* [PATCH] move dma_consistent_dma_mask to the generic device
@ 2004-02-26 1:34 James Bottomley
2004-02-26 2:06 ` Andrew Morton
2004-03-03 15:29 ` Jes Sorensen
0 siblings, 2 replies; 4+ messages in thread
From: James Bottomley @ 2004-02-26 1:34 UTC (permalink / raw)
To: Andrew Morton, Linus Torvalds, willy, Jes Sorensen; +Cc: Linux Kernel
pci_dev.consistent_dma_mask was introduced to get around problems in the
IA64 Altix machine.
Now, we have a use for it in x86: the aacraid needs coherent memory in a
31 bit address range (2GB). Unfortunately, x86 is converted to the dma
model, so it can't see the pci_dev by the time coherent memory is
allocated.
The solution to all of this is to move pci_dev.consistent_dma_mask to
dev.coherent_dma_mask and make x86 use it in the dma_alloc_coherent()
calls.
This should allow me to make the aacraid set the coherent mask instead
of using it's current dma_mask juggling.
I copied Matthew because parisc also has a parisc_device that needed to
be converted over.
James
===== arch/i386/kernel/pci-dma.c 1.11 vs edited =====
--- 1.11/arch/i386/kernel/pci-dma.c Mon Jan 13 10:28:47 2003
+++ edited/arch/i386/kernel/pci-dma.c Wed Feb 25 14:58:56 2004
@@ -20,8 +20,10 @@
/* ignore region specifiers */
gfp &= ~(__GFP_DMA | __GFP_HIGHMEM);
- if (dev == NULL || (*dev->dma_mask < 0xffffffff))
+ if (dev == NULL || (dev->coherent_dma_mask < 0xffffffff)) {
+ printk("COHERENT MASK 0x%llx\n", dev->coherent_dma_mask);
gfp |= GFP_DMA;
+ }
ret = (void *)__get_free_pages(gfp, get_order(size));
if (ret != NULL) {
===== arch/parisc/kernel/drivers.c 1.10 vs edited =====
--- 1.10/arch/parisc/kernel/drivers.c Fri Feb 6 08:23:39 2004
+++ edited/arch/parisc/kernel/drivers.c Wed Feb 25 15:11:50 2004
@@ -618,6 +618,7 @@
tmp1);
/* make the generic dma mask a pointer to the parisc one */
dev->dev.dma_mask = &dev->dma_mask;
+ dev->dev.coherent_dma_mask = dev->dma_mask;
pr_debug("device_register(%s)\n", dev->dev.bus_id);
device_register(&dev->dev);
}
===== arch/parisc/kernel/pci-dma.c 1.7 vs edited =====
--- 1.7/arch/parisc/kernel/pci-dma.c Tue Feb 3 23:41:56 2004
+++ edited/arch/parisc/kernel/pci-dma.c Wed Feb 25 15:15:00 2004
@@ -372,7 +372,7 @@
** ISA cards will certainly only support 24-bit DMA addressing.
** Not clear if we can, want, or need to support ISA.
*/
- if (!dev || *dev->dma_mask != 0xffffffff)
+ if (!dev || *dev->coherent_dma_mask < 0xffffffff)
gfp |= GFP_DMA;
#endif
return (void *)vaddr;
===== drivers/eisa/eisa-bus.c 1.14 vs edited =====
--- 1.14/drivers/eisa/eisa-bus.c Sun Oct 5 03:07:57 2003
+++ edited/drivers/eisa/eisa-bus.c Wed Feb 25 14:58:57 2004
@@ -187,6 +187,7 @@
edev->dev.parent = root->dev;
edev->dev.bus = &eisa_bus_type;
edev->dev.dma_mask = &edev->dma_mask;
+ edev->dev.coherent_dma_mask = edev->dma_mask;
sprintf (edev->dev.bus_id, "%02X:%02X", root->bus_nr, slot);
for (i = 0; i < EISA_MAX_RESOURCES; i++) {
===== drivers/mca/mca-bus.c 1.5 vs edited =====
--- 1.5/drivers/mca/mca-bus.c Sun Aug 17 13:44:15 2003
+++ edited/drivers/mca/mca-bus.c Wed Feb 25 14:58:58 2004
@@ -106,6 +106,7 @@
sprintf (mca_dev->dev.bus_id, "%02d:%02X", bus, mca_dev->slot);
mca_dev->dma_mask = mca_bus->default_dma_mask;
mca_dev->dev.dma_mask = &mca_dev->dma_mask;
+ mca_dev->dev.coherent_dma_mask = mca_dev->dma_mask;
if (device_register(&mca_dev->dev))
return 0;
===== drivers/pci/pci.c 1.61 vs edited =====
--- 1.61/drivers/pci/pci.c Tue Feb 10 12:01:13 2004
+++ edited/drivers/pci/pci.c Wed Feb 25 14:58:58 2004
