On 31/01/2019 08:19, Christoffer Dall wrote: > On Mon, Jan 28, 2019 at 03:42:42PM +0000, Julien Thierry wrote: >> Hi James, >> >> On 28/01/2019 11:48, James Morse wrote: >>> Hi Julien, >>> >>> On 21/01/2019 15:33, Julien Thierry wrote: >>>> When using VHE, the host needs to clear HCR_EL2.TGE bit in order >>>> to interract with guest TLBs, switching from EL2&0 translation regime >>> >>> (interact) >>> >>> >>>> to EL1&0. >>>> >>>> However, some non-maskable asynchronous event could happen while TGE is >>>> cleared like SDEI. Because of this address translation operations >>>> relying on EL2&0 translation regime could fail (tlb invalidation, >>>> userspace access, ...). >>>> >>>> Fix this by properly setting HCR_EL2.TGE when entering NMI context and >>>> clear it if necessary when returning to the interrupted context. >>> >>> Yes please. This would not have been fun to debug! >>> >>> Reviewed-by: James Morse <james.morse@xxxxxxx> >>> >>> >> >> Thanks. >> >>> >>> I was looking for why we need core code to do this, instead of updating the >>> arch's call sites. Your 'irqdesc: Add domain handlers for NMIs' patch (pointed >>> to from the cover letter) is the reason: core-code calls nmi_enter()/nmi_exit() >>> itself. >>> >> >> Yes, that's the main reason. >> > I wondered the same thing, but I don't understand the explanation :( > > Why can't we do a local_daif_mask() around the (very small) calls that > clear TGE instead? > That would protect against the pseudo-NMIs, but you can still get an SDEI at that point even with all daif bits set. Or did I misunderstand how SDEI works? Thanks, -- Julien Thierry
