linux-kernel - Re: [PATCH 1/3] i2c: imx: Fix reset of I2SR

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]

Date:   Fri, 2 Oct 2020 10:01:30 +0200
From:   Christian Eggers <ceggers@...i.de>
To:     Uwe Kleine-König 
        <u.kleine-koenig@...gutronix.de>
CC:     Shawn Guo <shawnguo@...nel.org>,
        Sascha Hauer <s.hauer@...gutronix.de>,
        <linux-kernel@...r.kernel.org>, <stable@...r.kernel.org>,
        Oleksij Rempel <linux@...pel-privat.de>,
        NXP Linux Team <linux-imx@....com>,
        "Pengutronix Kernel Team" <kernel@...gutronix.de>,
        Fabio Estevam <festevam@...il.com>,
        <linux-arm-kernel@...ts.infradead.org>, <linux-i2c@...r.kernel.org>
Subject: Re: [PATCH 1/3] i2c: imx: Fix reset of I2SR_IAL flag

Hi Uwe,

On Friday, 25 September 2020, 10:11:01 CEST, Uwe Kleine-König wrote:
> On Thu, Sep 17, 2020 at 04:13:50PM +0200, Christian Eggers wrote:
> > On Thursday, 17 September 2020, 16:02:35 CEST, Uwe Kleine-König wrote:
> > > On Thu, Sep 17, 2020 at 02:20:27PM +0200, Christian Eggers wrote:
> > > ...
> > > 
> > > >             /* check for arbitration lost */
> > > >             if (temp & I2SR_IAL) {
> > > >             
> > > >                     temp &= ~I2SR_IAL;
> > > > 
> > > > +                   temp |= (i2c_imx->hwdata->i2sr_clr_opcode & I2SR_IAL);
> > > > 
> > > >                     imx_i2c_write_reg(temp, i2c_imx, IMX_I2C_I2SR);
> > > >                     return -EAGAIN;
> > > 
> > > ...
> > > 
> > > This looks strange. First the flag is cleared and then it is (in some
> > > cases) set again.
> > 
> > i.MX controllers require writing a 0 to clear these bits. Vybrid
> > controllers need writing a 1 for the same.
> 
> Yes, I understood that.
> 
> > > If I2SR_IIF is set in temp you ack this irq without handling it. (Which
> > > might happen if atomic is set and irqs are off?!)
> > 
> > This patch is only about using the correct processor specific value for
> > acknowledging an IRQ... But I think that returning EAGAIN (which aborts
> > the
> > transfer) should be handling enough. At the next transfer, the controller
> > will be set back to master mode.
> 
> Either you didn't understand what I meant, or I don't understand why you
> consider your patch right anyhow. 
Probably both.

> So I try to explain in other and more words.
> 
> IMHO the intention here (and also what happens on i.MX) is that exactly
> the AL interrupt pending bit should be cleared and the IF irq is
> supposed to be untouched.
Yes.

> Given there are only two irq flags in the I2SR register (which is called
> IBSR on Vybrid) ...

Vybrid:
=======
+-------+-----+------+-----+------+-----+-----+------+------+
| BIT   |  7  |  6   |  5  |  4   |  3  |  2  |  1   |  0   |
+-------+-----+------+-----+------+-----+-----+------+------+
| READ  | TCF | IAAS | IBB | IBAL |  0  | SRW | IBIF | RXAK |
+-------+-----+------+-----+------+-----+-----+------+------+
| WRITE |  -  |  -   |  -  | W1C  |  -  |  -  | W1C  |  -   |
+-------+-----+------+-----+-^^^--+-----+-----+-^^^--+------+

i.MX:
=======
+-------+-----+------+-----+------+-----+-----+------+------+
| BIT   |  7  |  6   |  5  |  4   |  3  |  2  |  1   |  0   |
+-------+-----+------+-----+------+-----+-----+------+------+
| READ  | ICF | IAAS | IBB | IAL  |  0  | SRW | IIF  | RXAK |
+-------+-----+------+-----+------+-----+-----+------+------+
| WRITE |  -  |  -   |  -  | W0C  |  -  |  -  | W0C  |  -   |
+-------+-----+------+-----+-^^^--+-----+-----+-^^^--+------+


> ... the status quo (i.e. without your patch) is:
> 
>   On i.MX IAL is cleared
Yes

>   On Vybrid IIF (which is called IBIF there) is cleared.
If IBIF is set, then it's cleared (probably by accident).
But in the "if (temp & I2SR_IAL)" condition, I focus on 
the IBAL flag, not IBIF.

> With your patch we get:
> 
>   On i.MX IAL is cleared
Yes

>   On Vybrid both IIF (aka IBIF) and IAL (aka IBAL) are cleared.
Agree. IBAL is cleared by intention, IBIF by accident (if set).
Do you see any problem if IBIF is also cleared?


> To get it right for both SoC types you have to do (e.g.):
> 
> 	temp = ~i2c_imx->hwdata->i2sr_clr_opcode ^ I2SR_IAL;
Sorry, even if this is correct, it looks hard to understand for me.

> (and in i2c_imx_isr() the same using I2SR_IIF instead of I2SR_IAL
> because there currently IAL might be cleared by mistake on Vybrid).
> 
> I considered creating a patch, but as I don't have a Vybrid on my desk
> and on i.MX there is no change, I let you do this.
I also don't own a Vybrid system. I consider my patch correct in terms of
clearing the IBAL flag (which was wrong before). Additional work may be
useful for NOT clearing the other status flag. I also would like to keep
this task for somebody who owns a Vybrid system. But the other patches
in this series fixes some more important problems, so maybe you could
give your acknowledge anyhow.

Best regards
Christian