[Qemu-devel] [PATCH] hw/core/generic-loader: Compile only once, not for each target

[Qemu-devel] [PATCH] hw/core/generic-loader: Compile only once, not for each target

[Thomas Huth]

The generic-loader is currently compiled target specific due to one
single "#ifdef TARGET_WORDS_BIGENDIAN" in the file. We've already
got a function called target_words_bigendian() for this instead, so
we can put the generic-loader into common-obj to save some compilation
time.

Signed-off-by: Thomas Huth <thuth@redhat.com>
---
 exec.c                   | 1 -
 hw/core/Makefile.objs    | 2 +-
 hw/core/generic-loader.c | 6 +-----
 hw/virtio/virtio.c       | 1 -
 include/qom/cpu.h        | 2 ++
 qom/cpu.c                | 1 -
 6 files changed, 4 insertions(+), 9 deletions(-)

diff --git a/exec.c b/exec.c
index d0821e6..3ae0e54 100644
--- a/exec.c
+++ b/exec.c
@@ -3910,7 +3910,6 @@ int qemu_target_page_bits_min(void)
  * A helper function for the _utterly broken_ virtio device model to find out if
  * it's running on a big endian machine. Don't do this at home kids!
  */
-bool target_words_bigendian(void);
 bool target_words_bigendian(void)
 {
 #if defined(TARGET_WORDS_BIGENDIAN)
diff --git a/hw/core/Makefile.objs b/hw/core/Makefile.objs
index eb88ca9..b736ce2 100644
--- a/hw/core/Makefile.objs
+++ b/hw/core/Makefile.objs
@@ -20,6 +20,6 @@ common-obj-$(CONFIG_SOFTMMU) += register.o
 common-obj-$(CONFIG_SOFTMMU) += or-irq.o
 common-obj-$(CONFIG_SOFTMMU) += split-irq.o
 common-obj-$(CONFIG_PLATFORM_BUS) += platform-bus.o
+common-obj-$(CONFIG_SOFTMMU) += generic-loader.o
 
-obj-$(CONFIG_SOFTMMU) += generic-loader.o
 obj-$(CONFIG_SOFTMMU) += null-machine.o
diff --git a/hw/core/generic-loader.c b/hw/core/generic-loader.c
index be29ae1..fbae05f 100644
--- a/hw/core/generic-loader.c
+++ b/hw/core/generic-loader.c
@@ -130,11 +130,7 @@ static void generic_loader_realize(DeviceState *dev, Error **errp)
         s->cpu = first_cpu;
     }
 
-#ifdef TARGET_WORDS_BIGENDIAN
-    big_endian = 1;
-#else
-    big_endian = 0;
-#endif
+    big_endian = target_words_bigendian();
 
     if (s->file) {
         AddressSpace *as = s->cpu ? s->cpu->as :  NULL;
diff --git a/hw/virtio/virtio.c b/hw/virtio/virtio.c
index 94f5c8e..4e61944 100644
--- a/hw/virtio/virtio.c
+++ b/hw/virtio/virtio.c
@@ -1169,7 +1169,6 @@ int virtio_set_status(VirtIODevice *vdev, uint8_t val)
     return 0;
 }
 
-bool target_words_bigendian(void);
 static enum virtio_device_endian virtio_default_endian(void)
 {
     if (target_words_bigendian()) {
diff --git a/include/qom/cpu.h b/include/qom/cpu.h
index dc130cd..0e2ce80 100644
--- a/include/qom/cpu.h
+++ b/include/qom/cpu.h
@@ -1085,6 +1085,8 @@ void cpu_exec_initfn(CPUState *cpu);
 void cpu_exec_realizefn(CPUState *cpu, Error **errp);
 void cpu_exec_unrealizefn(CPUState *cpu);
 
+bool target_words_bigendian(void);
+
 #ifdef NEED_CPU_H
 
 #ifdef CONFIG_SOFTMMU
diff --git a/qom/cpu.c b/qom/cpu.c
index 92599f3..f774654 100644
--- a/qom/cpu.c
+++ b/qom/cpu.c
@@ -194,7 +194,6 @@ static bool cpu_common_debug_check_watchpoint(CPUState *cpu, CPUWatchpoint *wp)
     return true;
 }
 
