[PATCH] AGPGART 3/5: support arches where CPU doesn't go through GART

Bjorn Helgaas (helgaas@fc.hp.com)
Fri, 27 Sep 2002 22:34:54 -0600


On some architectures, CPU accesses aren't translated by the GART.
Currently only IA64 ZX1 and 460GX fall into this category, and they
set "cant_use_aperture = 1". This patch adds infrastructure to
support them. Behavior when "cant_use_aperture == 0" is unchanged.

# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher.
# This patch includes the following deltas:
# ChangeSet 1.685 -> 1.686
# drivers/char/agp/agpgart_be.c 1.37 -> 1.38
# include/linux/agp_backend.h 1.13 -> 1.14
#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 02/09/19 bjorn_helgaas@hp.com 1.686
# Add support for AGP bridges where CPU accesses don't go
# through the GART aperture.
# --------------------------------------------
#
diff -Nru a/drivers/char/agp/agpgart_be.c b/drivers/char/agp/agpgart_be.c
--- a/drivers/char/agp/agpgart_be.c Thu Sep 19 13:15:48 2002
+++ b/drivers/char/agp/agpgart_be.c Thu Sep 19 13:15:48 2002
@@ -43,6 +43,9 @@
#include <asm/uaccess.h>
#include <asm/io.h>
#include <asm/page.h>
+#include <asm/pgalloc.h>
+#include <asm/pgtable.h>
+#include <asm/smplock.h>

#include <linux/agp_backend.h>
#include "agp.h"
@@ -207,10 +210,14 @@
agp_bridge.free_by_type(curr);
return;
}
- if (curr->page_count != 0) {
- for (i = 0; i < curr->page_count; i++) {
- agp_bridge.agp_destroy_page((unsigned long)
- phys_to_virt(curr->memory[i]));
+ if (agp_bridge.cant_use_aperture) {
+ vfree(curr->vmptr);
+ } else {
+ if (curr->page_count != 0) {
+ for (i = 0; i < curr->page_count; i++) {
+ agp_bridge.agp_destroy_page((unsigned long)
+ phys_to_virt(curr->memory[i]));
+ }
}
}
agp_free_key(curr->key);
@@ -219,6 +226,37 @@
MOD_DEC_USE_COUNT;
}

+#define IN_VMALLOC(_x) (((_x) >= VMALLOC_START) && ((_x) < VMALLOC_END))
+
+/*
+ * Look up and return the pte corresponding to addr. We only do this for
+ * agp_ioremap'ed addresses.
+ */
+static pte_t *agp_lookup_pte(unsigned long addr)
+{
+ pgd_t *dir;
+ pmd_t *pmd;
+ pte_t *pte;
+
+ if (!IN_VMALLOC(addr))
+ return NULL;
+
+ dir = pgd_offset_k(addr);
+ pmd = pmd_offset(dir, addr);
+
+ if (pmd) {
+ pte = pte_offset(pmd, addr);
+
+ if (pte) {
+ return pte;
+ } else {
+ return NULL;
+ }
+ } else {
+ return NULL;
+ }
+}
+
#define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))

agp_memory *agp_allocate_memory(size_t page_count, u32 type)
@@ -253,18 +291,44 @@
MOD_DEC_USE_COUNT;
return NULL;
}
- for (i = 0; i < page_count; i++) {
- new->memory[i] = agp_bridge.agp_alloc_page();

- if (new->memory[i] == 0) {
- /* Free this structure */
- agp_free_memory(new);
+ if (agp_bridge.cant_use_aperture) {
+ void *vmblock;
+ unsigned long vaddr;
+ pte_t *pte;
+
+ vmblock = __vmalloc(page_count << PAGE_SHIFT, GFP_KERNEL, PAGE_KERNEL);
+ if (vmblock == NULL) {
+ MOD_DEC_USE_COUNT;
return NULL;
}
- new->memory[i] = virt_to_phys((void *) new->memory[i]);
- new->page_count++;
- }

+ new->vmptr = vmblock;
+ vaddr = (unsigned long) vmblock;
+
+ for (i = 0; i < page_count; i++, vaddr += PAGE_SIZE) {
+ pte = agp_lookup_pte(vaddr);
+ if (pte == NULL) {
+ MOD_DEC_USE_COUNT;
+ return NULL;
+ }
+ new->memory[i] = virt_to_phys(page_address(pte_page(*pte)));
+ }
+
+ new->page_count = page_count;
+ } else {
+ for (i = 0; i < page_count; i++) {
+ new->memory[i] = agp_bridge.agp_alloc_page();
+
+ if (new->memory[i] == 0) {
+ /* Free this structure */
+ agp_free_memory(new);
+ return NULL;
+ }
+ new->memory[i] = virt_to_phys((void *) new->memory[i]);
+ new->page_count++;
+ }
+ }
return new;
}

diff -Nru a/include/linux/agp_backend.h b/include/linux/agp_backend.h
--- a/include/linux/agp_backend.h Thu Sep 19 13:15:48 2002
+++ b/include/linux/agp_backend.h Thu Sep 19 13:15:48 2002
@@ -120,6 +120,7 @@
size_t page_count;
int num_scratch_pages;
unsigned long *memory;
+ void *vmptr;
off_t pg_start;
u32 type;
u32 physical;
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/