crypto: hisilicon - avoid unused function warning
Subsystems (3)
PATCH v2 1/2 crypto: hisilicon - avoid unused function warning
The only caller of hisi_zip_vf_q_assign() is hidden in an #ifdef, so the function causes a warning when CONFIG_PCI_IOV is disabled:
drivers/crypto/hisilicon/zip/zip_main.c:740:12: error: unused function 'hisi_zip_vf_q_assign' [-Werror,-Wunused-function]
Replace the #ifdef with an IS_ENABLED() check that leads to the function being dropped based on the configuration.
--- drivers/crypto/hisilicon/zip/zip_main.c | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-)
drivers/crypto/hisilicon/zip/zip_main.c +2 -5
@@ -785,7 +785,6 @@ static int hisi_zip_clear_vft_config(struct hisi_zip *hisi_zip)static int hisi_zip_sriov_enable (struct pci_dev *pdev,int max_vfs) { -#ifdef CONFIG_PCI_IOV struct hisi_zip *hisi_zip =pci_get_drvdata (pdev);int pre_existing_vfs, num_vfs, ret;
@@ -815,9 +814,6 @@ static int hisi_zip_sriov_enable(struct pci_dev *pdev, int max_vfs) }return num_vfs; -#else -return 0 ; -#endif }static int hisi_zip_sriov_disable (struct pci_dev *pdev)
@@ -948,7 +944,8 @@ static struct pci_driver hisi_zip_pci_driver = { .id_table = hisi_zip_dev_ids, .probe = hisi_zip_probe, .remove = hisi_zip_remove, - .sriov_configure = hisi_zip_sriov_configure, + .sriov_configure = IS_ENABLED(CONFIG_PCI_IOV) ? + hisi_zip_sriov_configure : 0, .err_handler = &hisi_zip_err_handler, };
Signature · 1 line
2.20.0
PATCH 2/2 crypto: hisilicon - allow compile-testing on x86
To avoid missing arm64 specific warnings that get introduced in this driver, allow compile-testing on all 64-bit architectures.
The only actual arm64 specific code in this driver is an open- coded 128 bit MMIO write. On non-arm64 the same can be done using memcpy_toio. What I also noticed is that the mmio store (either one) is not endian-safe, this will only work on little- endian configurations, so I also add a Kconfig dependency on that, regardless of the architecture.
Finally, a depenndecy on CONFIG_64BIT is needed because of the writeq().
--- drivers/crypto/hisilicon/Kconfig | 8 +++++--- drivers/crypto/hisilicon/qm.c | 6 ++++++ 2 files changed, 11 insertions(+), 3 deletions(-)
drivers/crypto/hisilicon/Kconfig +5 -3
@@ -16,14 +16,15 @@ config CRYPTO_DEV_HISI_SEC config CRYPTO_DEV_HISI_QM tristate - depends on ARM64 && PCI && PCI_MSI + depends on ARM64 || COMPILE_TEST + depends on PCI && PCI_MSI help HiSilicon accelerator engines use a common queue management interface. Specific engine driver may use this module. config CRYPTO_HISI_SGL tristate - depends on ARM64 + depends on ARM64 || COMPILE_TEST help HiSilicon accelerator engines use a common hardware scatterlist interface for data format. Specific engine driver may use this
@@ -31,7 +32,8 @@ config CRYPTO_HISI_SGL config CRYPTO_DEV_HISI_ZIP tristate "Support for HiSilicon ZIP accelerator" - depends on ARM64 && PCI && PCI_MSI + depends on PCI && PCI_MSI + depends on ARM64 || (COMPILE_TEST && 64BIT) select CRYPTO_DEV_HISI_QM select CRYPTO_HISI_SGL select SG_SPLIT
drivers/crypto/hisilicon/qm.c +6 -0
@@ -331,6 +331,12 @@ static void qm_mb_write(struct hisi_qm *qm, const void *src)void __iomem *fun_base = qm->io_base +QM_MB_CMD_SEND_BASE ;unsigned long tmp0 =0 , tmp1 =0 ; +if (!IS_ENABLED (CONFIG_ARM64 )) { +memcpy_toio (fun_base, src,16 ); +wmb (); +return ; + } + asmvolatile ("ldp %0, %1, %3\n" "stp %0, %1, %2\n" "dsb sy\n"
Signature · 1 line
2.20.0
PATCH v2 2/2 crypto: hisilicon - allow compile-testing on x86
To avoid missing arm64 specific warnings that get introduced in this driver, allow compile-testing on all 64-bit architectures.
