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 rppt@linux.ibm.com Date: Mon, 2 Mar 2020 15:53:17 +0200 Subject: [PATCH] arm64/mm: implement unmap_hotplug_p4d_range
Signed-off-by: Mike Rapoport rppt@linux.ibm.com
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);}
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.