arm: Error handling for cacheflush

-----Original Message-----
From: Will Deacon [mailto:will.deacon at arm.com]
Sent: Thursday, November 13, 2014 3:04 AM
To: Chanho Min
Cc: '???'; 'Jongsung Kim'; 'HyoJun Im'
Subject: Re: arm: Error handling for cacheflush

On Wed, Nov 12, 2014 at 10:38:28AM +0000, Chanho Min wrote:

quoted

Dear Will Deacon,

Hello,

quoted

Since commit 28256d612726 ("ARM: cacheflush: split user cache-flushing
into interruptible chunks"), cacheflush can return -EINTR if it is interrupted
by signal and process has user-defined signal handler.

But, most user of cacheflush haven't handled -EINTR and no man page or guide
is not founded. So, Would you check that cacheflush's behavior is changed to
to as the follows or you have another intention?

int cacheflush(void *addr, size_t nbytes, int cache)
{
        int err;
        do err = syscall(__ARM_NR_cacheflush, addr, addr + nbytes, cache);
        while (err < 0 && errno == EINTR);
        return err;
}

Well, the system call could always have failed, so applications should be
checking for that anyway. It's true that -EINTR is a new error code for
this system call, but it was needed to prevent the machine being stalled
from userspace for large ranges.

I fully understand the intention of this patch.
But, In this case, cacheflush should be re-invoked from the start of address.
(restart_syscall is not available in userspace)
So Handling of -EINTR may cause regression.

I don't know of any manpage for the ARM cacheflush system call, as it's
private to the architecture.

Will

Chanho