linux-kernel - Re: [PATCH V1 09/15] spmi: pmic-arb: check apid enabled before calling the handler

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] [thread-next>] [day] [month] [year] [list]

Message-ID: <09e72f239b5cbf615ab828a32f34f9b5@codeaurora.org>
Date:   Wed, 14 Jun 2017 21:08:53 +0530
From:   kgunda@...eaurora.org
To:     Stephen Boyd <sboyd@...eaurora.org>
Cc:     Abhijeet Dharmapurikar <adharmap@...eaurora.org>,
        David Collins <collinsd@...eaurora.org>,
        Subbaraman Narayanamurthy <subbaram@...eaurora.org>,
        Christophe JAILLET <christophe.jaillet@...adoo.fr>,
        linux-kernel@...r.kernel.org, linux-arm-msm@...r.kernel.org,
        adharmap@...cinc.com, aghayal@....qualcomm.com,
        linux-arm-msm-owner@...r.kernel.org
Subject: Re: [PATCH V1 09/15] spmi: pmic-arb: check apid enabled before
 calling the handler

On 2017-06-01 02:09, Stephen Boyd wrote:
> On 05/30, Kiran Gunda wrote:
>> From: Abhijeet Dharmapurikar <adharmap@...eaurora.org>
>> 
>> The driver currently invokes the apid handler (periph_handler())
> 
> You mean periph_interrupt()?
> 
Yes.
>> once it sees that the summary status bit for that apid is set.
>> 
>> However the hardware is designed to set that bit even if the apid
>> interrupts are disabled. The driver should check whether the apid
>> is indeed enabled before calling the apid handler.
> 
> Really? Wow that is awful. Or is this because ACC_ENABLE bit is
> always set now and never cleared?
> 
Yes. It is awful. It is not because of the ACC_ENABLE bit is set.
>> 
>> Signed-off-by: Abhijeet Dharmapurikar <adharmap@...eaurora.org>
>> Signed-off-by: Kiran Gunda <kgunda@...eaurora.org>
>> ---
>>  drivers/spmi/spmi-pmic-arb.c | 10 +++++++---
>>  1 file changed, 7 insertions(+), 3 deletions(-)
>> 
>> diff --git a/drivers/spmi/spmi-pmic-arb.c 
>> b/drivers/spmi/spmi-pmic-arb.c
>> index ad34491..f8638fa 100644
>> --- a/drivers/spmi/spmi-pmic-arb.c
>> +++ b/drivers/spmi/spmi-pmic-arb.c
>> @@ -536,8 +536,8 @@ static void pmic_arb_chained_irq(struct irq_desc 
>> *desc)
>>  	void __iomem *intr = pa->intr;
>>  	int first = pa->min_apid >> 5;
>>  	int last = pa->max_apid >> 5;
>> -	u32 status;
>> -	int i, id;
>> +	u32 status, enable;
>> +	int i, id, apid;
>> 
>>  	chained_irq_enter(chip, desc);
>> 
>> @@ -547,7 +547,11 @@ static void pmic_arb_chained_irq(struct irq_desc 
>> *desc)
>>  		while (status) {
>>  			id = ffs(status) - 1;
>>  			status &= ~BIT(id);
>> -			periph_interrupt(pa, id + i * 32);
>> +			apid = id + i * 32;
>> +			enable = readl_relaxed(intr +
>> +					pa->ver_ops->acc_enable(apid));
> 
> Do we need to read the hardware to figure this out? After earlier
> patches in this series we would never clear the
> SPMI_PIC_ACC_ENABLE_BIT after one of the irqs in a peripheral is
> unmasked for the first time (which looks to be fixing a bug in
> the existing driver BTW). So in practice, this should almost
> always be true.
> 
yes. We have removed clearing the SPMI_PIC_ACC_ENABLE_BIT from the 
irq_mask,
because if we disable this BIT it disables all the peripheral IRQs, 
which we don't want.

Once the peripheral fires the interrupt the summary status bit for that 
peripheral
is set even though the SPMI_PIC_ACC_ENABLE_BIT is not enabled. That's 
why we have to
read this register to not service the interrupt that is not 
requested/enabled yet.
This SPMI_PIC_ACC_ENABLE_BIT is enabled during the irq_unmask which is 
called from request_irq.

> In the one case that it isn't true, we'll be handling some other
> irq for another peripheral and then hardware will tell us there's
> an interrupt for a peripheral that doesn't have any interrupts
> unmasked. We would call periph_interrupt() and then that
> shouldn't see any interrupts in the status register for that
> APID. So we do some more work, but nothing happens still. Did I
> miss something? What is this fixing?

Yes. As you said this fixes the issue of calling the periph_interrupt
for some other irq that is not yet requested and enabled yet.