[PATCH] Do not require PCI code for RayeR SPI driver

[Carl-Daniel Hailfinger]

This patch depends on
Re: [flashrom] [PATCH] List affected features for missing libraries

CONFIG_RAYER_SPI only needs raw hardware access, but not libpci.

Signed-off-by: Carl-Daniel Hailfinger

diff -ur flashrom-libpci_libusb_report_requirements/hwaccess.h flashrom-rayer_do_not_require_libpci/hwaccess.h
--- flashrom-libpci_libusb_report_requirements/hwaccess.h	2016-02-17 00:26:07.221475948 +0100
+++ flashrom-rayer_do_not_require_libpci/hwaccess.h	2016-02-17 01:09:53.270497512 +0100
@@ -196,7 +196,7 @@
 #define le_to_cpu32 cpu_to_le32
 #define le_to_cpu64 cpu_to_le64

-#if NEED_PCI == 1
+#if NEED_RAW_ACCESS == 1
 #if IS_X86

/* sys/io.h provides iopl(2) and x86 I/O port access functions (inb, outb etc).
@@ -376,6 +376,6 @@
 #error Unknown architecture, please check if it supports PCI port IO.

#endif /* IS_* */
-#endif /* NEED_PCI == 1 */
+#endif /* NEED_RAW_ACCESS == 1 */

#endif /* !__HWACCESS_H__ */
diff -ur flashrom-libpci_libusb_report_requirements/Makefile flashrom-rayer_do_not_require_libpci/Makefile
--- flashrom-libpci_libusb_report_requirements/Makefile	2016-02-17 00:26:37.497626067 +0100
+++ flashrom-rayer_do_not_require_libpci/Makefile	2016-02-17 00:52:46.457405821 +0100
@@ -599,8 +599,7 @@
 ifeq ($(CONFIG_RAYER_SPI), yes)
 FEATURE_CFLAGS += -D'CONFIG_RAYER_SPI=1'
 PROGRAMMER_OBJS += rayer_spi.o
-# Actually, NEED_LIBPCI is wrong. NEED_IOPORT_ACCESS would be more correct.
-NEED_LIBPCI += CONFIG_RAYER_SPI
+NEED_RAW_ACCESS += CONFIG_RAYER_SPI
 endif

ifeq ($(CONFIG_PONY_SPI), yes)
@@ -782,10 +781,14 @@

ifneq ($(NEED_LIBPCI), )
 CHECK_LIBPCI = yes
+# This is a dirty hack, but it saves us from checking all PCI drivers and all platforms manually.
+# libpci may need raw memory, MSR or PCI port I/O on some platforms.
+# Individual drivers might have the same needs as well.
+NEED_RAW_ACCESS += $(NEED_LIBPCI)
 FEATURE_CFLAGS += -D'NEED_PCI=1'
 FEATURE_CFLAGS += $(call debug_shell,grep -q "OLD_PCI_GET_DEV := yes" .libdeps && printf "%s" "-D'OLD_PCI_GET_DEV=1'")

-PROGRAMMER_OBJS += pcidev.o physmap.o hwaccess.o
+PROGRAMMER_OBJS += pcidev.o
 ifeq ($(TARGET_OS), NetBSD)
 # The libpci we want is called libpciutils on NetBSD and needs NetBSD libpci.
 PCILIBS += -lpciutils -lpci
@@ -807,6 +810,11 @@
 endif
 endif

+ifneq ($(NEED_RAW_ACCESS), )
+FEATURE_CFLAGS += -D'NEED_RAW_ACCESS=1'
+PROGRAMMER_OBJS += physmap.o hwaccess.o
+endif
+
 ifneq ($(NEED_LIBUSB0), )
 CHECK_LIBUSB0 = yes
 FEATURE_CFLAGS += -D'NEED_LIBUSB0=1'

[Stefan Tauner]