@@ -691,7 +691,7 @@
if (!pci_dma_supported(dev, mask))
return -EIO;
- dev->consistent_dma_mask = mask;
+ dev->dev.coherent_dma_mask = mask;
return 0;
}
===== drivers/pci/probe.c 1.59 vs edited =====
--- 1.59/drivers/pci/probe.c Wed Feb 18 21:42:58 2004
+++ edited/drivers/pci/probe.c Wed Feb 25 14:59:00 2004
@@ -566,7 +566,6 @@
/* Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer)
set this higher, assuming the system even supports it. */
dev->dma_mask = 0xffffffff;
- dev->consistent_dma_mask = 0xffffffff;
if (pci_setup_device(dev) < 0) {
kfree(dev);
return NULL;
@@ -578,6 +577,7 @@
pci_name_device(dev);
dev->dev.dma_mask = &dev->dma_mask;
+ dev->dev.coherent_dma_mask = 0xffffffffull;
return dev;
}
===== include/linux/device.h 1.115 vs edited =====
--- 1.115/include/linux/device.h Mon Feb 9 17:40:25 2004
+++ edited/include/linux/device.h Wed Feb 25 14:59:35 2004
@@ -285,6 +285,12 @@
detached from its driver. */
u64 *dma_mask; /* dma mask (if dma'able device) */
+ u64 coherent_dma_mask;/* Like dma_mask, but for
+ alloc_coherent mappings as
+ not all hardware supports
+ 64 bit addresses for consistent
+ allocations such descriptors. */
+
struct list_head dma_pools; /* dma pools (if dma'ble) */
void (*release)(struct device * dev);
===== include/linux/pci.h 1.117 vs edited =====
--- 1.117/include/linux/pci.h Tue Feb 10 11:51:08 2004
+++ edited/include/linux/pci.h Wed Feb 25 14:59:03 2004
@@ -392,11 +392,6 @@
this if your device has broken DMA
or supports 64-bit transfers. */
- u64 consistent_dma_mask;/* Like dma_mask, but for
- pci_alloc_consistent mappings as
- not all hardware supports
- 64 bit addresses for consistent
- allocations such descriptors. */
u32 current_state; /* Current operating state. In ACPI-speak,
this is D0-D3, D0 being fully functional,
and D3 being off. */
^ permalink raw reply [flat|nested] 4+ messages in thread* Re: [PATCH] move dma_consistent_dma_mask to the generic device
2004-02-26 1:34 [PATCH] move dma_consistent_dma_mask to the generic device James Bottomley
@ 2004-02-26 2:06 ` Andrew Morton
2004-02-26 2:55 ` James Bottomley
2004-03-03 15:29 ` Jes Sorensen
1 sibling, 1 reply; 4+ messages in thread
From: Andrew Morton @ 2004-02-26 2:06 UTC (permalink / raw)
To: James Bottomley; +Cc: torvalds, willy, jes, linux-kernel
James Bottomley <James.Bottomley@SteelEye.com> wrote:
>
> pci_dev.consistent_dma_mask was introduced to get around problems in the
> IA64 Altix machine.
>
> Now, we have a use for it in x86: the aacraid needs coherent memory in a
> 31 bit address range (2GB). Unfortunately, x86 is converted to the dma
> model, so it can't see the pci_dev by the time coherent memory is
> allocated.
>
> The solution to all of this is to move pci_dev.consistent_dma_mask to
> dev.coherent_dma_mask and make x86 use it in the dma_alloc_coherent()
> calls.
A bit more grepping is needed.
sound/core/memalloc.c: In function `snd_pci_hack_alloc_consistent':
sound/core/memalloc.c:103: structure has no member named `consistent_dma_mask'
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] move dma_consistent_dma_mask to the generic device
2004-02-26 2:06 ` Andrew Morton
@ 2004-02-26 2:55 ` James Bottomley
0 siblings, 0 replies; 4+ messages in thread
From: James Bottomley @ 2004-02-26 2:55 UTC (permalink / raw)
To: Andrew Morton; +Cc: Linus Torvalds, willy, Jes Sorensen, Linux Kernel
On Wed, 2004-02-25 at 20:06, Andrew Morton wrote:
> A bit more grepping is needed.
>
> sound/core/memalloc.c: In function `snd_pci_hack_alloc_consistent':
> sound/core/memalloc.c:103: structure has no member named `consistent_dma_mask'
Mea culpa, I don't have any boxes that define this hack, so I didn't see
it. This sound memalloc is horribly illegal under the API...I suppose
we're just lucky the Altix doesn't have a sound card ;-)
This incremental patch should fix it up (I compiled it on parisc by
including it in the hacked architectures).
James
===== sound/core/memalloc.c 1.16 vs edited =====
--- 1.16/sound/core/memalloc.c Mon Jan 19 04:37:35 2004
+++ edited/sound/core/memalloc.c Wed Feb 25 20:51:28 2004
@@ -100,13 +100,13 @@
if (hwdev == NULL)
return pci_alloc_consistent(hwdev, size, dma_handle);
dma_mask = hwdev->dma_mask;
- cdma_mask = hwdev->consistent_dma_mask;
+ cdma_mask = hwdev->dev.coherent_dma_mask;
mask = (unsigned long)dma_mask && (unsigned long)cdma_mask;
hwdev->dma_mask = 0xffffffff; /* do without masking */
- hwdev->consistent_dma_mask = 0xffffffff; /* do without masking */
+ hwdev->dev.coherent_dma_mask = 0xffffffff; /* do without masking */
ret = pci_alloc_consistent(hwdev, size, dma_handle);
hwdev->dma_mask = dma_mask; /* restore */
- hwdev->consistent_dma_mask = cdma_mask; /* restore */
+ hwdev->dev.coherent_dma_mask = cdma_mask; /* restore */
if (ret) {
/* obtained address is out of range? */
if (((unsigned long)*dma_handle + size - 1) & ~mask) {
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] move dma_consistent_dma_mask to the generic device
2004-02-26 1:34 [PATCH] move dma_consistent_dma_mask to the generic device James Bottomley
2004-02-26 2:06 ` Andrew Morton
@ 2004-03-03 15:29 ` Jes Sorensen
1 sibling, 0 replies; 4+ messages in thread
From: Jes Sorensen @ 2004-03-03 15:29 UTC (permalink / raw)
To: James Bottomley; +Cc: Andrew Morton, Linus Torvalds, willy, Linux Kernel
>>>>> "James" == James Bottomley <James.Bottomley@steeleye.com> writes:
James> pci_dev.consistent_dma_mask was introduced to get around
James> problems in the IA64 Altix machine.
James> Now, we have a use for it in x86: the aacraid needs coherent
James> memory in a 31 bit address range (2GB). Unfortunately, x86 is
James> converted to the dma model, so it can't see the pci_dev by the
James> time coherent memory is allocated.
James,
Could you add this one to your patchset, it fixes the sn2 code to work
with the new location of the mask.
Thanks,
Jes
--- arch/ia64/sn/io/machvec/pci_dma.c~ Wed Mar 3 06:47:36 2004
+++ arch/ia64/sn/io/machvec/pci_dma.c Wed Mar 3 07:23:34 2004
@@ -152,7 +152,7 @@
* pcibr_dmatrans_addr ignores a missing PCIIO_DMA_A64 flag on
* PCI-X buses.
*/
- if (hwdev->consistent_dma_mask == ~0UL)
+ if (hwdev->dev.coherent_dma_mask == ~0UL)
*dma_handle = pcibr_dmatrans_addr(vhdl, NULL, phys_addr, size,
PCIIO_DMA_CMD | PCIIO_DMA_A64);
else {
@@ -169,7 +169,7 @@
}
}
- if (!*dma_handle || *dma_handle > hwdev->consistent_dma_mask) {
+ if (!*dma_handle || *dma_handle > hwdev->dev.coherent_dma_mask) {
if (dma_map) {
pcibr_dmamap_done(dma_map);
pcibr_dmamap_free(dma_map);
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2004-03-03 15:33 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2004-02-26 1:34 [PATCH] move dma_consistent_dma_mask to the generic device James Bottomley
2004-02-26 2:06 ` Andrew Morton
2004-02-26 2:55 ` James Bottomley
2004-03-03 15:29 ` Jes Sorensen
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox