Received-SPF: pass (google.com: domain of linux-kernel+bounces-210190-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Message-ID: <b7d247a8759982f3fda64ad250151672452aed84.camel@linux.ibm.com>
Subject: Re: [PATCH v3 1/3] s390/pci: Fix s390_mmio_read/write syscall page
 fault handling
From: Niklas Schnelle <schnelle@linux.ibm.com>
To: David Hildenbrand <david@redhat.com>,
        Gerald Schaefer
 <gerald.schaefer@linux.ibm.com>,
        Heiko Carstens <hca@linux.ibm.com>, Vasily
 Gorbik <gor@linux.ibm.com>,
        Alexander Gordeev <agordeev@linux.ibm.com>,
        Christian Borntraeger <borntraeger@linux.ibm.com>,
        Sven Schnelle
 <svens@linux.ibm.com>,
        Alex Williamson <alex.williamson@redhat.com>,
        Gerd
 Bayer <gbayer@linux.ibm.com>,
        Matthew Rosato <mjrosato@linux.ibm.com>,
        Jason Gunthorpe <jgg@ziepe.ca>, Suren Baghdasaryan <surenb@google.com>
Cc: linux-s390@vger.kernel.org, linux-kernel@vger.kernel.org,
        kvm@vger.kernel.org
Date: Tue, 11 Jun 2024 17:56:46 +0200
In-Reply-To: <89c74380-6a60-4091-ba57-93c75d9a37d7@redhat.com>
References: <20240529-vfio_pci_mmap-v3-0-cd217d019218@linux.ibm.com>
	 <20240529-vfio_pci_mmap-v3-1-cd217d019218@linux.ibm.com>
	 <98de56b1ba37f51639b9a2c15a745e19a45961a0.camel@linux.ibm.com>
	 <30ecb17b7a3414aeb605c51f003582c7f2cf6444.camel@linux.ibm.com>
	 <db10735e74d5a89aed73ad3268e0be40394efc31.camel@linux.ibm.com>
	 <ce7b9655-aaeb-4a13-a3ac-bd4a70bbd173@redhat.com>
	 <32b515269a31e177779f4d2d4fe2c05660beccc4.camel@linux.ibm.com>
	 <89c74380-6a60-4091-ba57-93c75d9a37d7@redhat.com>
Autocrypt: addr=schnelle@linux.ibm.com; prefer-encrypt=mutual;
 keydata=mQINBGHm3M8BEAC+MIQkfoPIAKdjjk84OSQ8erd2OICj98+GdhMQpIjHXn/RJdCZLa58k
 /ay5x0xIHkWzx1JJOm4Lki7WEzRbYDexQEJP0xUia0U+4Yg7PJL4Dg/W4Ho28dRBROoJjgJSLSHwc
 3/1pjpNlSaX/qg3ZM8+/EiSGc7uEPklLYu3gRGxcWV/944HdUyLcnjrZwCn2+gg9ncVJjsimS0ro/
 2wU2RPE4ju6NMBn5Go26sAj1owdYQQv9t0d71CmZS9Bh+2+cLjC7HvyTHKFxVGOznUL+j1a45VrVS
 XQ+nhTVjvgvXR84z10bOvLiwxJZ/00pwNi7uCdSYnZFLQ4S/JGMs4lhOiCGJhJ/9FR7JVw/1t1G9a
 UlqVp23AXwzbcoV2fxyE/CsVpHcyOWGDahGLcH7QeitN6cjltf9ymw2spBzpRnfFn80nVxgSYVG1d
 w75ksBAuQ/3e+oTQk4GAa2ShoNVsvR9GYn7rnsDN5pVILDhdPO3J2PGIXa5ipQnvwb3EHvPXyzakY
 tK50fBUPKk3XnkRwRYEbbPEB7YT+ccF/HioCryqDPWUivXF8qf6Jw5T1mhwukUV1i+QyJzJxGPh19
 /N2/GK7/yS5wrt0Lwxzevc5g+jX8RyjzywOZGHTVu9KIQiG8Pqx33UxZvykjaqTMjo7kaAdGEkrHZ
 dVHqoPZwhCsgQARAQABtChOaWtsYXMgU2NobmVsbGUgPHNjaG5lbGxlQGxpbnV4LmlibS5jb20+iQ
 JXBBMBCABBAhsBBQsJCAcCBhUKCQgLAgQWAgMBAh4BAheAAhkBFiEEnbAAstJ1IDCl9y3cr+Q/Fej
 CYJAFAmWVooIFCQWP+TMACgkQr+Q/FejCYJCmLg/+OgZD6wTjooE77/ZHmW6Egb5nUH6DU+2nMHMH
 UupkE3dKuLcuzI4aEf/6wGG2xF/LigMRrbb1iKRVk/VG/swyLh/OBOTh8cJnhdmURnj3jhaefzslA
 1wTHcxeH4wMGJWVRAhOfDUpMMYV2J5XoroiA1+acSuppelmKAK5voVn9/fNtrVr6mgBXT5RUnmW60
 UUq5z6a1zTMOe8lofwHLVvyG9zMgv6Z9IQJc/oVnjR9PWYDUX4jqFL3yO6DDt5iIQCN8WKaodlNP6
 1lFKAYujV8JY4Ln+IbMIV2h34cGpIJ7f76OYt2XR4RANbOd41+qvlYgpYSvIBDml/fT2vWEjmncm7
