From mboxrd@z Thu Jan  1 00:00:00 1970
From: rabin@rab.in (Rabin Vincent)
Date: Wed, 2 Nov 2016 14:27:14 +0100
Subject: flush_dcache_page() in ARM vs ARM64
Message-ID: <20161102132714.GA1326@lnxartpec.se.axis.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

ARMv7-A and ARMv8-A are, as far as I can see, identical in which cache
behaviours they support.  The data cache has to behave as PIPT while for
the instruction cache, PIPT, VIPT, and ASIC-tagged VIVT behaviours are
supported.  See section B3.11 of the ARMv7-A ARM and section D4.9 of the
ARMv8-A ARM.

Both ARMv7-A with Multiprocessing Extensions and ARMv8-A broadcast cache
maintenance operations to other cores.  See B2.2.5 of the ARMv7-A ARM
and D7.2.57 of the ARMv8-A ARM.

Both arch/arm/ (for ARMv6+) and arch/arm64/ define PG_arch_1 to be
PG_dcache_clean and use it to postpone flushing from flush_dcache_page()
to set_pte_at().  See arch/{arm,arm64}/mm/flush.c.

However, arch/arm64/'s flush_dcache_page() is implemented like this:

	void flush_dcache_page(struct page *page)
	{
		if (test_bit(PG_dcache_clean, &page->flags))
			clear_bit(PG_dcache_clean, &page->flags);
	}

while arch/arm/ has this:

	void flush_dcache_page(struct page *page)
	{
		struct address_space *mapping;
	
		/*
		 * The zero page is never written to, so never has any dirty
		 * cache lines, and therefore never needs to be flushed.
		 */
		if (page == ZERO_PAGE(0))
			return;
	
		mapping = page_mapping(page);
	
		if (!cache_ops_need_broadcast() &&
		    mapping && !page_mapcount(page))
			clear_bit(PG_dcache_clean, &page->flags);
		else {
			__flush_dcache_page(mapping, page);
			if (mapping && cache_is_vivt())
				__flush_dcache_aliases(mapping, page);
			else if (mapping)
				__flush_icache_all();
			set_bit(PG_dcache_clean, &page->flags);
		}
	}

Why does arch/arm/ flush the data cache area in flush_dcache_page() for
the (!mapping || page_mapcount(page)) case even on ARMv7+ME, while
arch/arm64/ doesn't for ARMv8?

Why does arch/arm/ invalidate the instruction cache in
flush_dcache_page() for the (mapping && page_count(page)) case even for
ARMv7+ME, while arch/arm64/ doesn't for ARMv8?

What would break with the following patch?

diff --git a/arch/arm/mm/flush.c b/arch/arm/mm/flush.c
index 3cced84..f1e6190 100644
--- a/arch/arm/mm/flush.c
+++ b/arch/arm/mm/flush.c
@@ -327,6 +327,12 @@ void flush_dcache_page(struct page *page)
 	if (page == ZERO_PAGE(0))
 		return;
 
+	if (!cache_ops_need_broadcast() && cache_is_vipt_nonaliasing()) {
+		if (test_bit(PG_dcache_clean, &page->flags))
+			clear_bit(PG_dcache_clean, &page->flags);
+		return;
+	}
+
 	mapping = page_mapping(page);
 
 	if (!cache_ops_need_broadcast() &&