Message-ID: <20120807110156.4d0c2571@pyramind.ukuu.org.uk>
Date:	Tue, 7 Aug 2012 11:01:56 +0100
From:	Alan Cox <alan@...rguk.ukuu.org.uk>
To:	Arun Murthy <arun.murthy@...ricsson.com>
Cc:	<linux-kernel@...r.kernel.org>, <netdev@...r.kernel.org>,
	<linux-doc@...r.kernel.org>, <gregkh@...uxfoundation.org>,
	<sjur.brandeland@...ricsson.com>
Subject: Re: [PATCHv2 3/4] modem_shm: u8500-shm: U8500 Shared Memory Driver

> +#define SIZE_OF_FIFO (512*1024)
> +
> +static u8 message_fifo[ISA_DEVICES][SIZE_OF_FIFO];

Thats a huge amount of static memory that gets allocated regardless of
whether the device is open or being used ?

> +static int major;
> +module_param(major, int, 0);
> +MODULE_PARM_DESC(major, "Major device number");

Is this really needed still - it looks like an escaped debugging aid

> +/* global fops mutex */
> +static DEFINE_MUTEX(isa_lock);
> +
> +/**
> + * shrm_get_cdev_index() - return the index mapped to l2 header
> + * @l2_header:	L2 header
> + *
> + * struct map_device maps the index(count) with the device L2 header.
> + * This function returns the index for the provided L2 header in case
> + * of success else -ve value.
> + */
> +int shrm_get_cdev_index(u8 l2_header)
> +{
> +	u8 cnt;
> +	for (cnt = 0; cnt < ISA_DEVICES; cnt++) {
> +		if (map_dev[cnt].l2_header == l2_header)
> +			return map_dev[cnt].idx;
> +	}
> +	return -EINVAL;
> +}

Would be good to document the locking assumptions on this lot

> +		/* reset the msg queue pointers */
> +		q->size = SIZE_OF_FIFO;
> +		q->readptr = 0;
> +		q->writeptr = 0;

Is there a reason for not using the existing kfifo layer for this work or
does it need to handle other things kfifo cannot.

> +static u32 isa_select(struct file *filp,
> +				struct poll_table_struct *wait)
> +{
> +	struct isadev_context *isadev = filp->private_data;
> +	struct shrm_dev *shrm = isadev->dl_queue.shrm;
> +	struct message_queue *q;
> +	u32 mask = 0;
> +	u32 m = iminor(filp->f_path.dentry->d_inode);
> +	u8 idx = shrm_get_cdev_index(m);
> +
> +	dev_dbg(shrm->dev, "%s IN\n", __func__);
> +
> +	if (shrm->msr_flag)
> +		return -ENODEV;

This test has no associated locking. Please explain what stop sit
becoming true during the rest of this function ?

> +/**
> + * isa_read() - Read from device
> + * @filp:	file descriptor
> + * @buf:	user buffer pointer
> + * @len:	size of requested data transfer
> + * @ppos:	not used
> + *
> + * It reads a oldest message from queue and copies it into user buffer and
> + * returns its size.
> + * If there is no message present in queue, then it blocks until new data is
> + * available.
> + */
> +ssize_t isa_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
> +{
> +	u32 size = 0;
> +	int ret;
> +	char *psrc;
> +	struct isadev_context *isadev = (struct isadev_context *)
> +							filp->private_data;

Coding style - don't need the casts for void *

> +	struct shrm_dev *shrm = isadev->dl_queue.shrm;
> +	struct message_queue *q;
> +	u32 msgsize;
> +
> +	dev_dbg(shrm->dev, "%s IN\n", __func__);
> +
> +	if (len <= 0)
> +		return -EFAULT;

How can this occur ?
> +
> +	q = &isadev->dl_queue;
> +
> +	if (shrm->msr_flag) {
> +		atomic_set(&q->q_rp, 0);
> +		return -ENODEV;
> +	}