-bool target_words_bigendian(void);
 static bool cpu_common_virtio_is_big_endian(CPUState *cpu)
 {
     return target_words_bigendian();
-- 
1.8.3.1

Re: [Qemu-devel] [PATCH] hw/core/generic-loader: Compile only once, not for each target

[Peter Maydell]

On 5 October 2018 at 11:15, Thomas Huth <thuth@redhat.com> wrote:
> The generic-loader is currently compiled target specific due to one
> single "#ifdef TARGET_WORDS_BIGENDIAN" in the file. We've already
> got a function called target_words_bigendian() for this instead, so
> we can put the generic-loader into common-obj to save some compilation
> time.
>
> Signed-off-by: Thomas Huth <thuth@redhat.com>
> ---
>  exec.c                   | 1 -
>  hw/core/Makefile.objs    | 2 +-
>  hw/core/generic-loader.c | 6 +-----
>  hw/virtio/virtio.c       | 1 -
>  include/qom/cpu.h        | 2 ++
>  qom/cpu.c                | 1 -
>  6 files changed, 4 insertions(+), 9 deletions(-)
>
> diff --git a/exec.c b/exec.c
> index d0821e6..3ae0e54 100644
> --- a/exec.c
> +++ b/exec.c
> @@ -3910,7 +3910,6 @@ int qemu_target_page_bits_min(void)
>   * A helper function for the _utterly broken_ virtio device model to find out if
>   * it's running on a big endian machine. Don't do this at home kids!
>   */
> -bool target_words_bigendian(void);
>  bool target_words_bigendian(void)
>  {
>  #if defined(TARGET_WORDS_BIGENDIAN)

> --- a/include/qom/cpu.h
> +++ b/include/qom/cpu.h
> @@ -1085,6 +1085,8 @@ void cpu_exec_initfn(CPUState *cpu);
>  void cpu_exec_realizefn(CPUState *cpu, Error **errp);
>  void cpu_exec_unrealizefn(CPUState *cpu);
>
> +bool target_words_bigendian(void);
> +

Can we have a doc comment for a new global function in a header
file, please? In particular this one needs a big health warning
that using it is almost always the wrong thing (compare the
comment in the .c file).

thanks
-- PMM

Re: [Qemu-devel] [PATCH] hw/core/generic-loader: Compile only once, not for each target

[Philippe Mathieu-Daudé]

Hi Thomas,

On 10/5/18 12:15 PM, Thomas Huth wrote:
> The generic-loader is currently compiled target specific due to one
> single "#ifdef TARGET_WORDS_BIGENDIAN" in the file. We've already
> got a function called target_words_bigendian() for this instead, so
> we can put the generic-loader into common-obj to save some compilation
> time.

Please do this in 2 patches:
- clean the target_words_bigendian() mess, if possible documenting the
new function added in "qom/cpu.h"
- then move generic-loader to common

Is the comment added in 8e4a424b305 still valid?

/*
 * A helper function for the _utterly broken_ virtio device model to
find out if
 * it's running on a big endian machine. Don't do this at home kids!
 */
bool target_words_bigendian(void);

Thanks,

Phil.

