On 14/03/2019 17:12, Guenter Roeck wrote:
On Fri, Mar 15, 2019 at 12:55:15AM +0800, John Garry wrote:
Currently when we request an IO port region, the request is made directly
to the top resource, ioport_resource.
There is an issue here, in that drivers may request an IO port region even
if the IO port region has not even been mapped in (in pci_remap_iospace()).
This may lead to crashes when the PCI host has not enumerated prior to
accessing the IO port region, as below:
root@(none)$ insmod f71805f.ko
[ 32.264016] Unable to handle kernel paging request at virtual address ffff7dfffee0002e
[ 32.279936] Mem abort info:
[ 32.285536] ESR = 0x96000046
[ 32.291657] Exception class = DABT (current EL), IL = 32 bits
[ 32.303542] SET = 0, FnV = 0
[ 32.309661] EA = 0, S1PTW = 0
[ 32.315957] Data abort info:
[ 32.321728] ISV = 0, ISS = 0x00000046
[ 32.329420] CM = 0, WnR = 1
[ 32.335367] swapper pgtable: 4k pages, 48-bit VAs, pgdp = (____ptrval____)
[ 32.349174] [ffff7dfffee0002e] pgd=0000000001392003, pud=0000000001393003, pmd=0000000000000000
[ 32.366652] Internal error: Oops: 96000046 [#1] PREEMPT SMP
[ 32.377835] Modules linked in: f71805f(+)
[ 32.385876] CPU: 4 PID: 2681 Comm: insmod Not tainted 5.0.0-00003-ga6247cc0f8f2 #52
[ 32.401252] Hardware name: Huawei Taishan 2280 /D05, BIOS Hisilicon D05 IT21 Nemo 2.0 RC0 04/18/2018
[ 32.419597] pstate: 80000005 (Nzcv daif -PAN -UAO)
[ 32.429213] pc : logic_outb+0x54/0xb8
[ 32.436556] lr : f71805f_find+0x2c/0x1b8 [f71805f]
[ 32.446167] sp : ffff0000256bba90
[ 32.452808] x29: ffff0000256bba90 x28: ffff000008b544d0
[ 32.463468] x27: ffff0000256bbdf0 x26: 0000000000000100
[ 32.474127] x25: 000000000000002c x24: ffff000011396000
[ 32.484787] x23: ffff0000256bbb3e x22: ffff0000256bbb40
[ 32.495446] x21: ffff000008b591b8 x20: 0000000000000087
[ 32.506106] x19: 000000000000002e x18: ffffffffffffffff
[ 32.516765] x17: 0000000000000000 x16: 0000000000000000
[ 32.527424] x15: 0000000000000400 x14: 0000000000000400
[ 32.538084] x13: 0000000000001832 x12: 0000000000000000
[ 32.548743] x11: ffff801ffbf08ac0 x10: 0000801fead02000
[ 32.559403] x9 : ffff801ffbffe840 x8 : 0000000000000001
[ 32.570062] x7 : 0000000000210d00 x6 : 0000000000000000
[ 32.580721] x5 : ffff801fb3eac380 x4 : ffff801ffbef4b20
[ 32.591381] x3 : 0000000000ffbffe x2 : ffff0000256bbb40
[ 32.602040] x1 : ffff7dfffee0002e x0 : ffff7dfffee00000
[ 32.612701] Process insmod (pid: 2681, stack limit = 0x(____ptrval____))
[ 32.626155] Call trace:
[ 32.631050] logic_outb+0x54/0xb8
[ 32.637693] f71805f_find+0x2c/0x1b8 [f71805f]
[ 32.646607] f71805f_init+0x38/0xe48 [f71805f]
[ 32.655521] do_one_initcall+0x5c/0x178
[ 32.663212] do_init_module+0x58/0x1b0
[ 32.670728] load_module+0x1dc8/0x2178
[ 32.678243] __se_sys_init_module+0x14c/0x1e8
[ 32.686981] __arm64_sys_init_module+0x18/0x20
[ 32.695894] el0_svc_common+0x60/0x100
[ 32.703409] el0_svc_handler+0x2c/0x80
[ 32.710924] el0_svc+0x8/0xc
[ 32.716693] Code: d2bfdc00 f2cfbfe0 f2ffffe0 8b000021 (39000034)
[ 32.728925] ---[ end trace ddb5e493ee686685 ]---
Segmentation fault
root@(none)$
This issue was originally reported in [1].
This patch changes the functionality of request_region() to request a
region of the direct children of the top ioport_resource.
In this, if the IO port region has not been mapped for a particular IO
region, a suitable child region will not exist, and, as such,
request_region() calls will fail.
A side note: there are many drivers in the kernel which fail to even call
request_region() prior to IO port accesses, and they also need to be fixed
(to call request_region() as appropriate).
Hi Guenter,
Isn't that orthogonal to this problem ?
Yes, it is really. I only added the f71805f driver fixup to illustrate
the problem in resource.c and the potential system crash, and how this
patch would fix it.
[1] https://www.spinics.net/lists/linux-pci/msg49821.html
Signed-off-by: John Garry <john.garry@xxxxxxxxxx>
---
include/linux/ioport.h | 6 +++++-
kernel/resource.c | 19 +++++++++++++++++++
2 files changed, 24 insertions(+), 1 deletion(-)
diff --git a/include/linux/ioport.h b/include/linux/ioport.h
index da0ebaec25f0..cf40e1ed8211 100644
--- a/include/linux/ioport.h
+++ b/include/linux/ioport.h
@@ -217,7 +217,7 @@ static inline bool resource_contains(struct resource *r1, struct resource *r2)
/* Convenience shorthand with allocation */
-#define request_region(start,n,name) __request_region(&ioport_resource, (start), (n), (name), 0)
+#define request_region(start,n,name) __request_region_from_children(&ioport_resource, (start), (n), (name), 0)
#define request_muxed_region(start,n,name) __request_region(&ioport_resource, (start), (n), (name), IORESOURCE_MUXED)
#define __request_mem_region(start,n,name, excl) __request_region(&iomem_resource, (start), (n), (name), excl)
#define request_mem_region(start,n,name) __request_region(&iomem_resource, (start), (n), (name), 0)
I can't comment on the merits of the patch (though I am a bit concerned
about its impact and potential source of regressions),but I don't really
understand why it would only apply to request_region() and not to
request_muxed_region(). Is this an oversight or on purpose ?
It's an oversight, which would be fixed.
Thanks,
John
Guenter
.