The only actual arm64 specific code in this driver is an open- coded 128 bit MMIO write. On non-arm64 the same can be done using memcpy_toio. What I also noticed is that the mmio store (either one) is not endian-safe, this will only work on little- endian configurations, so I also add a Kconfig dependency on that, regardless of the architecture.
Finally, a depenndecy on CONFIG_64BIT is needed because of the writeq().
---
v2: actually add !CPU_BIG_ENDIAN dependency as described in the changelog
--- drivers/crypto/hisilicon/Kconfig | 9 ++++++--- drivers/crypto/hisilicon/qm.c | 6 ++++++ 2 files changed, 12 insertions(+), 3 deletions(-)
drivers/crypto/hisilicon/Kconfig +6 -3
@@ -16,14 +16,15 @@ config CRYPTO_DEV_HISI_SEC config CRYPTO_DEV_HISI_QM tristate - depends on ARM64 && PCI && PCI_MSI + depends on ARM64 || COMPILE_TEST + depends on PCI && PCI_MSI help HiSilicon accelerator engines use a common queue management interface. Specific engine driver may use this module. config CRYPTO_HISI_SGL tristate - depends on ARM64 + depends on ARM64 || COMPILE_TEST help HiSilicon accelerator engines use a common hardware scatterlist interface for data format. Specific engine driver may use this
@@ -31,7 +32,9 @@ config CRYPTO_HISI_SGL config CRYPTO_DEV_HISI_ZIP tristate "Support for HiSilicon ZIP accelerator" - depends on ARM64 && PCI && PCI_MSI + depends on PCI && PCI_MSI + depends on ARM64 || (COMPILE_TEST && 64BIT) + depends on !CPU_BIG_ENDIAN || COMPILE_TEST select CRYPTO_DEV_HISI_QM select CRYPTO_HISI_SGL select SG_SPLIT
drivers/crypto/hisilicon/qm.c +6 -0
@@ -331,6 +331,12 @@ static void qm_mb_write(struct hisi_qm *qm, const void *src)void __iomem *fun_base = qm->io_base +QM_MB_CMD_SEND_BASE ;unsigned long tmp0 =0 , tmp1 =0 ; +if (!IS_ENABLED (CONFIG_ARM64 )) { +memcpy_toio (fun_base, src,16 ); +wmb (); +return ; + } + asmvolatile ("ldp %0, %1, %3\n" "stp %0, %1, %2\n" "dsb sy\n"
Signature · 1 line
2.20.0
Re: [PATCH 2/2] [v2] crypto: hisilicon - allow compile-testing on x86
Quoted 10 lines · Arnd Bergmann
To avoid missing arm64 specific warnings that get introduced in this driver, allow compile-testing on all 64-bit architectures.
The only actual arm64 specific code in this driver is an open- coded 128 bit MMIO write. On non-arm64 the same can be done using
memcpy_toio. What I also noticed is that the mmio store (either one) is not endian-safe, this will only work on little- endian configurations, so I also add aKconfigdependency on that, regardless of the architecture.
