SPL_SPI=y CONFIG_SYS_LOAD_ADDR=0x800800 CONFIG_PCI=y CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_SPL_FIT_SIGNATURE=y CONFIG_BOOTSTAGE=y CONFIG_BOOTSTAGE_REPORT=y @@ -37,6 +39,7 @@ CONFIG_SPL_BSS_MAX_SIZE=0x2000 CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x10000 CONFIG_SPL_SPI_LOAD=y +CONFIG_SYS_SPI_U_BOOT_OFFS=0x60000 CONFIG_TPL=y CONFIG_CMD_BOOTZ=y CONFIG_CMD_GPT=y -- 2.40.1
On Wed, May 17, 2023 at 7:40 PM Jonas Karlman jonas@kwiboo.se wrote:
The eMMC performance on RK3399 was reduced sigificant by the commit 2cc6cde647e2 ("mmc: rockchip_sdhci: Limit number of blocks read in a single command").
That workaround should only have been applied to RK3568 and RK3588. This series fixes that and also help boost eMMC performance on two RK3399 boards even more by enabling use of SDMA.
Is there a reason to do this on just two devices?
There is also an extra commit to help build a u-boot-rockchip-spi.bin image that can be used for SPI flash boot on RockPro64.
Changes in v2:
- Rebase on top of defconfig and spi v2 series
- Collect r-b and t-b tags
Jonas Karlman (5): mmc: rockchip_sdhci: Skip blocks read workaround on RK3399 mmc: rockchip_sdhci: Disable DMA mode using a device tree property rockchip: rockpro64: Use SDMA to boost eMMC performance rockchip: rock-pi-4: Use SDMA to boost eMMC performance rockchip: rockpro64: Build u-boot-rockchip-spi.bin
arch/arm/dts/rk3399-rock-pi-4-u-boot.dtsi | 6 ++++++ arch/arm/dts/rk3399-rockpro64-u-boot.dtsi | 5 +++++ arch/arm/dts/rk3399-u-boot.dtsi | 1 + arch/arm/dts/rk3588s-u-boot.dtsi | 1 + configs/rock-pi-4-rk3399_defconfig | 2 ++ configs/rock5b-rk3588_defconfig | 1 - configs/rockpro64-rk3399_defconfig | 5 +++++ drivers/mmc/rockchip_sdhci.c | 12 +++++++++++- 8 files changed, 31 insertions(+), 2 deletions(-)
-- 2.40.1
Hi Peter, On 2023-05-17 21:05, Peter Robinson wrote:
On Wed, May 17, 2023 at 7:40 PM Jonas Karlman jonas@kwiboo.se wrote:
The eMMC performance on RK3399 was reduced sigificant by the commit 2cc6cde647e2 ("mmc: rockchip_sdhci: Limit number of blocks read in a single command").
That workaround should only have been applied to RK3568 and RK3588. This series fixes that and also help boost eMMC performance on two RK3399 boards even more by enabling use of SDMA.
Is there a reason to do this on just two devices?
DMA use can and should probably be enabled on other RK3399 devices, these two where the only ones I could runtime test on.
Regards, Jonas
There is also an extra commit to help build a u-boot-rockchip-spi.bin image that can be used for SPI flash boot on RockPro64.
Changes in v2:
- Rebase on top of defconfig and spi v2 series
- Collect r-b and t-b tags
Jonas Karlman (5): mmc: rockchip_sdhci: Skip blocks read workaround on RK3399 mmc: rockchip_sdhci: Disable DMA mode using a device tree property rockchip: rockpro64: Use SDMA to boost eMMC performance rockchip: rock-pi-4: Use SDMA to boost eMMC performance rockchip: rockpro64: Build u-boot-rockchip-spi.bin
arch/arm/dts/rk3399-rock-pi-4-u-boot.dtsi | 6 ++++++ arch/arm/dts/rk3399-rockpro64-u-boot.dtsi | 5 +++++ arch/arm/dts/rk3399-u-boot.dtsi | 1 + arch/arm/dts/rk3588s-u-boot.dtsi | 1 + configs/rock-pi-4-rk3399_defconfig | 2 ++ configs/rock5b-rk3588_defconfig | 1 - configs/rockpro64-rk3399_defconfig | 5 +++++ drivers/mmc/rockchip_sdhci.c | 12 +++++++++++- 8 files changed, 31 insertions(+), 2 deletions(-)
-- 2.40.1
participants (2)
-
Jonas Karlman -
Peter Robinson