Thread (5 messages) 5 messages, 2 authors, 2015-10-30

[PATCH v2 4/4] arm64: document the cache policy behavior

From: Will Deacon <hidden>
Date: 2015-10-30 12:08:25
Also in: stable

On Thu, Oct 29, 2015 at 04:20:42PM -0700, Alex Van Brunt wrote:
Add a comment that clairfies how the kernel should behave given the cache
clarifies
quoted hunk ↗ jump to hunk
policy reported by the CPU.

Signed-off-by: Alex Van Brunt <redacted>
Cc: <redacted>
---
 arch/arm64/kernel/cpuinfo.c | 8 ++++++++
 1 file changed, 8 insertions(+)
diff --git a/arch/arm64/kernel/cpuinfo.c b/arch/arm64/kernel/cpuinfo.c
index ae04ac1..bf7e5e2 100644
--- a/arch/arm64/kernel/cpuinfo.c
+++ b/arch/arm64/kernel/cpuinfo.c
@@ -49,6 +49,14 @@ static void cpuinfo_detect_icache_policy(struct cpuinfo_arm64 *info)
 	unsigned int cpu = smp_processor_id();
 	u32 l1ip = CTR_L1IP(info->reg_ctr);
 
+	/*
+	 * The ARM architecture defines PIPT, VIPT and AIVIVT in terms of the
-the
+	 * the observable behavior not how the CPU implements the policy.
s/not/rather than/
+	 * Specifically, the policies differentiate the correct way to
+	 * invalidate the cache. The definitions say that the only
+	 * architecturally guaranteed way to invalidate a VIPT or AIVIVT
+	 * instruction cache is to invalidate the entire instruction cache.
+	 */
 	if (l1ip != ICACHE_POLICY_PIPT)
 		set_bit(ICACHEF_ALIASING, &__icache_flags);
 	if (l1ip == ICACHE_POLICY_AIVIVT)
With the minor cosmetic changes:

  Acked-by: Will Deacon [off-list ref]

although I don't see how we can really apply this given that we're not
planning to revert the cache geometry stuff. Maybe you could spin a
separate series just addressing the aliasing I-cache detection, then we
can build on top of that?

Will
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help