cvs commit: src/sys/alpha/isa isa_dma.c src/sys/amd64/isa isa_dma.c src/sys/i386/isa isa_dma.c src/sys/ia64/isa isa_dma.c src/sys/pc98/pc98 isa_dma.c src/sys/isa isavar.h src/sys/vm vm_page.c

Brian Fundakowski Feldman green at freebsd.org
Wed Sep 15 13:06:39 PDT 2004


On Wed, Sep 15, 2004 at 01:49:18PM -0600, M. Warner Losh wrote:
> In message: <20040915152822.GD36708 at green.homeunix.org>
>             Brian Fundakowski Feldman <green at freebsd.org> writes:
> : On Wed, Sep 15, 2004 at 12:09:51PM +0000, Poul-Henning Kamp wrote:
> : > phk         2004-09-15 12:09:51 UTC
> : > 
> : >   FreeBSD src repository
> : > 
> : >   Modified files:
> : >     sys/alpha/isa        isa_dma.c 
> : >     sys/amd64/isa        isa_dma.c 
> : >     sys/i386/isa         isa_dma.c 
> : >     sys/ia64/isa         isa_dma.c 
> : >     sys/pc98/pc98        isa_dma.c 
> : >     sys/isa              isavar.h 
> : >     sys/vm               vm_page.c 
> : >   Log:
> : >   Add new a function isa_dma_init() which returns an errno when it fails
> : >   and which takes a M_WAITOK/M_NOWAIT flag argument.
> : >   
> : >   Add compatibility isa_dmainit() macro which whines loudly if
> : >   isa_dma_init() fails.
> : >   
> : >   Problem uncovered by:   tegge
> : 
> : David Yeske is running into a problem with it failing in fdc(4).  I am not
> : sure how 40KB of memory is too big for it to allocate... but I think it
> : might be appropriate to try some sort of fallback.
> : 
> : --- /usr/src/sys/dev/fdc/fdc.c	Fri Aug 27 19:15:22 2004
> : +++ tmp/fdc.c	Tue Sep  7 21:44:26 2004
> : @@ -1334,7 +1334,11 @@
> :  		if (!(fdc->flags & FDC_NODMA) && fdc->dmacnt++ == 0) {
> :  			mtx_lock(&Giant);
> :  			isa_dma_acquire(fdc->dmachan);
> : -			isa_dmainit(fdc->dmachan, MAX_BYTES_PER_CYL);
> : +			if (isa_dmainit(fdc->dmachan, MAX_BYTES_PER_CYL) != 0) {
> : +				isa_dma_release(fdc->dmachan);
> : +				fdc->dmacnt--;
> : +				fdc->flags |= FDC_NODMA;
> : +			}
> :  			mtx_unlock(&Giant);
> :  		}
> :  	}
> 
> I have patches, improved by Tor and phk, that move the acquisision of
> the dma buffer to attach.
> 
> The problem is that 40k isn't too big, but 40k below 16MB is very hard
> after boon on large memory machines due to the amount of RAM allocated
> in the boot process.

Well, 40kb below 16MB _during_boot_ shouldn't be too hard, certainly
more so with large memory machines though, indeed...

-- 
Brian Fundakowski Feldman                           \'[ FreeBSD ]''''''''''\
  <> green at FreeBSD.org                               \  The Power to Serve! \
 Opinions expressed are my own.                       \,,,,,,,,,,,,,,,,,,,,,,\


More information about the cvs-src mailing list