[PATCH] x86-64: disable the GART early

[PATCH] x86-64: disable the GART early

[Yinghai Lu]

this one is against to linus tree.
in case some one need to use kexec to new kernel from old kernel with gart_shutdown.

YH

[PATCH] x86-64: disable the GART early

For K8 system: 4G RAM with memory hole remapping enabled, or more than 4G RAM
installed.

when try to use kexec second kernel, and the first doesn't include gart_shutdown.
the second kernel could have different aper position than the first kernel. and second
kernel could use that hole as RAM that is still used by GART set by the first kernel.
esp. when try to kexec 2.6.24 with sparse mem enable from previous kernel (from RHEL 5
or SLES 10). the new kernel will use aper by GART (set by first kernel) for vmemmap.
and after new kernel setting one new GART. the position will be real RAM. the _mapcount
set is lost.


Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13

Call Trace:
 [<ffffffff8026401f>] bad_page+0x63/0x8d
 [<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
 [<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
 [<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
 [<ffffffff80ba3461>] mem_init+0x3b/0x152
 [<ffffffff80b959d3>] start_kernel+0x236/0x2c2
 [<ffffffff80b9511a>] _sinittext+0x11a/0x121

and
 [ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000

RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB

new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000

So try to disable that GART as early as possible.


Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>

Index: linux-2.6/arch/x86/kernel/aperture_64.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/aperture_64.c
+++ linux-2.6/arch/x86/kernel/aperture_64.c
@@ -204,6 +204,73 @@ static __u32 __init search_agp_bridge(u3
 	return 0;
 }
 
+void __init early_gart_iommu_disable(void)
+{
+	/*
+	 * disable it in case it is enabled before, esp for kexec/kdump,
+	 * previous kernel already enable that. otherwise memset called
+	 * by allocate_aperture/__alloc_bootmem_nopanic cause restart.
+	 * or second kernel have different position for GART hole. and new
+	 * kernel could use hole as RAM that is still used by GART set by
+	 * first kernel
+	 */
+	int fix, num;
+	u32 ctl;
+        u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0;
+        u64 aper_base, last_aper_base = 0;
+        int valid_agp = 0;
+
+        if (!early_pci_allowed())
+                return;
+
+	fix = 0;
+        for (num = 24; num < 32; num++) {
+                if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
+                        continue;
+
+                aper_order = (read_pci_config(0, num, 3, 0x90) >> 1) & 7;
+                aper_size = (32 * 1024 * 1024) << aper_order;
+                aper_base = read_pci_config(0, num, 3, 0x94) & 0x7fff;
+                aper_base <<= 25;
+
+                if (!aperture_valid(aper_base, aper_size)) {
+                        fix = 1;
+                        break;
+                }
+
+                if ((last_aper_order && aper_order != last_aper_order) ||
+                    (last_aper_base && aper_base != last_aper_base)) {
+                        fix = 1;
+                        break;
+                }
+                last_aper_order = aper_order;
+                last_aper_base = aper_base;
+        }
+
+	if (!fix)
+		return;
+
+	aper_alloc = search_agp_bridge(&aper_order, &valid_agp);
+
+	if (aper_alloc) {
+		/*
+		 * could get aper later or before from the AGP bridge,
+		 * so it is safe
+		 */
+		return;
+	}
+
+	for (num = 24; num < 32; num++) {
+		if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
+			continue;
+
+		ctl = read_pci_config(0, num, 3, 0x90);
+		ctl &= ~1;
+		write_pci_config(0, num, 3, 0x90, 0);
+	}
+
+}
+
 void __init gart_iommu_hole_init(void)
 { 
 	int fix, num; 
Index: linux-2.6/arch/x86/kernel/setup_64.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/setup_64.c
+++ linux-2.6/arch/x86/kernel/setup_64.c
@@ -256,6 +256,8 @@ static void discover_ebda(void)
 		ebda_size = 64*1024;
 }
 
+extern void early_gart_iommu_disable(void);
+
 void __init setup_arch(char **cmdline_p)
 {
 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
@@ -309,6 +311,8 @@ void __init setup_arch(char **cmdline_p)
 
 	discover_ebda();
 
+	early_gart_iommu_disable();
+
 	init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
 
 	dmi_scan_machine();

[PATCH] x86-64: disable the GART early v2

[Yinghai Lu]

please check the one that can be applied to x86.git -mm

YH

[PATCH] x86-64: disable the GART early v2

For K8 system: 4G RAM with memory hole remapping enabled, or more than 4G RAM
installed.

when try to use kexec second kernel, and the first doesn't include gart_shutdown.
the second kernel could have different aper position than the first kernel. and second
kernel could use that hole as RAM that is still used by GART set by the first kernel.
esp. when try to kexec 2.6.24 with sparse mem enable from previous kernel (from RHEL 5
or SLES 10). the new kernel will use aper by GART (set by first kernel) for vmemmap.
and after new kernel setting one new GART. the position will be real RAM. the _mapcount
set is lost.


Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13

Call Trace:
 [<ffffffff8026401f>] bad_page+0x63/0x8d
 [<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
 [<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
 [<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
 [<ffffffff80ba3461>] mem_init+0x3b/0x152
 [<ffffffff80b959d3>] start_kernel+0x236/0x2c2
 [<ffffffff80b9511a>] _sinittext+0x11a/0x121

and
 [ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000

RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB

new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000

So try to disable that GART as early as possible.


Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>


diff --git a/arch/x86/kernel/aperture_64.c b/arch/x86/kernel/aperture_64.c
index 52d2bea..6e7af45 100644
--- a/arch/x86/kernel/aperture_64.c
+++ b/arch/x86/kernel/aperture_64.c
@@ -218,6 +218,73 @@ static __u32 __init search_agp_bridge(u32 *order, int *valid_agp)
 	return 0;
 }
 
+void __init early_gart_iommu_disable(void)
+{
+	/*
+	 * disable it in case it is enabled before, esp for kexec/kdump,
+	 * previous kernel already enable that. otherwise memset called
+	 * by allocate_aperture/__alloc_bootmem_nopanic cause restart.
+	 * or second kernel have different position for GART hole. and new
+	 * kernel could use hole as RAM that is still used by GART set by
+	 * first kernel
+	 */
+	int fix, num;
+	u32 ctl;
+	u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0;
+	u64 aper_base, last_aper_base = 0;
+	int valid_agp = 0;
+
+	if (!early_pci_allowed())
+		return;
+
+	fix = 0;
+	for (num = 24; num < 32; num++) {
+		if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
+			continue;
+
+		aper_order = (read_pci_config(0, num, 3, 0x90) >> 1) & 7;
+		aper_size = (32 * 1024 * 1024) << aper_order;
+		aper_base = read_pci_config(0, num, 3, 0x94) & 0x7fff;
+		aper_base <<= 25;
+
+		if (!aperture_valid(aper_base, aper_size)) {
+			fix = 1;
+			break;
+		}
+
+		if ((last_aper_order && aper_order != last_aper_order) ||
+		    (last_aper_base && aper_base != last_aper_base)) {
+			fix = 1;
+			break;
+		}
+		last_aper_order = aper_order;
+		last_aper_base = aper_base;
+	}
+
+	if (!fix)
+		return;
+
+	aper_alloc = search_agp_bridge(&aper_order, &valid_agp);
+
+	if (aper_alloc) {
+		/*
+		 * could get aper later or before from the AGP bridge,
+		 * so it is safe
+		 */
+		return;
+	}
+
+	for (num = 24; num < 32; num++) {
+		if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
+			continue;
+
+		ctl = read_pci_config(0, num, 3, 0x90);
+		ctl &= ~1;
+		write_pci_config(0, num, 3, 0x90, ctl);
+	}
+
+}
+
 void __init gart_iommu_hole_init(void)
 {
 	u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0;
diff --git a/arch/x86/kernel/setup_64.c b/arch/x86/kernel/setup_64.c
index 998ce03..da82065 100644
--- a/arch/x86/kernel/setup_64.c
+++ b/arch/x86/kernel/setup_64.c
@@ -54,6 +54,7 @@
 #include <video/edid.h>
 #include <asm/e820.h>
 #include <asm/dma.h>
+#include <asm/gart.h>
 #include <asm/mpspec.h>
 #include <asm/mmu_context.h>
 #include <asm/proto.h>
@@ -358,6 +359,8 @@ void __init setup_arch(char **cmdline_p)
 
 	discover_ebda();
 
