Some sub-1MB memory regions may be reserved by EFI boot services, and the
memory regions will be released later in the efi_free_boot_services().
Currently, always reserve all sub-1MB memory regions when the crashkernel
option is specified, but unfortunately EFI boot services may have already
reserved some sub-1MB memory regions before the crash_reserve_low_1M() is
called, which makes that the crash_reserve_low_1M() only own the
remaining sub-1MB memory regions, not all sub-1MB memory regions, because,
subsequently EFI boot services will free its own sub-1MB memory regions.
Eventually, DMA will be able to allocate memory from the sub-1MB area and
cause the following error:
crash> kmem -s |grep invalid
kmem: dma-kmalloc-512: slab: ffffd52c40001900 invalid freepointer: ffff9403c0067300
kmem: dma-kmalloc-512: slab: ffffd52c40001900 invalid freepointer: ffff9403c0067300
crash> vtop ffff9403c0067300
VIRTUAL PHYSICAL
ffff9403c0067300 67300 --->The physical address falls into this range [0x0000000000063000-0x000000000008efff]
kernel debugging log:
...
[ 0.008927] memblock_reserve: [0x0000000000010000-0x0000000000013fff] efi_reserve_boot_services+0x85/0xd0
[ 0.008930] memblock_reserve: [0x0000000000063000-0x000000000008efff] efi_reserve_boot_services+0x85/0xd0
...
[ 0.009425] memblock_reserve: [0x0000000000000000-0x00000000000fffff] crash_reserve_low_1M+0x2c/0x49
...
[ 0.010586] Zone ranges:
[ 0.010587] DMA [mem 0x0000000000001000-0x0000000000ffffff]
[ 0.010589] DMA32 [mem 0x0000000001000000-0x00000000ffffffff]
[ 0.010591] Normal [mem 0x0000000100000000-0x0000000c7fffffff]
[ 0.010593] Device empty
...
[ 8.814894] __memblock_free_late: [0x0000000000063000-0x000000000008efff] efi_free_boot_services+0x14b/0x23b
[ 8.815793] __memblock_free_late: [0x0000000000010000-0x0000000000013fff] efi_free_boot_services+0x14b/0x23b
Do not release sub-1MB memory regions even though they are reserved by
EFI boot services, so that always reserve all sub-1MB memory regions when
the crashkernel option is specified.
Signed-off-by: Lianbo Jiang <[email protected]>
---
arch/x86/platform/efi/quirks.c | 14 ++++++++++++++
1 file changed, 14 insertions(+)
diff --git a/arch/x86/platform/efi/quirks.c b/arch/x86/platform/efi/quirks.c
index 67d93a243c35..637f932c4fd4 100644
--- a/arch/x86/platform/efi/quirks.c
+++ b/arch/x86/platform/efi/quirks.c
@@ -18,6 +18,7 @@
#include <asm/cpu_device_id.h>
#include <asm/realmode.h>
#include <asm/reboot.h>
+#include <asm/cmdline.h>
#define EFI_MIN_RESERVE 5120
@@ -303,6 +304,19 @@ void __init efi_arch_mem_reserve(phys_addr_t addr, u64 size)
*/
static __init bool can_free_region(u64 start, u64 size)
{
+ /*
+ * Some sub-1MB memory regions may be reserved by EFI boot
+ * services, and these memory regions will be released later
+ * in the efi_free_boot_services().
+ *
+ * Do not release sub-1MB memory regions even though they are
+ * reserved by EFI boot services, because, always reserve all
+ * sub-1MB memory when the crashkernel option is specified.
+ */
+ if (cmdline_find_option(boot_command_line, "crashkernel", NULL, 0) > 0
+ && (start + size < (1<<20)))
+ return false;
+
if (start + size > __pa_symbol(_text) && start <= __pa_symbol(_end))
return false;
--
2.17.1
On 04/07/21 at 10:03pm, Lianbo Jiang wrote:
> Some sub-1MB memory regions may be reserved by EFI boot services, and the
> memory regions will be released later in the efi_free_boot_services().
>
> Currently, always reserve all sub-1MB memory regions when the crashkernel
> option is specified, but unfortunately EFI boot services may have already
> reserved some sub-1MB memory regions before the crash_reserve_low_1M() is
> called, which makes that the crash_reserve_low_1M() only own the
> remaining sub-1MB memory regions, not all sub-1MB memory regions, because,
> subsequently EFI boot services will free its own sub-1MB memory regions.
> Eventually, DMA will be able to allocate memory from the sub-1MB area and
> cause the following error:
>
So this patch is fixing a problem found in crash utility. We ever met
the similar issue, later fixed by always reserving low 1M in commit
6f599d84231fd27 ("x86/kdump: Always reserve the low 1M when the crashkernel
option is specified"). Seems the commit is not fixing it completely.
> crash> kmem -s |grep invalid
> kmem: dma-kmalloc-512: slab: ffffd52c40001900 invalid freepointer: ffff9403c0067300
> kmem: dma-kmalloc-512: slab: ffffd52c40001900 invalid freepointer: ffff9403c0067300
> crash> vtop ffff9403c0067300
> VIRTUAL PHYSICAL
> ffff9403c0067300 67300 --->The physical address falls into this range [0x0000000000063000-0x000000000008efff]
>
> kernel debugging log:
> ...
