2020-03-02 08:28:56

by Naresh Kamboju

[permalink] [raw]
Subject: Linux-next-20200302: arm64 build failed

Linux-Next 20200302 arm64 build failed due to below errors,
Suspecting patch causing this build break.

87d900aef3e2 arm/arm64: add support for folded p4d page tables

Error log,
-------------
arch/arm64/mm/mmu.c: In function 'unmap_hotplug_pud_range':
include/linux/compiler.h:284:1: error: incompatible type for argument
1 of 'p4d_page_paddr'
({ \
^
arch/arm64/include/asm/memory.h:270:45: note: in definition of macro
'__phys_to_virt'
#define __phys_to_virt(x) ((unsigned long)((x) - physvirt_offset))
^
arch/arm64/include/asm/pgtable.h:629:42: note: in expansion of macro '__va'
#define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
^~~~
include/linux/compiler.h:293:22: note: in expansion of macro '__READ_ONCE'
#define READ_ONCE(x) __READ_ONCE(x, 1)
^~~~~~~~~~~
arch/arm64/include/asm/pgtable.h:628:52: note: in expansion of macro 'READ_ONCE'
#define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir)))
+ pud_index(addr) * sizeof(pud_t))
^~~~~~~~~
arch/arm64/include/asm/pgtable.h:629:47: note: in expansion of macro
'pud_offset_phys'
#define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
^~~~~~~~~~~~~~~
arch/arm64/mm/mmu.c:827:10: note: in expansion of macro 'pud_offset'
pudp = pud_offset(pgdp, addr);
^~~~~~~~~~

ref:
https://ci.linaro.org/view/lkft/job/openembedded-lkft-linux-next/DISTRO=lkft,MACHINE=juno,label=docker-lkft/716/consoleText

--
Linaro LKFT
https://lkft.linaro.org


2020-03-02 10:49:00

by Will Deacon

[permalink] [raw]
Subject: Re: Linux-next-20200302: arm64 build failed

[+Anshuman and Catalin]

On Mon, Mar 02, 2020 at 01:58:26PM +0530, Naresh Kamboju wrote:
> Linux-Next 20200302 arm64 build failed due to below errors,
> Suspecting patch causing this build break.
>
> 87d900aef3e2 arm/arm64: add support for folded p4d page tables
>
> Error log,
> -------------
> arch/arm64/mm/mmu.c: In function 'unmap_hotplug_pud_range':
> include/linux/compiler.h:284:1: error: incompatible type for argument
> 1 of 'p4d_page_paddr'
> ({ \
> ^
> arch/arm64/include/asm/memory.h:270:45: note: in definition of macro
> '__phys_to_virt'
> #define __phys_to_virt(x) ((unsigned long)((x) - physvirt_offset))
> ^
> arch/arm64/include/asm/pgtable.h:629:42: note: in expansion of macro '__va'
> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
> ^~~~
> include/linux/compiler.h:293:22: note: in expansion of macro '__READ_ONCE'
> #define READ_ONCE(x) __READ_ONCE(x, 1)
> ^~~~~~~~~~~
> arch/arm64/include/asm/pgtable.h:628:52: note: in expansion of macro 'READ_ONCE'
> #define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir)))
> + pud_index(addr) * sizeof(pud_t))
> ^~~~~~~~~
> arch/arm64/include/asm/pgtable.h:629:47: note: in expansion of macro
> 'pud_offset_phys'
> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
> ^~~~~~~~~~~~~~~
> arch/arm64/mm/mmu.c:827:10: note: in expansion of macro 'pud_offset'
> pudp = pud_offset(pgdp, addr);
> ^~~~~~~~~~

Looks like we need an implementation of unmap_hotplug_p4d_range() to
walk the dummy p4d level. Unfortunately, we don't have the folded p4d
patches in the arm64 tree so we'll either need a common branch or the
hotplug patches will need to be dropped for the moment.

Will

2020-03-02 13:49:05

by Anshuman Khandual

[permalink] [raw]
Subject: Re: Linux-next-20200302: arm64 build failed



On 03/02/2020 04:17 PM, Will Deacon wrote:
> [+Anshuman and Catalin]
>
> On Mon, Mar 02, 2020 at 01:58:26PM +0530, Naresh Kamboju wrote:
>> Linux-Next 20200302 arm64 build failed due to below errors,
>> Suspecting patch causing this build break.
>>
>> 87d900aef3e2 arm/arm64: add support for folded p4d page tables
>>
>> Error log,
>> -------------
>> arch/arm64/mm/mmu.c: In function 'unmap_hotplug_pud_range':
>> include/linux/compiler.h:284:1: error: incompatible type for argument
>> 1 of 'p4d_page_paddr'
>> ({ \
>> ^
>> arch/arm64/include/asm/memory.h:270:45: note: in definition of macro
>> '__phys_to_virt'
>> #define __phys_to_virt(x) ((unsigned long)((x) - physvirt_offset))
>> ^
>> arch/arm64/include/asm/pgtable.h:629:42: note: in expansion of macro '__va'
>> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
>> ^~~~
>> include/linux/compiler.h:293:22: note: in expansion of macro '__READ_ONCE'
>> #define READ_ONCE(x) __READ_ONCE(x, 1)
>> ^~~~~~~~~~~
>> arch/arm64/include/asm/pgtable.h:628:52: note: in expansion of macro 'READ_ONCE'
>> #define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir)))
>> + pud_index(addr) * sizeof(pud_t))
>> ^~~~~~~~~
>> arch/arm64/include/asm/pgtable.h:629:47: note: in expansion of macro
>> 'pud_offset_phys'
>> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
>> ^~~~~~~~~~~~~~~
>> arch/arm64/mm/mmu.c:827:10: note: in expansion of macro 'pud_offset'
>> pudp = pud_offset(pgdp, addr);
>> ^~~~~~~~~~
>
> Looks like we need an implementation of unmap_hotplug_p4d_range() to
> walk the dummy p4d level. Unfortunately, we don't have the folded p4d
> patches in the arm64 tree so we'll either need a common branch or the
> hotplug patches will need to be dropped for the moment.

If we decide to get a common branch, will try to get this sorted with
an unmap_hotplug_p4d_range() implementation as you have suggested.

>
> Will
>

2020-03-02 13:55:49

by Mike Rapoport

[permalink] [raw]
Subject: Re: Linux-next-20200302: arm64 build failed

On Mon, Mar 02, 2020 at 10:47:27AM +0000, Will Deacon wrote:
> [+Anshuman and Catalin]
>
> On Mon, Mar 02, 2020 at 01:58:26PM +0530, Naresh Kamboju wrote:
> > Linux-Next 20200302 arm64 build failed due to below errors,
> > Suspecting patch causing this build break.
> >
> > 87d900aef3e2 arm/arm64: add support for folded p4d page tables
> >
> > Error log,
> > -------------
> > arch/arm64/mm/mmu.c: In function 'unmap_hotplug_pud_range':
> > include/linux/compiler.h:284:1: error: incompatible type for argument
> > 1 of 'p4d_page_paddr'
> > ({ \
> > ^
> > arch/arm64/include/asm/memory.h:270:45: note: in definition of macro
> > '__phys_to_virt'
> > #define __phys_to_virt(x) ((unsigned long)((x) - physvirt_offset))
> > ^
> > arch/arm64/include/asm/pgtable.h:629:42: note: in expansion of macro '__va'
> > #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
> > ^~~~
> > include/linux/compiler.h:293:22: note: in expansion of macro '__READ_ONCE'
> > #define READ_ONCE(x) __READ_ONCE(x, 1)
> > ^~~~~~~~~~~
> > arch/arm64/include/asm/pgtable.h:628:52: note: in expansion of macro 'READ_ONCE'
> > #define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir)))
> > + pud_index(addr) * sizeof(pud_t))
> > ^~~~~~~~~
> > arch/arm64/include/asm/pgtable.h:629:47: note: in expansion of macro
> > 'pud_offset_phys'
> > #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
> > ^~~~~~~~~~~~~~~
> > arch/arm64/mm/mmu.c:827:10: note: in expansion of macro 'pud_offset'
> > pudp = pud_offset(pgdp, addr);
> > ^~~~~~~~~~
>
> Looks like we need an implementation of unmap_hotplug_p4d_range() to
> walk the dummy p4d level. Unfortunately, we don't have the folded p4d
> patches in the arm64 tree so we'll either need a common branch or the
> hotplug patches will need to be dropped for the moment.

unmap_hotplug_p4d_range() is easy :)

From c7a5d08ff51ca2057b6b0289c4423bdfd7643518 Mon Sep 17 00:00:00 2001
From: Mike Rapoport <[email protected]>
Date: Mon, 2 Mar 2020 15:53:17 +0200
Subject: [PATCH] arm64/mm: implement unmap_hotplug_p4d_range

Signed-off-by: Mike Rapoport <[email protected]>
---
arch/arm64/mm/mmu.c | 20 +++++++++++++++++++-
1 file changed, 19 insertions(+), 1 deletion(-)

diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
index 05ec8e5f1436..c76b11577558 100644
--- a/arch/arm64/mm/mmu.c
+++ b/arch/arm64/mm/mmu.c
@@ -840,6 +840,24 @@ static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
} while (addr = next, addr < end);
}

+static void unmap_hotplug_p4d_range(pgd_t *pgd, unsigned long addr,
+ unsigned long end, bool free_mapped)
+{
+ unsigned long next;
+ pgd_t *p4dp, p4d;
+
+ do {
+ next = p4d_addr_end(addr, end);
+ p4dp = p4d_offset(pgd, addr);
+ p4d = READ_ONCE(*p4dp);
+ if (p4d_none(p4d))
+ continue;
+
+ WARN_ON(!p4d_present(p4d));
+ unmap_hotplug_pud_range(p4dp, addr, next, free_mapped);
+ } while (addr = next, addr < end);
+}
+
static void unmap_hotplug_range(unsigned long addr, unsigned long end,
bool free_mapped)
{
@@ -854,7 +872,7 @@ static void unmap_hotplug_range(unsigned long addr, unsigned long end,
continue;

WARN_ON(!pgd_present(pgd));
- unmap_hotplug_pud_range(pgdp, addr, next, free_mapped);
+ unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped);
} while (addr = next, addr < end);
}

--
2.21.1


> Will

--
Sincerely yours,
Mike.

2020-03-02 17:47:17

by Catalin Marinas

[permalink] [raw]
Subject: Re: Linux-next-20200302: arm64 build failed

On Mon, Mar 02, 2020 at 03:54:43PM +0200, Mike Rapoport wrote:
> On Mon, Mar 02, 2020 at 10:47:27AM +0000, Will Deacon wrote:
> > [+Anshuman and Catalin]
> >
> > On Mon, Mar 02, 2020 at 01:58:26PM +0530, Naresh Kamboju wrote:
> > > Linux-Next 20200302 arm64 build failed due to below errors,
> > > Suspecting patch causing this build break.
> > >
> > > 87d900aef3e2 arm/arm64: add support for folded p4d page tables
> > >
> > > Error log,
> > > -------------
> > > arch/arm64/mm/mmu.c: In function 'unmap_hotplug_pud_range':
> > > include/linux/compiler.h:284:1: error: incompatible type for argument
> > > 1 of 'p4d_page_paddr'
> > > ({ \
> > > ^
> > > arch/arm64/include/asm/memory.h:270:45: note: in definition of macro
> > > '__phys_to_virt'
> > > #define __phys_to_virt(x) ((unsigned long)((x) - physvirt_offset))
> > > ^
> > > arch/arm64/include/asm/pgtable.h:629:42: note: in expansion of macro '__va'
> > > #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
> > > ^~~~
> > > include/linux/compiler.h:293:22: note: in expansion of macro '__READ_ONCE'
> > > #define READ_ONCE(x) __READ_ONCE(x, 1)
> > > ^~~~~~~~~~~
> > > arch/arm64/include/asm/pgtable.h:628:52: note: in expansion of macro 'READ_ONCE'
> > > #define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir)))
> > > + pud_index(addr) * sizeof(pud_t))
> > > ^~~~~~~~~
> > > arch/arm64/include/asm/pgtable.h:629:47: note: in expansion of macro
> > > 'pud_offset_phys'
> > > #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
> > > ^~~~~~~~~~~~~~~
> > > arch/arm64/mm/mmu.c:827:10: note: in expansion of macro 'pud_offset'
> > > pudp = pud_offset(pgdp, addr);
> > > ^~~~~~~~~~
> >
> > Looks like we need an implementation of unmap_hotplug_p4d_range() to
> > walk the dummy p4d level. Unfortunately, we don't have the folded p4d
> > patches in the arm64 tree so we'll either need a common branch or the
> > hotplug patches will need to be dropped for the moment.
>
> unmap_hotplug_p4d_range() is easy :)
>
> From c7a5d08ff51ca2057b6b0289c4423bdfd7643518 Mon Sep 17 00:00:00 2001
> From: Mike Rapoport <[email protected]>
> Date: Mon, 2 Mar 2020 15:53:17 +0200
> Subject: [PATCH] arm64/mm: implement unmap_hotplug_p4d_range
>
> Signed-off-by: Mike Rapoport <[email protected]>
> ---
> arch/arm64/mm/mmu.c | 20 +++++++++++++++++++-
> 1 file changed, 19 insertions(+), 1 deletion(-)
>
> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> index 05ec8e5f1436..c76b11577558 100644
> --- a/arch/arm64/mm/mmu.c
> +++ b/arch/arm64/mm/mmu.c
> @@ -840,6 +840,24 @@ static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
> } while (addr = next, addr < end);
> }
>
> +static void unmap_hotplug_p4d_range(pgd_t *pgd, unsigned long addr,
> + unsigned long end, bool free_mapped)
> +{
> + unsigned long next;
> + pgd_t *p4dp, p4d;
> +
> + do {
> + next = p4d_addr_end(addr, end);
> + p4dp = p4d_offset(pgd, addr);
> + p4d = READ_ONCE(*p4dp);
> + if (p4d_none(p4d))
> + continue;
> +
> + WARN_ON(!p4d_present(p4d));
> + unmap_hotplug_pud_range(p4dp, addr, next, free_mapped);
> + } while (addr = next, addr < end);
> +}
> +
> static void unmap_hotplug_range(unsigned long addr, unsigned long end,
> bool free_mapped)
> {
> @@ -854,7 +872,7 @@ static void unmap_hotplug_range(unsigned long addr, unsigned long end,
> continue;
>
> WARN_ON(!pgd_present(pgd));
> - unmap_hotplug_pud_range(pgdp, addr, next, free_mapped);
> + unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped);
> } while (addr = next, addr < end);
> }

Thanks Mike. With the additional diff below, I can get it to build with
and without the p4d clean-up patches in -next. If Anshuman confirms that
they work, I can add them on top of the arm64 for-next/memory-hotremove
branch

----------8<------------------------
diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
index 56bd9227937f..2faf11860097 100644
--- a/arch/arm64/mm/mmu.c
+++ b/arch/arm64/mm/mmu.c
@@ -816,7 +816,7 @@ static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
} while (addr = next, addr < end);
}

-static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
+static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
unsigned long end, bool free_mapped)
{
unsigned long next;
@@ -824,7 +824,7 @@ static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,

do {
next = pud_addr_end(addr, end);
- pudp = pud_offset(pgdp, addr);
+ pudp = pud_offset(p4dp, addr);
pud = READ_ONCE(*pudp);
if (pud_none(pud))
continue;
@@ -848,15 +848,15 @@ static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
} while (addr = next, addr < end);
}

-static void unmap_hotplug_p4d_range(pgd_t *pgd, unsigned long addr,
- unsigned long end, bool free_mapped)
+static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
+ unsigned long end, bool free_mapped)
{
unsigned long next;
- pgd_t *p4dp, p4d;
+ p4d_t *p4dp, p4d;

do {
next = p4d_addr_end(addr, end);
- p4dp = p4d_offset(pgd, addr);
+ p4dp = p4d_offset(pgdp, addr);
p4d = READ_ONCE(*p4dp);
if (p4d_none(p4d))
continue;
@@ -961,7 +961,7 @@ static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
free_hotplug_pgtable_page(virt_to_page(pmdp));
}

-static void free_empty_pud_table(pgd_t *pgdp, unsigned long addr,
+static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
unsigned long end, unsigned long floor,
unsigned long ceiling)
{
@@ -970,7 +970,7 @@ static void free_empty_pud_table(pgd_t *pgdp, unsigned long addr,

do {
next = pud_addr_end(addr, end);
- pudp = pud_offset(pgdp, addr);
+ pudp = pud_offset(p4dp, addr);
pud = READ_ONCE(*pudp);
if (pud_none(pud))
continue;
@@ -990,17 +990,36 @@ static void free_empty_pud_table(pgd_t *pgdp, unsigned long addr,
* entries are empty. Overlap with other regions have been
* handled by the floor/ceiling check.
*/
- pudp = pud_offset(pgdp, 0UL);
+ pudp = pud_offset(p4dp, 0UL);
for (i = 0; i < PTRS_PER_PUD; i++) {
if (!pud_none(READ_ONCE(pudp[i])))
return;
}

- pgd_clear(pgdp);
+ p4d_clear(p4dp);
__flush_tlb_kernel_pgtable(start);
free_hotplug_pgtable_page(virt_to_page(pudp));
}

+static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
+ unsigned long end, unsigned long floor,
+ unsigned long ceiling)
+{
+ p4d_t *p4dp, p4d;
+ unsigned long next;
+
+ do {
+ next = p4d_addr_end(addr, end);
+ p4dp = p4d_offset(pgdp, addr);
+ p4d = READ_ONCE(*p4dp);
+ if (p4d_none(p4d))
+ continue;
+
+ WARN_ON(!p4d_present(p4d));
+ free_empty_pud_table(p4dp, addr, next, floor, ceiling);
+ } while (addr = next, addr < end);
+}
+
static void free_empty_tables(unsigned long addr, unsigned long end,
unsigned long floor, unsigned long ceiling)
{
@@ -1015,7 +1034,7 @@ static void free_empty_tables(unsigned long addr, unsigned long end,
continue;

WARN_ON(!pgd_present(pgd));
- free_empty_pud_table(pgdp, addr, next, floor, ceiling);
+ free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
} while (addr = next, addr < end);
}
#endif

2020-03-02 22:24:35

by Andrew Morton

[permalink] [raw]
Subject: Re: Linux-next-20200302: arm64 build failed

On Mon, 2 Mar 2020 17:45:53 +0000 Catalin Marinas <[email protected]> wrote:

> > static void unmap_hotplug_range(unsigned long addr, unsigned long end,
> > bool free_mapped)
> > {
> > @@ -854,7 +872,7 @@ static void unmap_hotplug_range(unsigned long addr, unsigned long end,
> > continue;
> >
> > WARN_ON(!pgd_present(pgd));
> > - unmap_hotplug_pud_range(pgdp, addr, next, free_mapped);
> > + unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped);
> > } while (addr = next, addr < end);
> > }
>
> Thanks Mike. With the additional diff below, I can get it to build with
> and without the p4d clean-up patches in -next. If Anshuman confirms that
> they work, I can add them on top of the arm64 for-next/memory-hotremove
> branch

Can't I simply fold these into the offending -mm
arm-arm64-add-support-for-folded-p4d-page-tables.patch?

2020-03-03 04:06:22

by Anshuman Khandual

[permalink] [raw]
Subject: Re: Linux-next-20200302: arm64 build failed


On 03/02/2020 11:15 PM, Catalin Marinas wrote:
> On Mon, Mar 02, 2020 at 03:54:43PM +0200, Mike Rapoport wrote:
>> On Mon, Mar 02, 2020 at 10:47:27AM +0000, Will Deacon wrote:
>>> [+Anshuman and Catalin]
>>>
>>> On Mon, Mar 02, 2020 at 01:58:26PM +0530, Naresh Kamboju wrote:
>>>> Linux-Next 20200302 arm64 build failed due to below errors,
>>>> Suspecting patch causing this build break.
>>>>
>>>> 87d900aef3e2 arm/arm64: add support for folded p4d page tables
>>>>
>>>> Error log,
>>>> -------------
>>>> arch/arm64/mm/mmu.c: In function 'unmap_hotplug_pud_range':
>>>> include/linux/compiler.h:284:1: error: incompatible type for argument
>>>> 1 of 'p4d_page_paddr'
>>>> ({ \
>>>> ^
>>>> arch/arm64/include/asm/memory.h:270:45: note: in definition of macro
>>>> '__phys_to_virt'
>>>> #define __phys_to_virt(x) ((unsigned long)((x) - physvirt_offset))
>>>> ^
>>>> arch/arm64/include/asm/pgtable.h:629:42: note: in expansion of macro '__va'
>>>> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
>>>> ^~~~
>>>> include/linux/compiler.h:293:22: note: in expansion of macro '__READ_ONCE'
>>>> #define READ_ONCE(x) __READ_ONCE(x, 1)
>>>> ^~~~~~~~~~~
>>>> arch/arm64/include/asm/pgtable.h:628:52: note: in expansion of macro 'READ_ONCE'
>>>> #define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir)))
>>>> + pud_index(addr) * sizeof(pud_t))
>>>> ^~~~~~~~~
>>>> arch/arm64/include/asm/pgtable.h:629:47: note: in expansion of macro
>>>> 'pud_offset_phys'
>>>> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
>>>> ^~~~~~~~~~~~~~~
>>>> arch/arm64/mm/mmu.c:827:10: note: in expansion of macro 'pud_offset'
>>>> pudp = pud_offset(pgdp, addr);
>>>> ^~~~~~~~~~
>>>
>>> Looks like we need an implementation of unmap_hotplug_p4d_range() to
>>> walk the dummy p4d level. Unfortunately, we don't have the folded p4d
>>> patches in the arm64 tree so we'll either need a common branch or the
>>> hotplug patches will need to be dropped for the moment.
>>
>> unmap_hotplug_p4d_range() is easy :)
>>
>> From c7a5d08ff51ca2057b6b0289c4423bdfd7643518 Mon Sep 17 00:00:00 2001
>> From: Mike Rapoport <[email protected]>
>> Date: Mon, 2 Mar 2020 15:53:17 +0200
>> Subject: [PATCH] arm64/mm: implement unmap_hotplug_p4d_range
>>
>> Signed-off-by: Mike Rapoport <[email protected]>
>> ---
>> arch/arm64/mm/mmu.c | 20 +++++++++++++++++++-
>> 1 file changed, 19 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
>> index 05ec8e5f1436..c76b11577558 100644
>> --- a/arch/arm64/mm/mmu.c
>> +++ b/arch/arm64/mm/mmu.c
>> @@ -840,6 +840,24 @@ static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
>> } while (addr = next, addr < end);
>> }
>>
>> +static void unmap_hotplug_p4d_range(pgd_t *pgd, unsigned long addr,
>> + unsigned long end, bool free_mapped)
>> +{
>> + unsigned long next;
>> + pgd_t *p4dp, p4d;
>> +
>> + do {
>> + next = p4d_addr_end(addr, end);
>> + p4dp = p4d_offset(pgd, addr);
>> + p4d = READ_ONCE(*p4dp);
>> + if (p4d_none(p4d))
>> + continue;
>> +
>> + WARN_ON(!p4d_present(p4d));
>> + unmap_hotplug_pud_range(p4dp, addr, next, free_mapped);
>> + } while (addr = next, addr < end);
>> +}
>> +
>> static void unmap_hotplug_range(unsigned long addr, unsigned long end,
>> bool free_mapped)
>> {
>> @@ -854,7 +872,7 @@ static void unmap_hotplug_range(unsigned long addr, unsigned long end,
>> continue;
>>
>> WARN_ON(!pgd_present(pgd));
>> - unmap_hotplug_pud_range(pgdp, addr, next, free_mapped);
>> + unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped);
>> } while (addr = next, addr < end);
>> }
>
> Thanks Mike. With the additional diff below, I can get it to build with
> and without the p4d clean-up patches in -next. If Anshuman confirms that
> they work, I can add them on top of the arm64 for-next/memory-hotremove
> branch.

These two patches applied on next-20200302 works fine for hot-remove. As the
second patch also fixes the first one, IMHO both should be folded into a
single one instead. Just wondering if this combined patch which enables P4D
page table should be posted on the list or do I need to respin original hot
remove patches again.

- Anshuman

>
> ----------8<------------------------
> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> index 56bd9227937f..2faf11860097 100644
> --- a/arch/arm64/mm/mmu.c
> +++ b/arch/arm64/mm/mmu.c
> @@ -816,7 +816,7 @@ static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
> } while (addr = next, addr < end);
> }
>
> -static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
> +static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
> unsigned long end, bool free_mapped)
> {
> unsigned long next;
> @@ -824,7 +824,7 @@ static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
>
> do {
> next = pud_addr_end(addr, end);
> - pudp = pud_offset(pgdp, addr);
> + pudp = pud_offset(p4dp, addr);
> pud = READ_ONCE(*pudp);
> if (pud_none(pud))
> continue;
> @@ -848,15 +848,15 @@ static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
> } while (addr = next, addr < end);
> }
>
> -static void unmap_hotplug_p4d_range(pgd_t *pgd, unsigned long addr,
> - unsigned long end, bool free_mapped)
> +static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
> + unsigned long end, bool free_mapped)
> {
> unsigned long next;
> - pgd_t *p4dp, p4d;
> + p4d_t *p4dp, p4d;
>
> do {
> next = p4d_addr_end(addr, end);
> - p4dp = p4d_offset(pgd, addr);
> + p4dp = p4d_offset(pgdp, addr);
> p4d = READ_ONCE(*p4dp);
> if (p4d_none(p4d))
> continue;
> @@ -961,7 +961,7 @@ static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
> free_hotplug_pgtable_page(virt_to_page(pmdp));
> }
>
> -static void free_empty_pud_table(pgd_t *pgdp, unsigned long addr,
> +static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
> unsigned long end, unsigned long floor,
> unsigned long ceiling)
> {
> @@ -970,7 +970,7 @@ static void free_empty_pud_table(pgd_t *pgdp, unsigned long addr,
>
> do {
> next = pud_addr_end(addr, end);
> - pudp = pud_offset(pgdp, addr);
> + pudp = pud_offset(p4dp, addr);
> pud = READ_ONCE(*pudp);
> if (pud_none(pud))
> continue;
> @@ -990,17 +990,36 @@ static void free_empty_pud_table(pgd_t *pgdp, unsigned long addr,
> * entries are empty. Overlap with other regions have been
> * handled by the floor/ceiling check.
> */
> - pudp = pud_offset(pgdp, 0UL);
> + pudp = pud_offset(p4dp, 0UL);
> for (i = 0; i < PTRS_PER_PUD; i++) {
> if (!pud_none(READ_ONCE(pudp[i])))
> return;
> }
>
> - pgd_clear(pgdp);
> + p4d_clear(p4dp);
> __flush_tlb_kernel_pgtable(start);
> free_hotplug_pgtable_page(virt_to_page(pudp));
> }
>
> +static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
> + unsigned long end, unsigned long floor,
> + unsigned long ceiling)
> +{
> + p4d_t *p4dp, p4d;
> + unsigned long next;
> +
> + do {
> + next = p4d_addr_end(addr, end);
> + p4dp = p4d_offset(pgdp, addr);
> + p4d = READ_ONCE(*p4dp);
> + if (p4d_none(p4d))
> + continue;
> +
> + WARN_ON(!p4d_present(p4d));
> + free_empty_pud_table(p4dp, addr, next, floor, ceiling);
> + } while (addr = next, addr < end);
> +}
> +
> static void free_empty_tables(unsigned long addr, unsigned long end,
> unsigned long floor, unsigned long ceiling)
> {
> @@ -1015,7 +1034,7 @@ static void free_empty_tables(unsigned long addr, unsigned long end,
> continue;
>
> WARN_ON(!pgd_present(pgd));
> - free_empty_pud_table(pgdp, addr, next, floor, ceiling);
> + free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
> } while (addr = next, addr < end);
> }
> #endif
>

2020-03-03 16:29:53

by Catalin Marinas

[permalink] [raw]
Subject: Re: Linux-next-20200302: arm64 build failed

On Tue, Mar 03, 2020 at 09:34:45AM +0530, Anshuman Khandual wrote:
> On 03/02/2020 11:15 PM, Catalin Marinas wrote:
> > On Mon, Mar 02, 2020 at 03:54:43PM +0200, Mike Rapoport wrote:
> >> On Mon, Mar 02, 2020 at 10:47:27AM +0000, Will Deacon wrote:
> >>> [+Anshuman and Catalin]
> >>>
> >>> On Mon, Mar 02, 2020 at 01:58:26PM +0530, Naresh Kamboju wrote:
> >>>> Linux-Next 20200302 arm64 build failed due to below errors,
> >>>> Suspecting patch causing this build break.
> >>>>
> >>>> 87d900aef3e2 arm/arm64: add support for folded p4d page tables
> >>>>
> >>>> Error log,
> >>>> -------------
> >>>> arch/arm64/mm/mmu.c: In function 'unmap_hotplug_pud_range':
> >>>> include/linux/compiler.h:284:1: error: incompatible type for argument
> >>>> 1 of 'p4d_page_paddr'
> >>>> ({ \
> >>>> ^
> >>>> arch/arm64/include/asm/memory.h:270:45: note: in definition of macro
> >>>> '__phys_to_virt'
> >>>> #define __phys_to_virt(x) ((unsigned long)((x) - physvirt_offset))
> >>>> ^
> >>>> arch/arm64/include/asm/pgtable.h:629:42: note: in expansion of macro '__va'
> >>>> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
> >>>> ^~~~
> >>>> include/linux/compiler.h:293:22: note: in expansion of macro '__READ_ONCE'
> >>>> #define READ_ONCE(x) __READ_ONCE(x, 1)
> >>>> ^~~~~~~~~~~
> >>>> arch/arm64/include/asm/pgtable.h:628:52: note: in expansion of macro 'READ_ONCE'
> >>>> #define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir)))
> >>>> + pud_index(addr) * sizeof(pud_t))
> >>>> ^~~~~~~~~
> >>>> arch/arm64/include/asm/pgtable.h:629:47: note: in expansion of macro
> >>>> 'pud_offset_phys'
> >>>> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
> >>>> ^~~~~~~~~~~~~~~
> >>>> arch/arm64/mm/mmu.c:827:10: note: in expansion of macro 'pud_offset'
> >>>> pudp = pud_offset(pgdp, addr);
> >>>> ^~~~~~~~~~
> >>>
> >>> Looks like we need an implementation of unmap_hotplug_p4d_range() to
> >>> walk the dummy p4d level. Unfortunately, we don't have the folded p4d
> >>> patches in the arm64 tree so we'll either need a common branch or the
> >>> hotplug patches will need to be dropped for the moment.
> >>
> >> unmap_hotplug_p4d_range() is easy :)
> >>
> >> From c7a5d08ff51ca2057b6b0289c4423bdfd7643518 Mon Sep 17 00:00:00 2001
> >> From: Mike Rapoport <[email protected]>
> >> Date: Mon, 2 Mar 2020 15:53:17 +0200
> >> Subject: [PATCH] arm64/mm: implement unmap_hotplug_p4d_range
> >>
> >> Signed-off-by: Mike Rapoport <[email protected]>
> >> ---
> >> arch/arm64/mm/mmu.c | 20 +++++++++++++++++++-
> >> 1 file changed, 19 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> >> index 05ec8e5f1436..c76b11577558 100644
> >> --- a/arch/arm64/mm/mmu.c
> >> +++ b/arch/arm64/mm/mmu.c
> >> @@ -840,6 +840,24 @@ static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
> >> } while (addr = next, addr < end);
> >> }
> >>
> >> +static void unmap_hotplug_p4d_range(pgd_t *pgd, unsigned long addr,
> >> + unsigned long end, bool free_mapped)
> >> +{
> >> + unsigned long next;
> >> + pgd_t *p4dp, p4d;
> >> +
> >> + do {
> >> + next = p4d_addr_end(addr, end);
> >> + p4dp = p4d_offset(pgd, addr);
> >> + p4d = READ_ONCE(*p4dp);
> >> + if (p4d_none(p4d))
> >> + continue;
> >> +
> >> + WARN_ON(!p4d_present(p4d));
> >> + unmap_hotplug_pud_range(p4dp, addr, next, free_mapped);
> >> + } while (addr = next, addr < end);
> >> +}
> >> +
> >> static void unmap_hotplug_range(unsigned long addr, unsigned long end,
> >> bool free_mapped)
> >> {
> >> @@ -854,7 +872,7 @@ static void unmap_hotplug_range(unsigned long addr, unsigned long end,
> >> continue;
> >>
> >> WARN_ON(!pgd_present(pgd));
> >> - unmap_hotplug_pud_range(pgdp, addr, next, free_mapped);
> >> + unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped);
> >> } while (addr = next, addr < end);
> >> }
> >
> > Thanks Mike. With the additional diff below, I can get it to build with
> > and without the p4d clean-up patches in -next. If Anshuman confirms that
> > they work, I can add them on top of the arm64 for-next/memory-hotremove
> > branch.
>
> These two patches applied on next-20200302 works fine for hot-remove.

Do they also work on top of the vanilla kernel + your hotremove patches
(i.e. not on top of -next)?

> As the second patch also fixes the first one, IMHO both should be
> folded into a single one instead. Just wondering if this combined
> patch which enables P4D page table should be posted on the list or do
> I need to respin original hot remove patches again.

If your unmap patches plus the fixes from Mike and me work fine on top
of 5.6-rc3 (as well as when combined with linux-next), I'd prefer you
respin your patches to include the p4d support from start. Otherwise, we
create a single patch that Andrew can merge on top of the -mm tree.

--
Catalin

2020-03-04 02:28:13

by Anshuman Khandual

[permalink] [raw]
Subject: Re: Linux-next-20200302: arm64 build failed



On 03/03/2020 09:58 PM, Catalin Marinas wrote:
> On Tue, Mar 03, 2020 at 09:34:45AM +0530, Anshuman Khandual wrote:
>> On 03/02/2020 11:15 PM, Catalin Marinas wrote:
>>> On Mon, Mar 02, 2020 at 03:54:43PM +0200, Mike Rapoport wrote:
>>>> On Mon, Mar 02, 2020 at 10:47:27AM +0000, Will Deacon wrote:
>>>>> [+Anshuman and Catalin]
>>>>>
>>>>> On Mon, Mar 02, 2020 at 01:58:26PM +0530, Naresh Kamboju wrote:
>>>>>> Linux-Next 20200302 arm64 build failed due to below errors,
>>>>>> Suspecting patch causing this build break.
>>>>>>
>>>>>> 87d900aef3e2 arm/arm64: add support for folded p4d page tables
>>>>>>
>>>>>> Error log,
>>>>>> -------------
>>>>>> arch/arm64/mm/mmu.c: In function 'unmap_hotplug_pud_range':
>>>>>> include/linux/compiler.h:284:1: error: incompatible type for argument
>>>>>> 1 of 'p4d_page_paddr'
>>>>>> ({ \
>>>>>> ^
>>>>>> arch/arm64/include/asm/memory.h:270:45: note: in definition of macro
>>>>>> '__phys_to_virt'
>>>>>> #define __phys_to_virt(x) ((unsigned long)((x) - physvirt_offset))
>>>>>> ^
>>>>>> arch/arm64/include/asm/pgtable.h:629:42: note: in expansion of macro '__va'
>>>>>> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
>>>>>> ^~~~
>>>>>> include/linux/compiler.h:293:22: note: in expansion of macro '__READ_ONCE'
>>>>>> #define READ_ONCE(x) __READ_ONCE(x, 1)
>>>>>> ^~~~~~~~~~~
>>>>>> arch/arm64/include/asm/pgtable.h:628:52: note: in expansion of macro 'READ_ONCE'
>>>>>> #define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir)))
>>>>>> + pud_index(addr) * sizeof(pud_t))
>>>>>> ^~~~~~~~~
>>>>>> arch/arm64/include/asm/pgtable.h:629:47: note: in expansion of macro
>>>>>> 'pud_offset_phys'
>>>>>> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
>>>>>> ^~~~~~~~~~~~~~~
>>>>>> arch/arm64/mm/mmu.c:827:10: note: in expansion of macro 'pud_offset'
>>>>>> pudp = pud_offset(pgdp, addr);
>>>>>> ^~~~~~~~~~
>>>>>
>>>>> Looks like we need an implementation of unmap_hotplug_p4d_range() to
>>>>> walk the dummy p4d level. Unfortunately, we don't have the folded p4d
>>>>> patches in the arm64 tree so we'll either need a common branch or the
>>>>> hotplug patches will need to be dropped for the moment.
>>>>
>>>> unmap_hotplug_p4d_range() is easy :)
>>>>
>>>> From c7a5d08ff51ca2057b6b0289c4423bdfd7643518 Mon Sep 17 00:00:00 2001
>>>> From: Mike Rapoport <[email protected]>
>>>> Date: Mon, 2 Mar 2020 15:53:17 +0200
>>>> Subject: [PATCH] arm64/mm: implement unmap_hotplug_p4d_range
>>>>
>>>> Signed-off-by: Mike Rapoport <[email protected]>
>>>> ---
>>>> arch/arm64/mm/mmu.c | 20 +++++++++++++++++++-
>>>> 1 file changed, 19 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
>>>> index 05ec8e5f1436..c76b11577558 100644
>>>> --- a/arch/arm64/mm/mmu.c
>>>> +++ b/arch/arm64/mm/mmu.c
>>>> @@ -840,6 +840,24 @@ static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
>>>> } while (addr = next, addr < end);
>>>> }
>>>>
>>>> +static void unmap_hotplug_p4d_range(pgd_t *pgd, unsigned long addr,
>>>> + unsigned long end, bool free_mapped)
>>>> +{
>>>> + unsigned long next;
>>>> + pgd_t *p4dp, p4d;
>>>> +
>>>> + do {
>>>> + next = p4d_addr_end(addr, end);
>>>> + p4dp = p4d_offset(pgd, addr);
>>>> + p4d = READ_ONCE(*p4dp);
>>>> + if (p4d_none(p4d))
>>>> + continue;
>>>> +
>>>> + WARN_ON(!p4d_present(p4d));
>>>> + unmap_hotplug_pud_range(p4dp, addr, next, free_mapped);
>>>> + } while (addr = next, addr < end);
>>>> +}
>>>> +
>>>> static void unmap_hotplug_range(unsigned long addr, unsigned long end,
>>>> bool free_mapped)
>>>> {
>>>> @@ -854,7 +872,7 @@ static void unmap_hotplug_range(unsigned long addr, unsigned long end,
>>>> continue;
>>>>
>>>> WARN_ON(!pgd_present(pgd));
>>>> - unmap_hotplug_pud_range(pgdp, addr, next, free_mapped);
>>>> + unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped);
>>>> } while (addr = next, addr < end);
>>>> }
>>>
>>> Thanks Mike. With the additional diff below, I can get it to build with
>>> and without the p4d clean-up patches in -next. If Anshuman confirms that
>>> they work, I can add them on top of the arm64 for-next/memory-hotremove
>>> branch.
>>
>> These two patches applied on next-20200302 works fine for hot-remove.
>
> Do they also work on top of the vanilla kernel + your hotremove patches
> (i.e. not on top of -next)?

Yes, they do work on current vanilla kernel (8b614cb8f1dcac8ca77cf4dd85f46)
and v13 hotremove series.

>
>> As the second patch also fixes the first one, IMHO both should be
>> folded into a single one instead. Just wondering if this combined
>> patch which enables P4D page table should be posted on the list or do
>> I need to respin original hot remove patches again.
>
> If your unmap patches plus the fixes from Mike and me work fine on top
> of 5.6-rc3 (as well as when combined with linux-next), I'd prefer you
Yes, they do work on both.

> respin your patches to include the p4d support from start. Otherwise, we

Okay. I will be sending V14 on v5.6-rc3 (OR v.5.6-rc4 is preferred ?) with
p4d support. I will add yours and Mike's Signed-off-by as well.

> create a single patch that Andrew can merge on top of the -mm tree.
>

2020-03-04 04:33:03

by Anshuman Khandual

[permalink] [raw]
Subject: Re: Linux-next-20200302: arm64 build failed



On 03/04/2020 07:56 AM, Anshuman Khandual wrote:
>
>
> On 03/03/2020 09:58 PM, Catalin Marinas wrote:
>> On Tue, Mar 03, 2020 at 09:34:45AM +0530, Anshuman Khandual wrote:
>>> On 03/02/2020 11:15 PM, Catalin Marinas wrote:
>>>> On Mon, Mar 02, 2020 at 03:54:43PM +0200, Mike Rapoport wrote:
>>>>> On Mon, Mar 02, 2020 at 10:47:27AM +0000, Will Deacon wrote:
>>>>>> [+Anshuman and Catalin]
>>>>>>
>>>>>> On Mon, Mar 02, 2020 at 01:58:26PM +0530, Naresh Kamboju wrote:
>>>>>>> Linux-Next 20200302 arm64 build failed due to below errors,
>>>>>>> Suspecting patch causing this build break.
>>>>>>>
>>>>>>> 87d900aef3e2 arm/arm64: add support for folded p4d page tables
>>>>>>>
>>>>>>> Error log,
>>>>>>> -------------
>>>>>>> arch/arm64/mm/mmu.c: In function 'unmap_hotplug_pud_range':
>>>>>>> include/linux/compiler.h:284:1: error: incompatible type for argument
>>>>>>> 1 of 'p4d_page_paddr'
>>>>>>> ({ \
>>>>>>> ^
>>>>>>> arch/arm64/include/asm/memory.h:270:45: note: in definition of macro
>>>>>>> '__phys_to_virt'
>>>>>>> #define __phys_to_virt(x) ((unsigned long)((x) - physvirt_offset))
>>>>>>> ^
>>>>>>> arch/arm64/include/asm/pgtable.h:629:42: note: in expansion of macro '__va'
>>>>>>> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
>>>>>>> ^~~~
>>>>>>> include/linux/compiler.h:293:22: note: in expansion of macro '__READ_ONCE'
>>>>>>> #define READ_ONCE(x) __READ_ONCE(x, 1)
>>>>>>> ^~~~~~~~~~~
>>>>>>> arch/arm64/include/asm/pgtable.h:628:52: note: in expansion of macro 'READ_ONCE'
>>>>>>> #define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir)))
>>>>>>> + pud_index(addr) * sizeof(pud_t))
>>>>>>> ^~~~~~~~~
>>>>>>> arch/arm64/include/asm/pgtable.h:629:47: note: in expansion of macro
>>>>>>> 'pud_offset_phys'
>>>>>>> #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr))))
>>>>>>> ^~~~~~~~~~~~~~~
>>>>>>> arch/arm64/mm/mmu.c:827:10: note: in expansion of macro 'pud_offset'
>>>>>>> pudp = pud_offset(pgdp, addr);
>>>>>>> ^~~~~~~~~~
>>>>>>
>>>>>> Looks like we need an implementation of unmap_hotplug_p4d_range() to
>>>>>> walk the dummy p4d level. Unfortunately, we don't have the folded p4d
>>>>>> patches in the arm64 tree so we'll either need a common branch or the
>>>>>> hotplug patches will need to be dropped for the moment.
>>>>>
>>>>> unmap_hotplug_p4d_range() is easy :)
>>>>>
>>>>> From c7a5d08ff51ca2057b6b0289c4423bdfd7643518 Mon Sep 17 00:00:00 2001
>>>>> From: Mike Rapoport <[email protected]>
>>>>> Date: Mon, 2 Mar 2020 15:53:17 +0200
>>>>> Subject: [PATCH] arm64/mm: implement unmap_hotplug_p4d_range
>>>>>
>>>>> Signed-off-by: Mike Rapoport <[email protected]>
>>>>> ---
>>>>> arch/arm64/mm/mmu.c | 20 +++++++++++++++++++-
>>>>> 1 file changed, 19 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
>>>>> index 05ec8e5f1436..c76b11577558 100644
>>>>> --- a/arch/arm64/mm/mmu.c
>>>>> +++ b/arch/arm64/mm/mmu.c
>>>>> @@ -840,6 +840,24 @@ static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr,
>>>>> } while (addr = next, addr < end);
>>>>> }
>>>>>
>>>>> +static void unmap_hotplug_p4d_range(pgd_t *pgd, unsigned long addr,
>>>>> + unsigned long end, bool free_mapped)
>>>>> +{
>>>>> + unsigned long next;
>>>>> + pgd_t *p4dp, p4d;
>>>>> +
>>>>> + do {
>>>>> + next = p4d_addr_end(addr, end);
>>>>> + p4dp = p4d_offset(pgd, addr);
>>>>> + p4d = READ_ONCE(*p4dp);
>>>>> + if (p4d_none(p4d))
>>>>> + continue;
>>>>> +
>>>>> + WARN_ON(!p4d_present(p4d));
>>>>> + unmap_hotplug_pud_range(p4dp, addr, next, free_mapped);
>>>>> + } while (addr = next, addr < end);
>>>>> +}
>>>>> +
>>>>> static void unmap_hotplug_range(unsigned long addr, unsigned long end,
>>>>> bool free_mapped)
>>>>> {
>>>>> @@ -854,7 +872,7 @@ static void unmap_hotplug_range(unsigned long addr, unsigned long end,
>>>>> continue;
>>>>>
>>>>> WARN_ON(!pgd_present(pgd));
>>>>> - unmap_hotplug_pud_range(pgdp, addr, next, free_mapped);
>>>>> + unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped);
>>>>> } while (addr = next, addr < end);
>>>>> }
>>>>
>>>> Thanks Mike. With the additional diff below, I can get it to build with
>>>> and without the p4d clean-up patches in -next. If Anshuman confirms that
>>>> they work, I can add them on top of the arm64 for-next/memory-hotremove
>>>> branch.
>>>
>>> These two patches applied on next-20200302 works fine for hot-remove.
>>
>> Do they also work on top of the vanilla kernel + your hotremove patches
>> (i.e. not on top of -next)?
>
> Yes, they do work on current vanilla kernel (8b614cb8f1dcac8ca77cf4dd85f46)
> and v13 hotremove series.
>
>>
>>> As the second patch also fixes the first one, IMHO both should be
>>> folded into a single one instead. Just wondering if this combined
>>> patch which enables P4D page table should be posted on the list or do
>>> I need to respin original hot remove patches again.
>>
>> If your unmap patches plus the fixes from Mike and me work fine on top
>> of 5.6-rc3 (as well as when combined with linux-next), I'd prefer you
> Yes, they do work on both.
>
>> respin your patches to include the p4d support from start. Otherwise, we
>
> Okay. I will be sending V14 on v5.6-rc3 (OR v.5.6-rc4 is preferred ?) with
> p4d support. I will add yours and Mike's Signed-off-by as well.

Posted V14 (https://patchwork.kernel.org/project/linux-mm/list/?series=250939)
for the arm64 hot remove series on v5.6-rc3, though it also applies and works
on 5.6-rc4 as well.

>
>> create a single patch that Andrew can merge on top of the -mm tree.
>>