Again - tests without locking
> +
> +	spin_lock_bh(&q->update_lock);
> +	if (list_empty(&q->msg_list)) {
> +		spin_unlock_bh(&q->update_lock);
> +		dev_dbg(shrm->dev, "Waiting for Data\n");
> +		if (wait_event_interruptible(q->wq_readable,
> +				atomic_read(&q->q_rp) == 1))
> +			return -ERESTARTSYS;
> +	} else
> +		spin_unlock_bh(&q->update_lock);
> +
> +	if (shrm->msr_flag) {
> +		atomic_set(&q->q_rp, 0);
> +		return -ENODEV;
> +	}
> +
> +	msgsize = get_size_of_new_msg(q);
> +
> +	if (len < msgsize)
> +		return -EINVAL;

What happens with two parallel reads - I don't see what prevents
corruption if that occurs or one racing read freeing the message before
another has finished processing it.

> +
> +	if ((q->readptr+msgsize) >= q->size) {

Minor style comment - you seen to randomly switch between a+b and a + b
styles 8)

> +		dev_dbg(shrm->dev, "Inside Loop Back\n");
> +		psrc = (char *)buf;

Probably best to use u8 * here, your buffer contents are not signed I
assume ?

> +		size = (q->size-q->readptr);
> +		/* Copy First Part of msg */
> +		if (copy_to_user(psrc,
> +				(u8 *)(q->fifo_base+q->readptr),
> +				size)) {
> +			dev_err(shrm->dev, "copy_to_user failed\n");
> +			return -EFAULT;
> +		}
> +		psrc += size;
> +		/* Copy Second Part of msg at the top of fifo */
> +		if (copy_to_user(psrc,
> +				(u8 *)(q->fifo_base),
> +				(msgsize-size))) {
> +			dev_err(shrm->dev, "copy_to_user failed\n");
> +			return -EFAULT;
> +		}
> +	} else {
> +		if (copy_to_user(buf,
> +				(u8 *)(q->fifo_base + q->readptr),
> +				msgsize)) {
> +			dev_err(shrm->dev, "copy_to_user failed\n");
> +			return -EFAULT;
> +		}

Pedantically you should return the number of bytes successfully copied if
you consume data. As this code stands it's not a big deal as far as I can
see since you don't remove the message in this situation. I suspect that
may need to change ?


> +/**
> + * isa_write() - Write to shrm char device
> + * @filp:	file descriptor
> + * @buf:	user buffer pointer
> + * @len:	size of requested data transfer
> + * @ppos:	not used
> + *
> + * It checks if there is space available in queue, and copies the message
> + * inside queue. If there is no space, it blocks until space becomes available.
> + * It also schedules transfer thread to transmit the newly added message.
> + */
> +ssize_t isa_write(struct file *filp, const char __user *buf,
> +				 size_t len, loff_t *ppos)
> +{
> +	struct isadev_context *isadev = filp->private_data;
> +	struct shrm_dev *shrm = isadev->dl_queue.shrm;
> +	struct message_queue *q;
> +	void *addr = 0;
> +	int err, l2_header;
> +	int ret = 0;
> +
> +	dev_dbg(shrm->dev, "%s IN\n", __func__);
> +
> +	if (len <= 0 || buf == NULL)
> +		return -EFAULT;

len < 0 cannot occur, buf == NULL is not an error

> +	q = &isadev->dl_queue;
> +	l2_header = shrm_get_cdev_l2header(isadev->device_id);
> +	if (l2_header < 0) {
> +		dev_err(shrm->dev, "failed to get L2 header\n");
> +		return l2_header;
> +	}
> +
> +	switch (l2_header) {

Why keep going through switches and lookup tables - can't you cache this
in the isadev or in some file private data ?

> +	if (copy_from_user(addr, buf, len)) {
> +		dev_err(shrm->dev, "copy_from_user failed\n");

This allows an user to fill the logs with crap - should be dev_dbg

Also what happens on two parallel writers ?


