| Message ID | 20160904053331.7264-1-stefan@agner.ch (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Sat, Sep 03, 2016 at 10:33:31PM -0700, Stefan Agner wrote: > The cachepolicy variable gets initialized using a masked pmd > So far, the pmd has been masked with flags valid for the 2-page > table format. In the LPAE case, this lead to a wrong assumption > of what the initial cachepolicy has been used. Later a check > forces the cache policy to writealloc and prints the following > warning: > Forcing write-allocate cache policy for SMP > > This patch uses PMD_SECT_WBWA to mask all cache setting flags. > The define represents the complete mask of the cache relevant > flags for both page table formats. PMD_SECT_WBWA is just one possible combination, it's not a bit-mask. We need a new definition.
On 2016-09-04 15:09, Russell King - ARM Linux wrote: > On Sat, Sep 03, 2016 at 10:33:31PM -0700, Stefan Agner wrote: >> The cachepolicy variable gets initialized using a masked pmd >> So far, the pmd has been masked with flags valid for the 2-page >> table format. In the LPAE case, this lead to a wrong assumption >> of what the initial cachepolicy has been used. Later a check >> forces the cache policy to writealloc and prints the following >> warning: >> Forcing write-allocate cache policy for SMP >> >> This patch uses PMD_SECT_WBWA to mask all cache setting flags. >> The define represents the complete mask of the cache relevant >> flags for both page table formats. > > PMD_SECT_WBWA is just one possible combination, it's not a bit-mask. > We need a new definition. Agreed, it just happens to be the same as the bit-mask in both page-table layouts. Does PMD_SECT_CACHE_MASK sounds like a reasonable identifier for it? -- Stefan
On Sun, Sep 04, 2016 at 04:31:40PM -0700, Stefan Agner wrote: > On 2016-09-04 15:09, Russell King - ARM Linux wrote: > > On Sat, Sep 03, 2016 at 10:33:31PM -0700, Stefan Agner wrote: > >> The cachepolicy variable gets initialized using a masked pmd > >> So far, the pmd has been masked with flags valid for the 2-page > >> table format. In the LPAE case, this lead to a wrong assumption > >> of what the initial cachepolicy has been used. Later a check > >> forces the cache policy to writealloc and prints the following > >> warning: > >> Forcing write-allocate cache policy for SMP > >> > >> This patch uses PMD_SECT_WBWA to mask all cache setting flags. > >> The define represents the complete mask of the cache relevant > >> flags for both page table formats. > > > > PMD_SECT_WBWA is just one possible combination, it's not a bit-mask. > > We need a new definition. > > Agreed, it just happens to be the same as the bit-mask in both > page-table layouts. > > Does PMD_SECT_CACHE_MASK sounds like a reasonable identifier for it? Yep, thanks.
diff --git a/arch/arm/mm/mmu.c b/arch/arm/mm/mmu.c index 724d6be..241e5e2 100644 --- a/arch/arm/mm/mmu.c +++ b/arch/arm/mm/mmu.c @@ -137,7 +137,7 @@ void __init init_default_cache_policy(unsigned long pmd) initial_pmd_value = pmd; - pmd &= PMD_SECT_TEX(1) | PMD_SECT_BUFFERABLE | PMD_SECT_CACHEABLE; + pmd &= PMD_SECT_WBWA; for (i = 0; i < ARRAY_SIZE(cache_policies); i++) if (cache_policies[i].pmd == pmd) {
The cachepolicy variable gets initialized using a masked pmd So far, the pmd has been masked with flags valid for the 2-page table format. In the LPAE case, this lead to a wrong assumption of what the initial cachepolicy has been used. Later a check forces the cache policy to writealloc and prints the following warning: Forcing write-allocate cache policy for SMP This patch uses PMD_SECT_WBWA to mask all cache setting flags. The define represents the complete mask of the cache relevant flags for both page table formats. Signed-off-by: Stefan Agner <stefan@agner.ch> --- arch/arm/mm/mmu.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)