RE: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO
Hello Eugen, I have successfully tested the binary on sama5d2_icp board. Board is booting OK the zImage from fat (SD-card). Tested-by: Mihai Sain Best regards, Mihai Sain -Original Message- From: Eugen Hristev Sent: Monday, July 24, 2023 2:08 PM To: u-boot@lists.denx.de; Mihai Sain - M19926 Cc: Pali Rohár ; Stefan Roese Subject: Re: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe On 4/27/23 11:59, Stefan Roese wrote: > Adding just a tiny bit more code for sama5d2_icp_mmc leads to a SRAM > image overflow. Fix this by enabling LTO for this board, so that such > changes still can be made to the common U-Boot code. > > Signed-off-by: Stefan Roese > Cc: Tudor Ambarus > Cc: Eugen Hristev > Cc: Sergiu Moga > Cc: Pali Rohár > --- > configs/sama5d2_icp_mmc_defconfig | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > diff --git a/configs/sama5d2_icp_mmc_defconfig > b/configs/sama5d2_icp_mmc_defconfig > index e1b602d8e5ec..a3c57a3f1250 100644 > --- a/configs/sama5d2_icp_mmc_defconfig > +++ b/configs/sama5d2_icp_mmc_defconfig > @@ -9,9 +9,11 @@ CONFIG_SPL_LIBCOMMON_SUPPORT=y > CONFIG_SPL_LIBGENERIC_SUPPORT=y > CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y > CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0x20003ef0 > +CONFIG_SF_DEFAULT_SPEED=6600 > CONFIG_ENV_SIZE=0x4000 > CONFIG_DM_GPIO=y > CONFIG_DEFAULT_DEVICE_TREE="at91-sama5d2_icp" > +CONFIG_OF_LIBFDT_OVERLAY=y > CONFIG_SPL_MMC=y > CONFIG_SPL_SERIAL=y > CONFIG_SPL_DRIVERS_MISC=y > @@ -24,6 +26,7 @@ CONFIG_SPL_FS_FAT=y > CONFIG_SPL_LIBDISK_SUPPORT=y > CONFIG_SYS_LOAD_ADDR=0x2200 > CONFIG_DEBUG_UART=y > +CONFIG_LTO=y > CONFIG_ENV_VARS_UBOOT_CONFIG=y > CONFIG_SYS_MONITOR_LEN=524288 > CONFIG_FIT=y > @@ -86,7 +89,6 @@ CONFIG_MMC_SDHCI=y > CONFIG_MMC_SDHCI_ATMEL=y > CONFIG_DM_SPI_FLASH=y > CONFIG_SF_DEFAULT_BUS=2 > -CONFIG_SF_DEFAULT_SPEED=6600 > CONFIG_SPI_FLASH_SFDP_SUPPORT=y > CONFIG_SPI_FLASH_ATMEL=y > CONFIG_SPI_FLASH_MACRONIX=y > @@ -110,5 +112,4 @@ CONFIG_TIMER=y > CONFIG_SPL_TIMER=y > CONFIG_ATMEL_TCB_TIMER=y > CONFIG_SPL_ATMEL_TCB_TIMER=y > -CONFIG_OF_LIBFDT_OVERLAY=y > # CONFIG_EFI_LOADER_HII is not set Mihai (added to the thread) tested this patch on sama5d2_icp. Mihai, can you add your Tested-by tag then? This is the original LTO patch for sama5d2_icp that we discussed about. Thanks, Eugen
Re: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO
On 4/27/23 11:59, Stefan Roese wrote: Adding just a tiny bit more code for sama5d2_icp_mmc leads to a SRAM image overflow. Fix this by enabling LTO for this board, so that such changes still can be made to the common U-Boot code. Signed-off-by: Stefan Roese Cc: Tudor Ambarus Cc: Eugen Hristev Cc: Sergiu Moga Cc: Pali Rohár --- configs/sama5d2_icp_mmc_defconfig | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/configs/sama5d2_icp_mmc_defconfig b/configs/sama5d2_icp_mmc_defconfig index e1b602d8e5ec..a3c57a3f1250 100644 --- a/configs/sama5d2_icp_mmc_defconfig +++ b/configs/sama5d2_icp_mmc_defconfig @@ -9,9 +9,11 @@ CONFIG_SPL_LIBCOMMON_SUPPORT=y CONFIG_SPL_LIBGENERIC_SUPPORT=y CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0x20003ef0 +CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_ENV_SIZE=0x4000 CONFIG_DM_GPIO=y CONFIG_DEFAULT_DEVICE_TREE="at91-sama5d2_icp" +CONFIG_OF_LIBFDT_OVERLAY=y CONFIG_SPL_MMC=y CONFIG_SPL_SERIAL=y CONFIG_SPL_DRIVERS_MISC=y @@ -24,6 +26,7 @@ CONFIG_SPL_FS_FAT=y CONFIG_SPL_LIBDISK_SUPPORT=y CONFIG_SYS_LOAD_ADDR=0x2200 CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_ENV_VARS_UBOOT_CONFIG=y CONFIG_SYS_MONITOR_LEN=524288 CONFIG_FIT=y @@ -86,7 +89,6 @@ CONFIG_MMC_SDHCI=y CONFIG_MMC_SDHCI_ATMEL=y CONFIG_DM_SPI_FLASH=y CONFIG_SF_DEFAULT_BUS=2 -CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_SPI_FLASH_SFDP_SUPPORT=y CONFIG_SPI_FLASH_ATMEL=y CONFIG_SPI_FLASH_MACRONIX=y @@ -110,5 +112,4 @@ CONFIG_TIMER=y CONFIG_SPL_TIMER=y CONFIG_ATMEL_TCB_TIMER=y CONFIG_SPL_ATMEL_TCB_TIMER=y -CONFIG_OF_LIBFDT_OVERLAY=y # CONFIG_EFI_LOADER_HII is not set Mihai (added to the thread) tested this patch on sama5d2_icp. Mihai, can you add your Tested-by tag then? This is the original LTO patch for sama5d2_icp that we discussed about. Thanks, Eugen
