Re: [PATCH V3 11/11] MIPS: Add multiplatform BMIPS target

[Arnd Bergmann]

On Monday 24 November 2014 13:07:46 Kevin Cernekee wrote:
> On Mon, Nov 24, 2014 at 6:00 AM, Arnd Bergmann <arnd@xxxxxxxx> wrote:
> > It mihgt be good to split this into multiple patches
> 
> OK.  For V4 I could submit the arch/mips/bmips code changes in one
> patch, and then add the DTS files in the next patch?

Sounds good.

> >> --- /dev/null
> >> +++ b/arch/mips/bmips/Kconfig
> >> @@ -0,0 +1,50 @@
> >> +choice
> >> +     prompt "Built-in device tree"
> >> +     help
> >> +       Legacy bootloaders do not pass a DTB pointer to the kernel, so
> >> +       if a "wrapper" is not being used, the kernel will need to include
> >> +       a device tree that matches the target board.
> >> +
> >> +       The builtin DTB will only be used if the firmware does not supply
> >> +       a valid DTB.
> >> +
> >> +config DT_NONE
> >> +     bool "None"
> >> +
> >> +config DT_BCM93384WVG
> >> +     bool "BCM93384WVG Zephyr CPU"
> >> +     select BUILTIN_DTB
> >> +
> >> +config DT_BCM93384WVG_VIPER
> >> +     bool "BCM93384WVG Viper CPU (EXPERIMENTAL)"
> >> +     select BUILTIN_DTB
> >
> > Why do you have to pick just one? I liked the suggestion of just
> > appending the dtb to the zImage as we do on ARM, so you can build
> > a combined kernel and then run it on multiple machines.
> 
> I'm not sure this exists yet on MIPS.  Perhaps we can discuss it as a
> possible future enhancement to the general arch/mips code?

I don't remember who said it, but someone commented on v2 that they
had a patch for this.
 
> One possible complication is that we're mostly using ELF images on
> MIPS, not zImage.  Is it easy to calculate the end address of an ELF
> file?
> 
> If there is no PT_LOAD segment for the blob, will the bootloader even
> copy it into memory?

Don't know

> >> +unsigned int get_c0_compare_int(void)
> >> +{
> >> +     return CP0_LEGACY_COMPARE_IRQ;
> >> +}
> >
> > Could this just become a function pointer instead of a global
> > variable?
> 
> >> +void __init prom_free_prom_memory(void)
> >> +{
> >> +}
> >
> > This in turn could live outside of the platform codefor anything
> > that is "multiplatform".
> 
> >> +#define R4600_V1_INDEX_ICACHEOP_WAR  0
> >> +#define R4600_V1_HIT_CACHEOP_WAR     0
> >> +#define R4600_V2_HIT_CACHEOP_WAR     0
> >> ...
> >
> > As mentioned before, it seems like you are simply defining these all to zero,
> > like most other platforms do too. Why not add this file as
> > arch/mips/include/asm/mach-generic/war.h and delete all identical copies?
> 
> Likewise - currently every existing MIPS machine type implements it this way.
> 
> Perhaps a future patch series can generalize the way these definitions
> are handled on MIPS?

I'd like to hear what Ralf thinks about it, it's really his decision.
What I was pointing out here are things that are still in the way of
a real "multiplatform" implementation. None of these are really hard
to do, but I don't know where you are heading with MIPS.

I think in case of the last one, it's really just a matter of moving the
file, you could delete the other copies later.


> >> +OF_DECLARE_2(irqchip, mips_cpu_intc, "mti,cpu-interrupt-controller",
> >> +          mips_cpu_irq_of_init);
> >
> > OF_DECLARE_2 really wasn't meant to be used directly. Can you move this
> > code into drivers/irqchip and make it use IRQCHIP_DECLARE()?
> 
> Perhaps arch/mips/kernel/irq_cpu.c could be moved under
> drivers/irqchip in a future commit?  We'll probably have to change the
> way arch/mips/ralink invokes it too.

Possibly, but that seems unrelated. Moving this file is required
in order to use IRQCHIP_DECLARE, which is defined in
drivers/irqchip/irqchip.h.

> >> +void __init prom_init(void)
> >> +{
> >> +     register_bmips_smp_ops();
> >> +}
> >
> > This seems to be the wrong place for calling this function.
> 
> Hmm, looking around the tree:
> 
> cavium-octeon, ip27, loongson, netlogic xlr, paravirt, and sibyte call
> register_smp_ops() from prom_init().
> 
> netlogic xlp calls it from plat_mem_setup().

I guess prom_init() doesn't do what I expected it to on MIPS then.
On powerpc, this is where we call into Open Firmware. No point changing
it now I guess.

For a real multiplatform kernel, the function would likely have to
be matched up with a root DT compatible property like your fixups.

> >> +     /* Intended to somewhat resemble ARM; see Documentation/arm/Booting */
> >> +     if (fw_arg0 == 0 && fw_arg1 == 0xffffffff)
> >> +             dtb = phys_to_virt(fw_arg2);
> >> +     else if (__dtb_start != __dtb_end)
> >> +             dtb = (void *)__dtb_start;
> >> +     else
> >> +             panic("no dtb found");
> >> +
> >> +     __dt_setup_arch(dtb);
> >> +     strlcpy(arcs_cmdline, boot_command_line, COMMAND_LINE_SIZE);
> >
> > Is this intended to become a generic MIPS boot interface? Better
> > document it in Documentation/mips/
> 
> Not sure yet.  It's currently limited to BCM3384.
> 
> For V4 I can add an "Entry point for arch/mips/bmips" or even an
> "Entry point for arch/mips" section to
> Documentation/devicetree/booting-without-of.txt.  Any preferences?

If the goal is being able to have a multiplatform kernel
that can cover more than just BMIPS, I think this would have
to be documented as the only way for MIPS multiplatform.

If that isn't possible, most of my other comments here are
moot, but then you shouldn't call it "multiplatform" but just
"generic BMIPS" or something like that.

> >> +void __init device_tree_init(void)
> >> +{
> >> +     struct device_node *np;
> >> +
> >> +     unflatten_and_copy_device_tree();
> >> +
> >> +     /* Disable SMP boot unless both CPUs are listed in DT and !disabled */
> >> +     np = of_find_node_by_name(NULL, "cpus");
> >> +     if (np && of_get_available_child_count(np) <= 1)
> >> +             bmips_smp_enabled = 0;
> >> +     of_node_put(np);
> >> +}
> >
> > this looks also like it should be in platform
> > independent code
> 
> Hmm.  Do we want to add some logic to kernel/smp.c or kernel/cpu.c
> that says something like: if there is a "cpus" node, clear
> cpu_possible_mask and then set the corresponding bit for each cpu@N
> listed in the DT?
> 
> Unfortunately, if enabled by default, this could impact a lot of other
> systems too.
> 
> Another option is to make a generic helper function that individual
> platforms can call to say "populate my CPU bitmap from DT," maybe
> something like of_set_possible_cpus().

I would do it for all "multiplatform" implementations, i.e. BMIPS
for now and then any other platform that is ready to convert to that.

The existing platforms already provide their own device_tree_init()
function already.

> >> +const char *get_system_type(void)
> >> +{
> >> +     return "BMIPS multiplatform kernel";
> >> +}
> >
> > You could set the string from bmips_quirk_list and make this generic
> > as well.
> 
> The way it currently looks is:
> 
> # cat /proc/cpuinfo
> system type        : BMIPS multiplatform kernel
> machine            : Broadcom BCM97425SVMB
> processor        : 0
> cpu model        : Broadcom BMIPS5000 V1.1  FPU V0.1
> BogoMIPS        : 866.30
> wait instruction    : yes
> microsecond timers    : yes
> tlb_entries        : 64
> extra interrupt vector    : yes
> hardware watchpoint    : no
> isa            : mips1 mips2 mips32r1
> [...]
> 
> So this indicates:
> 
> system type: the kernel build
> machine: the board name from DT
> cpu model: the detected CPU type
> 
> I would rather not tie quirks to the system identity, as sane
> legacy-free systems (such as bcm3384-zephyr) probably shouldn't have
> quirks.

Ok, got it. So in a multiplatform kernel, this would be another function
pointer I guess, but it would always return this string for any BMIPS
system.

> >> diff --git a/arch/mips/include/asm/mach-bmips/dma-coherence.h b/arch/mips/include/asm/mach-bmips/dma-coherence.h
> >> new file mode 100644
> >> index 000000000000..5481a4d1bbbf
> >> --- /dev/null
> >> +++ b/arch/mips/include/asm/mach-bmips/dma-coherence.h
> >> @@ -0,0 +1,45 @@
> >> +#ifndef __ASM_MACH_BMIPS_DMA_COHERENCE_H
> >> +#define __ASM_MACH_BMIPS_DMA_COHERENCE_H
> >> +
> >> +struct device;
> >> +
> >> +extern dma_addr_t plat_map_dma_mem(struct device *dev, void *addr, size_t size);
> >> +extern dma_addr_t plat_map_dma_mem_page(struct device *dev, struct page *page);
> >> +extern unsigned long plat_dma_addr_to_phys(struct device *dev,
> >> +     dma_addr_t dma_addr);
> >> +extern void plat_unmap_dma_mem(struct device *dev, dma_addr_t dma_addr,
> >> +     size_t size, enum dma_data_direction direction);
> >
> > I think you could just add these to
> > arch/mips/include/asm/mach-generic/dma-coherence.h and get rid of the
> > header file, after adding a Kconfig symbol.
> 
> Some platforms mix and match inline definitions versus externs in this file.
> 
> Maybe Ralf can comment on whether we should move to an "all or nothing" model?

To clarify where I was getting to here: In a generic multiplatform kernel,
you would probably want to always look at the dma-ranges properties here,
at least if there are one or more platforms built into the kernel that
don't just have a flat mapping that the current mach-generic header
provides.

> > Would either
> > this value or the 0xfffe0000 from
> > arch/mips/include/asm/mach-generic/spaces.h would everywhere?
> 
> I would have to defer to the other MIPS guys...
> 
> The other question is what "everywhere" means in this context.  Maybe
> we want to just use 0xff00_0000 for a multiplatform kernel, and any
> system that is too obscure or too "weird" would continue using its own
> build.

I mean every system that today uses the plain mach-generic/spaces.h.

	Arnd
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/