Re: [edk2-devel] [PATCH v2 2/6] OvmfPkg: add 'initrd' shell command to expose Linux initrd via device path

["Laszlo Ersek" <lersek@redhat.com>]

On 02/26/20 00:43, Laszlo Ersek wrote:
> On 02/25/20 10:39, Ard Biesheuvel wrote:

>> +"  2. The initrd is not unloaded when the shell exits, and will remain active\r\n"
>> +"     until it is unloaded again by a different invocation of the shell.\r\n"
>> +"     Consumers of the LoadFile2 protocol on the LINUX_EFI_INITRD_MEDIA_GUID\r\n"
>> +"     device path that are started via means other than the shell will be able\r\n"
>> +"     to locate the protocol and invoke it.\r\n"
>>
> 
> (16) So, I don't see how paragraph#2 works here. When the shell exits,
> all our global variables disappear. (LoadFile2 disappears too.) The next
> time the shell is launched (in the same "UEFI session", so to say), we
> won't know that we had loaded an initrd before. Is that right?

Sorry, I guess I missed the core idea behind dynamic shell commands :(

The dynamic shell command seems like a separate DXE driver, after all.
The shell can call into it (via EFI_SHELL_DYNAMIC_COMMAND_PROTOCOL), but
its lifecycle is independent of the shell's. The shell can also unload
it (like the shell can unload any other driver).

So please ignore my point (16).

(But then, I do think we leak the initrd, if one has been loaded, in
LinuxInitrdDynamicShellCommandUnload()). I.e., pls see my point (15).)

Thanks
Laszlo