Tim, was it really intended for "x86/hvm: use unlocked p2m lookups in hvmemul_rep_movs()" to special case p2m_mmio_dm but not p2m_mmio_direct? Jan
At 12:46 +0100 on 20 Sep (1379681171), Jan Beulich wrote:> Tim, > > was it really intended for "x86/hvm: use unlocked p2m lookups in > hvmemul_rep_movs()" to special case p2m_mmio_dm but not > p2m_mmio_direct?Hmm. It certainly doesn''t seem to handle that case very well now, but I''m not sure the code before was any better. AFAICT it would have passed mmio_direct accesses to hvmemul_do_mmio(), which would send them to qemu. Tim.
>>> On 20.09.13 at 14:05, Tim Deegan <tim@xen.org> wrote: > At 12:46 +0100 on 20 Sep (1379681171), Jan Beulich wrote: >> Tim, >> >> was it really intended for "x86/hvm: use unlocked p2m lookups in >> hvmemul_rep_movs()" to special case p2m_mmio_dm but not >> p2m_mmio_direct? > > Hmm. It certainly doesn''t seem to handle that case very well now, but > I''m not sure the code before was any better. AFAICT it would have > passed mmio_direct accesses to hvmemul_do_mmio(), which would send them > to qemu.Hmm, wait - if MMIO of a passed through device, other than for its port I/O, doesn''t get intercepted at all, but instead gets taken care of by there being a valid gfn->mfn translation in place, then indeed before and after your change things aren''t handled well. Perhaps we should bail from there if either side is p2m_mmio_direct as well as if both sides are p2m_mmio_dm: --- a/xen/arch/x86/hvm/emulate.c +++ b/xen/arch/x86/hvm/emulate.c @@ -799,6 +799,10 @@ static int hvmemul_rep_movs( (void) get_gfn_query_unlocked(current->domain, sgpa >> PAGE_SHIFT, &sp2mt); (void) get_gfn_query_unlocked(current->domain, dgpa >> PAGE_SHIFT, &dp2mt); + if ( sp2mt == p2m_mmio_direct || dp2mt == p2m_mmio_direct || + (sp2mt == p2m_mmio_dm && dp2mt == p2m_mmio_dm) ) + return X86EMUL_UNHANDLEABLE; + if ( sp2mt == p2m_mmio_dm ) return hvmemul_do_mmio( sgpa, reps, bytes_per_rep, dgpa, IOREQ_READ, df, NULL); Jan
At 13:41 +0100 on 20 Sep (1379684515), Jan Beulich wrote:> >>> On 20.09.13 at 14:05, Tim Deegan <tim@xen.org> wrote: > > At 12:46 +0100 on 20 Sep (1379681171), Jan Beulich wrote: > >> Tim, > >> > >> was it really intended for "x86/hvm: use unlocked p2m lookups in > >> hvmemul_rep_movs()" to special case p2m_mmio_dm but not > >> p2m_mmio_direct? > > > > Hmm. It certainly doesn''t seem to handle that case very well now, but > > I''m not sure the code before was any better. AFAICT it would have > > passed mmio_direct accesses to hvmemul_do_mmio(), which would send them > > to qemu. > > Hmm, wait - if MMIO of a passed through device, other than for > its port I/O, doesn''t get intercepted at all, but instead gets taken > care of by there being a valid gfn->mfn translation in place, then > indeed before and after your change things aren''t handled well. > Perhaps we should bail from there if either side is p2m_mmio_direct > as well as if both sides are p2m_mmio_dm:Sounds OK to me. Clearly mmio-mmio string operations aren''t something we need to make go fast. :) Reviewed-by: Tim Deegan <tim@xen.org>> --- a/xen/arch/x86/hvm/emulate.c > +++ b/xen/arch/x86/hvm/emulate.c > @@ -799,6 +799,10 @@ static int hvmemul_rep_movs( > (void) get_gfn_query_unlocked(current->domain, sgpa >> PAGE_SHIFT, &sp2mt); > (void) get_gfn_query_unlocked(current->domain, dgpa >> PAGE_SHIFT, &dp2mt); > > + if ( sp2mt == p2m_mmio_direct || dp2mt == p2m_mmio_direct || > + (sp2mt == p2m_mmio_dm && dp2mt == p2m_mmio_dm) ) > + return X86EMUL_UNHANDLEABLE; > + > if ( sp2mt == p2m_mmio_dm ) > return hvmemul_do_mmio( > sgpa, reps, bytes_per_rep, dgpa, IOREQ_READ, df, NULL); > > Jan >
>>> On 20.09.13 at 14:47, Tim Deegan <tim@xen.org> wrote: > At 13:41 +0100 on 20 Sep (1379684515), Jan Beulich wrote: >> >>> On 20.09.13 at 14:05, Tim Deegan <tim@xen.org> wrote: >> > At 12:46 +0100 on 20 Sep (1379681171), Jan Beulich wrote: >> >> Tim, >> >> >> >> was it really intended for "x86/hvm: use unlocked p2m lookups in >> >> hvmemul_rep_movs()" to special case p2m_mmio_dm but not >> >> p2m_mmio_direct? >> > >> > Hmm. It certainly doesn''t seem to handle that case very well now, but >> > I''m not sure the code before was any better. AFAICT it would have >> > passed mmio_direct accesses to hvmemul_do_mmio(), which would send them >> > to qemu. >> >> Hmm, wait - if MMIO of a passed through device, other than for >> its port I/O, doesn''t get intercepted at all, but instead gets taken >> care of by there being a valid gfn->mfn translation in place, then >> indeed before and after your change things aren''t handled well. >> Perhaps we should bail from there if either side is p2m_mmio_direct >> as well as if both sides are p2m_mmio_dm: > > Sounds OK to me. Clearly mmio-mmio string operations aren''t something > we need to make go fast. :)Who knows: Frame buffer internal moves might e.g. be desirable to get accelerated. But if we want to do such, be need to honor the original access size at the very least.> Reviewed-by: Tim Deegan <tim@xen.org>Actually the final patch is to be bigger - we need to do something similar in rep_ins and rep_outs. Will send the whole thing in a few minutes (after smoke testing). Jan>> --- a/xen/arch/x86/hvm/emulate.c >> +++ b/xen/arch/x86/hvm/emulate.c >> @@ -799,6 +799,10 @@ static int hvmemul_rep_movs( >> (void) get_gfn_query_unlocked(current->domain, sgpa >> PAGE_SHIFT, &sp2mt); >> (void) get_gfn_query_unlocked(current->domain, dgpa >> PAGE_SHIFT, &dp2mt); >> >> + if ( sp2mt == p2m_mmio_direct || dp2mt == p2m_mmio_direct || >> + (sp2mt == p2m_mmio_dm && dp2mt == p2m_mmio_dm) ) >> + return X86EMUL_UNHANDLEABLE; >> + >> if ( sp2mt == p2m_mmio_dm ) >> return hvmemul_do_mmio( >> sgpa, reps, bytes_per_rep, dgpa, IOREQ_READ, df, NULL); >> >> Jan >>
Reasonably Related Threads
- [PATCH 0 of 3] RFC Paging support for AMD NPT V2
- [PATCH 0 of 3] Miscellaneous populate-on-demand bugs
- [PATCH 00/11] Add virtual EPT support Xen.
- xen-unstable, winxp32 very poor performance on AMD FX-8150, I bisected and changeset is 24770:7f79475d3de7
- [PATCH 0 of 4 RFC] Populate-on-demand: Check pages being returned by the balloon driver