Finally, a depenndecy onCONFIG_64BITis needed because of the
nit: spelling mistake
Quoted 49 lines
writeq().Signed-off-by Arnd Bergmann <[email protected]>---
v2: actually add !CPU_BIG_ENDIAN dependency as described in the changelog--- drivers/crypto/hisilicon/Kconfig | 9 ++++++--- drivers/crypto/hisilicon/qm.c | 6 ++++++ 2 files changed, 12 insertions(+), 3 deletions(-)drivers/crypto/hisilicon/Kconfig +6
-3@@ -16,14 +16,15 @@ config CRYPTO_DEV_HISI_SEC config CRYPTO_DEV_HISI_QM tristate - depends on ARM64 && PCI && PCI_MSI + depends on ARM64 || COMPILE_TEST + depends on PCI && PCI_MSI help HiSilicon accelerator engines use a common queue management interface. Specific engine driver may use this module. config CRYPTO_HISI_SGL tristate - depends on ARM64 + depends on ARM64 || COMPILE_TEST help HiSilicon accelerator engines use a common hardware scatterlist interface for data format. Specific engine driver may use this@@ -31,7 +32,9 @@ config CRYPTO_HISI_SGL config CRYPTO_DEV_HISI_ZIP tristate "Support for HiSilicon ZIP accelerator" - depends on ARM64 && PCI && PCI_MSI + depends on PCI && PCI_MSI + depends on ARM64 || (COMPILE_TEST && 64BIT) + depends on !CPU_BIG_ENDIAN || COMPILE_TEST select CRYPTO_DEV_HISI_QM select CRYPTO_HISI_SGL select SG_SPLITdrivers/crypto/hisilicon/qm.c +0
-0
drivers/crypto/hisilicon/qm.c static void qm_mb_write(struct hisi_qm *qm, const void *src) · line 333
Quoted 3 lines
drivers/crypto/hisilicon/qm.c +0
-0@@ -331,6 +331,12 @@ static void qm_mb_write(struct hisi_qm *qm, const void *src)void __iomem *fun_base = qm->io_base +QM_MB_CMD_SEND_BASE ;unsigned long tmp0 =0 , tmp1 =0 ;
Hi Arnd,
drivers/crypto/hisilicon/qm.c static void qm_mb_write(struct hisi_qm *qm, const void *src) · line 342
Quoted 10 lines
drivers/crypto/hisilicon/qm.c +6
-0+if (!IS_ENABLED (CONFIG_ARM64 )) { +memcpy_toio (fun_base, src,16 ); +wmb (); +return ; + } + asmvolatile ("ldp %0, %1, %3\n" "stp %0, %1, %2\n" "dsb sy\n"
As I understand, this operation needs to be done atomically. So - even though your change is just for compile testing - the memcpy_to_io() may not do the same thing on other archs, right?
I just wonder if it's right to make that change, or at least warn the imaginary user of possible malfunction for !arm64.
Thanks,
John
Re: [PATCH 2/2] [v2] crypto: hisilicon - allow compile-testing on x86
Quoted 11 lines · John Garry
drivers/crypto/hisilicon/qm.c static void qm_mb_write(struct hisi_qm *qm, const void *src) · line 342
Quoted 10 lines
drivers/crypto/hisilicon/qm.c +6
-0+if (!IS_ENABLED (CONFIG_ARM64 )) { +memcpy_toio (fun_base, src,16 ); +wmb (); +return ; + } + asmvolatile ("ldp %0, %1, %3\n" "stp %0, %1, %2\n" "dsb sy\n"
Quoted 6 lines
As I understand, this operation needs to be done atomically. So - even though your change is just for compile testing - the
memcpy_to_io()may not do the same thing on other archs, right?I just wonder if it's right to make that change, or at least warn the imaginary user of possible malfunction for !arm64.
It's probably not necessary here. From what I can tell from the documentation, this is only safe on ARMv8.4 or higher anyway, earlier ARMv8.x implementations don't guarantee that an stp arrives on the bus in one piece either.
Usually, hardware like this has no hard requirement on an atomic store, it just needs the individual bits to arrive in a particular order, and then triggers the update on the last bit that gets stored. If that is the case here as well, it might actually be better to use two writeq_relaxed() and a barrier. This would also solve the endianess issue.
Arnd
Re: [PATCH 2/2] [v2] crypto: hisilicon - allow compile-testing on x86
Quoted 25 lines · Arnd Bergmann
Quoted 11 lines · John Garry
drivers/crypto/hisilicon/qm.c static void qm_mb_write(struct hisi_qm *qm, const void *src) · line 342
Quoted 10 lines
drivers/crypto/hisilicon/qm.c +6
-0+if (!IS_ENABLED (CONFIG_ARM64 )) { +memcpy_toio (fun_base, src,16 ); +wmb (); +return ; + } + asmvolatile ("ldp %0, %1, %3\n" "stp %0, %1, %2\n" "dsb sy\n"Quoted 6 lines
As I understand, this operation needs to be done atomically. So - even though your change is just for compile testing - the
memcpy_to_io()may not do the same thing on other archs, right?I just wonder if it's right to make that change, or at least warn the imaginary user of possible malfunction for !arm64.
It's probably not necessary here. From what I can tell from the documentation, this is only safe on ARMv8.4 or higher anyway, earlier ARMv8.x implementations don't guarantee that an stp arrives on the bus in one piece either.
Quoted 5 lines
Usually, hardware like this has no hard requirement on an atomic store, it just needs the individual bits to arrive in a particular order, and then triggers the update on the last bit that gets stored. If that is the case here as well, it might actually be better to use two
writeq_relaxed()and a barrier. This would also solve the endianess issue.
See also https://lkml.org/lkml/2018/1/26/554 for a previous attempt to introduce 128-bit MMIO accessors, this got rejected since they are not atomic even on ARMv8.4.
Arnd
Re: [PATCH 2/2] [v2] crypto: hisilicon - allow compile-testing on x86
Quoted 15 lines · Arnd Bergmann
Quoted 14 lines · Arnd Bergmann
Quoted 11 lines · John Garry
Quoted 10 lines
drivers/crypto/hisilicon/qm.c +6
-0+if (!IS_ENABLED (CONFIG_ARM64 )) { +memcpy_toio (fun_base, src,16 ); +wmb (); +return ; + } + asmvolatile ("ldp %0, %1, %3\n" "stp %0, %1, %2\n" "dsb sy\n"
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
As I understand, this operation needs to be done atomically. So - even though your change is just for compile testing - the
memcpy_to_io()may not do the same thing on other archs, right?I just wonder if it's right to make that change, or at least warn the imaginary user of possible malfunction for !arm64.
Hi Arnd,
Quoted 10 lines
Quoted 9 lines
It's probably not necessary here. From what I can tell from the documentation, this is only safe on ARMv8.4 or higher anyway, earlier ARMv8.x implementations don't guarantee that an stp arrives on the bus in one piece either.
Usually, hardware like this has no hard requirement on an atomic store, it just needs the individual bits to arrive in a particular order, and then triggers the update on the last bit that gets stored. If that is the case here as well, it might actually be better to use two
writeq_relaxed()and a barrier. This would also solve the endianess issue.
Quoted 3 lines
See also https://lkml.org/lkml/2018/1/26/554 for a previous attempt to introduce 128-bit MMIO accessors, this got rejected since they are not atomic even on ARMv8.4.
So this is proprietary IP integrated with a proprietary ARMv8 implementation, so there could be a tight coupling, the like of which Will mentioned in that thread, but I'm doubtful.
I'm looking at the electronically translated documentation on this HW, and it reads "The Mailbox operation performed by the CPU cannot be interleaved", and then tells that software should lock against concurrent accesses or alternatively use a 128-bit access. So it seems that the 128b op used is only to guarantee software is atomic.
Wang Zhou can confirm my understanding.
If true, I see that we seem to be already guaranteeing mutual exclusion in qm_mb(), in taking a mutex.
Thanks,
John
Quoted 5 lines
Arnd
.
Zhou Wang
maintainerRe: [PATCH 2/2] [v2] crypto: hisilicon - allow compile-testing on x86
Quoted 44 lines · John Garry
Quoted 15 lines · Arnd Bergmann
Quoted 14 lines · Arnd Bergmann
Quoted 11 lines · John Garry
Quoted 10 lines
drivers/crypto/hisilicon/qm.c +6
-0+if (!IS_ENABLED (CONFIG_ARM64 )) { +memcpy_toio (fun_base, src,16 ); +wmb (); +return ; + } + asmvolatile ("ldp %0, %1, %3\n" "stp %0, %1, %2\n" "dsb sy\n"Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
As I understand, this operation needs to be done atomically. So - even though your change is just for compile testing - the
memcpy_to_io()may not do the same thing on other archs, right?I just wonder if it's right to make that change, or at least warn the imaginary user of possible malfunction for !arm64.