On Wed, 17 Feb 2016 01:10:41 +0100
Carl-Daniel Hailfinger  wrote:
> This patch depends on
> Re: [flashrom] [PATCH] List affected features for missing libraries
>
> CONFIG_RAYER_SPI only needs raw hardware access, but not libpci.
>
> Signed-off-by: Carl-Daniel Hailfinger
>
> diff -ur flashrom-libpci_libusb_report_requirements/hwaccess.h flashrom-rayer_do_not_require_libpci/hwaccess.h
> --- flashrom-libpci_libusb_report_requirements/hwaccess.h	2016-02-17 00:26:07.221475948 +0100
> +++ flashrom-rayer_do_not_require_libpci/hwaccess.h	2016-02-17 01:09:53.270497512 +0100
> @@ -196,7 +196,7 @@
>  #define le_to_cpu32 cpu_to_le32
>  #define le_to_cpu64 cpu_to_le64
>
> -#if NEED_PCI == 1
> +#if NEED_RAW_ACCESS == 1
>  #if IS_X86
>
> /* sys/io.h provides iopl(2) and x86 I/O port access functions (inb, outb etc).
> @@ -376,6 +376,6 @@
>  #error Unknown architecture, please check if it supports PCI port IO.
>
> #endif /* IS_* */
> -#endif /* NEED_PCI == 1 */
> +#endif /* NEED_RAW_ACCESS == 1 */
>
> #endif /* !__HWACCESS_H__ */
> diff -ur flashrom-libpci_libusb_report_requirements/Makefile flashrom-rayer_do_not_require_libpci/Makefile
> --- flashrom-libpci_libusb_report_requirements/Makefile	2016-02-17 00:26:37.497626067 +0100
> +++ flashrom-rayer_do_not_require_libpci/Makefile	2016-02-17 00:52:46.457405821 +0100
> @@ -599,8 +599,7 @@
>  ifeq ($(CONFIG_RAYER_SPI), yes)
>  FEATURE_CFLAGS += -D'CONFIG_RAYER_SPI=1'
>  PROGRAMMER_OBJS += rayer_spi.o
> -# Actually, NEED_LIBPCI is wrong. NEED_IOPORT_ACCESS would be more correct.

What's wrong with NEED_IOPORT_ACCESS? I'd prefer that (not only because
I have read this wannabe FIXME a few dozen times by now... :)
> -NEED_LIBPCI += CONFIG_RAYER_SPI
> +NEED_RAW_ACCESS += CONFIG_RAYER_SPI
>  endif
>
> ifeq ($(CONFIG_PONY_SPI), yes)
> @@ -782,10 +781,14 @@
>
> ifneq ($(NEED_LIBPCI), )
>  CHECK_LIBPCI = yes
> +# This is a dirty hack, but it saves us from checking all PCI drivers and all platforms manually.
> +# libpci may need raw memory, MSR or PCI port I/O on some platforms.
> +# Individual drivers might have the same needs as well.
> +NEED_RAW_ACCESS += $(NEED_LIBPCI)
>  FEATURE_CFLAGS += -D'NEED_PCI=1'
>  FEATURE_CFLAGS += $(call debug_shell,grep -q "OLD_PCI_GET_DEV := yes" .libdeps && printf "%s" "-D'OLD_PCI_GET_DEV=1'")
>
> -PROGRAMMER_OBJS += pcidev.o physmap.o hwaccess.o
> +PROGRAMMER_OBJS += pcidev.o
>  ifeq ($(TARGET_OS), NetBSD)
>  # The libpci we want is called libpciutils on NetBSD and needs NetBSD libpci.
>  PCILIBS += -lpciutils -lpci
> @@ -807,6 +810,11 @@
>  endif
>  endif
>
> +ifneq ($(NEED_RAW_ACCESS), )
> +FEATURE_CFLAGS += -D'NEED_RAW_ACCESS=1'
> +PROGRAMMER_OBJS += physmap.o hwaccess.o
> +endif
>
> ifneq ($(NEED_LIBUSB0), )
>  CHECK_LIBUSB0 = yes
>  FEATURE_CFLAGS += -D'NEED_LIBUSB0=1'

Apparently this somehow breaks some includes as shown by the buildbot
(as discussed on IRC)...

[Carl-Daniel Hailfinger]

On 20.02.2016 01:24, Stefan Tauner wrote:
> On Wed, 17 Feb 2016 01:10:41 +0100
> Carl-Daniel Hailfinger  wrote:
> > --- flashrom-libpci_libusb_report_requirements/Makefile	2016-02-17 00:26:37.497626067 +0100
> > +++ flashrom-rayer_do_not_require_libpci/Makefile	2016-02-17 00:52:46.457405821 +0100
> > @@ -599,8 +599,7 @@
> >  ifeq ($(CONFIG_RAYER_SPI), yes)
> >  FEATURE_CFLAGS += -D'CONFIG_RAYER_SPI=1'
> >  PROGRAMMER_OBJS += rayer_spi.o
> > -# Actually, NEED_LIBPCI is wrong. NEED_IOPORT_ACCESS would be more correct.
>
> What's wrong with NEED_IOPORT_ACCESS? I'd prefer that (not only because
> I have read this wannabe FIXME a few dozen times by now... :)

I didn't want to add NEED_IOPORT_ACCESS, NEED_MEM_ACCESS and
NEED_MSR_ACCESS.
The sane way was to combine all three into NEED_RAW_ACCESS.
> > -NEED_LIBPCI += CONFIG_RAYER_SPI
> > +NEED_RAW_ACCESS += CONFIG_RAYER_SPI
> >  endif
> >
> > ifeq ($(CONFIG_PONY_SPI), yes)
> > @@ -807,6 +810,11 @@
> >  endif
> >  endif
> >
> > +ifneq ($(NEED_RAW_ACCESS), )
> > +FEATURE_CFLAGS += -D'NEED_RAW_ACCESS=1'
> > +PROGRAMMER_OBJS += physmap.o hwaccess.o
> > +endif
> >
> > ifneq ($(NEED_LIBUSB0), )
> >  CHECK_LIBUSB0 = yes
> >  FEATURE_CFLAGS += -D'NEED_LIBUSB0=1'
>
> Apparently this somehow breaks some includes as shown by the buildbot
> (as discussed on IRC)...

