Re: [PATCH] platform/x86: intel_pmc_core: Do not ioremap RAM
On Fri, Aug 16, 2019 at 4:42 AM M. Vefa Bicakci wrote:
>
> On a Xen-based PVH virtual machine with more than 4 GiB of RAM,
> intel_pmc_core fails initialization with the following warning message
> from the kernel, indicating that the driver is attempting to ioremap
> RAM:
>
> [ cut here ]
> ioremap on RAM at 0xfe00 - 0xfe001fff
> WARNING: CPU: 1 PID: 434 at arch/x86/mm/ioremap.c:186
> __ioremap_caller.constprop.0+0x2aa/0x2c0
> Modules linked in: intel_pmc_core_pltdrv(+) xen_netfront(+) intel_pmc_core \
> crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel \
> intel_rapl_perf pcspkr binfmt_misc u2mfn(O) xenfs xen_gntdev \
> xen_gntalloc xen_blkback xen_privcmd xen_evtchn ip_tables overlay \
> xen_blkfront
> CPU: 1 PID: 434 Comm: systemd-udevd Tainted: G O
> 5.3.0-rc3-next-20190809-1 #2
> RIP: 0010:__ioremap_caller.constprop.0+0x2aa/0x2c0
> Code: 05 0f c9 1b 01 49 09 c6 e9 8e fe ff ff 48 8d 54 24 28 48 c7 c7 68 87
> 0a 82 c6 05 c4 23 2b 01 01 48 8d 74 24 18 e8 d4 4a 06 00 <0f> 0b 45 31 ff e9
> fe fe ff ff e8 e7 47 06 00 0f 1f 80 00 00 00 00
> RSP: 0018:c92dfa20 EFLAGS: 00010282
> RAX: RBX: c0107000 RCX: 0006
> RDX: 0007 RSI: 0092 RDI: 88813bb17540
> RBP: fe00 R08: 015f R09: 000a
> R10: 000a R11: c92df8da R12: 2000
> R13: R14: 0002 R15: 822da3a0
> FS: 72ecce666940() GS:88813bb0() knlGS:
> CS: 0010 DS: ES: CR0: 80050033
> CR2: 785be7346d20 CR3: 0d872001 CR4: 003606e0
> Call Trace:
>? pmc_core_probe+0x87/0x2d0 [intel_pmc_core]
>pmc_core_probe+0x87/0x2d0 [intel_pmc_core]
>platform_drv_probe+0x35/0x80
>really_probe+0x298/0x3c0
>driver_probe_device+0x7a/0x100
>? driver_allows_async_probing+0x50/0x50
>bus_for_each_drv+0x8f/0xd0
>__device_attach+0xf0/0x170
>bus_probe_device+0x8e/0xa0
>device_add+0x692/0x7e0
>platform_device_add+0xe9/0x240
>? 0xc00ec000
>do_one_initcall+0x59/0x214
>do_init_module+0x5c/0x230
>load_module+0x258d/0x2840
>? __do_sys_finit_module+0xbb/0x120
>__do_sys_finit_module+0xbb/0x120
>do_syscall_64+0x5f/0x1a0
>entry_SYSCALL_64_after_hwframe+0x44/0xa9
> RIP: 0033:0x72eccf662fad
> Code: 00 c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7
> 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff
> 73 01 c3 48 8b 0d ab 9e 0c 00 f7 d8 64 89 01 48
> RSP: 002b:7ffd3710d018 EFLAGS: 0246 ORIG_RAX: 0139
> RAX: ffda RBX: 60ab8fda2530 RCX: 72eccf662fad
> RDX: RSI: 72eccf2bd84d RDI: 0017
> RBP: 0002 R08: R09: 0007
> R10: 0017 R11: 0246 R12: 72eccf2bd84d
> R13: R14: 60ab8fd9b5e0 R15: 0007
> ---[ end trace 5e5421608729d6f4 ]---
>
> This issue appears to manifest itself because of the following fallback
> mechanism in the driver:
>
> if (lpit_read_residency_count_address(_s0_addr))
> pmcdev->base_addr = PMC_BASE_ADDR_DEFAULT;
>
> The validity of address PMC_BASE_ADDR_DEFAULT (i.e., 0xFE00) is not
> verified by the driver, which is what this patch introduces. With this
> patch, if address PMC_BASE_ADDR_DEFAULT is in RAM, then the driver will
> not attempt to ioremap the aforementioned address.
>
Pushed to my review and testing queue, thanks!
P.S. I cut unnecessary output off.
> Signed-off-by: M. Vefa Bicakci
> ---
> drivers/platform/x86/intel_pmc_core.c | 8 ++--
> 1 file changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/platform/x86/intel_pmc_core.c
> b/drivers/platform/x86/intel_pmc_core.c
> index c510d0d72475..3b6b8dcc4767 100644
> --- a/drivers/platform/x86/intel_pmc_core.c
> +++ b/drivers/platform/x86/intel_pmc_core.c
> @@ -878,10 +878,14 @@ static int pmc_core_probe(struct platform_device *pdev)
> if (pmcdev->map == _reg_map && !pci_dev_present(pmc_pci_ids))
> pmcdev->map = _reg_map;
>
> - if (lpit_read_residency_count_address(_s0_addr))
> + if (lpit_read_residency_count_address(_s0_addr)) {
> pmcdev->base_addr = PMC_BASE_ADDR_DEFAULT;
> - else
> +
> + if (page_is_ram(PHYS_PFN(pmcdev->base_addr)))
> + return -ENODEV;
> + } else {
> pmcdev->base_addr = slp_s0_addr - pmcdev->map->slp_s0_offset;
> + }
>
> pmcdev->regbase = ioremap(pmcdev->base_addr,
> pmcdev->map->regmap_length);
> --
> 2.21.0
>
--
With Best Regards,
Andy Shevchenko
Re: [PATCH] platform/x86: intel_pmc_core: Do not ioremap RAM
On Fri, 16 Aug 2019 12:21:58 +0300 Andy Shevchenko wrote: > On Fri, Aug 16, 2019 at 4:42 AM M. Vefa Bicakci wrote: > > > > On a Xen-based PVH virtual machine with more than 4 GiB of RAM, > > intel_pmc_core fails initialization with the following warning message > > from the kernel, indicating that the driver is attempting to ioremap > > RAM: > > > > [ cut here ] > > ioremap on RAM at 0xfe00 - 0xfe001fff > > > This issue appears to manifest itself because of the following fallback > > mechanism in the driver: > > > > if (lpit_read_residency_count_address(_s0_addr)) > > pmcdev->base_addr = PMC_BASE_ADDR_DEFAULT; > > > > The validity of address PMC_BASE_ADDR_DEFAULT (i.e., 0xFE00) is not > > verified by the driver, which is what this patch introduces. With this > > patch, if address PMC_BASE_ADDR_DEFAULT is in RAM, then the driver will > > not attempt to ioremap the aforementioned address. > > Thank you for the patch. Hello Andy, Thank you for reviewing the patch! > Is there anything preventing us to use memremap() in such case? I re-read the documentation for memremap a few times along with taking a look at its code, but I think I am missing an important piece of information. As I understand it, depending on its flags, memremap would allow a section of RAM to be mapped for the PMC driver. The intention with this patch is to prevent the driver from being instantiated when the default/fallback memory address is in RAM, as this issue occurs with a non-administrative virtual machine (domU in Xen terminology) that does not simulate or pass-through a corresponding PMC device. I think that I have misunderstood your review comment though, so I would apppreciate it if you could elaborate. Thanks again for reviewing the patch, Vefa (Please note that my next reply may be delayed by about 10 hours.)
Re: [PATCH] platform/x86: intel_pmc_core: Do not ioremap RAM
On Fri, Aug 16, 2019 at 4:42 AM M. Vefa Bicakci wrote: > > On a Xen-based PVH virtual machine with more than 4 GiB of RAM, > intel_pmc_core fails initialization with the following warning message > from the kernel, indicating that the driver is attempting to ioremap > RAM: > > [ cut here ] > ioremap on RAM at 0xfe00 - 0xfe001fff > This issue appears to manifest itself because of the following fallback > mechanism in the driver: > > if (lpit_read_residency_count_address(_s0_addr)) > pmcdev->base_addr = PMC_BASE_ADDR_DEFAULT; > > The validity of address PMC_BASE_ADDR_DEFAULT (i.e., 0xFE00) is not > verified by the driver, which is what this patch introduces. With this > patch, if address PMC_BASE_ADDR_DEFAULT is in RAM, then the driver will > not attempt to ioremap the aforementioned address. Thank you for the patch. Is there anything preventing us to use memremap() in such case? -- With Best Regards, Andy Shevchenko
[PATCH] platform/x86: intel_pmc_core: Do not ioremap RAM
On a Xen-based PVH virtual machine with more than 4 GiB of RAM,
intel_pmc_core fails initialization with the following warning message
from the kernel, indicating that the driver is attempting to ioremap
RAM:
[ cut here ]
ioremap on RAM at 0xfe00 - 0xfe001fff
WARNING: CPU: 1 PID: 434 at arch/x86/mm/ioremap.c:186
__ioremap_caller.constprop.0+0x2aa/0x2c0
Modules linked in: intel_pmc_core_pltdrv(+) xen_netfront(+) intel_pmc_core \
crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel \
intel_rapl_perf pcspkr binfmt_misc u2mfn(O) xenfs xen_gntdev \
xen_gntalloc xen_blkback xen_privcmd xen_evtchn ip_tables overlay \
xen_blkfront
CPU: 1 PID: 434 Comm: systemd-udevd Tainted: G O
5.3.0-rc3-next-20190809-1 #2
RIP: 0010:__ioremap_caller.constprop.0+0x2aa/0x2c0
Code: 05 0f c9 1b 01 49 09 c6 e9 8e fe ff ff 48 8d 54 24 28 48 c7 c7 68 87 0a
82 c6 05 c4 23 2b 01 01 48 8d 74 24 18 e8 d4 4a 06 00 <0f> 0b 45 31 ff e9 fe fe
ff ff e8 e7 47 06 00 0f 1f 80 00 00 00 00
RSP: 0018:c92dfa20 EFLAGS: 00010282
RAX: RBX: c0107000 RCX: 0006
RDX: 0007 RSI: 0092 RDI: 88813bb17540
RBP: fe00 R08: 015f R09: 000a
R10: 000a R11: c92df8da R12: 2000
R13: R14: 0002 R15: 822da3a0
FS: 72ecce666940() GS:88813bb0() knlGS:
CS: 0010 DS: ES: CR0: 80050033
CR2: 785be7346d20 CR3: 0d872001 CR4: 003606e0
Call Trace:
? pmc_core_probe+0x87/0x2d0 [intel_pmc_core]
pmc_core_probe+0x87/0x2d0 [intel_pmc_core]
platform_drv_probe+0x35/0x80
really_probe+0x298/0x3c0
driver_probe_device+0x7a/0x100
? driver_allows_async_probing+0x50/0x50
bus_for_each_drv+0x8f/0xd0
__device_attach+0xf0/0x170
bus_probe_device+0x8e/0xa0
device_add+0x692/0x7e0
platform_device_add+0xe9/0x240
? 0xc00ec000
do_one_initcall+0x59/0x214
do_init_module+0x5c/0x230
load_module+0x258d/0x2840
? __do_sys_finit_module+0xbb/0x120
__do_sys_finit_module+0xbb/0x120
do_syscall_64+0x5f/0x1a0
entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x72eccf662fad
Code: 00 c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7 48
89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01
c3 48 8b 0d ab 9e 0c 00 f7 d8 64 89 01 48
RSP: 002b:7ffd3710d018 EFLAGS: 0246 ORIG_RAX: 0139
RAX: ffda RBX: 60ab8fda2530 RCX: 72eccf662fad
RDX: RSI: 72eccf2bd84d RDI: 0017
RBP: 0002 R08: R09: 0007
R10: 0017 R11: 0246 R12: 72eccf2bd84d
R13: R14: 60ab8fd9b5e0 R15: 0007
---[ end trace 5e5421608729d6f4 ]---
This issue appears to manifest itself because of the following fallback
mechanism in the driver:
if (lpit_read_residency_count_address(_s0_addr))
pmcdev->base_addr = PMC_BASE_ADDR_DEFAULT;
The validity of address PMC_BASE_ADDR_DEFAULT (i.e., 0xFE00) is not
verified by the driver, which is what this patch introduces. With this
patch, if address PMC_BASE_ADDR_DEFAULT is in RAM, then the driver will
not attempt to ioremap the aforementioned address.
Signed-off-by: M. Vefa Bicakci
---
drivers/platform/x86/intel_pmc_core.c | 8 ++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/drivers/platform/x86/intel_pmc_core.c
b/drivers/platform/x86/intel_pmc_core.c
index c510d0d72475..3b6b8dcc4767 100644
--- a/drivers/platform/x86/intel_pmc_core.c
+++ b/drivers/platform/x86/intel_pmc_core.c
@@ -878,10 +878,14 @@ static int pmc_core_probe(struct platform_device *pdev)
if (pmcdev->map == _reg_map && !pci_dev_present(pmc_pci_ids))
pmcdev->map = _reg_map;
- if (lpit_read_residency_count_address(_s0_addr))
+ if (lpit_read_residency_count_address(_s0_addr)) {
pmcdev->base_addr = PMC_BASE_ADDR_DEFAULT;
- else
+
+ if (page_is_ram(PHYS_PFN(pmcdev->base_addr)))
+ return -ENODEV;
+ } else {
pmcdev->base_addr = slp_s0_addr - pmcdev->map->slp_s0_offset;
+ }
pmcdev->regbase = ioremap(pmcdev->base_addr,
pmcdev->map->regmap_length);
--
2.21.0