Re: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO
On 03/05/2023 at 11:51, Eugen Hristev wrote: > [You don't often get email from eugen.hris...@collabora.com. Learn why this > is important at https://aka.ms/LearnAboutSenderIdentification ] > > EXTERNAL EMAIL: Do not click links or open attachments unless you know the > content is safe > > On 4/27/23 19:55, Pali Rohár wrote: >> On Thursday 27 April 2023 12:30:43 Stefan Roese wrote: >>> On 4/27/23 11:51, Eugen Hristev wrote: On 4/27/23 12:26, Stefan Roese wrote: > Hi Eugen, > > On 4/27/23 11:19, Eugen Hristev wrote: >> Hi Stefan, >> >> Thank you for the patch. >> >> This I guess is a workaround such that you can add a bit more of code. >> In the end, it's not scalable, and we have to find a better way, >> probably by removing some of the code to make the SPL smaller. > > U-Boot image size increase resulting in overflowing some limits is > a common problem, especially in SPL. Enabling LTO gives quite some good > improvements in image size decrease. So I don't think it's an > workaround. If this was not needed until today, and not adding any new functionality, I would call this a workaround to avoid shrinking the size to fit in the SRAM. When we are adding more and more, and eventually hit this problem again, LTO already enabled, what we will do ? That's why I call this a workaround because we are not solving the problem, just postponing so we can add more things today. >>> >>> This is what's happening since many years. But okay, let's call it a >>> workaround. >>> > >> How does this impact the size? How much we are gaining ? > > I did not measure this. I just checked that this target compiles clean > again with LTO enabled and the MMC related patches applied. > > Could you (or some college?) please investigate here, how the results > are in image size? No, you are submitting the patch, I assume you could give some numbers to support your patch. >>> >>> Sorry, my time is limited and frankly, I don't feel very much motivated >>> (any more) to do additional work here. Even if it's not that much of >>> effort. >>> > >> We can perhaps have a look to see which code is removed and >> guard it by #ifndef SPL_BUILD and that might lower the size. (if >> ofcourse, this code should really be removed) > > Sure, other improvements in image size decrease are of course always a > good idea. > >> Also, I don't have a board at hand to test this, so it has to be >> tested first to make sure the board doesn't break. > > Agreed. I assume/hope that one of your colleges will be able to test > this? Someone from Microchip can, or other people using the board from the community I no longer work for Microchip, but I am still maintaining the AT91 custodian tree >>> >>> Okay. Let's see, where this goes. I'm monitoring that, with help from Eugen. >> >> Well, if nobody wants to care about this board, go ahead with this >> change and if it is not enough that drop support for this board. Well, give us some time, please! > Hi Pali, > > I kind of dislike this attitude. If a patchset breaks a board, a > patchset should be changed. Or rejected. > I don't agree with removing boards just because in a few days nobody > tested one patch. > And applying untested patches is again something which I disagree upon. Thanks Eugen for your support. We didn't ask for this situation, so Pali, understand that we need to get organized before giving an answer / testing. Best regards, Nicolas >> On 4/27/23 11:59, Stefan Roese wrote: >>> Adding just a tiny bit more code for sama5d2_icp_mmc leads to a SRAM >>> image overflow. Fix this by enabling LTO for this board, so that such >>> changes still can be made to the common U-Boot code. >>> >>> Signed-off-by: Stefan Roese >>> Cc: Tudor Ambarus >>> Cc: Eugen Hristev >>> Cc: Sergiu Moga >>> Cc: Pali Rohár >>> --- >>> configs/sama5d2_icp_mmc_defconfig | 5 +++-- >>> 1 file changed, 3 insertions(+), 2 deletions(-) >>> >>> diff --git a/configs/sama5d2_icp_mmc_defconfig >>> b/configs/sama5d2_icp_mmc_defconfig >>> index e1b602d8e5ec..a3c57a3f1250 100644 >>> --- a/configs/sama5d2_icp_mmc_defconfig >>> +++ b/configs/sama5d2_icp_mmc_defconfig >>> @@ -9,9 +9,11 @@ CONFIG_SPL_LIBCOMMON_SUPPORT=y >>> CONFIG_SPL_LIBGENERIC_SUPPORT=y >>> CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y >>> CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0x20003ef0 >>> +CONFIG_SF_DEFAULT_SPEED=6600 >>> CONFIG_ENV_SIZE=0x4000 >>> CONFIG_DM_GPIO=y >>> CONFIG_DEFAULT_DEVICE_TREE="at91-sama5d2_icp" >>> +CONFIG_OF_LIBFDT_OVERLAY=y >>> CONFIG_SPL_MMC=y >>> CONFIG_SPL_SERIAL=y >>> CONFIG_SPL_DRIVERS_MISC=y >>> @@ -24,6 +26,7 @@ CONFIG_SPL_FS_FAT=y >>>
Re: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO
On 4/27/23 19:55, Pali Rohár wrote: On Thursday 27 April 2023 12:30:43 Stefan Roese wrote: On 4/27/23 11:51, Eugen Hristev wrote: On 4/27/23 12:26, Stefan Roese wrote: Hi Eugen, On 4/27/23 11:19, Eugen Hristev wrote: Hi Stefan, Thank you for the patch. This I guess is a workaround such that you can add a bit more of code. In the end, it's not scalable, and we have to find a better way, probably by removing some of the code to make the SPL smaller. U-Boot image size increase resulting in overflowing some limits is a common problem, especially in SPL. Enabling LTO gives quite some good improvements in image size decrease. So I don't think it's an workaround. If this was not needed until today, and not adding any new functionality, I would call this a workaround to avoid shrinking the size to fit in the SRAM. When we are adding more and more, and eventually hit this problem again, LTO already enabled, what we will do ? That's why I call this a workaround because we are not solving the problem, just postponing so we can add more things today. This is what's happening since many years. But okay, let's call it a workaround. How does this impact the size? How much we are gaining ? I did not measure this. I just checked that this target compiles clean again with LTO enabled and the MMC related patches applied. Could you (or some college?) please investigate here, how the results are in image size? No, you are submitting the patch, I assume you could give some numbers to support your patch. Sorry, my time is limited and frankly, I don't feel very much motivated (any more) to do additional work here. Even if it's not that much of effort. We can perhaps have a look to see which code is removed and guard it by #ifndef SPL_BUILD and that might lower the size. (if ofcourse, this code should really be removed) Sure, other improvements in image size decrease are of course always a good idea. Also, I don't have a board at hand to test this, so it has to be tested first to make sure the board doesn't break. Agreed. I assume/hope that one of your colleges will be able to test this? Someone from Microchip can, or other people using the board from the community I no longer work for Microchip, but I am still maintaining the AT91 custodian tree Okay. Let's see, where this goes. Well, if nobody wants to care about this board, go ahead with this change and if it is not enough that drop support for this board. Hi Pali, I kind of dislike this attitude. If a patchset breaks a board, a patchset should be changed. Or rejected. I don't agree with removing boards just because in a few days nobody tested one patch. And applying untested patches is again something which I disagree upon. Thanks, Stefan Eugen Thanks, Stefan Eugen On 4/27/23 11:59, Stefan Roese wrote: Adding just a tiny bit more code for sama5d2_icp_mmc leads to a SRAM image overflow. Fix this by enabling LTO for this board, so that such changes still can be made to the common U-Boot code. Signed-off-by: Stefan Roese Cc: Tudor Ambarus Cc: Eugen Hristev Cc: Sergiu Moga Cc: Pali Rohár --- configs/sama5d2_icp_mmc_defconfig | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/configs/sama5d2_icp_mmc_defconfig b/configs/sama5d2_icp_mmc_defconfig index e1b602d8e5ec..a3c57a3f1250 100644 --- a/configs/sama5d2_icp_mmc_defconfig +++ b/configs/sama5d2_icp_mmc_defconfig @@ -9,9 +9,11 @@ CONFIG_SPL_LIBCOMMON_SUPPORT=y CONFIG_SPL_LIBGENERIC_SUPPORT=y CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0x20003ef0 +CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_ENV_SIZE=0x4000 CONFIG_DM_GPIO=y CONFIG_DEFAULT_DEVICE_TREE="at91-sama5d2_icp" +CONFIG_OF_LIBFDT_OVERLAY=y CONFIG_SPL_MMC=y CONFIG_SPL_SERIAL=y CONFIG_SPL_DRIVERS_MISC=y @@ -24,6 +26,7 @@ CONFIG_SPL_FS_FAT=y CONFIG_SPL_LIBDISK_SUPPORT=y CONFIG_SYS_LOAD_ADDR=0x2200 CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_ENV_VARS_UBOOT_CONFIG=y CONFIG_SYS_MONITOR_LEN=524288 CONFIG_FIT=y @@ -86,7 +89,6 @@ CONFIG_MMC_SDHCI=y CONFIG_MMC_SDHCI_ATMEL=y CONFIG_DM_SPI_FLASH=y CONFIG_SF_DEFAULT_BUS=2 -CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_SPI_FLASH_SFDP_SUPPORT=y CONFIG_SPI_FLASH_ATMEL=y CONFIG_SPI_FLASH_MACRONIX=y @@ -110,5 +112,4 @@ CONFIG_TIMER=y CONFIG_SPL_TIMER=y CONFIG_ATMEL_TCB_TIMER=y CONFIG_SPL_ATMEL_TCB_TIMER=y -CONFIG_OF_LIBFDT_OVERLAY=y # CONFIG_EFI_LOADER_HII is not set Viele Grüße, Stefan Roese Viele Grüße, Stefan Roese -- DENX Software Engineering GmbH, Managing Director: Erika Unter HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-51 Fax: (+49)-8142-66989-80 Email: s...@denx.de
Re: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO
On Thursday 27 April 2023 12:30:43 Stefan Roese wrote: > On 4/27/23 11:51, Eugen Hristev wrote: > > On 4/27/23 12:26, Stefan Roese wrote: > > > Hi Eugen, > > > > > > On 4/27/23 11:19, Eugen Hristev wrote: > > > > Hi Stefan, > > > > > > > > Thank you for the patch. > > > > > > > > This I guess is a workaround such that you can add a bit more of code. > > > > In the end, it's not scalable, and we have to find a better way, > > > > probably by removing some of the code to make the SPL smaller. > > > > > > U-Boot image size increase resulting in overflowing some limits is > > > a common problem, especially in SPL. Enabling LTO gives quite some good > > > improvements in image size decrease. So I don't think it's an > > > workaround. > > > > If this was not needed until today, and not adding any new > > functionality, I would call this a workaround to avoid shrinking the > > size to fit in the SRAM. > > When we are adding more and more, and eventually hit this problem again, > > LTO already enabled, what we will do ? > > That's why I call this a workaround because we are not solving the > > problem, just postponing so we can add more things today. > > This is what's happening since many years. But okay, let's call it a > workaround. > > > > > > > > How does this impact the size? How much we are gaining ? > > > > > > I did not measure this. I just checked that this target compiles clean > > > again with LTO enabled and the MMC related patches applied. > > > > > > Could you (or some college?) please investigate here, how the results > > > are in image size? > > > > No, you are submitting the patch, I assume you could give some numbers > > to support your patch. > > Sorry, my time is limited and frankly, I don't feel very much motivated > (any more) to do additional work here. Even if it's not that much of > effort. > > > > > > > > We can perhaps have a look to see which code is removed and > > > > guard it by #ifndef SPL_BUILD and that might lower the size. (if > > > > ofcourse, this code should really be removed) > > > > > > Sure, other improvements in image size decrease are of course always a > > > good idea. > > > > > > > Also, I don't have a board at hand to test this, so it has to be > > > > tested first to make sure the board doesn't break. > > > > > > Agreed. I assume/hope that one of your colleges will be able to test > > > this? > > > > Someone from Microchip can, or other people using the board from the > > community > > > > I no longer work for Microchip, but I am still maintaining the AT91 > > custodian tree > > Okay. Let's see, where this goes. Well, if nobody wants to care about this board, go ahead with this change and if it is not enough that drop support for this board. > Thanks, > Stefan > > > Eugen > > > > > > Thanks, > > > Stefan > > > > > > > Eugen > > > > > > > > > > > > On 4/27/23 11:59, Stefan Roese wrote: > > > > > Adding just a tiny bit more code for sama5d2_icp_mmc leads to a SRAM > > > > > image overflow. Fix this by enabling LTO for this board, so that such > > > > > changes still can be made to the common U-Boot code. > > > > > > > > > > Signed-off-by: Stefan Roese > > > > > Cc: Tudor Ambarus > > > > > Cc: Eugen Hristev > > > > > Cc: Sergiu Moga > > > > > Cc: Pali Rohár > > > > > --- > > > > > configs/sama5d2_icp_mmc_defconfig | 5 +++-- > > > > > 1 file changed, 3 insertions(+), 2 deletions(-) > > > > > > > > > > diff --git a/configs/sama5d2_icp_mmc_defconfig > > > > > b/configs/sama5d2_icp_mmc_defconfig > > > > > index e1b602d8e5ec..a3c57a3f1250 100644 > > > > > --- a/configs/sama5d2_icp_mmc_defconfig > > > > > +++ b/configs/sama5d2_icp_mmc_defconfig > > > > > @@ -9,9 +9,11 @@ CONFIG_SPL_LIBCOMMON_SUPPORT=y > > > > > CONFIG_SPL_LIBGENERIC_SUPPORT=y > > > > > CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y > > > > > CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0x20003ef0 > > > > > +CONFIG_SF_DEFAULT_SPEED=6600 > > > > > CONFIG_ENV_SIZE=0x4000 > > > > > CONFIG_DM_GPIO=y > > > > > CONFIG_DEFAULT_DEVICE_TREE="at91-sama5d2_icp" > > > > > +CONFIG_OF_LIBFDT_OVERLAY=y > > > > > CONFIG_SPL_MMC=y > > > > > CONFIG_SPL_SERIAL=y > > > > > CONFIG_SPL_DRIVERS_MISC=y > > > > > @@ -24,6 +26,7 @@ CONFIG_SPL_FS_FAT=y > > > > > CONFIG_SPL_LIBDISK_SUPPORT=y > > > > > CONFIG_SYS_LOAD_ADDR=0x2200 > > > > > CONFIG_DEBUG_UART=y > > > > > +CONFIG_LTO=y > > > > > CONFIG_ENV_VARS_UBOOT_CONFIG=y > > > > > CONFIG_SYS_MONITOR_LEN=524288 > > > > > CONFIG_FIT=y > > > > > @@ -86,7 +89,6 @@ CONFIG_MMC_SDHCI=y > > > > > CONFIG_MMC_SDHCI_ATMEL=y > > > > > CONFIG_DM_SPI_FLASH=y > > > > > CONFIG_SF_DEFAULT_BUS=2 > > > > > -CONFIG_SF_DEFAULT_SPEED=6600 > > > > > CONFIG_SPI_FLASH_SFDP_SUPPORT=y > > > > > CONFIG_SPI_FLASH_ATMEL=y > > > > > CONFIG_SPI_FLASH_MACRONIX=y > > > > > @@ -110,5 +112,4 @@ CONFIG_TIMER=y > > > > > CONFIG_SPL_TIMER=y > > > > > CONFIG_ATMEL_TCB_TIMER=y > > > > > CONFIG_SPL_ATMEL_TCB_TIMER=y > > > > > -CONFIG_OF_LIBFDT_OVERLAY=y
Re: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO
On Thu, Apr 27, 2023 at 11:26:46AM +0200, Stefan Roese wrote: > Hi Eugen, > > On 4/27/23 11:19, Eugen Hristev wrote: > > Hi Stefan, > > > > Thank you for the patch. > > > > This I guess is a workaround such that you can add a bit more of code. > > In the end, it's not scalable, and we have to find a better way, > > probably by removing some of the code to make the SPL smaller. > > U-Boot image size increase resulting in overflowing some limits is > a common problem, especially in SPL. Enabling LTO gives quite some good > improvements in image size decrease. So I don't think it's an > workaround. > > > How does this impact the size? How much we are gaining ? > > I did not measure this. I just checked that this target compiles clean > again with LTO enabled and the MMC related patches applied. Note that buildman already has flags to tell you how much the size changed and in what functions between commits. -- Tom signature.asc Description: PGP signature
Re: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO
On 4/27/23 11:51, Eugen Hristev wrote: On 4/27/23 12:26, Stefan Roese wrote: Hi Eugen, On 4/27/23 11:19, Eugen Hristev wrote: Hi Stefan, Thank you for the patch. This I guess is a workaround such that you can add a bit more of code. In the end, it's not scalable, and we have to find a better way, probably by removing some of the code to make the SPL smaller. U-Boot image size increase resulting in overflowing some limits is a common problem, especially in SPL. Enabling LTO gives quite some good improvements in image size decrease. So I don't think it's an workaround. If this was not needed until today, and not adding any new functionality, I would call this a workaround to avoid shrinking the size to fit in the SRAM. When we are adding more and more, and eventually hit this problem again, LTO