Hi Arnd,
Quoted 10 lines
Quoted 9 lines
It's probably not necessary here. From what I can tell from the documentation, this is only safe on ARMv8.4 or higher anyway, earlier ARMv8.x implementations don't guarantee that an stp arrives on the bus in one piece either.
Usually, hardware like this has no hard requirement on an atomic store, it just needs the individual bits to arrive in a particular order, and then triggers the update on the last bit that gets stored. If that is the case here as well, it might actually be better to use two
writeq_relaxed()and a barrier. This would also solve the endianess issue.Quoted 3 lines
See also https://lkml.org/lkml/2018/1/26/554 for a previous attempt to introduce 128-bit MMIO accessors, this got rejected since they are not atomic even on ARMv8.4.
So this is proprietary IP integrated with a proprietary ARMv8 implementation, so there could be a tight coupling, the like of which Will mentioned in that thread, but I'm doubtful.
Quoted 6 lines
I'm looking at the electronically translated documentation on this HW, and it reads "The Mailbox operation performed by the CPU cannot be interleaved", and then tells that software should lock against concurrent accesses or alternatively use a 128-bit access. So it seems that the 128b op used is only to guarantee software is atomic.
Wang Zhou can confirm my understanding
We have to do a 128bit atomic write here to trigger a mailbox. The reason is that one QM hardware entity in one accelerator servers QM mailbox MMIO interfaces in related PF and VFs.
A mutex can not lock different processing flows in different functions.
As Arnd mentioned, v8.4 extends the support for 16 bytes atomic stp to some kinds of normal memory, but for device memory, it is still implementation defined. For this SoC(Kunpeng920) which has QM/ZIP, if the address is 128bit aligned, stp will be atomic. The offset of QM mailbox is 128bit aligned, so it is safe here.
Best,
Zhou
Quoted 18 lines
If true, I see that we seem to be already guaranteeing mutual exclusion in
qm_mb(), in taking a mutex.Thanks,
JohnQuoted 5 lines
Arnd
.
.
Re: [PATCH 2/2] [v2] crypto: hisilicon - allow compile-testing on x86
Quoted 6 lines · Zhou Wang
Quoted 5 lines · John Garry
Quoted 4 lines · Arnd Bergmann
Quoted 3 lines · Arnd Bergmann
On Fri, Sep 20, 2019 at 10:34 AM John Garry <[email protected]> wrote:
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
drivers/crypto/hisilicon/qm.c static void qm_mb_write(struct hisi_qm *qm, const void *src) · line 342
Quoted 10 lines
drivers/crypto/hisilicon/qm.c +6
-0+if (!IS_ENABLED (CONFIG_ARM64 )) { +memcpy_toio (fun_base, src,16 ); +wmb (); +return ; + } + asmvolatile ("ldp %0, %1, %3\n" "stp %0, %1, %2\n" "dsb sy\n"
***
Quoted 28 lines
Quoted 28 lines
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
As I understand, this operation needs to be done atomically. So - even though your change is just for compile testing - the
memcpy_to_io()may not do the same thing on other archs, right?I just wonder if it's right to make that change, or at least warn the imaginary user of possible malfunction for !arm64.
Hi Arnd,
Quoted 10 lines
Quoted 9 lines
It's probably not necessary here. From what I can tell from the documentation, this is only safe on ARMv8.4 or higher anyway, earlier ARMv8.x implementations don't guarantee that an stp arrives on the bus in one piece either.
Usually, hardware like this has no hard requirement on an atomic store, it just needs the individual bits to arrive in a particular order, and then triggers the update on the last bit that gets stored. If that is the case here as well, it might actually be better to use two
writeq_relaxed()and a barrier. This would also solve the endianess issue.Quoted 3 lines
See also https://lkml.org/lkml/2018/1/26/554 for a previous attempt to introduce 128-bit MMIO accessors, this got rejected since they are not atomic even on ARMv8.4.