> +/**
> + * isa_ioctl() - To handle different ioctl commands supported by driver.
> + * @inode:	structure is used by the kernel internally to represent files
> + * @filp:	file descriptor pointer
> + * @cmd:	ioctl command
> + * @arg:	input param

This appears incomplete - it's just debug statements


> +}
> +/**
> + * isa_mmap() - Maps kernel queue memory to user space.
> + * @filp:	file descriptor pointer
> + * @vma:	virtual area memory structure.
> + *
> + * This function maps kernel FIFO into user space. This function
> + * shall be called twice to map both uplink and downlink buffers.

Again this appears incomplete

> +	if (atomic_dec_and_test(&isa_context->is_open[idx])) {
> +		atomic_inc(&isa_context->is_open[idx]);

This isn't safe I'd point out.. although it seems to jut be a "can't
happen" recovery

> +		dev_err(shrm->dev, "Device not opened yet\n");
> +		mutex_unlock(&isa_lock);
> +		return -ENODEV;
> +	}
> +	atomic_set(&isa_context->is_open[idx], 1);

How do you know it will always be one. Also given it's within the mutex
in all uses I can see why is it an atomic ?

> +
> +	switch (m) {
> +	case RPC_MESSAGING:
> +		dev_info(shrm->dev, "Close RPC_MESSAGING Device\n");
> +		break;
> +	case AUDIO_MESSAGING:
> +		dev_info(shrm->dev, "Close AUDIO_MESSAGING Device\n");
> +		break;
> +	case SECURITY_MESSAGING:
> +		dev_info(shrm->dev, "CLose SECURITY_MESSAGING Device\n");
> +		break;
> +	case COMMON_LOOPBACK_MESSAGING:
> +		kfree(isadev->addr);
> +		dev_info(shrm->dev, "Close COMMON_LOOPBACK_MESSAGING Device\n");
> +		break;
> +	case AUDIO_LOOPBACK_MESSAGING:
> +		kfree(isadev->addr);
> +		dev_info(shrm->dev, "Close AUDIO_LOOPBACK_MESSAGING Device\n");
> +		break;
> +	case CIQ_MESSAGING:
> +		kfree(isadev->addr);
> +		dev_info(shrm->dev, "Close CIQ_MESSAGING Device\n");
> +		break;
> +	case RTC_CAL_MESSAGING:
> +		dev_info(shrm->dev, "Close RTC_CAL_MESSAGING Device\n");
> +		break;

kfree(NULL) is a no-op and guaranteed so, providing you see isadev->addr
= NULL properly in the open path you can delete all of this code and just
do
kfree(isadev->addr);
mutex_unlock(...)
return 0;

> +	if (get_boot_state() != BOOT_DONE) {
> +		dev_err(shrm->dev, "Boot is not done\n");
> +		return -EBUSY;
> +	}

Is it guaranteed that this is a one way path - ie a device never goes
back into BOOT state ?

> +	if (!atomic_dec_and_test(&isa_context->is_open[idx])) {
> +		atomic_inc(&isa_context->is_open[idx]);
> +		dev_err(shrm->dev, "Device already opened\n");
> +		mutex_unlock(&isa_lock);
> +		return -EBUSY;

See comments earlier about the atomics

> +	}
> +	isadev = &isa_context->isadev[idx];
> +	if (filp != NULL)
> +		filp->private_data = isadev;

How can filp be NULL ?

> +
> +	switch (m) {
> +	case RPC_MESSAGING:
> +		dev_info(shrm->dev, "Open RPC_MESSAGING Device\n");
> +		break;
> +	case AUDIO_MESSAGING:
> +		dev_info(shrm->dev, "Open AUDIO_MESSAGING Device\n");
> +		break;
> +	case SECURITY_MESSAGING:
> +		dev_info(shrm->dev, "Open SECURITY_MESSAGING Device\n");
> +		break;
> +	case COMMON_LOOPBACK_MESSAGING:
> +		isadev->addr = kzalloc(10 * 1024, GFP_KERNEL);
> +		if (!isadev->addr) {
> +			mutex_unlock(&isa_lock);
> +			return -ENOMEM;
> +		}
> +		dev_info(shrm->dev, "Open COMMON_LOOPBACK_MESSAGING Device\n");
> +		break;
> +	case AUDIO_LOOPBACK_MESSAGING:
> +		isadev->addr = kzalloc(10 * 1024, GFP_KERNEL);
> +		if (!isadev->addr) {
> +			mutex_unlock(&isa_lock);
> +			return -ENOMEM;
> +		}
> +		dev_info(shrm->dev, "Open AUDIO_LOOPBACK_MESSAGING Device\n");
> +		break;
> +	case CIQ_MESSAGING:
> +		isadev->addr = kzalloc(10 * 1024, GFP_KERNEL);
> +		if (!isadev->addr) {
> +			mutex_unlock(&isa_lock);
> +			return -ENOMEM;
> +		}
> +		dev_info(shrm->dev, "Open CIQ_MESSAGING Device\n");
> +		break;
> +	case RTC_CAL_MESSAGING:
> +		dev_info(shrm->dev, "Open RTC_CAL_MESSAGING Device\n");
> +		break;
> +	};

So all of this could just be

	dev_dbg("Open device %d\n", m);
	switch (m) {
	case a
	case b
	case c
		isadev->addr = ....
		break;
	case e
	case f
		isadev->addr = NULL;
		break;
	default:
		error path

also in your error case right now if you run out of memory you don't fix
up the open counters.

> +const struct file_operations isa_fops = {

static


> +	for (no_dev = 0; no_dev < ISA_DEVICES; no_dev++) {
> +		atomic_set(&isa_context->is_open[no_dev], 1);
> +		device_create(isa_context->shm_class, NULL,
> +				MKDEV(MAJOR(dev_id),
> +				map_dev[no_dev].l2_header), NULL,
> +				map_dev[no_dev].name);
> +	}

What happens if I open the device right here... ?

> +
> +	isa_context->isadev = kzalloc(sizeof
> +				(struct isadev_context)*ISA_DEVICES,
> +				GFP_KERNEL);
> +	if (isa_context->isadev == NULL) {
> +		dev_err(shrm->dev, "Failed to alloc memory\n");
> +		return -ENOMEM;

You just leaked all the other stuff you created, left device pointers to
invalid memory registered.

There is a reason kernel drivers use the goto type error handling - it
avoids missing these things.


> diff --git a/drivers/modem_shm/u8500_shm/shrm_fifo.c b/drivers/modem_shm/u8500_shm/shrm_fifo.c
> new file mode 100644
> index 0000000..ad63cc4
> --- /dev/null
> +++ b/drivers/modem_shm/u8500_shm/shrm_fifo.c
> @@ -0,0 +1,837 @@
> +/*
> + * Copyright (C) ST-Ericsson SA 2010
> + *
> + * Author: Arun Murthy <arun.murthy@...ricsson.com>
> + *	Kumar Sanghvi for ST-Ericsson
> + *
> + * License terms: GNU General Public License (GPL) version 2
> + */
> +
> +#include <linux/modem_shm/u8500_shm/shrm.h>
> +#include <linux/modem_shm/u8500_shm/shrm_driver.h>
> +#include <linux/modem_shm/u8500_shm/shrm_private.h>
> +#include <linux/modem_shm/u8500_shm/shrm_net.h>
> +#include <linux/mfd/dbx500-prcmu.h>
> +
> +#define L1_BOOT_INFO_REQ	1
> +#define L1_BOOT_INFO_RESP	2
> +#define L1_NORMAL_MSG		3
> +#define L1_HEADER_MASK		28
> +#define L1_MAPID_MASK		0xF0000000
> +#define CONFIG_OFFSET		8
> +#define COUNTER_OFFSET		20
> +#define L2_HEADER_SIZE		4
> +#define L2_HEADER_OFFSET	24
> +#define MASK_0_15_BIT		0xFF
> +#define MASK_16_31_BIT		0xFF00
> +#define MASK_16_27_BIT		0xFFF0000
> +#define MASK_0_39_BIT		0xFFFFF
> +#define MASK_40_55_BIT		0xFF00000
> +#define MASK_8_16_BIT           0x0000FF00
> +#define MSG_LEN_OFFSET          16
> +#define SHRM_VER                2
> +#define ca_ist_inactivity_timer 25 /*25ms */
> +#define ca_csc_inactivity_timer 25 /*25ms */
> +
> +static u8 msg_audio_counter;
> +static u8 msg_common_counter;
> +
> +struct fifo_write_params ape_shm_fifo_0;
> +struct fifo_write_params ape_shm_fifo_1;
> +struct fifo_read_params cmt_shm_fifo_0;
> +struct fifo_read_params cmt_shm_fifo_1;

