Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, 31 May 2013, Matthew Garrett wrote: I agree that a revert is probably the right thing to do here, but the original patch was there to permit a more accurate calculation of the amount of nvram in use, not to provide additional debug information. Reverting it is going to differently break a different set of systems So differently break doesn't matter, if it's old breakage, and people thus don't really expect it to work. We need to fix bugs without *new* breakage, and quite frankly, I have been distressed by hearing the EFI specifications mentioned so many times in this thread. Firmware specs are pure and utter garbage. They are irrelevant. Firmware is buggy, and will always be buggy. The spec doesn't matter. We should use firmware for loading the kernel, and as little else as humanly possible. I'm very disappointed in how the EFI code doesn't seem to understand that. There's tons of these stupid EFI variable crap that simply shouldn't matter. Quite frankly, we'd be better off ignoring as much of it by default as at all possible. Exactly because the more of an EFI interface we have, the more we open us up to th einevitable firmware bugs. Anyway, I'm traveling with absolutely horrendous internet access, so can somebody please send a description of the revert with the relevant information, because I literally have a hard time extracting it all from this thread because my email access is so slow and flaky... Make it easy for me to do the revert with a good explanation message, please, Linus -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
* Russ Anderson r...@sgi.com wrote: On Thu, May 30, 2013 at 10:32:09PM +, Matthew Garrett wrote: On Thu, 2013-05-30 at 17:28 -0500, Russ Anderson wrote: On Thu, May 30, 2013 at 10:21:53PM +, Matthew Garrett wrote: On Thu, 2013-05-30 at 17:17 -0500, Russ Anderson wrote: That's a great idea. This patch moves the QueryVariableInfo() call from bootime to runtime, in efi_late_init(). The attached patch is consistent with the UEFI spec and avoids the problem. No, that defeats the entire point of the original patch. How so? It is still calling QueryVariableInfo() before the data is used. We want to know how much space is used by variables that aren't visible at runtime. We want to boot. We could boot up through 3.9-rc7. Knowing how much space is used by variables that aren't visible at runtime it moot if you can't boot. Exactly - fixing a boot regression is _WAY_ more important than pretty much any other concern. and the boot breakage is not limited to UV systems - the thread mentioned a couple of other systems as well. So it's an absolute no-brainer that this change should be reverted or fixed via your patch. Once a safer mechanism is implemented to call QueryVariableInfo() earlier (Boris's patches?) the change can be reintroduced. Thanks, Ingo -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, 31 May 2013, Ingo Molnar wrote: So this change needs to be reverted or fixed. I don't think anyone is arguing against that. My remark was purely to describe the current status quo and help to understand what exactly is happening, i.e.: - QueryVariableInfo() should be a valid thing to do from inside boot environment, according to the spec - now we see that at least SGI bios (an probably other incarnations) think otherwise - if we are not able to fix / work around the bug in BIOS (*), we have to make a choice between two evils -- either increase likelyhood of bricking certain machines due to filling the EFI storage space, or break booting on broken BIOSen - the theory is that Borislav's 1:1 mapping patches will work this around; one of the supporting arguments being that it's probably what Windows is doing. I believe Borislav is in the process of testing this. But the patches are not ready for mainline yet. (*) If one would be naive enough, he'd believe that the pressure should be put on the BIOS writers instead of OS to fix the bug :) -- Jiri Kosina SUSE Labs -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
* Borislav Petkov b...@alien8.de wrote: On Fri, May 31, 2013 at 01:06:09PM +0200, Jiri Kosina wrote: (*) If one would be naive enough, he'd believe that the pressure should be put on the BIOS writers instead of OS to fix the bug :) Oh, definitely pressure on BIOS dudes. If they're in violation of the spec and they still can fix it in time, they better. I'm sick and tired of having to deal with BIOS idiocy in kernel code. I'm all for some BIOS quality bashing, but the reality is: 1) It's not just about SGI/UV systems but apparently about several different types of x86 laptops produce the same boot crash pattern: most of which come from manufacturers that simply don't care about Linux all that much. So by not reverting we'd screw our users, not put any recognizable pressure on any BIOS writers or manufacturers. 2) Obviously Windows does not crash, and that's what most laptops test. So our realistic 'spec target' is not some sort of pure 'EFI spec', but EFI implementations _tested under Windows_. Consider it an 'extended EFI compatibility spec'. 3) There's a better, more robust firmware environment approach being worked on (by you?) that avoids such 1:1 physical mapping assumption crashes. That's something worth doing anyway, so why not delay the early QueryVariableInfo() call change to when that enviroment is properly implemented? 4) The revert is easy, and the functionality the original patch provided was a marginal increase in debug output to begin with... So to me the right approach seems to be: A: revert now for v3.10 B: implement 1:1 mappings environment for firmware, for v3.11 C: reintroduce the early QueryVariableInfo() call again, in v3.11 Thanks, Ingo -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, May 31, 2013 at 02:43:56PM +0200, Ingo Molnar wrote: 4) The revert is easy, and the functionality the original patch provided was a marginal increase in debug output to begin with... I agree that a revert is probably the right thing to do here, but the original patch was there to permit a more accurate calculation of the amount of nvram in use, not to provide additional debug information. Reverting it is going to differently break a different set of systems -- Matthew Garrett | mj...