+	early_gart_iommu_disable();
+
 	init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
 	if (efi_enabled)
 		efi_init();
diff --git a/include/asm-x86/gart.h b/include/asm-x86/gart.h
index f704c50..2ca68c6 100644
--- a/include/asm-x86/gart.h
+++ b/include/asm-x86/gart.h
@@ -9,6 +9,7 @@ extern int iommu_detected;
 extern void gart_iommu_init(void);
 extern void gart_iommu_shutdown(void);
 extern void __init gart_parse_options(char *);
+extern void early_gart_iommu_disable(void);
 extern void gart_iommu_hole_init(void);
 extern int fallback_aper_order;
 extern int fallback_aper_force;
@@ -20,6 +21,10 @@ extern int fix_aperture;
 #define gart_iommu_aperture 0
 #define gart_iommu_aperture_allowed 0
 
+static inline void early_gart_iommu_disable(void)
+{
+}
+
 static inline void gart_iommu_shutdown(void)
 {
 }

Re: [PATCH] x86-64: disable the GART early v2

[Ingo Molnar]

* Yinghai Lu <Yinghai.Lu@Sun.COM> wrote:

> --- a/arch/x86/kernel/aperture_64.c
> +++ b/arch/x86/kernel/aperture_64.c
> @@ -218,6 +218,73 @@ static __u32 __init search_agp_bridge(u32 *order, int *valid_agp)
>  	return 0;
>  }
>  
> +void __init early_gart_iommu_disable(void)
> +{
> +	/*
> +	 * disable it in case it is enabled before, esp for kexec/kdump,
> +	 * previous kernel already enable that. otherwise memset called
> +	 * by allocate_aperture/__alloc_bootmem_nopanic cause restart.
> +	 * or second kernel have different position for GART hole. and new
> +	 * kernel could use hole as RAM that is still used by GART set by
> +	 * first kernel
> +	 */

hm, i'm wondering, instead of modifying the GART, why dont we simply 
_detect_ whatever GART settings we have inherited, and propagate that 
into our e820 maps? I.e. if there's inconsistency, then punch that out 
from the memory maps and just dont use that memory.

that way it would not matter whether the GART settings came from a [old 
or crashing] Linux kernel that has not called gart_iommu_shutdown(), or 
whether it's a BIOS that has set up an aperture hole inconsistent with 
the memory map it passed. (or the memory map we _think_ i tried to pass 
us)

it would also be more robust to only read and do a memory map quirk 
based on that, than actively trying to change the GART so early in the 
bootup. Later on we have to re-enable the GART _anyway_ and have to 
punch a hole for it.

and as a bonus, we would have shored up our defenses against crappy 
BIOSes as well.

	Ingo

Re: [PATCH] x86-64: disable the GART early v2

[Yinghai Lu]

On Jan 11, 2008 12:14 AM, Ingo Molnar <mingo@elte.hu> wrote:
>
> * Yinghai Lu <Yinghai.Lu@Sun.COM> wrote:
>
> > --- a/arch/x86/kernel/aperture_64.c
> > +++ b/arch/x86/kernel/aperture_64.c
> > @@ -218,6 +218,73 @@ static __u32 __init search_agp_bridge(u32 *order, int *valid_agp)
> >       return 0;
> >  }
> >
> > +void __init early_gart_iommu_disable(void)
> > +{
> > +     /*
> > +      * disable it in case it is enabled before, esp for kexec/kdump,
> > +      * previous kernel already enable that. otherwise memset called
> > +      * by allocate_aperture/__alloc_bootmem_nopanic cause restart.
> > +      * or second kernel have different position for GART hole. and new
> > +      * kernel could use hole as RAM that is still used by GART set by
> > +      * first kernel
> > +      */
>
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
good, i will update the patch to check that hole with e820 map.

that also make code short so don't need to check if agp is there or not.

YH

[PATCH] x86-64: fix e820 for GART or disable the GART early

[Yinghai Lu]

[PATCH] x86-64: fix e820 for GART or disable the GART early

For K8 system: 4G RAM with memory hole remapping enabled, or more than 4G RAM
installed.

when try to use kexec second kernel, and the first doesn't include gart_shutdown.
the second kernel could have different aper position than the first kernel. and second
kernel could use that hole as RAM that is still used by GART set by the first kernel.
esp. when try to kexec 2.6.24 with sparse mem enable from previous kernel (from RHEL 5
or SLES 10). the new kernel will use aper by GART (set by first kernel) for vmemmap.
and after new kernel setting one new GART. the position will be real RAM. the _mapcount
set is lost.


Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13

Call Trace:
 [<ffffffff8026401f>] bad_page+0x63/0x8d
 [<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
 [<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
 [<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
 [<ffffffff80ba3461>] mem_init+0x3b/0x152
 [<ffffffff80b959d3>] start_kernel+0x236/0x2c2
 [<ffffffff80b9511a>] _sinittext+0x11a/0x121

and
 [ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000

RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB

new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000

So could try to disable that GART if possible.

According to Ingo

> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
> 
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
> 
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
> 
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.

add e820 modification for gart inconsistent setting.

gart_fix_e820=off could be used to disable e820 fix.

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>

 Documentation/kernel-parameters.txt |    3 +
 arch/x86/kernel/aperture_64.c       |   86 ++++++++++++++++++++++++++++++++++++
 arch/x86/kernel/e820_64.c           |   12 +++++
 arch/x86/kernel/setup_64.c          |    3 +
 include/asm-x86/e820_64.h           |    1 
 include/asm-x86/gart.h              |    5 ++
 6 files changed, 110 insertions(+)

diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index 7357f24..f2a7fa4 100644
--- a/Documentation/kernel-parameters.txt
+++ b/Documentation/kernel-parameters.txt
@@ -652,6 +652,9 @@ and is between 256 and 4096 characters. It is defined in the file
 
 	gamma=		[HW,DRM]
 
+	gart_fix_e820=  [X86_64] disable the fix e820 for K8 GART
+			Format: off
+
 	gdth=		[HW,SCSI]
 			See header of drivers/scsi/gdth.c.
 
diff --git a/arch/x86/kernel/aperture_64.c b/arch/x86/kernel/aperture_64.c
index 52d2bea..53cbcae 100644
--- a/arch/x86/kernel/aperture_64.c
+++ b/arch/x86/kernel/aperture_64.c
@@ -218,6 +218,92 @@ static __u32 __init search_agp_bridge(u32 *order, int *valid_agp)
 	return 0;
 }
 
+static int gart_fix_e820 __initdata = 1;
+
+static int __init parse_gart_mem(char *p)
+{
+	if (!p)
+		return -EINVAL;
+
+	if (!strncmp(p, "off", 3))
+		gart_fix_e820 = 0;
+
+        return 0;
+}
+early_param("gart_fix_e820", parse_gart_mem);
+
+void __init early_gart_iommu_check(void)
+{
+	/*
+	 * in case it is enabled before, esp for kexec/kdump,
+	 * previous kernel already enable that. memset called
+	 * by allocate_aperture/__alloc_bootmem_nopanic cause restart.
+	 * or second kernel have different position for GART hole. and new
+	 * kernel could use hole as RAM that is still used by GART set by
+	 * first kernel
+	 * or BIOS forget to put that in reserved.
+	 * try to update e820 to make that region as reserved.
+	 */
+	int fix, num;
+	u32 ctl;
+	u32 aper_size = 0, aper_order = 0, last_aper_order = 0;
+	u64 aper_base = 0, last_aper_base = 0;
+	int aper_enabled = 0, last_aper_enabled = 0;
+
+	if (!early_pci_allowed())
+		return;
+
+	fix = 0;
+	for (num = 24; num < 32; num++) {
+		if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
+			continue;
+
+		ctl = read_pci_config(0, num, 3, 0x90);
+		aper_enabled = ctl & 1;
+		aper_order = (ctl >> 1) & 7;
+		aper_size = (32 * 1024 * 1024) << aper_order;
+		aper_base = read_pci_config(0, num, 3, 0x94) & 0x7fff;
+		aper_base <<= 25;
+
+		if ((last_aper_order && aper_order != last_aper_order) ||
+		    (last_aper_base && aper_base != last_aper_base) ||
+		    (last_aper_enabled && aper_enabled != last_aper_enabled)) {
+			fix = 1;
+			break;
+		}
+		last_aper_order = aper_order;
+		last_aper_base = aper_base;
+		last_aper_enabled = aper_enabled;
+	}
+
+	if (!fix && !aper_enabled)
+		return;
+
+	if (!aper_base || !aper_size || aper_base + aper_size > 0x100000000UL)
+                fix = 1;
+
+	if (gart_fix_e820 && !fix && aper_enabled &&
+	    e820_any_mapped(aper_base, aper_base + aper_size, E820_RAM)) {
+		/* reserved that, so we can resuse it in second kernel */
+		printk(KERN_INFO "update e820 for GART\n");
+		add_memory_region(aper_base, aper_size, E820_RESERVED);
+		update_e820();
+		return;
+	}
+
+	/* different nodes have different setting, disable them all at first*/
+
+	for (num = 24; num < 32; num++) {
+		if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
+			continue;
+
+		ctl = read_pci_config(0, num, 3, 0x90);
+		ctl &= ~1;
+		write_pci_config(0, num, 3, 0x90, ctl);
+	}
+
+}
+
 void __init gart_iommu_hole_init(void)
 {
 	u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0;
diff --git a/arch/x86/kernel/e820_64.c b/arch/x86/kernel/e820_64.c
index abc473b..07cfaae 100644
--- a/arch/x86/kernel/e820_64.c
+++ b/arch/x86/kernel/e820_64.c
@@ -728,6 +728,18 @@ void __init finish_e820_parsing(void)
 	}
 }
 