static or fix the naming
> +
> +
> +static u8 cmt_read_notif_0_send;
> +static u8 cmt_read_notif_1_send;
> +
> +void shm_fifo_init(struct shrm_dev *shrm)

Fix the naming shm_ is the kernel shared memory layer, and we don't want
clashes. You seem to use shrm_ elsewhere

 +u8 read_boot_info_req(struct shrm_dev *shrm,

Static or naming

> +void write_boot_info_resp(struct shrm_dev *shrm, u32 config,
> +							u32 version)
> +{

Ditto

> +		skb_copy_to_linear_data(skb,
> +				(u8 *)(q->fifo_base + q->readptr), size);
> +		skb_put(skb, size);

Is this coming from host or ioremap memory ?

> +	skb->dev = dev;/*kmalloc(sizeof(struct net_device), GFP_ATOMIC);*/

Wants cleaning up

> +	struct shrm_net_iface_priv *net_iface_priv =
> +			(struct shrm_net_iface_priv *)netdev_priv(dev);

[style: unneeded casts - lot of thise in this file]



> +struct sock *shrm_nl_sk;

Better to make globals visible than bury them in a mass of statics

(actually all of this raises a big question - what happens if you ever
need to have two modems - you are basically hardcoding one per system.
That's fine if you are *very* sure it will remain true)

> +/* Spin lock and tasklet declaration */
> +DECLARE_TASKLET(shm_ca_0_tasklet, shm_ca_msgpending_0_tasklet, 0);
> +DECLARE_TASKLET(shm_ca_1_tasklet, shm_ca_msgpending_1_tasklet, 0);
> +DECLARE_TASKLET(shm_ac_read_0_tasklet, shm_ac_read_notif_0_tasklet, 0);
> +DECLARE_TASKLET(shm_ac_read_1_tasklet, shm_ac_read_notif_1_tasklet, 0);

static


> +#ifdef CONFIG_U8500_SHRM_MODEM_SILENT_RESET
> +static int shrm_modem_reset_sequence(void)

> +	/* reset the boot state */
> +	spin_lock_irqsave(&boot_lock, flags);
> +	boot_state = BOOT_INIT;

Your open method locking earlier assumes this cannot go backwards...


> +		dev_info(shm_dev->dev, "Initiating Modem silent reset\n");

Seems an odd use of dev_info.. if its a bug workaround then perhaps
dev_warn, and if its not important dev_dbg ?

> +DECLARE_TASKLET(shrm_sw_reset_callback, shrm_modem_reset_callback,
> +		IRQ_PRCMU_MODEM_SW_RESET_REQ);
> +

static 

Lots of other stuff here that should be shrm_ or static...

There also seems to be places where you copy or touch the fifo memory
which was ioremapped but don't use readb/writeb/memcpy_toio etc, or
handle endianness. While endianness might be the kind of thing you can be
sure of in some embedded platforms not using readb and the like means the
compiler can do interesting optimisations you won't want and is something
that will break on some architectures and potentially on others in future.

Alan
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists