[U-Boot] [PATCH v5 2/2] common: Generic firmware loader for file system

[Lothar Waßmann]

Hi,

On Fri, 22 Dec 2017 01:43:38 +0000 Chee, Tien Fong wrote:
> On Kha, 2017-12-21 at 16:08 +0100, Lothar Waßmann wrote:
> > Hi,
> > 
> > On Thu, 21 Dec 2017 12:48:53 +0000 Chee, Tien Fong wrote:
> > > 
> > > On Kha, 2017-12-21 at 12:53 +0100, Lothar Waßmann wrote:
> > > > 
> > > > Hi,
> > > > 
> > > > On Thu, 21 Dec 2017 09:36:41 +0000 Chee, Tien Fong wrote:
> > > > > 
> > > > > 
> > > > > On Kha, 2017-12-21 at 09:48 +0100, Lothar Waßmann wrote:
> > > > > > 
> > > > > > 
> > > > > > Hi,
> > > > > > 
> > > > > > On Thu, 21 Dec 2017 15:25:29 +0800 tien.fong.chee at intel.com
> > > > > > wrote:
> > > > > > > 
> > > > > > > 
> > > > > > > 
> > > > > > > From: Tien Fong Chee <tien.fong.chee at intel.com>
> > > > > > > 
> > > > > > > This is file system generic loader which can be used to
> > > > > > > load
> > > > > > > the file image from the storage into target such as memory.
> > > > > > > The consumer driver would then use this loader to program
> > > > > > > whatever,
> > > > > > > ie. the FPGA device.
> > > > > > > 
> > > > > > > Signed-off-by: Tien Fong Chee <tien.fong.chee at intel.com>
> > > > > > > ---
> > > > > > >  common/Makefile            |   1 +
> > > > > > >  common/fs_loader.c         | 311
> > > > > > > +++++++++++++++++++++++++++++++++++++++++++++
> > > > > > >  doc/README.firmware_loader |  76 +++++++++++
> > > > > > >  include/fs_loader.h        |  28 ++++
> > > > > > >  4 files changed, 416 insertions(+)
> > > > > > >  create mode 100644 common/fs_loader.c
> > > > > > >  create mode 100644 doc/README.firmware_loader
> > > > > > >  create mode 100644 include/fs_loader.h
> > > > > > > 
> > > > > > > diff --git a/common/Makefile b/common/Makefile
> > > > > > > index cec506f..2934221 100644
> > > > > > > --- a/common/Makefile
> > > > > > > +++ b/common/Makefile
> > > > > > > @@ -130,3 +130,4 @@ obj-$(CONFIG_CMD_DFU) += dfu.o
> > > > > > >  obj-y += command.o
> > > > > > >  obj-y += s_record.o
> > > > > > >  obj-y += xyzModem.o
> > > > > > > +obj-y += fs_loader.o
> > > > > > > diff --git a/common/fs_loader.c b/common/fs_loader.c
> > > > > > > new file mode 100644
> > > > > > > index 0000000..ddfce58
> > > > > > > --- /dev/null
> > > > > > > +++ b/common/fs_loader.c
> > > > > > > @@ -0,0 +1,311 @@
> > > > > > > +/*
> > > > > > > + * Copyright (C) 2017 Intel Corporation <www.intel.com>
> > > > > > > + *
> > > > > > > + * SPDX-License-Identifier:    GPL-2.0
> > > > > > > + */
> > > > > > > +
> > > > > > > +#include <common.h>
> > > > > > > +#include <errno.h>
> > > > > > > +#include <fs.h>
> > > > > > > +#include <fs_loader.h>
> > > > > > > +#include <nand.h>
> > > > > > > +#include <sata.h>
> > > > > > > +#include <spi.h>
> > > > > > > +#include <spi_flash.h>
> > > > > > > +#include <spl.h>
> > > > > > > +#include <linux/string.h>
> > > > > > > +#include <usb.h>
> > > > > > > +
> > > > > > > +struct firmware_priv {
> > > > > > > +	const char *name;	/* Filename */
> > > > > > > +	u32 offset;		/* Offset of reading a
> > > > > > > file
> > > > > > > */
