x86/boot/compressed: Exclude 'top_pgtable' from relocation
PATCH 1/1 x86/boot/compressed: Exclude 'top_pgtable' from relocation
startup_64() copies kernel (including .data section) to the new place. It's required for safe in-place decompression.
This is a problem if the original place is referenced: by mistake I've put 'top_pgtable' into .data section and the address is loaded into CR3. If the original place gets overwritten during image decompression the kernel will crash and the machine will be rebooted.
Move 'top_pgtable' into .pgtable section where the rest of page tables are. This section is not subject for relocation.
--- arch/x86/boot/compressed/head_64.S | 8 ++++++++ arch/x86/boot/compressed/pgtable_64.c | 4 +--- 2 files changed, 9 insertions(+), 3 deletions(-)
arch/x86/boot/compressed/head_64.S +8 -0
@@ -649,3 +649,11 @@ boot_stack_end: .balign 4096 pgtable: .fill BOOT_PGT_SIZE, 1, 0 + +/* + * The page table is going to be used instead of page table in the trampoline + * memory. + */ + .global top_pgtable +top_pgtable: + .fill PAGE_SIZE, 1, 0
arch/x86/boot/compressed/pgtable_64.c +1 -3
@@ -25,10 +25,8 @@ static char trampoline_save[TRAMPOLINE_32BIT_SIZE];/* * The page table is going to be used instead of page table in the trampoline * memory. -* -* It must not be in BSS as BSS is cleared after cleanup_trampoline(). */ -static char top_pgtable[PAGE_SIZE] __aligned(PAGE_SIZE) __section(.data); +extern char *top_pgtable;/* * Trampoline address will be printed by extract_kernel() for debugging
Signature · 1 line
2.17.0
Quoted 8 lines · Kirill A. Shutemov
startup_64()copies kernel (including .data section) to the new place. It's required for safe in-place decompression.This is a problem if the original place is referenced: by mistake I've put '
top_pgtable' into .data section and the address is loaded into CR3. If the original place gets overwritten during image decompression the kernel will crash and the machine will be rebooted.
Quoted 5 lines
Move '
top_pgtable' into .pgtable section where the rest of page tables are. This section is not subject for relocation.Signed-off-by Kirill A. Shutemov <[email protected]>Fixes e9d0e6330eb8 ("x86/boot/compressed/64: Prepare new top-level page table for trampoline")
Thanks for the Cc, Kirill, which I presume was because I'd mentioned to you that I was unable to boot 4.17-rc on laptop or workstation.
Which is still so with 4.17-rc3, and I'm sorry to say still so with this
patch: even if I add the fix which I think this patch needs, see below.
I did bisect on Monday, and the first bad was your commit 194a9749c73d "x86/boot/compressed/64: Handle 5-level paging boot if kernel is above 4G". (Cc'ing Dave since his PTI Global work was my other suspect, but that's off the hook - if I revert just 194a9749c73d then I have no trouble.)
Am I really the only one getting immediate reboot on x86_64? Perhaps everyone else has machines with 5-level page tables now ?-)
I've looked at the changes a little, and tried a few things (hoping to avoid a long back and forth describing and trying things for you); but no success yet, and rather out of my depth with these changes - I've not had to delve into boot/compressed before.
(I did briefly get excited by thetrampoline_32bit + TRAMPOLINE_32BIT_PGTABLE_OFFSET
in cleanup_trampoline(), which lacks a "/ sizeof(unsigned long)"; but since ...PGTABLE_OFFSET is 0 anyway, that's nothing but cosmetic.)
Hugh
Quoted 26 lines
--- arch/x86/boot/compressed/head_64.S | 8 ++++++++ arch/x86/boot/compressed/pgtable_64.c | 4 +--- 2 files changed, 9 insertions(+), 3 deletions(-)arch/x86/boot/compressed/head_64.S +8
-0@@ -649,3 +649,11 @@ boot_stack_end: .balign 4096 pgtable: .fill BOOT_PGT_SIZE, 1, 0 + +/* + * The page table is going to be used instead of page table in the trampoline + * memory. + */ + .global top_pgtable +top_pgtable: + .fill PAGE_SIZE, 1, 0arch/x86/boot/compressed/pgtable_64.c +0
-0
arch/x86/boot/compressed/pgtable_64.c static char trampoline_save[TRAMPOLINE_32BIT_SIZE]; · line 29
Quoted 8 lines
arch/x86/boot/compressed/pgtable_64.c +1
-3@@ -25,10 +25,8 @@ static char trampoline_save[TRAMPOLINE_32BIT_SIZE];/* * The page table is going to be used instead of page table in the trampoline * memory. -* -* It must not be in BSS as BSS is cleared after cleanup_trampoline(). */ -static char top_pgtable[PAGE_SIZE] __aligned(PAGE_SIZE) __section(.data); +extern char *top_pgtable;
Doesn't that need to be extern char top_pgtable[] ?
Quoted 5 lines
arch/x86/boot/compressed/pgtable_64.c +0
-0/* * Trampoline address will be printed by extract_kernel() for debugging Signature · 1 line
2.17.0
Quoted 10 lines · Hugh Dickins
Quoted 8 lines · Kirill A. Shutemov
startup_64()copies kernel (including .data section) to the new place. It's required for safe in-place decompression.This is a problem if the original place is referenced: by mistake I've put '
top_pgtable' into .data section and the address is loaded into CR3. If the original place gets overwritten during image decompression the kernel will crash and the machine will be rebooted.
Quoted 8 lines
Quoted 5 lines
Move '
top_pgtable' into .pgtable section where the rest of page tables are. This section is not subject for relocation.Signed-off-by Kirill A. Shutemov <[email protected]>Fixes e9d0e6330eb8 ("x86/boot/compressed/64: Prepare new top-level page table for trampoline")Thanks for the Cc, Kirill, which I presume was because I'd mentioned to you that I was unable to boot 4.17-rc on laptop or workstation.
Right.
Quoted 2 lines
Which is still so with 4.17-rc3, and I'm sorry to say still so with this
patch: even if I add the fix which I think this patch needs, see below.
Hm.. Could you share your config?
IIRC, you use legacy boot. What bootloader?
Quoted 6 lines
I did bisect on Monday, and the first bad was your commit
194a9749c73d"x86/boot/compressed/64: Handle 5-level paging boot if kernel is above 4G". (Cc'ing Dave since his PTI Global work was my other suspect, but that's off the hook - if I revert just194a9749c73dthen I have no trouble.)Am I really the only one getting immediate reboot on
x86_64?
There was one more thread:
http://lkml.kernel.org/r/[email protected]
But no firm conclusion, only blaming GCC without a good reason. BTW, what compiler do you use?
Quoted 1 line
Perhaps everyone else has machines with 5-level page tables now ?-)
No :)
Quoted 9 lines
I've looked at the changes a little, and tried a few things (hoping to avoid a long back and forth describing and trying things for you); but no success yet, and rather out of my depth with these changes - I've not had to delve into boot/compressed before.
(I did briefly get excited by the
trampoline_32bit+TRAMPOLINE_32BIT_PGTABLE_OFFSET
incleanup_trampoline(), which lacks a "/ sizeof(unsigned long)"; but since ...PGTABLE_OFFSET is 0 anyway, that's nothing but cosmetic.)
It worth fixing anyway. Thanks for pointing it out.
Quoted 28 lines
Quoted 28 lines
--- arch/x86/boot/compressed/head_64.S | 8 ++++++++ arch/x86/boot/compressed/pgtable_64.c | 4 +--- 2 files changed, 9 insertions(+), 3 deletions(-)arch/x86/boot/compressed/head_64.S +8
-0@@ -649,3 +649,11 @@ boot_stack_end: .balign 4096 pgtable: .fill BOOT_PGT_SIZE, 1, 0 + +/* + * The page table is going to be used instead of page table in the trampoline + * memory. + */ + .global top_pgtable +top_pgtable: + .fill PAGE_SIZE, 1, 0arch/x86/boot/compressed/pgtable_64.c +0
-0@@ -25,10 +25,8 @@ static char trampoline_save[TRAMPOLINE_32BIT_SIZE];/* * The page table is going to be used instead of page table in the trampoline
Quoted 8 lines
arch/x86/boot/compressed/pgtable_64.c static char trampoline_save[TRAMPOLINE_32BIT_SIZE]; · line 29
Quoted 6 lines
arch/x86/boot/compressed/pgtable_64.c +1
-3* memory. -* -* It must not be in BSS as BSS is cleared after cleanup_trampoline(). */ -static char top_pgtable[PAGE_SIZE] __aligned(PAGE_SIZE) __section(.data); +extern char *top_pgtable;Doesn't that need to be extern char top_pgtable[] ?
Ouch. That's embarrassing.
So in my case the top_pgtable is zero and apparently it's good enough place to put the page table. It boots :P
The patch is bogus and I still don't understand what is going on.
Could you please check if bypassing cleanup_trampoline() altogether fixes the issue for you:
arch/x86/boot/compressed/head_64.S +0 -10
@@ -367,16 +367,6 @@ trampoline_return: /* Restore the stack, the 32-bit trampoline uses its own stack */ leaq boot_stack_end(%rbx), %rsp - /* - * cleanup_trampoline() would restore trampoline memory. - * - * RSI holds real mode data and needs to be preserved across - * this function call. - */ - pushq %rsi - call cleanup_trampoline - popq %rsi - /* Zero EFLAGS */ pushq $0 popfq
Signature · 1 line
Kirill A. Shutemov
Quoted 1 line · Kirill A. Shutemov
The patch is bogus and I still don't understand what is going on.
I think I found the issue. Could you check the patch below:
arch/x86/boot/compressed/head_64.S +12 -0
@@ -370,10 +370,13 @@ trampoline_return: /* * cleanup_trampoline() would restore trampoline memory. * + * RDI is address of the page table to use (if required). + * * RSI holds real mode data and needs to be preserved across * this function call. */ pushq %rsi + leaq top_pgtable(%rbx), %rdi call cleanup_trampoline popq %rsi
@@ -647,5 +650,14 @@ boot_stack_end: */ .section ".pgtable","a",@nobits .balign 4096 + .global pgtable pgtable: .fill BOOT_PGT_SIZE, 1, 0 + +/* + * The page table is going to be used instead of page table in the trampoline + * memory. + */ + .global top_pgtable +top_pgtable: + .fill PAGE_SIZE, 1, 0
arch/x86/boot/compressed/pgtable_64.c +3 -11
@@ -22,14 +22,6 @@ struct paging_config {/* Buffer to preserve trampoline memory */ static char trampoline_save[TRAMPOLINE_32BIT_SIZE ]; -/* -* The page table is going to be used instead of page table in the trampoline -* memory. -* -* It must not be in BSS as BSS is cleared after cleanup_trampoline(). -*/ -static char top_pgtable[PAGE_SIZE] __aligned(PAGE_SIZE) __section(.data); -/* * Trampoline address will be printed by extract_kernel() for debugging * purposes.
@@ -134,7 +126,7 @@ struct paging_config paging_prepare(void)return paging_config; } -void cleanup_trampoline (void ) +void cleanup_trampoline (void *pgtable) {void *trampoline_pgtable;
@@ -145,8 +137,8 @@ void cleanup_trampoline(void)* if it's there. */ if ((void *)__native_read_cr3 () == trampoline_pgtable) { -memcpy (top_pgtable, trampoline_pgtable,PAGE_SIZE ); -native_write_cr3 ((unsigned long )top_pgtable); +memcpy (pgtable, trampoline_pgtable,PAGE_SIZE ); +native_write_cr3 ((unsigned long )pgtable); }/* Restore trampoline memory */
Signature · 1 line
Kirill A. Shutemov
Quoted 4 lines · Kirill A. Shutemov
Quoted 1 line · Kirill A. Shutemov
The patch is bogus and I still don't understand what is going on.
I think I found the issue. Could you check the patch below:
Sorry, no good on either machine, immediate reboot as before. I'm gathering the info you asked, will send privately in an hour or so.
Hugh
Quoted 76 lines
arch/x86/boot/compressed/head_64.S +12
-0@@ -370,10 +370,13 @@ trampoline_return: /* * cleanup_trampoline() would restore trampoline memory. * + * RDI is address of the page table to use (if required). + * * RSI holds real mode data and needs to be preserved across * this function call. */ pushq %rsi + leaq top_pgtable(%rbx), %rdi call cleanup_trampoline popq %rsi@@ -647,5 +650,14 @@ boot_stack_end: */ .section ".pgtable","a",@nobits .balign 4096 + .global pgtable pgtable: .fill BOOT_PGT_SIZE, 1, 0 + +/* + * The page table is going to be used instead of page table in the trampoline + * memory. + */ + .global top_pgtable +top_pgtable: + .fill PAGE_SIZE, 1, 0arch/x86/boot/compressed/pgtable_64.c +3
-11@@ -22,14 +22,6 @@ struct paging_config {/* Buffer to preserve trampoline memory */ static char trampoline_save[TRAMPOLINE_32BIT_SIZE ]; -/* -* The page table is going to be used instead of page table in the trampoline -* memory. -* -* It must not be in BSS as BSS is cleared after cleanup_trampoline(). -*/ -static char top_pgtable[PAGE_SIZE] __aligned(PAGE_SIZE) __section(.data); -/* * Trampoline address will be printed by extract_kernel() for debugging * purposes. @@ -134,7 +126,7 @@ struct paging_config paging_prepare(void)return paging_config; } -void cleanup_trampoline (void ) +void cleanup_trampoline (void *pgtable) {void *trampoline_pgtable;@@ -145,8 +137,8 @@ void cleanup_trampoline(void)* if it's there. */ if ((void *)__native_read_cr3 () == trampoline_pgtable) { -memcpy (top_pgtable, trampoline_pgtable,PAGE_SIZE ); -native_write_cr3 ((unsigned long )top_pgtable); +memcpy (pgtable, trampoline_pgtable,PAGE_SIZE ); +native_write_cr3 ((unsigned long )pgtable); }/* Restore trampoline memory */ Signature · 1 line
Kirill A. Shutemov
Quoted 9 lines · Hugh Dickins
Quoted 4 lines · Kirill A. Shutemov
Quoted 1 line · Kirill A. Shutemov
The patch is bogus and I still don't understand what is going on.
I think I found the issue. Could you check the patch below:
Sorry, no good on either machine, immediate reboot as before. I'm gathering the info you asked, will send privately in an hour or so.
Hi, any update on this bug?
Kirill: mind (re-)sending a series of any resulting patches that you have pending?
If there's no fix, is there any list of SHA1's to revert?
Thanks,
Ingo
Quoted 10 lines · Ingo Molnar
Quoted 7 lines · Hugh Dickins
Quoted 4 lines · Kirill A. Shutemov
Quoted 1 line · Kirill A. Shutemov
The patch is bogus and I still don't understand what is going on.
I think I found the issue. Could you check the patch below:
Quoted 8 lines
Quoted 2 lines
Sorry, no good on either machine, immediate reboot as before. I'm gathering the info you asked, will send privately in an hour or so.
Hi, any update on this bug?
Kirill: mind (re-)sending a series of any resulting patches that you have pending?
If there's no fix, is there any list of SHA1's to revert?
The root cause of the bug was fixed in v4.17 by
5c9b0b1c4988 ("x86/boot/compressed/64: Set up GOT for paging_prepare() and cleanup_trampoline()").
The only pending patch I have at the moment is
http://lkml.kernel.org/r/[email protected]
Do you want me to resend it again?
Signature · 1 line
Kirill A. Shutemov
Quoted 5 lines · Kirill A. Shutemov
Quoted 1 line
Kirill: mind (re-)sending a series of any resulting patches that you have pending?
The root cause of the bug was fixed in v4.17 by
5c9b0b1c4988("x86/boot/compressed/64: Set up GOT forpaging_prepare()andcleanup_trampoline()").
Ok, cool!
Quoted 5 lines
The only pending patch I have at the moment is
http://lkml.kernel.org/r/[email protected]
Do you want me to resend it again?
No need, I think Thomas is handling that one.
Thanks,
Ingo