@srcf.ucam.org -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On 05/29/2013 01:16 PM, Russ Anderson wrote: To be more specific (now that I've dug through the code), grub2 (used by rhel7) uses EFI boot stubs. grub and elilo apparently do not use EFI boot stubs, so they don't hit the problem (at least on my test systems). Grub2 *can* use the EFI boot stub... as far as I know it doesn't by default, but the way it is used in RHEL does. Either way, we need to fix this and fix it soon. -hpa -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, 2013-05-31 at 15:34 +0100, Matthew Garrett wrote: On Fri, May 31, 2013 at 02:43:56PM +0200, Ingo Molnar wrote: 4) The revert is easy, and the functionality the original patch provided was a marginal increase in debug output to begin with... I agree that a revert is probably the right thing to do here, but the original patch was there to permit a more accurate calculation of the amount of nvram in use, not to provide additional debug information. Reverting it is going to differently break a different set of systems The only ones that are broken are the Samsung ones. Samsung claims to have fixed their UEFI firmware, so we could refer any problems to them. The signature of the Samsung failure, which this is guarding against is that the laptop gets bricked, so it really is a nasty choice of poisons we have to pick... Could we hedge the QueryVariableInfo checks with a test for Samsung in the UEFI identity strings? James -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, May 31, 2013 at 07:42:37AM -0700, James Bottomley wrote: On Fri, 2013-05-31 at 15:34 +0100, Matthew Garrett wrote: I agree that a revert is probably the right thing to do here, but the original patch was there to permit a more accurate calculation of the amount of nvram in use, not to provide additional debug information. Reverting it is going to differently break a different set of systems The only ones that are broken are the Samsung ones. Samsung claims to have fixed their UEFI firmware, so we could refer any problems to them. No, reverting this gets us back to the old state of refusing any writes if more than 50% of the variable store *appears* to be used, regardless of whether it's actually used. Which, unfortunately, makes it impossible to install Linux on most UEFI machines. In any case, Samsung clearly haven't fixed this problem on a pile of machines that have already shipped. Could we hedge the QueryVariableInfo checks with a test for Samsung in the UEFI identity strings? We could, but apparently some Lenovos also have a similar problem. We just don't have the information we need to implement a comprehensive blacklist, and if we get it wrong we're back to destroying people's hardware. -- Matthew Garrett | mj...@srcf.ucam.org -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, May 31, 2013 at 03:48:26PM +0100, Matthew Garrett wrote: On Fri, May 31, 2013 at 07:42:37AM -0700, James Bottomley wrote: On Fri, 2013-05-31 at 15:34 +0100, Matthew Garrett wrote: I agree that a revert is probably the right thing to do here, but the original patch was there to permit a more accurate calculation of the amount of nvram in use, not to provide additional debug information. Reverting it is going to differently break a different set of systems The only ones that are broken are the Samsung ones. Samsung claims to have fixed their UEFI firmware, so we could refer any problems to them. No, reverting this gets us back to the old state of refusing any writes if more than 50% of the variable store *appears* to be used, regardless of whether it's actually used. Which, unfortunately, makes it impossible to install Linux on most UEFI machines. When did writing EFI variables to nvram become necessary to boot on UEFI? And if it is necessary, why is it that only linux boot loaders that use EFI stubs (generally grub2) need it? The current kernel boots using EFI/grub and EFI/elilo. It is just when EFI stubs are used that the boot fails. I'm missing the background on why linux needs to write so many EFI variables to nvram that it fills up nvram. What is that all about? In any case, Samsung clearly haven't fixed this problem on a pile of machines that have already shipped. Which means the previous patch(es) that caused the bricking should get pulled, too. -- Russ Anderson, OS RAS/Partitioning Project Lead SGI - Silicon Graphics Inc r...@sgi.com -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, May 31, 2013 at 10:43:49AM -0500, Russ Anderson wrote: When did writing EFI variables to nvram become necessary to boot on UEFI? And if it is necessary, why is it that only linux boot loaders that use EFI stubs (generally grub2) need it? The current kernel boots using EFI/grub and EFI/elilo. It is just when EFI stubs are used that the boot fails. I think you've misunderstood the problem. If nvram becaomes full, some systems crash during firmware initialisation. So we can't let nvram become full. The obvious thing to do here is to look at the values from QueryVariableInfo, but many systems won't perform any garbage collection until they're almost out of space and so variables that have been deleted still show up as used space. We can work around that by adding up the size of the variables ourselves, but that only gives us the value for runtime-visible variables. We also need to know how much space is used by variables that are only visible during boot, hence calling QueryVariableInfo before ExitBootServices. Which means the previous patch(es) that caused the bricking should get pulled, too. There are no patches that cause the bricking. -- Matthew Garrett | mj...@srcf.ucam.org -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, 2013-05-31 at 17:28 +0100, Matthew Garrett wrote: On Fri, May 31, 2013 at 10:43:49AM -0500, Russ Anderson wrote: When did writing EFI variables to nvram become necessary to boot on UEFI? And if it is necessary, why is it that only linux boot loaders that use EFI stubs (generally grub2) need it? The current kernel boots using EFI/grub and EFI/elilo. It is just when EFI stubs are used that the boot fails. I think you've misunderstood the problem. If nvram becaomes full, some systems crash during firmware initialisation. So we can't let nvram become full. The obvious thing to do here is to look at the values from QueryVariableInfo, but many systems won't perform any garbage collection until they're almost out of space and so variables that have been deleted still show up as used space. We can work around that by adding up the size of the variables ourselves, but that only gives us the value for runtime-visible variables. We also need to know how much space is used by variables that are only visible during boot, hence calling QueryVariableInfo before ExitBootServices. Which means the previous patch(es) that caused the bricking should get pulled, too. There are no patches that cause the bricking. This is the description of the original problem: http://www.h-online.com/open/news/item/Booting-Linux-using-UEFI-can-brick-Samsung-laptops-1793958.html And the further investigation: http://mjg59.dreamwidth.org/22855.html If you read the latter, it shows you why we have to use QueryVariableInfo to try to work out how much space we have available. James -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, May 31, 2013 at 05:28:16PM +0100, Matthew Garrett wrote: On Fri, May 31, 2013 at 10:43:49AM -0500, Russ Anderson wrote: When did writing EFI variables to nvram become necessary to boot on UEFI? And if it is necessary, why is it that only linux boot loaders that use EFI stubs (generally grub2) need it? The current kernel boots using EFI/grub and EFI/elilo. It is just when EFI stubs are used that the boot fails. I think you've misunderstood the problem. The problem for me is the system not booting. I was not familiar with the problem you are trying to solve, but have now looked at the links. If nvram becaomes full, some systems crash during firmware initialisation. So we can't let nvram become full. The obvious thing to do here is to look at the values from