Dedicated SV copying code in place of Perl_sv_setsv_flags #23202

richardleach · 2025-04-15T23:29:12Z

Perl_sv_setsv_flags is the heavyweight function for assigning the value(s) of
a source SV to a destination SV. It contains many branches for preparing the
destination SV prior to assignment. However:

If the destination SV has just been created, much of that logic isn't needed.
When cloning a SV, simple assignments (particularly IVs and PVs) dominate.

This set of commits:

Extracts the "is this CoWable?" test from Perl_sv_setsv_flags into a macro.
Adds Perl_sv_freshcopy_flags and two static helper functions.
Modifies Perl_newSVsv_flags and Perl_sv_mortalcopy_flags to use them.
Standardizes a number of call sites that did their own things but really
should use Perl_newSVsv_flags or Perl_sv_mortalcopy_flags.

Using perl's test harness as a guide:

Bodyless code handles 45% of calls to Perl_newSVsv_flags and
57% of calls to Perl_sv_mortalcopy_flags.
The SVt_PV/SVp_POK code handles 32% of calls to
Perl_newSVsv_flags and 36% of calls to Perl_sv_mortalcopy_flags.
S_sv_freshcopy_flags code handles 95% of the remainder in
Perl_newSVsv_flags and 91% of the remainder in to Perl_sv_mortalcopy_flags.

With these changes compared with a build of blead:

perl -e 'for (1..100_000) { my $x = [ (1) x 1000 ]; }' runs 10% faster
perl -e 'for (1..100_000) { my $x = [ ("Perl") x 250 ]; }' runs 45% faster

This set of changes does require a perldelta entry and I'll write one post-merge.

pp.c

pp_hot.c

sv.c

Leont · 2025-04-29T20:52:21Z

Cloning is rather unfortunate choice of words, given that it has a very specific meaning in our codebase that is quite different from what this PR is about. Renaming the PR may be helpful.

richardleach · 2025-05-08T17:06:30Z

I've made a lot of changes following earlier comments - thanks for those - and have finally force-pushed.

These changes aren't complete. For example:

Measured performance seems worse than in the PR version, so I need to look into that
Might change sflag handling/ SvFLAGS(dsv) setting
Not settled on struct membet initialisation
Might still rename the function that is currently Perl_newSVsv_flags and have newSVsv_flags be a macro that checks (ssv) before calling the sv.c function.

richardleach · 2025-05-08T17:13:02Z

sv.c

+                   if (sflags & SVf_IsCOW) {
+                        sv_buf_to_rw(ssv);
+                    }


I'm unclear if there's a bug here carried over from Perl_sv_setsv_flags.

If (!(sflags & SVf_IsCOW)) above, sflags isn't updated to reflect that SVf_Is_COW is now set on ssv, but why would we not want to do sv_buf_to_rw(ssv) in that situation?

I'm going to make another PR about a very sketchy function in Perl 2003-2025 with 1 parent caller, its name is sv_setsv_cow(). I might be mistaken, but your code highlight, and what I think 1 branch in sv_setsv_cow() is doing, in my eyes, is a virgin never been HW ROed buffer, that for the first time will be getting HW RO perms added, therefore no ring 0 call is needed to mark it RW perms. verify I am making any sense, this feature isn't available on Windows and I don't see the point/personal necessity to port it to Windows. It would take me 2-5 hours to do that, somewhere between easy and monotonous/unchallenging on the difficulty scale.

Edit:

Since my Win32::API::Callback for many years already has MsMalloc() blocks that are inter-ithread safe, clone safe, Perl ref counted, executable perm-ed (x86/x64 mach code) read only perm-ed (DEP/being a good citizen). If I wrote such an API, I would probably write the patch for WinPerl to have HW RO COW buffers for production grade default WinPerl stable. Win32::API::Callback flips the RO RW perm bit manually, with minor help from the MS-only Public API HeapCreate().

AFAIK Father C's (???) #define PERL_DEBUG_READONLY_COW is un-usable in a production stable perl b/c extreme memory waste. I think it give each 'Newx/malloc()block an entire 4096 byte long page valgrind/dr memory style. Win 7's undocumented turn-key HW RO malloc feature,RtlProtectHeap()has no !!! memory/RAM overhead, since the blocks still have WinOS's default 8 byte long headers. RtlProtectHeap()'s runtime CPU overhead, ...., well ....., that is a different story. RtlProtectHeap()internally is a large tree of if/else branches and some loops. B/c its closed source, I can't describe exactly what all the asm code logic does. ButRtlProtectHeap()is very simple internally, the only 2 functions it is capable of calling areZwQueryVirtualMemory()andZwProtectVirtualMemory()which areinterrupt 0x2e` system calls.

More realistically, Perl 5 needs to just verbatim copy Chrome Browser's hardware RO COW/Heap implementation. I'm sure by now Webkit and Mozilla also use Chrome's HW RO COW/Heap design.

Remember Chrome's design is penetration tested and extensively fuzzed, to prevent rootkit-ed/exploited chroot jailed OS processes aka same origin browser tabs aka V8 isolates (my_perl in Perl-ese) aka JS VMs instances, where blackhat javascript code escaped the javascript sandbox and now has C lang/Asm machine code control over an OS thread in a chroot jailed OS process.

Chrome's design is simple, 2 mmaps at different address ranges, or in 2 different OS processes, to the same phy RAM pages, or to the same pagefile virtual RAM pages. The JS runtime/runloop engine only knows the address of RO-perm-ed view or address range of mmap() object. The hypervisor, which is a same process different OS thread, or a different OS process, knows the location of BOTH address ranges, the RO public address range, and the RW secret address range. The "hypervisor" (in Perl-ese that means the "proprietary undocumented internals of sv.c and libperl.so"), so the "hypervisor", gets called from untrusted JS bytecode, to RC++ or RC-- an JS data-ish object or a JS Function object, and the hypervisor uses the secret RW memory window to change the reference count on a ref counted/color counted JS object, Untrusted byte code can read the RC and other JS object header fields (SV * header), but untrusted JS byte code, even if exploited to machine code/C level, can't rewrite a COWed/RO/immutable/static/const JS object.

Since libperl.so and PP code fundamentally, as designed, are a https://en.wikipedia.org/wiki/Arbitrary_code_execution exploit, P5P's only duty is user error/dev error/idiot mitigation. Perl and P5P have minimal security obligation to think of, P5P would just embed the RO ptr mmap to RW ptr mmap conversion equation into macro #define share_hek_hek(hek) and mission accomplished,

If a user has suffered damages from Perl or /bin/sh being an arbitrary code execution exploit, please send all legal threats to https://kernel.org/ and https://www.opengroup.org and https://canonical.com/ .

Since around the 5.10 era, `Perl_sv_setsv_flags` has unconditionally set `SvPOK_only(dsv)` in the `SVp_POK` branch. The associated comment reads: /* Whichever path we take through the next code, we want this true, and doing it now facilitates the COW check. */ Things have changed since 5.10 though, in particular using the `‎SVf_POK` to distinguish between a value that started off as a string from one that was originally an integer/float and later stringified. This commit: * Removes the `SvPOK_only(dsv)` in favour of `SvOK_off(dsv)` and hoisting the copying of `sflags` over. * Transforms the subsequent now-redundant `SVf_POK` toggles into asserts (to help reduce) the chance of inadvertent behaviour changes.

`Perl_sv_setsv_flags` is a hot function that contains liberal sprinklings of `SvOK_off()`. This commit changes two instances, where the operand SV cannot possibly be using the OOK hack, to do direct flag twiddling instead. `SvOK_off()` does two things: 1. Toggles off some flags: SvFLAGS(sv) &= ~(SVf_OK|SVf_IVisUV|SVf_UTF8) 2. Checks for use of the OOK hack and undoes it: ((void)(SvOOK(sv) && (sv_backoff(sv),0))) At least some compilers seem to struggle to figure out when `SvOOK(sv)` cannot be true and to then elide the call to `sv_backoff()`. This is desirable when: 1. ssv & dsv are both lower types than SVt_PV and cannot support OOK 2. inside a block following a conditional check that OOK is not in use In the two cases identified, the flag toggling is now done explicitly.

Perl_sv_freshcopy_flags creates a fresh SV that contains the values of it source SV argument. It's like calling `new_SV(dsv)` followed by `sv_setsv_flags(dsv, ssv, flags`, but is optimized for a brand new destination SV and the most common code paths. The intended initial users for this new function were: * Perl_sv_mortalcopy_flags (still in sv.c) * Perl_newSVsv_flags (now a simple function in sv_inline.h) Perl_sv_freshcopy_flags handles the following cases: * SVt_NULL * SVt_IV * SVt_NV * SVt_PV * SVt_LAST For everything else, it calls S_sv_freshcopy_PVxx which handles: * SVt_INVLIST * SVt_PVIV * SVt_PVNV * SVt_PVMG - with no GET magic For everything else, there's a fall back to sv_setsv_flags. S_sv_freshcopy_POK is a dedicated helper for string swipe/COW/copy logic and is called from both Perl_sv_freshcopy_flags and S_sv_freshcopy_PVxx. With these changes compared with the previous commit: * `perl -e 'for (1..100_000_0) { my $x = { (1) x 1000 }; }'` runs about 20% faster * `perl -e 'for (1..100_000_0) { my $x = { ("Perl") x 250 }' runs about 40% faster * `perl -e 'for (1..100_000_0) { my $x = { a => 1, b => 2, c => 3, d => 4, e => 5 }; }'` is a touch faster, but within the margin for error * `perl -e 'for (1..100_000_0) { my $x = { a => "Perl", b => "Perl", c => "Perl", d => "Perl", e => "Perl" } ; }'` runs about 17% faster

Besides using the just-introduced faster path for SV copying, this allows the check for SV_GMAGIC to be pushed into the called function without having to worry about SV leaks. Two additional micro-optimizations are also in this commit: * A pointer to xav_fill is cached for use in the loop. This can be used directly to update AvFILLp(av), rather than having to get there from av's SV* each time. * The value of the loop iterator, i, is directly written into xav_fill, rather than getting the value in that slot, incrementing it (to get the same value as i), and writing it back.

pp.c

sv.c

bulk88 · 2025-06-13T23:46:41Z

sv.c

-        (void)SvPOK_only(dsv);
+        (void)SvOK_off(dsv);
+        SvFLAGS(dsv) |= sflags &
+            (SVf_POK|SVp_POK|SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);


why isnt SVF_ISUV copied over from sflags? did anything earlier wipe the SVf_UTF8 inside SvFLAGS(dsv)

Thanks for pointing this out, I think the flag handling around this bit of code could be tidied up further. Will look at it again.

Remember (void)SvOK_off(dsv); being a func call was IIRC a Father C 2014 or 2012 quick fix for random memory corruption/leaks involving SvOOK() back when the OOK delta count was stored in a SVt_PVIV inside field SvIVX(). Someone later on changed OOK()'s design to be a BER/ASN.1/SMI/compressed integer stored in the PV buffer. Eliminating the former design limitation, that an OOK POK SV, can NEVER EVER EVER be a dual var with POK + OOK + IOK.

I'm not gonna touch cactus that is sv_backoff()/SvOK_off(dsv) unless I'm at a YAPC or a Toolchain summit with another qualified Perl C human 3-6 feet away.

bulk88 · 2025-06-13T23:52:15Z

sv.c

+        /* Passes the swipe test.  */
+        SvLEN_set(dsv, len);
+        SvCUR_set(dsv, cur);
+        SvPV_set(dsv, SvPVX_mutable(ssv));


split mem read SvPVX_mutable(ssv) from SvPV_set(dsv, ), and move the mem read SvPVX_mutable(ssv) and save it to a C auto, right before SvLEN_set(dsv, len);. pointer aliasing rules in C. C abstract machine doesn't know if SvCUR_set(dsv, cur) will write ontop of 8 bytes of memory backing SvPVX_mutable(ssv).

This doesn't make any difference on gcc builds. What do you see on MSVC builds?

bulk88 · 2025-06-13T23:56:04Z

sv.c

+        /* (void)SvOK_off(ssv); but without the superfluous SvOOK_off(ssv)) */
+        SvFLAGS(ssv) &= ~(SVf_OK|SVf_IVisUV|SVf_UTF8|SVs_TEMP);
+
+        SvPV_set(ssv, NULL);


move this line to below SvCUR_set(ssv, 0);, pointer aliasing, ILP, pipeline stall reasons. The value of SvANY(ssv) is unknown until SvPV_set(ssv, NULL); is 100% completed by the CPU pipeline/ CPU conveyor belt.

Move which of these two lines? Neither of them need to know SvANY(ssv), so I'm confused.

bulk88 · 2025-06-14T00:08:16Z

sv.c

+#else
+            dsv->sv_u.svu_rv = SvREFCNT_inc( old->sv_u.svu_rv );
+#endif
+            SvFLAGS(dsv) = SVt_IV|SVf_ROK;


move SvFLAGS(dsv) = SVt_IV|SVf_ROK; up to between these 2 lines

assert(SvANY(dsv)); /////////////// here #if defined (DEBUGGING) || defined (PERL_DEBUG_COW)

that way

dsv->sv_u.svu_iv = old->sv_u.svu_iv;

dsv->sv_u.svu_rv = SvREFCNT_inc( old->sv_u.svu_rv );

are likely to get some cpu ops deduped by the CC

GCC adds in more movs and a few additional branches with that change. I'd need more convincing to put it in.

bulk88 · 2025-06-14T00:09:36Z

sv.c

+#if defined (DEBUGGING) || defined (PERL_DEBUG_COW)
+            dsv->sv_u.svu_rv = SvREFCNT_inc(SvRV(old));
+#else
+            dsv->sv_u.svu_rv = SvREFCNT_inc( old->sv_u.svu_rv );


ADD SvREFCNT_inc_simple_NN(sv) its faster

It's 100% safe to assume that (SvRV(old)) is true?

bulk88 · 2025-06-14T00:13:12Z

sv.c

+ *        taint and vstring magic, which are also handled here. */
+
+static void
+S_sv_freshcopy_PVxx(pTHX_ SV* dsv, SV* ssv, const I32 flags)


Change retval type to SV* and return SV* dsv as the retval. Never waste a free register. Tailcall or reducing liveness or reducing peak non volatile reg usage opportunities exist now, or will exist in the future.

I did try it and (IIRC) it performed either worse or no better. I'll look again before force-pushing changes for some of the other comments above though.

It wouldn't be benchmark able, Its just a provision for the future. A human still needs to single step the asm code and manually look for callsites where that "setter" call is very close or right above a possible return some_size_t; statement.

I have a never finished branch turning all to Perl_sv_set**v**_*mg*() to return their SV* arg, sv_2mortal() style. Doing the change I described would allow 25%-50% of all XSUBs on CPAN, to use 100% volatile CPU registers, and zero reads/writes to the C stack, and zero save/restores of non-vol registers, since that 25%-50% group of XSUBs don't need to carry any C land variables, around any function call in their body.

bulk88 · 2025-06-14T00:22:20Z

sv.c

+ * This function is marked for inlining, also to benefit the hot SVt_PV case.
+ */
+PERL_STATIC_FORCE_INLINE void
+S_sv_freshcopy_POK(pTHX_ SV* dsv, SV* ssv, const I32 flags)


change void to SV*, return as retval SV* dsv see oth cmt 4 rational

See reply to other comment. ;) I will still look at it again though.

richardleach added the defer-next-dev This PR should not be merged yet, but await the next development cycle label Apr 15, 2025