RE: [PATCH] megaraid driver fix for 2.5.70

[James Bottomley]

On Fri, 2003-06-06 at 09:28, Mukker, Atul wrote:
> Coming back to main issue, declaring complete mailbox would be superfluous
> since driver uses 16 bytes at most. The following patch should fix the panic
> 
>  	mbox = (mbox_t *)raw_mbox;
>  
> -	memset(mbox, 0, sizeof(*mbox));
> +	memset(mbox, 0, 16);
>  
>  	memset((void *)adapter->mega_buffer, 0, MEGA_BUFFER_SIZE);
>  

This, I think, is a bad idea.  It looks intrinsically wrong to allocate
storage and assign a pointer to it of a type that is longer than the
allocated storage.  The initial buffer overrun was due to problems with
this.

I think the correct solution is to define your mailbox like this:

typedef struct {
	/* 0x0 */ u8 cmd;
	/* 0x1 */ u8 cmdid;
	/* 0x2 */ u16 numsectors;
	/* 0x4 */ u32 lba;
	/* 0x8 */ u32 xferaddr;
	/* 0xC */ u8 logdrv;
	/* 0xD */ u8 numsgelements;
	/* 0xE */ u8 resvd;
	/* 0xF */ volatile u8 busy;
} __attribute__ ((packed)) user_mbox_t;

typedef struct {
	user_mbox_t mbox_out
	/* 0x10 */ volatile u8 numstatus;
	/* 0x11 */ volatile u8 status;
	/* 0x12 */ volatile u8 completed[MAX_FIRMWARE_STATUS];
	volatile u8 poll;
	volatile u8 ack;
} __attribute__ ((packed)) mbox_t;

and then re-define the issue_scb..() routines to use user_mbox_t which
is always the correct size.

Thus, you can throw away the raw_mbox and just do

user_mbox_t mbox;
memset(&mbox, 0, sizeof(mbox));

of course, your ->busy references become ->mbox_out.busy.

James