QueryVariableInfo, but many systems won't perform any garbage collection until they're almost out of space and so variables that have been deleted still show up as used space. OK. I get nvram looks full due to lack of garbage collection on some systems. Does QueryVariableInfo (at runtime) tell you it is full? Is the problem that it says it is full when it is not, or does not tell you it is full when it is? We can work around that by adding up the size of the variables ourselves, but that only gives us the value for runtime-visible variables. We also need to know how much space is used by variables that are only visible during boot, Is it valid to assume that only the kernel writes to nvram at runtime? Can bios log error information to nvram on an SMM interrupt (for example)? hence calling QueryVariableInfo before ExitBootServices. Correct me if I am wrong, but that is called from EFI stubs, which is only used by some bootloaders (sometimes grub2). Does that mean EFI/grub and EFI/elilo will not hit the problem, or will not have your fix? Which means the previous patch(es) that caused the bricking should get pulled, too. There are no patches that cause the bricking. I thought that was the problem you were trying to avoid. -- Russ Anderson, OS RAS/Partitioning Project Lead SGI - Silicon Graphics Inc r...@sgi.com -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Sat, Jun 01, 2013 at 01:03:11AM +0100, Matthew Garrett wrote: On Fri, May 31, 2013 at 05:57:31PM -0500, Russ Anderson wrote: On Fri, May 31, 2013 at 05:28:16PM +0100, Matthew Garrett wrote: If nvram becaomes full, some systems crash during firmware initialisation. So we can't let nvram become full. The obvious thing to do here is to look at the values from QueryVariableInfo, but many systems won't perform any garbage collection until they're almost out of space and so variables that have been deleted still show up as used space. OK. I get nvram looks full due to lack of garbage collection on some systems. Does QueryVariableInfo (at runtime) tell you it is full? Is the problem that it says it is full when it is not, or does not tell you it is full when it is? QueryVariableInfo reports the amount used *including garbage*. This makes it useless for determining how much space is available for the firmware to use. So when QueryVariableInfo reports there is space available, the write works and everything is fine. And when QueryVariableInfo reports insufficient space, don't write and everything is fine. But when QueryVariableInfo reports insufficient space and you write anyway, the system can get bricked. Is that the problem? According to the UEFI spec for QueryVariableInfo() RemainingVariableStorageSize Returns the remaining size of the storage space available for EFI variables associated with the attributes specified. It also says: The returned MaximumVariableStorageSize, RemainingVariableStorageSize, MaximumVariableSize information may change immediately after the call based on other runtime activities including asynchronous error events. Also, these values associated with different attributes are not additive in nature. Note that other runtime activities including asynchronous error events. That means runtime services may be using nvram without the OS knowing about it. Some bios implementation may be *including garbage*, but can you definitively say ALL do? The spec explicitly says the implementation of variable storage is not defined in this specification. You can't assume any form of garbage collection. 7.2 Variable Services Although the implementation of variable storage is not defined in this specification, variables must be persistent in most cases. This implies that the EFI implementation on a platform must arrange it so that variables passed in for storage are retained and available for use each time the system boots, at least until they are explicitly deleted or overwritten. Provision of this type of nonvolatile storage may be very limited on some platforms, so variables should be used sparingly in cases where other means of communicating information cannot be used. I don't see anything in here about the OS being free to use more nvram than QueryVariableInfo() reported as being available. What happened to following the spec? We can work around that by adding up the size of the variables ourselves, but that only gives us the value for runtime-visible variables. We also need to know how much space is used by variables that are only visible during boot, Is it valid to assume that only the kernel writes to nvram at runtime? Can bios log error information to nvram on an SMM interrupt (for example)? There's a limited number of situations where the firmware can write to nvram, but they're basically uninteresting. Practically speaking, it's always via the kernel once ExitBootServices() has been called. How can you say they're basically uninteresting??? Do you know what EVERY bios implementation does??? hence calling QueryVariableInfo before ExitBootServices. Correct me if I am wrong, but that is called from EFI stubs, which is only used by some bootloaders (sometimes grub2). Does that mean EFI/grub and EFI/elilo will not hit the problem, or will not have your fix? Will not have the fix. Boot EFI systems via the EFI stub. Thanks for that clarification. Boot EFI systems via the EFI stub. Fortunately all our shipping systems are EFI/grub and EFI/elilo right now, so they boot. -- Russ Anderson, OS RAS/Partitioning Project Lead SGI - Silicon Graphics Inc r...@sgi.com -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Thu, May 30, 2013 at 10:16:12AM +0800, joeyli wrote:
於 四，2013-05-30 於 00:53 +0200，Jiri Kosina 提到：
On Wed, 29 May 2013, Russ Anderson wrote:
Yes, but this call is clearly happening way before ExitBootServices()
--
see the surrounding code, see for example this in efi_main():
[ ... snip ... ]
setup_efi_vars(boot_params);
setup_efi_pci(boot_params);
status = efi_call_phys3(sys_table-boottime-allocate_pool,
EFI_LOADER_DATA, sizeof(*gdt),
(void **)gdt);
if (status != EFI_SUCCESS) {
efi_printk(Failed to alloc mem for gdt structure\n);
goto fail;
}
[ ... snip ... ]
Yes. Note the failing call is sys_table-runtime while all the
other calls are sys_table-boottime and seem to work. Not sure
why the sys_table-runtime call has a problem but it may be
a clue. Could something in the runtime path not be set up???
That was my original idea early today as well. My understanding of the
UEFI spec is admittedly limited, but afaics calling runtime method from
boot environment should be a valid thing to do ... ?
QueryVariableInfo() is a runtime services, all runtime services should
available bother on boot time and runtime:
UEFI spec 2.3.1 P.109:
Runtime Services
Functions that are available before and after any call to
ExitBootServices(). These functions are described in Section 7.
That's a great idea. This patch moves the QueryVariableInfo()
call from bootime to runtime, in efi_late_init(). The attached
patch is consistent with the UEFI spec and avoids the problem.
Thanks,
--
Russ Anderson, OS RAS/Partitioning Project Lead
SGI - Silicon Graphics Inc r...@sgi.com
Move query_variable_info call to runtime to avoid bios issues.