Thanks for the review!

New version, fixes OpenBSD, NetBSD, FreeBSD, OS X.
Should compile fine on everything except Android, which was broken since
r1898 (sys/io.h was suddenly included for all Linux targets instead of
just for glibc, and Android doesn't have it).

CONFIG_RAYER_SPI only needs raw hardware access, but not libpci.

Signed-off-by: Carl-Daniel Hailfinger

Index: flashrom-rayer_do_not_require_libpci/Makefile
===================================================================
--- flashrom-rayer_do_not_require_libpci/Makefile	(Revision 1931)
+++ flashrom-rayer_do_not_require_libpci/Makefile	(Arbeitskopie)
@@ -599,8 +599,7 @@
 ifeq ($(CONFIG_RAYER_SPI), yes)
 FEATURE_CFLAGS += -D'CONFIG_RAYER_SPI=1'
 PROGRAMMER_OBJS += rayer_spi.o
-# Actually, NEED_LIBPCI is wrong. NEED_IOPORT_ACCESS would be more correct.
-NEED_LIBPCI += CONFIG_RAYER_SPI
+NEED_RAW_ACCESS += CONFIG_RAYER_SPI
 endif

ifeq ($(CONFIG_PONY_SPI), yes)
@@ -782,19 +781,30 @@

ifneq ($(NEED_LIBPCI), )
 CHECK_LIBPCI = yes
+# This is a dirty hack, but it saves us from checking all PCI drivers and all platforms manually.
+# libpci may need raw memory, MSR or PCI port I/O on some platforms.
+# Individual drivers might have the same needs as well.
+NEED_RAW_ACCESS += $(NEED_LIBPCI)
 FEATURE_CFLAGS += -D'NEED_PCI=1'
 FEATURE_CFLAGS += $(call debug_shell,grep -q "OLD_PCI_GET_DEV := yes" .libdeps && printf "%s" "-D'OLD_PCI_GET_DEV=1'")

-PROGRAMMER_OBJS += pcidev.o physmap.o hwaccess.o
+PROGRAMMER_OBJS += pcidev.o
 ifeq ($(TARGET_OS), NetBSD)
 # The libpci we want is called libpciutils on NetBSD and needs NetBSD libpci.
 PCILIBS += -lpciutils -lpci
+else
+PCILIBS += -lpci
+endif
+endif
+
+ifneq ($(NEED_RAW_ACCESS), )
+FEATURE_CFLAGS += -D'NEED_RAW_ACCESS=1'
+PROGRAMMER_OBJS += physmap.o hwaccess.o
+
+ifeq ($(TARGET_OS), NetBSD)
 # For (i386|x86_64)_iopl(2).
 PCILIBS += -l$(shell uname -p)
 else
-
-PCILIBS += -lpci
-
 ifeq ($(TARGET_OS), OpenBSD)
 # For (i386|amd64)_iopl(2).
 PCILIBS += -l$(shell uname -m)
@@ -805,6 +815,7 @@
 endif
 endif
 endif
+
 endif

ifneq ($(NEED_LIBUSB0), )
Index: flashrom-rayer_do_not_require_libpci/hwaccess.h
===================================================================
--- flashrom-rayer_do_not_require_libpci/hwaccess.h	(Revision 1931)
+++ flashrom-rayer_do_not_require_libpci/hwaccess.h	(Arbeitskopie)
@@ -24,6 +24,8 @@
 #ifndef __HWACCESS_H__
 #define __HWACCESS_H__ 1

+#include <sys/types.h>
+
 #include "platform.h"

#if NEED_PCI == 1
@@ -196,7 +198,7 @@
 #define le_to_cpu32 cpu_to_le32
 #define le_to_cpu64 cpu_to_le64

-#if NEED_PCI == 1
+#if NEED_RAW_ACCESS == 1
 #if IS_X86

/* sys/io.h provides iopl(2) and x86 I/O port access functions (inb, outb etc).
@@ -376,6 +378,6 @@
 #error Unknown architecture, please check if it supports PCI port IO.

