linux-kernel - Re: [RFC PATCH v1 0/2] cxl/cli: HPA-ordered destroy-region teardown

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <aS-wqvEjYUGEUs-s@aschofie-mobl2.lan>
Date: Tue, 2 Dec 2025 19:38:18 -0800
From: Alison Schofield <alison.schofield@...el.com>
To: Pawel Mielimonka <pawel.mielimonka@...itsu.com>
CC: <dan.j.williams@...el.com>, <Smita.KoralahalliChannabasappa@....com>,
	<linux-cxl@...r.kernel.org>, <linux-kernel@...r.kernel.org>,
	<dave@...olabs.net>, <jonathan.cameron@...wei.com>, <dave.jiang@...el.com>,
	<vishal.l.verma@...el.com>, <ira.weiny@...el.com>
Subject: Re: [RFC PATCH v1 0/2] cxl/cli: HPA-ordered destroy-region teardown

On Tue, Nov 25, 2025 at 11:38:22PM +0900, Pawel Mielimonka wrote:
> This series addresses an issue in destroy-region, where
> region teardown relied on libcxl enumeration order, which is not
> guaranteed to match the increasing HPA order exposed by the kernel. 
> When a decoder window is fully populated, attempting to destroy a
> non-last region causes kernel-side validation to fail (e.g.
> set_dpa_size(..., 0) returns an error), and subsequent destroy/create
> sequences may become impossible.
> 
> The CXL specification requires that decoder programming (and the
> implicit teardown path) must preserve continuous HPA coverage and
> proceed strictly in order: decoder m before decoder m+1, with each
> covering an HPA range below the next one. Practically, this means that
> region teardown must follow HPA-descending order and must stop as soon
> as a gap in the requested suffix is encountered.
> 
> The patch introduces destroy_multiple_regions(), which collects all
> regions under a given root decoder, sorts them by HPA, and destroys
> only the suffix requested by the user (or all regions in the case of
> “all”). The implementation guarantees that only valid teardown
> sequences are attempted and prevents decoder state inconsistencies
> observed during repeated destroy/create cycles.
> 
> Enable/disable paths and all existing bus/port/decoder filtering
> remain unchanged.

Hi Pawel,

Thanks - this is much needed!

It's very valuable that now trying to destroy any region out of 
order will fail 'gracefully'.

This worked for 'all' and needs a small fixup for the decoder option.
That's noted in Patch 2 reply. I didn't test the other filtering yet
but started updating a unit test to cover these cases.

See https://github.com/pmem/ndctl/blob/main/CONTRIBUTING.md for
a few housekeeping tips on submitting to NDCTL.

-- Alison


> 
> Pawel Mielimonka (2):
>   cxl/cli: add helpers to collect and sort regions by HPA
>   cxl/cli: enforce HPA-descending teardown order for destroy-region
> 
>  cxl/region.c | 114 +++++++++++++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 111 insertions(+), 3 deletions(-)
> 
> -- 
> 2.45.1.windows.1
> 
>