x86: do not enable efi boot for sgi uv machine

kexec/kdump uefi support depends on the 1:1 mapping scheme in 3.14 kernel but SGI reported that 1:1 mapping does not work on their UV system. thus there's below commit to switch back to old ioremap way for sgi uv machines.

So for kexec-tools before we got the 1:1 mapping support on those machines we should also use old way that means boot 2nd kernel without uefi.

Tested on a uv1 machine.

Here is the kernel commit to quirk out SGI UV:
commit a5d90c923bcfb9632d998ed06e9569216ad695f3

Author: Borislav Petkov <[email protected]>
Date:   Tue Mar 4 17:02:17 2014 +0100

    x86/efi: Quirk out SGI UV
    
    Alex reported hitting the following BUG after the EFI 1:1 virtual
    mapping work was merged,
    
     kernel BUG at arch/x86/mm/init_64.c:351!
     invalid opcode: 0000 [#1] SMP
     Call Trace:
      [<ffffffff818aa71d>] init_extra_mapping_uc+0x13/0x15
      [<ffffffff818a5e20>] uv_system_init+0x22b/0x124b
      [<ffffffff8108b886>] ? clockevents_register_device+0x138/0x13d
      [<ffffffff81028dbb>] ? setup_APIC_timer+0xc5/0xc7
      [<ffffffff8108b620>] ? clockevent_delta2ns+0xb/0xd
      [<ffffffff818a3a92>] ? setup_boot_APIC_clock+0x4a8/0x4b7
      [<ffffffff8153d955>] ? printk+0x72/0x74
      [<ffffffff818a1757>] native_smp_prepare_cpus+0x389/0x3d6
      [<ffffffff818957bc>] kernel_init_freeable+0xb7/0x1fb
      [<ffffffff81535530>] ? rest_init+0x74/0x74
      [<ffffffff81535539>] kernel_init+0x9/0xff
      [<ffffffff81541dfc>] ret_from_fork+0x7c/0xb0
      [<ffffffff81535530>] ? rest_init+0x74/0x74
    
    Getting this thing to work with the new mapping scheme would need more
    work, so automatically switch to the old memmap layout for SGI UV.

Signed-off-by Dave Young <[email protected]>

---
 kexec/arch/i386/x86-linux-setup.c |   15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

kexec/arch/i386/x86-linux-setup.c +11 -0

@@ -817,12 +817,25 @@ out:
 	return;
 }
 
+static int is_sgi_uv_machine(void)
+{
+	DIR *sgi_dir;
+
+	sgi_dir = opendir("/sys/firmware/sgi_uv");
+	if (sgi_dir) {
+		closedir(sgi_dir);
+		return 1;
+	}
+
+	return 0;

kexec/kdump uefi support depends on the 1:1 mapping scheme in 3.14 kernel but SGI reported that 1:1 mapping does not work on their UV system. thus there's below commit to switch back to old ioremap way for sgi uv machines.

So for kexec-tools before we got the 1:1 mapping support on those machines we should also use old way that means boot 2nd kernel without uefi.

Tested on a uv1 machine.

Here is the kernel commit to quirk out SGI UV:
commit a5d90c923bcfb9632d998ed06e9569216ad695f3

Author: Borislav Petkov <[email protected]>
Date:   Tue Mar 4 17:02:17 2014 +0100

    x86/efi: Quirk out SGI UV
    
    Alex reported hitting the following BUG after the EFI 1:1 virtual
    mapping work was merged,
    
