Re: linux-next: Tree for Apr 6 (arch/x86/mm/init_64.c)

Re: linux-next: Tree for Apr 6 (arch/x86/mm/init_64.c)

[Randy Dunlap]

On 4/6/21 5:31 AM, Stephen Rothwell wrote:
> Hi all,
> 
> Changes since 20210401:
> 

on x86_64:
# CONFIG_SPARSEMEM_VMEMMAP is not set


gcc (SUSE Linux) 7.5.0


../arch/x86/mm/init_64.c: In function 'remove_pmd_table':
../arch/x86/mm/init_64.c:1127:8: error: implicit declaration of function 'vmemmap_pmd_is_unused'; did you mean 'vmemmap_pmd_populate'? [-Werror=implicit-function-declaration]
        vmemmap_pmd_is_unused(addr, next)) {
        ^~~~~~~~~~~~~~~~~~~~~
        vmemmap_pmd_populate


Looks like that compound "if" is too much for gcc 7.5.0 to handle:

			} else if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP) &&
				   vmemmap_pmd_is_unused(addr, next)) {
					free_hugepage_table(pmd_page(*pmd),
							    altmap);
					spin_lock(&init_mm.page_table_lock);
					pmd_clear(pmd);
					spin_unlock(&init_mm.page_table_lock);
			}


-- 
~Randy
Reported-by: Randy Dunlap <rdunlap@infradead.org>

Re: linux-next: Tree for Apr 6 (arch/x86/mm/init_64.c)

[Randy Dunlap]

On 4/6/21 11:32 AM, Randy Dunlap wrote:
> On 4/6/21 5:31 AM, Stephen Rothwell wrote:
>> Hi all,
>>
>> Changes since 20210401:
>>
> 
> on x86_64:
> # CONFIG_SPARSEMEM_VMEMMAP is not set
> 
> 
> gcc (SUSE Linux) 7.5.0
> 
> 
> ../arch/x86/mm/init_64.c: In function 'remove_pmd_table':
> ../arch/x86/mm/init_64.c:1127:8: error: implicit declaration of function 'vmemmap_pmd_is_unused'; did you mean 'vmemmap_pmd_populate'? [-Werror=implicit-function-declaration]
>         vmemmap_pmd_is_unused(addr, next)) {
>         ^~~~~~~~~~~~~~~~~~~~~
>         vmemmap_pmd_populate
> 
> 
> Looks like that compound "if" is too much for gcc 7.5.0 to handle:
> 
> 			} else if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP) &&
> 				   vmemmap_pmd_is_unused(addr, next)) {
> 					free_hugepage_table(pmd_page(*pmd),
> 							    altmap);
> 					spin_lock(&init_mm.page_table_lock);
> 					pmd_clear(pmd);
> 					spin_unlock(&init_mm.page_table_lock);
> 			}
> 
> 

This is what I am using for now:

---
---
 arch/x86/mm/init_64.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

--- linux-next-20210406.orig/arch/x86/mm/init_64.c
+++ linux-next-20210406/arch/x86/mm/init_64.c
@@ -1123,14 +1123,16 @@ remove_pmd_table(pmd_t *pmd_start, unsig
 				pmd_clear(pmd);
 				spin_unlock(&init_mm.page_table_lock);
 				pages++;
-			} else if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP) &&
-				   vmemmap_pmd_is_unused(addr, next)) {
+			}
+#ifdef CONFIG_SPARSEMEM_VMEMMAP
+			else if (vmemmap_pmd_is_unused(addr, next)) {
 					free_hugepage_table(pmd_page(*pmd),
 							    altmap);
 					spin_lock(&init_mm.page_table_lock);
 					pmd_clear(pmd);
 					spin_unlock(&init_mm.page_table_lock);
 			}
+#endif
 
 			continue;
 		}

Re: linux-next: Tree for Apr 6 (arch/x86/mm/init_64.c)

[Oscar Salvador]

On Tue, Apr 06, 2021 at 11:39:47AM -0700, Randy Dunlap wrote:
 
> > Looks like that compound "if" is too much for gcc 7.5.0 to handle:
> > 
> > 			} else if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP) &&
> > 				   vmemmap_pmd_is_unused(addr, next)) {
> > 					free_hugepage_table(pmd_page(*pmd),
> > 							    altmap);
> > 					spin_lock(&init_mm.page_table_lock);
> > 					pmd_clear(pmd);
> > 					spin_unlock(&init_mm.page_table_lock);
> > 			}
> > 
> > 
> 
> This is what I am using for now:

Hi Randy

Yeah, that is what v4 was using [1].
We decided to get rid of the ifdef for costimetic reasons but it seems
it does not do the trick.

I will ask Andrew to squash that on top.

[1] https://patchwork.kernel.org/project/linux-mm/patch/20210301083230.30924-4-osalvador@suse.de/

Thanks

> 
> ---
> ---
>  arch/x86/mm/init_64.c |    6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> --- linux-next-20210406.orig/arch/x86/mm/init_64.c
> +++ linux-next-20210406/arch/x86/mm/init_64.c
> @@ -1123,14 +1123,16 @@ remove_pmd_table(pmd_t *pmd_start, unsig
>  				pmd_clear(pmd);
>  				spin_unlock(&init_mm.page_table_lock);
>  				pages++;
> -			} else if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP) &&
> -				   vmemmap_pmd_is_unused(addr, next)) {
> +			}
> +#ifdef CONFIG_SPARSEMEM_VMEMMAP
> +			else if (vmemmap_pmd_is_unused(addr, next)) {
>  					free_hugepage_table(pmd_page(*pmd),
>  							    altmap);
>  					spin_lock(&init_mm.page_table_lock);
>  					pmd_clear(pmd);
>  					spin_unlock(&init_mm.page_table_lock);
>  			}
> +#endif
>  
>  			continue;
>  		}
> 

-- 
Oscar Salvador
SUSE L3

Re: linux-next: Tree for Apr 6 (arch/x86/mm/init_64.c)

[Matthew Wilcox]

On Tue, Apr 06, 2021 at 09:57:03PM +0200, Oscar Salvador wrote:
> On Tue, Apr 06, 2021 at 11:39:47AM -0700, Randy Dunlap wrote:
>  
> > > Looks like that compound "if" is too much for gcc 7.5.0 to handle:
> > > 
> > > 			} else if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP) &&
> > > 				   vmemmap_pmd_is_unused(addr, next)) {
> > > 					free_hugepage_table(pmd_page(*pmd),
> > > 							    altmap);
> > > 					spin_lock(&init_mm.page_table_lock);
> > > 					pmd_clear(pmd);
> > > 					spin_unlock(&init_mm.page_table_lock);
> > > 			}
> > > 
> > > 
> > 
> > This is what I am using for now:
> 
> Hi Randy
> 
> Yeah, that is what v4 was using [1].
> We decided to get rid of the ifdef for costimetic reasons but it seems
> it does not do the trick.
> 
> I will ask Andrew to squash that on top.

This is pretty ugly.  What I was doing was this:

+++ b/arch/x86/mm/init_64.c
@@ -862,6 +862,9 @@ static bool __meminit vmemmap_pmd_is_unused(unsigned long addr, unsigned long en
 
        return !memchr_inv((void *)start, PAGE_UNUSED, PMD_SIZE);
 }
+#else
+static inline bool vmemmap_pmd_is_unused(unsigned long addr, unsigned long end)
+{ return false; }
 #endif
 
 static void __meminit __vmemmap_use_sub_pmd(unsigned long start)
@@ -924,6 +927,9 @@ static void __meminit vmemmap_use_new_sub_pmd(unsigned long start, unsigned long
        if (!IS_ALIGNED(end, PMD_SIZE))
                unused_pmd_start = end;
 }
+#else
+static inline bool vmemmap_pmd_is_unused(unsigned long addr, unsigned long end)
+{ return false; }
 #endif
 
 /*

(whitespace damaged, but ...)