Note: I took over the work on this feature from Sadre. This series is based on top of v6.14-rc6 and the following series from Stephan Gerhold: https://lore.kernel.org/all/20250212-bam-dma-fixes-v1-0-f560889e65d8@xxxxxxxxxx/ There's also an outstanding problem with the iommus property of the crypto and cryptobam platforms (identified at least on sm8650 and sc7280). This can be worked around by removing them like: diff --git a/arch/arm64/boot/dts/qcom/sc7280.dtsi b/arch/arm64/boot/dts/qcom/sc7280.dtsi index 0f2caf36910b..2dd378d4c299 100644 --- a/arch/arm64/boot/dts/qcom/sc7280.dtsi +++ b/arch/arm64/boot/dts/qcom/sc7280.dtsi @@ -2408,8 +2408,6 @@ cryptobam: dma-controller@1dc4000 { reg = <0x0 0x01dc4000 0x0 0x28000>; interrupts = <GIC_SPI 272 IRQ_TYPE_LEVEL_HIGH>; #dma-cells = <1>; - iommus = <&apps_smmu 0x4e4 0x0011>, - <&apps_smmu 0x4e6 0x0011>; qcom,ee = <0>; qcom,controlled-remotely; num-channels = <16>; @@ -2421,8 +2419,6 @@ crypto: crypto@1dfa000 { reg = <0x0 0x01dfa000 0x0 0x6000>; dmas = <&cryptobam 4>, <&cryptobam 5>; dma-names = "rx", "tx"; - iommus = <&apps_smmu 0x4e4 0x0011>, - <&apps_smmu 0x4e4 0x0011>; interconnects = <&aggre2_noc MASTER_CRYPTO 0 &mc_virt SLAVE_EBI1 0>; interconnect-names = "memory"; }; But a proper fix is being worked on. I'm sending this series out for review anyway but it's v6.16 material. Currently the QCE crypto driver accesses the crypto engine registers directly via CPU. Trust Zone may perform crypto operations simultaneously resulting in a race condition. To avoid this behavior we need to use BAM HW LOCK/UNLOCK feature on BAM pipes, This is achieved by sending a command descriptor performing a dummy write with a relevant flag set. This feature was tested with tcrypt.ko and "libkcapi" with all the AES algorithms supported by QCE crypto engine. Tested on IPQ9574, qcm6490.LE, sc7280 and sm8650. Testing: insmod tcrypt.ko mode=101 insmod tcrypt.ko mode=102 insmod tcrypt.ko mode=155 insmod tcrypt.ko mode=180 insmod tcrypt.ko mode=181 insmod tcrypt.ko mode=182 insmod tcrypt.ko mode=185 insmod tcrypt.ko mode=186 insmod tcrypt.ko mode=212 insmod tcrypt.ko mode=216 insmod tcrypt.ko mode=403 insmod tcrypt.ko mode=404 insmod tcrypt.ko mode=500 insmod tcrypt.ko mode=501 insmod tcrypt.ko mode=502 insmod tcrypt.ko mode=600 insmod tcrypt.ko mode=601 insmod tcrypt.ko mode=602 Encryption command line: ./kcapi -x 1 -e -c "cbc(aes)" -k 8d7dd9b0170ce0b5f2f8e1aa768e01e91da8bfc67fd486d081b28254c99eb423 -i 7fbc02ebf5b93322329df9bfccb635af -p 48981da18e4bb9ef7e2e3162d16b1910 * 8b19050f66582cb7f7e4b6c873819b71 * Decryption command line: * $ ./kcapi -x 1 -c "cbc(aes)" -k 3023b2418ea59a841757dcf07881b3a8def1c97b659a4dad -i 95aa5b68130be6fcf5cabe7d9f898a41 -q c313c6b50145b69a77b33404cb422598 * 836de0065f9d6f6a3dd2c53cd17e33a * $ ./kcapi -x 3 -c sha256 -p 38f86d * cc42f645c5aa76ac3154b023359b665375fc3ae42f025fe961fb0f65205ad70e * $ ./kcapi -x 3 -c sha256 -p bbb300ac5eda9d * 61f7b48577a613fbdfe0d6d90b49985e07a42c99e7a439b6efb76d5ec71b3d30 ./kcapi -x 12 -c "hmac(sha256)" -k 0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b -i 000102030405060708090a0b0c -p f0f1f2f3f4f5f6f7f8f9 -b 42 * 3cb25f25faacd57a90434f64d0362f2a2d2d0a90cf1a5a4c5db02d56ecc4c5bf3400720 8d5b887185865 Paraller test with two different EE's (Execution Environment) EE1 (Trust Zone) EE2 (HLOS) There is a TZ application which "libkcapi" or "tcrypt.ko" will run in will do continuous enc/dec with continuous loop to do enc/dec with different AES algorithm supported different algorithm supported QCE by QCE crypto engine. crypto engine. 