> > > > > > > +};
> > > > > > > +
> > > > > > > +static struct device_location default_locations[] = {
> > > > > > > +	{
> > > > > > > +		.name = "mmc",
> > > > > > > +		.devpart = "0:1",
> > > > > > > +	},
> > > > > > > +	{
> > > > > > > +		.name = "usb",
> > > > > > > +		.devpart = "0:1",
> > > > > > > +	},
> > > > > > > +	{
> > > > > > > +		.name = "sata",
> > > > > > > +		.devpart = "0:1",
> > > > > > > +	},
> > > > > > > +};
> > > > > > > +
> > > > > > > +/* USB build is not supported yet in SPL */
> > > > > > > +#ifndef CONFIG_SPL_BUILD
> > > > > > > +#ifdef CONFIG_USB_STORAGE
> > > > > > > +static int init_usb(void)
> > > > > > > +{
> > > > > > > +	int err;
> > > > > > > +
> > > > > > > +	err = usb_init();
> > > > > > > +	if (err)
> > > > > > > +		return err;
> > > > > > > +
> > > > > > > +#ifndef CONFIG_DM_USB
> > > > > > > +	err = usb_stor_scan(1) < 0 ? -ENODEV : 0;
> > > > > > > +#endif
> > > > > > > +
> > > > > > > +	return err;
> > > > > > > +}
> > > > > > > +#else
> > > > > > > +static int init_usb(void)
> > > > > > > +{
> > > > > > > +	printf("Error: Cannot load flash image: no USB
> > > > > > > support\n");
> > > > > > > +	return -ENOSYS;
> > > > > > > +}
> > > > > > > +#endif
> > > > > > > +#endif
> > > > > > > +
> > > > > > > +#ifdef CONFIG_SATA
> > > > > > > +static int init_storage_sata(void)
> > > > > > > +{
> > > > > > > +	return sata_probe(0);
> > > > > > > +}
> > > > > > > +#else
> > > > > > > +static int init_storage_sata(void)
> > > > > > > +{
> > > > > > > +	printf("Error: Cannot load image: no SATA
> > > > > > > support\n");
> > > > > > > +	return -ENOSYS;
> > > > > > > +}
> > > > > > > +#endif
> > > > > > > +
> > > > > > > +#ifdef CONFIG_CMD_UBIFS
> > > > > > > +static int mount_ubifs(struct device_location *location)
> > > > > > > +{
> > > > > > > +	int ret;
> > > > > > > +	char cmd[32];
> > > > > > > +
> > > > > > > +	sprintf(cmd, "ubi part %s", location->mtdpart);
> > > > > > > +
> > > > > > > +	ret = run_command(cmd, 0);
> > > > > > > +	if (ret)
> > > > > > > +		return ret;
> > > > > > > +
> > > > > > > +	sprintf(cmd, "ubifsmount %s", location->ubivol);
> > > > > > > +
> > > > > > > +	ret = run_command(cmd, 0);
> > > > > > > +
> > > > > > > +	return ret;
> > > > > > > +}
> > > > > > > +
> > > > > > > +static int umount_ubifs(void)
> > > > > > > +{
> > > > > > > +	return run_command("ubifsumount", 0);
> > > > > > > +}
> > > > > > > +#else
> > > > > > > +static int mount_ubifs(struct device_location *location)
> > > > > > > +{
> > > > > > > +	printf("Error: Cannot load image: no UBIFS
> > > > > > > support\n");
> > > > > > > +	return -ENOSYS;
> > > > > > > +}
> > > > > > > +#endif
> > > > > > > +
> > > > > > > +#if defined(CONFIG_SPL_MMC_SUPPORT) &&
> > > > > > > defined(CONFIG_SPL_BUILD)
> > > > > > > +static int init_mmc(void)
> > > > > > > +{
> > > > > > > +	/* Just for the case MMC is not yet initialized */
> > > > > > > +	struct mmc *mmc = NULL;
> > > > > > > +	int err;
> > > > > > > +
> > > > > > > +	spl_mmc_find_device(&mmc, spl_boot_device());
> > > > > > > +
> > > > > > > +	err = mmc_init(mmc);
> > > > > > > +	if (err) {
> > > > > > > +		printf("spl: mmc init failed with error:
> > > > > > > %d\n",
> > > > > > > err);
> > > > > > > +		return err;
> > > > > > > +	}
> > > > > > > +
> > > > > > > +	return err;
> > > > > > > +}
> > > > > > > +#else
> > > > > > > +static int init_mmc(void)
> > > > > > > +{
> > > > > > > +	/* Expect somewhere already initialize MMC */
> > > > > > > +	return 0;
> > > > > > > +}
> > > > > > > +#endif
> > > > > > > +
> > > > > > > +static int select_fs_dev(struct device_location *location)
> > > > > > > +{
> > > > > > > +	int ret;
> > > > > > > +
> > > > > > > +	if (!strcmp("mmc", location->name)) {
> > > > > > > +		ret = fs_set_blk_dev("mmc", location-
> > > > > > > >devpart,
> > > > > > > FS_TYPE_ANY);
> > > > > > > +	} else if (!strcmp("usb", location->name)) {
> > > > > > > +		ret = fs_set_blk_dev("usb", location-
> > > > > > > >devpart,
> > > > > > > FS_TYPE_ANY);
> > > > > > > +	} else if (!strcmp("sata", location->name)) {
> > > > > > > +		ret = fs_set_blk_dev("sata", location-
> > > > > > > > 
> > > > > > > > devpart,
> > > > > > > FS_TYPE_ANY);
> > > > > > > +	} else if (!strcmp("ubi", location->name)) {
> > > > > > > +		if (location->ubivol != NULL)
> > > > > > > +			ret = fs_set_blk_dev("ubi", NULL,
> > > > > > > FS_TYPE_UBIFS);
> > > > > > > +		else
> > > > > > > +			ret = -ENODEV;
> > > > > > > +	} else {
> > > > > > > +		printf("Error: unsupported location
> > > > > > > storage.\n");
> > > > > > > +		return -ENODEV;
> > > > > > > +	}
> > > > > > > +
> > > > > > > +	if (ret)
> > > > > > > +		printf("Error: could not access
> > > > > > > storage.\n");
> > > > > > > +
> > > > > > > +	return ret;
> > > > > > > +}
> > > > > > > +
> > > > > > > +static int init_storage_device(struct device_location
> > > > > > > *location)
> > > > > > > +{
> > > > > > > +	int ret;
> > > > > > > +
> > > > > > > +	if (!strcmp("mmc", location->name)) {
> > > > > > > +		ret = init_mmc();
> > > > > > > +	} else if (!strcmp("sata", location->name)) {
> > > > > > > +		ret = init_storage_sata();
> > > > > > > +	} else if (location->ubivol != NULL) {
> > > > > > > +		ret = mount_ubifs(location);
> > > > > > > +#ifndef CONFIG_SPL_BUILD
> > > > > > > +	/* USB build is not supported yet in SPL */
> > > > > > > +	} else if (!strcmp("usb", location->name)) {
> > > > > > > +		ret = init_usb();
> > > > > > > +#endif
> > > > > > > +	} else {
> > > > > > > +		printf("Error: no supported storage device
> > > > > > > is
> > > > > > > available.\n");
> > > > > > > +		ret = -ENODEV;
> > > > > > > +	}
> > > > > > > +
> > > > > > > +	return ret;
> > > > > > > +}
> > > > > > > +
> > > > > > > +static void set_storage_devpart(char *name, char *devpart)
> > > > > > > +{
> > > > > > > +	size_t i;
> > > > > > > +
> > > > > > > +	for (i = 0; i < ARRAY_SIZE(default_locations);
> > > > > > > i++) {
> > > > > > > +		if (!strcmp(default_locations[i].name,
> > > > > > > name))
> > > > > > > +			default_locations[i].devpart =
> > > > > > > devpart;
> > > > > > > +	}
> > > > > > > +}
> > > > > > > +
> > > > > > > +/*
> > > > > > > + * Prepare firmware struct;
> > > > > > > + * return -ve if fail.
> > > > > > > + */
> > > > > > > +static int _request_firmware_prepare(struct firmware
> > > > > > > **firmware_p,
> > > > > > > +				     const char *name,
> > > > > > > void
> > > > > > > *dbuf,
> > > > > > > +				     size_t size, u32
> > > > > > > offset)
> > > > > > > +{
> > > > > > > +	struct firmware *firmware;
> > > > > > > +	struct firmware_priv *fw_priv;
> > > > > > > +
> > > > > > > +	*firmware_p = NULL;
> > > > > > > +
> > > > > > > +	if (!name || name[0] == '\0')
> > > > > > > +		return -EINVAL;
> > > > > > > +
> > > > > > > +	firmware = calloc(1, sizeof(*firmware));
> > > > > > > +	if (!firmware) {
> > > > > > > +		printf("%s: calloc(struct firmware)
> > > > > > > failed\n",
> > > > > > > __func__);
> > > > > > > +		return -ENOMEM;
> > > > > > > +	}
> > > > > > > +
> > > > > > > +	fw_priv = calloc(1, sizeof(*fw_priv));
> > > > > > > +	if (!fw_priv) {
> > > > > > > +		printf("%s: calloc(struct fw_priv)
> > > > > > > failed\n",
> > > > > > > __func__);
> > > > > > > +		free(firmware);
> > > > > > > +		return -ENOMEM;
> > > > > > > +	}
> > > > > > > +
> > > > > > > +	fw_priv->name = name;
> > > > > > > +	fw_priv->offset = offset;
> > > > > > > +	firmware->data = dbuf;
> > > > > > > +	firmware->size = size;
> > > > > > > +	firmware->priv = fw_priv;
> > > > > > > +	*firmware_p = firmware;
> > > > > > > +
> > > > > > > +	return 0;
> > > > > > > +}
> > > > > > > +
> > > > > > > +/*
> > > > > > > + * fw_get_filesystem_firmware - load firmware into an
> > > > > > > allocated
> > > > > > > buffer
> > > > > > > + * @location: An array of supported firmware location
> > > > > > > + * @firmware_p: pointer to firmware image
> > > > > > > + *
> > > > > > > + * @return: size of total read
> > > > > > > + *	    -ve when error
> > > > > > > + */
> > > > > > > +static int fw_get_filesystem_firmware(struct
> > > > > > > device_location
> > > > > > > *location,
> > > > > > > +				      struct firmware
> > > > > > > *firmware_p)
> > > > > > > +{
> > > > > > > +	struct firmware_priv *fw_priv = NULL;
> > > > > > > +	loff_t actread;
> > > > > > > +	char *dev_part;
> > > > > > > +	int ret;
> > > > > > > +
> > > > > > > +	dev_part = env_get("fw_dev_part");
> > > > > > > +	if (dev_part)
> > > > > > > +		set_storage_devpart(location->name,
> > > > > > > dev_part);
> > > > > > > +
> > > > > > > +	ret = init_storage_device(location);
> > > > > > > +	if (ret)
> > > > > > > +		goto out;
> > > > > > > +
> > > > > > > +	select_fs_dev(location);
> > > > > > > 
> > > > > > 'ret = ' is missing.
> > > > > > 
> > > > > Okay.
> > > > > > 
> > > > > > 
> > > > > > > 
> > > > > > > 
> > > > > > > 
> > > > > > > +	if (ret)
> > > > > > > +		goto out;
> > > > > > > +
> > > > > > > +	fw_priv = firmware_p->priv;
> > > > > > > +
> > > > > > > +	ret = fs_read(fw_priv->name, (ulong)firmware_p-
> > > > > > > >data,
> > > > > > > fw_priv->offset,
> > > > > > > +		     firmware_p->size, &actread);
> > > > > > > +
> > > > > > > +	if (ret) {
> > > > > > > +		printf("Error: %d Failed to read %s from
> > > > > > > flash
> > > > > > > %lld != %d.\n",
> > > > > > > +		      ret, fw_priv->name, actread,
> > > > > > > firmware_p-
> > > > > > > > 
> > > > > > > > 
> > > > > > > > size);
> > > > > > > +		return ret;
> > > > > > Y
> > > > > > Shouldn't this be 'goto out', do do the umount_ubifs() as in
> > > > > > all
> > > > > > other
> > > > > > error cases?
> > > > > > 
> > > > > okay.
> > > > > > 
> > > > > > 
> > > > > > > 
> > > > > > > 
> > > > > > > 
> > > > > > > +	} else {
> > > > > > > +		ret = actread;
> > > > > > > +	}
> > > > > > > 
> > > > > > What if actread != firmware_p->size?
> > > > > > You handled it as an error before, now you happily return the
> > > > > > number
> > > > > > of
> > > > > > bytes read in case of a short read.
> > > > > > Operation not permitted
> > > > > May be i misunderstand on ur previosly comment. There is not
> > > > > much
> > > > > 
> > > > My comment was about returning -EPERM in this case.
> > > > First of all you discarded the possible error code returned from
> > > > fs_read() and secondly -EIO would be more appropriate than EPERM
> > > > in
> > > > case of reading less data than expected. 
> > > > 
> > > One of the reason I prefer to use -EIO instead of return from
> > > fs_read,
> > > because i'm not sure when "actread != firmware_p->size", the return
> > > from the fs_read is still -ve or not.
