[RESEND] [PATCH v2 1/2] arm: devtree: Set system_rev from DT revision

[Arnd Bergmann]

On Wednesday 06 May 2015 12:37:52 Pali Rohár wrote:
> On Wednesday 06 May 2015 11:31:15 Arnd Bergmann wrote:
> > On Wednesday 06 May 2015 10:49:01 Pali Rohár wrote:
> > > With this patch "revision" DT string entry is used to set global system_rev
> > > variable. DT "revision" is expected to be string with one hexadecimal number.
> > > So "Revision" line in /proc/cpuinfo will be same as "revision" DT value.
> > > 
> > > Signed-off-by: Pali Rohár <pali.rohar at gmail.com>
> > > Acked-by: Pavel Machek <pavel at ucw.cz>
> > 
> > +devicetree mailing list
> > 
> > The property needs to be specified in a binding somewhere.
> > 
> > > @@ -246,5 +247,14 @@ const struct machine_desc * __init setup_machine_fdt(unsigned int dt_phys)
> > >     /* Change machine number to match the mdesc we're using */
> > >     __machine_arch_type = mdesc->nr;
> > >  
> > > +   /* Set system revision from DT */
> > > +   prop = of_get_flat_dt_prop(dt_root, "revision", &size);
> > > +   if (prop && size > 0) {
> > > +           char revision[11];
> > > +           strlcpy(revision, prop, min(size, (int)sizeof(revision)));
> > > +           if (kstrtouint(revision, 16, &system_rev) != 0)
> > > +                   system_rev = 0;
> > > +   }
> > > +
> > >     return mdesc;
> > >  }
> > > 
> > 
> > What is the reason for doing it this early? I think it would be nicer to do
> > it after unflattening the DT.
> > 
> 
> It needs to be done in this code, so "system_rev" variable is set
> properly...

What I mean is which code accesses this variable that early?

> > Also, it seems strange to have a string property and then use kstrtouint
> > to convert it into a number. I think it should either be specified in a DT
> > binding to be a string and then have the kernel not assume that it is a number,
> > or we should define it to be binary.
> > 
> >       Arnd
> 
> Variable "system_rev" is number and it always was. So chaning type will
> break more parts.
> 
> And it is string DT property to be human readable. Some other developers
> suggested for v2 to change it to string (from number).

Both of them would be human readable, you just use something else to
read them ;-)

If we have a string here, we should just change all uses of system_rev
in the kernel accordingly, there are only a few of them:

$ git grep -w system_rev
arch/arm/include/asm/system_info.h:extern unsigned int system_rev;
arch/arm/kernel/atags_parse.c:  system_rev = tag->u.revision.rev;
arch/arm/kernel/setup.c:unsigned int system_rev;
arch/arm/kernel/setup.c:EXPORT_SYMBOL(system_rev);
arch/arm/kernel/setup.c:        seq_printf(m, "Revision\t: %04x\n", system_rev);
arch/arm/mach-clps711x/devices.c:       system_rev = SYSFLG1_VERID(readl(base + SYSFLG1));
arch/arm/mach-clps711x/devices.c:       soc_dev_attr->revision = kasprintf(GFP_KERNEL, "%u", system_rev);
arch/arm/mach-davinci/board-da850-evm.c:        switch (system_rev & 0xF) {
arch/arm/mach-imx/mach-imx27_visstrim_m10.c:    system_rev = 0x27000;
arch/arm/mach-imx/mach-imx27_visstrim_m10.c:    system_rev |= (mo_version << MOTHERBOARD_SHIFT);
arch/arm/mach-imx/mach-imx27_visstrim_m10.c:    system_rev |= (exp_version << EXPBOARD_SHIFT);
arch/arm/mach-imx/mach-imx27_visstrim_m10.c:    mo_version = (system_rev >> MOTHERBOARD_SHIFT) & VERSION_MASK;
arch/arm/mach-ixp4xx/goramo_mlr.c:              system_rev = __raw_readl(flash + CFG_REV);
arch/arm/mach-omap2/board-rx51-peripherals.c:   if ((system_rev >= SYSTEM_REV_S_USES_VAUX3 && system_rev < 0x100) ||
arch/arm/mach-omap2/board-rx51-peripherals.c:       system_rev >= SYSTEM_REV_B_USES_VAUX3) {
arch/arm/mach-orion5x/dns323-setup.c:   if (machine_is_dns323() && system_rev == DNS323_REV_A1)
arch/arm/mach-orion5x/dns323-setup.c:   system_rev = dns323_identify_rev();
arch/arm/mach-orion5x/dns323-setup.c:   pr_info("DNS-323: Identified HW revision %c1\n", 'A' + system_rev);
arch/arm/mach-orion5x/dns323-setup.c:   switch(system_rev) {
arch/arm/mach-orion5x/dns323-setup.c:   switch(system_rev) {
arch/arm/mach-orion5x/dns323-setup.c:   switch(system_rev) {
arch/arm/mach-pxa/cm-x300.c:    if (system_rev < 130)
arch/arm/mach-pxa/cm-x300.c:    if (system_rev < 130) {
arch/arm/mach-pxa/cm-x300.c:    if (system_rev < 130)
arch/arm/mach-pxa/magician.c:           if (system_rev < 3)
arch/arm/mach-pxa/magician.c:           if (system_rev < 3)
arch/arm/mach-pxa/magician.c:           system_rev = board_id & 0x7;
arch/arm/mach-pxa/magician.c:           if (lcd_select && (system_rev < 3))
arch/arm/mach-pxa/raumfeld.c:   if ((system_rev & 0xff) == 2) {
arch/arm/mach-pxa/raumfeld.c:   if ((system_rev & 0xff) > 1) {
arch/arm/mach-pxa/viper.c:              system_rev = (VIPER_BOARD_VERSION(version) << 8) |
arch/arm/mach-pxa/zeus.c:       system_rev = __raw_readw(ZEUS_CPLD_VERSION);
arch/arm/mach-pxa/zeus.c:       pr_info("Zeus CPLD V%dI%d\n", (system_rev & 0xf0) >> 4, (system_rev & 0x0f));
arch/arm/mach-zynq/common.c:    system_rev = zynq_get_revision();
arch/arm/mach-zynq/common.c:    soc_dev_attr->revision = kasprintf(GFP_KERNEL, "0x%x", system_rev);

In fact, half the uses of this actually assign the revision number themselves.
code outside of arch/arm/mach-* and /proc/cpuinfo currently uses the variable.

	Arnd