So this is proprietary IP integrated with a proprietary ARMv8 implementation, so there could be a tight coupling, the like of which Will mentioned in that thread, but I'm doubtful.
Quoted 7 lines
Quoted 6 lines
I'm looking at the electronically translated documentation on this HW, and it reads "The Mailbox operation performed by the CPU cannot be interleaved", and then tells that software should lock against concurrent accesses or alternatively use a 128-bit access. So it seems that the 128b op used is only to guarantee software is atomic.
Wang Zhou can confirm my understanding
Just to add a few more details here:
Quoted 5 lines
We have to do a 128bit atomic write here to trigger a mailbox. The reason is that one QM hardware entity in one accelerator servers QM mailbox MMIO interfaces in related PF and VFs.
A mutex can not lock different processing flows in different functions.
This means that we can have userspace drivers and the kernel driver simultaneously accessing these mailboxes.
Quoted 5 lines
As Arnd mentioned, v8.4 extends the support for 16 bytes atomic stp to some kinds of normal memory, but for device memory, it is still implementation defined. For this SoC(Kunpeng920) which has QM/ZIP, if the address is 128bit aligned, stp will be atomic. The offset of QM mailbox is 128bit aligned, so it is safe here.
The strange thing (to me) about this hw is that we cannot interleave mailbox accesses on different functions - with 2x 64b accesses, for example. This is the reason that we require 128b accesses.
The upshot is that we can't switch to generic code unfortunately.
But I am not sure if the change is right, as I mentioned originally, above ***.
I'll leave to Arnd's better judgment :)
Thanks,
John
Quoted 26 lines
Best,
ZhouQuoted 9 lines
If true, I see that we seem to be already guaranteeing mutual exclusion in
qm_mb(), in taking a mutex.Thanks,
JohnQuoted 1 line
Quoted 9 lines
Quoted 4 lines
Arnd
.
.
.
Herbert Xu
maintainerRe: [PATCH 2/2] [v2] crypto: hisilicon - allow compile-testing on x86
In reply to Arnd BergmannQuoted 12 lines · Arnd Bergmann
To avoid missing arm64 specific warnings that get introduced in this driver, allow compile-testing on all 64-bit architectures.
The only actual arm64 specific code in this driver is an open- coded 128 bit MMIO write. On non-arm64 the same can be done using
memcpy_toio. What I also noticed is that the mmio store (either one) is not endian-safe, this will only work on little- endian configurations, so I also add aKconfigdependency on that, regardless of the architecture.
Finally, a depenndecy onCONFIG_64BITis needed because of thewriteq().
Quoted 8 lines
Signed-off-by Arnd Bergmann <[email protected]>---
v2: actually add !CPU_BIG_ENDIAN dependency as described in the changelog--- drivers/crypto/hisilicon/Kconfig | 9 ++++++--- drivers/crypto/hisilicon/qm.c | 6 ++++++ 2 files changed, 12 insertions(+), 3 deletions(-)
Patch applied. Thanks.
Signature · 3 lines
Email: Herbert Xu <[email protected]> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
Herbert Xu
maintainerQuoted 5 lines · Arnd Bergmann
The only caller of
hisi_zip_vf_q_assign()is hidden in an #ifdef, so the function causes a warning whenCONFIG_PCI_IOVis disabled:
drivers/crypto/hisilicon/zip/zip_main.c:740:12: error: unused function 'hisi_zip_vf_q_assign' [-Werror,-Wunused-function]
Quoted 8 lines
Replace the #ifdef with an
IS_ENABLED()check that leads to the function being dropped based on the configuration.Fixes 79e09f30eeba ("crypto: hisilicon - add SRIOV support for ZIP")Signed-off-by Arnd Bergmann <[email protected]>--- drivers/crypto/hisilicon/zip/zip_main.c | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-)
Patch applied. Thanks.
Signature · 3 lines
Email: Herbert Xu <[email protected]> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt