From: Sean Anderson <sean.anderson@seco.com>
To: Andrew Lunn <andrew@lunn.ch>
Cc: Heiner Kallweit <hkallweit1@gmail.com>,
Russell King <linux@armlinux.org.uk>,
netdev@vger.kernel.org, "David S . Miller" <davem@davemloft.net>,
Vladimir Oltean <olteanv@gmail.com>,
linux-kernel@vger.kernel.org, Paolo Abeni <pabeni@redhat.com>,
Eric Dumazet <edumazet@google.com>,
Florian Fainelli <f.fainelli@gmail.com>,
Tobias Waldekranz <tobias@waldekranz.com>,
Jakub Kicinski <kuba@kernel.org>
Subject: Re: [PATCH net-next] net: mdio: Add netlink interface
Date: Tue, 7 Mar 2023 11:16:11 -0500 [thread overview]
Message-ID: <f947e5e2-770e-4b12-67ae-8abf5866e250@seco.com> (raw)
In-Reply-To: <7a02294e-bf50-4399-9e68-1235ba24a381@lunn.ch>
On 3/7/23 09:22, Andrew Lunn wrote:
>> To prevent userspace phy drivers, writes are disabled by default, and can
>> only be enabled by editing the source. This is the same strategy used by
>> regmap for debugfs writes. Unfortunately, this disallows several useful
>> features, including
>>
>> - Register writes (obviously)
>> - C45-over-C22
>
> You could add C45-over-C22 as another op.
>
> This tool is dangerous, even in its read only mode, just like the
> IOCTL interface. Interrupt status registers are often clear on read,
> so you can loose interrupts. Statistics counters are sometimes clear
> on read. BMSR link bit is also latching, so a read of it could mean
> you miss link events, etc. Adding C45-over-C22 is just as dangerous,
> you can mess up MDIO switches which use the registers for other
> things, but by deciding to use this tool you have decided to take the
> risk of blowing your foot off.
Yes, and I should probably have commented on this in the commit message.
IMO the things you listed are... iffy but unlikely to cause a
malfunction. Tainting on read would be fine, since it is certainly
possible to imagine hardware which would malfunction on read. I suppose
regmap gets away with this by sticking the whole thing in debugfs.
>> - Atomic access to paged registers
>> - Better MDIO emulation for e.g. QEMU
>>
>> However, the read-only interface remains broadly useful for debugging.
>
> I would say it is broadly useful for PHYs. But not Ethernet switches,
> when in read only mode.
>
>> +static int mdio_nl_open(struct mdio_nl_xfer *xfer);
>> +static int mdio_nl_close(struct mdio_nl_xfer *xfer, bool last, int xerr);
>
> I guess i never did a proper review of this code before, due to not
> liking the concept....
>
> Move the code around so these are not needed, unless there are
> functions which are mutually recursive.
They do indeed call each other (although by my analysis no recursion
results). The forward declaration of mdio_nl_open is unnecessary, so I
will rearrange things to avoid that.
>> +static inline u16 *__arg_r(u32 arg, u16 *regs)
>> +{
>> + WARN_ON_ONCE(arg >> 16 != MDIO_NL_ARG_REG);
>> +
>> + return ®s[arg & 0x7];
>> +}
>
> No inline functions in C files. Leave the compiler to decide.
OK
>> +static int mdio_nl_eval(struct mdio_nl_xfer *xfer)
>> +{
>> + struct mdio_nl_insn *insn;
>> + unsigned long timeout;
>> + u16 regs[8] = { 0 };
>> + int pc, ret = 0;
>> + int phy_id, reg, prtad, devad, val;
>> +
>> + timeout = jiffies + msecs_to_jiffies(xfer->timeout_ms);
>> +
>> + mutex_lock(&xfer->mdio->mdio_lock);
>
> Should timeout be set inside the lock, for when you have two
> applications running in parallel and each take a while?
That seems reasonable.
>> +
>> + for (insn = xfer->prog, pc = 0;
>> + pc < xfer->prog_len;
>> + insn = &xfer->prog[++pc]) {
>> + if (time_after(jiffies, timeout)) {
>> + ret = -ETIMEDOUT;
>> + break;
>> + }
>> +
>> + switch ((enum mdio_nl_op)insn->op) {
>> + case MDIO_NL_OP_READ:
>> + phy_id = __arg_ri(insn->arg0, regs);
>> + prtad = mdio_phy_id_prtad(phy_id);
>> + devad = mdio_phy_id_devad(phy_id);
>> + reg = __arg_ri(insn->arg1, regs);
>> +
>> + if (mdio_phy_id_is_c45(phy_id))
>> + ret = __mdiobus_c45_read(xfer->mdio, prtad,
>> + devad, reg);
>> + else
>> + ret = __mdiobus_read(xfer->mdio, phy_id, reg);
>
> The application should say if it want to do C22 or C45.
The phy_id comes from the application. So it sets MDIO_PHY_ID_C45 if it wants
to use C45.
> As you said in
> the cover note, the ioctl interface is limiting when there is no PHY,
> so you are artificially adding the same restriction here.
I don't follow.
> Also, you
> might want to do C45 on a C22 PHY, e.g. to access EEE registers. Plus
> you could consider adding C45 over C22 here.
As Russell noted, this is dangerous in the general case.
>> +
>> + if (ret < 0)
>> + goto exit;
>> + *__arg_r(insn->arg2, regs) = ret;
>> + ret = 0;
>> + break;
>> +
>> + case MDIO_NL_OP_WRITE:
>> + phy_id = __arg_ri(insn->arg0, regs);
>> + prtad = mdio_phy_id_prtad(phy_id);
>> + devad = mdio_phy_id_devad(phy_id);
>> + reg = __arg_ri(insn->arg1, regs);
>> + val = __arg_ri(insn->arg2, regs);
>> +
>> +#ifdef MDIO_NETLINK_ALLOW_WRITE
>> + add_taint(TAINT_USER, LOCKDEP_STILL_OK);
>
> I don't know, but maybe taint on read as well.
>
>> + if (mdio_phy_id_is_c45(phy_id))
>> + ret = __mdiobus_c45_write(xfer->mdio, prtad,
>> + devad, reg, val
>> + else
>> + ret = __mdiobus_write(xfer->mdio, dev, reg,
>> + val);
>> +#else
>> + ret = -EPERM;
>
> EPERM is odd, EOPNOTSUPP would be better. EPERM suggests you can run
> it as root and it should work.
Well, EPERM is what you get when trying to write a 444 file, which is
effectively what we're enforcing here.
--Sean
next prev parent reply other threads:[~2023-03-07 16:16 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-03-06 20:45 Sean Anderson
2023-03-06 22:48 ` Russell King (Oracle)
2023-03-06 23:39 ` Sean Anderson
2023-03-07 13:47 ` Andrew Lunn
2023-03-07 16:41 ` Sean Anderson
2023-03-07 0:05 ` kernel test robot
2023-03-07 11:23 ` Michael Walle
2023-03-07 13:49 ` Andrew Lunn
2023-03-07 14:05 ` Vladimir Oltean
2023-03-07 14:33 ` Andrew Lunn
2023-03-07 15:00 ` Russell King (Oracle)
2023-03-07 12:26 ` Tobias Waldekranz
2023-03-07 16:30 ` Sean Anderson
2023-03-07 14:22 ` Andrew Lunn
2023-03-07 14:50 ` Russell King (Oracle)
2023-03-07 16:16 ` Sean Anderson [this message]
2023-03-07 17:23 ` Andrew Lunn
2023-03-07 17:42 ` Sean Anderson
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=f947e5e2-770e-4b12-67ae-8abf5866e250@seco.com \
--to=sean.anderson@seco.com \
--cc=andrew@lunn.ch \
--cc=davem@davemloft.net \
--cc=edumazet@google.com \
--cc=f.fainelli@gmail.com \
--cc=hkallweit1@gmail.com \
--cc=kuba@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux@armlinux.org.uk \
--cc=netdev@vger.kernel.org \
--cc=olteanv@gmail.com \
--cc=pabeni@redhat.com \
--cc=tobias@waldekranz.com \
/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
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®