> 
> Signed-off-by: Thomas Huth <thuth@redhat.com>
> ---
>  exec.c                   | 1 -
>  hw/core/Makefile.objs    | 2 +-
>  hw/core/generic-loader.c | 6 +-----
>  hw/virtio/virtio.c       | 1 -
>  include/qom/cpu.h        | 2 ++
>  qom/cpu.c                | 1 -
>  6 files changed, 4 insertions(+), 9 deletions(-)
> 
> diff --git a/exec.c b/exec.c
> index d0821e6..3ae0e54 100644
> --- a/exec.c
> +++ b/exec.c
> @@ -3910,7 +3910,6 @@ int qemu_target_page_bits_min(void)
>   * A helper function for the _utterly broken_ virtio device model to find out if
>   * it's running on a big endian machine. Don't do this at home kids!
>   */
> -bool target_words_bigendian(void);
>  bool target_words_bigendian(void)
>  {
>  #if defined(TARGET_WORDS_BIGENDIAN)
> diff --git a/hw/core/Makefile.objs b/hw/core/Makefile.objs
> index eb88ca9..b736ce2 100644
> --- a/hw/core/Makefile.objs
> +++ b/hw/core/Makefile.objs
> @@ -20,6 +20,6 @@ common-obj-$(CONFIG_SOFTMMU) += register.o
>  common-obj-$(CONFIG_SOFTMMU) += or-irq.o
>  common-obj-$(CONFIG_SOFTMMU) += split-irq.o
>  common-obj-$(CONFIG_PLATFORM_BUS) += platform-bus.o
> +common-obj-$(CONFIG_SOFTMMU) += generic-loader.o
>  
> -obj-$(CONFIG_SOFTMMU) += generic-loader.o
>  obj-$(CONFIG_SOFTMMU) += null-machine.o
> diff --git a/hw/core/generic-loader.c b/hw/core/generic-loader.c
> index be29ae1..fbae05f 100644
> --- a/hw/core/generic-loader.c
> +++ b/hw/core/generic-loader.c
> @@ -130,11 +130,7 @@ static void generic_loader_realize(DeviceState *dev, Error **errp)
>          s->cpu = first_cpu;
>      }
>  
> -#ifdef TARGET_WORDS_BIGENDIAN
> -    big_endian = 1;
> -#else
> -    big_endian = 0;
> -#endif
> +    big_endian = target_words_bigendian();
>  
>      if (s->file) {
>          AddressSpace *as = s->cpu ? s->cpu->as :  NULL;
> diff --git a/hw/virtio/virtio.c b/hw/virtio/virtio.c
> index 94f5c8e..4e61944 100644
> --- a/hw/virtio/virtio.c
> +++ b/hw/virtio/virtio.c
> @@ -1169,7 +1169,6 @@ int virtio_set_status(VirtIODevice *vdev, uint8_t val)
>      return 0;
>  }
>  
> -bool target_words_bigendian(void);
>  static enum virtio_device_endian virtio_default_endian(void)
>  {
>      if (target_words_bigendian()) {
> diff --git a/include/qom/cpu.h b/include/qom/cpu.h
> index dc130cd..0e2ce80 100644
> --- a/include/qom/cpu.h
> +++ b/include/qom/cpu.h
> @@ -1085,6 +1085,8 @@ void cpu_exec_initfn(CPUState *cpu);
>  void cpu_exec_realizefn(CPUState *cpu, Error **errp);
>  void cpu_exec_unrealizefn(CPUState *cpu);
>  
> +bool target_words_bigendian(void);
> +
>  #ifdef NEED_CPU_H
>  
>  #ifdef CONFIG_SOFTMMU
> diff --git a/qom/cpu.c b/qom/cpu.c
> index 92599f3..f774654 100644
> --- a/qom/cpu.c
> +++ b/qom/cpu.c
> @@ -194,7 +194,6 @@ static bool cpu_common_debug_check_watchpoint(CPUState *cpu, CPUWatchpoint *wp)
>      return true;
>  }
>  
> -bool target_words_bigendian(void);
>  static bool cpu_common_virtio_is_big_endian(CPUState *cpu)
>  {
>      return target_words_bigendian();
> 

Re: [Qemu-devel] [PATCH] hw/core/generic-loader: Compile only once, not for each target

[Laszlo Ersek]

On 10/05/18 12:49, Philippe Mathieu-Daudé wrote:
> Hi Thomas,
> 
> On 10/5/18 12:15 PM, Thomas Huth wrote:
>> The generic-loader is currently compiled target specific due to one
>> single "#ifdef TARGET_WORDS_BIGENDIAN" in the file. We've already
>> got a function called target_words_bigendian() for this instead, so
>> we can put the generic-loader into common-obj to save some compilation
>> time.
> 
> Please do this in 2 patches:
> - clean the target_words_bigendian() mess, if possible documenting the
> new function added in "qom/cpu.h"
> - then move generic-loader to common
> 
> Is the comment added in 8e4a424b305 still valid?
> 
> /*
>  * A helper function for the _utterly broken_ virtio device model to
> find out if
>  * it's running on a big endian machine. Don't do this at home kids!
>  */
> bool target_words_bigendian(void);

Yeah, that comment irks me too.

