diff mbox series

[2/2] drm/i915: reword documentation of possible pci_device_id struct

Message ID 20180828174146.29894-2-lucas.demarchi@intel.com (mailing list archive)
State New, archived
Headers show
Series [1/2] drm/i915: make field unsigned | expand

Commit Message

Lucas De Marchi Aug. 28, 2018, 5:41 p.m. UTC
Document it like a real struct for ease of copy and paste, remove
comment of C99 compatibility and document that in some cases the first 2
fields can be u16.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
---
 include/drm/i915_pciids.h | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)

Comments

Chris Wilson Aug. 28, 2018, 6:05 p.m. UTC | #1
Quoting Lucas De Marchi (2018-08-28 18:41:46)
> Document it like a real struct for ease of copy and paste, remove
> comment of C99 compatibility and document that in some cases the first 2

I do recall that we couldn't use either C99 or class due to userspace
compatibility... The essence is that we need a reminder that we can't
assume the relaxed nature of kcc here.
-Chris
Ville Syrjälä Aug. 28, 2018, 6:06 p.m. UTC | #2
On Tue, Aug 28, 2018 at 10:41:46AM -0700, Lucas De Marchi wrote:
> Document it like a real struct for ease of copy and paste, remove
> comment of C99 compatibility and document that in some cases the first 2
> fields can be u16.
> 
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
> ---
>  include/drm/i915_pciids.h | 16 +++++++++-------
>  1 file changed, 9 insertions(+), 7 deletions(-)
> 
> diff --git a/include/drm/i915_pciids.h b/include/drm/i915_pciids.h
> index 754ce4b10129..0c2cc43f916c 100644
> --- a/include/drm/i915_pciids.h
> +++ b/include/drm/i915_pciids.h
> @@ -26,14 +26,16 @@
>  #define _I915_PCIIDS_H
>  
>  /*
> - * A pci_device_id struct {
> - *	__u32 vendor, device;
> - *      __u32 subvendor, subdevice;
> - *	__u32 class, class_mask;
> - *	kernel_ulong_t driver_data;
> + * These macros can be used with a struct declared like this:
> + *
> + * struct pci_device_id {
> + * 	__u32 vendor, device;
> + * 	__u32 subvendor, subdevice;
> + * 	__u32 class, class_mask;
> + * 	kernel_ulong_t driver_data;
>   * };
> - * Don't use C99 here because "class" is reserved and we want to
> - * give userspace flexibility.
> + *
> + * First two fields may be __u16 if PCI_DEVICE_ANY is not used

PCI_DEVICE_ANY undefined?

Also you can surely use u16 just fine as long as you're careful when
comparing with ~0?

>   */
>  #define INTEL_VGA_DEVICE(id, info) {		\
>  	0x8086,	id,				\
> -- 
> 2.17.1
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Lucas De Marchi Aug. 28, 2018, 7:30 p.m. UTC | #3
On Tue, Aug 28, 2018 at 07:05:46PM +0100, Chris Wilson wrote:
> Quoting Lucas De Marchi (2018-08-28 18:41:46)
> > Document it like a real struct for ease of copy and paste, remove
> > comment of C99 compatibility and document that in some cases the first 2
> 
> I do recall that we couldn't use either C99 or class due to userspace

you can't actually use a c++ compiler.

For C it works with any of -std=c99, gnu99, c11, gnu11, c17, gnu17.
Tested with both gcc and clang. I've never heard of class being a
reserved keyword and section 6.4.5 of said standard doesn't list it
neither.

Here the struct definition is in a *comment*... i.e. the user will copy
and paste somewhere else and probably change __u16 to uint16_t in
userspace. If he's building with g++, he can name the field to something
else.

If it was something we were defining in this header than I would agree
with you... to retain compatibility with c++, not c99.

Lucas De Marchi

> compatibility... The essence is that we need a reminder that we can't
> assume the relaxed nature of kcc here.
> -Chris
Lucas De Marchi Aug. 28, 2018, 7:34 p.m. UTC | #4
On Tue, Aug 28, 2018 at 09:06:15PM +0300, Ville Syrjälä wrote:
> On Tue, Aug 28, 2018 at 10:41:46AM -0700, Lucas De Marchi wrote:
> > Document it like a real struct for ease of copy and paste, remove
> > comment of C99 compatibility and document that in some cases the first 2
> > fields can be u16.
> > 
> > Cc: Chris Wilson <chris@chris-wilson.co.uk>
> > Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
> > ---
> >  include/drm/i915_pciids.h | 16 +++++++++-------
> >  1 file changed, 9 insertions(+), 7 deletions(-)
> > 
> > diff --git a/include/drm/i915_pciids.h b/include/drm/i915_pciids.h
> > index 754ce4b10129..0c2cc43f916c 100644
> > --- a/include/drm/i915_pciids.h
> > +++ b/include/drm/i915_pciids.h
> > @@ -26,14 +26,16 @@
> >  #define _I915_PCIIDS_H
> >  
> >  /*
> > - * A pci_device_id struct {
> > - *	__u32 vendor, device;
> > - *      __u32 subvendor, subdevice;
> > - *	__u32 class, class_mask;
> > - *	kernel_ulong_t driver_data;
> > + * These macros can be used with a struct declared like this:
> > + *
> > + * struct pci_device_id {
> > + * 	__u32 vendor, device;
> > + * 	__u32 subvendor, subdevice;
> > + * 	__u32 class, class_mask;
> > + * 	kernel_ulong_t driver_data;
> >   * };
> > - * Don't use C99 here because "class" is reserved and we want to
> > - * give userspace flexibility.
> > + *
> > + * First two fields may be __u16 if PCI_DEVICE_ANY is not used
> 
> PCI_DEVICE_ANY undefined?

this should be PCI_ANY_ID, but now I'm not sure if I should even mention
it since it's defined somewhere that's private to kernel.

> 
> Also you can surely use u16 just fine as long as you're careful when
> comparing with ~0?

nops, since ~0u is unsigned int (~0 being int before patch 1), it will
overflow and cause a warning due to -Woverflow. This is what happens in
igt if I do what you said:

../intel/i915_pciids.h:40:2: warning: conversion from ‘unsigned int’ to ‘short unsigned int’ changes value from ‘4294967295’ to ‘65535’ [-Woverflow]
  ~0u, ~0u,    \
  ^


Lucas De Marchi

> 
> >   */
> >  #define INTEL_VGA_DEVICE(id, info) {		\
> >  	0x8086,	id,				\
> > -- 
> > 2.17.1
> > 
> > _______________________________________________
> > Intel-gfx mailing list
> > Intel-gfx@lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
> 
> -- 
> Ville Syrjälä
> Intel
Jani Nikula Sept. 4, 2018, 1:24 p.m. UTC | #5
On Tue, 28 Aug 2018, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
> On Tue, Aug 28, 2018 at 07:05:46PM +0100, Chris Wilson wrote:
>> Quoting Lucas De Marchi (2018-08-28 18:41:46)
>> > Document it like a real struct for ease of copy and paste, remove
>> > comment of C99 compatibility and document that in some cases the first 2
>> 
>> I do recall that we couldn't use either C99 or class due to userspace
>
> you can't actually use a c++ compiler.
>
> For C it works with any of -std=c99, gnu99, c11, gnu11, c17, gnu17.
> Tested with both gcc and clang. I've never heard of class being a
> reserved keyword and section 6.4.5 of said standard doesn't list it
> neither.
>
> Here the struct definition is in a *comment*... i.e. the user will copy
> and paste somewhere else and probably change __u16 to uint16_t in
> userspace. If he's building with g++, he can name the field to something
> else.
>
> If it was something we were defining in this header than I would agree
> with you... to retain compatibility with c++, not c99.

I always thought the comment told you not to use designated initializers
(introduced in C99), which would both impose a minimum C version
requirement as well as bring .class = foo in the code, which requires
more than just renaming the field with C++.

BR,
Jani.

>
> Lucas De Marchi
>
>> compatibility... The essence is that we need a reminder that we can't
>> assume the relaxed nature of kcc here.
>> -Chris
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
diff mbox series

Patch

diff --git a/include/drm/i915_pciids.h b/include/drm/i915_pciids.h
index 754ce4b10129..0c2cc43f916c 100644
--- a/include/drm/i915_pciids.h
+++ b/include/drm/i915_pciids.h
@@ -26,14 +26,16 @@ 
 #define _I915_PCIIDS_H
 
 /*
- * A pci_device_id struct {
- *	__u32 vendor, device;
- *      __u32 subvendor, subdevice;
- *	__u32 class, class_mask;
- *	kernel_ulong_t driver_data;
+ * These macros can be used with a struct declared like this:
+ *
+ * struct pci_device_id {
+ * 	__u32 vendor, device;
+ * 	__u32 subvendor, subdevice;
+ * 	__u32 class, class_mask;
+ * 	kernel_ulong_t driver_data;
  * };
- * Don't use C99 here because "class" is reserved and we want to
- * give userspace flexibility.
+ *
+ * First two fields may be __u16 if PCI_DEVICE_ANY is not used
  */
 #define INTEL_VGA_DEVICE(id, info) {		\
 	0x8086,	id,				\