Ignore MCFG if the mmconfig area isn't reserved in the e820 table
PATCH 1/1 Ignore MCFG if the mmconfig area isn't reserved in the e820 table
Hi,
There have been several machines that don't have a working MMCONFIG, often because of a buggy MCFG table in the ACPI bios. This patch adds a simple sanity check that detects a whole bunch of these cases, and when it detects it, linux now boots rather than crash-and-burns.
arch/i386/kernel/setup.c +25 -0
@@ -1377,6 +1377,31 @@ static void __init register_memory(void) pci_mem_start, gapstart, gapsize); } + + +/* +* Check if an address is reserved in the e820 map +*/ +int is_e820_reserved (u64 address) +{ +int i; + i = e820.nr_map; +while (--i >=0 ) { +unsigned long long start = e820.map[i].addr; +unsigned long long end = start + e820.map[i].size; + +if (address <=end && address >= start) { +if (e820.map[i].type ==E820_RESERVED ) +return 1 ; +else +return 0 ; + } + } +return 0 ; +} + + +/* Use inline assembly to define this because the nops are defined as inline assembly strings in the include files and we cannot get them easily into strings. */
arch/i386/pci/mmconfig.c +12 -0
@@ -12,6 +12,7 @@#include <linux/pci.h> #include <linux/init.h> #include <linux/acpi.h> +#include <asm/e820.h> #include "pci.h" #define mmcfg_virt_addr ((void __iomem *) fix_to_virt(FIX_PCIE_MCFG))
@@ -183,6 +184,17 @@ static int __init pci_mmcfg_init(void) (pci_mmcfg_config[0].base_address == 0)) goto out; +/* +* several bioses have a buggy MCFG table. While this is hard +* to test for conclusively, we know the value is defective +* if the memory isn't marked reserved in the e820 table +*/ +if (!is_e820_reserved (pci_mmcfg_config[0 ].base_address)) { +printk (KERN_INFO "PCI: BIOS Bug: MCFG area is not reserved\n" ); +printk (KERN_INFO "PCI: Not using MMCONFIG.\n" ); + goto out; + } +printk (KERN_INFO "PCI: Using MMCONFIG\n" ); raw_pci_ops = &pci_mmcfg; pci_probe = (pci_probe & ~PCI_PROBE_MASK ) |PCI_PROBE_MMCONF ;
arch/x86_64/kernel/e820.c +22 -0
@@ -639,3 +639,25 @@ __init void e820_setup_gap(void)printk (KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n" , pci_mem_start, gapstart, gapsize); } + +/* +* Check if an address is reserved in the e820 map +*/ +int is_e820_reserved (u64 address) +{ +int i; + i = e820.nr_map; +while (--i >=0 ) { +unsigned long long start = e820.map[i].addr; +unsigned long long end = start + e820.map[i].size; + +if (address <=end && address >= start) { +if (e820.map[i].type ==E820_RESERVED ) +return 1 ; +else +return 0 ; + } + } +return 0 ; +} +
arch/x86_64/pci/mmconfig.c +14 -0
@@ -9,6 +9,7 @@#include <linux/init.h> #include <linux/acpi.h> #include <linux/bitmap.h> +#include <asm/e820.h> #include "pci.h" #define MMCONFIG_APER_SIZE (256*1024*1024)
@@ -161,6 +162,19 @@ static int __init pci_mmcfg_init(void) (pci_mmcfg_config[0].base_address == 0))return 0 ; + +/* +* several bioses have a buggy MCFG table. While this is hard +* to test for conclusively, we know the value is defective +* if the memory isn't marked reserved in the e820 table +*/ +if (!is_e820_reserved (pci_mmcfg_config[0 ].base_address)) { +printk (KERN_INFO "PCI: BIOS Bug: MCFG area is not reserved\n" ); +printk (KERN_INFO "PCI: Not using MMCONFIG.\n" ); +return 0 ; + } + +/* RED-PEN i386 doesn't do _nocache right now */ pci_mmcfg_virt =kmalloc (sizeof (*pci_mmcfg_virt) * pci_mmcfg_config_num,GFP_KERNEL );if (pci_mmcfg_virt ==NULL ) {
include/asm-i386/e820.h +3 -0
@@ -35,6 +35,9 @@ struct e820map { };extern struct e820map e820; + +extern int is_e820_reserved (u64 address); +#endif /*!__ASSEMBLY__*/ #endif /*__E820_HEADER*/
include/asm-x86_64/e820.h +1 -0
@@ -47,6 +47,7 @@ extern void contig_e820_setup(void);extern unsigned long e820_end_of_ram (void );extern void e820_reserve_resources (void );extern void e820_print_map (char *who); +extern int is_e820_reserved (u64 address);extern int e820_mapped (unsigned long start,unsigned long end,unsigned type);extern void e820_bootmem_free (pg_data_t *pgdat,unsigned long start,unsigned long end);
Quoted 6 lines · Arjan van de Ven
Hi,
There have been several machines that don't have a working MMCONFIG, often because of a buggy MCFG table in the ACPI bios. This patch adds a simple sanity check that detects a whole bunch of these cases, and when it detects it, linux now boots rather than crash-and-burns.
Yes, MCFG is a pain recently. Looks like we did the grave mistake of using something in the BIOS before Windows again.
Quoted 11 lines
arch/i386/kernel/setup.c +11
-0+ +/* +* Check if an address is reserved in the e820 map +*/ +int is_e820_reserved (u64 address) +{ +int i; + i = e820.nr_map; +while (--i >=0 ) { +unsigned long long start = e820.map[i].addr; +unsigned long long end = start + e820.map[i].size;
arch/i386/kernel/setup.c static void __init register_memory(void) · line 1401
Quoted 10 lines
arch/i386/kernel/setup.c +10
-0+ +if (address <=end && address >= start) { +if (e820.map[i].type ==E820_RESERVED ) +return 1 ; +else +return 0 ; + } + } +return 0 ; +}
That is e820_mapped(address, address+size, E820_RESERVED)
And not having a size is definitely wrong on i386 too.
-Andi
Quoted 4 lines
That is
e820_mapped(address, address+size, E820_RESERVED)And not having a size is definitely wrong on i386 too.
s/wrong/not selective enough/
and e820_mapped doesn't check this either anyway, at least not the way you imply it does.
I'll do a new patch using this for x86_64 though, no need to make a second function like this.
Quoted 11 lines · Arjan van de Ven
Quoted 3 lines
That is
e820_mapped(address, address+size, E820_RESERVED)And not having a size is definitely wrong on i386 too.
s/wrong/not selective enough/
and
e820_mappeddoesn't check this either anyway, at least not the way you imply it does.I'll do a new patch using this for
x86_64though, no need to make a second function like this.
There have been several machines that don't have a working MMCONFIG, often because of a buggy MCFG table in the ACPI bios. This patch adds a simple sanity check that detects a whole bunch of these cases, and when it detects it, linux now boots rather than crash-and-burns. The accuracy of this detection can in principle be improved if there was a "is this entire range in e820 with THIS attribute", but no such function exist and the complexity needed for this is not really worth it; this simple check already catches most cases anyway.
arch/i386/kernel/setup.c +21 -0
@@ -1377,6 +1377,27 @@ static void __init register_memory(void) pci_mem_start, gapstart, gapsize); } +/* +* Check if an address is reserved in the e820 map +*/ +int is_e820_reserved (u64 address) +{ +int i; + i = e820.nr_map; +while (--i >=0 ) { +unsigned long long start = e820.map[i].addr; +unsigned long long end = start + e820.map[i].size; + +if (address <=end && address >= start) { +if (e820.map[i].type ==E820_RESERVED ) +return 1 ; +else +return 0 ; + } + } +return 0 ; +} +/* Use inline assembly to define this because the nops are defined as inline assembly strings in the include files and we cannot get them easily into strings. */
arch/i386/pci/mmconfig.c +12 -0
@@ -12,6 +12,7 @@#include <linux/pci.h> #include <linux/init.h> #include <linux/acpi.h> +#include <asm/e820.h> #include "pci.h" #define mmcfg_virt_addr ((void __iomem *) fix_to_virt(FIX_PCIE_MCFG))
@@ -183,6 +184,17 @@ static int __init pci_mmcfg_init(void) (pci_mmcfg_config[0].base_address == 0)) goto out; +/* +* several bioses have a buggy MCFG table. While this is hard +* to test for conclusively, we know the value is defective +* if the memory isn't marked reserved in the e820 table +*/ +if (!is_e820_reserved (pci_mmcfg_config[0 ].base_address)) { +printk (KERN_INFO "PCI: BIOS Bug: MCFG area is not reserved\n" ); +printk (KERN_INFO "PCI: Not using MMCONFIG.\n" ); + goto out; + } +printk (KERN_INFO "PCI: Using MMCONFIG\n" ); raw_pci_ops = &pci_mmcfg; pci_probe = (pci_probe & ~PCI_PROBE_MASK ) |PCI_PROBE_MMCONF ;
arch/x86_64/kernel/e820.c +5 -0
@@ -80,6 +80,11 @@ static inline int bad_addr(unsigned longreturn 0 ; } +/* +* This function returns 1 if any part of the <start, end> range is in the +* E820 map having "type". There may be parts in this range that are not in +* E820 at all and/or parts with different types in addition. +*/ int __init e820_mapped (unsigned long start,unsigned long end,unsigned type) {int i;
arch/x86_64/pci/mmconfig.c +14 -0
@@ -9,6 +9,7 @@#include <linux/init.h> #include <linux/acpi.h> #include <linux/bitmap.h> +#include <asm/e820.h> #include "pci.h" #define MMCONFIG_APER_SIZE (256*1024*1024)
@@ -161,6 +162,19 @@ static int __init pci_mmcfg_init(void) (pci_mmcfg_config[0].base_address == 0))return 0 ; +/* +* several bioses have a buggy MCFG table. While this is hard +* to test for conclusively, we know the value is defective +* if the memory isn't marked reserved in the e820 table +*/ +if (!e820_mapped (pci_mmcfg_config[0 ].base_address, + pci_mmcfg_config[0 ].base_address +MMCONFIG_APER_SIZE , +E820_RESERVED )) { +printk (KERN_INFO "PCI: BIOS Bug: MCFG area is not reserved\n" ); +printk (KERN_INFO "PCI: Not using MMCONFIG.\n" ); +return 0 ; + } +/* RED-PEN i386 doesn't do _nocache right now */ pci_mmcfg_virt =kmalloc (sizeof (*pci_mmcfg_virt) * pci_mmcfg_config_num,GFP_KERNEL );if (pci_mmcfg_virt ==NULL ) {
include/asm-i386/e820.h +3 -0
@@ -35,6 +35,9 @@ struct e820map { };extern struct e820map e820; + +extern int is_e820_reserved (u64 address); +#endif /*!__ASSEMBLY__*/ #endif /*__E820_HEADER*/
Quoted 14 lines · Arjan van de Ven
Quoted 11 lines · Arjan van de Ven
Quoted 3 lines
That is
e820_mapped(address, address+size, E820_RESERVED)And not having a size is definitely wrong on i386 too.
s/wrong/not selective enough/
and
e820_mappeddoesn't check this either anyway, at least not the way you imply it does.I'll do a new patch using this for
x86_64though, no need to make a second function like this.
Quoted 8 lines
There have been several machines that don't have a working MMCONFIG, often because of a buggy MCFG table in the ACPI bios. This patch adds a simple sanity check that detects a whole bunch of these cases, and when it detects it, linux now boots rather than crash-and-burns. The accuracy of this detection can in principle be improved if there was a "is this entire range in e820 with THIS attribute", but no such function exist and the complexity needed for this is not really worth it; this simple check already catches most cases anyway.
I added the patch to my patchkit now. I also have an older patch (needs a bit more cleanup) that checks for all busses if they are reachable using MCFG Still needs some more work and interaction check with PCI hotplug though.
-Andi
Quoted 1 line · Andi Kleen
I added the patch to my patchkit now.
thanks
Quoted 4 lines
I also have an older patch (needs a bit
more cleanup) that checks for all busses if they are reachable using MCFG Still needs some more work and interaction check with PCI hotplug though.
yes more advanced tests are going to be useful; at least this one was simple and catches a large portion of the problem cases.
Quoted 25 lines · Andi Kleen
Quoted 14 lines · Arjan van de Ven
Quoted 11 lines · Arjan van de Ven
Quoted 3 lines
That is
e820_mapped(address, address+size, E820_RESERVED)And not having a size is definitely wrong on i386 too.
s/wrong/not selective enough/
and
e820_mappeddoesn't check this either anyway, at least not the way you imply it does.I'll do a new patch using this for
x86_64though, no need to make a second function like this.Quoted 8 lines
There have been several machines that don't have a working MMCONFIG, often because of a buggy MCFG table in the ACPI bios. This patch adds a simple sanity check that detects a whole bunch of these cases, and when it detects it, linux now boots rather than crash-and-burns. The accuracy of this detection can in principle be improved if there was a "is this entire range in e820 with THIS attribute", but no such function exist and the complexity needed for this is not really worth it; this simple check already catches most cases anyway.
Quoted 3 lines
I added the patch to my patchkit now. I also have an older patch (needs a bit more cleanup) that checks for all busses if they are reachable using MCFG Still needs some more work and interaction check with PCI hotplug though.
If you need help with that, please let me know.
Otherwise I'll let you push this to Linus when you feel it's ready :)
thanks,
greg k-h
Re: [patch] Ignore MCFG if the mmconfig area isn't reserved in thee820 table
In reply to Arjan van de VenQuoted 7 lines · Arjan van de Ven
Quoted 4 lines
I'll do a new patch using this for
x86_64though, no need to make a second function like this.int
__inite820_mapped(unsigned long start, unsigned long end,
unsigned type)
Why not use the same type of function like x86_64 as well instead of the newly added is_820_mapped()? If the purpose of both functions is the same, i386 could benefit with same style code instead of a slight variant.
Signature · 3 lines
Cheers, Ashok Raj - Open Source Technology Center
Re: [patch] Ignore MCFG if the mmconfig area isn't reserved in thee820 table
Quoted 11 lines · Ashok Raj
Quoted 6 lines · Arjan van de Ven
Quoted 3 lines
I'll do a new patch using this for
x86_64though, no need to make a second function like this.int
__inite820_mapped(unsigned long start, unsigned long end,
unsigned type)Why not use the same type of function like
x86_64as well instead of the newly
addedis_820_mapped()? If the purpose of both functions is the same, i386 could benefit with same style code instead of a slight variant.
the purpose is not the same. the e820_mapped function is far less strict in its check (I'm still afraid it is too weak for this purpose actually)
and it's not is_e820_mapped but is_e820_reserved()
Re: [patch] Ignore MCFG if the mmconfig area isn't reserved in thee820 table
Quoted 10 lines · Arjan van de Ven
Quoted 9 lines · Ashok Raj
Quoted 6 lines · Arjan van de Ven
Quoted 3 lines
I'll do a new patch using this for
x86_64though, no need to make a second function like this.int
__inite820_mapped(unsigned long start, unsigned long end,
unsigned type)
Quoted 6 lines
Quoted 3 lines
Why not use the same type of function like
x86_64as well instead of the newly addedis_820_mapped()? If the purpose of both functions is the same, i386 could benefit with same style code instead of a slight variant.the purpose is not the same. the
e820_mappedfunction is far less strict in its check (I'm still afraid it is too weak for this purpose actually)
In theory they should be the same. What do you think is different?
Quoted 2 lines
and it's not
is_e820_mappedbutis_e820_reserved()
That's just a special case.
-Andi
Re: [patch] Ignore MCFG if the mmconfig area isn't reserved in thee820 table
Quoted 1 line · Andi Kleen
In theory they should be the same. What do you think is different?
in practice the x86-64 version returns "success" if there is one byte in the entire memory range that complies with the requested type, even if the rest of the range is of another type. What the ideal is for the purpose here is "is the entire range reserved", but for now I'll settle for "is the start address reserved".
(and yes you can express the "is the start address reserved" as a question to the current function for a 1 byte range, I probably should do that I suppose)
Re: [patch] Ignore MCFG if the mmconfig area isn't reserved in thee820 table
Quoted 10 lines · Arjan van de Ven
Quoted 1 line · Andi Kleen
In theory they should be the same. What do you think is different?
in practice the x86-64 version returns "success" if there is one byte in the entire memory range that complies with the requested type, even if the rest of the range is of another type. What the ideal is for the purpose here is "is the entire range reserved", but for now I'll settle for "is the start address reserved".
(and yes you can express the "is the start address reserved" as a question to the current function for a 1 byte range, I probably should do that I suppose)
or why not check
if (type == ei->type && start >= ei->addr && end <= (ei->addr + ei->size))
return 1;
will this make the range check check stricter?
Signature · 3 lines
Cheers, Ashok Raj - Open Source Technology Center
Re: [patch] Ignore MCFG if the mmconfig area isn't reserved in thee820 table
Quoted 11 lines · Ashok Raj
Quoted 9 lines · Arjan van de Ven
Quoted 1 line · Andi Kleen
In theory they should be the same. What do you think is different?
in practice the x86-64 version returns "success" if there is one byte in the entire memory range that complies with the requested type, even if the rest of the range is of another type. What the ideal is for the purpose here is "is the entire range reserved", but for now I'll settle for "is the start address reserved".
(and yes you can express the "is the start address reserved" as a question to the current function for a 1 byte range, I probably should do that I suppose)
Quoted 6 lines
or why not check
if (type ==
ei->type&& start >=ei->addr&& end <= (ei->addr+ei->size))return 1;will this make the range check check stricter?
that's not going to cut it; you can have 2 e820 entries that together span the range
Re: [patch] Ignore MCFG if the mmconfig area isn't reserved in thee820 table
In reply to Arjan van de VenQuoted 6 lines · Arjan van de Ven
Quoted 1 line · Andi Kleen
In theory they should be the same. What do you think is different?
in practice the x86-64 version returns "success" if there is one byte in the entire memory range that complies with the requested type, even if the rest of the range is of another type.
I would consider that a bug. Please send fix.
-Andi
Re: [patch] Ignore MCFG if the mmconfig area isn't reserved in thee820 table
Quoted 8 lines · Andi Kleen
Quoted 5 lines · Arjan van de Ven
Quoted 1 line · Andi Kleen
In theory they should be the same. What do you think is different?
in practice the x86-64 version returns "success" if there is one byte in the entire memory range that complies with the requested type, even if the rest of the range is of another type.
I would consider that a bug. Please send fix.
I'm less sure. It's what the function does, and I can see very valid usage models for it; to detect that a certain type is NOT present. And the code is clearly written with that goal in mind at least. I'm tempted to write a real range function but I also was hoping to avoid doing that, since for the MCFG test it really is a bit overkill.