x86: don't crash when mapping a page using EFI runtime page tables
authorRoss Lagerwall <ross.lagerwall@citrix.com>
Mon, 1 Jun 2015 09:59:14 +0000 (11:59 +0200)
committerJan Beulich <jbeulich@suse.com>
Mon, 1 Jun 2015 09:59:14 +0000 (11:59 +0200)
commit591e1e357c29589e9d6121d8faadc4f4d3b9013e
tree219af12eca1437395328bd1f978066a095f1c2d4
parent47ec25a3c8cdd7a057af0a05e8e00257ef950437
x86: don't crash when mapping a page using EFI runtime page tables

When an interrupt is received during an EFI runtime service call, Xen
may call map_domain_page() while using the EFI runtime page tables.
This fails because, although the EFI runtime page tables are a
copy of the idle domain's page tables, current points at a different
domain's vCPU.

To fix this, return NULL from mapcache_current_vcpu() when using the EFI
runtime page tables which is treated equivalently to running in an idle
vCPU.

This issue can be reproduced by repeatedly calling GetVariable() from
dom0 while using VT-d, since VT-d frequently maps a page from interrupt
context.

Example call trace:
[<ffff82d0801615dc>] __find_next_zero_bit+0x28/0x60
[<ffff82d08016a10e>] map_domain_page+0x4c6/0x4eb
[<ffff82d080156ae6>] map_vtd_domain_page+0xd/0xf
[<ffff82d08015533a>] msi_msg_read_remap_rte+0xe3/0x1d8
[<ffff82d08014e516>] iommu_read_msi_from_ire+0x31/0x34
[<ffff82d08016ff6c>] set_msi_affinity+0x134/0x17a
[<ffff82d0801737b5>] move_masked_irq+0x5c/0x98
[<ffff82d080173816>] move_native_irq+0x25/0x36
[<ffff82d08016ffcb>] ack_nonmaskable_msi_irq+0x19/0x20
[<ffff82d08016ffdb>] ack_maskable_msi_irq+0x9/0x37
[<ffff82d080173e8b>] do_IRQ+0x251/0x635
[<ffff82d080234502>] common_interrupt+0x62/0x70
[<00000000df7ed2be>] 00000000df7ed2be

Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
xen/arch/x86/domain_page.c
xen/arch/x86/efi/stub.c
xen/common/efi/runtime.c
xen/include/xen/efi.h