     kernel BUG at arch/x86/mm/init_64.c:351!
     invalid opcode: 0000 [#1] SMP
     Call Trace:
      [<ffffffff818aa71d>] init_extra_mapping_uc+0x13/0x15
      [<ffffffff818a5e20>] uv_system_init+0x22b/0x124b
      [<ffffffff8108b886>] ? clockevents_register_device+0x138/0x13d
      [<ffffffff81028dbb>] ? setup_APIC_timer+0xc5/0xc7
      [<ffffffff8108b620>] ? clockevent_delta2ns+0xb/0xd
      [<ffffffff818a3a92>] ? setup_boot_APIC_clock+0x4a8/0x4b7
      [<ffffffff8153d955>] ? printk+0x72/0x74
      [<ffffffff818a1757>] native_smp_prepare_cpus+0x389/0x3d6
      [<ffffffff818957bc>] kernel_init_freeable+0xb7/0x1fb
      [<ffffffff81535530>] ? rest_init+0x74/0x74
      [<ffffffff81535539>] kernel_init+0x9/0xff
      [<ffffffff81541dfc>] ret_from_fork+0x7c/0xb0
      [<ffffffff81535530>] ? rest_init+0x74/0x74
    
    Getting this thing to work with the new mapping scheme would need more
    work, so automatically switch to the old memmap layout for SGI UV.

Signed-off-by Dave Young <[email protected]>

---
 kexec/arch/i386/x86-linux-setup.c |   15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

kexec/arch/i386/x86-linux-setup.c +11 -0

@@ -817,12 +817,25 @@ out:
 	return;
 }
 
+static int is_sgi_uv_machine(void)
+{
+	DIR *sgi_dir;
+
+	sgi_dir = opendir("/sys/firmware/sgi_uv");
+	if (sgi_dir) {
+		closedir(sgi_dir);
+		return 1;
+	}
+
+	return 0;

Hi,

Vivek, replying with webmail, sorry for the formatting issue.

Another way is export the mapping scheme in sysfs. So if kernel boot with efi=old_map we can detect it as well. OTOH in future there's probably other quirk system.

So kexec-tools can switch to use old way if kernel use ioremap.

Adding Matt and Boris to cc, to see if they have idea.

Thanks
Dave

----- Original Message -----
From: "Vivek Goyal" <[email protected]>
To: "Dave Young" <[email protected]>
Cc: [email protected], [email protected]
Sent: Friday, May 23, 2014 9:28:34 PM
Subject: Re: [PATCH] x86: do not enable efi boot for sgi uv machine

Quoted 9 lines · Dave Young

kexec/kdump uefi support depends on the 1:1 mapping scheme in 3.14 kernel but SGI reported that 1:1 mapping does not work on their UV system. thus there's below commit to switch back to old ioremap way for sgi uv machines.

So for kexec-tools before we got the 1:1 mapping support on those machines we should also use old way that means boot 2nd kernel without uefi.

Quoted 56 lines

Tested on a uv1 machine.

Here is the kernel commit to quirk out SGI UV:
commit a5d90c923bcfb9632d998ed06e9569216ad695f3

Author: Borislav Petkov <[email protected]>
Date:   Tue Mar 4 17:02:17 2014 +0100

    x86/efi: Quirk out SGI UV
    
    Alex reported hitting the following BUG after the EFI 1:1 virtual
    mapping work was merged,
    
     kernel BUG at arch/x86/mm/init_64.c:351!
     invalid opcode: 0000 [#1] SMP
     Call Trace:
      [<ffffffff818aa71d>] init_extra_mapping_uc+0x13/0x15
      [<ffffffff818a5e20>] uv_system_init+0x22b/0x124b
      [<ffffffff8108b886>] ? clockevents_register_device+0x138/0x13d
      [<ffffffff81028dbb>] ? setup_APIC_timer+0xc5/0xc7
      [<ffffffff8108b620>] ? clockevent_delta2ns+0xb/0xd
      [<ffffffff818a3a92>] ? setup_boot_APIC_clock+0x4a8/0x4b7
      [<ffffffff8153d955>] ? printk+0x72/0x74
      [<ffffffff818a1757>] native_smp_prepare_cpus+0x389/0x3d6
      [<ffffffff818957bc>] kernel_init_freeable+0xb7/0x1fb
      [<ffffffff81535530>] ? rest_init+0x74/0x74
      [<ffffffff81535539>] kernel_init+0x9/0xff
      [<ffffffff81541dfc>] ret_from_fork+0x7c/0xb0
      [<ffffffff81535530>] ? rest_init+0x74/0x74
    
    Getting this thing to work with the new mapping scheme would need more
    work, so automatically switch to the old memmap layout for SGI UV.

