[PATCH] PPC: Don't sleep in flush_dcache_icache_page()

[PATCH] PPC: Don't sleep in flush_dcache_icache_page()

[Roland Dreier]

flush_dcache_icache_page() will be called on an instruction page
fault.  We can't sleep in the fault handler, so use kmap_atomic()
instead of just kmap() for the Book-E case.

Signed-off-by: Roland Dreier <rolandd@cisco.com>

diff --git a/arch/ppc/mm/init.c b/arch/ppc/mm/init.c
--- a/arch/ppc/mm/init.c
+++ b/arch/ppc/mm/init.c
@@ -560,8 +560,16 @@ void flush_dcache_page(struct page *page
 void flush_dcache_icache_page(struct page *page)
 {
 #ifdef CONFIG_BOOKE
-	__flush_dcache_icache(kmap(page));
-	kunmap(page);
+	unsigned long flags;
+	void *start;
+
+	local_irq_save(flags);
+
+	start = kmap_atomic(page, KM_PPC_SYNC_PAGE);
+	__flush_dcache_icache(start);
+	kunmap_atomic(start, KM_PPC_SYNC_PAGE);
+
+	local_irq_restore(flags);
 #elif CONFIG_8xx
 	/* On 8xx there is no need to kmap since highmem is not supported */
 	__flush_dcache_icache(page_address(page)); 

Re: [PATCH] PPC: Don't sleep in flush_dcache_icache_page()

[Marcelo Tosatti]

Hi Roland,

On Tue, Aug 16, 2005 at 01:56:49PM -0700, Roland Dreier wrote:
> flush_dcache_icache_page() will be called on an instruction page
> fault.  We can't sleep in the fault handler, so use kmap_atomic()
> instead of just kmap() for the Book-E case.
> 
> Signed-off-by: Roland Dreier <rolandd@cisco.com>

Why do you need to disable interrupts during the kmap_atomic/flush_dcache_icache
operation ? 

I fail to see how an interrupt could have any reference to the data
being dealt with here (the user page).

> diff --git a/arch/ppc/mm/init.c b/arch/ppc/mm/init.c
> --- a/arch/ppc/mm/init.c
> +++ b/arch/ppc/mm/init.c
> @@ -560,8 +560,16 @@ void flush_dcache_page(struct page *page
>  void flush_dcache_icache_page(struct page *page)
>  {
>  #ifdef CONFIG_BOOKE
> -	__flush_dcache_icache(kmap(page));
> -	kunmap(page);
> +	unsigned long flags;
> +	void *start;
> +
> +	local_irq_save(flags);
> +
> +	start = kmap_atomic(page, KM_PPC_SYNC_PAGE);
> +	__flush_dcache_icache(start);
> +	kunmap_atomic(start, KM_PPC_SYNC_PAGE);
> +
> +	local_irq_restore(flags);
>  #elif CONFIG_8xx
>  	/* On 8xx there is no need to kmap since highmem is not supported */
>  	__flush_dcache_icache(page_address(page)); 

Re: [PATCH] PPC: Don't sleep in flush_dcache_icache_page()

[Matt Porter]

On Thu, Aug 18, 2005 at 02:56:42PM -0300, Marcelo Tosatti wrote:
> 
> Hi Roland,
> 
> On Tue, Aug 16, 2005 at 01:56:49PM -0700, Roland Dreier wrote:
> > flush_dcache_icache_page() will be called on an instruction page
> > fault.  We can't sleep in the fault handler, so use kmap_atomic()
> > instead of just kmap() for the Book-E case.
> > 
> > Signed-off-by: Roland Dreier <rolandd@cisco.com>
> 
> Why do you need to disable interrupts during the kmap_atomic/flush_dcache_icache
> operation ? 
> 
> I fail to see how an interrupt could have any reference to the data
> being dealt with here (the user page).

We just took care of this offline.  The original patch is sharing
a kmap slot with another kmap_atomic user I put in before...the
sync page user.  If an interrupt came in causing the DMA API to
be used, we would have a problem.

The clean solution was to use a different kmap slot.

-Matt