#endif /* IS_* */
-#endif /* NEED_PCI == 1 */
+#endif /* NEED_RAW_ACCESS == 1 */

#endif /* !__HWACCESS_H__ */

[Carl-Daniel Hailfinger]

On 20.02.2016 02:36, Carl-Daniel Hailfinger wrote:
> On 20.02.2016 01:24, Stefan Tauner wrote:
> > On Wed, 17 Feb 2016 01:10:41 +0100
> > Carl-Daniel Hailfinger  wrote:
> > > --- flashrom-libpci_libusb_report_requirements/Makefile	2016-02-17 00:26:37.497626067 +0100
> > > +++ flashrom-rayer_do_not_require_libpci/Makefile	2016-02-17 00:52:46.457405821 +0100
> > > @@ -599,8 +599,7 @@
> > >  ifeq ($(CONFIG_RAYER_SPI), yes)
> > >  FEATURE_CFLAGS += -D'CONFIG_RAYER_SPI=1'
> > >  PROGRAMMER_OBJS += rayer_spi.o
> > > -# Actually, NEED_LIBPCI is wrong. NEED_IOPORT_ACCESS would be more correct.
> >
> > What's wrong with NEED_IOPORT_ACCESS? I'd prefer that (not only because
> > I have read this wannabe FIXME a few dozen times by now... :)
>
> I didn't want to add NEED_IOPORT_ACCESS, NEED_MEM_ACCESS and
> NEED_MSR_ACCESS.
> The sane way was to combine all three into NEED_RAW_ACCESS.
> > > -NEED_LIBPCI += CONFIG_RAYER_SPI
> > > +NEED_RAW_ACCESS += CONFIG_RAYER_SPI
> > >  endif
> > >
> > > ifeq ($(CONFIG_PONY_SPI), yes)
> > > @@ -807,6 +810,11 @@
> > >  endif
> > >  endif
> > >
> > > +ifneq ($(NEED_RAW_ACCESS), )
> > > +FEATURE_CFLAGS += -D'NEED_RAW_ACCESS=1'
> > > +PROGRAMMER_OBJS += physmap.o hwaccess.o
> > > +endif
> > >
> > > ifneq ($(NEED_LIBUSB0), )
> > >  CHECK_LIBUSB0 = yes
> > >  FEATURE_CFLAGS += -D'NEED_LIBUSB0=1'
> >
> > Apparently this somehow breaks some includes as shown by the buildbot
> > (as discussed on IRC)...
>
> Thanks for the review!
>
> New version, fixes OpenBSD, NetBSD, FreeBSD, OS X.
> Should compile fine on everything except Android, which was broken since
> r1898 (sys/io.h was suddenly included for all Linux targets instead of
> just for glibc, and Android doesn't have it).
>
> CONFIG_RAYER_SPI only needs raw hardware access, but not libpci.
>
> Signed-off-by: Carl-Daniel Hailfinger

New version, fixes Android by disabling CONFIG_RAYER_SPI. Apparently
Android has iopl(), but not inb/outb.

CONFIG_RAYER_SPI only needs raw hardware access, but not libpci.
Disable CONFIG_RAYER_SPI on Android because of missing inb/outb.
Introduce NEED_RAW_ACCESS for raw memory access, PCI port I/O
and MSR access.

Signed-off-by: Carl-Daniel Hailfinger

Index: flashrom-rayer_do_not_require_libpci/Makefile
===================================================================
--- flashrom-rayer_do_not_require_libpci/Makefile	(Revision 1931)
+++ flashrom-rayer_do_not_require_libpci/Makefile	(Arbeitskopie)
@@ -349,6 +349,15 @@
 endif
 endif

+ifeq ($(TARGET_OS), Android)
+# Android on x86 (currently) does not provide raw PCI port I/O operations
+ifeq ($(CONFIG_RAYER_SPI), yes)
+UNSUPPORTED_FEATURES += CONFIG_RAYER_SPI=yes
+else
+override CONFIG_RAYER_SPI = no
+endif
+endif
+
 ###############################################################################
 # General architecture-specific settings.
 # Like above for the OS, below we verify user-supplied options depending on the target architecture.
@@ -599,8 +608,7 @@
 ifeq ($(CONFIG_RAYER_SPI), yes)
 FEATURE_CFLAGS += -D'CONFIG_RAYER_SPI=1'
 PROGRAMMER_OBJS += rayer_spi.o
-# Actually, NEED_LIBPCI is wrong. NEED_IOPORT_ACCESS would be more correct.
-NEED_LIBPCI += CONFIG_RAYER_SPI
+NEED_RAW_ACCESS += CONFIG_RAYER_SPI
 endif

ifeq ($(CONFIG_PONY_SPI), yes)
@@ -782,19 +790,31 @@

ifneq ($(NEED_LIBPCI), )
 CHECK_LIBPCI = yes
+# This is a dirty hack, but it saves us from checking all PCI drivers and all platforms manually.
+# libpci may need raw memory, MSR or PCI port I/O on some platforms.
+# Individual drivers might have the same needs as well.
+NEED_RAW_ACCESS += $(NEED_LIBPCI)
 FEATURE_CFLAGS += -D'NEED_PCI=1'
 FEATURE_CFLAGS += $(call debug_shell,grep -q "OLD_PCI_GET_DEV := yes" .libdeps && printf "%s" "-D'OLD_PCI_GET_DEV=1'")

