Date: Tue, 22 Nov 2005 07:00:56 GMT From: Arthur Hartwig <arthur.hartwig@nokia.com> To: freebsd-i386@FreeBSD.org Subject: Re: i386/89296: Spurious atapci1: failed to enable memory mapping! on ICH7 Message-ID: <200511220700.jAM70uHP068304@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
The following reply was made to PR i386/89296; it has been noted by GNATS.
From: Arthur Hartwig <arthur.hartwig@nokia.com>
To: bug-followup@FreeBSD.org, Francis.Dupont@enst-bretagne.fr
Cc:
Subject: Re: i386/89296: Spurious atapci1: failed to enable memory mapping!
on ICH7
Date: Tue, 22 Nov 2005 16:51:40 +1000
I also saw this on my Gigabyte GA-*i945P-G motherboard which has an ICH7.
The message is printed as a result of a bus_alloc_resource_any() call in
ata_intel_chipinit() in ata-chipset.c. The offending call is
/* SATA parts can be either compat or AHCI */
else {
/* if we have BAR(5) as a memory resource we should use AHCI mode */
ctlr->r_type2 = SYS_RES_MEMORY;
ctlr->r_rid2 =PCIR_BAR(5);
if ((ctlr->r_res2 = bus_alloc_resource_any(dev, ctlr->r_type2,
&ctlr->r_rid2,
RF_ACTIVE))) {
The bus_alloc_resource_any() call winds up in pci_alloc_resource() in
dev/pci/pci.c which sees the RF_ACTIVE flag is set and triies to enable
memory space access by calling PCI_ENABLE_IO() to set the appropriate
bit in the PCIR_COMMAND register of the ATA controller. Unfortunately
the memory space enable bit in the ICH7 ATA controller on my motherboard
is hardwired to 0 unless the SCRAE bit is one. So the ATA driver could
be changed to stop this message appearing but I think the logic in
pci_alloc_resource() is flawed - it should not be attempting to enable
memory or i/o access UNTIL it has successfully allocated the requested
resource. Suggested fix:
change pci_alloc_resource() to read as follows:
struct resource *
pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
u_long start, u_long end, u_long count, u_int flags)
{
struct pci_devinfo *dinfo = device_get_ivars(child);
struct resource_list *rl = &dinfo->resources;
struct resource_list_entry *rle;
pcicfgregs *cfg = &dinfo->cfg;
/*
* Perform lazy resource allocation
*/
if (device_get_parent(child) == dev) {
switch (type) {
case SYS_RES_IRQ:
/*
* If the child device doesn't have an
* interrupt routed and is deserving of an
* interrupt, try to assign it one.
*/
if (!PCI_INTERRUPT_VALID(cfg->intline) &&
(cfg->intpin != 0))
pci_assign_interrupt(dev, child, 0);
break;
case SYS_RES_IOPORT:
case SYS_RES_MEMORY:
rle = resource_list_find(rl, type, *rid);
if (rle == NULL)
return (pci_alloc_map(dev, child, type, rid,
start, end, count, flags));
break;
}
/*
* If we've already allocated the resource, then
* return it now. But first we may need to activate
* it, since we don't allocate the resource as active
* above. Normally this would be done down in the
* nexus, but since we short-circuit that path we have
* to do its job here. Not sure if we should free the
* resource if it fails to activate.
*/
rle = resource_list_find(rl, type, *rid);
if (rle != NULL && rle->res != NULL) {
if (bootverbose)
device_printf(child,
"Reserved %#lx bytes for rid %#x type %d at
%#lx\n",
rman_get_size(rle->res), *rid, type,
rman_get_start(rle->res));
/*
* Enable IO mode only if the matching resource is
* successfully activated else the PCI_ENABLE_IO
might
* fail and report the failre when it doesn't
matter.
*/
if (flags & RF_ACTIVE) {
if (bus_generic_activate_resource(dev,
child, type,
*rid, rle->res) != 0)
return NULL;
else
PCI_ENABLE_IO(dev, child, type);
}
return (rle->res);
}
}
return (resource_list_alloc(rl, dev, child, type, rid,
start, end, count, flags));
}
I built a kernel with this change, booted it and the memory mapping
failure message was gone. All devices seemed to work but I was left
wondering if the return in the SYS_RES_MEMORY and SYS_RES_IOPORT case
also needs a call to bus_generic_activate_resource() and PCI_ENABLE_IO().
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200511220700.jAM70uHP068304>