[PATCH 0/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[PATCH 0/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[Heinrich Schuchardt]

Our MMC driver for JH7110 boards only supports reading to the low 4 GiB of
memory. Booting boards with more memory requires EFI_LOADER_BOUNCE_BUFFER.

As a prerequisite we must allow CONFIG_EFI_LOADER_BOUNCE_BUFFER on RISC-V.

Heinrich Schuchardt (2):
  efi_loader: allow EFI_LOADER_BOUNCE_BUFFER on all architectures
  configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

 configs/starfive_visionfive2_defconfig | 1 +
 lib/efi_loader/Kconfig                 | 1 -
 2 files changed, 1 insertion(+), 1 deletion(-)

-- 
2.45.2

[PATCH 1/2] efi_loader: allow EFI_LOADER_BOUNCE_BUFFER on all architectures

[Heinrich Schuchardt]

Commit 775f7657ba58 ("Kconfig: clean up the efi configuration status")
by mistake revoked commit dcd1b63b7072 ("efi_loader: allow
EFI_LOADER_BOUNCE_BUFFER on all architectures").

Fixes: 775f7657ba58 ("Kconfig: clean up the efi configuration status")
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
---
 lib/efi_loader/Kconfig | 1 -
 1 file changed, 1 deletion(-)

diff --git a/lib/efi_loader/Kconfig b/lib/efi_loader/Kconfig
index 58d49789f12..d93f28b8422 100644
--- a/lib/efi_loader/Kconfig
+++ b/lib/efi_loader/Kconfig
@@ -481,7 +481,6 @@ endmenu
 menu "Misc options"
 config EFI_LOADER_BOUNCE_BUFFER
 	bool "EFI Applications use bounce buffers for DMA operations"
-	depends on ARM64
 	help
 	  Some hardware does not support DMA to full 64bit addresses. For this
 	  hardware we can create a bounce buffer so that payloads don't have to
-- 
2.45.2

[PATCH 2/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[Heinrich Schuchardt]

Our MMC driver for JH7110 boards only supports reading to the low 4 GiB of
memory. Booting boards with more memory requires EFI_LOADER_BOUNCE_BUFFER.

Reported-by: E Shattow <lucent@gmail.com>
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
---
 configs/starfive_visionfive2_defconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/configs/starfive_visionfive2_defconfig b/configs/starfive_visionfive2_defconfig
index 1c70d1d4b70..20f89ae6796 100644
--- a/configs/starfive_visionfive2_defconfig
+++ b/configs/starfive_visionfive2_defconfig
@@ -32,6 +32,7 @@ CONFIG_CMODEL_MEDANY=y
 CONFIG_RISCV_SMODE=y
 # CONFIG_OF_BOARD_FIXUP is not set
 # CONFIG_SYS_MALLOC_CLEAR_ON_INIT is not set
+CONFIG_EFI_LOADER_BOUNCE_BUFFER=y
 CONFIG_FIT=y
 CONFIG_BOOTSTD_DEFAULTS=y
 CONFIG_BOOTSTAGE=y
-- 
2.45.2

Re: [PATCH 0/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[Andreas Schwab]

On Nov 21 2024, Heinrich Schuchardt wrote:

> Our MMC driver for JH7110 boards only supports reading to the low 4 GiB of
> memory. Booting boards with more memory requires EFI_LOADER_BOUNCE_BUFFER.

That doesn't change the situation with the booting problem I have.

-- 
Andreas Schwab, SUSE Labs, schwab@suse.de
GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE  1748 E4D4 88E3 0EEA B9D7
"And now for something completely different."

Re: [PATCH 0/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[Heinrich Schuchardt]

On 21.11.24 16:24, Andreas Schwab wrote:
> On Nov 21 2024, Heinrich Schuchardt wrote:
> 
>> Our MMC driver for JH7110 boards only supports reading to the low 4 GiB of
>> memory. Booting boards with more memory requires EFI_LOADER_BOUNCE_BUFFER.
> 
> That doesn't change the situation with the booting problem I have.
> 

Do you relate to
https://lore.kernel.org/u-boot/mvm8qtnv310.fsf@suse.de/
where you had some problem reading the FAT file system and which was for 
as system with only 4 GiB of memory?