-PROGRAMMER_OBJS += pcidev.o physmap.o hwaccess.o
+PROGRAMMER_OBJS += pcidev.o
 ifeq ($(TARGET_OS), NetBSD)
 # The libpci we want is called libpciutils on NetBSD and needs NetBSD libpci.
 PCILIBS += -lpciutils -lpci
+else
+PCILIBS += -lpci
+endif
+endif
+
+ifneq ($(NEED_RAW_ACCESS), )
+# Raw memory, MSR or PCI port I/O access.
+FEATURE_CFLAGS += -D'NEED_RAW_ACCESS=1'
+PROGRAMMER_OBJS += physmap.o hwaccess.o
+
+ifeq ($(TARGET_OS), NetBSD)
 # For (i386|x86_64)_iopl(2).
 PCILIBS += -l$(shell uname -p)
 else
-
-PCILIBS += -lpci
-
 ifeq ($(TARGET_OS), OpenBSD)
 # For (i386|amd64)_iopl(2).
 PCILIBS += -l$(shell uname -m)
@@ -805,6 +825,7 @@
 endif
 endif
 endif
+
 endif

ifneq ($(NEED_LIBUSB0), )
Index: flashrom-rayer_do_not_require_libpci/hwaccess.h
===================================================================
--- flashrom-rayer_do_not_require_libpci/hwaccess.h	(Revision 1931)
+++ flashrom-rayer_do_not_require_libpci/hwaccess.h	(Arbeitskopie)
@@ -24,6 +24,8 @@
 #ifndef __HWACCESS_H__
 #define __HWACCESS_H__ 1

+#include <sys/types.h>
+
 #include "platform.h"

#if NEED_PCI == 1
@@ -196,7 +198,7 @@
 #define le_to_cpu32 cpu_to_le32
 #define le_to_cpu64 cpu_to_le64

-#if NEED_PCI == 1
+#if NEED_RAW_ACCESS == 1
 #if IS_X86

/* sys/io.h provides iopl(2) and x86 I/O port access functions (inb, outb etc).
@@ -376,6 +378,6 @@
 #error Unknown architecture, please check if it supports PCI port IO.

#endif /* IS_* */
-#endif /* NEED_PCI == 1 */
+#endif /* NEED_RAW_ACCESS == 1 */

#endif /* !__HWACCESS_H__ */

[Carl-Daniel Hailfinger]

On 20.02.2016 21:46, Carl-Daniel Hailfinger wrote:
> On 20.02.2016 02:36, Carl-Daniel Hailfinger wrote:
> > On 20.02.2016 01:24, Stefan Tauner wrote:
> > > On Wed, 17 Feb 2016 01:10:41 +0100
> > > Carl-Daniel Hailfinger  wrote:
> > > > --- flashrom-libpci_libusb_report_requirements/Makefile	2016-02-17 00:26:37.497626067 +0100
> > > > +++ flashrom-rayer_do_not_require_libpci/Makefile	2016-02-17 00:52:46.457405821 +0100
> > > > @@ -599,8 +599,7 @@
> > > >  ifeq ($(CONFIG_RAYER_SPI), yes)
> > > >  FEATURE_CFLAGS += -D'CONFIG_RAYER_SPI=1'
> > > >  PROGRAMMER_OBJS += rayer_spi.o
> > > > -# Actually, NEED_LIBPCI is wrong. NEED_IOPORT_ACCESS would be more correct.
> > >
> > > What's wrong with NEED_IOPORT_ACCESS? I'd prefer that (not only because
> > > I have read this wannabe FIXME a few dozen times by now... :)
> >
> > I didn't want to add NEED_IOPORT_ACCESS, NEED_MEM_ACCESS and
> > NEED_MSR_ACCESS.
> > The sane way was to combine all three into NEED_RAW_ACCESS.
> > > > -NEED_LIBPCI += CONFIG_RAYER_SPI
> > > > +NEED_RAW_ACCESS += CONFIG_RAYER_SPI
> > > >  endif
> > > >
> > > > ifeq ($(CONFIG_PONY_SPI), yes)
> > > > @@ -807,6 +810,11 @@
> > > >  endif
> > > >  endif
> > > >
> > > > +ifneq ($(NEED_RAW_ACCESS), )
> > > > +FEATURE_CFLAGS += -D'NEED_RAW_ACCESS=1'
> > > > +PROGRAMMER_OBJS += physmap.o hwaccess.o
> > > > +endif
> > > >
> > > > ifneq ($(NEED_LIBUSB0), )
> > > >  CHECK_LIBUSB0 = yes
> > > >  FEATURE_CFLAGS += -D'NEED_LIBUSB0=1'
> > >
> > > Apparently this somehow breaks some includes as shown by the buildbot
> > > (as discussed on IRC)...
> >
> > Thanks for the review!
> >
> > New version, fixes OpenBSD, NetBSD, FreeBSD, OS X.
> > Should compile fine on everything except Android, which was broken since
> > r1898 (sys/io.h was suddenly included for all Linux targets instead of
> > just for glibc, and Android doesn't have it).
> >
> > CONFIG_RAYER_SPI only needs raw hardware access, but not libpci.
> >
> > Signed-off-by: Carl-Daniel Hailfinger
>
> New version, fixes Android by disabling CONFIG_RAYER_SPI. Apparently
> Android has iopl(), but not inb/outb.