already enabled, what we will do ? That's why I call this a workaround because we are not solving the problem, just postponing so we can add more things today. This is what's happening since many years. But okay, let's call it a workaround. How does this impact the size? How much we are gaining ? I did not measure this. I just checked that this target compiles clean again with LTO enabled and the MMC related patches applied. Could you (or some college?) please investigate here, how the results are in image size? No, you are submitting the patch, I assume you could give some numbers to support your patch. Sorry, my time is limited and frankly, I don't feel very much motivated (any more) to do additional work here. Even if it's not that much of effort. We can perhaps have a look to see which code is removed and guard it by #ifndef SPL_BUILD and that might lower the size. (if ofcourse, this code should really be removed) Sure, other improvements in image size decrease are of course always a good idea. Also, I don't have a board at hand to test this, so it has to be tested first to make sure the board doesn't break. Agreed. I assume/hope that one of your colleges will be able to test this? Someone from Microchip can, or other people using the board from the community I no longer work for Microchip, but I am still maintaining the AT91 custodian tree Okay. Let's see, where this goes. Thanks, Stefan Eugen Thanks, Stefan Eugen On 4/27/23 11:59, Stefan Roese wrote: Adding just a tiny bit more code for sama5d2_icp_mmc leads to a SRAM image overflow. Fix this by enabling LTO for this board, so that such changes still can be made to the common U-Boot code. Signed-off-by: Stefan Roese Cc: Tudor Ambarus Cc: Eugen Hristev Cc: Sergiu Moga Cc: Pali Rohár --- configs/sama5d2_icp_mmc_defconfig | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/configs/sama5d2_icp_mmc_defconfig b/configs/sama5d2_icp_mmc_defconfig index e1b602d8e5ec..a3c57a3f1250 100644 --- a/configs/sama5d2_icp_mmc_defconfig +++ b/configs/sama5d2_icp_mmc_defconfig @@ -9,9 +9,11 @@ CONFIG_SPL_LIBCOMMON_SUPPORT=y CONFIG_SPL_LIBGENERIC_SUPPORT=y CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0x20003ef0 +CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_ENV_SIZE=0x4000 CONFIG_DM_GPIO=y CONFIG_DEFAULT_DEVICE_TREE="at91-sama5d2_icp" +CONFIG_OF_LIBFDT_OVERLAY=y CONFIG_SPL_MMC=y CONFIG_SPL_SERIAL=y CONFIG_SPL_DRIVERS_MISC=y @@ -24,6 +26,7 @@ CONFIG_SPL_FS_FAT=y CONFIG_SPL_LIBDISK_SUPPORT=y CONFIG_SYS_LOAD_ADDR=0x2200 CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_ENV_VARS_UBOOT_CONFIG=y CONFIG_SYS_MONITOR_LEN=524288 CONFIG_FIT=y @@ -86,7 +89,6 @@ CONFIG_MMC_SDHCI=y CONFIG_MMC_SDHCI_ATMEL=y CONFIG_DM_SPI_FLASH=y CONFIG_SF_DEFAULT_BUS=2 -CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_SPI_FLASH_SFDP_SUPPORT=y CONFIG_SPI_FLASH_ATMEL=y CONFIG_SPI_FLASH_MACRONIX=y @@ -110,5 +112,4 @@ CONFIG_TIMER=y CONFIG_SPL_TIMER=y CONFIG_ATMEL_TCB_TIMER=y CONFIG_SPL_ATMEL_TCB_TIMER=y -CONFIG_OF_LIBFDT_OVERLAY=y # CONFIG_EFI_LOADER_HII is not set Viele Grüße, Stefan Roese Viele Grüße, Stefan Roese -- DENX Software Engineering GmbH, Managing Director: Erika Unter HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-51 Fax: (+49)-8142-66989-80 Email: s...@denx.de
Re: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO
On 4/27/23 12:26, Stefan Roese wrote: Hi Eugen, On 4/27/23 11:19, Eugen Hristev wrote: Hi Stefan, Thank you for the patch. This I guess is a workaround such that you can add a bit more of code. In the end, it's not scalable, and we have to find a better way, probably by removing some of the code to make the SPL smaller. U-Boot image size increase resulting in overflowing some limits is a common problem, especially in SPL. Enabling LTO gives quite some good improvements in image size decrease. So I don't think it's an workaround. If this was not needed until today, and not adding any new functionality, I would call this a workaround to avoid shrinking the size to fit in the SRAM. When we are adding more and more, and eventually hit this problem again, LTO already enabled, what we will do ? That's why I call this a workaround because we are not solving the problem, just postponing so we can add more things today. How does this impact the size? How much we are gaining ? I did not measure this. I just checked that this target compiles clean again with LTO enabled and the MMC related patches applied. Could you (or some college?) please investigate here, how the results are in image size? No, you are submitting the patch, I assume you could give some numbers to support your patch. We can perhaps have a look to see which code is removed and guard it by #ifndef SPL_BUILD and that might lower the size. (if ofcourse, this code should really be removed) Sure, other improvements in image size decrease are of course always a good idea. Also, I don't have a board at hand to test this, so it has to be tested first to make sure the board doesn't break. Agreed. I assume/hope that one of your colleges will be able to test this? Someone from Microchip can, or other people using the board from the community I no longer work for Microchip, but I am still maintaining the AT91 custodian tree Eugen Thanks, Stefan Eugen On 4/27/23 11:59, Stefan Roese wrote: Adding just a tiny bit more code for sama5d2_icp_mmc leads to a SRAM image overflow. Fix this by enabling LTO for this board, so that such changes still can be made to the common U-Boot code. Signed-off-by: Stefan Roese Cc: Tudor Ambarus Cc: Eugen Hristev Cc: Sergiu Moga Cc: Pali Rohár --- configs/sama5d2_icp_mmc_defconfig | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/configs/sama5d2_icp_mmc_defconfig b/configs/sama5d2_icp_mmc_defconfig index e1b602d8e5ec..a3c57a3f1250 100644 --- a/configs/sama5d2_icp_mmc_defconfig +++ b/configs/sama5d2_icp_mmc_defconfig @@ -9,9 +9,11 @@ CONFIG_SPL_LIBCOMMON_SUPPORT=y CONFIG_SPL_LIBGENERIC_SUPPORT=y CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0x20003ef0 +CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_ENV_SIZE=0x4000 CONFIG_DM_GPIO=y CONFIG_DEFAULT_DEVICE_TREE="at91-sama5d2_icp" +CONFIG_OF_LIBFDT_OVERLAY=y CONFIG_SPL_MMC=y CONFIG_SPL_SERIAL=y CONFIG_SPL_DRIVERS_MISC=y @@ -24,6 +26,7 @@ CONFIG_SPL_FS_FAT=y CONFIG_SPL_LIBDISK_SUPPORT=y CONFIG_SYS_LOAD_ADDR=0x2200 CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_ENV_VARS_UBOOT_CONFIG=y CONFIG_SYS_MONITOR_LEN=524288 CONFIG_FIT=y @@ -86,7 +89,6 @@ CONFIG_MMC_SDHCI=y CONFIG_MMC_SDHCI_ATMEL=y CONFIG_DM_SPI_FLASH=y CONFIG_SF_DEFAULT_BUS=2 -CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_SPI_FLASH_SFDP_SUPPORT=y CONFIG_SPI_FLASH_ATMEL=y CONFIG_SPI_FLASH_MACRONIX=y @@ -110,5 +112,4 @@ CONFIG_TIMER=y CONFIG_SPL_TIMER=y CONFIG_ATMEL_TCB_TIMER=y CONFIG_SPL_ATMEL_TCB_TIMER=y -CONFIG_OF_LIBFDT_OVERLAY=y # CONFIG_EFI_LOADER_HII is not set Viele Grüße, Stefan Roese
Re: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO
Hi Eugen, On 4/27/23 11:19, Eugen Hristev wrote: Hi Stefan, Thank you for the patch. This I guess is a workaround such that you can add a bit more of code. In the end, it's not scalable, and we have to find a better way, probably by removing some of the code to make the SPL smaller. U-Boot image size increase resulting in overflowing some limits is a common problem, especially in SPL. Enabling LTO gives quite some good improvements in image size decrease. So I don't think it's an workaround. How does this impact the size? How much we are gaining ? I did not measure this. I just checked that this target compiles clean again with LTO enabled and the MMC related patches applied. Could you (or some college?) please investigate here, how the results are in image size? We can perhaps have a look to see which code is removed and guard it by #ifndef SPL_BUILD and that might lower the size. (if ofcourse, this code should really be removed) Sure, other improvements in image size decrease are of course always a good idea. Also, I don't have a board at hand to test this, so it has to be tested first to make sure the board doesn't break. Agreed. I assume/hope that one of your colleges will be