Skip to contentSkip to conversation

efi_loader: efi_runtime_detach()?

U-Boot AKASHI Takahiro 3 participants 4 messages May 22, 2019 – May 29, 2019

lore
Share & tools
lore
#0 lore raw

Alex,

It seems to me that the current efi_runtime_detach() has two meanings:
* changes relating to the transition to virtual address mode
* changes relating to exiting boot services

In the current implementation, efi_runtime_detach() is
called from efi_set_virtual_address_map() as those two purposes are apparently the same.
For instance, however, systab.con_in/out/std_err/boottime are set to NULL in exit_boot_services() as well.
Logically, efi_get_variable() won't work as U-Boot environment data can be potentially re-used at any time after exit_boot_services(). The only meaningful change is to invalidate efi_set_virtual_address_map().

So I believe that we should distinguish the two purposes and make appropriate changes in efi_set_virtual_address_map() and and exit_boot_services() respectively.

This will be crucial particularly when I will add
runtime variable support using "cache".

Any thoughts?

Thanks,
-Takahiro Akashi

#1 lore raw
Quoted 16 lines · AKASHI Takahiro

Alex,

It seems to me that the current efi_runtime_detach() has two meanings:
* changes relating to the transition to virtual address mode
* changes relating to exiting boot services

In the current implementation, efi_runtime_detach() is
called from efi_set_virtual_address_map() as those two purposes are apparently the same.
For instance, however, systab.con_in/out/std_err/boottime are set to NULL in exit_boot_services() as well.
Logically, efi_get_variable() won't work as U-Boot environment data can be potentially re-used at any time after exit_boot_services(). The only meaningful change is to invalidate efi_set_virtual_address_map().

Quoted 8 lines

So I believe that we should distinguish the two purposes and make appropriate changes in efi_set_virtual_address_map() and and exit_boot_services() respectively.

This will be crucial particularly when I will add
runtime variable support using "cache".

Any thoughts?

Please, consider this pending patch from Alex:
https://patchwork.ozlabs.org/patch/1054074/
efi_loader: Patch non-runtime code out at ExitBootServices already

Unfortunately Travis CI is still failing with the patch and QEMU arm64 does not reboot.

Best regards

Heinrich

Quoted 4 lines

Thanks,
-Takahiro Akashi

#2 lore raw
Quoted 21 lines · Heinrich Schuchardt
Quoted 20 lines · AKASHI Takahiro

Alex,

It seems to me that the current efi_runtime_detach() has two meanings:
* changes relating to the transition to virtual address mode
* changes relating to exiting boot services

In the current implementation, efi_runtime_detach() is
called from efi_set_virtual_address_map() as those two purposes are apparently the same.
For instance, however, systab.con_in/out/std_err/boottime are set to NULL in exit_boot_services() as well.
Logically, efi_get_variable() won't work as U-Boot environment data can be potentially re-used at any time after exit_boot_services(). The only meaningful change is to invalidate efi_set_virtual_address_map().

So I believe that we should distinguish the two purposes and make appropriate changes in efi_set_virtual_address_map() and and exit_boot_services() respectively.

Quoted 8 lines
Quoted 4 lines

This will be crucial particularly when I will add
runtime variable support using "cache".

Any thoughts?

Please, consider this pending patch from Alex:
https://patchwork.ozlabs.org/patch/1054074/
efi_loader: Patch non-runtime code out at ExitBootServices already

@Alex, what is the current status of this patch?

Quoted 2 lines

Unfortunately Travis CI is still failing with the patch and QEMU arm64 does not reboot.

What error did you see?
After what, did qemu arm64 fail to reboot?

-Takahiro Akashi

Quoted 9 lines

Best regards

Heinrich

Quoted 4 lines

Thanks,
-Takahiro Akashi

#3 lore raw
Quoted 22 lines · AKASHI Takahiro
Quoted 21 lines · Heinrich Schuchardt
Quoted 20 lines · AKASHI Takahiro

Alex,

It seems to me that the current efi_runtime_detach() has two meanings:
* changes relating to the transition to virtual address mode
* changes relating to exiting boot services

In the current implementation, efi_runtime_detach() is
called from efi_set_virtual_address_map() as those two purposes are apparently the same.
For instance, however, systab.con_in/out/std_err/boottime are set to NULL in exit_boot_services() as well.
Logically, efi_get_variable() won't work as U-Boot environment data can be potentially re-used at any time after exit_boot_services(). The only meaningful change is to invalidate efi_set_virtual_address_map().

So I believe that we should distinguish the two purposes and make appropriate changes in efi_set_virtual_address_map() and and exit_boot_services() respectively.

Quoted 8 lines
Quoted 7 lines
Quoted 4 lines

This will be crucial particularly when I will add
runtime variable support using "cache".

Any thoughts?

Please, consider this pending patch from Alex:
https://patchwork.ozlabs.org/patch/1054074/
efi_loader: Patch non-runtime code out at ExitBootServices already

@Alex, what is the current status of this patch?

I don't think I touched it ever since :). It's the right thing to do though. Care to debug why it breaks?

Alex