powerpc/eeh: Create PHB PEs after EEH is initialized
Subsystems (2)
Otherwise we end up not yet having computed the right diag data size on powernv where EEH initialization is delayed, thus causing memory corruption later on when calling OPAL.
---
Russell, what do you think ? The end result is that the PEs for the PHBs are created much later. I suppose that might cause changes of behaviour if, for example, we hit EEH while probing before we call eeh_init() again. Hopefully we have all the appropriate NULL checks to deal with it though...
Another option would be to break up eeh_init() between calling the backend init, which we would unconditionally do early (ie removing the special powernv test) and a new eeh_create_pe()'s which would explicitely be called by the backend at the "right" time...
Without either fix, we are currently corrupting kernel memory when hitting EEH on a PHB PE (fences for example).
If this ends up being the right approach, then we need a CC stable as well.
arch/powerpc/kernel/eeh.c | 4 ++++ arch/powerpc/kernel/eeh_dev.c | 18 ------------------ 2 files changed, 4 insertions(+), 18 deletions(-)
arch/powerpc/kernel/eeh.c +4 -0
@@ -1018,6 +1018,10 @@ int eeh_init(void) }else if ((ret = eeh_ops->init ()))return ret; +/* Initialize PHB PEs */ +list_for_each_entry_safe (hose, tmp, &hose_list, list_node) +eeh_dev_phb_init_dynamic (hose); +/* Initialize EEH event */ ret =eeh_event_init ();if (ret)
arch/powerpc/kernel/eeh_dev.c +0 -18
@@ -83,21 +83,3 @@ void eeh_dev_phb_init_dynamic(struct pci_controller *phb)/* EEH PE for PHB */ eeh_phb_pe_create (phb); } - -/** -* eeh_dev_phb_init - Create EEH devices for devices included in existing PHBs -* -* Scan all the existing PHBs and create EEH devices for their OF -* nodes and their children OF nodes -*/ -static int __init eeh_dev_phb_init (void ) -{ -struct pci_controller *phb, *tmp; - -list_for_each_entry_safe (phb, tmp, &hose_list, list_node) -eeh_dev_phb_init_dynamic (phb); - -return 0 ; -} - -core_initcall (eeh_dev_phb_init);