[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] REGRESSION: Xen 4.13 RC5 fails to bootstrap Dom0 on ARM



Hi,

On 20/12/2019 00:01, Stefano Stabellini wrote:
On Thu, 19 Dec 2019, Julien Grall wrote:
In fact most of people on Arm are using GRUB rather than EFI directly as
this is more friendly to use.

Regarding the devicetree, Xen and Linux will completely ignore the
memory nodes in Xen if using EFI. This because the EFI memory map will
give you an overview of the platform with the EFI regions included.

Aha! So in that sense it is a bug in Xen after all, right? (that's what
you're
referring to when you say you now understand what needs to get fixed).

Yes. The EFI memory map is a list of existing memory with a type associated to
it (Conventional, BootServiceCodes, MemoryMappedIO...).

The OS/Hypervisor will have to go through them and check which regions are
usuable. Compare to Linux, Xen has limited itself to only a few types.

However, I think we can be on a par with Linux here.

I gave a look at the Linux implementation, the interesting bit is
drivers/firmware/efi/arm-init.c:is_usable_memory as far as I can tell.
I also gave a look at the Xen side, which is
xen/arch/arm/efi/efi-boot.h:efi_process_memory_map_bootinfo. As guessed,
the two are not quite the same.

One of the main differences is that Linux uses as "System RAM" even
regions that were marked as EFI_BOOT_SERVICES_CODE/DATA and
EFI_LOADER_CODE/DATA because they will get freed anyway. Xen doesn't
do that unless map_bs is set.

Well, map_bs is used to request to map the boot services into Xen PT. In other words, you can't consider them as usuable if that option is set.


I wrote a quick patch to implement the Linux behavior on Xen, only
lightly tested. I can confirm that I see more memory this way. However,
I am not sure we actually want to import the Linux behavior wholesale.

This is not what I had in mind, we still need to keep Xen behavior for boot services and ensure the region are recorded/skipped as expected (see more below).


Anyway, Roman, could you please let me know if this patch solves the
issue?


Please see a couple of comments below.



diff --git a/xen/arch/arm/efi/efi-boot.h b/xen/arch/arm/efi/efi-boot.h
index ca655ff003..ad18ff3669 100644
--- a/xen/arch/arm/efi/efi-boot.h
+++ b/xen/arch/arm/efi/efi-boot.h
@@ -149,10 +149,14 @@ static EFI_STATUS __init 
efi_process_memory_map_bootinfo(EFI_MEMORY_DESCRIPTOR *
for ( Index = 0; Index < (mmap_size / desc_size); Index++ )
      {
-        if ( desc_ptr->Type == EfiConventionalMemory ||
-             (!map_bs &&
-              (desc_ptr->Type == EfiBootServicesCode ||
-               desc_ptr->Type == EfiBootServicesData)) )
+        if ( desc_ptr->Attribute == EFI_MEMORY_WB &&

This is not quite the same as Linux. They use:

desc_ptr->Attribute & EFI_MEMORY_WB

But I don't see why Attribute should be strictly equal to EFI_MEMORY_WB.

+             (desc_ptr->Type == EfiConventionalMemory ||
+              desc_ptr->Type == EfiLoaderCode ||
+              desc_ptr->Type == EfiLoaderData ||
+              desc_ptr->Type == EfiACPIReclaimMemory ||

Linux will explictly reserve the ACPI region (see the caller of is_usable_memory()). For us this is done in the "else if" part.

+              desc_ptr->Type == EfiPersistentMemory ||
+              desc_ptr->Type == EfiBootServicesCode ||
+              desc_ptr->Type == EfiBootServicesData) )
          {
              if ( !meminfo_add_bank(&bootinfo.mem, desc_ptr) )
              {
diff --git a/xen/include/efi/efidef.h b/xen/include/efi/efidef.h
index 86a7e111bf..f46207840f 100644
--- a/xen/include/efi/efidef.h
+++ b/xen/include/efi/efidef.h
@@ -147,6 +147,7 @@ typedef enum {
      EfiMemoryMappedIO,
      EfiMemoryMappedIOPortSpace,
      EfiPalCode,
+    EfiPersistentMemory,
      EfiMaxMemoryType
  } EFI_MEMORY_TYPE;

Cheers,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.