From: Bartosz Golaszewski bartosz.golaszewski@linaro.org
[ Upstream commit 87be3e7a2d0030cda6314d2ec96b37991f636ccd ]
Now that qcom_scm_shm_bridge_enable() uses the struct device passed to it as argument to make the QCOM_SCM_MP_SHM_BRIDGE_ENABLE SCM call, we can move the TZMem initialization before the assignment of the __scm pointer in the SCM driver (which marks SCM as ready to users) thus fixing the potential race between consumer calls and the memory pool initialization.
Reported-by: Johan Hovold johan+linaro@kernel.org Closes: https://lore.kernel.org/all/20250120151000.13870-1-johan+linaro@kernel.org/ Signed-off-by: Bartosz Golaszewski bartosz.golaszewski@linaro.org Link: https://lore.kernel.org/r/20250630-qcom-scm-race-v2-3-fa3851c98611@linaro.or... Signed-off-by: Bjorn Andersson andersson@kernel.org Signed-off-by: Sasha Levin sashal@kernel.org ---
LLM Generated explanations, may be completely bogus:
YES
This commit should be backported to stable kernel trees. The commit fixes a critical race condition in the Qualcomm SCM (Secure Channel Manager) driver that can lead to system crashes.
**Analysis of the race condition:**
Looking at the code changes, the commit moves the TrustZone memory (TZMem) pool initialization from **after** the SCM is marked as available (line 2254 in the original code: `smp_store_release(&__scm, scm)`) to **before** it. This is a critical ordering fix because:
1. **The race window**: Once `__scm` is assigned via `smp_store_release(&__scm, scm)` at line 2254, the SCM API becomes available to all kernel consumers through `qcom_scm_is_available()` which checks this pointer.
2. **The problem**: Between lines 2254-2298 in the original code, the SCM is marked as available but the TZMem pool (`__scm->mempool`) hasn't been initialized yet. If any SCM consumer makes an API call during this window that requires memory allocation from the TZMem pool, it will access an uninitialized pointer, causing a crash.
3. **The fix**: The commit moves the TZMem initialization (lines 2286-2298 in original) to lines 2253-2277 in the patched version, ensuring the memory pool is fully initialized before marking SCM as available.
**Why this qualifies for stable backport:**
1. **Fixes a real bug**: This addresses a genuine race condition that can cause kernel crashes, as reported by Johan Hovold.
2. **Security-critical subsystem**: The SCM driver handles secure communication with TrustZone firmware on Qualcomm platforms, making stability crucial.
3. **Small, contained fix**: The change is minimal - it simply reorders initialization steps without changing functionality or adding features.
4. **No architectural changes**: This is purely a bug fix that corrects initialization ordering.
5. **Clear problem and solution**: The race condition is well-defined, and the fix is straightforward and obvious.
6. **Minimal regression risk**: Moving initialization earlier in the probe sequence is a safe change that doesn't affect the driver's operation once initialized.
The commit message also references a specific bug report, indicating this is a real issue encountered in production, not a theoretical problem. For stable kernel trees supporting Qualcomm platforms, this fix prevents potential crashes during system initialization.
drivers/firmware/qcom/qcom_scm.c | 53 ++++++++++++++++---------------- 1 file changed, 26 insertions(+), 27 deletions(-)
diff --git a/drivers/firmware/qcom/qcom_scm.c b/drivers/firmware/qcom/qcom_scm.c index f63b716be5b0..4e510eb70c79 100644 --- a/drivers/firmware/qcom/qcom_scm.c +++ b/drivers/firmware/qcom/qcom_scm.c @@ -2250,7 +2250,32 @@ static int qcom_scm_probe(struct platform_device *pdev) if (ret) return ret;
- /* Paired with smp_load_acquire() in qcom_scm_is_available(). */ + ret = of_reserved_mem_device_init(scm->dev); + if (ret && ret != -ENODEV) + return dev_err_probe(scm->dev, ret, + "Failed to setup the reserved memory region for TZ mem\n"); + + ret = qcom_tzmem_enable(scm->dev); + if (ret) + return dev_err_probe(scm->dev, ret, + "Failed to enable the TrustZone memory allocator\n"); + + memset(&pool_config, 0, sizeof(pool_config)); + pool_config.initial_size = 0; + pool_config.policy = QCOM_TZMEM_POLICY_ON_DEMAND; + pool_config.max_size = SZ_256K; + + scm->mempool = devm_qcom_tzmem_pool_new(scm->dev, &pool_config); + if (IS_ERR(scm->mempool)) + return dev_err_probe(scm->dev, PTR_ERR(scm->mempool), + "Failed to create the SCM memory pool\n"); + + /* + * Paired with smp_load_acquire() in qcom_scm_is_available(). + * + * This marks the SCM API as ready to accept user calls and can only + * be called after the TrustZone memory pool is initialized. + */ smp_store_release(&__scm, scm);
irq = platform_get_irq_optional(pdev, 0); @@ -2283,32 +2308,6 @@ static int qcom_scm_probe(struct platform_device *pdev) if (of_property_read_bool(pdev->dev.of_node, "qcom,sdi-enabled") || !download_mode) qcom_scm_disable_sdi();
- ret = of_reserved_mem_device_init(__scm->dev); - if (ret && ret != -ENODEV) { - dev_err_probe(__scm->dev, ret, - "Failed to setup the reserved memory region for TZ mem\n"); - goto err; - } - - ret = qcom_tzmem_enable(__scm->dev); - if (ret) { - dev_err_probe(__scm->dev, ret, - "Failed to enable the TrustZone memory allocator\n"); - goto err; - } - - memset(&pool_config, 0, sizeof(pool_config)); - pool_config.initial_size = 0; - pool_config.policy = QCOM_TZMEM_POLICY_ON_DEMAND; - pool_config.max_size = SZ_256K; - - __scm->mempool = devm_qcom_tzmem_pool_new(__scm->dev, &pool_config); - if (IS_ERR(__scm->mempool)) { - ret = dev_err_probe(__scm->dev, PTR_ERR(__scm->mempool), - "Failed to create the SCM memory pool\n"); - goto err; - } - /* * Initialize the QSEECOM interface. *