New version, include sys/types.h on FreeBSD specifically instead of for
all OS.

CONFIG_RAYER_SPI only needs raw hardware access, but not libpci.
Disable CONFIG_RAYER_SPI on Android because of missing inb/outb.
Introduce NEED_RAW_ACCESS for raw memory access, PCI port I/O
and MSR access.

Signed-off-by: Carl-Daniel Hailfinger

Index: flashrom-rayer_do_not_require_libpci/Makefile
===================================================================
--- flashrom-rayer_do_not_require_libpci/Makefile	(Revision 1931)
+++ flashrom-rayer_do_not_require_libpci/Makefile	(Arbeitskopie)
@@ -349,6 +349,15 @@
 endif
 endif

+ifeq ($(TARGET_OS), Android)
+# Android on x86 (currently) does not provide raw PCI port I/O operations
+ifeq ($(CONFIG_RAYER_SPI), yes)
+UNSUPPORTED_FEATURES += CONFIG_RAYER_SPI=yes
+else
+override CONFIG_RAYER_SPI = no
+endif
+endif
+
 ###############################################################################
 # General architecture-specific settings.
 # Like above for the OS, below we verify user-supplied options depending on the target architecture.
@@ -599,8 +608,7 @@
 ifeq ($(CONFIG_RAYER_SPI), yes)
 FEATURE_CFLAGS += -D'CONFIG_RAYER_SPI=1'
 PROGRAMMER_OBJS += rayer_spi.o
-# Actually, NEED_LIBPCI is wrong. NEED_IOPORT_ACCESS would be more correct.
-NEED_LIBPCI += CONFIG_RAYER_SPI
+NEED_RAW_ACCESS += CONFIG_RAYER_SPI
 endif

ifeq ($(CONFIG_PONY_SPI), yes)
@@ -782,19 +790,31 @@

ifneq ($(NEED_LIBPCI), )
 CHECK_LIBPCI = yes
+# This is a dirty hack, but it saves us from checking all PCI drivers and all platforms manually.
+# libpci may need raw memory, MSR or PCI port I/O on some platforms.
+# Individual drivers might have the same needs as well.
+NEED_RAW_ACCESS += $(NEED_LIBPCI)
 FEATURE_CFLAGS += -D'NEED_PCI=1'
 FEATURE_CFLAGS += $(call debug_shell,grep -q "OLD_PCI_GET_DEV := yes" .libdeps && printf "%s" "-D'OLD_PCI_GET_DEV=1'")

-PROGRAMMER_OBJS += pcidev.o physmap.o hwaccess.o
+PROGRAMMER_OBJS += pcidev.o
 ifeq ($(TARGET_OS), NetBSD)
 # The libpci we want is called libpciutils on NetBSD and needs NetBSD libpci.
 PCILIBS += -lpciutils -lpci
+else
+PCILIBS += -lpci
+endif
+endif
+
+ifneq ($(NEED_RAW_ACCESS), )
+# Raw memory, MSR or PCI port I/O access.
+FEATURE_CFLAGS += -D'NEED_RAW_ACCESS=1'
+PROGRAMMER_OBJS += physmap.o hwaccess.o
+
+ifeq ($(TARGET_OS), NetBSD)
 # For (i386|x86_64)_iopl(2).
 PCILIBS += -l$(shell uname -p)
 else
-
-PCILIBS += -lpci
-
 ifeq ($(TARGET_OS), OpenBSD)
 # For (i386|amd64)_iopl(2).
 PCILIBS += -l$(shell uname -m)
@@ -805,6 +825,7 @@
 endif
 endif
 endif
+
 endif

ifneq ($(NEED_LIBUSB0), )
Index: flashrom-rayer_do_not_require_libpci/hwaccess.h
===================================================================
--- flashrom-rayer_do_not_require_libpci/hwaccess.h	(Revision 1931)
+++ flashrom-rayer_do_not_require_libpci/hwaccess.h	(Arbeitskopie)
@@ -24,6 +24,8 @@
 #ifndef __HWACCESS_H__
 #define __HWACCESS_H__ 1

+#include <sys/types.h>
+
 #include "platform.h"

#if NEED_PCI == 1
@@ -196,7 +198,7 @@
 #define le_to_cpu32 cpu_to_le32
 #define le_to_cpu64 cpu_to_le64

-#if NEED_PCI == 1
+#if NEED_RAW_ACCESS == 1
 #if IS_X86

/* sys/io.h provides iopl(2) and x86 I/O port access functions (inb, outb etc).
@@ -376,6 +378,6 @@
 #error Unknown architecture, please check if it supports PCI port IO.

