bcmwl 6.30.223.271+bdcom build fails with kernel 5.8.0-34/36 Ubuntu 20.04.1 LTS

[Paul Wilson]

Ubuntu 20.04.1 LTS
bcmwl-kernel-source_6.30.223.271

I believe the important part of the /var/lib/dkms/bcmwl/6.30.223.271+bdcom/build/make.log is below.

/var/lib/dkms/bcmwl/6.30.223.271+bdcom/build/src/wl/sys/wl_linux.c: In function ‘wl_reg_proc_entry’:
/var/lib/dkms/bcmwl/6.30.223.271+bdcom/build/src/wl/sys/wl_linux.c:3376:58: error: passing argument 4 of ‘proc_create_data’ from incompatible pointe
r type [-Werror=incompatible-pointer-types]
 3376 | if ((wl->proc_entry = proc_create_data(tmp, 0644, NULL, &wl_fops, wl)) == NULL) {
      | ^~~~~~~~
      | |
      | const struct file_operations *
In file included from /var/lib/dkms/bcmwl/6.30.223.271+bdcom/build/src/wl/sys/wl_linux.c:38:
./include/linux/proc_fs.h:102:31: note: expected ‘const struct proc_ops *’ but argument is of type ‘const struct file_operations *’
  102 | extern struct proc_dir_entry *proc_create_data(const char *, umode_t,
      | ^~~~~~~~~~~~~~~~

It looks like /var/lib/dkms/bcmwl/6.30.223.271+bdcom/build/src/wl/sys/wl_linux.c uses the file_operations structure.

lines 3359-3381 from wl_linux.c

#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 10, 0)
static const struct file_operations wl_fops = {
 .owner = THIS_MODULE,
 .read = wl_proc_read,
 .write = wl_proc_write,
};
#endif

static int
wl_reg_proc_entry(wl_info_t *wl)
{
 char tmp[32];
 sprintf(tmp, "%s%d", HYBRID_PROC, wl->pub->unit);
#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 10, 0)
 if ((wl->proc_entry = create_proc_entry(tmp, 0644, NULL)) == NULL) {
  WL_ERROR(("%s: create_proc_entry %s failed\n", __FUNCTION__, tmp));
#else
 if ((wl->proc_entry = proc_create_data(tmp, 0644, NULL, &wl_fops, wl)) == NULL) {
  WL_ERROR(("%s: proc_create_data %s failed\n", __FUNCTION__, tmp));
#endif
  ASSERT(0);
  return -1;
 }

However, it looks like proc_fs.h uses a different structure (proc_ops).

/usr/src/linux-headers-5.8.0-34-generic/include/linux/proc_fs.h
/usr/src/linux-hwe-5.8-headers-5.8.0-34/include/linux/proc_fs.h

kernel 5.8.0-34
proc_fs.h
lines 29-43

struct proc_ops {
 unsigned int proc_flags;
 int (*proc_open)(struct inode *, struct file *);
 ssize_t (*proc_read)(struct file *, char __user *, size_t, loff_t *);
 ssize_t (*proc_write)(struct file *, const char __user *, size_t, loff_t *);
 loff_t (*proc_lseek)(struct file *, loff_t, int);
 int (*proc_release)(struct inode *, struct file *);
 __poll_t (*proc_poll)(struct file *, struct poll_table_struct *);
 long (*proc_ioctl)(struct file *, unsigned int, unsigned long);
#ifdef CONFIG_COMPAT
 long (*proc_compat_ioctl)(struct file *, unsigned int, unsigned long);
#endif
 int (*proc_mmap)(struct file *, struct vm_area_struct *);
 unsigned long (*proc_get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
} __randomize_layout;

and lines 102-107

extern struct proc_dir_entry *proc_create_data(const char *, umode_t,
            struct proc_dir_entry *,
            const struct proc_ops *,
            void *);

struct proc_dir_entry *proc_create(const char *name, umode_t mode, struct proc_dir_entry *parent, const struct proc_ops *proc_ops);

Looking back at the linux 5.4.0-59 headers, proc_fs.h appears to use the file_operations structure.

/usr/src/linux-headers-5.4.0-59-generic/include/linux/proc_fs.h
/usr/src/linux-headers-5.4.0-59/include/linux/proc_fs.h

kernel 5.4.0-59
proc_fs.h
lines 44-49

extern struct proc_dir_entry *proc_create_data(const char *, umode_t,
            struct proc_dir_entry *,
            const struct file_operations *,
            void *);

struct proc_dir_entry *proc_create(const char *name, umode_t mode, struct proc_dir_entry *parent, const struct file_operations *proc_fops);

If this is where the problem lies, I am guessing it is better to try to fix the Broadcom package/ wl_linux.c than the Linux 5.8.0-34 headers?

The information above, was prompted by:

sudo apt-get install --reinstall bcmwl-kernel-source
...
Building initial module for 5.8.0-34-generic
ERROR: Cannot create report: [Errno 17] File exists: '/var/crash/bcmwl-kernel-source.0.crash'
Error! Bad return status for module build on kernel: 5.8.0-34-generic (x86_64)
Consult /var/lib/dkms/bcmwl/6.30.223.271+bdcom/build/make.log for more information.
dpkg: error processing package bcmwl-kernel-source (--configure):
 installed bcmwl-kernel-source package post-installation script subprocess returned error exit status 10
Errors were encountered while processing:
 bcmwl-kernel-source
E: Sub-process /usr/bin/dpkg returned an error code (1)

[Paul Wilson (paulwil)]

From what I have been reading, it sounds like linux is converting to proc_ops.

From this, I am guessing lines 3360-3364 in wl_linux.c will need to changed from using the file_operations structure to the proc_ops structure.

I do realize this is easier said than done.

[Paul Wilson (paulwil)]

Looking at the source code for the broadcom-sta package, which is working, I am wondering if

lines 3359-3381 from wl_linux.c in the bcmwl package

#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 10, 0)
static const struct file_operations wl_fops = {
 .owner = THIS_MODULE,
 .read = wl_proc_read,
 .write = wl_proc_write,
};
#endif

should be updated to something similar to what is in the broadcom-sta package in order to convert to using proc_ops that is found in the new 5.8.0-34/36 headers.

lines 3359-3381 from wl_linux.c in the broadcom-sta package

#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 10, 0)
#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 6, 0)
static const struct proc_ops wl_fops = {
 .proc_read = wl_proc_read,
 .proc_write = wl_proc_write,
};
#else
static const struct file_operations wl_fops = {
 .owner = THIS_MODULE,
 .read = wl_proc_read,
 .write = wl_proc_write,
};
#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(5, 6, 0) */
#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(3, 10, 0) */

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[ZdravkoG (zdravko-g)]

Hi @paulwil,
Your bug seems duplicate of Bug #1910555. Take a look there for workaround till Canonical fix gets released. 😉 Just a obsolete function ioremap_nocache (and one more thing) together with delayed new release of bcmwl-kernel-source (who knows why), where new patch has added, dealing with the changes.
Hope this helps.

[Paul Wilson (paulwil)]

Hi @zdravko-g

Thank you for the update. I have temporarily removed bcmwl-kernel-source and have installed broadcom-sta until the new release comes out. WiFi is currently running.

It will be interesting to how the file_operations to proc_ops conversion is dealt with in the new release (wl_linux.c) or if I am chasing the wrong rabbit.