> [ 0.008927] memblock_reserve: [0x0000000000010000-0x0000000000013fff] efi_reserve_boot_services+0x85/0xd0
> [ 0.008930] memblock_reserve: [0x0000000000063000-0x000000000008efff] efi_reserve_boot_services+0x85/0xd0
> ...
> [ 0.009425] memblock_reserve: [0x0000000000000000-0x00000000000fffff] crash_reserve_low_1M+0x2c/0x49
> ...
> [ 0.010586] Zone ranges:
> [ 0.010587] DMA [mem 0x0000000000001000-0x0000000000ffffff]
> [ 0.010589] DMA32 [mem 0x0000000001000000-0x00000000ffffffff]
> [ 0.010591] Normal [mem 0x0000000100000000-0x0000000c7fffffff]
> [ 0.010593] Device empty
> ...
> [ 8.814894] __memblock_free_late: [0x0000000000063000-0x000000000008efff] efi_free_boot_services+0x14b/0x23b
> [ 8.815793] __memblock_free_late: [0x0000000000010000-0x0000000000013fff] efi_free_boot_services+0x14b/0x23b
In commit 6f599d84231fd27, we call crash_reserve_low_1M() to lock the
whole low 1M area if crashkernel is specified in kernel cmdline.
But earlier efi_reserve_boot_services() invokation will break the
intention of the whole low 1M reserving. In efi_reserve_boot_services(),
if any memory under low 1M hasn't been reserved, it will call
memblock_reserve() to reserve it and leave it to
efi_free_boot_services() to free.
Hi Lianbo,
Please correct me if I am wrong or anything is missed. IIUC, can we move
efi_reserve_boot_services() after reserve_real_mode() to fix this bug?
Or move reserve_real_mode() before efi_reserve_boot_services() since
those real mode regions are all under 1M? Assume efi boot code/data
won't rely on low 1M area any more at this moment.
Thanks
Baoquan
>
> Do not release sub-1MB memory regions even though they are reserved by
> EFI boot services, so that always reserve all sub-1MB memory regions when
> the crashkernel option is specified.
>
> Signed-off-by: Lianbo Jiang <[email protected]>
> ---
> arch/x86/platform/efi/quirks.c | 14 ++++++++++++++
> 1 file changed, 14 insertions(+)
>
> diff --git a/arch/x86/platform/efi/quirks.c b/arch/x86/platform/efi/quirks.c
> index 67d93a243c35..637f932c4fd4 100644
> --- a/arch/x86/platform/efi/quirks.c
> +++ b/arch/x86/platform/efi/quirks.c
> @@ -18,6 +18,7 @@
> #include <asm/cpu_device_id.h>
> #include <asm/realmode.h>
> #include <asm/reboot.h>
> +#include <asm/cmdline.h>
>
> #define EFI_MIN_RESERVE 5120
>
> @@ -303,6 +304,19 @@ void __init efi_arch_mem_reserve(phys_addr_t addr, u64 size)
> */
> static __init bool can_free_region(u64 start, u64 size)
> {
> + /*
> + * Some sub-1MB memory regions may be reserved by EFI boot
> + * services, and these memory regions will be released later
> + * in the efi_free_boot_services().
> + *
> + * Do not release sub-1MB memory regions even though they are
> + * reserved by EFI boot services, because, always reserve all
> + * sub-1MB memory when the crashkernel option is specified.
> + */
> + if (cmdline_find_option(boot_command_line, "crashkernel", NULL, 0) > 0
> + && (start + size < (1<<20)))
> + return false;
> +
> if (start + size > __pa_symbol(_text) && start <= __pa_symbol(_end))
> return false;
>
> --
> 2.17.1
>
Hi, Baoquan
Thank you for the comment.
在 2021年04月09日 20:44, Baoquan He 写道:
> On 04/07/21 at 10:03pm, Lianbo Jiang wrote:
>> Some sub-1MB memory regions may be reserved by EFI boot services, and the
>> memory regions will be released later in the efi_free_boot_services().
>>
>> Currently, always reserve all sub-1MB memory regions when the crashkernel
>> option is specified, but unfortunately EFI boot services may have already
>> reserved some sub-1MB memory regions before the crash_reserve_low_1M() is
>> called, which makes that the crash_reserve_low_1M() only own the
>> remaining sub-1MB memory regions, not all sub-1MB memory regions, because,
>> subsequently EFI boot services will free its own sub-1MB memory regions.
>> Eventually, DMA will be able to allocate memory from the sub-1MB area and
>> cause the following error:
>>
>
> So this patch is fixing a problem found in crash utility. We ever met
> the similar issue, later fixed by always reserving low 1M in commit
> 6f599d84231fd27 ("x86/kdump: Always reserve the low 1M when the crashkernel
> option is specified"). Seems the commit is not fixing it completely.
>
Maybe I should add the "Fixes: 6f599d84231f" in front of 'Signed-off-by' as below:
Fixes: 6f599d84231f ("x86/kdump: Always reserve the low 1M when the crashkernel option is specified")
>> crash> kmem -s |grep invalid
>> kmem: dma-kmalloc-512: slab: ffffd52c40001900 invalid freepointer: ffff9403c0067300
>> kmem: dma-kmalloc-512: slab: ffffd52c40001900 invalid freepointer: ffff9403c0067300
>> crash> vtop ffff9403c0067300
>> VIRTUAL PHYSICAL
>> ffff9403c0067300 67300 --->The physical address falls into this range [0x0000000000063000-0x000000000008efff]
>>
>> kernel debugging log:
>> ...
>> [ 0.008927] memblock_reserve: [0x0000000000010000-0x0000000000013fff] efi_reserve_boot_services+0x85/0xd0
>> [ 0.008930] memblock_reserve: [0x0000000000063000-0x000000000008efff] efi_reserve_boot_services+0x85/0xd0
>> ...
>> [ 0.009425] memblock_reserve: [0x0000000000000000-0x00000000000fffff] crash_reserve_low_1M+0x2c/0x49
>> ...
>> [ 0.010586] Zone ranges:
>> [ 0.010587] DMA [mem 0x0000000000001000-0x0000000000ffffff]
>> [ 0.010589] DMA32 [mem 0x0000000001000000-0x00000000ffffffff]
>> [ 0.010591] Normal [mem 0x0000000100000000-0x0000000c7fffffff]
>> [ 0.010593] Device empty
>> ...
>> [ 8.814894] __memblock_free_late: [0x0000000000063000-0x000000000008efff] efi_free_boot_services+0x14b/0x23b
>> [ 8.815793] __memblock_free_late: [0x0000000000010000-0x0000000000013fff] efi_free_boot_services+0x14b/0x23b
>
>
> In commit 6f599d84231fd27, we call crash_reserve_low_1M() to lock the
> whole low 1M area if crashkernel is specified in kernel cmdline.
> But earlier efi_reserve_boot_services() invokation will break the
> intention of the whole low 1M reserving. In efi_reserve_boot_services(),
> if any memory under low 1M hasn't been reserved, it will call
> memblock_reserve() to reserve it and leave it to
> efi_free_boot_services() to free.
>
Good understanding.
> Hi Lianbo,
>
> Please correct me if I am wrong or anything is missed. IIUC, can we move
> efi_reserve_boot_services() after reserve_real_mode() to fix this bug?
What do you think about the following changes?
patch [1]:
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index 5ecd69a48393..c343de3178ec 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -1064,12 +1064,6 @@ void __init setup_arch(char **cmdline_p)
efi_esrt_init();
efi_mokvar_table_init();
- /*
- * The EFI specification says that boot service code won't be
- * called after ExitBootServices(). This is, in fact, a lie.
- */
- efi_reserve_boot_services();
-
/* preallocate 4k for mptable mpc */
e820__memblock_alloc_reserved_mpc_new();
@@ -1087,6 +1081,12 @@ void __init setup_arch(char **cmdline_p)
trim_platform_memory_ranges();
trim_low_memory_range();
+ /*
+ * The EFI specification says that boot service code won't be
+ * called after ExitBootServices(). This is, in fact, a lie.
+ */
+ efi_reserve_boot_services();
+
init_mem_mapping();
idt_setup_early_pf();
> Or move reserve_real_mode() before efi_reserve_boot_services() since
> those real mode regions are all under 1M? Assume efi boot code/data
Or patch [2]
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index 5ecd69a48393..ceec5af0dfab 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -1058,6 +1058,7 @@ void __init setup_arch(char **cmdline_p)
sev_setup_arch();
reserve_bios_regions();
+ reserve_real_mode();
efi_fake_memmap();
efi_find_mirror();
@@ -1082,8 +1083,6 @@ void __init setup_arch(char **cmdline_p)
(max_pfn_mapped<<PAGE_SHIFT) - 1);
#endif
- reserve_real_mode();
-
trim_platform_memory_ranges();
trim_low_memory_range();
I tested the above two changes on my machine, both work well.
But I'm not very sure if the reordering of the code may affect the startup(boot)
on other machines. Any comments about this?
Thanks.
Lianbo
> won't rely on low 1M area any more at this moment.
>
> Thanks
> Baoquan
>
>>
>> Do not release sub-1MB memory regions even though they are reserved by
>> EFI boot services, so that always reserve all sub-1MB memory regions when
>> the crashkernel option is specified.
>>
>> Signed-off-by: Lianbo Jiang <[email protected]>
>> ---
>> arch/x86/platform/efi/quirks.c | 14 ++++++++++++++
>> 1 file changed, 14 insertions(+)
>>
>> diff --git a/arch/x86/platform/efi/quirks.c b/arch/x86/platform/efi/quirks.c
>> index 67d93a243c35..637f932c4fd4 100644
>> --- a/arch/x86/platform/efi/quirks.c
>> +++ b/arch/x86/platform/efi/quirks.c
>> @@ -18,6 +18,7 @@
>> #include <asm/cpu_device_id.h>
>> #include <asm/realmode.h>
>> #include <asm/reboot.h>
>> +#include <asm/cmdline.h>
>>
>> #define EFI_MIN_RESERVE 5120
>>
>> @@ -303,6 +304,19 @@ void __init efi_arch_mem_reserve(phys_addr_t addr, u64 size)
>> */
>> static __init bool can_free_region(u64 start, u64 size)
>> {
>> + /*
>> + * Some sub-1MB memory regions may be reserved by EFI boot
>> + * services, and these memory regions will be released later
>> + * in the efi_free_boot_services().
>> + *
>> + * Do not release sub-1MB memory regions even though they are
>> + * reserved by EFI boot services, because, always reserve all
>> + * sub-1MB memory when the crashkernel option is specified.
>> + */
>> + if (cmdline_find_option(boot_command_line, "crashkernel", NULL, 0) > 0
>> + && (start + size < (1<<20)))
>> + return false;
>> +
>> if (start + size > __pa_symbol(_text) && start <= __pa_symbol(_end))
>> return false;
>>
>> --
>> 2.17.1
>>
On 04/09/21 at 07:59pm, H. Peter Anvin wrote:
> Why don't we do this unconditionally? At the very best we gain half a megabyte of memory (except the trampoline, which has to live there, but it is only a few kilobytes.)
This is a great suggestion, thanks. I think we can fix it in this way to
make code simpler. Then the specific caring of real mode in
efi_free_boot_services() can be removed too.
Thanks
Baoquan
> On Apr 11, 2021, at 6:14 PM, Baoquan He <[email protected]> wrote:
>
> On 04/09/21 at 07:59pm, H. Peter Anvin wrote:
>> Why don't we do this unconditionally? At the very best we gain half a megabyte of memory (except the trampoline, which has to live there, but it is only a few kilobytes.)
>
> This is a great suggestion, thanks. I think we can fix it in this way to
> make code simpler. Then the specific caring of real mode in
> efi_free_boot_services() can be removed too.
>
This whole situation makes me think that the code is buggy before and buggy after.
The issue here (I think) is that various pieces of code want to reserve specific pieces of otherwise-available low memory for their own nefarious uses. I don’t know *why* crash kernel needs this, but that doesn’t matter too much.
I propose that the right solution is to give low-memory-reserving code paths two chances to do what they need: once at the very beginning and once after EFI boot services are freed.
Alternatively, just reserve *all* otherwise unused sub 1M memory up front, then release it right after releasing boot services, and then invoke the special cases exactly once.
In either case, the result is that the crashkernel mess gets unified with the trampoline mess. One way the result is called twice and needs to be more careful, and the other way it’s called only once.
Just skipping freeing boot services seems wrong. It doesn’t unmap boot services, and skipping that is incorrect, I think. And it seems to result in a bogus memory map in which the system thinks that some crashkernel memory is EFI memory instead.
Let’s please just fix the problem instead of papering over it with more hacks.
> Thanks
> Baoquan
>
On 04/11/21 at 06:49pm, Andy Lutomirski wrote:
>
>
> > On Apr 11, 2021, at 6:14 PM, Baoquan He <[email protected]> wrote:
> >
> > On 04/09/21 at 07:59pm, H. Peter Anvin wrote:
> >> Why don't we do this unconditionally? At the very best we gain half a megabyte of memory (except the trampoline, which has to live there, but it is only a few kilobytes.)
> >
> > This is a great suggestion, thanks. I think we can fix it in this way to
> > make code simpler. Then the specific caring of real mode in
> > efi_free_boot_services() can be removed too.
> >
>
> This whole situation makes me think that the code is buggy before and buggy after.
>
> The issue here (I think) is that various pieces of code want to reserve specific pieces of otherwise-available low memory for their own nefarious uses. I don’t know *why* crash kernel needs this, but that doesn’t matter too much.
Kdump kernel also need go through real mode code path during bootup. It
is not different than normal kernel except that it skips the firmware
resetting. So kdump kernel needs low 1M as system RAM just as normal
kernel does. Here we reserve the whole low 1M with memblock_reserve()
to avoid any later kernel or driver data reside in this area. Otherwise,
we need dump the content of this area to vmcore. As we know, when crash
happened, the old memory of 1st kernel should be untouched until vmcore
dumping read out its content. Meanwhile, kdump kernel need reuse low 1M.
In the past, we used a back up region to copy out the low 1M area, and
map the back up region into the low 1M area in vmcore elf file. In
6f599d84231fd27 ("x86/kdump: Always reserve the low 1M when the crashkernel
option is specified"), we changed to lock the whole low 1M to avoid
writting any kernel data into, like this we can skip this area when
dumping vmcore.
Above is why we try to memblock reserve the whole low 1M. We don't want
to use it, just don't want anyone to use it in 1st kernel.
>
> I propose that the right solution is to give low-memory-reserving code paths two chances to do what they need: once at the very beginning and once after EFI boot services are freed.
>
> Alternatively, just reserve *all* otherwise unused sub 1M memory up front, then release it right after releasing boot services, and then invoke the special cases exactly once.
I am not sure if I got both suggested ways clearly. They look a little
complicated in our case. As I explained at above, we want the whole low
1M locked up, not one piece or some pieces of it.
>
> In either case, the result is that the crashkernel mess gets unified with the trampoline mess. One way the result is called twice and needs to be more careful, and the other way it’s called only once.
>
> Just skipping freeing boot services seems wrong. It doesn’t unmap boot services, and skipping that is incorrect, I think. And it seems to result in a bogus memory map in which the system thinks that some crashkernel memory is EFI memory instead.
I like hpa's thought to lock the whole low 1M unconditionally since only
a few KB except of trampoline area is there. Rethinking about it, doing
it in can_free_region() may be risky because efi memory region could
cross the 1M boundary, e.g [640K, 100M] with type of
EFI_BOOT_SERVICES_CODE|EFI_BOOT_SERVICES_DATA, it could cause loss of memory.
Just a wild guess, not very sure if the 1M boundary corssing can really
happen. efi_reserve_boot_services() won't split regions.
If moving efi_reserve_boot_services() after reserve_real_mode() is not
accepted, maybe we can call efi_mem_reserve(0, 1M) just as
efi_esrt_init() has done.
Thank you for the comment, H. Peter, Andy and Baoquan.
在 2021年04月12日 17:52, Baoquan He 写道:
> On 04/11/21 at 06:49pm, Andy Lutomirski wrote:
>>
>>
>>> On Apr 11, 2021, at 6:14 PM, Baoquan He <[email protected]> wrote:
>>>
>>> On 04/09/21 at 07:59pm, H. Peter Anvin wrote:
>>>> Why don't we do this unconditionally? At the very best we gain half a megabyte of memory (except the trampoline, which has to live there, but it is only a few kilobytes.)
>>>
>>> This is a great suggestion, thanks. I think we can fix it in this way to
>>> make code simpler. Then the specific caring of real mode in
>>> efi_free_boot_services() can be removed too.
>>>
>>
>> This whole situation makes me think that the code is buggy before and buggy after.
>>
>> The issue here (I think) is that various pieces of code want to reserve specific pieces of otherwise-available low memory for their own nefarious uses. I don’t know *why* crash kernel needs this, but that doesn’t matter too much.
>
> Kdump kernel also need go through real mode code path during bootup. It
> is not different than normal kernel except that it skips the firmware
> resetting. So kdump kernel needs low 1M as system RAM just as normal
> kernel does. Here we reserve the whole low 1M with memblock_reserve()
> to avoid any later kernel or driver data reside in this area. Otherwise,
> we need dump the content of this area to vmcore. As we know, when crash
> happened, the old memory of 1st kernel should be untouched until vmcore
> dumping read out its content. Meanwhile, kdump kernel need reuse low 1M.
> In the past, we used a back up region to copy out the low 1M area, and
> map the back up region into the low 1M area in vmcore elf file. In
> 6f599d84231fd27 ("x86/kdump: Always reserve the low 1M when the crashkernel
> option is specified"), we changed to lock the whole low 1M to avoid
> writting any kernel data into, like this we can skip this area when
> dumping vmcore.
>
> Above is why we try to memblock reserve the whole low 1M. We don't want
> to use it, just don't want anyone to use it in 1st kernel.
>
>>
>> I propose that the right solution is to give low-memory-reserving code paths two chances to do what they need: once at the very beginning and once after EFI boot services are freed.
>>
>> Alternatively, just reserve *all* otherwise unused sub 1M memory up front, then release it right after releasing boot services, and then invoke the special cases exactly once.
>
After EFI boot services are freed, I'm worried that it's a bit late. All sub-1M memory regions need to be reserved early as soon as possible.
> I am not sure if I got both suggested ways clearly. They look a little
> complicated in our case. As I explained at above, we want the whole low
> 1M locked up, not one piece or some pieces of it.
>
>>
>> In either case, the result is that the crashkernel mess gets unified with the trampoline mess. One way the result is called twice and needs to be more careful, and the other way it’s called only once.
>>
That may still have a chance to allocate memory from sub-1M regions at some point, because EFI boot services will be freed after EFI enters virtual mode, it looks late.
>> Just skipping freeing boot services seems wrong. It doesn’t unmap boot services, and skipping that is incorrect, I think. And it seems to result in a bogus memory map in which the system thinks that some crashkernel memory is EFI memory instead.
>
> I like hpa's thought to lock the whole low 1M unconditionally since only
> a few KB except of trampoline area is there. Rethinking about it, doing
> it in can_free_region() may be risky because efi memory region could
> cross the 1M boundary, e.g [640K, 100M] with type of
> EFI_BOOT_SERVICES_CODE|EFI_BOOT_SERVICES_DATA, it could cause loss of memory.
Theoretically, yes. But so far I haven't seen the situation of crossing the 1M boundary.
Thanks.
Lianbo
> Just a wild guess, not very sure if the 1M boundary corssing can really
> happen. efi_reserve_boot_services() won't split regions.
>
> If moving efi_reserve_boot_services() after reserve_real_mode() is not
> accepted, maybe we can call efi_mem_reserve(0, 1M) just as
> efi_esrt_init() has done.
>
On Mon, Apr 12, 2021 at 2:52 AM Baoquan He <[email protected]> wrote:
>
> On 04/11/21 at 06:49pm, Andy Lutomirski wrote:
> >
> >
> > > On Apr 11, 2021, at 6:14 PM, Baoquan He <[email protected]> wrote:
> > >
> > > On 04/09/21 at 07:59pm, H. Peter Anvin wrote:
> > >> Why don't we do this unconditionally? At the very best we gain half a megabyte of memory (except the trampoline, which has to live there, but it is only a few kilobytes.)
> > >
> > > This is a great suggestion, thanks. I think we can fix it in this way to
> > > make code simpler. Then the specific caring of real mode in
> > > efi_free_boot_services() can be removed too.
> > >
> >
> > This whole situation makes me think that the code is buggy before and buggy after.
> >
> > The issue here (I think) is that various pieces of code want to reserve specific pieces of otherwise-available low memory for their own nefarious uses. I don’t know *why* crash kernel needs this, but that doesn’t matter too much.
>
> Kdump kernel also need go through real mode code path during bootup. It
> is not different than normal kernel except that it skips the firmware
> resetting. So kdump kernel needs low 1M as system RAM just as normal
> kernel does. Here we reserve the whole low 1M with memblock_reserve()
> to avoid any later kernel or driver data reside in this area. Otherwise,
> we need dump the content of this area to vmcore. As we know, when crash
> happened, the old memory of 1st kernel should be untouched until vmcore
> dumping read out its content. Meanwhile, kdump kernel need reuse low 1M.
> In the past, we used a back up region to copy out the low 1M area, and
> map the back up region into the low 1M area in vmcore elf file. In
> 6f599d84231fd27 ("x86/kdump: Always reserve the low 1M when the crashkernel
> option is specified"), we changed to lock the whole low 1M to avoid
> writting any kernel data into, like this we can skip this area when
> dumping vmcore.
>
> Above is why we try to memblock reserve the whole low 1M. We don't want
> to use it, just don't want anyone to use it in 1st kernel.
>
> >
> > I propose that the right solution is to give low-memory-reserving code paths two chances to do what they need: once at the very beginning and once after EFI boot services are freed.
> >
> > Alternatively, just reserve *all* otherwise unused sub 1M memory up front, then release it right after releasing boot services, and then invoke the special cases exactly once.
>
> I am not sure if I got both suggested ways clearly. They look a little
> complicated in our case. As I explained at above, we want the whole low
> 1M locked up, not one piece or some pieces of it.
My second suggestion is probably the better one. Here it is, concretely:
The early (pre-free_efi_boot_services) code just reserves all
available sub-1M memory unconditionally, but it specially marks it as
reserved-but-available-later. We stop allocating the trampoline page
at this stage.
In free_efi_boot_services, instead of *freeing* the sub-1M memory, we
stick it in the pile of reserved memory created in the early step.
This may involve splitting a block, kind of like the current
trampoline late allocation works.
Then, *after* free_efi_boot_services(), we run a single block of code
that lets everything that wants sub-1M code claim some. This means
that the trampoline gets allocated and, if crashkernel wants to claim
everything else, it can. After that, everything still unclaimed gets
freed.
Does that make sense?
--Andy
On 04/12/21 at 08:24am, Andy Lutomirski wrote:
> On Mon, Apr 12, 2021 at 2:52 AM Baoquan He <[email protected]> wrote:
> >
> > On 04/11/21 at 06:49pm, Andy Lutomirski wrote:
> > >
> > >
> > > > On Apr 11, 2021, at 6:14 PM, Baoquan He <[email protected]> wrote:
> > > >
> > > > On 04/09/21 at 07:59pm, H. Peter Anvin wrote:
> > > >> Why don't we do this unconditionally? At the very best we gain half a megabyte of memory (except the trampoline, which has to live there, but it is only a few kilobytes.)
> > > >
> > > > This is a great suggestion, thanks. I think we can fix it in this way to
> > > > make code simpler. Then the specific caring of real mode in
> > > > efi_free_boot_services() can be removed too.
> > > >
> > >
> > > This whole situation makes me think that the code is buggy before and buggy after.
> > >
> > > The issue here (I think) is that various pieces of code want to reserve specific pieces of otherwise-available low memory for their own nefarious uses. I don’t know *why* crash kernel needs this, but that doesn’t matter too much.
> >
> > Kdump kernel also need go through real mode code path during bootup. It
> > is not different than normal kernel except that it skips the firmware
> > resetting. So kdump kernel needs low 1M as system RAM just as normal
> > kernel does. Here we reserve the whole low 1M with memblock_reserve()
> > to avoid any later kernel or driver data reside in this area. Otherwise,
> > we need dump the content of this area to vmcore. As we know, when crash
> > happened, the old memory of 1st kernel should be untouched until vmcore
> > dumping read out its content. Meanwhile, kdump kernel need reuse low 1M.
> > In the past, we used a back up region to copy out the low 1M area, and
> > map the back up region into the low 1M area in vmcore elf file. In
> > 6f599d84231fd27 ("x86/kdump: Always reserve the low 1M when the crashkernel
> > option is specified"), we changed to lock the whole low 1M to avoid
> > writting any kernel data into, like this we can skip this area when
> > dumping vmcore.
> >
> > Above is why we try to memblock reserve the whole low 1M. We don't want
> > to use it, just don't want anyone to use it in 1st kernel.
> >
> > >
> > > I propose that the right solution is to give low-memory-reserving code paths two chances to do what they need: once at the very beginning and once after EFI boot services are freed.
> > >
> > > Alternatively, just reserve *all* otherwise unused sub 1M memory up front, then release it right after releasing boot services, and then invoke the special cases exactly once.
> >
> > I am not sure if I got both suggested ways clearly. They look a little
> > complicated in our case. As I explained at above, we want the whole low
> > 1M locked up, not one piece or some pieces of it.
>
> My second suggestion is probably the better one. Here it is, concretely:
>
> The early (pre-free_efi_boot_services) code just reserves all
> available sub-1M memory unconditionally, but it specially marks it as
> reserved-but-available-later. We stop allocating the trampoline page
> at this stage.
>
> In free_efi_boot_services, instead of *freeing* the sub-1M memory, we
> stick it in the pile of reserved memory created in the early step.
> This may involve splitting a block, kind of like the current
> trampoline late allocation works.
>
> Then, *after* free_efi_boot_services(), we run a single block of code
> that lets everything that wants sub-1M code claim some. This means
> that the trampoline gets allocated and, if crashkernel wants to claim
> everything else, it can. After that, everything still unclaimed gets
> freed.
void __init setup_arch(char **cmdline_p)
{
...
efi_reserve_boot_services();
e820__memblock_alloc_reserved_mpc_new();
#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
setup_bios_corruption_check();
#endif
reserve_real_mode();
trim_platform_memory_ranges();
trim_low_memory_range();
...
}
After efi_reserve_boot_services(), there are several function calling to
require memory reservation under low 1M.
asmlinkage __visible void __init __no_sanitize_address start_kernel(void)
{
...
setup_arch(&command_line);
...
mm_init();
--> mem_init();
-->memblock_free_all();
...
#ifdef CONFIG_X86
if (efi_enabled(EFI_RUNTIME_SERVICES))
efi_enter_virtual_mode();
-->efi_free_boot_services();
-->memblock_free_late();
#endif
...
}
So from the code flow, we can see that buddy allocator is built in
mm_init() which puts all memory from memblock.memory excluding
memblock.reserved into buddy. And much later, we call
efi_free_boot_services() to release those reserved efi boot memory into
buddy too.
Are you suggesting we should do the memory reservation from low 1M
after efi_free_boot_services()? To require memory pages from buddy for
them? Please help point out my misunderstanding if have any.
With my understanding, in non-efi case, we have done the memory
reservation with memblock_reserve(), e.g
e820__memblock_alloc_reserved_mpc_new, reserve_real_mode() are calling
to do. Just efi_reserve|free_boot_services() break them when efi is
enabled. We can do them again in efi_free_boot_services() just like the
real_mode reservation does.
Thanks
Baoquan
Hi Lianbo,
On 05/24/21 at 11:00am, lijiang wrote:
> Also add mail lists and more people in the cc list.
>
> Thanks.
> Lianbo
>
>
> On Fri, May 21, 2021 at 8:36 PM lijiang <[email protected]> wrote:
>
> > Hi, Baoquan, Andy
> > Sorry for the late reply.
> >
> > On Tue, Apr 13, 2021 at 5:45 PM Baoquan He <[email protected]> wrote:
> >
> >> On 04/12/21 at 08:24am, Andy Lutomirski wrote:
> >> > On Mon, Apr 12, 2021 at 2:52 AM Baoquan He <[email protected]> wrote:
> >> > >
> >> > > On 04/11/21 at 06:49pm, Andy Lutomirski wrote:
> >> > > >
> >> > > >
> >> > > > > On Apr 11, 2021, at 6:14 PM, Baoquan He <[email protected]> wrote:
> >> > > > >
> >> > > > > On 04/09/21 at 07:59pm, H. Peter Anvin wrote:
> >> > > > >> Why don't we do this unconditionally? At the very best we gain
> >> half a megabyte of memory (except the trampoline, which has to live there,
> >> but it is only a few kilobytes.)
> >> > > > >
> >> > > > > This is a great suggestion, thanks. I think we can fix it in this
> >> way to
> >> > > > > make code simpler. Then the specific caring of real mode in
> >> > > > > efi_free_boot_services() can be removed too.
> >> > > > >
> >> > > >
> >> > > > This whole situation makes me think that the code is buggy before
> >> and buggy after.
> >> > > >
> >> > > > The issue here (I think) is that various pieces of code want to
> >> reserve specific pieces of otherwise-available low memory for their own
> >> nefarious uses. I don’t know *why* crash kernel needs this, but that
> >> doesn’t matter too much.
> >> > >
> >> > > Kdump kernel also need go through real mode code path during bootup.
> >> It
> >> > > is not different than normal kernel except that it skips the firmware
> >> > > resetting. So kdump kernel needs low 1M as system RAM just as normal
> >> > > kernel does. Here we reserve the whole low 1M with memblock_reserve()
> >> > > to avoid any later kernel or driver data reside in this area.
> >> Otherwise,
> >> > > we need dump the content of this area to vmcore. As we know, when
> >> crash
> >> > > happened, the old memory of 1st kernel should be untouched until
> >> vmcore
> >> > > dumping read out its content. Meanwhile, kdump kernel need reuse low
> >> 1M.
> >> > > In the past, we used a back up region to copy out the low 1M area, and
> >> > > map the back up region into the low 1M area in vmcore elf file. In
> >> > > 6f599d84231fd27 ("x86/kdump: Always reserve the low 1M when the
> >> crashkernel
> >> > > option is specified"), we changed to lock the whole low 1M to avoid
> >> > > writting any kernel data into, like this we can skip this area when
> >> > > dumping vmcore.
> >> > >
> >> > > Above is why we try to memblock reserve the whole low 1M. We don't
> >> want
> >> > > to use it, just don't want anyone to use it in 1st kernel.
> >> > >
> >> > > >
> >> > > > I propose that the right solution is to give low-memory-reserving
> >> code paths two chances to do what they need: once at the very beginning and
> >> once after EFI boot services are freed.
> >> > > >
> >> > > > Alternatively, just reserve *all* otherwise unused sub 1M memory up
> >> front, then release it right after releasing boot services, and then invoke
> >> the special cases exactly once.
> >> > >
> >> > > I am not sure if I got both suggested ways clearly. They look a little
> >> > > complicated in our case. As I explained at above, we want the whole
> >> low
> >> > > 1M locked up, not one piece or some pieces of it.
> >> >
> >> > My second suggestion is probably the better one. Here it is,
> >> concretely:
> >> >
> >> > The early (pre-free_efi_boot_services) code just reserves all
> >> > available sub-1M memory unconditionally, but it specially marks it as
> >> > reserved-but-available-later. We stop allocating the trampoline page
> >> > at this stage.
