x86/HVM: use available linear->phys translations in REP MOVS/STOS handling If we have the translation result available already, we should also use is here. In my tests with Linux guests this eliminates all calls to hvmemul_linear_to_phys() out of the two functions being changed. Signed-off-by: Jan Beulich --- a/xen/arch/x86/hvm/emulate.c +++ b/xen/arch/x86/hvm/emulate.c @@ -1156,6 +1156,7 @@ static int hvmemul_rep_movs( { struct hvm_emulate_ctxt *hvmemul_ctxt = container_of(ctxt, struct hvm_emulate_ctxt, ctxt); + struct hvm_vcpu_io *vio = ¤t->arch.hvm_vcpu.hvm_io; unsigned long saddr, daddr, bytes; paddr_t sgpa, dgpa; uint32_t pfec = PFEC_page_present; @@ -1178,16 +1179,43 @@ static int hvmemul_rep_movs( if ( hvmemul_ctxt->seg_reg[x86_seg_ss].attr.fields.dpl == 3 ) pfec |= PFEC_user_mode; - rc = hvmemul_linear_to_phys( - saddr, &sgpa, bytes_per_rep, reps, pfec, hvmemul_ctxt); - if ( rc != X86EMUL_OKAY ) - return rc; + bytes = PAGE_SIZE - (saddr & ~PAGE_MASK); + if ( vio->mmio_access.read_access && + (vio->mmio_gva == (saddr & PAGE_MASK)) && + bytes >= bytes_per_rep ) + { + sgpa = pfn_to_paddr(vio->mmio_gpfn) | (saddr & ~PAGE_MASK); + if ( *reps * bytes_per_rep > bytes ) + *reps = bytes / bytes_per_rep; + } + else + { + rc = hvmemul_linear_to_phys(saddr, &sgpa, bytes_per_rep, reps, pfec, + hvmemul_ctxt); + if ( rc != X86EMUL_OKAY ) + return rc; - rc = hvmemul_linear_to_phys( - daddr, &dgpa, bytes_per_rep, reps, - pfec | PFEC_write_access, hvmemul_ctxt); - if ( rc != X86EMUL_OKAY ) - return rc; + latch_linear_to_phys(vio, saddr, sgpa, 0); + } + + bytes = PAGE_SIZE - (daddr & ~PAGE_MASK); + if ( vio->mmio_access.write_access && + (vio->mmio_gva == (daddr & PAGE_MASK)) && + bytes >= bytes_per_rep ) + { + dgpa = pfn_to_paddr(vio->mmio_gpfn) | (daddr & ~PAGE_MASK); + if ( *reps * bytes_per_rep > bytes ) + *reps = bytes / bytes_per_rep; + } + else + { + rc = hvmemul_linear_to_phys(daddr, &dgpa, bytes_per_rep, reps, + pfec | PFEC_write_access, hvmemul_ctxt); + if ( rc != X86EMUL_OKAY ) + return rc; + + latch_linear_to_phys(vio, daddr, dgpa, 1); + } /* Check for MMIO ops */ (void) get_gfn_query_unlocked(current->domain, sgpa >> PAGE_SHIFT, &sp2mt); @@ -1279,25 +1307,40 @@ static int hvmemul_rep_stos( { struct hvm_emulate_ctxt *hvmemul_ctxt = container_of(ctxt, struct hvm_emulate_ctxt, ctxt); - unsigned long addr; + struct hvm_vcpu_io *vio = ¤t->arch.hvm_vcpu.hvm_io; + unsigned long addr, bytes; paddr_t gpa; p2m_type_t p2mt; bool_t df = !!(ctxt->regs->eflags & X86_EFLAGS_DF); int rc = hvmemul_virtual_to_linear(seg, offset, bytes_per_rep, reps, hvm_access_write, hvmemul_ctxt, &addr); - if ( rc == X86EMUL_OKAY ) + if ( rc != X86EMUL_OKAY ) + return rc; + + bytes = PAGE_SIZE - (addr & ~PAGE_MASK); + if ( vio->mmio_access.write_access && + (vio->mmio_gva == (addr & PAGE_MASK)) && + bytes >= bytes_per_rep ) + { + gpa = pfn_to_paddr(vio->mmio_gpfn) | (addr & ~PAGE_MASK); + if ( *reps * bytes_per_rep > bytes ) + *reps = bytes / bytes_per_rep; + } + else { uint32_t pfec = PFEC_page_present | PFEC_write_access; if ( hvmemul_ctxt->seg_reg[x86_seg_ss].attr.fields.dpl == 3 ) pfec |= PFEC_user_mode; - rc = hvmemul_linear_to_phys( - addr, &gpa, bytes_per_rep, reps, pfec, hvmemul_ctxt); + rc = hvmemul_linear_to_phys(addr, &gpa, bytes_per_rep, reps, pfec, + hvmemul_ctxt); + if ( rc != X86EMUL_OKAY ) + return rc; + + latch_linear_to_phys(vio, addr, gpa, 1); } - if ( rc != X86EMUL_OKAY ) - return rc; /* Check for MMIO op */ (void)get_gfn_query_unlocked(current->domain, gpa >> PAGE_SHIFT, &p2mt);