[Xen-devel] [PATCH] Fix acpi_dmar_zap/reinstate() (fixes S3 regression)

[Tomasz Wroblewski <tomasz.wroblewski@xxxxxxxxxx>]

Hello all,

Changeset  23013:65d26504e843 (ACPI: large cleanup) modified 
acpi_dmar_reinstate() and acpi_dmar_zap() to use global "dmar_table" 
pointer instead of fetching it dynamically via acpi_get_table (and it 
removed the get_dmar() function which was used to this). This global 
pointer is initialised once when parsing the acpi table.
However, this seems incorrect due to how acpi_get_table() and underlying 
__acpi_map_table() is implemented. It pretty much always returns the 
same virtual pointer but instead remaps the fixmap underlying that 
virtual pointer. So storing this virtual pointer in global variable is 
incorrect because the pointer is invalidated next time acpi_get_table() 
is called.  Therefore acpi_dmar_reinstate()/zap() actually modify not 
the DMAR table but the table which was last accessed by acpi_get_table 
(which happens to not be DMAR at least on some Lenovos we tested but 
likely more platforms). This causes the affected table corruption, its 
checksum corruption, and failure to resume from S3 second consecutive time.
Attached patch restores the previous behaviour before cset 
23013:65d26504e843 of fetching the dmar_table pointer dynamically. Some 
__init annotations needed to be removed as the acpi_get_table() function 
is now again used on suspend/resume path
Signed-off-by: Tomasz Wroblewski <tomasz.wroblewski@xxxxxxxxxx>

Attachment: fix-dmar-zap-reinstate.patch
Description: Text Data

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel