A memblock patch uses a binary search and cites a drop from about 268 seconds to about 50 milliseconds. Phoronix says that patch went out for Linux 7.4.
Tarun Sahu's memblock patch reports about 268 seconds falling to about 50 milliseconds. Phoronix says the 7.4 pull was sent. A sent pull is not a tag.
The short version
The October 8 patch from Tarun Sahu says a binary search in memblock_add_range and memblock_isolate_range cut one KHO retrieval from about 268 seconds to about 50 milliseconds, on about 393,000 pages. Phoronix on October 24 says that work was sent in the memblock pull for Linux 7.4. This page does not call a sent pull a merged tag. One sentence in Phoronix says 238 seconds. The patch text says about 268.
The patch is older than the headline
Phoronix's October 24, 2026 story says a large speedup in Linux's early physical-memory allocator went out as part of the memblock pull for Linux 7.4. The change it describes is not a patch written that morning. Tarun Sahu's version 4, message dated October 8, 2026, is the text. It says the work uses a binary search in memblock_add_range and memblock_isolate_range. The lore copy and the IU archive are the same message on two hosts. Phoronix is the outside account of that message going into a pull. Two copies of one email are one document. This page says so.the short slice is the other 7.4 change already filed.
268 seconds is the patch's number
The patch describes a KHO retrieval of about 393,000 pages that took about 268 seconds and then about 50 milliseconds with the search. That is one workload the author measured, not a general claim about boot time on every machine. Phoronix's headline uses the 268 second figure. Another sentence in the same article says about 238 seconds. This page does not average them. The patch is the measurement. The 238 is a slip unless a later message corrects the patch, and this page has not seen that correction.systemd's Rust leaves are not this allocator.
A pull sent is not a release
Phoronix says the work was sent for the 7.4 memblock pull. Sending a pull during a merge window is a maintainer's request that Linus take the commits. It is not a tag, and it is not the kernel a distribution will ship next week. The LWN note on the first part of the 7.4 merge window is a separate article, and this page does not treat a paywalled overview as confirmation that this particular hunk landed. Until a pull is in the tree, the status is the one Phoronix used: sent.Node's October release is a different project.
What the search is allowed to mean
The patch says the allocator was walking ranges in a way that got expensive when a lot of pages had to be put back. A binary search over the existing memblock structure is the author's description of the fix. This page names those two functions because they are the subject of the patch, and it does not paste the diff, the helper, or a benchmark harness. A reader who wants the C can open the message. A reader who wants the result can stop at 268 seconds to 50 milliseconds, on that retrieval, as the author wrote it.a CVE identifier is not a kernel timing.
Who measured it
The figure belongs to Tarun Sahu's message. Pratyush Yadav is the reviewer named on the version this page is using. Google is the affiliation on the message. None of that makes the number a Google product benchmark or a number this paper re-ran. KHO, in the patch's usage, is the retrieval path being timed. This page does not expand that into a boot tutorial. Asked directly, Which number should a reader keep, 268 seconds or 238? The patch says about 268 seconds down to about 50 milliseconds. Phoronix writes 268 in the headline and 238 in one sentence. This page keeps the patch.
Why it is not the short slice
This paper already filed a Linux 7.4 change that shortens a short slice of another path. That brief stays where it is. Memblock's search is a different patch, a different author, and a different measurement. Linking them is a way to keep the merge window from looking like one commit. It is not a claim that one review covered both. Asked directly, Is the memblock speedup the same change as the Linux 7.4 short slice? No. The short slice is a different patch this paper already filed. This page is the memblock search. Same kernel series, different change.
Who is in scope
The people who feel this, if the pull lands, are machines that hit the slow path the author timed, which the patch presents as a large KHO retrieval rather than an ordinary desktop boot. Everyone else should not expect a 268 second pause to disappear from a laptop they never saw pause. Distributions that have not taken 7.4 are not running it. The sent pull is the status on October 24.Claude's proof notes cover a different binary.
The archives
The IU link and the lore link are the same October 8 message. Citing both is how a reader finds a copy if one host is awkward, not how this page pretends two researchers agreed. Phoronix is the only outside writeup in the sources. If the pull is later dropped, the patch still says what it said on October 8, and the news story still says it was sent.
The limit
No diff. No claim of a merged tag. No 238 second figure promoted to the result. No mixing with the short-slice brief. The record is the patch's 268 seconds and 50 milliseconds, and Phoronix's statement that the pull was sent for 7.4. The short answer this page will stand on is this: The October 8 patch from Tarun Sahu says a binary search in memblock_add_range and memblock_isolate_range cut one KHO retrieval from about 268 seconds to about 50 milliseconds, on about 393,000 pages. Phoronix on October 24 says that work was sent in the memblock pull for Linux 7.4. This page does not call a sent pull a merged tag. One sentence in Phoronix says 238 seconds. The patch text says about 268.
Sources
The reports this brief is filing. Futureweb did not republish them.
Questions
Is the memblock speedup the same change as the Linux 7.4 short slice?
No. The short slice is a different patch this paper already filed. This page is the memblock search. Same kernel series, different change.
Which number should a reader keep, 268 seconds or 238?
The patch says about 268 seconds down to about 50 milliseconds. Phoronix writes 268 in the headline and 238 in one sentence. This page keeps the patch.
Did Linux 7.4 release with this memblock change?
Phoronix says the pull was sent for 7.4. A pull request sent during a merge window is not, by itself, a released kernel. This page does not point at a tag.
Who wrote the memblock patch?
Tarun Sahu, at Google, in a version 4 posted October 8, 2026. Pratyush Yadav is named as a reviewer on that message. The timing figure is the author's.
Does this page include the memblock diff?
No. The functions named are the ones the patch says it changes. The source listing stays on the list.