linux-kernel - Re: [PATCH 2/3] PCI: endpoint: pci-epf-test: Use dedicated pci_epf

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: <aX49Ucwd1PalCcGr@fedora>
Date: Sat, 31 Jan 2026 18:35:13 +0100
From: Niklas Cassel <cassel@...nel.org>
To: Koichiro Den <den@...inux.co.jp>
Cc: mani@...nel.org, kwilczynski@...nel.org, kishon@...nel.org,
	bhelgaas@...gle.com, corbet@....net, jingoohan1@...il.com,
	lpieralisi@...nel.org, robh@...nel.org, Frank.Li@....com,
	linux-pci@...r.kernel.org, linux-doc@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH 2/3] PCI: endpoint: pci-epf-test: Use dedicated
 pci_epf_bar for subrange mapping

On Sat, Jan 31, 2026 at 10:36:54PM +0900, Koichiro Den wrote:
> The BAR subrange setup/clear paths in pci-epf-test used to update
> epf->bar[barno].submap in place and free/restore the submap around
> pci_epc_set_bar() calls.
>
> Some EPC drivers may keep a reference to the struct pci_epf_bar passed
> to pci_epc_set_bar(). Mutating or freeing the same bar descriptor after
> a successful set_bar() can therefore lead to unexpected behaviour.
>
> Use a dedicated pci_epf_bar instance for the subrange mapping test and
> only free the allocated submap after restoring the BAR mapping back to
> the default epf->bar[barno] descriptor.
>
> Fixes: 6c5e6101423b ("PCI: endpoint: pci-epf-test: Add BAR subrange mapping test support")
> Suggested-by: Niklas Cassel <cassel@...nel.org>
> Signed-off-by: Koichiro Den <den@...inux.co.jp>
> ---
>  drivers/pci/endpoint/functions/pci-epf-test.c | 32 ++++++-------------
>  1 file changed, 10 insertions(+), 22 deletions(-)
>
> diff --git a/drivers/pci/endpoint/functions/pci-epf-test.c b/drivers/pci/endpoint/functions/pci-epf-test.c
> index 6952ee418622..fd6452d1dcc7 100644
> --- a/drivers/pci/endpoint/functions/pci-epf-test.c
> +++ b/drivers/pci/endpoint/functions/pci-epf-test.c
> @@ -86,6 +86,7 @@ struct pci_epf_test {
>	bool			dma_private;
>	const struct pci_epc_features *epc_features;
>	struct pci_epf_bar	db_bar;
> +	struct pci_epf_bar	subrange_bar[PCI_STD_NUM_BARS];

If we compare your test:
pci_epf_test_bar_subrange_setup(), the host side decides which BAR you
want to configure.

For pci_epf_test_enable_doorbell(), the function itself uses
pci_epc_get_next_free_bar(), so the EP side decides which BAR to use.

This is a difference, but I think your way is fine.

Another difference is that you have:
struct pci_epf_bar      subrange_bar[PCI_STD_NUM_BARS];

while the doorbell test case has:

struct pci_epf_bar      db_bar;

Looking at the code, you allow multiple BARs to be configured in subrange
mapping mode (even though the selftest itself will only enable+disable it
for one BAR one by one, but I guess someone could theoretically write their
own test program that puts all the BARs in subrange mapping mode at the same
time).

This is another difference from enable_doorbell(), but again I think your
way is also fine.

Looking at the pci-epf-test code, I realize that, because:
struct pci_epf_bar      db_bar;
is just a single struct, doing ioctl ENABLE_DOORBELL multiple times will
just overwrite the existing db_bar struct... Not very nice...

Since it is only one db_bar, pci-epf-test should return an error if
ENABLE_DOORBELL is called multiple times in a row, rather than just silently
overwrite pci_epf_bar db_bar, leaving the previous BAR still configured
while programming yet another BAR for a HW doorbell... This means that
calling DISABLE_DOORBELL will incorrectly just cleanup one BAR and not two...
This is not your bug however...

TL;DR: I think your code looks fine, even though it is different that
the doorbell test case in a few ways.

Also the doorbell test case is buggy, but that is not really your problem.

Kind regards,
Niklas