| Message ID | 20240702072510.248272-7-ebiggers@kernel.org (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | Basic inline encryption support for ufs-exynos | expand |
On 7/2/24 12:25 AM, Eric Biggers wrote: > From: Eric Biggers <ebiggers@google.com> > > Add support for Flash Memory Protector (FMP), which is the inline > encryption hardware on Exynos and Exynos-based SoCs. > > Specifically, add support for the "traditional FMP mode" that works on > many Exynos-based SoCs including gs101. This is the mode that uses > "software keys" and is compatible with the upstream kernel's existing > inline encryption framework in the block and filesystem layers. I plan > to add support for the wrapped key support on gs101 at a later time. > > Tested on gs101 (specifically Pixel 6) by running the 'encrypt' group of > xfstests on a filesystem mounted with the 'inlinecrypt' mount option. > > Signed-off-by: Eric Biggers <ebiggers@google.com> > --- > drivers/ufs/host/ufs-exynos.c | 228 +++++++++++++++++++++++++++++++++- > 1 file changed, 222 insertions(+), 6 deletions(-) > > diff --git a/drivers/ufs/host/ufs-exynos.c b/drivers/ufs/host/ufs-exynos.c > index 88d125d1ee3c..dd545ef7c361 100644 > --- a/drivers/ufs/host/ufs-exynos.c > +++ b/drivers/ufs/host/ufs-exynos.c > @@ -6,10 +6,13 @@ > * Author: Seungwon Jeon <essuuj@gmail.com> > * Author: Alim Akhtar <alim.akhtar@samsung.com> > * > */ > > +#include <asm/unaligned.h> > +#include <crypto/aes.h> > +#include <linux/arm-smccc.h> > #include <linux/clk.h> > #include <linux/delay.h> > #include <linux/module.h> > #include <linux/of.h> > #include <linux/of_address.h> > @@ -23,16 +26,18 @@ > #include <ufs/ufshci.h> > #include <ufs/unipro.h> > > #include "ufs-exynos.h" > > +#define DATA_UNIT_SIZE 4096 > +#define LOG2_DATA_UNIT_SIZE 12 If this series has to be reposted, please consider changing "12" into "ilog2(DATA_UNIT_SIZE)". I think that the ilog2() macro generates a constant expression if its argument is a constant. In case it wouldn't be clear, I'm fine with this patch with or without that change. Thanks, Bart.
Hi Eric, Thanks for your contribution, it's great to see new features like this being posted upstream for gs101 :) On Tue, 2 Jul 2024 at 08:28, Eric Biggers <ebiggers@kernel.org> wrote: > > From: Eric Biggers <ebiggers@google.com> > > Add support for Flash Memory Protector (FMP), which is the inline > encryption hardware on Exynos and Exynos-based SoCs. > > Specifically, add support for the "traditional FMP mode" that works on > many Exynos-based SoCs including gs101. This is the mode that uses > "software keys" and is compatible with the upstream kernel's existing > inline encryption framework in the block and filesystem layers. I plan > to add support for the wrapped key support on gs101 at a later time. > > Tested on gs101 (specifically Pixel 6) by running the 'encrypt' group of > xfstests on a filesystem mounted with the 'inlinecrypt' mount option. > > Signed-off-by: Eric Biggers <ebiggers@google.com> > --- > drivers/ufs/host/ufs-exynos.c | 228 +++++++++++++++++++++++++++++++++- > 1 file changed, 222 insertions(+), 6 deletions(-) > > diff --git a/drivers/ufs/host/ufs-exynos.c b/drivers/ufs/host/ufs-exynos.c > index 88d125d1ee3c..dd545ef7c361 100644 > --- a/drivers/ufs/host/ufs-exynos.c > +++ b/drivers/ufs/host/ufs-exynos.c > @@ -6,10 +6,13 @@ > * Author: Seungwon Jeon <essuuj@gmail.com> > * Author: Alim Akhtar <alim.akhtar@samsung.com> > * > */ > > +#include <asm/unaligned.h> > +#include <crypto/aes.h> > +#include <linux/arm-smccc.h> > #include <linux/clk.h> > #include <linux/delay.h> > #include <linux/module.h> > #include <linux/of.h> > #include <linux/of_address.h> > @@ -23,16 +26,18 @@ > #include <ufs/ufshci.h> > #include <ufs/unipro.h> > > #include "ufs-exynos.h" > > +#define DATA_UNIT_SIZE 4096 > +#define LOG2_DATA_UNIT_SIZE 12 > + > /* > * Exynos's Vendor specific registers for UFSHCI > */ > #define HCI_TXPRDT_ENTRY_SIZE 0x00 > #define PRDT_PREFECT_EN BIT(31) > -#define PRDT_SET_SIZE(x) ((x) & 0x1F) > #define HCI_RXPRDT_ENTRY_SIZE 0x04 > #define HCI_1US_TO_CNT_VAL 0x0C > #define CNT_VAL_1US_MASK 0x3FF > #define HCI_UTRL_NEXUS_TYPE 0x40 > #define HCI_UTMRL_NEXUS_TYPE 0x44 > @@ -1041,12 +1046,12 @@ static int exynos_ufs_post_link(struct ufs_hba *hba) > > exynos_ufs_establish_connt(ufs); > exynos_ufs_fit_aggr_timeout(ufs); > > hci_writel(ufs, 0xa, HCI_DATA_REORDER); > - hci_writel(ufs, PRDT_SET_SIZE(12), HCI_TXPRDT_ENTRY_SIZE); > - hci_writel(ufs, PRDT_SET_SIZE(12), HCI_RXPRDT_ENTRY_SIZE); > + hci_writel(ufs, LOG2_DATA_UNIT_SIZE, HCI_TXPRDT_ENTRY_SIZE); > + hci_writel(ufs, LOG2_DATA_UNIT_SIZE, HCI_RXPRDT_ENTRY_SIZE); > hci_writel(ufs, (1 << hba->nutrs) - 1, HCI_UTRL_NEXUS_TYPE); > hci_writel(ufs, (1 << hba->nutmrs) - 1, HCI_UTMRL_NEXUS_TYPE); > hci_writel(ufs, 0xf, HCI_AXIDMA_RWDATA_BURST_LEN); > > if (ufs->opts & EXYNOS_UFS_OPT_SKIP_CONNECTION_ESTAB) > @@ -1149,10 +1154,218 @@ static inline void exynos_ufs_priv_init(struct ufs_hba *hba, > ufs->rx_sel_idx = 0; > hba->priv = (void *)ufs; > hba->quirks = ufs->drv_data->quirks; > } > > +#ifdef CONFIG_SCSI_UFS_CRYPTO > + > +/* > + * Support for Flash Memory Protector (FMP), which is the inline encryption > + * hardware on Exynos and Exynos-based SoCs. The interface to this hardware is > + * not compatible with the standard UFS crypto. It requires that encryption be > + * configured in the PRDT using a nonstandard extension. > + */ > + > +enum fmp_crypto_algo_mode { > + FMP_BYPASS_MODE = 0, > + FMP_ALGO_MODE_AES_CBC = 1, > + FMP_ALGO_MODE_AES_XTS = 2, > +}; > +enum fmp_crypto_key_length { > + FMP_KEYLEN_256BIT = 1, > +}; > + > +/** > + * struct fmp_sg_entry - nonstandard format of PRDT entries when FMP is enabled > + * > + * @base: The standard PRDT entry, but with nonstandard bitfields in the high > + * bits of the 'size' field, i.e. the last 32-bit word. When these > + * nonstandard bitfields are zero, the data segment won't be encrypted or > + * decrypted. Otherwise they specify the algorithm and key length with > + * which the data segment will be encrypted or decrypted. > + * @file_iv: The initialization vector (IV) with all bytes reversed > + * @file_enckey: The first half of the AES-XTS key with all bytes reserved > + * @file_twkey: The second half of the AES-XTS key with all bytes reserved > + * @disk_iv: Unused > + * @reserved: Unused > + */ > +struct fmp_sg_entry { > + struct ufshcd_sg_entry base; > + __be64 file_iv[2]; > + __be64 file_enckey[4]; > + __be64 file_twkey[4]; > + __be64 disk_iv[2]; > + __be64 reserved[2]; > +}; > + > +#define SMC_CMD_FMP_SECURITY \ > + ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL, ARM_SMCCC_SMC_64, \ > + ARM_SMCCC_OWNER_SIP, 0x1810) > +#define SMC_CMD_SMU \ > + ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL, ARM_SMCCC_SMC_64, \ > + ARM_SMCCC_OWNER_SIP, 0x1850) > +#define SMC_CMD_FMP_SMU_RESUME \ > + ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL, ARM_SMCCC_SMC_64, \ > + ARM_SMCCC_OWNER_SIP, 0x1860) > +#define SMU_EMBEDDED 0 > +#define SMU_INIT 0 > +#define CFG_DESCTYPE_3 3 > + > +static void exynos_ufs_fmp_init(struct ufs_hba *hba) > +{ > + struct blk_crypto_profile *profile = &hba->crypto_profile; > + struct arm_smccc_res res; > + int err; > + > + /* > + * Check for the standard crypto support bit, since it's available even > + * though the rest of the interface to FMP is nonstandard. > + * > + * This check should have the effect of preventing the driver from > + * trying to use FMP on old Exynos SoCs that don't have FMP. > + */ > + if (!(ufshcd_readl(hba, REG_CONTROLLER_CAPABILITIES) & > + MASK_CRYPTO_SUPPORT)) > + return; > + Do you know how these FMP registers (FMPSECURITY0 etc) relate to the UFSPR* registers set in the existing exynos_ufs_config_smu()? The UFS_LINK spec talks about UFSPR(FMP), so I had assumed the FMP support would be writing these same registers but via SMC call. I think by the looks of things #define UFSPRSECURITY 0x010 #define UFSPSBEGIN0 0x200 #define UFSPSEND0 0x204 #define UFSPSLUN0 0x208 #define UFSPSCTRL0 0x20C relates to the following registers in gs101 spec FMPSECURITY0 0x0010 FMPSBEGIN0 0x2000 FMPSEND0 0x2004 FMPSLUN0 0x2008 FMPSCTRL0 0x200C And the SMC calls your calling set those same registers as exynos_ufs_config_smu() function. Although it is hard to be certain as I don't have access to the firmware code. Certainly the comment below about FMPSECURITY0 implies that :) With that in mind I think exynos_ufs_fmp_init() function in this patch needs to be better integrated with the EXYNOS_UFS_OPT_UFSPR_SECURE flag and the existing exynos_ufs_config_smu() function that is currently just disabling decryption on platforms where it can access the UFSPR(FMP) regs via mmio. Thanks, Peter p.s. Also whilst reviewing this I noticed a bug where I don't check EXYNOS_UFS_OPT_UFSPR_SECURE flag in exynos_ufs_resume() before calling exynos_ufs_config_smu(). I'll send a patch to fix that. [..]
Hi Peter, On Thu, Jul 04, 2024 at 02:26:05PM +0100, Peter Griffin wrote: > Do you know how these FMP registers (FMPSECURITY0 etc) relate to the > UFSPR* registers set in the existing exynos_ufs_config_smu()? The > UFS_LINK spec talks about UFSPR(FMP), so I had assumed the FMP support > would be writing these same registers but via SMC call. > > I think by the looks of things > > #define UFSPRSECURITY 0x010 > #define UFSPSBEGIN0 0x200 > #define UFSPSEND0 0x204 > #define UFSPSLUN0 0x208 > #define UFSPSCTRL0 0x20C > > relates to the following registers in gs101 spec > > FMPSECURITY0 0x0010 > FMPSBEGIN0 0x2000 > FMPSEND0 0x2004 > FMPSLUN0 0x2008 > FMPSCTRL0 0x200C > > And the SMC calls your calling set those same registers as > exynos_ufs_config_smu() function. Although it is hard to be certain as > I don't have access to the firmware code. Certainly the comment below > about FMPSECURITY0 implies that :) > > With that in mind I think exynos_ufs_fmp_init() function in this patch > needs to be better integrated with the EXYNOS_UFS_OPT_UFSPR_SECURE > flag and the existing exynos_ufs_config_smu() function that is > currently just disabling decryption on platforms where it can access > the UFSPR(FMP) regs via mmio. I think that is all correct. For some reason, on gs101 the FMP registers are not accessible by the "normal world", and SMC calls need to be used instead. The sequences of SMC calls originated from Samsung's Linux driver code for FMP. So I know they are the magic incantations that are needed, but I don't have access to the source code or documentation for them. It does seem clear that one of the things they must do is write the needed values to the FMP registers. I'd hope that these same SMC calls also work on Exynos-based SoCs that do make the FMP registers accessible to the "normal world", and therefore they can just be used on all Exynos-based SoCs and ufs-exynos won't need two different code paths. But I don't have a way to confirm this myself. Until someone is able to confirm this, I think we need to make the FMP support depend on EXYNOS_UFS_OPT_UFSPR_SECURE so that it doesn't conflict with exynos_ufs_config_smu() which runs when !EXYNOS_UFS_OPT_UFSPR_SECURE. - Eric
On Mon, Jul 08, 2024 at 01:26:30PM -0700, Eric Biggers wrote: > Hi Peter, > > On Thu, Jul 04, 2024 at 02:26:05PM +0100, Peter Griffin wrote: > > Do you know how these FMP registers (FMPSECURITY0 etc) relate to the > > UFSPR* registers set in the existing exynos_ufs_config_smu()? The > > UFS_LINK spec talks about UFSPR(FMP), so I had assumed the FMP support > > would be writing these same registers but via SMC call. > > > > I think by the looks of things > > > > #define UFSPRSECURITY 0x010 > > #define UFSPSBEGIN0 0x200 > > #define UFSPSEND0 0x204 > > #define UFSPSLUN0 0x208 > > #define UFSPSCTRL0 0x20C > > > > relates to the following registers in gs101 spec > > > > FMPSECURITY0 0x0010 > > FMPSBEGIN0 0x2000 > > FMPSEND0 0x2004 > > FMPSLUN0 0x2008 > > FMPSCTRL0 0x200C > > > > And the SMC calls your calling set those same registers as > > exynos_ufs_config_smu() function. Although it is hard to be certain as > > I don't have access to the firmware code. Certainly the comment below > > about FMPSECURITY0 implies that :) > > > > With that in mind I think exynos_ufs_fmp_init() function in this patch > > needs to be better integrated with the EXYNOS_UFS_OPT_UFSPR_SECURE > > flag and the existing exynos_ufs_config_smu() function that is > > currently just disabling decryption on platforms where it can access > > the UFSPR(FMP) regs via mmio. > > I think that is all correct. For some reason, on gs101 the FMP registers are > not accessible by the "normal world", and SMC calls need to be used instead. > The sequences of SMC calls originated from Samsung's Linux driver code for FMP. > So I know they are the magic incantations that are needed, but I don't have > access to the source code or documentation for them. It does seem clear that > one of the things they must do is write the needed values to the FMP registers. > > I'd hope that these same SMC calls also work on Exynos-based SoCs that do make > the FMP registers accessible to the "normal world", and therefore they can just > be used on all Exynos-based SoCs and ufs-exynos won't need two different code > paths. But I don't have a way to confirm this myself. Until someone is able to > confirm this, I think we need to make the FMP support depend on > EXYNOS_UFS_OPT_UFSPR_SECURE so that it doesn't conflict with > exynos_ufs_config_smu() which runs when !EXYNOS_UFS_OPT_UFSPR_SECURE. > These same SMC calls can be found in the downstream source for other Exynos-based SoCs. I suspect that exynos_ufs_config_smu() should be removed, and exynos_ufs_fmp_init() should run regardless of EXYNOS_UFS_OPT_UFSPR_SECURE. It still would need to be tested, though, which I'm not able to do. (And especially as a cryptography feature, this *must* be tested...) So for now I'm going to make the FMP support conditional on EXYNOS_UFS_OPT_UFSPR_SECURE. - Eric
Hello Eric, > -----Original Message----- > From: Eric Biggers <ebiggers@kernel.org> > Sent: Tuesday, July 9, 2024 5:19 AM > To: Peter Griffin <peter.griffin@linaro.org> > Cc: linux-scsi@vger.kernel.org; linux-samsung-soc@vger.kernel.org; linux- > fscrypt@vger.kernel.org; Alim Akhtar <alim.akhtar@samsung.com>; Avri > Altman <avri.altman@wdc.com>; Bart Van Assche <bvanassche@acm.org>; > Martin K . Petersen <martin.petersen@oracle.com>; André Draszik > <andre.draszik@linaro.org>; William McVicker <willmcvicker@google.com> > Subject: Re: [PATCH v2 6/6] scsi: ufs: exynos: Add support for Flash Memory > Protector (FMP) > > On Mon, Jul 08, 2024 at 01:26:30PM -0700, Eric Biggers wrote: > > Hi Peter, > > > > On Thu, Jul 04, 2024 at 02:26:05PM +0100, Peter Griffin wrote: > > > Do you know how these FMP registers (FMPSECURITY0 etc) relate to the > > > UFSPR* registers set in the existing exynos_ufs_config_smu()? The > > > UFS_LINK spec talks about UFSPR(FMP), so I had assumed the FMP > > > support would be writing these same registers but via SMC call. > > > > > > I think by the looks of things > > > > > > #define UFSPRSECURITY 0x010 > > > #define UFSPSBEGIN0 0x200 > > > #define UFSPSEND0 0x204 > > > #define UFSPSLUN0 0x208 > > > #define UFSPSCTRL0 0x20C > > > > > > relates to the following registers in gs101 spec > > > > > > FMPSECURITY0 0x0010 > > > FMPSBEGIN0 0x2000 > > > FMPSEND0 0x2004 > > > FMPSLUN0 0x2008 > > > FMPSCTRL0 0x200C > > > > > > And the SMC calls your calling set those same registers as > > > exynos_ufs_config_smu() function. Although it is hard to be certain > > > as I don't have access to the firmware code. Certainly the comment > > > below about FMPSECURITY0 implies that :) > > > > > > With that in mind I think exynos_ufs_fmp_init() function in this > > > patch needs to be better integrated with the > > > EXYNOS_UFS_OPT_UFSPR_SECURE flag and the existing > > > exynos_ufs_config_smu() function that is currently just disabling > > > decryption on platforms where it can access the UFSPR(FMP) regs via > mmio. > > > > I think that is all correct. For some reason, on gs101 the FMP > > registers are not accessible by the "normal world", and SMC calls need to > be used instead. > > The sequences of SMC calls originated from Samsung's Linux driver code > for FMP. > > So I know they are the magic incantations that are needed, but I don't > > have access to the source code or documentation for them. It does > > seem clear that one of the things they must do is write the needed values > to the FMP registers. > > > > I'd hope that these same SMC calls also work on Exynos-based SoCs that > > do make the FMP registers accessible to the "normal world", and > > therefore they can just be used on all Exynos-based SoCs and > > ufs-exynos won't need two different code paths. But I don't have a > > way to confirm this myself. Until someone is able to confirm this, I > > think we need to make the FMP support depend on > > EXYNOS_UFS_OPT_UFSPR_SECURE so that it doesn't conflict with > > exynos_ufs_config_smu() which runs when > !EXYNOS_UFS_OPT_UFSPR_SECURE. > > > > These same SMC calls can be found in the downstream source for other > Exynos-based SoCs. I suspect that exynos_ufs_config_smu() should be > removed, and exynos_ufs_fmp_init() should run regardless of > EXYNOS_UFS_OPT_UFSPR_SECURE. > It still would need to be tested, though, which I'm not able to do. (And > especially as a cryptography feature, this *must* be tested...) So for now I'm > going to make the FMP support conditional on > EXYNOS_UFS_OPT_UFSPR_SECURE. > SMU controls the security access aspect of the FMP, one can have a usecase where one wants to enable inline encryption using FMP in a non-secure mode/world after a secure boot of the system and in another case, configure FMP in secure mode/world during secure boot. I am not sure how it is designed in gs101 though. Currently, exynos_ufs_config_smu() just allows SMU registers modification by non-secure world. > - Eric
On Wed, Jul 10, 2024 at 11:22:52AM +0530, Alim Akhtar wrote: > Hello Eric, > > > -----Original Message----- > > From: Eric Biggers <ebiggers@kernel.org> > > Sent: Tuesday, July 9, 2024 5:19 AM > > To: Peter Griffin <peter.griffin@linaro.org> > > Cc: linux-scsi@vger.kernel.org; linux-samsung-soc@vger.kernel.org; linux- > > fscrypt@vger.kernel.org; Alim Akhtar <alim.akhtar@samsung.com>; Avri > > Altman <avri.altman@wdc.com>; Bart Van Assche <bvanassche@acm.org>; > > Martin K . Petersen <martin.petersen@oracle.com>; André Draszik > > <andre.draszik@linaro.org>; William McVicker <willmcvicker@google.com> > > Subject: Re: [PATCH v2 6/6] scsi: ufs: exynos: Add support for Flash > Memory > > Protector (FMP) > > > > On Mon, Jul 08, 2024 at 01:26:30PM -0700, Eric Biggers wrote: > > > Hi Peter, > > > > > > On Thu, Jul 04, 2024 at 02:26:05PM +0100, Peter Griffin wrote: > > > > Do you know how these FMP registers (FMPSECURITY0 etc) relate to the > > > > UFSPR* registers set in the existing exynos_ufs_config_smu()? The > > > > UFS_LINK spec talks about UFSPR(FMP), so I had assumed the FMP > > > > support would be writing these same registers but via SMC call. > > > > > > > > I think by the looks of things > > > > > > > > #define UFSPRSECURITY 0x010 > > > > #define UFSPSBEGIN0 0x200 > > > > #define UFSPSEND0 0x204 > > > > #define UFSPSLUN0 0x208 > > > > #define UFSPSCTRL0 0x20C > > > > > > > > relates to the following registers in gs101 spec > > > > > > > > FMPSECURITY0 0x0010 > > > > FMPSBEGIN0 0x2000 > > > > FMPSEND0 0x2004 > > > > FMPSLUN0 0x2008 > > > > FMPSCTRL0 0x200C > > > > > > > > And the SMC calls your calling set those same registers as > > > > exynos_ufs_config_smu() function. Although it is hard to be certain > > > > as I don't have access to the firmware code. Certainly the comment > > > > below about FMPSECURITY0 implies that :) > > > > > > > > With that in mind I think exynos_ufs_fmp_init() function in this > > > > patch needs to be better integrated with the > > > > EXYNOS_UFS_OPT_UFSPR_SECURE flag and the existing > > > > exynos_ufs_config_smu() function that is currently just disabling > > > > decryption on platforms where it can access the UFSPR(FMP) regs via > > mmio. > > > > > > I think that is all correct. For some reason, on gs101 the FMP > > > registers are not accessible by the "normal world", and SMC calls need > to > > be used instead. > > > The sequences of SMC calls originated from Samsung's Linux driver code > > for FMP. > > > So I know they are the magic incantations that are needed, but I don't > > > have access to the source code or documentation for them. It does > > > seem clear that one of the things they must do is write the needed > values > > to the FMP registers. > > > > > > I'd hope that these same SMC calls also work on Exynos-based SoCs that > > > do make the FMP registers accessible to the "normal world", and > > > therefore they can just be used on all Exynos-based SoCs and > > > ufs-exynos won't need two different code paths. But I don't have a > > > way to confirm this myself. Until someone is able to confirm this, I > > > think we need to make the FMP support depend on > > > EXYNOS_UFS_OPT_UFSPR_SECURE so that it doesn't conflict with > > > exynos_ufs_config_smu() which runs when > > !EXYNOS_UFS_OPT_UFSPR_SECURE. > > > > > > > These same SMC calls can be found in the downstream source for other > > Exynos-based SoCs. I suspect that exynos_ufs_config_smu() should be > > removed, and exynos_ufs_fmp_init() should run regardless of > > EXYNOS_UFS_OPT_UFSPR_SECURE. > > It still would need to be tested, though, which I'm not able to do. (And > > especially as a cryptography feature, this *must* be tested...) So for > now I'm > > going to make the FMP support conditional on > > EXYNOS_UFS_OPT_UFSPR_SECURE. > > > SMU controls the security access aspect of the FMP, one can have a usecase > where one wants to enable inline encryption using FMP in a non-secure > mode/world after a secure boot of the system > and in another case, configure FMP in secure mode/world during secure boot. > I am not sure how it is designed in gs101 though. > Currently, exynos_ufs_config_smu() just allows SMU registers modification by > non-secure world. > Apparently, gs101 has two levels of access control for FMP. Access to the range configuration registers like FMPSBEGIN0 (UFSPSBEGIN0 in the upstream source) is controlled by FMPSECURITY0.NSSMU (UFSPRSECURITY.NSSMU in the upstream source). But access to FMPSECURITY0 itself is controlled by TZPC.NSFMP, which is writable only by the "secure world". In the current upstream source, exynos_ufs_config_smu() writes to FMPSECURITY0, and this crashes on gs101 if actually executed (it's currently disabled on gs101 for that reason). So the secure world software on gs101 must be setting TZPC.NSFMP = 0. I expect that this isn't specific to gs101, and some other Exynos-based SoCs use this same design. This would explain the presence of the SMC calls in the downstream source used on other Exynos-based SoCs. So it seems that for now ufs-exynos has to use the SMC calls, as this patch does. This is similar to how ufs-qcom and ufs-mediatek similarly use their SoC's respective set of SMC calls to configure inline encryption, in order to work around similar designs where the inline encryption hardware can only be configured by secure world software. (I don't know why so many SoC vendors are choosing this design, given that in practice only the normal world wants to configure inline encryption. The detour to the secure world seems unnecessary.) - Eric
diff --git a/drivers/ufs/host/ufs-exynos.c b/drivers/ufs/host/ufs-exynos.c index 88d125d1ee3c..dd545ef7c361 100644 --- a/drivers/ufs/host/ufs-exynos.c +++ b/drivers/ufs/host/ufs-exynos.c @@ -6,10 +6,13 @@ * Author: Seungwon Jeon <essuuj@gmail.com> * Author: Alim Akhtar <alim.akhtar@samsung.com> * */ +#include <asm/unaligned.h> +#include <crypto/aes.h> +#include <linux/arm-smccc.h> #include <linux/clk.h> #include <linux/delay.h> #include <linux/module.h> #include <linux/of.h> #include <linux/of_address.h> @@ -23,16 +26,18 @@ #include <ufs/ufshci.h> #include <ufs/unipro.h> #include "ufs-exynos.h" +#define DATA_UNIT_SIZE 4096 +#define LOG2_DATA_UNIT_SIZE 12 + /* * Exynos's Vendor specific registers for UFSHCI */ #define HCI_TXPRDT_ENTRY_SIZE 0x00 #define PRDT_PREFECT_EN BIT(31) -#define PRDT_SET_SIZE(x) ((x) & 0x1F) #define HCI_RXPRDT_ENTRY_SIZE 0x04 #define HCI_1US_TO_CNT_VAL 0x0C #define CNT_VAL_1US_MASK 0x3FF #define HCI_UTRL_NEXUS_TYPE 0x40 #define HCI_UTMRL_NEXUS_TYPE 0x44 @@ -1041,12 +1046,12 @@ static int exynos_ufs_post_link(struct ufs_hba *hba) exynos_ufs_establish_connt(ufs); exynos_ufs_fit_aggr_timeout(ufs); hci_writel(ufs, 0xa, HCI_DATA_REORDER); - hci_writel(ufs, PRDT_SET_SIZE(12), HCI_TXPRDT_ENTRY_SIZE); - hci_writel(ufs, PRDT_SET_SIZE(12), HCI_RXPRDT_ENTRY_SIZE); + hci_writel(ufs, LOG2_DATA_UNIT_SIZE, HCI_TXPRDT_ENTRY_SIZE); + hci_writel(ufs, LOG2_DATA_UNIT_SIZE, HCI_RXPRDT_ENTRY_SIZE); hci_writel(ufs, (1 << hba->nutrs) - 1, HCI_UTRL_NEXUS_TYPE); hci_writel(ufs, (1 << hba->nutmrs) - 1, HCI_UTMRL_NEXUS_TYPE); hci_writel(ufs, 0xf, HCI_AXIDMA_RWDATA_BURST_LEN); if (ufs->opts & EXYNOS_UFS_OPT_SKIP_CONNECTION_ESTAB) @@ -1149,10 +1154,218 @@ static inline void exynos_ufs_priv_init(struct ufs_hba *hba, ufs->rx_sel_idx = 0; hba->priv = (void *)ufs; hba->quirks = ufs->drv_data->quirks; } +#ifdef CONFIG_SCSI_UFS_CRYPTO + +/* + * Support for Flash Memory Protector (FMP), which is the inline encryption + * hardware on Exynos and Exynos-based SoCs. The interface to this hardware is + * not compatible with the standard UFS crypto. It requires that encryption be + * configured in the PRDT using a nonstandard extension. + */ + +enum fmp_crypto_algo_mode { + FMP_BYPASS_MODE = 0, + FMP_ALGO_MODE_AES_CBC = 1, + FMP_ALGO_MODE_AES_XTS = 2, +}; +enum fmp_crypto_key_length { + FMP_KEYLEN_256BIT = 1, +}; + +/** + * struct fmp_sg_entry - nonstandard format of PRDT entries when FMP is enabled + * + * @base: The standard PRDT entry, but with nonstandard bitfields in the high + * bits of the 'size' field, i.e. the last 32-bit word. When these + * nonstandard bitfields are zero, the data segment won't be encrypted or + * decrypted. Otherwise they specify the algorithm and key length with + * which the data segment will be encrypted or decrypted. + * @file_iv: The initialization vector (IV) with all bytes reversed + * @file_enckey: The first half of the AES-XTS key with all bytes reserved + * @file_twkey: The second half of the AES-XTS key with all bytes reserved + * @disk_iv: Unused + * @reserved: Unused + */ +struct fmp_sg_entry { + struct ufshcd_sg_entry base; + __be64 file_iv[2]; + __be64 file_enckey[4]; + __be64 file_twkey[4]; + __be64 disk_iv[2]; + __be64 reserved[2]; +}; + +#define SMC_CMD_FMP_SECURITY \ + ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL, ARM_SMCCC_SMC_64, \ + ARM_SMCCC_OWNER_SIP, 0x1810) +#define SMC_CMD_SMU \ + ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL, ARM_SMCCC_SMC_64, \ + ARM_SMCCC_OWNER_SIP, 0x1850) +#define SMC_CMD_FMP_SMU_RESUME \ + ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL, ARM_SMCCC_SMC_64, \ + ARM_SMCCC_OWNER_SIP, 0x1860) +#define SMU_EMBEDDED 0 +#define SMU_INIT 0 +#define CFG_DESCTYPE_3 3 + +static void exynos_ufs_fmp_init(struct ufs_hba *hba) +{ + struct blk_crypto_profile *profile = &hba->crypto_profile; + struct arm_smccc_res res; + int err; + + /* + * Check for the standard crypto support bit, since it's available even + * though the rest of the interface to FMP is nonstandard. + * + * This check should have the effect of preventing the driver from + * trying to use FMP on old Exynos SoCs that don't have FMP. + */ + if (!(ufshcd_readl(hba, REG_CONTROLLER_CAPABILITIES) & + MASK_CRYPTO_SUPPORT)) + return; + + /* + * This call (which sets DESCTYPE to 0x3 in the FMPSECURITY0 register) + * is needed to make the hardware use the larger PRDT entry size. + */ + BUILD_BUG_ON(sizeof(struct fmp_sg_entry) != 128); + arm_smccc_smc(SMC_CMD_FMP_SECURITY, 0, SMU_EMBEDDED, CFG_DESCTYPE_3, + 0, 0, 0, 0, &res); + if (res.a0) { + dev_warn(hba->dev, + "SMC_CMD_FMP_SECURITY failed on init: %ld. Disabling FMP support.\n", + res.a0); + return; + } + ufshcd_set_sg_entry_size(hba, sizeof(struct fmp_sg_entry)); + + /* + * This is needed to initialize FMP. Without it, errors occur when + * inline encryption is used. + */ + arm_smccc_smc(SMC_CMD_SMU, SMU_INIT, SMU_EMBEDDED, 0, 0, 0, 0, 0, &res); + if (res.a0) { + dev_err(hba->dev, + "SMC_CMD_SMU(SMU_INIT) failed: %ld. Disabling FMP support.\n", + res.a0); + return; + } + + /* Advertise crypto capabilities to the block layer. */ + err = devm_blk_crypto_profile_init(hba->dev, profile, 0); + if (err) { + /* Only ENOMEM should be possible here. */ + dev_err(hba->dev, "Failed to initialize crypto profile: %d\n", + err); + return; + } + profile->max_dun_bytes_supported = AES_BLOCK_SIZE; + profile->dev = hba->dev; + profile->modes_supported[BLK_ENCRYPTION_MODE_AES_256_XTS] = + DATA_UNIT_SIZE; + + /* Advertise crypto support to ufshcd-core. */ + hba->caps |= UFSHCD_CAP_CRYPTO; + + /* Advertise crypto quirks to ufshcd-core. */ + hba->quirks |= UFSHCD_QUIRK_CUSTOM_CRYPTO_PROFILE | + UFSHCD_QUIRK_BROKEN_CRYPTO_ENABLE | + UFSHCD_QUIRK_KEYS_IN_PRDT; + +} + +static void exynos_ufs_fmp_resume(struct ufs_hba *hba) +{ + struct arm_smccc_res res; + + arm_smccc_smc(SMC_CMD_FMP_SECURITY, 0, SMU_EMBEDDED, CFG_DESCTYPE_3, + 0, 0, 0, 0, &res); + if (res.a0) + dev_err(hba->dev, + "SMC_CMD_FMP_SECURITY failed on resume: %ld\n", res.a0); + + arm_smccc_smc(SMC_CMD_FMP_SMU_RESUME, 0, SMU_EMBEDDED, 0, 0, 0, 0, 0, + &res); + if (res.a0) + dev_err(hba->dev, + "SMC_CMD_FMP_SMU_RESUME failed: %ld\n", res.a0); +} + +static inline __be64 fmp_key_word(const u8 *key, int j) +{ + return cpu_to_be64(get_unaligned_le64( + key + AES_KEYSIZE_256 - (j + 1) * sizeof(u64))); +} + +/* Fill the PRDT for a request according to the given encryption context. */ +static int exynos_ufs_fmp_fill_prdt(struct ufs_hba *hba, + const struct bio_crypt_ctx *crypt_ctx, + void *prdt, unsigned int num_segments) +{ + struct fmp_sg_entry *fmp_prdt = prdt; + const u8 *enckey = crypt_ctx->bc_key->raw; + const u8 *twkey = enckey + AES_KEYSIZE_256; + u64 dun_lo = crypt_ctx->bc_dun[0]; + u64 dun_hi = crypt_ctx->bc_dun[1]; + unsigned int i; + + /* If FMP wasn't enabled, we shouldn't get any encrypted requests. */ + if (WARN_ON_ONCE(!(hba->caps & UFSHCD_CAP_CRYPTO))) + return -EIO; + + /* Configure FMP on each segment of the request. */ + for (i = 0; i < num_segments; i++) { + struct fmp_sg_entry *prd = &fmp_prdt[i]; + int j; + + /* Each segment must be exactly one data unit. */ + if (prd->base.size != cpu_to_le32(DATA_UNIT_SIZE - 1)) { + dev_err(hba->dev, + "data segment is misaligned for FMP\n"); + return -EIO; + } + + /* Set the algorithm and key length. */ + prd->base.size |= cpu_to_le32((FMP_ALGO_MODE_AES_XTS << 28) | + (FMP_KEYLEN_256BIT << 26)); + + /* Set the IV. */ + prd->file_iv[0] = cpu_to_be64(dun_hi); + prd->file_iv[1] = cpu_to_be64(dun_lo); + + /* Set the key. */ + for (j = 0; j < AES_KEYSIZE_256 / sizeof(u64); j++) { + prd->file_enckey[j] = fmp_key_word(enckey, j); + prd->file_twkey[j] = fmp_key_word(twkey, j); + } + + /* Increment the data unit number. */ + dun_lo++; + if (dun_lo == 0) + dun_hi++; + } + return 0; +} + +#else /* CONFIG_SCSI_UFS_CRYPTO */ + +static void exynos_ufs_fmp_init(struct ufs_hba *hba) +{ +} + +static void exynos_ufs_fmp_resume(struct ufs_hba *hba) +{ +} + +#define exynos_ufs_fmp_fill_prdt NULL + +#endif /* !CONFIG_SCSI_UFS_CRYPTO */ + static int exynos_ufs_init(struct ufs_hba *hba) { struct device *dev = hba->dev; struct platform_device *pdev = to_platform_device(dev); struct exynos_ufs *ufs; @@ -1196,10 +1409,12 @@ static int exynos_ufs_init(struct ufs_hba *hba) goto out; } exynos_ufs_priv_init(hba, ufs); + exynos_ufs_fmp_init(hba); + if (ufs->drv_data->drv_init) { ret = ufs->drv_data->drv_init(dev, ufs); if (ret) { dev_err(dev, "failed to init drv-data\n"); goto out; @@ -1211,11 +1426,11 @@ static int exynos_ufs_init(struct ufs_hba *hba) goto out; exynos_ufs_specify_phy_time_attr(ufs); if (!(ufs->opts & EXYNOS_UFS_OPT_UFSPR_SECURE)) exynos_ufs_config_smu(ufs); - hba->host->dma_alignment = SZ_4K - 1; + hba->host->dma_alignment = DATA_UNIT_SIZE - 1; return 0; out: hba->priv = NULL; return ret; @@ -1330,11 +1545,11 @@ static int exynos_ufs_hce_enable_notify(struct ufs_hba *hba, * The maximum segment size must be set after scsi_host_alloc() * has been called and before LUN scanning starts * (ufshcd_async_scan()). Note: this callback may also be called * from other functions than ufshcd_init(). */ - hba->host->max_segment_size = SZ_4K; + hba->host->max_segment_size = DATA_UNIT_SIZE; if (ufs->drv_data->pre_hce_enable) { ret = ufs->drv_data->pre_hce_enable(ufs); if (ret) return ret; @@ -1430,11 +1645,11 @@ static int exynos_ufs_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op) if (!ufshcd_is_link_active(hba)) phy_power_on(ufs->phy); exynos_ufs_config_smu(ufs); - + exynos_ufs_fmp_resume(hba); return 0; } static int exynosauto_ufs_vh_link_startup_notify(struct ufs_hba *hba, enum ufs_notify_change_status status) @@ -1696,10 +1911,11 @@ static const struct ufs_hba_variant_ops ufs_hba_exynos_ops = { .setup_xfer_req = exynos_ufs_specify_nexus_t_xfer_req, .setup_task_mgmt = exynos_ufs_specify_nexus_t_tm_req, .hibern8_notify = exynos_ufs_hibern8_notify, .suspend = exynos_ufs_suspend, .resume = exynos_ufs_resume, + .fill_crypto_prdt = exynos_ufs_fmp_fill_prdt, }; static struct ufs_hba_variant_ops ufs_hba_exynosauto_vh_ops = { .name = "exynosauto_ufs_vh", .init = exynosauto_ufs_vh_init,