qemu-system-x86_64 generates ACPI tables with broken endianess when run on big-endian hosts

Bug #1724570 reported by Daniel Berrange
10
This bug affects 1 person
Affects Status Importance Assigned to Milestone
QEMU
Fix Released
Undecided
Unassigned

Bug Description

The bios-tables-test always fails when run on a big-endian host, which has iasl installed. When it calls iasl to dumps the AML files into ASL files, iasl complains

Intel ACPI Component Architecture
ASL+ Optimizing Compiler/Disassembler version 20170831
Copyright (c) 2000 - 2017 Intel Corporation

Input file aml-4L677Y, Length 0x38 (56) bytes
Table [TEPH] is too long for file - needs: 0x38000000, remaining in file: 0x38
Could not get ACPI tables from aml-4L677Y, AE_BAD_HEADER

At first I thought this was an iasl bug, but the latest version of iasl in rawhide is ported to big endian.

So I looked at the actual AML files that bios-tables-test extracts from the qemu-system-x86_64 memory space, when running on ppc64 host. These do indeed have different content from the AML files generated by qemu-system-x86_64 when running on an x86_64 host.

eg the AML file for the HPET shows

< 0000000 T E P H nul nul nul 8 soh etx B O C H S sp
< 4554 4850 0000 3800 0301 4f42 4843 2053
< 0000020 B X P C H P E T nul nul nul soh B X P C
< 5842 4350 5048 5445 0000 0100 5842 4350
< 0000040 nul nul nul soh soh " ack nul nul nul nul nul nul nul P ~
< 0000 0100 a201 8086 0000 0000 0000 fed0
---
> 0000000 H P E T 8 nul nul nul soh etx B O C H S sp
> 5048 5445 0038 0000 0301 4f42 4843 2053
> 0000020 B X P C H P E T soh nul nul nul B X P C
> 5842 4350 5048 5445 0001 0000 5842 4350
> 0000040 soh nul nul nul soh " ack nul nul nul nul nul nul nul P ~
> 0001 0000 a201 8086 0000 0000 0000 fed0

so not only is the table name inverted, but the lenght is inverted, and several fields later on are inverted too.

Other AML files for APIC and DSDT show similar brokenness

This is seen with QEMU 2.10.0

Revision history for this message
Daniel Berrange (berrange) wrote :
Revision history for this message
Daniel Berrange (berrange) wrote :
Revision history for this message
Daniel Berrange (berrange) wrote :
Revision history for this message
Daniel Berrange (berrange) wrote :
Revision history for this message
Daniel Berrange (berrange) wrote :
Revision history for this message
Daniel Berrange (berrange) wrote :
Revision history for this message
Thomas Huth (th-huth) wrote :

I think something like this should fix this issue:

diff a/tests/bios-tables-test.c b/tests/bios-tables-test.c
--- a/tests/bios-tables-test.c
+++ b/tests/bios-tables-test.c
@@ -279,8 +279,19 @@ static void dump_aml_files(test_data *data, bool rebuild)
         }
         g_assert(fd >= 0);

+ sdt->header.signature = cpu_to_le32(sdt->header.signature);
+ sdt->header.length = cpu_to_le32(sdt->header.length);
+ sdt->header.oem_revision = cpu_to_le32(sdt->header.oem_revision);
+ sdt->header.asl_compiler_revision = cpu_to_le32(sdt->header.asl_compiler_revision);
+
         ret = qemu_write_full(fd, sdt, sizeof(AcpiTableHeader));
         g_assert(ret == sizeof(AcpiTableHeader));
+
+ sdt->header.signature = le32_to_cpu(sdt->header.signature);
+ sdt->header.length = le32_to_cpu(sdt->header.length);
+ sdt->header.oem_revision = le32_to_cpu(sdt->header.oem_revision);
+ sdt->header.asl_compiler_revision = le32_to_cpu(sdt->header.asl_compiler_revision);
+
         ret = qemu_write_full(fd, sdt->aml, sdt->aml_len);
         g_assert(ret == sdt->aml_len);

Revision history for this message
Thomas Huth (th-huth) wrote :
Changed in qemu:
status: New → Fix Committed
Thomas Huth (th-huth)
Changed in qemu:
status: Fix Committed → Fix Released
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers