From: Stefan Berger <stefanb@linux.ibm.com> To: Rob Herring <robh@kernel.org> Cc: kexec@lists.infradead.org, devicetree@vger.kernel.org, Nayna Jain <nayna@linux.ibm.com>, nasastry@in.ibm.com, Frank Rowand <frowand.list@gmail.com>, Eric Biederman <ebiederm@xmission.com> Subject: Re: [PATCH 2/3] tpm/kexec: Duplicate TPM measurement log in of-tree for kexec Date: Wed, 15 Jun 2022 17:45:17 -0400 [thread overview] Message-ID: <08cccbf2-ac65-bd73-d51f-8ea52af2e975@linux.ibm.com> (raw) In-Reply-To: <20220615201453.GA1525994-robh@kernel.org> On 6/15/22 16:14, Rob Herring wrote: > On Wed, Jun 15, 2022 at 09:08:04AM -0400, Stefan Berger wrote: >> >> >> On 6/14/22 13:48, Rob Herring wrote: >>> (),On Tue, Jun 14, 2022 at 10:13 AM Stefan Berger <stefanb@linux.ibm.com> wrote: >>>> >>>> The memory area of the TPM measurement log is currently not properly >>>> duplicated for carrying it across kexec when an Open Firmware >>>> Devicetree is used. Therefore, the contents of the log get corrupted. >>>> Fix this for the kexec_file_load() syscall by allocating a buffer and >>>> copying the contents of the existing log into it. The new buffer is >>>> preserved across the kexec and a pointer to it is available when the new >>>> kernel is started. To achieve this, store the allocated buffer's address >>>> in the flattened device tree (fdt) under the name linux,tpm-kexec-buffer >>>> and search for this entry early in the kernel startup before the TPM >>>> subsystem starts up. Adjust the pointer in the of-tree stored under >>>> linux,sml-base to point to this buffer holding the preserved log. The >>>> TPM driver can then read the base address from this entry when making >>>> the log available. >>> >>> This series really needs a wider Cc list of folks that worry about >>> TPMs and kexec. >>> >>>> Signed-off-by: Stefan Berger <stefanb@linux.ibm.com> >>>> Cc: Rob Herring <robh+dt@kernel.org> >>>> Cc: Frank Rowand <frowand.list@gmail.com> >>>> Cc: Eric Biederman <ebiederm@xmission.com> >>>> --- >>>> drivers/of/device.c | 24 +++++ >>>> drivers/of/kexec.c | 190 +++++++++++++++++++++++++++++++++++++- >>>> include/linux/kexec.h | 6 ++ >>>> include/linux/of.h | 5 + >>>> include/linux/of_device.h | 3 + >>>> kernel/kexec_file.c | 6 ++ >>>> 6 files changed, 233 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/drivers/of/device.c b/drivers/of/device.c >>>> index 874f031442dc..0cbd47b1cabc 100644 >>>> --- a/drivers/of/device.c >>>> +++ b/drivers/of/device.c >>>> @@ -382,3 +382,27 @@ int of_device_uevent_modalias(struct device *dev, struct kobj_uevent_env *env) >>>> return 0; >>>> } >>>> EXPORT_SYMBOL_GPL(of_device_uevent_modalias); >>>> + >>>> +int of_tpm_get_sml_parameters(struct device_node *np, u64 *base, u32 *size) >>> >>> of/device.c is for functions that work on a struct device. This is not >>> the case here. >> >> Can I put it into platform.c? > > That's for struct platform_device things. > Should I create a new file then? >> I should have probably mentioned it but this function here is a copy of code >> from tpm_read_log_of() from here: https://elixir.bootlin.com/linux/latest/source/drivers/char/tpm/eventlog/of.c#L38 >> >> 3/3 refactors the code in tpm/eventlog/of.c to make use of this new function >> then to avoid code duplication. Therefore, this code here is more general >> than necessary at this point. Maybe I should do the move in a patch of its >> own? > > Maybe you should leave that function there and call it? The function would only be callable if TPM subsystem was enabled. That doesn't sound right to make kexec depend on the TPM subsystem. I could just implement a less generic function that only handles the vTPM case then and not target this function to be called from the TPM subsystem via the refactoring in 3/3. > > Generally, subsystem specific things go into the subsystems. However, > there's a few special cases like kexec now. That was added primarily to > avoid per arch duplication. > > I've never looked at the TPM code, so sorry, I don't have more specific > suggestions. > >>>> +{ >>>> + const u32 *sizep; >>>> + const u64 *basep; >>>> + >>>> + sizep = of_get_property(np, "linux,sml-size", NULL); >>>> + basep = of_get_property(np, "linux,sml-base", NULL); >>> >>> Any new properties need a schema. For chosen, that's located here[1]. >>> The more common pattern for similar properties is <base size>. >>> >>> Don't use of_get_property(), but the typed functions instead. >> >> I think this was done due to the little endian and big endian distinction >> below. > > > Right. > >>>> + if (sizep == NULL && basep == NULL) >>>> + return -ENODEV; >>>> + if (sizep == NULL || basep == NULL) >>>> + return -EIO; >>> >>> Do we really need the error distinction? >> >> As I mentioned, this code is a copy from the TPM subsystem. There it does >> make a distinction because similar functions for acpi and efi make a >> distinction as well although this particular function's return code doesn't >> matter in the end. >> >> The code I am referring to is this here: >> >> https://elixir.bootlin.com/linux/latest/source/drivers/char/tpm/eventlog/common.c#L85 >> >>> >>>> + >>>> + if (of_property_match_string(np, "compatible", "IBM,vtpm") < 0 && >>>> + of_property_match_string(np, "compatible", "IBM,vtpm20") < 0) { >>>> + *size = be32_to_cpup((__force __be32 *)sizep); >>>> + *base = be64_to_cpup((__force __be64 *)basep); >>>> + } else { >>>> + *size = *sizep; >>>> + *base = *basep; >>> >>> What? Some platforms aren't properly encoded? Fix the firmware. >> >> It's been like this for years... > > Great! :( > > I'm confused how IBM needs this? Only a LE machine with LE DT encoding > would need this. With Power being historically BE and only recently > (though I guess that's a few years now) supporting LE, how did the DT > encoding become LE for this yet not for everything else in DT? > For some reason the TPM log of a hardware TPM I believe is written in little endian. I am not aware of the history of it. > [...] > > >>>> +/** >>>> + * tpm_post_kexec - Make stored TPM log buffer available in of-tree >>>> + */ >>>> +static int __init tpm_post_kexec(void) >>>> +{ >>>> + struct property *newprop; >>>> + struct device_node *np; >>>> + void *phyaddr; >>>> + size_t size; >>>> + u32 oflogsize; >>>> + u64 unused; >>>> + int ret; >>>> + >>>> + ret = tpm_get_kexec_buffer(&phyaddr, &size); >>>> + if (ret) >>>> + return 0; >>>> + >>>> + /* >>>> + * If any one of the following steps fails then the next kexec will >>>> + * cause issues due to linux,sml-base pointing to a stale buffer. >>>> + */ >>>> + np = of_find_node_by_name(NULL, "vtpm"); >>> >>> This seems pretty IBM specific. >> >> Yes, it is and I am not sure what to do about it. Should I cover parts of >> the function with a #ifdef CONFIG_PPC_PSERIES ? > > #ifdef's aren't great. IS_ENABLED() is a bit better, but really put > implementation specific things in implementation specific code. > > Perhaps each TPM implementation needs its own hook to do stuff? > From what I know it's only vTPM with of-tree that needs this. >>>> + if (!np) >>>> + goto err_free_memblock; >>>> + >>>> + /* logsize must not have changed */ >>>> + if (of_tpm_get_sml_parameters(np, &unused, &oflogsize) < 0) >>>> + goto err_free_memblock; >>>> + if (oflogsize != size) >>>> + goto err_free_memblock; >>>> + >>>> + /* replace linux,sml-base with new physical address of buffer */ >>>> + ret = -ENOMEM; >>>> + newprop = kzalloc(sizeof(*newprop), GFP_KERNEL); >>>> + if (!newprop) >>>> + goto err_free_memblock; >>>> + >>>> + newprop->name = kstrdup("linux,sml-base", GFP_KERNEL); >>>> + if (!newprop->name) >>>> + goto err_free_newprop; >>>> + >>>> + newprop->length = sizeof(phyaddr); >>>> + >>>> + newprop->value = kmalloc(sizeof(u64), GFP_KERNEL); >>>> + if (!newprop->value) >>>> + goto err_free_newprop_struct; >>>> + >>>> + if (of_property_match_string(np, "compatible", "IBM,vtpm") < 0 && >>>> + of_property_match_string(np, "compatible", "IBM,vtpm20") < 0) { >>>> + ret = -ENODEV; >>>> + goto err_free_newprop_struct; >>>> + } else { >>>> + *(u64 *)newprop->value = (u64)phyaddr; >>>> + } >>>> + >>>> + ret = of_update_property(np, newprop); >>> >>> Just FYI for now, there's some work happening on a better API for >>> writing nodes and properties. >> >> Ok. >> >>> >>>> + if (ret) { >>>> + pr_err("Could not update linux,sml-base with new address"); >>>> + goto err_free_newprop_struct; >>>> + } >>>> + >>>> + goto exit; >>>> + >>>> +err_free_newprop_struct: >>>> + kfree(newprop->name); >>>> +err_free_newprop: >>>> + kfree(newprop); >>>> +err_free_memblock: >>>> + memblock_phys_free((phys_addr_t)phyaddr, size); >>>> +exit: >>>> + tpm_of_remove_kexec_buffer(); >>>> + >>>> + return ret; >>>> +} >>>> +postcore_initcall(tpm_post_kexec); >>> >>> Would be better if this is called explicitly at the right time when >>> needed rather than using an initcall. >> >> The 'when needed' would be the TPM subsystem. However, if I was to make it >> dependent on the TPM subsystem we would loose the TPM log if we were not to >> kexec into a kernel with TPM subsystem or the TPM driver wasn't activated. I >> wanted to be able to preserve the log even if a kexec'ed kernel didn't >> support or activate the TPM driver and then a subsequent one again has it >> activated... > > Sounds like a TPM problem the TPM code should deal with. Or a scenario > that just shouldn't be supported. IDK It only comes down to from where tpm_post_kexec() is called. The postcore_initcall() solves the issue this series is addressing in a more general way than if it's called from the TPM subsystem. If this initcall is unacceptable then I guess we would have to call if from the TPM subsystem - no other choice then. Thanks. Stefan
WARNING: multiple messages have this Message-ID (diff)
From: Stefan Berger <stefanb@linux.ibm.com> To: Rob Herring <robh@kernel.org> Cc: kexec@lists.infradead.org, devicetree@vger.kernel.org, Nayna Jain <nayna@linux.ibm.com>, nasastry@in.ibm.com, Frank Rowand <frowand.list@gmail.com>, Eric Biederman <ebiederm@xmission.com> Subject: Re: [PATCH 2/3] tpm/kexec: Duplicate TPM measurement log in of-tree for kexec Date: Wed, 15 Jun 2022 17:45:17 -0400 [thread overview] Message-ID: <08cccbf2-ac65-bd73-d51f-8ea52af2e975@linux.ibm.com> (raw) In-Reply-To: <20220615201453.GA1525994-robh@kernel.org> On 6/15/22 16:14, Rob Herring wrote: > On Wed, Jun 15, 2022 at 09:08:04AM -0400, Stefan Berger wrote: >> >> >> On 6/14/22 13:48, Rob Herring wrote: >>> (),On Tue, Jun 14, 2022 at 10:13 AM Stefan Berger <stefanb@linux.ibm.com> wrote: >>>> >>>> The memory area of the TPM measurement log is currently not properly >>>> duplicated for carrying it across kexec when an Open Firmware >>>> Devicetree is used. Therefore, the contents of the log get corrupted. >>>> Fix this for the kexec_file_load() syscall by allocating a buffer and >>>> copying the contents of the existing log into it. The new buffer is >>>> preserved across the kexec and a pointer to it is available when the new >>>> kernel is started. To achieve this, store the allocated buffer's address >>>> in the flattened device tree (fdt) under the name linux,tpm-kexec-buffer >>>> and search for this entry early in the kernel startup before the TPM >>>> subsystem starts up. Adjust the pointer in the of-tree stored under >>>> linux,sml-base to point to this buffer holding the preserved log. The >>>> TPM driver can then read the base address from this entry when making >>>> the log available. >>> >>> This series really needs a wider Cc list of folks that worry about >>> TPMs and kexec. >>> >>>> Signed-off-by: Stefan Berger <stefanb@linux.ibm.com> >>>> Cc: Rob Herring <robh+dt@kernel.org> >>>> Cc: Frank Rowand <frowand.list@gmail.com> >>>> Cc: Eric Biederman <ebiederm@xmission.com> >>>> --- >>>> drivers/of/device.c | 24 +++++ >>>> drivers/of/kexec.c | 190 +++++++++++++++++++++++++++++++++++++- >>>> include/linux/kexec.h | 6 ++ >>>> include/linux/of.h | 5 + >>>> include/linux/of_device.h | 3 + >>>> kernel/kexec_file.c | 6 ++ >>>> 6 files changed, 233 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/drivers/of/device.c b/drivers/of/device.c >>>> index 874f031442dc..0cbd47b1cabc 100644 >>>> --- a/drivers/of/device.c >>>> +++ b/drivers/of/device.c >>>> @@ -382,3 +382,27 @@ int of_device_uevent_modalias(struct device *dev, struct kobj_uevent_env *env) >>>> return 0; >>>> } >>>> EXPORT_SYMBOL_GPL(of_device_uevent_modalias); >>>> + >>>> +int of_tpm_get_sml_parameters(struct device_node *np, u64 *base, u32 *size) >>> >>> of/device.c is for functions that work on a struct device. This is not >>> the case here. >> >> Can I put it into platform.c? > > That's for struct platform_device things. > Should I create a new file then? >> I should have probably mentioned it but this function here is a copy of code >> from tpm_read_log_of() from here: https://elixir.bootlin.com/linux/latest/source/drivers/char/tpm/eventlog/of.c#L38 >> >> 3/3 refactors the code in tpm/eventlog/of.c to make use of this new function >> then to avoid code duplication. Therefore, this code here is more general >> than necessary at this point. Maybe I should do the move in a patch of its >> own? > > Maybe you should leave that function there and call it? The function would only be callable if TPM subsystem was enabled. That doesn't sound right to make kexec depend on the TPM subsystem. I could just implement a less generic function that only handles the vTPM case then and not target this function to be called from the TPM subsystem via the refactoring in 3/3. > > Generally, subsystem specific things go into the subsystems. However, > there's a few special cases like kexec now. That was added primarily to > avoid per arch duplication. > > I've never looked at the TPM code, so sorry, I don't have more specific > suggestions. > >>>> +{ >>>> + const u32 *sizep; >>>> + const u64 *basep; >>>> + >>>> + sizep = of_get_property(np, "linux,sml-size", NULL); >>>> + basep = of_get_property(np, "linux,sml-base", NULL); >>> >>> Any new properties need a schema. For chosen, that's located here[1]. >>> The more common pattern for similar properties is <base size>. >>> >>> Don't use of_get_property(), but the typed functions instead. >> >> I think this was done due to the little endian and big endian distinction >> below. > > > Right. > >>>> + if (sizep == NULL && basep == NULL) >>>> + return -ENODEV; >>>> + if (sizep == NULL || basep == NULL) >>>> + return -EIO; >>> >>> Do we really need the error distinction? >> >> As I mentioned, this code is a copy from the TPM subsystem. There it does >> make a distinction because similar functions for acpi and efi make a >> distinction as well although this particular function's return code doesn't >> matter in the end. >> >> The code I am referring to is this here: >> >> https://elixir.bootlin.com/linux/latest/source/drivers/char/tpm/eventlog/common.c#L85 >> >>> >>>> + >>>> + if (of_property_match_string(np, "compatible", "IBM,vtpm") < 0 && >>>> + of_property_match_string(np, "compatible", "IBM,vtpm20") < 0) { >>>> + *size = be32_to_cpup((__force __be32 *)sizep); >>>> + *base = be64_to_cpup((__force __be64 *)basep); >>>> + } else { >>>> + *size = *sizep; >>>> + *base = *basep; >>> >>> What? Some platforms aren't properly encoded? Fix the firmware. >> >> It's been like this for years... > > Great! :( > > I'm confused how IBM needs this? Only a LE machine with LE DT encoding > would need this. With Power being historically BE and only recently > (though I guess that's a few years now) supporting LE, how did the DT > encoding become LE for this yet not for everything else in DT? > For some reason the TPM log of a hardware TPM I believe is written in little endian. I am not aware of the history of it. > [...] > > >>>> +/** >>>> + * tpm_post_kexec - Make stored TPM log buffer available in of-tree >>>> + */ >>>> +static int __init tpm_post_kexec(void) >>>> +{ >>>> + struct property *newprop; >>>> + struct device_node *np; >>>> + void *phyaddr; >>>> + size_t size; >>>> + u32 oflogsize; >>>> + u64 unused; >>>> + int ret; >>>> + >>>> + ret = tpm_get_kexec_buffer(&phyaddr, &size); >>>> + if (ret) >>>> + return 0; >>>> + >>>> + /* >>>> + * If any one of the following steps fails then the next kexec will >>>> + * cause issues due to linux,sml-base pointing to a stale buffer. >>>> + */ >>>> + np = of_find_node_by_name(NULL, "vtpm"); >>> >>> This seems pretty IBM specific. >> >> Yes, it is and I am not sure what to do about it. Should I cover parts of >> the function with a #ifdef CONFIG_PPC_PSERIES ? > > #ifdef's aren't great. IS_ENABLED() is a bit better, but really put > implementation specific things in implementation specific code. > > Perhaps each TPM implementation needs its own hook to do stuff? > From what I know it's only vTPM with of-tree that needs this. >>>> + if (!np) >>>> + goto err_free_memblock; >>>> + >>>> + /* logsize must not have changed */ >>>> + if (of_tpm_get_sml_parameters(np, &unused, &oflogsize) < 0) >>>> + goto err_free_memblock; >>>> + if (oflogsize != size) >>>> + goto err_free_memblock; >>>> + >>>> + /* replace linux,sml-base with new physical address of buffer */ >>>> + ret = -ENOMEM; >>>> + newprop = kzalloc(sizeof(*newprop), GFP_KERNEL); >>>> + if (!newprop) >>>> + goto err_free_memblock; >>>> + >>>> + newprop->name = kstrdup("linux,sml-base", GFP_KERNEL); >>>> + if (!newprop->name) >>>> + goto err_free_newprop; >>>> + >>>> + newprop->length = sizeof(phyaddr); >>>> + >>>> + newprop->value = kmalloc(sizeof(u64), GFP_KERNEL); >>>> + if (!newprop->value) >>>> + goto err_free_newprop_struct; >>>> + >>>> + if (of_property_match_string(np, "compatible", "IBM,vtpm") < 0 && >>>> + of_property_match_string(np, "compatible", "IBM,vtpm20") < 0) { >>>> + ret = -ENODEV; >>>> + goto err_free_newprop_struct; >>>> + } else { >>>> + *(u64 *)newprop->value = (u64)phyaddr; >>>> + } >>>> + >>>> + ret = of_update_property(np, newprop); >>> >>> Just FYI for now, there's some work happening on a better API for >>> writing nodes and properties. >> >> Ok. >> >>> >>>> + if (ret) { >>>> + pr_err("Could not update linux,sml-base with new address"); >>>> + goto err_free_newprop_struct; >>>> + } >>>> + >>>> + goto exit; >>>> + >>>> +err_free_newprop_struct: >>>> + kfree(newprop->name); >>>> +err_free_newprop: >>>> + kfree(newprop); >>>> +err_free_memblock: >>>> + memblock_phys_free((phys_addr_t)phyaddr, size); >>>> +exit: >>>> + tpm_of_remove_kexec_buffer(); >>>> + >>>> + return ret; >>>> +} >>>> +postcore_initcall(tpm_post_kexec); >>> >>> Would be better if this is called explicitly at the right time when >>> needed rather than using an initcall. >> >> The 'when needed' would be the TPM subsystem. However, if I was to make it >> dependent on the TPM subsystem we would loose the TPM log if we were not to >> kexec into a kernel with TPM subsystem or the TPM driver wasn't activated. I >> wanted to be able to preserve the log even if a kexec'ed kernel didn't >> support or activate the TPM driver and then a subsequent one again has it >> activated... > > Sounds like a TPM problem the TPM code should deal with. Or a scenario > that just shouldn't be supported. IDK It only comes down to from where tpm_post_kexec() is called. The postcore_initcall() solves the issue this series is addressing in a more general way than if it's called from the TPM subsystem. If this initcall is unacceptable then I guess we would have to call if from the TPM subsystem - no other choice then. Thanks. Stefan _______________________________________________ kexec mailing list kexec@lists.infradead.org http://lists.infradead.org/mailman/listinfo/kexec
next prev parent reply other threads:[~2022-06-15 21:45 UTC|newest] Thread overview: 32+ messages / expand[flat|nested] mbox.gz Atom feed top 2022-06-14 16:12 [PATCH 0/3] tpm: Preserve TPM measurement log across kexec Stefan Berger 2022-06-14 16:12 ` Stefan Berger 2022-06-14 16:12 ` [PATCH 1/3] of: kexec: Refactor IMA buffer related functions to make them reusable Stefan Berger 2022-06-14 16:12 ` Stefan Berger 2022-06-14 16:35 ` Nageswara R Sastry 2022-06-14 16:35 ` Nageswara R Sastry 2022-06-14 16:12 ` [PATCH 2/3] tpm/kexec: Duplicate TPM measurement log in of-tree for kexec Stefan Berger 2022-06-14 16:12 ` Stefan Berger 2022-06-14 16:46 ` Nageswara R Sastry 2022-06-14 16:46 ` Nageswara R Sastry 2022-06-14 17:48 ` Rob Herring 2022-06-14 17:48 ` Rob Herring 2022-06-15 13:08 ` Stefan Berger 2022-06-15 13:08 ` Stefan Berger 2022-06-15 20:14 ` Rob Herring 2022-06-15 20:14 ` Rob Herring 2022-06-15 21:45 ` Stefan Berger [this message] 2022-06-15 21:45 ` Stefan Berger 2022-06-15 20:18 ` Rob Herring 2022-06-15 20:18 ` Rob Herring 2022-06-14 18:58 ` kernel test robot 2022-06-14 18:58 ` kernel test robot 2022-06-15 0:54 ` kernel test robot 2022-06-15 0:54 ` kernel test robot 2022-06-15 9:42 ` kernel test robot 2022-06-15 9:42 ` kernel test robot 2022-06-14 16:12 ` [PATCH 3/3] tpm: of: Call of_tpm_get_sml_parameters() to get base and size of log Stefan Berger 2022-06-14 16:12 ` Stefan Berger 2022-06-14 16:47 ` Nageswara R Sastry 2022-06-14 16:47 ` Nageswara R Sastry 2022-06-15 21:25 ` Jarkko Sakkinen 2022-06-15 21:25 ` Jarkko Sakkinen
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=08cccbf2-ac65-bd73-d51f-8ea52af2e975@linux.ibm.com \ --to=stefanb@linux.ibm.com \ --cc=devicetree@vger.kernel.org \ --cc=ebiederm@xmission.com \ --cc=frowand.list@gmail.com \ --cc=kexec@lists.infradead.org \ --cc=nasastry@in.ibm.com \ --cc=nayna@linux.ibm.com \ --cc=robh@kernel.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.