[RFC,v2,0/2] Attempt to fix DMA reentrancy issues

Message

Alexander Bulekov Dec. 17, 2021, 2:37 p.m. UTC

Here's my shot at fixing dma-reentracy issues. This patch adds a flag to
the DeviceState, which is set/checked when we call an accessor
associated with the device's IO MRs.

The problem, in short, as I understand it: For the vast majority of
cases, we want to prevent a device from accessing it's own PIO/MMIO
regions over DMA.

V2: Try to fix reentrancies initiated by DMA accesses in BHs

Cc: Philippe Mathieu-Daudé <[e-mail]>
Cc: Mauro Matteo Cascella <[e-mail]>
Cc: Qiuhao Li <[e-mail]>
Cc: Peter Xu <[e-mail]>
Cc: Jason Wang <[e-mail]>
Cc: David Hildenbrand <[e-mail]>
Cc: Gerd Hoffmann <[e-mail]>
Cc: Peter Maydell <[e-mail]>
Cc: Li Qiang <[e-mail]>
Cc: Thomas Huth <[e-mail]>
Cc: Laurent Vivier <[e-mail]>
Cc: Bandan Das <[e-mail]>
Cc: Edgar E. Iglesias <[e-mail]>
Cc: Darren Kenny <[e-mail]>
Cc: Bin Meng <[e-mail]>
Cc: Paolo Bonzini <[e-mail]>
Cc: Stefan Hajnoczi <[e-mail]>


Alexander Bulekov (2):
  memory: fix dma-reentrancy issues at the MMIO level
  memory: set engaged_in_io when a device calls DMA APIs

 include/hw/pci/pci.h   |  6 +++++-
 include/hw/qdev-core.h |  1 +
 softmmu/dma-helpers.c  |  2 ++
 softmmu/memory.c       | 15 +++++++++++++++
 softmmu/trace-events   |  1 +
 5 files changed, 24 insertions(+), 1 deletion(-)