Signed-off-by: Russ Anderson r...@sgi.com
---
arch/x86/boot/compressed/eboot.c | 49 ---
arch/x86/platform/efi/efi.c | 35 ---
2 files changed, 16 insertions(+), 68 deletions(-)
Index: linux/arch/x86/boot/compressed/eboot.c
===
--- linux.orig/arch/x86/boot/compressed/eboot.c 2013-05-30 11:02:19.034914824 -0500
+++ linux/arch/x86/boot/compressed/eboot.c 2013-05-30 16:53:50.512636568 -0500
@@ -251,53 +251,6 @@ static void find_bits(unsigned long mask
*size = len;
}
-static efi_status_t setup_efi_vars(struct boot_params *params)
-{
- struct setup_data *data;
- struct efi_var_bootdata *efidata;
- u64 store_size, remaining_size, var_size;
- efi_status_t status;
-
- if (sys_table-runtime-hdr.revision EFI_2_00_SYSTEM_TABLE_REVISION)
- return EFI_UNSUPPORTED;
-
- data = (struct setup_data *)(unsigned long)params-hdr.setup_data;
-
- while (data data-next)
- data = (struct setup_data *)(unsigned long)data-next;
-
- status = efi_call_phys4((void *)sys_table-runtime-query_variable_info,
-EFI_VARIABLE_NON_VOLATILE |
-EFI_VARIABLE_BOOTSERVICE_ACCESS |
-EFI_VARIABLE_RUNTIME_ACCESS, store_size,
-remaining_size, var_size);
-
- if (status != EFI_SUCCESS) { // RJA
- efi_printk(RJA: setup_efi_vars FAILED\n);
- return status;
- }
-
- status = efi_call_phys3(sys_table-boottime-allocate_pool,
-EFI_LOADER_DATA, sizeof(*efidata), efidata);
-
- if (status != EFI_SUCCESS)
- return status;
-
- efidata-data.type = SETUP_EFI_VARS;
- efidata-data.len = sizeof(struct efi_var_bootdata) -
- sizeof(struct setup_data);
- efidata-data.next = 0;
- efidata-store_size = store_size;
- efidata-remaining_size = remaining_size;
- efidata-max_var_size = var_size;
-
- if (data)
- data-next = (unsigned long)efidata;
- else
- params-hdr.setup_data = (unsigned long)efidata;
-
-}
-
static efi_status_t setup_efi_pci(struct boot_params *params)
{
efi_pci_io_protocol *pci;
@@ -1204,8 +1157,6 @@ struct boot_params *efi_main(void *handl
setup_graphics(boot_params);
- setup_efi_vars(boot_params);
-
setup_efi_pci(boot_params);
status = efi_call_phys3(sys_table-boottime-allocate_pool,
Index: linux/arch/x86/platform/efi/efi.c
===
--- linux.orig/arch/x86/platform/efi/efi.c 2013-05-30 11:02:19.034914824 -0500
+++ linux/arch/x86/platform/efi/efi.c 2013-05-30 17:05:38.140039879 -0500
@@ -786,9 +786,6 @@ void __init efi_init(void)
char vendor[100] = unknown;
int i = 0;
void *tmp;
- struct setup_data *data;
- struct efi_var_bootdata *efi_var_data;
- u64 pa_data;
#ifdef CONFIG_X86_32
if (boot_params.efi_info.efi_systab_hi ||
@@ -806,22 +803,6 @@ void __init efi_init(void)
if (efi_systab_init(efi_phys.systab))
return;
- pa_data = boot_params.hdr.setup_data;
- while (pa_data) {
- data = early_ioremap(pa_data, sizeof(*efi_var_data));
- if (data-type == SETUP_EFI_VARS) {
- efi_var_data = (struct efi_var_bootdata *)data;
-
-
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Thu, 2013-05-30 at 17:17 -0500, Russ Anderson wrote: That's a great idea. This patch moves the QueryVariableInfo() call from bootime to runtime, in efi_late_init(). The attached patch is consistent with the UEFI spec and avoids the problem. No, that defeats the entire point of the original patch. -- Matthew Garrett | mj...@srcf.ucam.org
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Thu, 30 May 2013, Russ Anderson wrote:
Yes, but this call is clearly happening way before ExitBootServices()
--
see the surrounding code, see for example this in efi_main():
[ ... snip ... ]
setup_efi_vars(boot_params);
setup_efi_pci(boot_params);
status = efi_call_phys3(sys_table-boottime-allocate_pool,
EFI_LOADER_DATA, sizeof(*gdt),
(void **)gdt);
if (status != EFI_SUCCESS) {
efi_printk(Failed to alloc mem for gdt structure\n);
goto fail;
}
[ ... snip ... ]
Yes. Note the failing call is sys_table-runtime while all the
other calls are sys_table-boottime and seem to work. Not sure
why the sys_table-runtime call has a problem but it may be
a clue. Could something in the runtime path not be set up???
That was my original idea early today as well. My understanding of the
UEFI spec is admittedly limited, but afaics calling runtime method from
boot environment should be a valid thing to do ... ?
QueryVariableInfo() is a runtime services, all runtime services should
available bother on boot time and runtime:
UEFI spec 2.3.1 P.109:
Runtime Services
Functions that are available before and after any call to
ExitBootServices(). These functions are described in Section 7.
That's a great idea. This patch moves the QueryVariableInfo()
call from bootime to runtime, in efi_late_init(). The attached
patch is consistent with the UEFI spec and avoids the problem.
Unfortunately that means that you can as well throw the patch away
completely.
The sole point is to run the QueryVariableInfo() from the boot
environment, in order to obtain more accurate information.
And it's a valid thing to do, according to UEFI specification.
--
Jiri Kosina
SUSE Labs
--
To unsubscribe from this list: send the line unsubscribe linux-efi in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Thu, May 30, 2013 at 10:21:53PM +, Matthew Garrett wrote: On Thu, 2013-05-30 at 17:17 -0500, Russ Anderson wrote: That's a great idea. This patch moves the QueryVariableInfo() call from bootime to runtime, in efi_late_init(). The attached patch is consistent with the UEFI spec and avoids the problem. No, that defeats the entire point of the original patch. How so? It is still calling QueryVariableInfo() before the data is used. -- Russ Anderson, OS RAS/Partitioning Project Lead SGI - Silicon Graphics Inc r...@sgi.com -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Thu, 2013-05-30 at 17:28 -0500, Russ Anderson wrote: On Thu, May 30, 2013 at 10:21:53PM +, Matthew Garrett wrote: On Thu, 2013-05-30 at 17:17 -0500, Russ Anderson wrote: That's a great idea. This patch moves the QueryVariableInfo() call from bootime to runtime, in efi_late_init(). The attached patch is consistent with the UEFI spec and avoids the problem. No, that defeats the entire point of the original patch. How so? It is still calling QueryVariableInfo() before the data is used. We want to know how much space is used by variables that aren't visible at runtime. -- Matthew Garrett | mj...@srcf.ucam.org
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, May 31, 2013 at 12:30:43AM +0200, Jiri Kosina wrote: On Thu, 30 May 2013, Russ Anderson wrote: That's a great idea. This patch moves the QueryVariableInfo() call from bootime to runtime, in efi_late_init(). The attached patch is consistent with the UEFI spec and avoids the problem. No, that defeats the entire point of the original patch. How so? It is still calling QueryVariableInfo() before the data is used. You lose information provided by QueryVariableInfo() about boot-only variables once the transition boot - runtime has happened. Is that information really more important than the ability to boot? Correct me if I'm wrong, but linux was able to boot without the boottime QueryVariableInfo() call up until 3.9-rc7, and it still does on systems that do not use EFI stubs (ie grub and elilo). It is only when linux uses EFI stubs (ie grub2) that linux makes the boottime QueryVariableInfo() call. So why is that call, or whatever is dependent on it, more important than booting? Thanks, -- Russ Anderson, OS RAS/Partitioning Project Lead SGI - Silicon Graphics Inc r...@sgi.com -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Thu, May 30, 2013 at 10:32:09PM +, Matthew Garrett wrote: On Thu, 2013-05-30 at 17:28 -0500, Russ Anderson wrote: On Thu, May 30, 2013 at 10:21:53PM +, Matthew Garrett wrote: On Thu, 2013-05-30 at 17:17 -0500, Russ Anderson wrote: That's a great idea. This patch moves the QueryVariableInfo() call from bootime to runtime, in efi_late_init(). The attached patch is consistent with the UEFI spec and avoids the problem. No, that defeats the entire point of the original patch. How so? It is still calling QueryVariableInfo() before the data is used. We want to know how much space is used by variables that aren't visible at runtime. We want to boot. We could boot up through 3.9-rc7. Knowing how much space is used by variables that aren't visible at runtime it moot if you can't boot. And again, maybe this is a bios bug - we have bios people looking into it - and maybe that call _should_ work, but the fact is the kernel booted without that change[1] and does not boot with it. [1] commit cc5a080c5d40c36089bb08a8a16fa3fc7047fe0f Thanks, -- Russ Anderson, OS RAS/Partitioning Project Lead SGI - Silicon Graphics Inc r...@sgi.com -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