able to test this? Thanks, Stefan Eugen On 4/27/23 11:59, Stefan Roese wrote: Adding just a tiny bit more code for sama5d2_icp_mmc leads to a SRAM image overflow. Fix this by enabling LTO for this board, so that such changes still can be made to the common U-Boot code. Signed-off-by: Stefan Roese Cc: Tudor Ambarus Cc: Eugen Hristev Cc: Sergiu Moga Cc: Pali Rohár --- configs/sama5d2_icp_mmc_defconfig | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/configs/sama5d2_icp_mmc_defconfig b/configs/sama5d2_icp_mmc_defconfig index e1b602d8e5ec..a3c57a3f1250 100644 --- a/configs/sama5d2_icp_mmc_defconfig +++ b/configs/sama5d2_icp_mmc_defconfig @@ -9,9 +9,11 @@ CONFIG_SPL_LIBCOMMON_SUPPORT=y CONFIG_SPL_LIBGENERIC_SUPPORT=y CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0x20003ef0 +CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_ENV_SIZE=0x4000 CONFIG_DM_GPIO=y CONFIG_DEFAULT_DEVICE_TREE="at91-sama5d2_icp" +CONFIG_OF_LIBFDT_OVERLAY=y CONFIG_SPL_MMC=y CONFIG_SPL_SERIAL=y CONFIG_SPL_DRIVERS_MISC=y @@ -24,6 +26,7 @@ CONFIG_SPL_FS_FAT=y CONFIG_SPL_LIBDISK_SUPPORT=y CONFIG_SYS_LOAD_ADDR=0x2200 CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_ENV_VARS_UBOOT_CONFIG=y CONFIG_SYS_MONITOR_LEN=524288 CONFIG_FIT=y @@ -86,7 +89,6 @@ CONFIG_MMC_SDHCI=y CONFIG_MMC_SDHCI_ATMEL=y CONFIG_DM_SPI_FLASH=y CONFIG_SF_DEFAULT_BUS=2 -CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_SPI_FLASH_SFDP_SUPPORT=y CONFIG_SPI_FLASH_ATMEL=y CONFIG_SPI_FLASH_MACRONIX=y @@ -110,5 +112,4 @@ CONFIG_TIMER=y CONFIG_SPL_TIMER=y CONFIG_ATMEL_TCB_TIMER=y CONFIG_SPL_ATMEL_TCB_TIMER=y -CONFIG_OF_LIBFDT_OVERLAY=y # CONFIG_EFI_LOADER_HII is not set Viele Grüße, Stefan Roese -- DENX Software Engineering GmbH, Managing Director: Erika Unter HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-51 Fax: (+49)-8142-66989-80 Email: s...@denx.de
Re: [PATCH] configs: at91: sama5d2_icp_mmc: Enable CONFIG_LTO
Hi Stefan, Thank you for the patch. This I guess is a workaround such that you can add a bit more of code. In the end, it's not scalable, and we have to find a better way, probably by removing some of the code to make the SPL smaller. How does this impact the size? How much we are gaining ? We can perhaps have a look to see which code is removed and guard it by #ifndef SPL_BUILD and that might lower the size. (if ofcourse, this code should really be removed) Also, I don't have a board at hand to test this, so it has to be tested first to make sure the board doesn't break. Eugen On 4/27/23 11:59, Stefan Roese wrote: Adding just a tiny bit more code for sama5d2_icp_mmc leads to a SRAM image overflow. Fix this by enabling LTO for this board, so that such changes still can be made to the common U-Boot code. Signed-off-by: Stefan Roese Cc: Tudor Ambarus Cc: Eugen Hristev Cc: Sergiu Moga Cc: Pali Rohár --- configs/sama5d2_icp_mmc_defconfig | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/configs/sama5d2_icp_mmc_defconfig b/configs/sama5d2_icp_mmc_defconfig index e1b602d8e5ec..a3c57a3f1250 100644 --- a/configs/sama5d2_icp_mmc_defconfig +++ b/configs/sama5d2_icp_mmc_defconfig @@ -9,9 +9,11 @@ CONFIG_SPL_LIBCOMMON_SUPPORT=y CONFIG_SPL_LIBGENERIC_SUPPORT=y CONFIG_HAS_CUSTOM_SYS_INIT_SP_ADDR=y CONFIG_CUSTOM_SYS_INIT_SP_ADDR=0x20003ef0 +CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_ENV_SIZE=0x4000 CONFIG_DM_GPIO=y CONFIG_DEFAULT_DEVICE_TREE="at91-sama5d2_icp" +CONFIG_OF_LIBFDT_OVERLAY=y CONFIG_SPL_MMC=y CONFIG_SPL_SERIAL=y CONFIG_SPL_DRIVERS_MISC=y @@ -24,6 +26,7 @@ CONFIG_SPL_FS_FAT=y CONFIG_SPL_LIBDISK_SUPPORT=y CONFIG_SYS_LOAD_ADDR=0x2200 CONFIG_DEBUG_UART=y +CONFIG_LTO=y CONFIG_ENV_VARS_UBOOT_CONFIG=y CONFIG_SYS_MONITOR_LEN=524288 CONFIG_FIT=y @@ -86,7 +89,6 @@ CONFIG_MMC_SDHCI=y CONFIG_MMC_SDHCI_ATMEL=y CONFIG_DM_SPI_FLASH=y CONFIG_SF_DEFAULT_BUS=2 -CONFIG_SF_DEFAULT_SPEED=6600 CONFIG_SPI_FLASH_SFDP_SUPPORT=y CONFIG_SPI_FLASH_ATMEL=y CONFIG_SPI_FLASH_MACRONIX=y @@ -110,5 +112,4 @@ CONFIG_TIMER=y CONFIG_SPL_TIMER=y CONFIG_ATMEL_TCB_TIMER=y CONFIG_SPL_ATMEL_TCB_TIMER=y -CONFIG_OF_LIBFDT_OVERLAY=y # CONFIG_EFI_LOADER_HII is not set