1) dummy write with LOCK bit set 1) dummy write with LOCK bit set 2) bam will lock all other pipes 2) bam will lock all other pipes which which not belongs to current not belongs to current EE's, i.e tz EE's, i.e HLOS pipe and keep pipe and keep handling current handling current pipe only. pipe only. 3) hlos prepare data descriptor and 3) tz prepare data descriptor submit to CE5 and submit to CE5 4) dummy write with UNLOCK bit set 4) dummy write with UNLOCK bit 5) bam will release all the locked set pipes 5) bam will release all the locked pipes Upon encountering a descriptor with Lock bit set, the BAM will lock all other pipes not related to the current pipe group, and keep handling the current pipe only until it sees the Un-Lock set (then it will release all locked pipes). The actual locking is done on the new descriptor fetching for publishing, i.e. locked pipe will not fetch new descriptors even if it got event/events adding more descriptors for this pipe. Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx> --- Changes in v7: - remove unused code: writing to multiple registers was not used in v6, neither were the functions for reading registers over BAM DMA- - remove - don't read the SW_VERSION register needlessly in the BAM driver, instead: encode the information on whether the IP supports BAM locking in device match data - shrink code where possible with logic modifications (for instance: change the implementation of qce_write() instead of replacing it everywhere with a new symbol) - remove duplicated error messages - rework commit messages - a lot of shuffling code around for easier review and a more streamlined series - Link to v6: https://lore.kernel.org/all/20250115103004.3350561-1-quic_mdalam@xxxxxxxxxxx/ Changes in v6: - change "BAM" to "DMA" - Ensured this series is compilable with the current Linux-next tip of the tree (TOT). Changes in v5: - Added DMA_PREP_LOCK and DMA_PREP_UNLOCK flag support in separate patch - Removed DMA_PREP_LOCK & DMA_PREP_UNLOCK flag - Added FIELD_GET and GENMASK macro to extract major and minor version Changes in v4: - Added feature description and test hardware with test command - Fixed patch version numbering - Dropped dt-binding patch - Dropped device tree changes - Added BAM_SW_VERSION register read - Handled the error path for the api dma_map_resource() in probe - updated the commit messages for batter redability - Squash the change where qce_bam_acquire_lock() and qce_bam_release_lock() api got introduce to the change where the lock/unlock flag get introced - changed cover letter subject heading to "dmaengine: qcom: bam_dma: add cmd descriptor support" - Added the very initial post for BAM lock/unlock patch link as v1 to track this feature Changes in v3: - https://lore.kernel.org/lkml/183d4f5e-e00a-8ef6-a589-f5704bc83d4a@xxxxxxxxxxx/ - Addressed all the comments from v2 - Added the dt-binding - Fix alignment issue - Removed type casting from qce_write_reg_dma() and qce_read_reg_dma() - Removed qce_bam_txn = dma->qce_bam_txn; line from qce_alloc_bam_txn() api and directly returning dma->qce_bam_txn Changes in v2: - https://lore.kernel.org/lkml/20231214114239.2635325-1-quic_mdalam@xxxxxxxxxxx/ - Initial set of patches for cmd descriptor support - Add client driver to use BAM lock/unlock feature - Added register read/write via BAM in QCE Crypto driver to use BAM lock/unlock feature --- Bartosz Golaszewski (4): dmaengine: qcom: bam_dma: extend the driver's device match data dmaengine: qcom: bam_dma: add bam_pipe_lock flag support crypto: qce - use devres to allocate the result buffer crypto: qce - Switch to using DMA Md Sadre Alam (4): dmaengine: add DMA_PREP_LOCK and DMA_PREP_UNLOCK flag crypto: qce - Map crypto memory for DMA crypto: qce - Add BAM DMA support for crypto register I/O crypto: qce - Add support for BAM locking Documentation/driver-api/dmaengine/provider.rst | 15 ++ drivers/crypto/qce/aead.c | 4 + drivers/crypto/qce/common.c | 47 +++++-- drivers/crypto/qce/common.h | 1 + drivers/crypto/qce/core.c | 24 +++- drivers/crypto/qce/core.h | 7 + drivers/crypto/qce/dma.c | 176 ++++++++++++++++++++++-- drivers/crypto/qce/dma.h | 19 +++ drivers/crypto/qce/sha.c | 4 + drivers/crypto/qce/skcipher.c | 4 + drivers/dma/qcom/bam_dma.c | 42 +++++- include/linux/dmaengine.h | 6 + 12 files changed, 317 insertions(+), 32 deletions(-) --- base-commit: adab295b226c77edf0a90c4c273d6bfed690ec4a change-id: 20250306-qce-cmd-descr-4b966dc29ac1 Best regards, -- Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