In E Shattow's case on a board with 8 GiB the file was read without 
error but then reported as not being an EFI binary.

Best regards

Heinrich

Re: [PATCH 0/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[E Shattow]

On Thu, Nov 21, 2024 at 8:05 AM Heinrich Schuchardt
<heinrich.schuchardt@canonical.com> wrote:
>
> On 21.11.24 16:24, Andreas Schwab wrote:
> > On Nov 21 2024, Heinrich Schuchardt wrote:
> >
> >> Our MMC driver for JH7110 boards only supports reading to the low 4 GiB of
> >> memory. Booting boards with more memory requires EFI_LOADER_BOUNCE_BUFFER.
> >
> > That doesn't change the situation with the booting problem I have.
> >
>
> Do you relate to
> https://lore.kernel.org/u-boot/mvm8qtnv310.fsf@suse.de/
> where you had some problem reading the FAT file system and which was for
> as system with only 4 GiB of memory?
>
> In E Shattow's case on a board with 8 GiB the file was read without
> error but then reported as not being an EFI binary.
>
> Best regards
>
> Heinrich
>
>

Hi,

Test failed for both boards here with origin/master then applying this
series, then starfive_visionfive2_defconfig target and set
CONFIG_BOOTSTD=n following with oldconfig target.

Testing failed on 4GB Star64:

Hit any key to stop autoboot:  0
Unknown command 'bootflow' - try 'help'
StarFive # bootefi bootmgr
Card did not respond to voltage select! : -110
Not a PE-COFF file
Loading Boot0000 'mmc 0' failed
Booting: nvme 0
error: no such partition.
grub rescue>

The grub EFI loader from EFI System Partition on eMMC fails to load.
The same EFI System Partition data on NVMe is successful.

Testing failed on 8GB Mars CM Lite:

Hit any key to stop autoboot:  0
Unknown command 'bootflow' - try 'help'
StarFive # bootefi bootmgr
Card did not respond to voltage select! : -110
Failed to load EFI variables
Error reading cluster
Loading Boot0000 'mmc 0' failed
EFI boot manager: Cannot load any image

We see that the grub EFI loader from this EFI System Partition fails
to load from MMC (SD Card).

Procedure that was used to update SPI NOR flash on either board
includes sanitizing the environment variable storages:
env erase; env default -a -f; sf probe; dhcp; env set serverip <ip
address of tftp server>
tftpboot $loadaddr u-boot-spl.bin.normal.out; sf update $loadaddr 0 $filesize
tftpboot $loadaddr u-boot.itb; sf update $loadaddr 100000 $filesize
fatrm mmc 0:1 ubootefi.var

This does not say anything for or against the validity of the patch
series as the root cause is a limitation of the dw mmc driver on
JH7110 not resolved here.

-E

Re: [PATCH 0/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[Sughosh Ganu]

On Fri, 22 Nov 2024 at 04:12, E Shattow <lucent@gmail.com> wrote:
>
> On Thu, Nov 21, 2024 at 8:05 AM Heinrich Schuchardt
> <heinrich.schuchardt@canonical.com> wrote:
> >
> > On 21.11.24 16:24, Andreas Schwab wrote:
> > > On Nov 21 2024, Heinrich Schuchardt wrote:
> > >
> > >> Our MMC driver for JH7110 boards only supports reading to the low 4 GiB of
> > >> memory. Booting boards with more memory requires EFI_LOADER_BOUNCE_BUFFER.
> > >
> > > That doesn't change the situation with the booting problem I have.
> > >
> >
> > Do you relate to
> > https://lore.kernel.org/u-boot/mvm8qtnv310.fsf@suse.de/
> > where you had some problem reading the FAT file system and which was for
> > as system with only 4 GiB of memory?
> >
> > In E Shattow's case on a board with 8 GiB the file was read without
> > error but then reported as not being an EFI binary.
> >
> > Best regards
> >
> > Heinrich
> >
> >
>
> Hi,
>
> Test failed for both boards here with origin/master then applying this
> series, then starfive_visionfive2_defconfig target and set
> CONFIG_BOOTSTD=n following with oldconfig target.

