Re: [PATCH-for-5.1 v2 1/1] accel/xen: Fix xen_enabled() behavior on target-agnostic objects

[Philippe Mathieu-Daudé <philmd@xxxxxxxxxx>]

Hi Paul,

On 8/4/20 9:57 AM, Paul Durrant wrote:
>> -----Original Message-----
>> From: Philippe Mathieu-Daudé <philmd@xxxxxxxxxx>
>> Sent: 04 August 2020 08:50
>> To: qemu-devel@xxxxxxxxxx
>> Cc: Peter Maydell <peter.maydell@xxxxxxxxxx>; Anthony Perard 
>> <anthony.perard@xxxxxxxxxx>; Paolo
>> Bonzini <pbonzini@xxxxxxxxxx>; Stefano Stabellini <sstabellini@xxxxxxxxxx>; 
>> xen-
>> devel@xxxxxxxxxxxxxxxxxxxx; Paul Durrant <paul@xxxxxxx>; Philippe 
>> Mathieu-Daudé <philmd@xxxxxxxxxx>;
>> Paul Durrant <pdurrant@xxxxxxxxxx>
>> Subject: [PATCH-for-5.1 v2 1/1] accel/xen: Fix xen_enabled() behavior on 
>> target-agnostic objects
>>
>> CONFIG_XEN is generated by configure and stored in "config-target.h",
>> which is (obviously) only include for target-specific objects.
>> This is a problem for target-agnostic objects as CONFIG_XEN is never
>> defined and xen_enabled() is always inlined as 'false'.
>>
>> Fix by following the KVM schema, defining CONFIG_XEN_IS_POSSIBLE
>> when we don't know to force the call of the non-inlined function,
>> returning the xen_allowed boolean.
>>
>> Fixes: da278d58a092 ("accel: Move Xen accelerator code under accel/xen/")
>> Reported-by: Paul Durrant <pdurrant@xxxxxxxxxx>
>> Suggested-by: Peter Maydell <peter.maydell@xxxxxxxxxx>
>> Signed-off-by: Philippe Mathieu-Daudé <philmd@xxxxxxxxxx>
>> ---
>>  include/sysemu/xen.h   | 18 ++++++++++++++----
>>  accel/stubs/xen-stub.c |  2 ++
>>  accel/xen/xen-all.c    |  7 +------
>>  3 files changed, 17 insertions(+), 10 deletions(-)
>>
>> diff --git a/include/sysemu/xen.h b/include/sysemu/xen.h
>> index 1ca292715e..2c2c429ea8 100644
>> --- a/include/sysemu/xen.h
>> +++ b/include/sysemu/xen.h
>> @@ -8,9 +8,19 @@
>>  #ifndef SYSEMU_XEN_H
>>  #define SYSEMU_XEN_H
>>
>> -#ifdef CONFIG_XEN
>> +#ifdef NEED_CPU_H
>> +# ifdef CONFIG_XEN
>> +#  define CONFIG_XEN_IS_POSSIBLE
>> +# endif
>> +#else
>> +# define CONFIG_XEN_IS_POSSIBLE
>> +#endif
>>
>> -bool xen_enabled(void);
>> +#ifdef CONFIG_XEN_IS_POSSIBLE
>> +
>> +extern bool xen_allowed;
>> +
>> +#define xen_enabled()           (xen_allowed)
> 
> Can this not move ahead of the #ifdef now (since xen_allowed is present in 
> both xen-stub and xen-all)? I think this is what Peter was saying in his 
> option '(2)'.

I think I respected Peter's option '(2)', following how KVM does, this
is the case with stub,

> 
>   Paul
> 
>>
>>  #ifndef CONFIG_USER_ONLY
>>  void xen_hvm_modified_memory(ram_addr_t start, ram_addr_t length);
>> @@ -18,7 +28,7 @@ void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size,
>>                     struct MemoryRegion *mr, Error **errp);
>>  #endif
>>
>> -#else /* !CONFIG_XEN */
>> +#else /* !CONFIG_XEN_IS_POSSIBLE */
>>
>>  #define xen_enabled() 0

    ^^^ here is the other case,

>>  #ifndef CONFIG_USER_ONLY
>> @@ -33,6 +43,6 @@ static inline void xen_ram_alloc(ram_addr_t ram_addr, 
>> ram_addr_t size,
>>  }
>>  #endif
>>
>> -#endif /* CONFIG_XEN */
>> +#endif /* CONFIG_XEN_IS_POSSIBLE */
>>
>>  #endif
>> diff --git a/accel/stubs/xen-stub.c b/accel/stubs/xen-stub.c
>> index dcca4e678a..8ae658acff 100644
>> --- a/accel/stubs/xen-stub.c
>> +++ b/accel/stubs/xen-stub.c
>> @@ -9,6 +9,8 @@
>>  #include "hw/xen/xen.h"
>>  #include "qapi/qapi-commands-misc.h"
>>
>> +bool xen_allowed;

here is the stub,

>> +
>>  void xenstore_store_pv_console_info(int i, Chardev *chr)
>>  {
>>  }
>> diff --git a/accel/xen/xen-all.c b/accel/xen/xen-all.c
>> index 0c24d4b191..60b971d0a8 100644
>> --- a/accel/xen/xen-all.c
>> +++ b/accel/xen/xen-all.c
>> @@ -32,12 +32,7 @@
>>      do { } while (0)
>>  #endif
>>
>> -static bool xen_allowed;
>> -
>> -bool xen_enabled(void)
>> -{
>> -    return xen_allowed;
>> -}
>> +bool xen_allowed;

here is the real variable.

>>
>>  xc_interface *xen_xc;
>>  xenforeignmemory_handle *xen_fmem;
>> --
>> 2.21.3
> 
>