+void __init update_e820(void)
+{
+	u8 nr_map;
+
+	nr_map = e820.nr_map;
+	if (sanitize_e820_map(e820.map, &nr_map))
+		return;
+	e820.nr_map = nr_map;
+	printk(KERN_INFO "modified physical RAM map:\n");
+	e820_print_map("modified");
+}
+
 unsigned long pci_mem_start = 0xaeedbabe;
 EXPORT_SYMBOL(pci_mem_start);
 
diff --git a/arch/x86/kernel/setup_64.c b/arch/x86/kernel/setup_64.c
index 998ce03..331f1f8 100644
--- a/arch/x86/kernel/setup_64.c
+++ b/arch/x86/kernel/setup_64.c
@@ -54,6 +54,7 @@
 #include <video/edid.h>
 #include <asm/e820.h>
 #include <asm/dma.h>
+#include <asm/gart.h>
 #include <asm/mpspec.h>
 #include <asm/mmu_context.h>
 #include <asm/proto.h>
@@ -346,6 +347,8 @@ void __init setup_arch(char **cmdline_p)
 
 	finish_e820_parsing();
 
+	early_gart_iommu_check();
+
 	e820_register_active_regions(0, 0, -1UL);
 	/*
 	 * partially used pages are not usable - thus
diff --git a/include/asm-x86/e820_64.h b/include/asm-x86/e820_64.h
index 1c7ba88..a56a0d5 100644
--- a/include/asm-x86/e820_64.h
+++ b/include/asm-x86/e820_64.h
@@ -35,6 +35,7 @@ extern void e820_register_active_regions(int nid,
 extern void finish_e820_parsing(void);
 
 extern struct e820map e820;
+extern void update_e820(void);
 
 extern unsigned ebda_addr, ebda_size;
 extern unsigned long nodemap_addr, nodemap_size;
diff --git a/include/asm-x86/gart.h b/include/asm-x86/gart.h
index f704c50..90958ed 100644
--- a/include/asm-x86/gart.h
+++ b/include/asm-x86/gart.h
@@ -9,6 +9,7 @@ extern int iommu_detected;
 extern void gart_iommu_init(void);
 extern void gart_iommu_shutdown(void);
 extern void __init gart_parse_options(char *);
+extern void early_gart_iommu_check(void);
 extern void gart_iommu_hole_init(void);
 extern int fallback_aper_order;
 extern int fallback_aper_force;
@@ -20,6 +21,10 @@ extern int fix_aperture;
 #define gart_iommu_aperture 0
 #define gart_iommu_aperture_allowed 0
 
+static inline void early_gart_iommu_check(void)
+{
+}
+
 static inline void gart_iommu_shutdown(void)
 {
 }

Re: [PATCH] x86-64: disable the GART early v2

[Yinghai Lu]

please check the one against x86.git
it will use fix e820 for gart.

YH

[PATCH] x86-64: fix e820 for GART or disable the GART early

For K8 system: 4G RAM with memory hole remapping enabled, or more than 4G RAM
installed.

when try to use kexec second kernel, and the first doesn't include gart_shutdown.
the second kernel could have different aper position than the first kernel. and second
kernel could use that hole as RAM that is still used by GART set by the first kernel.
esp. when try to kexec 2.6.24 with sparse mem enable from previous kernel (from RHEL 5
or SLES 10). the new kernel will use aper by GART (set by first kernel) for vmemmap.
and after new kernel setting one new GART. the position will be real RAM. the _mapcount
set is lost.


Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13

Call Trace:
 [<ffffffff8026401f>] bad_page+0x63/0x8d
 [<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
 [<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
 [<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
 [<ffffffff80ba3461>] mem_init+0x3b/0x152
 [<ffffffff80b959d3>] start_kernel+0x236/0x2c2
 [<ffffffff80b9511a>] _sinittext+0x11a/0x121

and
 [ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000

RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB

new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000

So could try to disable that GART if possible.

According to Ingo

> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
> 
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
> 
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
> 
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.

add e820 modification for gart inconsistent setting.

gart_fix_e820=off could be used to disable e820 fix.

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>

Index: linux-2.6/arch/x86/kernel/aperture_64.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/aperture_64.c
+++ linux-2.6/arch/x86/kernel/aperture_64.c
@@ -218,6 +218,95 @@ static __u32 __init search_agp_bridge(u3
 	return 0;
 }
 
+static int gart_fix_e820 __initdata = 1;
+
+static int __init parse_gart_mem(char *p)
+{
+	if (!p)
+		return -EINVAL;
+
+	if (!strncmp(p, "off", 3))
+		gart_fix_e820 = 0;
+	else if (!strncmp(p, "on", 2))
+		gart_fix_e820 = 1;
+
+        return 0;
+}
+early_param("gart_fix_e820", parse_gart_mem);
+
+void __init early_gart_iommu_check(void)
+{
+	/*
+	 * in case it is enabled before, esp for kexec/kdump,
+	 * previous kernel already enable that. memset called
+	 * by allocate_aperture/__alloc_bootmem_nopanic cause restart.
+	 * or second kernel have different position for GART hole. and new
+	 * kernel could use hole as RAM that is still used by GART set by
+	 * first kernel
+	 * or BIOS forget to put that in reserved.
+	 * try to update e820 to make that region as reserved.
+	 */
+	int fix, num;
+	u32 ctl;
+	u32 aper_size = 0, aper_order = 0, last_aper_order = 0;
+	u64 aper_base = 0, last_aper_base = 0;
+	int aper_enabled = 0, last_aper_enabled = 0;
+
+	if (!early_pci_allowed())
+		return;
+
+	fix = 0;
+	for (num = 24; num < 32; num++) {
+		if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
+			continue;
+
+		ctl = read_pci_config(0, num, 3, 0x90);
+		aper_enabled = ctl & 1;
+		aper_order = (ctl >> 1) & 7;
+		aper_size = (32 * 1024 * 1024) << aper_order;
+		aper_base = read_pci_config(0, num, 3, 0x94) & 0x7fff;
+		aper_base <<= 25;
+
+		if ((last_aper_order && aper_order != last_aper_order) ||
+		    (last_aper_base && aper_base != last_aper_base) ||
+		    (last_aper_enabled && aper_enabled != last_aper_enabled)) {
+			fix = 1;
+			break;
+		}
+		last_aper_order = aper_order;
+		last_aper_base = aper_base;
+		last_aper_enabled = aper_enabled;
+	}
+
+	if (!fix && !aper_enabled)
+		return;
+
+	if (!aper_base || !aper_size || aper_base + aper_size > 0x100000000UL)
+                fix = 1;
+
+	if (gart_fix_e820 && !fix && aper_enabled) {
+		if (e820_any_mapped(aper_base, aper_base + aper_size,
+				    E820_RAM)) {
+			/* reserved it, so we can resuse it in second kernel */
+			printk(KERN_INFO "update e820 for GART\n");
+			add_memory_region(aper_base, aper_size, E820_RESERVED);
+			update_e820();
+		}
+		return;
+	}
+
+	/* different nodes have different setting, disable them all at first*/
+	for (num = 24; num < 32; num++) {
+		if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
+			continue;
+
+		ctl = read_pci_config(0, num, 3, 0x90);
+		ctl &= ~1;
+		write_pci_config(0, num, 3, 0x90, ctl);
+	}
+
+}
+
 void __init gart_iommu_hole_init(void)
 {
 	u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0;
Index: linux-2.6/arch/x86/kernel/e820_64.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/e820_64.c
+++ linux-2.6/arch/x86/kernel/e820_64.c
@@ -728,6 +728,18 @@ void __init finish_e820_parsing(void)
 	}
 }
 
+void __init update_e820(void)
+{
+	u8 nr_map;
+
+	nr_map = e820.nr_map;
+	if (sanitize_e820_map(e820.map, &nr_map))
+		return;
+	e820.nr_map = nr_map;
+	printk(KERN_INFO "modified physical RAM map:\n");
+	e820_print_map("modified");
+}
+
 unsigned long pci_mem_start = 0xaeedbabe;
 EXPORT_SYMBOL(pci_mem_start);
 
Index: linux-2.6/arch/x86/kernel/setup_64.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/setup_64.c
+++ linux-2.6/arch/x86/kernel/setup_64.c
@@ -54,6 +54,7 @@
 #include <video/edid.h>
 #include <asm/e820.h>
 #include <asm/dma.h>
+#include <asm/gart.h>
 #include <asm/mpspec.h>
 #include <asm/mmu_context.h>
 #include <asm/proto.h>
@@ -346,6 +347,8 @@ void __init setup_arch(char **cmdline_p)
 
 	finish_e820_parsing();
 
+	early_gart_iommu_check();
+
 	e820_register_active_regions(0, 0, -1UL);
 	/*
 	 * partially used pages are not usable - thus
Index: linux-2.6/include/asm-x86/e820_64.h
===================================================================
--- linux-2.6.orig/include/asm-x86/e820_64.h
+++ linux-2.6/include/asm-x86/e820_64.h
@@ -35,6 +35,7 @@ extern void e820_register_active_regions
 extern void finish_e820_parsing(void);
 
 extern struct e820map e820;
+extern void update_e820(void);
 
 extern unsigned ebda_addr, ebda_size;
 extern unsigned long nodemap_addr, nodemap_size;
Index: linux-2.6/include/asm-x86/gart.h
===================================================================
--- linux-2.6.orig/include/asm-x86/gart.h
+++ linux-2.6/include/asm-x86/gart.h
@@ -9,6 +9,7 @@ extern int iommu_detected;
 extern void gart_iommu_init(void);
 extern void gart_iommu_shutdown(void);
 extern void __init gart_parse_options(char *);
+extern void early_gart_iommu_check(void);
 extern void gart_iommu_hole_init(void);
 extern int fallback_aper_order;
 extern int fallback_aper_force;
@@ -20,6 +21,10 @@ extern int fix_aperture;
 #define gart_iommu_aperture 0
 #define gart_iommu_aperture_allowed 0
 
+static inline void early_gart_iommu_check(void)
+{
+}
+
 static inline void gart_iommu_shutdown(void)
 {
 }
Index: linux-2.6/Documentation/kernel-parameters.txt
===================================================================
--- linux-2.6.orig/Documentation/kernel-parameters.txt
+++ linux-2.6/Documentation/kernel-parameters.txt
@@ -652,6 +652,10 @@ and is between 256 and 4096 characters. 
 
 	gamma=		[HW,DRM]
 
+	gart_fix_e820=  [X86_64] disable the fix e820 for K8 GART
+			Format: off | on
+			default: on
+
 	gdth=		[HW,SCSI]
 			See header of drivers/scsi/gdth.c.
 

Re: [PATCH] x86-64: disable the GART early v2

[Ingo Molnar]

* Yinghai Lu <Yinghai.Lu@Sun.COM> wrote:

> please check the one against x86.git it will use fix e820 for gart.

thanks, applied.

Sidenote - there were a few checkpatch failures:

 ERROR: use tabs not spaces
 #121: FILE: arch/x86/kernel/aperture_64.c:233:
 +        return 0;$

 ERROR: use tabs not spaces
 #173: FILE: arch/x86/kernel/aperture_64.c:285:
 +                fix = 1;$

i fixed them up. (please run "scripts/checkpatch.pl your.patch" in the 
future.)

	Ingo

[PATCH] x86_64: checking aperture report for node instead of cpu

[Yinghai Lu]

this one is against x86.git

[PATCH] x86_64: checking aperture report for node instead of cpu

currently when gart iommu is enabled by BIOS or previous we got

"
Checking aperture...
CPU 0: aperture @4000000 size 64MB
CPU 1: aperture @4000000 size 64MB
"
we should use use Node instead.

we will get
"
Checking aperture...
Node 0: aperture @4000000 size 64MB
Node 1: aperture @4000000 size 64MB
"

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>

Index: linux-2.6/arch/x86/kernel/aperture_64.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/aperture_64.c
+++ linux-2.6/arch/x86/kernel/aperture_64.c
@@ -312,6 +312,7 @@ void __init gart_iommu_hole_init(void)
 	u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0;
 	u64 aper_base, last_aper_base = 0;
 	int fix, num, valid_agp = 0;
+	int node;
 
 	if (gart_iommu_aperture_disabled || !fix_aperture ||
 	    !early_pci_allowed())
@@ -320,6 +321,7 @@ void __init gart_iommu_hole_init(void)
 	printk(KERN_INFO  "Checking aperture...\n");
 
 	fix = 0;
+	node = 0;
 	for (num = 24; num < 32; num++) {
 		if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
 			continue;
@@ -332,8 +334,8 @@ void __init gart_iommu_hole_init(void)
 		aper_base = read_pci_config(0, num, 3, 0x94) & 0x7fff;
 		aper_base <<= 25;
 
-		printk(KERN_INFO "CPU %d: aperture @ %Lx size %u MB\n",
-				num-24, aper_base, aper_size>>20);
+		printk(KERN_INFO "Node %d: aperture @ %Lx size %u MB\n",
+				node++, aper_base, aper_size >> 20);
 
 		if (!aperture_valid(aper_base, aper_size)) {
 			fix = 1;

Re: [PATCH] x86_64: checking aperture report for node instead of cpu

[Ingo Molnar]

* Yinghai Lu <Yinghai.Lu@Sun.COM> wrote:

> -		printk(KERN_INFO "CPU %d: aperture @ %Lx size %u MB\n",
> -				num-24, aper_base, aper_size>>20);
> +		printk(KERN_INFO "Node %d: aperture @ %Lx size %u MB\n",
> +				node++, aper_base, aper_size >> 20);

please dont put side-effects into printks. (put the node++ iteration out 
into a separate line) Your patch looks fine otherwise.

	Ingo

[PATCH] x86_64: checking aperture report for node instead of cpu v2

[Yinghai Lu]

[PATCH] x86_64: checking aperture report for node instead of cpu v2

currently when gart iommu is enabled by BIOS or previous we got

"
Checking aperture...
CPU 0: aperture @4000000 size 64MB
CPU 1: aperture @4000000 size 64MB
"
we should use use Node instead.

we will get
"
Checking aperture...
Node 0: aperture @4000000 size 64MB
Node 1: aperture @4000000 size 64MB
"

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>

Index: linux-2.6/arch/x86/kernel/aperture_64.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/aperture_64.c
+++ linux-2.6/arch/x86/kernel/aperture_64.c
@@ -312,6 +312,7 @@ void __init gart_iommu_hole_init(void)
 	u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0;
 	u64 aper_base, last_aper_base = 0;
 	int fix, num, valid_agp = 0;
+	int node;
 
 	if (gart_iommu_aperture_disabled || !fix_aperture ||
 	    !early_pci_allowed())
@@ -320,6 +321,7 @@ void __init gart_iommu_hole_init(void)
 	printk(KERN_INFO  "Checking aperture...\n");
 
 	fix = 0;
+	node = 0;
 	for (num = 24; num < 32; num++) {
 		if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
 			continue;
@@ -332,8 +334,9 @@ void __init gart_iommu_hole_init(void)
 		aper_base = read_pci_config(0, num, 3, 0x94) & 0x7fff;
 		aper_base <<= 25;
 
-		printk(KERN_INFO "CPU %d: aperture @ %Lx size %u MB\n",
-				num-24, aper_base, aper_size>>20);
+		printk(KERN_INFO "Node %d: aperture @ %Lx size %u MB\n",
+				node, aper_base, aper_size >> 20);
+		node++;
 
 		if (!aperture_valid(aper_base, aper_size)) {
 			fix = 1;

Re: [PATCH] x86_64: checking aperture report for node instead of?cpu v2

[Ingo Molnar]

* Yinghai Lu <Yinghai.Lu@Sun.COM> wrote:

> "
> Checking aperture...
> CPU 0: aperture @4000000 size 64MB
> CPU 1: aperture @4000000 size 64MB
> "
> we should use use Node instead.
> 
> we will get
> "
> Checking aperture...
> Node 0: aperture @4000000 size 64MB
> Node 1: aperture @4000000 size 64MB
> "

thanks, applied.

	Ingo