LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: NeilBrown <neil@brown.name>
To: James Hogan <jhogan@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>,
	Paul Burton <paul.burton@mips.com>,
	linux-mips@linux-mips.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2] MIPS: c-r4k: fix data corruption related to cache coherence.
Date: Mon, 07 May 2018 07:40:49 +1000	[thread overview]
Message-ID: <87lgcwcvj2.fsf@notabene.neil.brown.name> (raw)
In-Reply-To: <87vacdlf8d.fsf@notabene.neil.brown.name>

[-- Attachment #1: Type: text/plain, Size: 3574 bytes --]


Hi James,
 this hasn't appear in linux-next yet, or in any branch
 of
   git://git.kernel.org/pub/scm/linux/kernel/git/jhogan/mips.git

 Should I expect it to?

Thanks,
NeilBrown

On Fri, Apr 27 2018, NeilBrown wrote:

> When DMA will be performed to a MIPS32 1004K CPS, the
> L1-cache for the range needs to be flushed and invalidated
> first.
> The code currently takes one of two approaches.
> 1/ If the range is less than the size of the dcache, then
>    HIT type requests flush/invalidate cache lines for the
>    particular addresses.  HIT-type requests a globalised
>    by the CPS so this is safe on SMP.
>
> 2/ If the range is larger than the size of dcache, then
>    INDEX type requests flush/invalidate the whole cache.
>    INDEX type requests affect the local cache only. CPS
>    does not propagate them in any way.  So this invalidation
>    is not safe on SMP CPS systems.
>
> Data corruption due to '2' can quite easily be demonstrated by
> repeatedly "echo 3 > /proc/sys/vm/drop_caches" and then sha1sum
> a file that is several times the size of available memory.
> Dropping caches means that large contiguous extents (large than
> dcache) are more likely.
>
> This was not a problem before Linux-4.8 because option 2 was
> never used if CONFIG_MIPS_CPS was defined.  The commit
> which removed that apparently didn't appreciate the full
> consequence of the change.
>
> We could, in theory, globalize the INDEX based flush by sending an IPI
> to other cores.  These cache invalidation routines can be called with
> interrupts disabled and synchronous IPI require interrupts to be
> enabled.  Asynchronous IPI may not trigger writeback soon enough.
> So we cannot use IPI in practice.
>
> We can already test is IPI would be needed for an INDEX operation
> with r4k_op_needs_ipi(R4K_INDEX).  If this is True then we mustn't try
> the INDEX approach as we cannot use IPI.  If this is False (e.g. when
> there is only one core and hence one L1 cache) then it is safe to
> use the INDEX approach without IPI.
>
> This patch avoids options 2 if r4k_op_needs_ipi(R4K_INDEX), and so
> eliminates the corruption.
>
> Fixes: c00ab4896ed5 ("MIPS: Remove cpu_has_safe_index_cacheops")
> Cc: stable@vger.kernel.org # v4.8+
> Signed-off-by: NeilBrown <neil@brown.name>
> ---
>  arch/mips/mm/c-r4k.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/arch/mips/mm/c-r4k.c b/arch/mips/mm/c-r4k.c
> index 6f534b209971..e12dfa48b478 100644
> --- a/arch/mips/mm/c-r4k.c
> +++ b/arch/mips/mm/c-r4k.c
> @@ -851,9 +851,12 @@ static void r4k_dma_cache_wback_inv(unsigned long addr, unsigned long size)
>  	/*
>  	 * Either no secondary cache or the available caches don't have the
>  	 * subset property so we have to flush the primary caches
> -	 * explicitly
> +	 * explicitly.
> +	 * If we would need IPI to perform an INDEX-type operation, then
> +	 * we have to use the HIT-type alternative as IPI cannot be used
> +	 * here due to interrupts possibly being disabled.
>  	 */
> -	if (size >= dcache_size) {
> +	if (!r4k_op_needs_ipi(R4K_INDEX) && size >= dcache_size) {
>  		r4k_blast_dcache();
>  	} else {
>  		R4600_HIT_CACHEOP_WAR_IMPL;
> @@ -890,7 +893,7 @@ static void r4k_dma_cache_inv(unsigned long addr, unsigned long size)
>  		return;
>  	}
>  
> -	if (size >= dcache_size) {
> +	if (!r4k_op_needs_ipi(R4K_INDEX) && size >= dcache_size) {
>  		r4k_blast_dcache();
>  	} else {
>  		R4600_HIT_CACHEOP_WAR_IMPL;
> -- 
> 2.14.0.rc0.dirty

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]

  reply	other threads:[~2018-05-06 21:41 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-04-25  4:08 [PATCH] " NeilBrown
2018-04-25 21:46 ` James Hogan
2018-04-25 22:00   ` NeilBrown
2018-04-25 22:08     ` James Hogan
2018-04-26 23:28       ` [PATCH v2] " NeilBrown
2018-05-06 21:40         ` NeilBrown [this message]
2018-05-07 20:16           ` James Hogan
2018-05-08  1:22             ` NeilBrown
2018-05-11 21:56               ` James Hogan

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=87lgcwcvj2.fsf@notabene.neil.brown.name \
    --to=neil@brown.name \
    --cc=jhogan@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mips@linux-mips.org \
    --cc=paul.burton@mips.com \
    --cc=ralf@linux-mips.org \
    --subject='Re: [PATCH v2] MIPS: c-r4k: fix data corruption related to cache coherence.' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).