linux-kernel - Re: [PATCH 2/3] ARM: l2c: update prefetch bits in L2X0_AUX

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date:   Mon, 10 Aug 2020 13:34:33 +0100
From:   Guillaume Tucker <guillaume.tucker@...labora.com>
To:     Russell King - ARM Linux admin <linux@...linux.org.uk>
Cc:     Kukjin Kim <kgene@...nel.org>,
        Krzysztof Kozlowski <krzk@...nel.org>,
        Rob Herring <robh+dt@...nel.org>, kernel@...labora.com,
        devicetree@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        linux-samsung-soc@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 2/3] ARM: l2c: update prefetch bits in L2X0_AUX_CTRL using
 DT value

On 29/07/2020 17:22, Guillaume Tucker wrote:
> On 29/07/2020 15:18, Russell King - ARM Linux admin wrote:
>> On Wed, Jul 29, 2020 at 02:47:32PM +0100, Guillaume Tucker wrote:
>>> The L310_PREFETCH_CTRL register bits 28 and 29 to enable data and
>>> instruction prefetch respectively can also be accessed via the
>>> L2X0_AUX_CTRL register.  They appear to be actually wired together in
>>> hardware between the registers.  Changing them in the prefetch
>>> register only will get undone when restoring the aux control register
>>> later on.  For this reason, set these bits in both registers during
>>> initialisation according to the DT attributes.
>>
>> How will that happen?
>>
>> We write the auxiliary control register before the prefetch control
>> register, so the prefetch control register will take precedence.  See
>> l2c310_configure() - l2c_configure() writes the auxiliary control
>> register, and the function writes the prefetch control register later.
> 
> What I'm seeing is that outer_cache.configure() gets called, at
> least on exynos4412-odroidx2.  See l2c_enable():
> 
> 	if (outer_cache.configure)
> 		outer_cache.configure(&l2x0_saved_regs);
> 	else
> 		l2x0_data->configure(base);
> 
> Then instead of l2c310_configure(), exynos_l2_configure() gets
> called and writes prefetch_ctrl right before aux_ctrl.  Should
> exynos_l2_configure() be changed to swap the register writes?
> 
> 
>> I think the real issue is that Exynos has been modifying the prefetch
>> settings via its machine .aux_mask / .aux_val configuration, and the
>> opposite is actually true: the prefetch control register values will
>> overwrite the attempt to modify the auxiliary control values set through
>> the machine .aux_mask/.aux_val.
> 
> Yes with l2c310_configure() but not with exynos_l2_configure().
> 
> To be clear, this is what I've found to be happening, if you
> switch to using the device tree prefetch attributes and clear
> the bits in the default l2c_aux_val (see PATCH 3/3):
> 
> 1. l2x0_of_init() first gets called with the default aux_val
> 
> 2. l2c310_of_parse() sets the bits in l2x0_saved_regs.prefetch_ctrl
>    but not in aux_val (unless you apply this patch 2/3)
> 
> 3. l2c_enable() calls exynos_l2_configure() which writes
>    prefetch_ctrl and then aux_ctrl - thus setting the prefetch bits
>    and then clearing them just after
> 
> 4. l2c310_enable() reads back aux_ctrl and prefetch, both of which
>    now have the bits cleared (the pr_info() message about prefetch
>    enabled gets skipped)
> 
> 
> That's why I thought it would be safer to set the prefetch bits
> in both registers so it should work regardless if the
> initialisation sequence.  Also, if we want these bits to be
> changed, we should clear them in the aux_mask value to not get
> another error message about register corruption - so I'm doing
> that too.

I've kept this patch as-is in the v2 because I wasn't sure
whether you wanted the issue to be addressed differently in the
end.  I just made it a bit clearer in the commit message that
it's fixing an issue when using the DT prefetch properties.
Please let me know if you want me to rework this in any way.

Thanks,
Guillaume

>>> Fixes: ec3bd0e68a67 ("ARM: 8391/1: l2c: add options to overwrite prefetching behavior")
>>> Signed-off-by: Guillaume Tucker <guillaume.tucker@...labora.com>
>>> ---
>>>  arch/arm/mm/cache-l2x0.c | 16 ++++++++++++----
>>>  1 file changed, 12 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/arch/arm/mm/cache-l2x0.c b/arch/arm/mm/cache-l2x0.c
>>> index 12c26eb88afb..43d91bfd2360 100644
>>> --- a/arch/arm/mm/cache-l2x0.c
>>> +++ b/arch/arm/mm/cache-l2x0.c
>>> @@ -1249,20 +1249,28 @@ static void __init l2c310_of_parse(const struct device_node *np,
>>>  
>>>  	ret = of_property_read_u32(np, "prefetch-data", &val);
>>>  	if (ret == 0) {
>>> -		if (val)
>>> +		if (val) {
>>>  			prefetch |= L310_PREFETCH_CTRL_DATA_PREFETCH;
>>> -		else
>>> +			*aux_val |= L310_PREFETCH_CTRL_DATA_PREFETCH;
>>> +		} else {
>>>  			prefetch &= ~L310_PREFETCH_CTRL_DATA_PREFETCH;
>>> +			*aux_val &= ~L310_PREFETCH_CTRL_DATA_PREFETCH;
>>> +		}
>>> +		*aux_mask &= ~L310_PREFETCH_CTRL_DATA_PREFETCH;
>>>  	} else if (ret != -EINVAL) {
>>>  		pr_err("L2C-310 OF prefetch-data property value is missing\n");
>>>  	}
>>>  
>>>  	ret = of_property_read_u32(np, "prefetch-instr", &val);
>>>  	if (ret == 0) {
>>> -		if (val)
>>> +		if (val) {
>>>  			prefetch |= L310_PREFETCH_CTRL_INSTR_PREFETCH;
>>> -		else
>>> +			*aux_val |= L310_PREFETCH_CTRL_INSTR_PREFETCH;
>>> +		} else {
>>>  			prefetch &= ~L310_PREFETCH_CTRL_INSTR_PREFETCH;
>>> +			*aux_val &= ~L310_PREFETCH_CTRL_INSTR_PREFETCH;
>>> +		}
>>> +		*aux_mask &= ~L310_PREFETCH_CTRL_INSTR_PREFETCH;
>>>  	} else if (ret != -EINVAL) {
>>>  		pr_err("L2C-310 OF prefetch-instr property value is missing\n");
>>>  	}
>>> -- 
>>> 2.20.1
>>>
>>>
>>
>