#endif /* IS_* */
-#endif /* NEED_PCI == 1 */
+#endif /* NEED_RAW_ACCESS == 1 */

#endif /* !__HWACCESS_H__ */

[Stefan Tauner]

On Sat, 20 Feb 2016 22:23:06 +0100
Carl-Daniel Hailfinger  wrote:
> New version, include sys/types.h on FreeBSD specifically instead of for
> all OS.

liar! At least it is still in there below :)
> ifneq ($(NEED_LIBUSB0), )
> Index: flashrom-rayer_do_not_require_libpci/hwaccess.h
> ===================================================================
> --- flashrom-rayer_do_not_require_libpci/hwaccess.h	(Revision 1931)
> +++ flashrom-rayer_do_not_require_libpci/hwaccess.h	(Arbeitskopie)
> @@ -24,6 +24,8 @@
>  #ifndef __HWACCESS_H__
>  #define __HWACCESS_H__ 1
>
> +#include <sys/types.h>
>
> #include "platform.h"
>
> #if NEED_PCI == 1
> @@ -196,7 +198,7 @@
>  #define le_to_cpu32 cpu_to_le32
>  #define le_to_cpu64 cpu_to_le64
>
> -#if NEED_PCI == 1
> +#if NEED_RAW_ACCESS == 1
>  #if IS_X86
>
> /* sys/io.h provides iopl(2) and x86 I/O port access functions (inb, outb etc).
> @@ -376,6 +378,6 @@
>  #error Unknown architecture, please check if it supports PCI port IO.
>
> #endif /* IS_* */
> -#endif /* NEED_PCI == 1 */
> +#endif /* NEED_RAW_ACCESS == 1 */
>
> #endif /* !__HWACCESS_H__ */

But with that mentioned fix this is
Acked-by: Stefan Tauner 
Thanks!

[Carl-Daniel Hailfinger]

On 20.02.2016 22:35, Stefan Tauner wrote:
> On Sat, 20 Feb 2016 22:23:06 +0100
> Carl-Daniel Hailfinger  wrote:
> > New version, include sys/types.h on FreeBSD specifically instead of for
> > all OS.
>
> liar! At least it is still in there below :)
> > ifneq ($(NEED_LIBUSB0), )
> > Index: flashrom-rayer_do_not_require_libpci/hwaccess.h
> > ===================================================================
> > --- flashrom-rayer_do_not_require_libpci/hwaccess.h	(Revision 1931)
> > +++ flashrom-rayer_do_not_require_libpci/hwaccess.h	(Arbeitskopie)
> > @@ -24,6 +24,8 @@
> >  #ifndef __HWACCESS_H__
> >  #define __HWACCESS_H__ 1
> >
> > +#include <sys/types.h>
> >
> > #include "platform.h"
> >
> > #if NEED_PCI == 1
> > @@ -196,7 +198,7 @@
> >  #define le_to_cpu32 cpu_to_le32
> >  #define le_to_cpu64 cpu_to_le64
> >
> > -#if NEED_PCI == 1
> > +#if NEED_RAW_ACCESS == 1
> >  #if IS_X86
> >
> > /* sys/io.h provides iopl(2) and x86 I/O port access functions (inb, outb etc).
> > @@ -376,6 +378,6 @@
> >  #error Unknown architecture, please check if it supports PCI port IO.
> >
> > #endif /* IS_* */
> > -#endif /* NEED_PCI == 1 */
> > +#endif /* NEED_RAW_ACCESS == 1 */
> >
> > #endif /* !__HWACCESS_H__ */
>
> But with that mentioned fix this is
> Acked-by: Stefan Tauner 
> Thanks!

Indeed. I sent the wrong version. Correct version follows:

CONFIG_RAYER_SPI only needs raw hardware access, but not libpci.
Disable CONFIG_RAYER_SPI on Android because of missing inb/outb.
Introduce NEED_RAW_ACCESS for raw memory access, PCI port I/O
and MSR access.