Not that it isn't valid, necessarily. "hw/virtio/virtio.c" still
features an embedded declaration of this function. Just above
virtio_default_endian(), which calls it. And the latter function is
called in several places (mostly from commit 616a655219a92).

So I can't really tell what's "utterly broken" here -- the device model
(i.e., the virtio implementation in QEMU), or the legacy *spec* that
required QEMU to jump through such hoops?

(Either way, the embedded declaration of the function should go.)

BTW, I don't think angry comments are useful in *code*. They look & feel
great for about a week. In a year, they just look unprofessional. (NB,
this is not about "political correctness"; it's just that, when someone
reads such a comment, they receive the anger *before* they understand,
or recall, the underlying problem.)

(Don't mean to derail Thomas's patch with some bikeshedding here...)

Thanks,
Laszlo

>> Signed-off-by: Thomas Huth <thuth@redhat.com>
>> ---
>>  exec.c                   | 1 -
>>  hw/core/Makefile.objs    | 2 +-
>>  hw/core/generic-loader.c | 6 +-----
>>  hw/virtio/virtio.c       | 1 -
>>  include/qom/cpu.h        | 2 ++
>>  qom/cpu.c                | 1 -
>>  6 files changed, 4 insertions(+), 9 deletions(-)
>>
>> diff --git a/exec.c b/exec.c
>> index d0821e6..3ae0e54 100644
>> --- a/exec.c
>> +++ b/exec.c
>> @@ -3910,7 +3910,6 @@ int qemu_target_page_bits_min(void)
>>   * A helper function for the _utterly broken_ virtio device model to find out if
>>   * it's running on a big endian machine. Don't do this at home kids!
>>   */
>> -bool target_words_bigendian(void);
>>  bool target_words_bigendian(void)
>>  {
>>  #if defined(TARGET_WORDS_BIGENDIAN)
>> diff --git a/hw/core/Makefile.objs b/hw/core/Makefile.objs
>> index eb88ca9..b736ce2 100644
>> --- a/hw/core/Makefile.objs
>> +++ b/hw/core/Makefile.objs
>> @@ -20,6 +20,6 @@ common-obj-$(CONFIG_SOFTMMU) += register.o
>>  common-obj-$(CONFIG_SOFTMMU) += or-irq.o
>>  common-obj-$(CONFIG_SOFTMMU) += split-irq.o
>>  common-obj-$(CONFIG_PLATFORM_BUS) += platform-bus.o
>> +common-obj-$(CONFIG_SOFTMMU) += generic-loader.o
>>  
>> -obj-$(CONFIG_SOFTMMU) += generic-loader.o
>>  obj-$(CONFIG_SOFTMMU) += null-machine.o
>> diff --git a/hw/core/generic-loader.c b/hw/core/generic-loader.c
>> index be29ae1..fbae05f 100644
>> --- a/hw/core/generic-loader.c
>> +++ b/hw/core/generic-loader.c
>> @@ -130,11 +130,7 @@ static void generic_loader_realize(DeviceState *dev, Error **errp)
>>          s->cpu = first_cpu;
>>      }
>>  
>> -#ifdef TARGET_WORDS_BIGENDIAN
>> -    big_endian = 1;
>> -#else
>> -    big_endian = 0;
>> -#endif
>> +    big_endian = target_words_bigendian();
>>  
>>      if (s->file) {
>>          AddressSpace *as = s->cpu ? s->cpu->as :  NULL;
>> diff --git a/hw/virtio/virtio.c b/hw/virtio/virtio.c
>> index 94f5c8e..4e61944 100644
>> --- a/hw/virtio/virtio.c
>> +++ b/hw/virtio/virtio.c
>> @@ -1169,7 +1169,6 @@ int virtio_set_status(VirtIODevice *vdev, uint8_t val)
>>      return 0;
>>  }
>>  
>> -bool target_words_bigendian(void);
>>  static enum virtio_device_endian virtio_default_endian(void)
>>  {
>>      if (target_words_bigendian()) {
>> diff --git a/include/qom/cpu.h b/include/qom/cpu.h
>> index dc130cd..0e2ce80 100644
>> --- a/include/qom/cpu.h
>> +++ b/include/qom/cpu.h
>> @@ -1085,6 +1085,8 @@ void cpu_exec_initfn(CPUState *cpu);
>>  void cpu_exec_realizefn(CPUState *cpu, Error **errp);
>>  void cpu_exec_unrealizefn(CPUState *cpu);
>>  
>> +bool target_words_bigendian(void);
>> +
>>  #ifdef NEED_CPU_H
>>  
>>  #ifdef CONFIG_SOFTMMU
>> diff --git a/qom/cpu.c b/qom/cpu.c
>> index 92599f3..f774654 100644
>> --- a/qom/cpu.c
>> +++ b/qom/cpu.c
>> @@ -194,7 +194,6 @@ static bool cpu_common_debug_check_watchpoint(CPUState *cpu, CPUWatchpoint *wp)
>>      return true;
>>  }
>>  
>> -bool target_words_bigendian(void);
>>  static bool cpu_common_virtio_is_big_endian(CPUState *cpu)
>>  {
>>      return target_words_bigendian();
>>
> 