> > > 
> > The return code of fs_read() could be for example -ENOENT, if the
> > file
> > you try to load does not exist. Replacing it with -EIO will mislead
> > users about the cause of the error.
> > Thus, if fs_read()'s return code is < 0, you should return that!
> > If it is zero and you decide to treat actread =! firmware_p->size as
> > an
> > error, you should return -EIO only in that case!
> > 
> Okay.
> > > 
> > > > 
> > > > > 
> > > > > 
> > > > > description on fd_read, i just knowing that -ve is error from
> > > > > function
> > > > > based on other example codes in U-Boot as they check the error
> > > > > with
> > > > > "ret < 0" or "ret != 0". "actread != firmware_p->size" is the
> > > > > additonal
> > > > > checking i added myself because when i digging to the code, i
> > > > > found
> > > > > that such condition only with error printf inside the function
> > > > > without
> > > > > any error code return. So, i am not sure when "actread !=
> > > > > firmware_p-
> > > > > > 
> > > > > > 
> > > > > > size", the ret would be -ve too.
> > > > If 'size' is the actual size of the caller provided buffer and
> > > > not
> > > > the
> > > > exact size of the firmware to be loaded, the condition actread !=
> > > > size
> > > > would not be an error at all.
> > > > If 'size' designates the exact size of the firmware, that the
> > > > caller
> > > > may know by other means, it is obviously an error if actread !=
> > > > size.
> > > > Since you treated actread != size as an error in your original
> > > > patch
> > > > I
> > > > assumed the latter case.
> > > > 
> > > The size here refer to the size of firmware or chunk of firmware to
> > > read when the buffer is smaller than firmware.
> > > 
> > OK. In that case, you should not print a message when actread !=
> > firmware_p->size, since fs_read will have already printed a message
> > due
> > to this.
> > 
> The printed message with additional info such as ret, size and actread.
> > > 
> > > > 
> > > > AFAICT Linux uses the first semantics ('size' specifies the size
> > > > of
> > > > the
> > > > caller provided buffer as an upper limit for the read() call and
> > > > not
> > > > the exact amount of data to be read).
> > > > Unfortunately the fs_read() semantics is fundamentally broken as
> > > > it
> > > > interprets 'size == 0' as 'read as much data as you can get'
> > > > which
> > > > may
> > > > easily lead to buffer overflows and prints a very unhelpful error
> > > > message "<filename> shorter than offset + len" in case 'size'
> > > > does
> > > > not
> > > > match the actual amount of data read. 
> > > > 
> > > > You should also consider what happens, if the firmware file you
> > > > read
> > > > is
> > > > larger than the buffer you provided, as in that case fs_read()
> > > > will
> > > > stop reading at the provided buffer size and you won't be able to
> > > > notice, that the file was not completely loaded!
> > > > 
> > > I prefer to let the caller handles this, because this function can
> > > be
> > > used to read the firmware chunk by chunk when the allocated buffer
> > > is
> > > smaller than firmware.
> > > 
> > So, the caller should also check, whether the return value of the
> > function matches the requested size and you should omit the special
> > handling of actread != firmware_p->size in
> > fw_get_filesystem_firmware()
> > (as you did in the current version of your patch).
> > 
> Just to confirm we are in same page, the caller i means is function
> which call the request_firmware_into_buf. firmware_p->size is the size
> gonna be read from firmware, could be part of firmware with different
> offset.
> Would you mind to explain more why the special handling should be
> omitted? This special handling would return error code if fail,or 
> actread if pass to the request_firmware_into_buf.
> request_firmware_into_buf just need to know status of file reading is
> performed successfully or not.
>
The point is whether to define a short read as failure and return an
error code in this case or just let the caller know the actual amount
of data read. If you always return the actual number of bytes read (or
a negative error code), no matter whether it matches the requested size
or not, the caller has all information to handle the situation
correctly. It could just request a new chunk for the missing bytes in
case of a short read, or handle a short read as an error condition at
its own discretion. So there is no need to check the actual read size
in your function.

But probably this discussion is completely vain, since if you want to
load a piece of firmware, you would probably check for the existence
of the file and its size beforehand, allocate a sufficiently large
buffer and then call request_firmware_into_buf() with the correct size.
Thus the situation actread != size should never occur anyway in
practice.


Lothar Waßmann