Signed-off-by Dave Young <[email protected]>

---
 kexec/arch/i386/x86-linux-setup.c |   15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

kexec/arch/i386/x86-linux-setup.c +11 -0

@@ -817,12 +817,25 @@ out:
 	return;
 }
 
+static int is_sgi_uv_machine(void)
+{
+	DIR *sgi_dir;
+
+	sgi_dir = opendir("/sys/firmware/sgi_uv");
+	if (sgi_dir) {
+		closedir(sgi_dir);
+		return 1;
+	}
+
+	return 0;

kexec/arch/i386/x86-linux-setup.c out: · line 839

Quoted 10 lines

kexec/arch/i386/x86-linux-setup.c +3 -1

+}
+
 void setup_linux_system_parameters(struct kexec_info *info,
 				   struct x86_linux_param_header *real_mode)
 {
 	/* get subarch from running kernel */
 	setup_subarch(real_mode);
-	if (bzImage_support_efi_boot)
+	if (bzImage_support_efi_boot && !is_sgi_uv_machine())
 		setup_efi_info(info, real_mode);

Hi Dave,

I think instead of always falling back to old "noefi" method for SGI UV machines, we should probably provide a config option to override default behavior. Something like "--nouefi". And let user/distributions decide in what cases to use that parameter.

Reason being, that at some point of time this limitation will be fixed. And at that point of time if we modify kexec-tools, then it will stop working with older kernels. And then same issue will arise that kexec-tools is not working with older kernels.

If we just provide a way to override default, then distributions in their scripts should know what kernels work what way and automatically use or not use option accordingly.

Thanks
Vivek

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

Quoted 24 lines · Dave Young

Hi,

Vivek, replying with webmail, sorry for the formatting issue.

Another way is export the mapping scheme in sysfs. So if kernel boot with efi=old_map we can detect it as well. OTOH in future there's probably other quirk system.

So kexec-tools can switch to use old way if kernel use ioremap.

Adding Matt and Boris to cc, to see if they have idea.

Thanks
Dave

----- Original Message -----
From: "Vivek Goyal" <[email protected]>
To: "Dave Young" <[email protected]>
Cc: [email protected], [email protected]
Sent: Friday, May 23, 2014 9:28:34 PM
Subject: Re: [PATCH] x86: do not enable efi boot for sgi uv machine

Quoted 1 line · Dave Young

Quoted 8 lines

Quoted 8 lines

kexec/kdump uefi support depends on the 1:1 mapping scheme in 3.14 kernel but SGI reported that 1:1 mapping does not work on their UV system. thus there's below commit to switch back to old ioremap way for sgi uv machines.

So for kexec-tools before we got the 1:1 mapping support on those machines we should also use old way that means boot 2nd kernel without uefi.

Well, we did find the bug in a combined debug session and SGI will be issuing a fix soon. I don't know whether any quirk will be needed for systems which don't have the fix - I guess that's something for SGI people to decide. CCed and leaving in the rest for reference.

@Alex, Russ: this is about using kexec on UEFI boxes.

Quoted 89 lines

Quoted 55 lines

Tested on a uv1 machine.

Here is the kernel commit to quirk out SGI UV:
commit a5d90c923bcfb9632d998ed06e9569216ad695f3

Author: Borislav Petkov <[email protected]>
Date:   Tue Mar 4 17:02:17 2014 +0100

    x86/efi: Quirk out SGI UV
    
    Alex reported hitting the following BUG after the EFI 1:1 virtual
    mapping work was merged,
    
     kernel BUG at arch/x86/mm/init_64.c:351!
     invalid opcode: 0000 [#1] SMP
     Call Trace:
      [<ffffffff818aa71d>] init_extra_mapping_uc+0x13/0x15
      [<ffffffff818a5e20>] uv_system_init+0x22b/0x124b
      [<ffffffff8108b886>] ? clockevents_register_device+0x138/0x13d
      [<ffffffff81028dbb>] ? setup_APIC_timer+0xc5/0xc7
      [<ffffffff8108b620>] ? clockevent_delta2ns+0xb/0xd
      [<ffffffff818a3a92>] ? setup_boot_APIC_clock+0x4a8/0x4b7
      [<ffffffff8153d955>] ? printk+0x72/0x74
      [<ffffffff818a1757>] native_smp_prepare_cpus+0x389/0x3d6
      [<ffffffff818957bc>] kernel_init_freeable+0xb7/0x1fb
      [<ffffffff81535530>] ? rest_init+0x74/0x74
      [<ffffffff81535539>] kernel_init+0x9/0xff
      [<ffffffff81541dfc>] ret_from_fork+0x7c/0xb0
      [<ffffffff81535530>] ? rest_init+0x74/0x74
    
    Getting this thing to work with the new mapping scheme would need more
    work, so automatically switch to the old memmap layout for SGI UV.

Signed-off-by Dave Young <[email protected]>

---
 kexec/arch/i386/x86-linux-setup.c |   15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

kexec/arch/i386/x86-linux-setup.c +11 -0

@@ -817,12 +817,25 @@ out:
 	return;
 }
 
+static int is_sgi_uv_machine(void)
+{
+	DIR *sgi_dir;
+
+	sgi_dir = opendir("/sys/firmware/sgi_uv");
+	if (sgi_dir) {
+		closedir(sgi_dir);
+		return 1;
+	}
+
+	return 0;

kexec/arch/i386/x86-linux-setup.c out: · line 839

Quoted 10 lines

kexec/arch/i386/x86-linux-setup.c +3 -1

+}
+
 void setup_linux_system_parameters(struct kexec_info *info,
 				   struct x86_linux_param_header *real_mode)
 {
 	/* get subarch from running kernel */
 	setup_subarch(real_mode);
-	if (bzImage_support_efi_boot)
+	if (bzImage_support_efi_boot && !is_sgi_uv_machine())
 		setup_efi_info(info, real_mode);

Hi Dave,

I think instead of always falling back to old "noefi" method for SGI UV machines, we should probably provide a config option to override default behavior. Something like "--nouefi". And let user/distributions decide in what cases to use that parameter.

Reason being, that at some point of time this limitation will be fixed. And at that point of time if we modify kexec-tools, then it will stop working with older kernels. And then same issue will arise that kexec-tools is not working with older kernels.

If we just provide a way to override default, then distributions in their scripts should know what kernels work what way and automatically use or not use option accordingly.

Thanks
Vivek

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

Signature · 5 lines

Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--

Quoted 10 lines · Borislav Petkov

Quoted 9 lines · Dave Young

Hi,

Vivek, replying with webmail, sorry for the formatting issue.

Another way is export the mapping scheme in sysfs. So if kernel boot with efi=old_map we can detect it as well. OTOH in future there's probably other quirk system.

So kexec-tools can switch to use old way if kernel use ioremap.

Quoted 13 lines

Quoted 13 lines

Adding Matt and Boris to cc, to see if they have idea. Thanks
Dave

----- Original Message -----
From: "Vivek Goyal" <[email protected]>
To: "Dave Young" <[email protected]>
Cc: [email protected], [email protected]
Sent: Friday, May 23, 2014 9:28:34 PM
Subject: Re: [PATCH] x86: do not enable efi boot for sgi uv machine

Quoted 1 line · Dave Young

Quoted 8 lines

Quoted 8 lines

Quoted 8 lines

kexec/kdump uefi support depends on the 1:1 mapping scheme in 3.14 kernel but SGI reported that 1:1 mapping does not work on their UV system. thus there's below commit to switch back to old ioremap way for sgi uv machines.

So for kexec-tools before we got the 1:1 mapping support on those machines we should also use old way that means boot 2nd kernel without uefi.

Yes, the old workaround to allow the kdump kernel to boot was to add the noefi bootline option, which also meant passing all the memory ranges as kernel boot line options (since noefi meant no EFI memory map info). An ugly situation. That is why we want to get Boris' mapping changes working on SGI UV, so that kexec/kdump work without ugly workarounds.

Quoted 6 lines

Well, we did find the bug in a combined debug session and SGI will be issuing a fix soon. I don't know whether any quirk will be needed for systems which don't have the fix - I guess that's something for SGI people to decide. CCed and leaving in the rest for reference.

@Alex, Russ: this is about using kexec on UEFI boxes.

Thanks for the heads up. Our bios made an assumption about mapping that was valid with the old mapping scheme but not with the new. Fortunately the fix will work with both old and new kernels, so the fix can be added in the next bios update. But we still have to be careful when removing the quirk, to avoid someone trying a new kernel with old bios and not being able to boot.

Quoted 89 lines

Quoted 88 lines

Quoted 55 lines

Tested on a uv1 machine.

Here is the kernel commit to quirk out SGI UV:
commit a5d90c923bcfb9632d998ed06e9569216ad695f3

Author: Borislav Petkov <[email protected]>
Date:   Tue Mar 4 17:02:17 2014 +0100

    x86/efi: Quirk out SGI UV
    
    Alex reported hitting the following BUG after the EFI 1:1 virtual
    mapping work was merged,
    
     kernel BUG at arch/x86/mm/init_64.c:351!
     invalid opcode: 0000 [#1] SMP
     Call Trace:
      [<ffffffff818aa71d>] init_extra_mapping_uc+0x13/0x15
      [<ffffffff818a5e20>] uv_system_init+0x22b/0x124b
      [<ffffffff8108b886>] ? clockevents_register_device+0x138/0x13d
      [<ffffffff81028dbb>] ? setup_APIC_timer+0xc5/0xc7
      [<ffffffff8108b620>] ? clockevent_delta2ns+0xb/0xd
      [<ffffffff818a3a92>] ? setup_boot_APIC_clock+0x4a8/0x4b7
      [<ffffffff8153d955>] ? printk+0x72/0x74
      [<ffffffff818a1757>] native_smp_prepare_cpus+0x389/0x3d6
      [<ffffffff818957bc>] kernel_init_freeable+0xb7/0x1fb
      [<ffffffff81535530>] ? rest_init+0x74/0x74
      [<ffffffff81535539>] kernel_init+0x9/0xff
      [<ffffffff81541dfc>] ret_from_fork+0x7c/0xb0
      [<ffffffff81535530>] ? rest_init+0x74/0x74
    
    Getting this thing to work with the new mapping scheme would need more
    work, so automatically switch to the old memmap layout for SGI UV.

Signed-off-by Dave Young <[email protected]>

---
 kexec/arch/i386/x86-linux-setup.c |   15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

kexec/arch/i386/x86-linux-setup.c +11 -0

@@ -817,12 +817,25 @@ out:
 	return;
 }
 
+static int is_sgi_uv_machine(void)
+{
+	DIR *sgi_dir;
+
+	sgi_dir = opendir("/sys/firmware/sgi_uv");
+	if (sgi_dir) {
+		closedir(sgi_dir);
+		return 1;
+	}
+
+	return 0;

kexec/arch/i386/x86-linux-setup.c out: · line 839

Quoted 10 lines

kexec/arch/i386/x86-linux-setup.c +3 -1

+}
+
 void setup_linux_system_parameters(struct kexec_info *info,
 				   struct x86_linux_param_header *real_mode)
 {
 	/* get subarch from running kernel */
 	setup_subarch(real_mode);
-	if (bzImage_support_efi_boot)
+	if (bzImage_support_efi_boot && !is_sgi_uv_machine())
 		setup_efi_info(info, real_mode);

Hi Dave,

I think instead of always falling back to old "noefi" method for SGI UV machines, we should probably provide a config option to override default behavior. Something like "--nouefi". And let user/distributions decide in what cases to use that parameter.
Reason being, that at some point of time this limitation will be fixed. And at that point of time if we modify kexec-tools, then it will stop working with older kernels. And then same issue will arise that kexec-tools is not working with older kernels.

If we just provide a way to override default, then distributions in their scripts should know what kernels work what way and automatically use or not use option accordingly.

Thanks
Vivek

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

Quoted 6 lines

Signature · 5 lines

Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--

Signature · 2 lines

Russ Anderson,  Kernel and Performance Software Team Manager
SGI - Silicon Graphics Inc          [email protected]

Quoted 11 lines · Russ Anderson

Quoted 10 lines · Borislav Petkov

Quoted 9 lines · Dave Young

Hi,

Vivek, replying with webmail, sorry for the formatting issue.

Another way is export the mapping scheme in sysfs. So if kernel boot with efi=old_map we can detect it as well. OTOH in future there's probably other quirk system.

So kexec-tools can switch to use old way if kernel use ioremap.

This sounds like a good approach to use, but I think we should still have a way to force it to use the old way if necessary (can't think of a use case for that off the top of my head, but I'm sure someone can :)

I think a sort of hybrid approach between detecting the mapping scheme, and having a command-line option would probably be best:

- Detect which mapping method we used.
- Default to noefi if the old mapping method was used, default to efi if new mapping was used.

- Allow users to specify either --efi or --noefi when kexec'ing to override the suggested default, determined above.

Quoted 6 lines

Well, we did find the bug in a combined debug session and SGI will be issuing a fix soon. I don't know whether any quirk will be needed for systems which don't have the fix - I guess that's something for SGI people to decide. CCed and leaving in the rest for reference.

@Alex, Russ: this is about using kexec on UEFI boxes.

Thanks for the heads up. Our bios made an assumption about mapping that was valid with the old mapping scheme but not with the new. Fortunately the fix will work with both old and new kernels, so the fix can be added in the next bios update. But we still have to be careful when removing the quirk, to avoid someone trying a new kernel with old bios and not being able to boot.

Quoted 7 lines

Quoted 6 lines

Well, we did find the bug in a combined debug session and SGI will be issuing a fix soon. I don't know whether any quirk will be needed for systems which don't have the fix - I guess that's something for SGI people to decide. CCed and leaving in the rest for reference.

@Alex, Russ: this is about using kexec on UEFI boxes.

Quoted 7 lines

Thanks for the heads up. Our bios made an assumption about mapping that was valid with the old mapping scheme but not with the new. Fortunately the fix will work with both old and new kernels, so the fix can be added in the next bios update. But we still have to be careful when removing the quirk, to avoid someone trying a new kernel with old bios and not being able to boot.

The quirk will quirk out all uv system, will SGI people provide a patch to only add quirk for old firmware version or just removing the quirk after the update is out?

Thanks
Dave

Quoted 7 lines

Well, we did find the bug in a combined debug session and SGI will be issuing a fix soon. I don't know whether any quirk will be needed for systems which don't have the fix - I guess that's something for SGI people to decide. CCed and leaving in the rest for reference.

@Alex, Russ: this is about using kexec on UEFI boxes.

Thanks for the heads up. Our bios made an assumption about mapping that was valid with the old mapping scheme but not with the new. Fortunately the fix will work with both old and new kernels, so the fix can be added in the next bios update. But we still have to be careful when removing the quirk, to avoid someone trying a new kernel with old bios and not being able to boot.

Hi,

Vivek, replying with webmail, sorry for the formatting issue.

Another way is export the mapping scheme in sysfs. So if kernel boot with efi=old_map we can detect it as well. OTOH in future there's probably other quirk system.

So kexec-tools can switch to use old way if kernel use ioremap.

Adding Matt and Boris to cc, to see if they have idea.

Thanks
Dave

----- Original Message -----
From: "Vivek Goyal" <[email protected]>
To: "Dave Young" <[email protected]>
Cc: [email protected], [email protected]
Sent: Friday, May 23, 2014 9:28:34 PM
Subject: Re: [PATCH] x86: do not enable efi boot for sgi uv machine

Quoted 65 lines · Dave Young

kexec/kdump uefi support depends on the 1:1 mapping scheme in 3.14 kernel but SGI reported that 1:1 mapping does not work on their UV system. thus there's below commit to switch back to old ioremap way for sgi uv machines.

So for kexec-tools before we got the 1:1 mapping support on those machines we should also use old way that means boot 2nd kernel without uefi.

Tested on a uv1 machine.

Here is the kernel commit to quirk out SGI UV:
commit a5d90c923bcfb9632d998ed06e9569216ad695f3

Author: Borislav Petkov <[email protected]>
Date:   Tue Mar 4 17:02:17 2014 +0100

    x86/efi: Quirk out SGI UV
    
    Alex reported hitting the following BUG after the EFI 1:1 virtual
    mapping work was merged,
    
     kernel BUG at arch/x86/mm/init_64.c:351!
     invalid opcode: 0000 [#1] SMP
     Call Trace:
      [<ffffffff818aa71d>] init_extra_mapping_uc+0x13/0x15
      [<ffffffff818a5e20>] uv_system_init+0x22b/0x124b
      [<ffffffff8108b886>] ? clockevents_register_device+0x138/0x13d
      [<ffffffff81028dbb>] ? setup_APIC_timer+0xc5/0xc7
      [<ffffffff8108b620>] ? clockevent_delta2ns+0xb/0xd
      [<ffffffff818a3a92>] ? setup_boot_APIC_clock+0x4a8/0x4b7
      [<ffffffff8153d955>] ? printk+0x72/0x74
      [<ffffffff818a1757>] native_smp_prepare_cpus+0x389/0x3d6
      [<ffffffff818957bc>] kernel_init_freeable+0xb7/0x1fb
      [<ffffffff81535530>] ? rest_init+0x74/0x74
      [<ffffffff81535539>] kernel_init+0x9/0xff
      [<ffffffff81541dfc>] ret_from_fork+0x7c/0xb0
      [<ffffffff81535530>] ? rest_init+0x74/0x74
    
    Getting this thing to work with the new mapping scheme would need more
    work, so automatically switch to the old memmap layout for SGI UV.

Signed-off-by Dave Young <[email protected]>

---
 kexec/arch/i386/x86-linux-setup.c |   15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

kexec/arch/i386/x86-linux-setup.c +11 -0

@@ -817,12 +817,25 @@ out:
 	return;
 }
 
+static int is_sgi_uv_machine(void)
+{
+	DIR *sgi_dir;
+
+	sgi_dir = opendir("/sys/firmware/sgi_uv");
+	if (sgi_dir) {
+		closedir(sgi_dir);
+		return 1;
+	}
+
+	return 0;

kexec/arch/i386/x86-linux-setup.c out: · line 839

Quoted 10 lines

kexec/arch/i386/x86-linux-setup.c +3 -1

+}
+
 void setup_linux_system_parameters(struct kexec_info *info,
 				   struct x86_linux_param_header *real_mode)
 {
 	/* get subarch from running kernel */
 	setup_subarch(real_mode);
-	if (bzImage_support_efi_boot)
+	if (bzImage_support_efi_boot && !is_sgi_uv_machine())
 		setup_efi_info(info, real_mode);

Hi Dave,

I think instead of always falling back to old "noefi" method for SGI UV machines, we should probably provide a config option to override default behavior. Something like "--nouefi". And let user/distributions decide in what cases to use that parameter.

Reason being, that at some point of time this limitation will be fixed. And at that point of time if we modify kexec-tools, then it will stop working with older kernels. And then same issue will arise that kexec-tools is not working with older kernels.

If we just provide a way to override default, then distributions in their scripts should know what kernels work what way and automatically use or not use option accordingly.

Thanks
Vivek

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

Quoted 8 lines · Dave Young

Hi,

Vivek, replying with webmail, sorry for the formatting issue.

Another way is export the mapping scheme in sysfs. So if kernel boot with efi=old_map we can detect it as well. OTOH in future there's probably other quirk system.

Quoted 1 line

So kexec-tools can switch to use old way if kernel use ioremap.

Exporting type of mapping being used in current kernel and automatically deciding that whether to boot with efi or noefi sounds reasonable to me. I am assuming that once type of mapping got fixed in first kernel boot it can't be changed in subsequent kernel boots?