Re: Loading initrd above 4G causes freeze on boot

From: Michael Brown
Date: Wed Aug 20 2014 - 15:53:53 EST


On 20/08/14 20:05, Mantas MikulÄnas wrote:
On Wed, Aug 20, 2014 at 8:05 PM, Matt Fleming <matt@xxxxxxxxxxxxxxxxx> wrote:
On Wed, 13 Aug, at 07:44:49PM, Matt Fleming wrote:

At this point, I think modifying the max address is the best way to
debug this further, and figure out what address causes the hang.

Mantas, did you manage to get to the bottom of this issue?

I experimented with some things (like setting chunk size to a few kB
to see if it hangs earlier or only at the very end; etc.), and finally
found out that it stops freezing if I pad the initrd file to a
multiple of 512 bytes :/ That is, 5684268 bytes will freeze, 5684736
bytes will not.

...In other words, seems like it cannot read chunks that aren't
multiples of 512 into a location above 4 GB. Or something like that...

I haven't been following this thread closely, but that immediately sounds like a problem within the EFI_DISK_IO_PROTOCOL implementation (which is responsible for handling smaller-than-block-sized reads). Looking at the EDK2 implementation in MdeModulePkg/Universal/Disk/DiskIoDxe/DiskIo.c, the memory management does appear to be somewhat inventive. In particular, there's a frequent pattern in DiskIoCreateSubtaskList() equivalent to:

if ( blocking_io ) {
buffer = some_static_buffer;
} else {
buffer = malloc ( len );
if ( ! buffer )
goto single_shared_error_label;
}
... do not record whether or not buffer was dynamically allocated ...
... use buffer as part of an asynchronous I/O operation ...
... eventually choose whether or not to free buffer, and hope the choice is correct ...

It's not at all obvious that memory is freed correctly, especially under some of the error paths within that code.

I can't immediately see anything that should fail with a pointer above 4G, but I wouldn't be surprised to find a path that causes a double free or similar error.

Apologies if I've missed something critical earlier in the thread, making my ramblings are totally irrelevant.

Michael
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/