Re: [Qemu-devel] [PATCH] hw/core/generic-loader: Compile only once, not for each target

[Laszlo Ersek]

On 10/05/18 14:17, Laszlo Ersek wrote:

> Not that it isn't valid, necessarily. "hw/virtio/virtio.c" still
> features an embedded declaration of this function. Just above
> virtio_default_endian(), which calls it. [...]
> 
> (Either way, the embedded declaration of the function should go.)

(which is what the patch already does, of course!)
Laszlo

Re: [Qemu-devel] [PATCH] hw/core/generic-loader: Compile only once, not for each target

[Peter Maydell]

On 5 October 2018 at 13:17, Laszlo Ersek <lersek@redhat.com> wrote:
> On 10/05/18 12:49, Philippe Mathieu-Daudé wrote:
>> Is the comment added in 8e4a424b305 still valid?
>>
>> /*
>>  * A helper function for the _utterly broken_ virtio device model to
>> find out if
>>  * it's running on a big endian machine. Don't do this at home kids!
>>  */
>> bool target_words_bigendian(void);
>
> Yeah, that comment irks me too.
>
> Not that it isn't valid, necessarily. "hw/virtio/virtio.c" still
> features an embedded declaration of this function. Just above
> virtio_default_endian(), which calls it. And the latter function is
> called in several places (mostly from commit 616a655219a92).
>
> So I can't really tell what's "utterly broken" here -- the device model
> (i.e., the virtio implementation in QEMU), or the legacy *spec* that
> required QEMU to jump through such hoops?

AIUI, the brokenness here is that the virtio specification
requires that a device knows the endianness of the CPU,
which is an awkward layering violation [*]. The point of the
comment is really to say "virtio needs this, but that is
a special weird case, think twice or three times before
using this function anywhere else" (which I suspect is also
why the declaration is not in a header file, to make it less
easily accessible).

[*] Actually it's not even the endianness of the CPU, because
bi-endian CPUs like Arm can be in big-endian mode even though
TARGET_WORDS_BIGENDIAN is false...

I entirely agree that we should clean up the comment to be
clearer about what it's trying to say.

> BTW, I don't think angry comments are useful in *code*. They look & feel
> great for about a week. In a year, they just look unprofessional. (NB,
> this is not about "political correctness"; it's just that, when someone
> reads such a comment, they receive the anger *before* they understand,
> or recall, the underlying problem.

Agreed 100%.

thanks
-- PMM