Signed-off-by: Carl-Daniel Hailfinger 
Acked-by: Stefan Tauner

Index: flashrom-rayer_do_not_require_libpci/Makefile
===================================================================
--- flashrom-rayer_do_not_require_libpci/Makefile	(Revision 1933)
+++ flashrom-rayer_do_not_require_libpci/Makefile	(Arbeitskopie)
@@ -349,6 +349,15 @@
 endif
 endif

+ifeq ($(TARGET_OS), Android)
+# Android on x86 (currently) does not provide raw PCI port I/O operations
+ifeq ($(CONFIG_RAYER_SPI), yes)
+UNSUPPORTED_FEATURES += CONFIG_RAYER_SPI=yes
+else
+override CONFIG_RAYER_SPI = no
+endif
+endif
+
 ###############################################################################
 # General architecture-specific settings.
 # Like above for the OS, below we verify user-supplied options depending on the target architecture.
@@ -599,8 +608,7 @@
 ifeq ($(CONFIG_RAYER_SPI), yes)
 FEATURE_CFLAGS += -D'CONFIG_RAYER_SPI=1'
 PROGRAMMER_OBJS += rayer_spi.o
-# Actually, NEED_LIBPCI is wrong. NEED_IOPORT_ACCESS would be more correct.
-NEED_LIBPCI += CONFIG_RAYER_SPI
+NEED_RAW_ACCESS += CONFIG_RAYER_SPI
 endif

ifeq ($(CONFIG_PONY_SPI), yes)
@@ -782,19 +790,31 @@

ifneq ($(NEED_LIBPCI), )
 CHECK_LIBPCI = yes
+# This is a dirty hack, but it saves us from checking all PCI drivers and all platforms manually.
+# libpci may need raw memory, MSR or PCI port I/O on some platforms.
+# Individual drivers might have the same needs as well.
+NEED_RAW_ACCESS += $(NEED_LIBPCI)
 FEATURE_CFLAGS += -D'NEED_PCI=1'
 FEATURE_CFLAGS += $(call debug_shell,grep -q "OLD_PCI_GET_DEV := yes" .libdeps && printf "%s" "-D'OLD_PCI_GET_DEV=1'")

-PROGRAMMER_OBJS += pcidev.o physmap.o hwaccess.o
+PROGRAMMER_OBJS += pcidev.o
 ifeq ($(TARGET_OS), NetBSD)
 # The libpci we want is called libpciutils on NetBSD and needs NetBSD libpci.
 PCILIBS += -lpciutils -lpci
+else
+PCILIBS += -lpci
+endif
+endif
+
+ifneq ($(NEED_RAW_ACCESS), )
+# Raw memory, MSR or PCI port I/O access.
+FEATURE_CFLAGS += -D'NEED_RAW_ACCESS=1'
+PROGRAMMER_OBJS += physmap.o hwaccess.o
+
+ifeq ($(TARGET_OS), NetBSD)
 # For (i386|x86_64)_iopl(2).
 PCILIBS += -l$(shell uname -p)
 else
-
-PCILIBS += -lpci
-
 ifeq ($(TARGET_OS), OpenBSD)
 # For (i386|amd64)_iopl(2).
 PCILIBS += -l$(shell uname -m)
@@ -805,6 +825,7 @@
 endif
 endif
 endif
+
 endif

ifneq ($(NEED_LIBUSB0), )
Index: flashrom-rayer_do_not_require_libpci/hwaccess.h
===================================================================
--- flashrom-rayer_do_not_require_libpci/hwaccess.h	(Revision 1933)
+++ flashrom-rayer_do_not_require_libpci/hwaccess.h	(Arbeitskopie)
@@ -196,7 +196,7 @@
 #define le_to_cpu32 cpu_to_le32
 #define le_to_cpu64 cpu_to_le64

-#if NEED_PCI == 1
+#if NEED_RAW_ACCESS == 1
 #if IS_X86

/* sys/io.h provides iopl(2) and x86 I/O port access functions (inb, outb etc).
@@ -230,6 +230,7 @@
    * out[bwl] definitions in machine/cpufunc.h and sys/io.h at least in some
    * versions. Use machine/cpufunc.h only for plain FreeBSD/DragonFlyBSD.
    */
+  #include <sys/types.h>
   #include <machine/cpufunc.h>
   #define OUTB(x, y) do { u_int outb_tmp = (y); outb(outb_tmp, (x)); } while (0)
   #define OUTW(x, y) do { u_int outw_tmp = (y); outw(outw_tmp, (x)); } while (0)
@@ -382,6 +383,6 @@
 #error Unknown architecture, please check if it supports PCI port IO.

#endif /* IS_* */
-#endif /* NEED_PCI == 1 */
+#endif /* NEED_RAW_ACCESS == 1 */

#endif /* !__HWACCESS_H__ */

[Carl-Daniel Hailfinger]

On 20.02.2016 22:42, Carl-Daniel Hailfinger wrote:
> CONFIG_RAYER_SPI only needs raw hardware access, but not libpci.
> Disable CONFIG_RAYER_SPI on Android because of missing inb/outb.
> Introduce NEED_RAW_ACCESS for raw memory access, PCI port I/O
> and MSR access.
>
> Signed-off-by: Carl-Daniel Hailfinger 
> Acked-by: Stefan Tauner

Thanks for the review, committed in r1934.

Regards,
Carl-Daniel