 zzpVyPtCZlijV3npsTVerGbh0Ts/xC6ERQrB+rkUqN/fx+dGnTT9I7FLUQFBhK2pIuD+U1K+A+Egw
 UiTyiGtyRMqz12RdWzerRmWFo5Mmi8N1jhZRTs0yAUn3MSCdRHP1Nu3SMk/0oE+pVeni3ysdJ69Sl
 kCAZoaf1TMRdSlF71oT/fNgSnd90wkCHUK9pUJGRTUxgV9NjafZy7sx1Gz11s4QzJE6JBelClBUiF
 6QD4a+MzFh9TkUcpG0cPNsFfEGyxtGzuoeE86sL1tk3yO6ThJSLZyqFFLrZBIJvYK2UiD+6E7VWRW
 9y1OmPyyFBPBosOvmrkLlDtAtyfYInO0KU5pa2xhcyBTY2huZWxsZSA8bmlrbGFzLnNjaG5lbGxlQ
 GlibS5jb20+iQJUBBMBCAA+AhsBBQsJCAcCBhUKCQgLAgQWAgMBAh4BAheAFiEEnbAAstJ1IDCl9y
 3cr+Q/FejCYJAFAmWVoosFCQWP+TMACgkQr+Q/FejCYJB7oxAAksHYU+myhSZD0YSuYZl3oLDUEFP
 3fm9m6N9zgtiOg/GGI0jHc+Tt8qiQaLEtVeP/waWKgQnje/emHJOEDZTb0AdeXZk+T5/ydrKRLmYC
 6rPge3ue1yQUCiA+T72O3WfjZILI2yOstNwd1f0epQ32YaAvM+QbKDloJSmKhGWZlvdVUDXWkS6/m
 aUtUwZpddFY8InXBxsYCbJsqiKF3kPVD515/6keIZmZh1cTIFQ+Kc+UZaz0MxkhiCyWC4cH6HZGKR
 fiXLhPlmmAyW9FiZK9pwDocTLemfgMR6QXOiB0uisdoFnjhXNfp6OHSy7w7LTIHzCsJoHk+vsyvSp
 +fxkjCXgFzGRQaJkoX33QZwQj1mxeWl594QUfR4DIZ2KERRNI0OMYjJVEtB5jQjnD/04qcTrSCpJ5
 ZPtiQ6Umsb1c9tBRIJnL7gIslo/OXBe/4q5yBCtCZOoD6d683XaMPGhi/F6+fnGvzsi6a9qDBgVvt
 arI8ybayhXDuS6/StR8qZKCyzZ/1CUofxGVIdgkseDhts0dZ4AYwRVCUFQULeRtyoT4dKfEot7hPE
 /4wjm9qZf2mDPRvJOqss6jObTNuw1YzGlpe9OvDYtGeEfHgcZqEmHbiMirwfGLaTG2xKDx4g2jd2z
 Ocf83TCERFKJEhvZxB3tRiUQTd3dZ1TIaisv/o+y0K05pa2xhcyBTY2huZWxsZSA8bmlrbGFzLnNj
 aG5lbGxlQGdtYWlsLmNvbT6JAlQEEwEIAD4CGwEFCwkIBwIGFQoJCAsCBBYCAwECHgECF4AWIQSds
 ACy0nUgMKX3Ldyv5D8V6MJgkAUCZZWiiwUJBY/5MwAKCRCv5D8V6MJgkNVuEACo12niyoKhnXLQFt
 NaqxNZ+8p/MGA7g2XcVJ1bYMPoZ2Wh8zwX0sKX/dLlXVHIAeqelL5hIv6GoTykNqQGUN2Kqf0h/z7
 b85o3tHiqMAQV0dAB0y6qdIwdiB69SjpPNK5KKS1+AodLzosdIVKb+LiOyqUFKhLnablni1hiKlqY
 yDeD4k5hePeQdpFixf1YZclGZLFbKlF/A/0Q13USOHuAMYoA/iSgJQDMSUWkuC0mNxdhfVt/gVJnu
 Kq+uKUghcHflhK+yodqezlxmmRxg6HrPVqRG4pZ6YNYO7YXuEWy9JiEH7MmFYcjNdgjn+kxx4IoYU
 O0MJ+DjLpVCV1QP1ZvMy8qQxScyEn7pMpQ0aW6zfJBsvoV3EHCR1emwKYO6rJOfvtu1rElGCTe3sn
 sScV9Z1oXlvo8pVNH5a2SlnsuEBQe0RXNXNJ4RAls8VraGdNSHi4MxcsYEgAVHVaAdTLfJcXZNCIU
 cZejkOE+U2talW2n5sMvx+yURAEVsT/50whYcvomt0y81ImvCgUz4xN1axZ3PCjkgyhNiqLe+vzge
 xq7B2Kx2++hxIBDCKLUTn8JUAtQ1iGBZL9RuDrBy2rR7xbHcU2424iSbP0zmnpav5KUg4F1JVYG12
 vDCi5tq5lORCL28rjOQqE0aLHU1M1D2v51kjkmNuc2pgLDFzpvgLQhTmlrbGFzIFNjaG5lbGxlIDx
 uaWtzQGtlcm5lbC5vcmc+iQJUBBMBCAA+AhsBBQsJCAcCBhUKCQgLAgQWAgMBAh4BAheAFiEEnbAA
 stJ1IDCl9y3cr+Q/FejCYJAFAmWVoosFCQWP+TMACgkQr+Q/FejCYJAglRAAihbDxiGLOWhJed5cF
 kOwdTZz6MyYgazbr+2sFrfAhX3hxPFoG4ogY/BzsjkN0cevWpSigb2I8Y1sQD7BFWJ2OjpEpVQd0D
 sk5VbJBXEWIVDBQ4VMoACLUKgfrb0xiwMRg9C2h6KlwrPBlfgctfvrWWLBq7+oqx73CgxqTcGpfFy
 tD87R4ovR9W1doZbh7pjsH5Ae9xX5PnQFHruib3y35zC8+tvSgvYWv3Eg/8H4QWlrjLHHy2AfZDVl
 9F5t5RfGL8NRsiTdVg9VFYg/GDdck9WPEgdO3L/qoq3Iuk0SZccGl+Nj8vtWYPKNlu2UvgYEbB8cl
 UoWhg+SjjYQka7/p6tc+CCPZ8JUpkgkAdt7yXt6370wP1gct2VztS6SEGcmAE1qxtGhi5Kuln4ZJ/
 UO2yxhPHgoW99OuZw3IRHe0+mNR67JbIpSuFWDFNjZ0nckQcU1taSEUi0euWs7i4MEkm0NsOsVhbs
 4D2vMiC6kO/FqWOPmWZeAjyJw/KRUG4PaJAr5zJUx57nhKWgeTniW712n4DwCUh77D/PHY0nqBTG/
 B+QQCR/FYGpTFkO4DRVfapT8njDrsWyVpP9o64VNZP42S+DuRGWfUKCMAXsM/wPzRiDEVfnZMcUR9
 vwLSHeoV7MiIFC0xIrp5ES9R00t4UFgqtGc36DV71qjR+66Im0=
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.52.2 (3.52.2-1.fc40) 
Precedence: bulk
MIME-Version: 1.0

On Tue, 2024-06-11 at 17:10 +0200, David Hildenbrand wrote:
> > >=20
> > > which checks mmap_assert_write_locked().
> > >=20
> > > Setting VMA flags would be racy with the mmap lock in read mode.
> > >=20
> > >=20
> > > remap_pfn_range() documents: "this is only safe if the mm semaphore i=
s
> > > held when called." which doesn't spell out if it needs to be held in
> > > write mode (which I think it does) :)
> >=20
> > Logically this makes sense to me. At the same time it looks like
> > fixup_user_fault() expects the caller to only hold mmap_read_lock() as
> > I do here. In there it even retakes mmap_read_lock(). But then wouldn't
> > any fault handling by its nature need to hold the write lock?
>=20
> Well, if you're calling remap_pfn_range() right now the expectation is=
=20
> that we hold it in write mode. :)
>=20
> Staring at some random users, they all call it from mmap(), where you=20
> hold the mmap lock in write mode.
>=20
>=20
> I wonder why we are not seeing that splat with vfio all of the time?
>=20
> That mmap lock check was added "recently". In 1c71222e5f23 we started=20
> using vm_flags_set(). That (including the mmap_assert_write_locked())=20
> check was added via bc292ab00f6c almost 1.5 years ago.
>=20
> Maybe vfio is a bit special and was never really run with lockdep?
>=20
> >=20
> > >=20
> > >=20
> > > My best guess is: if you are using remap_pfn_range() from a fault
> > > handler (not during mmap time) you are doing something wrong, that's =
why
> > > you get that report.
> >=20
> > @Alex: I guess so far the vfio_pci_mmap_fault() handler is only ever
> > triggered by "normal"/"actual" page faults where this isn't a problem?
> > Or could it be a problem there too?
> >=20
>=20
> I think we should see it there as well, unless I am missing something.
>=20
> > >=20
> > > vmf_insert_pfn() and friends might be better alternatives, that make
> > > sure that the VMA already received the proper VMA flags at mmap time.
> > >=20
>=20
>=20
> There would be ways of silencing that check: for example, making sure at=
=20
> mmap time that these flags are already set, and skipping modifications=
=20
> if the flags are already set.
>=20
> But, we'll run into more similar checks in x86 VM_PAT code, where we=20
> would do vm_flags_set(vma, VM_PAT) from track_pfn_remap. Some of that=20
> code really doesn't want to be called concurrently (e.g., "vma->vm_pgoff=
=20
> =3D pfn;").
>=20
> I thought that we silenced some of these warnings in the past using=20
> __vm_flags_mod(). But it sounds hacky.
>=20
> CCing Sureen.
>=20

Before I forget it, thanks a lot for your incredible help David!