Re: [PATCH] arm64: kaslr: Use standard early random function

From: Mark Rutland
Date: Fri Aug 07 2020 - 06:17:31 EST


Hi Guenter,

On Thu, Aug 06, 2020 at 05:49:04PM -0700, Guenter Roeck wrote:
> Commit 585524081ecd ("random: random.h should include archrandom.h, not
> the other way around") tries to fix a problem with recursive inclusion
> of linux/random.h and arch/archrandom.h for arm64. Unfortunately, this
> results in the following compile error if ARCH_RANDOM is disabled.
>
> arch/arm64/kernel/kaslr.c: In function 'kaslr_early_init':
> arch/arm64/kernel/kaslr.c:128:6: error: implicit declaration of function
> '__early_cpu_has_rndr'; did you mean '__early_pfn_to_nid'?
> [-Werror=implicit-function-declaration]
> if (__early_cpu_has_rndr()) {
> ^~~~~~~~~~~~~~~~~~~~
> __early_pfn_to_nid
> arch/arm64/kernel/kaslr.c:131:7: error: implicit declaration of function
> '__arm64_rndr' [-Werror=implicit-function-declaration]
> if (__arm64_rndr(&raw))
> ^~~~~~~~~~~~
>
> Problem is that arch/archrandom.h is only included from linux/random.h if
> ARCH_RANDOM is enabled. If not, __arm64_rndr() and __early_cpu_has_rndr()
> are undeclared, causing the problem.
>
> Use arch_get_random_seed_long_early() instead of arm64 specific functions
> to solve the problem. As a side effect of this change, the code no longer
> bypasses ARCH_RANDOM, which I consider desirable (after all, ARCH_RANDOM
> was disabled for a reason).

There's no bypass of ARCH_RANDOM; the bits KASLR depends on are empty
stubs when ARCH_RANDOM is not selected. I added the common early
functions after this code was written.

> Reported-by: Qian Cai <cai@xxxxxx>
> Fixes: 585524081ecd ("random: random.h should include archrandom.h, not the other way around")

This is where things broke; there was no need to change kaslr.c's
include of <asm/archrandom.h>, since kaslr.c only depends on the RNDR
bits defined there./

> Fixes: 2e8e1ea88cbc ("arm64: Use v8.5-RNG entropy for KASLR seed")

I don't think this tag is necessary; this commit built and worked fine,
and there wasn't any ARCH_RANDOM bypass to speak of.

> Cc: Qian Cai <cai@xxxxxx>
> Cc: Mark Brown <broonie@xxxxxxxxxx>
> Signed-off-by: Guenter Roeck <linux@xxxxxxxxxxxx>

This patch itself looks fine, but as above I think the commit message is
misleading w.r.t. bypassing ARCH_RANDOM, and the second fixes tag isn't
necessary.

With those bits gone:

Reviewed-by: Mark Rutland <mark.rutland@xxxxxxx>

Mark.

> ---
> arch/arm64/kernel/kaslr.c | 12 +++++-------
> 1 file changed, 5 insertions(+), 7 deletions(-)
>
> diff --git a/arch/arm64/kernel/kaslr.c b/arch/arm64/kernel/kaslr.c
> index 9ded4237e1c1..b181e0544b79 100644
> --- a/arch/arm64/kernel/kaslr.c
> +++ b/arch/arm64/kernel/kaslr.c
> @@ -84,6 +84,7 @@ u64 __init kaslr_early_init(u64 dt_phys)
> void *fdt;
> u64 seed, offset, mask, module_range;
> const u8 *cmdline, *str;
> + unsigned long raw;
> int size;
>
> /*
> @@ -122,15 +123,12 @@ u64 __init kaslr_early_init(u64 dt_phys)
> }
>
> /*
> - * Mix in any entropy obtainable architecturally, open coded
> - * since this runs extremely early.
> + * Mix in any entropy obtainable architecturally if enabled
> + * and supported.
> */
> - if (__early_cpu_has_rndr()) {
> - unsigned long raw;
>
> - if (__arm64_rndr(&raw))
> - seed ^= raw;
> - }
> + if (arch_get_random_seed_long_early(&raw))
> + seed ^= raw;
>
> if (!seed) {
> kaslr_status = KASLR_DISABLED_NO_SEED;
> --
> 2.17.1
>