I think you need to apply another patch [1].

-sughosh

[1] - https://patchwork.ozlabs.org/project/uboot/patch/20241121142522.59059-2-heinrich.schuchardt@canonical.com/

>
> Testing failed on 4GB Star64:
>
> Hit any key to stop autoboot:  0
> Unknown command 'bootflow' - try 'help'
> StarFive # bootefi bootmgr
> Card did not respond to voltage select! : -110
> Not a PE-COFF file
> Loading Boot0000 'mmc 0' failed
> Booting: nvme 0
> error: no such partition.
> grub rescue>
>
> The grub EFI loader from EFI System Partition on eMMC fails to load.
> The same EFI System Partition data on NVMe is successful.
>
> Testing failed on 8GB Mars CM Lite:
>
> Hit any key to stop autoboot:  0
> Unknown command 'bootflow' - try 'help'
> StarFive # bootefi bootmgr
> Card did not respond to voltage select! : -110
> Failed to load EFI variables
> Error reading cluster
> Loading Boot0000 'mmc 0' failed
> EFI boot manager: Cannot load any image
>
> We see that the grub EFI loader from this EFI System Partition fails
> to load from MMC (SD Card).
>
> Procedure that was used to update SPI NOR flash on either board
> includes sanitizing the environment variable storages:
> env erase; env default -a -f; sf probe; dhcp; env set serverip <ip
> address of tftp server>
> tftpboot $loadaddr u-boot-spl.bin.normal.out; sf update $loadaddr 0 $filesize
> tftpboot $loadaddr u-boot.itb; sf update $loadaddr 100000 $filesize
> fatrm mmc 0:1 ubootefi.var
>
> This does not say anything for or against the validity of the patch
> series as the root cause is a limitation of the dw mmc driver on
> JH7110 not resolved here.
>
> -E

Re: [PATCH 0/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[E Shattow]

On Thu, Nov 21, 2024 at 10:32 PM Sughosh Ganu <sughosh.ganu@linaro.org> wrote:
...
> > Hi,
> >
> > Test failed for both boards here with origin/master then applying this
> > series, then starfive_visionfive2_defconfig target and set
> > CONFIG_BOOTSTD=n following with oldconfig target.
>
> I think you need to apply another patch [1].
>
> -sughosh
>
> [1] - https://patchwork.ozlabs.org/project/uboot/patch/20241121142522.59059-2-heinrich.schuchardt@canonical.com/
>

I did apply the whole series, patches 1 and 2 inclusive. Is there a
canary (printf? i.e.) I can add to know if this survives to the
compiled output and does what you think it should do?

-E

Re: [PATCH 0/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[Loic Devulder]

Hi guys!

I played with latest openSUSE TW image [1] for VisionFive2 and I also had an issue but not exactly
the same Andreas had. Booting the OS from an sdcard gave me the grub cli but not the menu. With
Heinrich's patch I was able to boot on grub menu as fully boot the OS as well.

Tested-by: Loic Devulder <ldevulder@suse.com>

-- 
Loic Devulder <ldevulder@suse.com>
Senior QA Engineer | SUSE Software Solutions

[1] -
https://download.opensuse.org/ports/riscv/tumbleweed/images/openSUSE-Tumbleweed-RISC-V-JeOS-starfivevisionfive2.riscv64-2024.11.10-Build1.19.raw.xz

On Thu, 2024-11-21 at 16:24 +0100, Andreas Schwab wrote:
> On Nov 21 2024, Heinrich Schuchardt wrote:
> 
> > Our MMC driver for JH7110 boards only supports reading to the low 4 GiB of
> > memory. Booting boards with more memory requires EFI_LOADER_BOUNCE_BUFFER.
> 
> That doesn't change the situation with the booting problem I have.
> 

Re: [PATCH 0/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[E Shattow]

On Fri, Nov 22, 2024 at 6:33 AM Loic Devulder <ldevulder@suse.com> wrote:
>
> Hi guys!
>
> I played with latest openSUSE TW image [1] for VisionFive2 and I also had an issue but not exactly
> the same Andreas had. Booting the OS from an sdcard gave me the grub cli but not the menu. With
> Heinrich's patch I was able to boot on grub menu as fully boot the OS as well.
>
> Tested-by: Loic Devulder <ldevulder@suse.com>
>
> --
> Loic Devulder <ldevulder@suse.com>
> Senior QA Engineer | SUSE Software Solutions
>
> [1] -
> https://download.opensuse.org/ports/riscv/tumbleweed/images/openSUSE-Tumbleweed-RISC-V-JeOS-starfivevisionfive2.riscv64-2024.11.10-Build1.19.raw.xz
>
> On Thu, 2024-11-21 at 16:24 +0100, Andreas Schwab wrote:
> > On Nov 21 2024, Heinrich Schuchardt wrote:
> >
> > > Our MMC driver for JH7110 boards only supports reading to the low 4 GiB of
> > > memory. Booting boards with more memory requires EFI_LOADER_BOUNCE_BUFFER.
> >
> > That doesn't change the situation with the booting problem I have.
> >

What was this series supposed to do and how to test for it properly?
I did not see any difference, still just failure on 4GB and 8GB DRAM
... is my testing method wrong?

-E

Re: [PATCH 0/2] configs: JH7110: enable EFI_LOADER_BOUNCE_BUFFER

[E Shattow]

On Mon, Nov 25, 2024 at 10:32 PM E Shattow <lucent@gmail.com> wrote:
>
> On Fri, Nov 22, 2024 at 6:33 AM Loic Devulder <ldevulder@suse.com> wrote:
> >
> > Hi guys!
> >
> > I played with latest openSUSE TW image [1] for VisionFive2 and I also had an issue but not exactly
> > the same Andreas had. Booting the OS from an sdcard gave me the grub cli but not the menu. With
> > Heinrich's patch I was able to boot on grub menu as fully boot the OS as well.
> >
> > Tested-by: Loic Devulder <ldevulder@suse.com>
> >
> > --
> > Loic Devulder <ldevulder@suse.com>
> > Senior QA Engineer | SUSE Software Solutions
> >
> > [1] -
> > https://download.opensuse.org/ports/riscv/tumbleweed/images/openSUSE-Tumbleweed-RISC-V-JeOS-starfivevisionfive2.riscv64-2024.11.10-Build1.19.raw.xz
> >
> > On Thu, 2024-11-21 at 16:24 +0100, Andreas Schwab wrote:
> > > On Nov 21 2024, Heinrich Schuchardt wrote:
> > >
> > > > Our MMC driver for JH7110 boards only supports reading to the low 4 GiB of
> > > > memory. Booting boards with more memory requires EFI_LOADER_BOUNCE_BUFFER.
> > >
> > > That doesn't change the situation with the booting problem I have.
> > >
>
> What was this series supposed to do and how to test for it properly?
> I did not see any difference, still just failure on 4GB and 8GB DRAM
> ... is my testing method wrong?
>
> -E

Postscript:  After some discussion with Sughosh we find that yes I am
testing this code path but "the logic of addr_aligned() needs to be
changed". This series is correct but the commit description is
confusing by omission of information: There is a 1GB offset in address
space to DRAM, and so any board with more than 3GB DRAM are affected.
If reading the commit message without knowing this information it
seems like 4GB DRAM board would not be affected but such address space
goes to 5GB. I was hoping for some reply from Heinrich to resolve why
testing results are not consistent and to offer a clearer commit
message before the pull request.

With that, and my thanks to Heinrich for speedy bug fixing,

Tested-by: E Shattow <lucent@gmail.com>