Re: [PATCH] ACP Modem (Mwave)

[Paul Schroeder]

Okay..  I cleaned things up...  I was more careful about the globals this
time...

Also...  The driver now builds as mwave.o instead of mwavedd.o...  The
driver now registers it's UART as a serial device (Thomas Hood)..  So there
is no need to run setserial anymore...

Cheers...Paul...

---
Paul B Schroeder  <[e-mail]>
Software Engineer, Linux Technology Center
IBM Corporation, Austin, TX

Alan Cox <[e-mail]> on 07/12/2001 02:55:00 PM

To:   Paul Schroeder/Austin/IBM@IBMUS
cc:   [e-mail], Mike Sullivan/Austin/IBM@IBMUS,
      [e-mail] (Alan Cox)
Subject:  Re: [PATCH] ACP Modem (Mwave)

> The patch has been updated...  The updates primarily consist of Alan's
> suggested changes below... (thank you)  It applies against the 2.4.6
> kernel...

A quick glance through it:

dsp3780I_WriteDStore still touches user space with a spinlock held
(also doesnt check the get_user return)

The ioctl handlers do not check copy_from_user/to_user returns

IOCTL_MW_UNREGISTER_IPC will oops if fed bogus info (ipcnum should be
unsigned)

The return should be -ENOTTY not -ENOIOCTLCMD  unless its internal code
that catches NOIOCTLCMD and changes it before user space sees it

mwave_Read should be -EINVAL not -ENOSYS (ENOSYS means the entire read
syscall
in the OS isnt there)

In debug mode mwave_write accesses user space directly and may crash
(buf[0])

Trivial item - coding style uses foo(void) not foo() to indicate functions
taking no arguments

Still have globals like "dspio" "uartio"  "ClaimResources" etc

whats wrong with tp3780_uart_io etc for globals ?

Otherwise it looks close to ready

Alan