> >> >
> >> > In free_efi_boot_services, instead of *freeing* the sub-1M memory, we
> >> > stick it in the pile of reserved memory created in the early step.
> >> > This may involve splitting a block, kind of like the current
> >> > trampoline late allocation works.
> >> >
> >> > Then, *after* free_efi_boot_services(), we run a single block of code
> >> > that lets everything that wants sub-1M code claim some. This means
> >> > that the trampoline gets allocated and, if crashkernel wants to claim
> >> > everything else, it can. After that, everything still unclaimed gets
> >> > freed.
> >>
> >> void __init setup_arch(char **cmdline_p)
> >> {
> >> ...
> >> efi_reserve_boot_services();
> >> e820__memblock_alloc_reserved_mpc_new();
> >> #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
> >> setup_bios_corruption_check();
> >> #endif
> >> reserve_real_mode();
> >>
> >>
> >> trim_platform_memory_ranges();
> >> trim_low_memory_range();
> >> ...
> >> }
> >>
> >> After efi_reserve_boot_services(), there are several function calling to
> >> require memory reservation under low 1M.
> >>
> >>
> >> asmlinkage __visible void __init __no_sanitize_address
> >> start_kernel(void)
> >>
> >> {
> >> ...
> >> setup_arch(&command_line);
> >> ...
> >> mm_init();
> >> --> mem_init();
> >> -->memblock_free_all();
> >>
> >> ...
> >> #ifdef CONFIG_X86
> >> if (efi_enabled(EFI_RUNTIME_SERVICES))
> >> efi_enter_virtual_mode();
> >> -->efi_free_boot_services();
> >> -->memblock_free_late();
> >> #endif
> >> ...
> >> }
> >>
> >> So from the code flow, we can see that buddy allocator is built in
> >> mm_init() which puts all memory from memblock.memory excluding
> >> memblock.reserved into buddy. And much later, we call
> >> efi_free_boot_services() to release those reserved efi boot memory into
> >> buddy too.
> >>
> >> Are you suggesting we should do the memory reservation from low 1M
> >> after efi_free_boot_services()? To require memory pages from buddy for
> >> them? Please help point out my misunderstanding if have any.
> >>
> >> With my understanding, in non-efi case, we have done the memory
> >> reservation with memblock_reserve(), e.g
> >> e820__memblock_alloc_reserved_mpc_new, reserve_real_mode() are calling
> >> to do. Just efi_reserve|free_boot_services() break them when efi is
> >>
> >
> > Yes. But Andy also suggested to reserve all available sub-1M memory
> > **unconditionally**
> > in the early code.
> >
> > enabled. We can do them again in efi_free_boot_services() just like the
> >> real_mode reservation does.
> >>
> >
> > Do you mean to call the memblock_reserve() in
> > the efi_free_boot_services()? Or anything else?
> > Would you mind sharing more details about this?
Hmm, maybe no. There's no chance to call memblock_reserve() after
memblock_free_late().
I suggested making change in efi_free_boot_services() because the
similar problem has been encountered by Andy himself and fixed in
efi_free_boot_services(). Please check below commit for reference.
The patch described the phenemenon, while explained why in code
comment.
commit 5bc653b7318217c54244a14f248f1f07abe0a865
Author: Andy Lutomirski <[email protected]>
Date: Wed Aug 10 02:29:17 2016 -0700
x86/efi: Allocate a trampoline if needed in efi_free_boot_services()
So we could handle it in the same place by extending the area to the
whole low-1M unconditionally. This is 1st way of three I can think of.
The other two are:
2) Move efi_reserve_boot_services() down to be above init_mem_mapping().
until init_mem_mapping(), we stop reserving memory from low-1M with
memblock_reserve(). The change is like below:
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index 72920af0b3c0..93b2106d2050 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -1067,7 +1067,6 @@ void __init setup_arch(char **cmdline_p)
* The EFI specification says that boot service code won't be
* called after ExitBootServices(). This is, in fact, a lie.
*/
- efi_reserve_boot_services();
/* preallocate 4k for mptable mpc */
e820__memblock_alloc_reserved_mpc_new();
@@ -1090,6 +1089,8 @@ void __init setup_arch(char **cmdline_p)
*/
trim_snb_memory();
+ efi_reserve_boot_services();
+
init_mem_mapping();
idt_setup_early_pf();
3) Do it in efi_reserve_boot_services() as you are doing in this patch.
I prefer the 1st and 2nd way.
And by the way, the original code of commit 5bc653b731821("x86/efi: Allocate
a trampoline if needed in efi_free_boot_services()") could be buggy. I
remember you ever pasted the boot log of system where you reproduced
issue and tested the tested patch, there are three separate pages from
low-1M reserved by boot services. If any of them falls into real mode
area, it will break the fix of commit 5bc653b731821.
Thanks
Baoquan
> >
> > diff --git a/arch/x86/platform/efi/quirks.c
> > b/arch/x86/platform/efi/quirks.c
> > index 7850111008a8..d02f12a60457 100644
> > --- a/arch/x86/platform/efi/quirks.c
> > +++ b/arch/x86/platform/efi/quirks.c
> > @@ -453,6 +453,8 @@ void __init efi_free_boot_services(void)
> > memblock_free_late(start, size);
> > }
> >
> > + memblock_reserve(0, 1<<20);
> > +
> > if (!num_entries)
> > return;
> >
> > Thanks.
> > Lianbo
> >
> >
> > Thanks
> >> Baoquan
> >>
> >