[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAL_JsqL2g69Duhkqd=f09BStoAKeakJz8Wgz3is37YU0+fc80A@mail.gmail.com>
Date: Mon, 29 Oct 2018 14:59:51 -0500
From: Rob Herring <robh+dt@...nel.org>
To: Florian Fainelli <f.fainelli@...il.com>,
Ard Biesheuvel <ard.biesheuvel@...aro.org>
Cc: "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
rppt@...ux.ibm.com, Catalin Marinas <catalin.marinas@....com>,
Will Deacon <will.deacon@....com>,
Frank Rowand <frowand.list@...il.com>,
Andrew Morton <akpm@...ux-foundation.org>,
Marc Zyngier <marc.zyngier@....com>,
Russell King <rmk+kernel@...linux.org.uk>,
aryabinin@...tuozzo.com, Andrey Konovalov <andreyknvl@...gle.com>,
Masahiro Yamada <yamada.masahiro@...ionext.com>,
Robin Murphy <robin.murphy@....com>,
Laura Abbott <labbott@...hat.com>,
Stefan Agner <stefan@...er.ch>,
Johannes Weiner <hannes@...xchg.org>, ghackmann@...roid.com,
Kristina Martsenko <kristina.martsenko@....com>,
chandan.vn@...sung.com,
"moderated list:ARM/FREESCALE IMX / MXC ARM ARCHITECTURE"
<linux-arm-kernel@...ts.infradead.org>, devicetree@...r.kernel.org,
Russell King <linux@...linux.org.uk>
Subject: Re: [PATCH 1/2 v5] arm64: Get rid of __early_init_dt_declare_initrd()
+Ard who last touched this.
On Mon, Oct 29, 2018 at 2:23 PM Florian Fainelli <f.fainelli@...il.com> wrote:
>
> ARM64 is the only architecture that re-defines
> __early_init_dt_declare_initrd() in order for that function to populate
> initrd_start/initrd_end with physical addresses instead of virtual
> addresses. Instead of having an override, just get rid of that
> implementation and perform the virtual to physical conversion of these
> addresses in arm64_memblock_init() where relevant.
>
> Signed-off-by: Florian Fainelli <f.fainelli@...il.com>
> Signed-off-by: Mike Rapoport <rppt@...ux.ibm.com>
> ---
> arch/arm64/include/asm/memory.h | 8 -------
> arch/arm64/mm/init.c | 42 +++++++++++++++++++++------------
> 2 files changed, 27 insertions(+), 23 deletions(-)
>
> diff --git a/arch/arm64/include/asm/memory.h b/arch/arm64/include/asm/memory.h
> index b96442960aea..dc3ca21ba240 100644
> --- a/arch/arm64/include/asm/memory.h
> +++ b/arch/arm64/include/asm/memory.h
> @@ -168,14 +168,6 @@
> #define IOREMAP_MAX_ORDER (PMD_SHIFT)
> #endif
>
> -#ifdef CONFIG_BLK_DEV_INITRD
> -#define __early_init_dt_declare_initrd(__start, __end) \
> - do { \
> - initrd_start = (__start); \
> - initrd_end = (__end); \
> - } while (0)
> -#endif
> -
> #ifndef __ASSEMBLY__
>
> #include <linux/bitops.h>
> diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c
> index 3cf87341859f..292570b08f85 100644
> --- a/arch/arm64/mm/init.c
> +++ b/arch/arm64/mm/init.c
> @@ -62,6 +62,8 @@
> s64 memstart_addr __ro_after_init = -1;
> phys_addr_t arm64_dma_phys_limit __ro_after_init;
>
> +static phys_addr_t phys_initrd_start, phys_initrd_end;
> +
> #ifdef CONFIG_BLK_DEV_INITRD
> static int __init early_initrd(char *p)
> {
> @@ -72,8 +74,8 @@ static int __init early_initrd(char *p)
> if (*endp == ',') {
> size = memparse(endp + 1, NULL);
>
> - initrd_start = start;
> - initrd_end = start + size;
> + phys_initrd_start = start;
> + phys_initrd_end = start + size;
> }
> return 0;
> }
> @@ -364,6 +366,7 @@ static void __init fdt_enforce_memory_region(void)
> void __init arm64_memblock_init(void)
> {
> const s64 linear_region_size = -(s64)PAGE_OFFSET;
> + u64 __maybe_unused base, size;
>
> /* Handle linux,usable-memory-range property */
> fdt_enforce_memory_region();
> @@ -408,14 +411,25 @@ void __init arm64_memblock_init(void)
> memblock_add(__pa_symbol(_text), (u64)(_end - _text));
> }
>
> - if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && initrd_start) {
> + if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) &&
> + (initrd_start || phys_initrd_start)) {
I've tried to explain already that this is broken. The problem is
__early_init_dt_declare_initrd using __va() which happens before this
function is called. __va() uses PHYS_OFFSET which in turn is defined
as memstart_addr. However, memstart_addr may be changed just above
this hunk, so the earlier conversion to a VA may not be valid at this
point. This is explained if you read Ard's commit that added all this
mess.
You could fix this by converting back to a PA before adjusting
memstart_addr, but that's 2 wrongs making a right and fragile. The
better solution is the other proposal making the DT code set
phys_initrd_* (whatever the ARM code calls them).
> /*
> * Add back the memory we just removed if it results in the
> * initrd to become inaccessible via the linear mapping.
> * Otherwise, this is a no-op
> */
> - u64 base = initrd_start & PAGE_MASK;
> - u64 size = PAGE_ALIGN(initrd_end) - base;
> + if (phys_initrd_start) {
> + /* Command line specified the initrd location */
> + initrd_start = __phys_to_virt(phys_initrd_start);
> + initrd_end = __phys_to_virt(phys_initrd_end);
> + } else if (initrd_start) {
> + /* FDT specified the initrd location */
> + phys_initrd_start = __pa(initrd_start);
> + phys_initrd_end = __pa(initrd_end);
Kind of inconsistent to mix __phys_to_virt and __pa flavors.
Rob
Powered by blists - more mailing lists