| Message ID | 20250407092243.2207837-1-xavier_qy@163.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [v1] mm/contpte: Optimize loop to reduce redundant operations | expand |
Thanks for the patch. Would the following change be better? diff --git a/arch/arm64/mm/contpte.c b/arch/arm64/mm/contpte.c index 55107d27d3f8..64eb3b2fbf06 100644 --- a/arch/arm64/mm/contpte.c +++ b/arch/arm64/mm/contpte.c @@ -174,6 +174,9 @@ pte_t contpte_ptep_get(pte_t *ptep, pte_t orig_pte) if (pte_young(pte)) orig_pte = pte_mkyoung(orig_pte); + + if (pte_young(orig_pte) && pte_dirty(orig_pte)) + break; } return orig_pte;
Hi Lance, Thanks for your feedback, my response is as follows. -- Thanks, Xavier At 2025-04-07 19:29:22, "Lance Yang" <ioworker0@gmail.com> wrote: >Thanks for the patch. Would the following change be better? > >diff --git a/arch/arm64/mm/contpte.c b/arch/arm64/mm/contpte.c >index 55107d27d3f8..64eb3b2fbf06 100644 >--- a/arch/arm64/mm/contpte.c >+++ b/arch/arm64/mm/contpte.c >@@ -174,6 +174,9 @@ pte_t contpte_ptep_get(pte_t *ptep, pte_t orig_pte) > > if (pte_young(pte)) > orig_pte = pte_mkyoung(orig_pte); >+ >+ if (pte_young(orig_pte) && pte_dirty(orig_pte)) >+ break; > } > > return orig_pte; >-- > >We can check the orig_pte flags directly instead of using extra boolean >variables, which gives us an early-exit when both dirty and young flags >are set. Your way of writing the code is indeed more concise. However, I think using boolean variables might be more efficient. Although it introduces additional variables, comparing boolean values is likely to be more efficient than checking bit settings. > >Also, is this optimization really needed for the common case? This function is on a high-frequency execution path. During debugging, I found that in most cases, the first few pages are already marked as both dirty and young. But currently, the program still has to complete the entire loop of 16 ptep iterations, which seriously reduces the efficiency. > >Thanks, >Lance
On Mon, Apr 7, 2025 at 8:56 PM Xavier <xavier_qy@163.com> wrote: > > > > Hi Lance, > > Thanks for your feedback, my response is as follows. > > -- > Thanks, > Xavier > > > > > > At 2025-04-07 19:29:22, "Lance Yang" <ioworker0@gmail.com> wrote: > >Thanks for the patch. Would the following change be better? > > > >diff --git a/arch/arm64/mm/contpte.c b/arch/arm64/mm/contpte.c > >index 55107d27d3f8..64eb3b2fbf06 100644 > >--- a/arch/arm64/mm/contpte.c > >+++ b/arch/arm64/mm/contpte.c > >@@ -174,6 +174,9 @@ pte_t contpte_ptep_get(pte_t *ptep, pte_t orig_pte) > > > > if (pte_young(pte)) > > orig_pte = pte_mkyoung(orig_pte); > >+ > >+ if (pte_young(orig_pte) && pte_dirty(orig_pte)) > >+ break; > > } > > > > return orig_pte; > >-- > > > >We can check the orig_pte flags directly instead of using extra boolean > >variables, which gives us an early-exit when both dirty and young flags > >are set. > Your way of writing the code is indeed more concise. However, I think > using boolean variables might be more efficient. Although it introduces > additional variables, comparing boolean values is likely to be more > efficient than checking bit settings. > > > > >Also, is this optimization really needed for the common case? > This function is on a high-frequency execution path. During debugging, > I found that in most cases, the first few pages are already marked as > both dirty and young. But currently, the program still has to complete > the entire loop of 16 ptep iterations, which seriously reduces the efficiency. Hmm... agreed that this patch helps when early PTEs are dirty/young, but for late-ones-only cases, it only introduces overhead with no benefit, IIUC. So, let's wait for folks to take a look ;) Thanks, Lance > > > >Thanks, > >Lance
Hi Xavier, On 07/04/25 7:01 pm, Lance Yang wrote: > On Mon, Apr 7, 2025 at 8:56 PM Xavier <xavier_qy@163.com> wrote: >> >> >> >> Hi Lance, >> >> Thanks for your feedback, my response is as follows. >> >> -- >> Thanks, >> Xavier >> >> >> >> >> >> At 2025-04-07 19:29:22, "Lance Yang" <ioworker0@gmail.com> wrote: >>> Thanks for the patch. Would the following change be better? >>> >>> diff --git a/arch/arm64/mm/contpte.c b/arch/arm64/mm/contpte.c >>> index 55107d27d3f8..64eb3b2fbf06 100644 >>> --- a/arch/arm64/mm/contpte.c >>> +++ b/arch/arm64/mm/contpte.c >>> @@ -174,6 +174,9 @@ pte_t contpte_ptep_get(pte_t *ptep, pte_t orig_pte) >>> >>> if (pte_young(pte)) >>> orig_pte = pte_mkyoung(orig_pte); >>> + >>> + if (pte_young(orig_pte) && pte_dirty(orig_pte)) >>> + break; >>> } Quite the coincidence, I was thinking of doing exactly this some days back and testing it out : ) Can you do a microanalysis whether this gets us a benefit or not? This looks like an optimization on paper but may not be one after all because CONT_PTES is only 16 and a simple loop without extra if-conditions may just be faster. >>> >>> return orig_pte; >>> -- >>> >>> We can check the orig_pte flags directly instead of using extra boolean >>> variables, which gives us an early-exit when both dirty and young flags >>> are set. >> Your way of writing the code is indeed more concise. However, I think >> using boolean variables might be more efficient. Although it introduces >> additional variables, comparing boolean values is likely to be more >> efficient than checking bit settings. >> >>> >>> Also, is this optimization really needed for the common case? >> This function is on a high-frequency execution path. During debugging, >> I found that in most cases, the first few pages are already marked as >> both dirty and young. But currently, the program still has to complete >> the entire loop of 16 ptep iterations, which seriously reduces the efficiency. > > Hmm... agreed that this patch helps when early PTEs are dirty/young, but > for late-ones-only cases, it only introduces overhead with no benefit, IIUC. > > So, let's wait for folks to take a look ;) > > Thanks, > Lance > >>> >>> Thanks, >>> Lance >
diff --git a/arch/arm64/mm/contpte.c b/arch/arm64/mm/contpte.c index bcac4f55f9c1..ca15d8f52d14 100644 --- a/arch/arm64/mm/contpte.c +++ b/arch/arm64/mm/contpte.c @@ -163,17 +163,26 @@ pte_t contpte_ptep_get(pte_t *ptep, pte_t orig_pte) pte_t pte; int i; + bool dirty = false; + bool young = false; ptep = contpte_align_down(ptep); for (i = 0; i < CONT_PTES; i++, ptep++) { pte = __ptep_get(ptep); - if (pte_dirty(pte)) + if (!dirty && pte_dirty(pte)) { + dirty = true; orig_pte = pte_mkdirty(orig_pte); + } - if (pte_young(pte)) + if (!young && pte_young(pte)) { + young = true; orig_pte = pte_mkyoung(orig_pte); + } + + if (dirty && young) + break; } return orig_pte;
This commit optimizes the contpte_ptep_get function by adding early termination logic. It checks if the dirty and young bits of orig_pte are already set and skips redundant bit-setting operations during the loop. This reduces unnecessary iterations and improves performance. Signed-off-by: Xavier <xavier_qy@163.com> --- arch/arm64/mm/contpte.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-)