On 7/28/20 11:56 AM, Philippe Mathieu-Daudé wrote:
> On 7/28/20 11:53 AM, Peter Maydell wrote:
>> On Tue, 28 Jul 2020 at 10:51, Philippe Mathieu-Daudé <phi...@redhat.com>
>> wrote:
>>> I'd rather uninline xen_enabled() but I'm not sure this has perf
>>> penalties. Paolo is that OK to uninline it?
>
> I suppose no because it is in various hot paths:
>
> exec.c:588: if (xen_enabled() && memory_access_is_direct(mr, is_write)) {
> exec.c:2243: if (xen_enabled()) {
> exec.c:2326: if (xen_enabled()) {
> exec.c:2478: } else if (xen_enabled()) {
> exec.c:2525: } else if (xen_enabled()) {
> exec.c:2576: if (xen_enabled() && block->host == NULL) {
> exec.c:2609: if (xen_enabled() && block->host == NULL) {
> exec.c:2657: if (xen_enabled()) {
> exec.c:3625: if (xen_enabled()) {
> exec.c:3717: if (xen_enabled()) {
> include/exec/ram_addr.h:295: if (!mask && !xen_enabled()) {
>
>>
>> Can we just follow the same working pattern we already have
>> for kvm_enabled() etc ?
>
> This was the idea... I'll look at what I missed.
Apparently kvm_enabled() checks CONFIG_KVM_IS_POSSIBLE instead
of CONFIG_KVM, I suppose to bypass this limitation (from osdep.h):
21 #ifdef NEED_CPU_H
22 # ifdef CONFIG_KVM
24 # define CONFIG_KVM_IS_POSSIBLE
25 # endif
26 #else
27 # define CONFIG_KVM_IS_POSSIBLE
28 #endif
29
30 #ifdef CONFIG_KVM_IS_POSSIBLE
...
Paolo do you confirm this is the reason?
I'll prepare a similar patch.
>
> Phil.
>