於 四，2013-05-30 於 21:17 -0500，Russ Anderson 提到： On Fri, May 31, 2013 at 12:30:43AM +0200, Jiri Kosina wrote: On Thu, 30 May 2013, Russ Anderson wrote: That's a great idea. This patch moves the QueryVariableInfo() call from bootime to runtime, in efi_late_init(). The attached patch is consistent with the UEFI spec and avoids the problem. No, that defeats the entire point of the original patch. How so? It is still calling QueryVariableInfo() before the data is used. You lose information provided by QueryVariableInfo() about boot-only variables once the transition boot - runtime has happened. Is that information really more important than the ability to boot? Correct me if I'm wrong, but linux was able to boot without the boottime QueryVariableInfo() call up until 3.9-rc7, and it still does on systems that do not use EFI stubs (ie grub and elilo). It is only when linux uses EFI stubs (ie grub2) that linux makes the boottime QueryVariableInfo() call. So why is that call, or whatever is dependent on it, more important than booting? Thanks, It related to BIOS's garbage collection behavior of UEFI variable storage. The used space of non volatile boottime variables is useful to us for calculate the active_size, please reference 31ff2f2 patch: https://lkml.org/lkml/2013/4/15/476 An earlier 68d9298 patch to avoid some machines bricked when more than 50% of the system flash is in use, because the garbage collection will not trigger on those machines. We need find out the size of system flash space indeed usage for avoid this problem. So, cc5a080c5 patch call QueryVariableInfo() to grab the usage information in boot time. Calling QueryVariableInfo() at boot time should not causes side effect until your issue show up. Before this issue happen, avoid bricking some machines is also important. Thanks a lot! Joey Lee -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
於 四，2013-05-30 於 00:53 +0200，Jiri Kosina 提到：
On Wed, 29 May 2013, Russ Anderson wrote:
Yes, but this call is clearly happening way before ExitBootServices() --
see the surrounding code, see for example this in efi_main():
[ ... snip ... ]
setup_efi_vars(boot_params);
setup_efi_pci(boot_params);
status = efi_call_phys3(sys_table-boottime-allocate_pool,
EFI_LOADER_DATA, sizeof(*gdt),
(void **)gdt);
if (status != EFI_SUCCESS) {
efi_printk(Failed to alloc mem for gdt structure\n);
goto fail;
}
[ ... snip ... ]
Yes. Note the failing call is sys_table-runtime while all the
other calls are sys_table-boottime and seem to work. Not sure
why the sys_table-runtime call has a problem but it may be
a clue. Could something in the runtime path not be set up???
That was my original idea early today as well. My understanding of the
UEFI spec is admittedly limited, but afaics calling runtime method from
boot environment should be a valid thing to do ... ?
QueryVariableInfo() is a runtime services, all runtime services should
available bother on boot time and runtime:
UEFI spec 2.3.1 P.109:
Runtime Services
Functions that are available before and after any call to
ExitBootServices(). These functions are described in Section 7.
We are calling QueryVariableInfo() in setup_efi_vars(), and later on
AllocatePool is being called (through boot table).
On my system the QueryVariableInfo() call fails, so AllocatePool()
is not called in setup_efi_vars().
But it's being called later on coming back to efi_main(). That was just a
poor man's demonstration attempt why this code is running before
ExitBootServices() has been called.
Yes, I agreed your point, the space information of
EFI_VARIABLE_BOOTSERVICE_ACCESS should still return by
QueryVariableInfo() because we call it before ExitBootServices():
arch/x86/boot/compressed/eboot.c
efi_main(void *handle, efi_system_table_t *_table, struct boot_params
*boot_params)
..
sys_table = _table;
/* Check if we were booted by the EFI firmware */
if (sys_table-hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
goto fail;
boot_params-secure_boot = get_secure_boot(sys_table)
/* check does BIOS in secure boot mode */
setup_graphics(boot_params);
setup_efi_vars(boot_params);
/* Pass boot services variable info to runtime code, call QueryVariableInfo() */
...
status = exit_boot(boot_params, handle);
/* call ExitBootServices() */
...
Thanks a lot!
Joey Lee
--
To unsubscribe from this list: send the line unsubscribe linux-efi in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
於 三，2013-05-29 於 17:46 -0500，Russ Anderson 提到：
On Thu, May 30, 2013 at 12:22:13AM +0200, Jiri Kosina wrote:
On Wed, 29 May 2013, Russ Anderson wrote:
What appears to be happening is that your the EFI runtime services code
is calling into the EFI boot services code, which is definitely a bug in
your firmware because we're at runtime, but we've seen other machines
that do similar things so we usually handle it just fine. However, what
makes your case different, and the reason you see the above splat, is
that it's using the physical address of the EFI boot services region,
not the virtual one we setup with SetVirtualAddressMap(). Which is a
second firmware bug. Again, we have seen other machines that access
physical addresses after SetVirtualAddressMap(), but until now we
haven't had any non-optional code that triggered them.
The only reason I can see that the offending commit would introduce this
problem is because it calls QueryVariableInfo() at boot time. I notice
that your machine is an SGI UV one, is there any chance you could get a
firmware fix for this? If possible, it would be also good to confirm
that it's this chunk of code in setup_efi_vars(),
status = efi_call_phys4(sys_table-runtime-query_variable_info,
EFI_VARIABLE_NON_VOLATILE |
EFI_VARIABLE_BOOTSERVICE_ACCESS |
EFI_VARIABLE_RUNTIME_ACCESS,
store_size,
remaining_size, var_size);
This does trigger the problem. Note that the definition of
QueryVariableInfo() in the UEFI spec says:
The returned MaximumVariableStorageSize, RemainingVariableStorageSize,
MaximumVariableSize information may change immediately after the call
based on other runtime activities including asynchronous error events.
Also, these values associated with different attributes are not
additive in nature.
Note the values may be accurate at the point in time when returned,
but may not be after that.
After the system has transitioned into runtime (after
ExitBootServices() is called), an implementation may not be able to
accurately return information about the Boot Services variable store.
In such cases, EFI_INVALID_PARAMETER should be returned.
It is not clear to me exactly when ExitBootServices() is called.
Our bios is returning a failing indication on the call.
Yes, but this call is clearly happening way before ExitBootServices() --
see the surrounding code, see for example this in efi_main():
[ ... snip ... ]
setup_efi_vars(boot_params);
setup_efi_pci(boot_params);
status = efi_call_phys3(sys_table-boottime-allocate_pool,
EFI_LOADER_DATA, sizeof(*gdt),
(void **)gdt);
if (status != EFI_SUCCESS) {
efi_printk(Failed to alloc mem for gdt structure\n);
goto fail;
}
[ ... snip ... ]
Yes. Note the failing call is sys_table-runtime while all the
other calls are sys_table-boottime and seem to work. Not sure
why the sys_table-runtime call has a problem but it may be
a clue. Could something in the runtime path not be set up???
Per UEFI spec Section 6, all runtime services should a available both on
boot time and runtime. And, we query the EFI_VARIABLE_BOOTSERVICE_ACCESS
space information before ExitBootServices(), that means we call it in
boot time, so, QueryVariableInfo() should return information to us.
Back to the kernel oops, the oops happened in runtime environment when
efivar_init running. As Matt's point out as following:
於 五，2013-05-24 於 08:43 +0100，Matt Fleming 提到：
On Thu, 23 May, at 03:32:34PM, Russ Anderson wrote:
efi: mem127: type=4, attr=0xf,
range=[0x6bb22000-0x7ca9c000) (271MB)
EFI_BOOT_SERVICES_CODE
Per UEFI 2.3.1 Section 6.2, type 4 is EfiBootServicesCode. This area available
for
OS using after ExitBootServices():
UEFI 2.3.1 P.133
EfiBootServicesData Memory available for general use.
So, any runtime services should not access this area, it's the first thing need
to check
the BIOS code for why.
efi: mem133: type=5, attr=0x800f,
range=[0x7daff000-0x7dbff000) (1MB)
EFI_RUNTIME_SERVICES_CODE
EFI Variables Facility v0.08 2004-May-17
BUG: unable to handle kernel paging request at 7ca95b10
IP: [88007dbf2140] 0x88007dbf213f
[...]
What appears to be happening is that your the EFI runtime services code
is calling into the EFI boot services code, which is definitely a bug in
your firmware because we're at runtime, but we've seen other machines
that do similar things so we usually handle it just fine. However, what
makes your case different, and the reason you see the above splat, is
that it's using the
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
Hi Dave, 於 五，2013-05-24 於 17:05 -0400，Dave Jones 提到： On Fri, May 24, 2013 at 12:02:15PM -0500, Russ Anderson wrote: On Fri, May 24, 2013 at 11:11:11AM -0500, Robin Holt wrote: Russ, Can we open a bug for the BIOS folks and see if we can get this addressed? I already talked with them. It is not in an area that we normally change, so if there is a bug may be in the Intel reference code. More investigation is needed to track down the actual problem, and that could take help from Intel. Regardless of that, it is a kernel patch that triggers the problem. This isn't the first time a kernel change does the right thing but trips across questionable bios/EFI/bootloader implementation. That still makes it a kernel bug. I'm still digging to better understand the root problem. When we rebased the Fedora 18 kernel to 3.9 we had a bunch of reports from people who can no longer boot with what looks like similar symptoms. https://bugzilla.redhat.com/show_bug.cgi?id=964335 Dave The oops on the above bug is similar to my problem on Acer machine, could you please ask the reporter try the eccaf52f patch in urgent branch on Matt's efi git tree: https://git.kernel.org/cgit/linux/kernel/git/mfleming/efi.git/commit/?h=urgentid=eccaf52fee8305d5207ff110950a82c100e459bc Thanks! Joey Lee -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
於 一，2013-05-27 於 12:27 +0800，joeyli 提到： Hi Dave, 於 五，2013-05-24 於 17:05 -0400，Dave Jones 提到： On Fri, May 24, 2013 at 12:02:15PM -0500, Russ Anderson wrote: On Fri, May 24, 2013 at 11:11:11AM -0500, Robin Holt wrote: Russ, Can we open a bug for the BIOS folks and see if we can get this addressed? I already talked with them. It is not in an area that we normally change, so if there is a bug may be in the Intel reference code. More investigation is needed to track down the actual problem, and that could take help from Intel. Regardless of that, it is a kernel patch that triggers the problem. This isn't the first time a kernel change does the right thing but trips across questionable bios/EFI/bootloader implementation. That still makes it a kernel bug. I'm still digging to better understand the root problem. When we rebased the Fedora 18 kernel to 3.9 we had a bunch of reports from people who can no longer boot with what looks like similar symptoms. https://bugzilla.redhat.com/show_bug.cgi?id=964335 Dave The oops on the above bug is similar to my problem on Acer machine, could you please ask the reporter try the eccaf52f patch in urgent branch on Matt's efi git tree: https://git.kernel.org/cgit/linux/kernel/git/mfleming/efi.git/commit/?h=urgentid=eccaf52fee8305d5207ff110950a82c100e459bc Thanks! Joey Lee Looks there have a couple of different machines on this bug. The oops similar to my machine is reported by josephhenryblack. Thanks Joey Lee -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Thu, 23 May, at 05:23:21PM, Russ Anderson wrote: Interesting data point. The failure is on a rhel7/grub2 root. The identical kernel on a rhel6/grub root boots. So maybe grub2 brings out the failure? I suspect Fedora19/grub2 on EFI should hit the problem (for someone looking to reproduce it). In both cases the kernel boot line options are the same. I'll bet that rhel7 is using the EFI handover protocol which uses the internal mechanisms of the EFI boot stub. I don't know whether anyone will be able to reproduce it, it looks like it's a bug that's specific to your firmware. -- Matt Fleming, Intel Open Source Technology Center -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, May 24, 2013 at 08:43:31AM +0100, Matt Fleming wrote: What appears to be happening is that your the EFI runtime services code is calling into the EFI boot services code, which is definitely a bug in your firmware because we're at runtime, but we've seen other machines that do similar things so we usually handle it just fine. However, what makes your case different, and the reason you see the above splat, is that it's using the physical address of the EFI boot services region, not the virtual one we setup with SetVirtualAddressMap(). Which is a second firmware bug. I'm speechless. Let's have someone else do the ranting this time: http://www.happyassassin.net/2013/05/03/a-day-in-the-life-of-a-firmware-engineer/ Again, we have seen other machines that access physical addresses after SetVirtualAddressMap(), but until now we haven't had any non-optional code that triggered them. The only reason I can see that the offending commit would introduce this problem is because it calls QueryVariableInfo() at boot time. I notice that your machine is an SGI UV one, is there any chance you could get a firmware fix for this? If possible, it would be also good to confirm that it's this chunk of code in setup_efi_vars(), status = efi_call_phys4(sys_table-runtime-query_variable_info, EFI_VARIABLE_NON_VOLATILE | EFI_VARIABLE_BOOTSERVICE_ACCESS | EFI_VARIABLE_RUNTIME_ACCESS, store_size, remaining_size, var_size); that later makes GetNextVariable() jump to the physical address of the EFI Boot Services region. Because if not, we need to do some more digging. Borislav, how are your 1:1 mapping patches coming along? In theory, once those are merged we can gracefully workaround these kinds of issues. What do you mean, map boot time functions 1:1 too? In any case, I think I have an idea about the bug I was discussing with hpa recently but I need to do more experimenting. I have the next week off, though, so don't hold your breath just yet :). -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
Russ, Can we open a bug for the BIOS folks and see if we can get this addressed? Robin On Fri, May 24, 2013 at 08:43:31AM +0100, Matt Fleming wrote: On Thu, 23 May, at 03:32:34PM, Russ Anderson wrote: efi: mem127: type=4, attr=0xf, range=[0x6bb22000-0x7ca9c000) (271MB) EFI_BOOT_SERVICES_CODE efi: mem133: type=5, attr=0x800f, range=[0x7daff000-0x7dbff000) (1MB) EFI_RUNTIME_SERVICES_CODE EFI Variables Facility v0.08 2004-May-17 BUG: unable to handle kernel paging request at 7ca95b10 IP: [88007dbf2140] 0x88007dbf213f This... Call Trace: [81139a34] ? __alloc_pages_nodemask+0x154/0x2f0 [81174f7d] ? alloc_page_interleave+0x9d/0xa0 [812fe192] ? put_dec+0x72/0x90 [812f6d53] ? ida_get_new_above+0xb3/0x220 [812f6174] ? sub_alloc+0x74/0x1d0 [812f6174] ? sub_alloc+0x74/0x1d0 [812f6d53] ? ida_get_new_above+0xb3/0x220 [814c8cc0] ? create_efivars_bin_attributes+0x150/0x150 is junk on the stack. [810499b3] ? efi_call3+0x43/0x80 [810492a7] ? virt_efi_get_next_variable+0x47/0x1c0 [814c8cc0] ? create_efivars_bin_attributes+0x150/0x150 [814c7b55] ? efivar_init+0xd5/0x390 [814c8ae0] ? efivar_update_sysfs_entries+0x90/0x90 [812f906b] ? kobject_uevent+0xb/0x10 [812f812b] ? kset_register+0x5b/0x70 [814c8cc0] ? create_efivars_bin_attributes+0x150/0x150 [814c8d47] ? efivars_sysfs_init+0x87/0xf0 [8100032a] ? do_one_initcall+0x15a/0x1b0 [81a17831] ? do_basic_setup+0xad/0xce [81a17ae3] ? kernel_init_freeable+0x291/0x291 [81a3708a] ? sched_init_smp+0x15b/0x162 [81a17a5f] ? kernel_init_freeable+0x20d/0x291 [81601eb0] ? rest_init+0x80/0x80 [81601ebe] ? kernel_init+0xe/0x180 [8162179c] ? ret_from_fork+0x7c/0xb0 [81601eb0] ? rest_init+0x80/0x80 Here's the real call stack leading up to the crash. What appears to be happening is that your the EFI runtime services code is calling into the EFI boot services code, which is definitely a bug in your firmware because we're at runtime, but we've seen other machines that do similar things so we usually handle it just fine. However, what makes your case different, and the reason you see the above splat, is that it's using the physical address of the EFI boot services region, not the virtual one we setup with SetVirtualAddressMap(). Which is a second firmware bug. Again, we have seen other machines that access physical addresses after SetVirtualAddressMap(), but until now we haven't had any non-optional code that triggered them. The only reason I can see that the offending commit would introduce this problem is because it calls QueryVariableInfo() at boot time. I notice that your machine is an SGI UV one, is there any chance you could get a firmware fix for this? If possible, it would be also good to confirm that it's this chunk of code in setup_efi_vars(), status = efi_call_phys4(sys_table-runtime-query_variable_info, EFI_VARIABLE_NON_VOLATILE | EFI_VARIABLE_BOOTSERVICE_ACCESS | EFI_VARIABLE_RUNTIME_ACCESS, store_size, remaining_size, var_size); that later makes GetNextVariable() jump to the physical address of the EFI Boot Services region. Because if not, we need to do some more digging. Borislav, how are your 1:1 mapping patches coming along? In theory, once those are merged we can gracefully workaround these kinds of issues. -- Matt Fleming, Intel Open Source Technology Center -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, May 24, 2013 at 08:43:31AM +0100, Matt Fleming wrote: On Thu, 23 May, at 03:32:34PM, Russ Anderson wrote: efi: mem127: type=4, attr=0xf, range=[0x6bb22000-0x7ca9c000) (271MB) EFI_BOOT_SERVICES_CODE efi: mem133: type=5, attr=0x800f, range=[0x7daff000-0x7dbff000) (1MB) EFI_RUNTIME_SERVICES_CODE EFI Variables Facility v0.08 2004-May-17 BUG: unable to handle kernel paging request at 7ca95b10 IP: [88007dbf2140] 0x88007dbf213f [810499b3] ? efi_call3+0x43/0x80 [810492a7] ? virt_efi_get_next_variable+0x47/0x1c0 [814c8cc0] ? create_efivars_bin_attributes+0x150/0x150 [814c7b55] ? efivar_init+0xd5/0x390 [814c8ae0] ? efivar_update_sysfs_entries+0x90/0x90 [812f906b] ? kobject_uevent+0xb/0x10 [812f812b] ? kset_register+0x5b/0x70 [814c8cc0] ? create_efivars_bin_attributes+0x150/0x150 [814c8d47] ? efivars_sysfs_init+0x87/0xf0 [8100032a] ? do_one_initcall+0x15a/0x1b0 [81a17831] ? do_basic_setup+0xad/0xce [81a17ae3] ? kernel_init_freeable+0x291/0x291 [81a3708a] ? sched_init_smp+0x15b/0x162 [81a17a5f] ? kernel_init_freeable+0x20d/0x291 [81601eb0] ? rest_init+0x80/0x80 [81601ebe] ? kernel_init+0xe/0x180 [8162179c] ? ret_from_fork+0x7c/0xb0 [81601eb0] ? rest_init+0x80/0x80 Here's the real call stack leading up to the crash. What appears to be happening is that your the EFI runtime services code is calling into the EFI boot services code, which is definitely a bug in your firmware because we're at runtime, but we've seen other machines that do similar things so we usually handle it just fine. However, what makes your case different, and the reason you see the above splat, is that it's using the physical address of the EFI boot services region, not the virtual one we setup with SetVirtualAddressMap(). Which is a second firmware bug. Again, we have seen other machines that access physical addresses after SetVirtualAddressMap(), but until now we haven't had any non-optional code that triggered them. The only reason I can see that the offending commit would introduce this problem is because it calls QueryVariableInfo() at boot time. I notice that your machine is an SGI UV one, is there any chance you could get a firmware fix for this? If possible, it would be also good to confirm that it's this chunk of code in setup_efi_vars(), status = efi_call_phys4(sys_table-runtime-query_variable_info, EFI_VARIABLE_NON_VOLATILE | EFI_VARIABLE_BOOTSERVICE_ACCESS | EFI_VARIABLE_RUNTIME_ACCESS, store_size, remaining_size, var_size); This call is failing, but not returning a valid EFI_* return status. setup_efi_vars() returns at that point. Maybe it is not set up to do GetNextVariable() later on??? Why call GetNextVariable() if the earlier call failed? that later makes GetNextVariable() jump to the physical address of the EFI Boot Services region. Because if not, we need to do some more digging. One other data point is if the query_variable_info call is hacked to remove one of the EFI flags (ie comment out EFI_VARIABLE_BOOTSERVICE_ACCESS) the efi_call_phys4() call fails with EFI_INVALID_PARAMETER and the system boots. Of course it does not create /sys/firmware/efivars entries and complains [Firmware Bug]: efi: Inconsistent initial sizes. But at least it boots. One of the BIOS guys will build a debug bios next week to help see what is going on in the query_variable_info() call. Borislav, how are your 1:1 mapping patches coming along? In theory, once those are merged we can gracefully workaround these kinds of issues. -- Matt Fleming, Intel Open Source Technology Center -- Russ Anderson, OS RAS/Partitioning Project Lead SGI - Silicon Graphics Inc r...@sgi.com -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, 2013-05-24 at 15:05 -0500, Russ Anderson wrote:
One other data point is if the query_variable_info call is hacked to
remove one of the EFI flags (ie comment out EFI_VARIABLE_BOOTSERVICE_ACCESS)
the efi_call_phys4() call fails with EFI_INVALID_PARAMETER and
the system boots. Of course it does not create /sys/firmware/efivars
entries and complains [Firmware Bug]: efi: Inconsistent initial sizes.
But at least it boots.
EFI_VARIABLE_RUNTIME_ACCESS is only legal if
EFI_VARIABLE_BOOTSERVICE_ACCESS is set, so it's correct to throw
EFI_INVALID_PARAMETER there.
--
Matthew Garrett | mj...@srcf.ucam.org
N�r��yb�X��ǧv�^�){.n�+{�y^n�r���z���h��� �G���h�(�階�ݢj���m��z�ޖ���f���h���~�m�
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, May 24, 2013 at 08:11:01PM +, Matthew Garrett wrote: On Fri, 2013-05-24 at 15:05 -0500, Russ Anderson wrote: One other data point is if the query_variable_info call is hacked to remove one of the EFI flags (ie comment out EFI_VARIABLE_BOOTSERVICE_ACCESS) the efi_call_phys4() call fails with EFI_INVALID_PARAMETER and the system boots. Of course it does not create /sys/firmware/efivars entries and complains [Firmware Bug]: efi: Inconsistent initial sizes. But at least it boots. EFI_VARIABLE_RUNTIME_ACCESS is only legal if EFI_VARIABLE_BOOTSERVICE_ACCESS is set, so it's correct to throw EFI_INVALID_PARAMETER there. Yes. The point of the experiment was to see if it returned a valid failure status (which it did) and see if the boot still failed (which it didn't). So something about going deeper into that call seems to trigger the failure. Why does the kernel still try to create /sys/firmware/efivars/ entries in the original failure even though efi_call_phys4() failed? Or does it always try to create those entries and GetNextVariable() blows up in the original failure but not in my experiment? Thanks, -- Russ Anderson, OS RAS/Partitioning Project Lead SGI - Silicon Graphics Inc r...@sgi.com -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Fri, May 24, 2013 at 12:02:15PM -0500, Russ Anderson wrote: On Fri, May 24, 2013 at 11:11:11AM -0500, Robin Holt wrote: Russ, Can we open a bug for the BIOS folks and see if we can get this addressed? I already talked with them. It is not in an area that we normally change, so if there is a bug may be in the Intel reference code. More investigation is needed to track down the actual problem, and that could take help from Intel. Regardless of that, it is a kernel patch that triggers the problem. This isn't the first time a kernel change does the right thing but trips across questionable bios/EFI/bootloader implementation. That still makes it a kernel bug. I'm still digging to better understand the root problem. When we rebased the Fedora 18 kernel to 3.9 we had a bunch of reports from people who can no longer boot with what looks like similar symptoms. https://bugzilla.redhat.com/show_bug.cgi?id=964335 Dave -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Wed, 22 May, at 11:27:47AM, Russ Anderson wrote: [6.062157] EFI Variables Facility v0.08 2004-May-17 [6.067731] BUG: unable to handle kernel paging request at 7ca95b10 [6.075519] IP: [88007dbf2140] 0x88007dbf213f This is a bit of a head scratcher. Could you paste the EFI memmap entries in your dmesg for the regions that cover 0x7ca95b10 and 0x7dbf2140? My guess would be that they're EFI runtime code regions, which would at least explain why we seem to be executing code in the direct mapping region (0x8800). Are you booting via the EFI boot stub? -- Matt Fleming, Intel Open Source Technology Center -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [regression, bisected] x86: efi: Pass boot services variable info to runtime code
On Thu, May 23, 2013 at 12:58:01PM +0100, Matt Fleming wrote: On Wed, 22 May, at 11:27:47AM, Russ Anderson wrote: [6.062157] EFI Variables Facility v0.08 2004-May-17 [6.067731] BUG: unable to handle kernel paging request at 7ca95b10 [6.075519] IP: [88007dbf2140] 0x88007dbf213f This is a bit of a head scratcher. Could you paste the EFI memmap entries in your dmesg for the regions that cover 0x7ca95b10 and 0x7dbf2140? My guess would be that they're EFI runtime code regions, which would at least explain why we seem to be executing code in the direct mapping region (0x8800). Are you booting via the EFI boot stub? Interesting data point. The failure is on a rhel7/grub2 root. The identical kernel on a rhel6/grub root boots. So maybe grub2 brings out the failure? I suspect Fedora19/grub2 on EFI should hit the problem (for someone looking to reproduce it). In both cases the kernel boot line options are the same. -- Russ Anderson, OS RAS/Partitioning Project Lead SGI - Silicon Graphics Inc r...@sgi.com -- To unsubscribe from this list: send the line unsubscribe linux-efi in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
